ISSN 0256-4602

IETE TECHNICAL REVIEW

The Institution of Electronics and Telecommunication Engineers

Let us do great things together

SUBSCRIBER COPY : NOT FOR RESALE

‘Interactive ECTs - A Focus on Rural India’

Special Issue
to commemorate
th
50 Annual Technical Convention

VOLUME 24 NUMBER 4 JULY-AUGUST 2007

IETE TECHNICAL REVIEW The Institution of Electronics and Telecommunication Engineers

President S Narayana IETE Technical Review is published bimonthly by

the Institution of Electronics and Telecommunication
Vice-Presidents A K Agarwal Engineers. All rights of publication are reserved by the
IETE.
R P Bajpai
Copyright and Reprint permission : Abstracting is
Davinder Kumar permitted with credit to the source. Libraries are permit-
ted to photocopy for private use of readers. Annual
Publications Committee Subscription : Subscription and Advertising rates are
available on request and can be seen on website: iete.org
Chairman H O Agrawal The IETE Technical Review invites articles
preferably readable without mathematical expressions,
Co-Chairman K S Prakash Rao state-of-the-art review papers on current and futuristic
technologies in the areas of electronics,
Members T K De telecommunication, computer science & engineering,
information technology (IT) and related disciplines. In
S N Gupta addition, informative and general interest articles
H Kaushal describing innovative products & applications, analysis
S K Kshirsagar of technical events, articles on technology assessment
& comparison, new & emerging topics of interest to
K Babu Rao
professionals are also welcome. While all the papers
T S Rathore submitted will go through the same detailed review
R P Singh process, short papers and Practical Designs will receive
special attention to enable early publication. Manuscripts
B P Srivastava
may please be submitted in triplicate to the Managing
Editor along with a soft copy on floppy/CD/e-mail.
Coopted A K Bhatnagar
Detailed guidelines to authors may be seen on IETE
Dilip Sahay Website : http://www.iete.org under the heading
Special Invitee M Jagadesh Kumar Publications.
Address for correspondence :
EDITORIAL BOARD Managing Editor
IETE
Chairman Dilip Sahay 2, Institutional Area
Members H O Agrawal Lodi Road,
New Delhi 110 003
A K Bhatnagar
R G Gupta Telephone : +91 (011)-2431820/30/40/50/ Ext:43
S S Motial Telefax : +91 (011) 24649429
email : ietend@gmail.com
Neeru Mohan Biswas
mgeditor@iete.org
H Kaushal Website : http://www.iete.org
Secretary General V K Panday FREE TO IETE CORPORATE MEMBERS
(Cost of Production: Rs.19.00)
Dy Managing Editor A P Sharma

COPYRIGHT
It is the IETE policy to own the copyright of the scientific and technical papers it publishes on behalf of
the authors and their employers, and to facilitate the appropriate reuse of this material by others. Authors
are required to sign an IETE copyright transfer form before publication. A copy of this form is available
in the most recent January issue of IETE publications, and online in IETE website:www.iete.org. IETE
retains the authors’ and their employers’ right to reuse their material for their own purposes.
 IETE TECHNICAL REVIEW
Published bimonthly by the Institution of Electronics and Telecommunication Engineers

July-August 2007 Vol 24 No 4

Special Issue
on
‘Interactive ECTs - A Focus on Rural India’

CONTENTS

192 Guest Editorial 271 Building a Strong Nation - The ECT way
B L Deekshatulu R Sreehari Rao

277 Satellite Technology Utilization for Rural

INVITED PAPERS
and Urban India
195 Broadband to Empower Rural India S Pal and V S Rao
Ashok Jhunjhunwala, David Koilpillai and
287 A Programmable Built-in Self-Test for
Bhaskar Ramamurthi
Embedded Memory Cores
203 MIMO Communications - Motivation and Shibaji Banerjee, Dipanwita Roy Chowdhury
a Practical Realization and Bhargab B Bhattacharya
G Kalyana Krishnan and V Umapathi Reddy
313 Space Enabled ICT Applications for Rural
Upliftment - Experience of Participatory
215 Role of Satellite Communication and
Watershed Development
Remote Sensing in Rural Development
P D Diwakar and V Jayaraman
A S Manjunath, D S Jain, Rajendra Kumar and R
V G Anjaneyulu
CONTRIBUTED PAPER
225 Design of a TDD Multisector TDM MAC for
the WiFiRe Proposal for Rural Broadband
323 Performance Analysis of SC DS-CDMA
Access
and MC DS-CDMA Systems over
Anitha Varghese and Anurag Kumar Nakagami-m Fading Channel
S Anuradha, S Srigowri, K S Rama Krishna and
243 Trends in VLSI Technology - Rural
K V V S Reddy
Applications Perspective
K Lal Kishore
SHORT PAPERS
249 Temporally Adaptive, Partially
Unsupervised Classifiers for Remote 331 Design of a FIR Filtering Core for High
Sensing Images Speed Application
Shilpa Inamdar and Subhasis Chaudhuri M Arif

257 Texture Feature Matching Methods for 335 Interactive Education through ECT - A
Content based Image Retrieval Focus on Rural India
Ivy Majumdar and B N Chatterji S Arumuga Perumal

Note : The Institution of Electronics and Telecommunication Engineers assumes no responsibility for the statements and
opinions expressed by individual authors and speakers.

191
 Guest Editorial
Electronics & Communication Technologies (ECTs) is a vast canvas covering several inter connected
topics and technologies, almost everything we need in our daily life: digital communications, VLSI, tele-medicine,
tele-education, networking, signal processing, information for all etc. The theme of this special issue is
“Interactive ECTs — A focus on rural India”. Of the many topics this issue could cover, IETE received 11
invited papers from experts in this field, besides three contributed/short articles.
Prof Jhunjhunwala et al, in their paper “Broad band to empower rural India” mentions about the urban/rural
divide and that the TV has made known this difference to the rural masses as to where they stand. The paper
presents many Broadband wireless technologies that could alleviate this gap to a great extent by providing higher
data speeds and system gain.
Dr V U Reddy et al, in their paper, deal with the spatial multiplexing aspects of Multiple Input Multiple
output (MIMO) communication systems-an exciting development. Various architectures are discussed. Also,
brought out how fading, considered usually as undesirable phenomena, can be exploited to attain high spectral
efficiencies under MIMO.
Advances in Satellite communications, Remote sensing and GIS (Geographic Information System) have
provided solutions to various national development programs, such as disaster management, tele-education, tele-
medicine, creative 3D image representations of human organs etc, besides land use planning and integrated
rural/urban development. Mr A S Manjunath et al discussed how advanced sensor technologies and data
processing techniques have made this possible.
Rural broadband access is developed with the help of the widely available and highly cost reduced WiFi chip
sets. Prof Anurag Kumar et al, discussed in detail, the design issues related to Medium Access Control (MAC)
for packet voice telephony and for Internet access.
Dr Lal Kishore has given a review of the various VLSI technologies, research work currently in vogue, and
the future perspective. Hosting of Websites/portals with information useful to rural areas is proving to be a boon.
VLSI has its significant contribution in this.
Remote sensing with its high resolution and multi spectral sensors is providing valuable data/ information for
land cover classification. Pre processing and data processing techniques have also commensurately advanced
for analyzing such data. Prof Subhasis Chaudhuri et al, describe a temporally adaptive partially unsupervised
classification technique for remote sensing data.
Retrieving images, similar to a given query image, from a huge image data bank is a challenging task. This
is because, ultimately human perception / understanding is involved in the interpretation of the word “similar” for
validation/ evaluation of the retrieved images. Prof B N Chatterjee et al have presented, with a few examples,
a host of techniques currently available in content-based image retrieval (CBIR).
Government’s vision to make India a developed nation by 2020 is not a simple task. Dr Srihari Rao
enumerates the various ECT technologies that empower the rural people and transform them to a knowledge
powered PURA (providing urban amenities to rural areas).
Dr Surendra Pal et al, describe in their paper “Satellite Technology utilization for rural and urban India” that
space communication has changed society beyond imagination. Many technological reversals have been seen:
like the telephone which should have been on the wired network has become wireless, while the TV which was
on wireless earlier, now works on cable. Satellite technology connects the total country irrespective of location.
This paper describes also the services provided through satellites.

192
 GUEST EDITORIAL 193

Testing in VLSI is important and complex. Prof Bhargav Bhattacharya et al developed a programmable
built in self-test for embedded memory cores for detecting neighborhood pattern sensitive faults including static
and active. A programmable BIST (built in self test) architecture is also designed.
Dr V Jayaraman et al, in their paper on “Use of GIS and MIS in participatory watershed development”
discussed about the Earth Observation (EO) inputs and spatial data modeling for integrated land and water
resource development at grass roots level. EO inputs are judicious mix of GIS and MIS (management
information system). Implementation and monitoring through webGIS tools in a project mode in an area in
Karnataka, has brought about significant impacts on natural resource conservation and living conditions in the
area.
Anuradha et al, in their contributed paper, discuss the simulation of Nakagami-m fading in wireless channels
using generalized SC DS-CDMA and MC DS-CDMA schemes, using binary phase shift keying modulation
techniques. It is observed that bit error rate decreases with increasing m for both SC DS-CDMA and MC DS-
CDMA.
This volume ends with two short papers, one on design of a FIR filtering core for high-speed application, and
another on interactive education through ECT for rural India.
I feel what we covered in this volume is only a drop in the ocean of the subject matter. We thank immensely
all the contributors of this special issue. Hope the reader gets a flavor of the ever changing/ growing ECT, and
the directions in which it is heading to reach the benefits to rural India.
Thanks to IETE for giving me this opportunity to be a guest editor and to interact with the many experts. The
publication wing of IETE headquarters have done an excellent job of getting this volume printed in time.

Prof B L Deekshatulu
Council Member & Distinguished Fellow IETE,
Guest Editor, IETE Technical Review,
Visiting Professor University of Hyderabad.
Residence Address : 10-3-123/3, East Maredpally,
Secunderabad 500 026 (AP).
Email deekshatulu@hotmail.com
Mobile: 9908499081
 Guest Editor
Dr B L Deekshatulu has
distinguished himself through his
research and technological contributions
in the field of Control systems, Digital
Image Processing and Remote Sensing.
Dr Deekshatulu obtained BSc (Engg -
Electrical) degree in 1958 from the
Banaras Hindu University and ME and
PhD degrees from the Indian Institute
of Science (IISc), Bangalore. He was
awarded Martin Foster Medal by IISc, for Best PhD thesis. He
joined as Lecturer in 1964 at IISc., Bangalore and became Professor
in 1970 and continued in that position till 1976. Dr Deekshatulu
visited USSR during 1968 as a Government of India delegate for
setting up School of Automation at IISc, Bangalore.
Dr Deekshatulu worked as Visiting Scientist at the IBM
Watson Research Centre, York Town Heights, New York, and
at the Environmental Research Institute of Michigan during
1971-72 on Digital Image Processing and Remote Sensing. He
designed and fabricated for the first time in India, Grey scale and
color Drum Scanners for Computer Picture processing which
has subsequently won him and his group a NRDC Award.
Dr Deekshatulu joined the National Remote Sensing Agency,
Hyderabad as Head, Technical Division in April 1976 and
became Director in January 1982 - promoted to the grade
“Outstanding Scientist” in July 1989 and “Distinguished
Scientist” (grade of Secretary) in July 1995 and retired from
NRSA in October 1996. He has been responsible for the
upbringing of the National Remote Sensing Agency in all its
facets and for executing National and State level projects in
many disciplines of Remote Sensing applications.
He has over 130 research publications to his credit. He has
guided 12 PhD Scholars and over 60 MTech. Students’
dissertations. He has visited 27 countries in the world. He is
Fellow of many Scientific/Engineering Academies such as FNA,
194
FASc, FNAE, FNASc, Distinguished Fellow IETE etc. Fellow
IEEE (USA), Fellow Third World Academy of Sciences (Italy).
His current interests are Remote Sensing Data Analysis, Digital
Image Processing and Neural Networks.
He is a recipient of many awards such as the Bharat Ratna,
Sir M Visveswaraya Award for “Outstanding Engineer” in
1984, NRDC Invention Awards in Jan 1986 and in Aug 1993,
and Dr Biren Roy Space Science Award in 1988, “Padamsri”
medal in Jan 1991 by President of India, Brahm Prakash Medal
for significant contributions to Engineering Technology, Om
Prakash Bhasin Award for Science and Technology for 1995.
Received “Sivananda Eminent Citizen Award” from Vice
President of India in Dec 1998 and Gold Medal from Indian
Geophysical Union 1997. He was Chairman, National Committee
on International Geosphere, Biosphere Programme (IGBP) during
1994-97, Chairman of Remote Sensing Applications Missions,
India 1987-96. Awarded Boon Indrambarya Gold Medal by
Thailand Remote Sensing and GIS Association in November
1999 in Hong Kong for contributions to Remote Sensing. Awarded
2002 Aryabhatta Award by Astronautical Society of India, for
lifetime contributions to Remote Sensing. Awarded Distinguished
Alumni Award from IISc, Bangalore.
Dr Deekshatulu was a UN / FAO consultant in Beijing
during November, 1981. He was the Government representative
in the UN/ESCAP/RSSP Directors’ meetings and Inter
Governmental Consultative Committee meetings from 1985-95.
He was a UN/ESCAP Senior Consultant during September-
November, 1996. He was Director of Centre for Space Science
and Technology Education in Asia and the Pacific (CSSTE-AP),
Affiliated to the United Nations, IIRS Campus, Dehra Dun,
from November 1995 to April 2002.
Currently, Dr Deekshatulu is a Visiting Professor in the
Dept of Computer & Information Sciences, University of
Hyderabad pursuing research and teaching in Image pro.
 INVITED PAPER

IETE Technical Review

Vol 24, No 4, July-August 2007, pp 195-201

Broadband to Empower Rural India

ASHOK JHUNJHUNWALA
ESB 331 A, TeNeT Office, 2nd Floor, Department of Electrical Engineering,
Indian Institute of Technology Madras, Chennai 600 036, India.
DAVID KOILPILLAI AND BHASKAR RAMAMURTHI
Department of Electrical Engineering,
Indian Institute of Technology Madras, Chennai 600 036, India.
email: ashok@ee.iitm.ac.in; koilpillai@ee.iitm.ac.in; bhaskar@tenet.res.in

1. INTRODUCTION comprises 250-300 households, and occupies an area

of 5 km2. Most of this is farmland, and typically the

U RBAN India has been on the move over the last

ten years and its growth has accelerated especially
during the last five. However, the same cannot be said
about rural India. Urban Indians are full of confidence,
but rural Indians do not see much of a future for
themselves. The only change in the lives of many rural
people is the availability of television, which in fact,
has created greater aspirations amongst them. They
can now clearly see the difference between life in
urban India and rural India and cannot understand
why they are being left so far behind.
In a democratic set up, where some form of
election (central, state or local) takes place constitute
almost every year and a half, the feeling of deprivation
amongst the rural people plays havoc, especially as
rural people constitute 70% of India’s population. Every
politician would be forced to promise more and more
to rectify this deprivation and Government policies
would be forced to be populist. Yet there are a few
constructive programs which can really change life in
rural India. Governments will be over-turned frequently,
as no populist measure would rectify the great divide
that is getting accentuated with India’s urban growth.
In fact, the policies required to sustain even urban
growth would be under threat and would not be
consistently pursued. The only answer is a quick and
urgent focus on rural areas towards rural
transformation.
2. RURAL INDIA TODAY
Rural India consists of approximately 700 million
people, living in 638000 villages. The average village
Paper No 125-A; Copyright © 2007 by the IETE.
*The paper is based on and borrows from some recent
articles on Telecom in rural India [1-3].
195
houses are in one or two clusters. Villages are thus
spaced 2-3 km apart, and spread out in all directions
from the market towns. The market centres are
typically spaced 30-40 km apart. Each such market
centre serves a catchment of around 250-300 villages
in a radius of about 20 km. As the population and the
economy grow, large villages morph continually into
towns and market centres. The villages are
characterized by low incomes. About 85% of the
households have an income less than Rs 3000/- per
month (amounting to Rs 600/- per month per person,
assuming a family of five). Two-thirds of the
households are dependent on agriculture for income,
and this is often seasonal and dependent on rainfall.
Rural India now has very little industry. Its people are
mostly under-employed in agriculture. At the same
time, agricultural growth in India has slowed down to
1% over the last decade, falling behind even the
population growth.
Fortunately, most of rural India has some form of
road connectivity today (even though often much of it
may be in a bad condition) and at least one bus would
ply to each village every day. A railway station is also
not very far off. Highways connect towns, which are
rarely farther than 15 kms from any village. Also, a
significant number of villages are on the electrical
grid. However, the grid supplies power only during
the period when the demand in urban (and industrial)
areas is low. During peak-demand period the urban
areas have the capability of sapping all the power
produced and the rural areas are supplied only whatever
is leftover. So, even when the power flows into the
rural grid (0-18 hours per day, depending on the
State), the voltage could fluctuate between 90V
(reflecting higher demand) to 440 V (during nights
when demand is low). Decentralized power generation
196 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
in rural India may be the only answer to this problem in
the short run.
Telecom technology has advanced very rapidly.
Even though only a small percentage of villages have
reliable telecommunications connectivity today, the
situation is changing fast. With the rural thrust, it is
reasonable to expect that most villages in India will
have mobile coverage as well as broadband Internet
connection within the next three to four years.
3. THE TELECOM SITUATION TODAY
The mobile revolution of the last five years has
seen base stations sprouting in most towns, owned and
operated by multiple operators, including the state-
owned company BSNL. The base stations of BSNL,
as well as those of the new operators are also
networked using optical fibre laid in the last five years.
There is a lot of dark fibre, and seemingly unlimited
scope for bandwidth expansion.
The solid telecom backbone that knits the country
ends abruptly at the towns and larger villages. Beyond
that, cellular coverage extends mobile telephone
connectivity only up to a radius of 5 km, and then the
telecommunication services simply peter out. Cellular
coverage can and will grow, but this will depend on the
rate at which infrastructure costs and operating costs
drop, and the rate at which rural incomes rise. Fixed
wireless telephones have been provided in tens of
thousands of villages, but it would be safe to conclude
that the telecommunications challenge in rural India
remains the “last ten miles”. This is particularly true if
one were to include broadband Internet access in
one’s scope, since the wireless technologies currently
being deployed can barely support dial-up speeds.
This then is the rural India in search of appropriate
broadband wireless technology: characterized by fat
optical-fiber POPa within 15-20 km of most villages,
fairly homogenous distribution of villages in the plains,
poor rural cellular coverage, and low incomes. The
last aspect makes the provision of basic
telecommunications as well as broadband internet
services all the more urgent, since ICT is an enabler
for wealth creation.
3.1. GSM and CDMA dominate today but
3G will be available in time
Before we look at broadband technologies for
rural India, let us take a look at mobile technologies of
today (we will mostly focus on GSM, though CDMA
systems are also present in India today). It may not be
readily apparent that the bottleneck in rolling out services
to rural areas is not the cost of electronic equipment,
but is actually due to the following:
i) The most significant cost component is the
site preparation and the erection of the tower.
Infrastructure like roads and electricity may
have to be first set up. The towers are about
40 m tall, and require considerable amounts of
expensive steel.
ii) The second highest contributor to the cost is
the electrical power infrastructure – RF cables
running to the top of the tower, the power
amplifiers, RF filtering and the transceivers.
Roughly 55% of the cost of the base station
equipment is in these RF components [4].
iii) The maintenance of cell site infrastructure
requires local personnel who should be trained
to deal with the problems that arise in wireless
equipment.
iv) Availability of ultra-low cost (ULC) mobile
phones at costs below Rs 1500/- with financing
packages.
v) Proper distribution infrastructure for phones,
SIMs, spares and accessories in the remote
areas, and availability of basic training to users
so that they can use the phones properly.
vi) Billing and collection infrastructure for pre-
and post-paid subscribers.
If one accepts these as the real bottleneck, then it
is immediately evident that as soon as there is sufficient
GSM voice coverage across India, we are already
past the key hurdles for upgrading. The cell sites and
towers are set up and maintenance, distribution, user
training and billing/collection infrastructure put in place.
One cannot afford to deploy any new cell sites,
but only add electronic equipment at existing cell sites.
To deploy 3G equipment at a cell site, the Node B has
to be installed (instead of, or in addition to, the GSM
BTS). The cost of Node B equipment has been falling
by approximately 40% each year over the last 4 years.
Taken together with the fact that 3G offers more
capacity than GSM, the 3G Node B is just 50% more
expensive today than the GSM BTS to deploy the
same voice capacity [4]. It has already been seen that
55% of the cost of base station equipment is in the RF.
Since a single 3G channel of 5 MHz replaces many
GSM channels of 200 kHz required achieving the
same capacity – the RF costs of 3G systems should,
over time, be lower than that of GSM systems. Thus
3G will eventually lead to cheaper equipment than
GSM, resulting in mobile broadband infrastructure in
India.
 ASHOK JHUNJHUNWALA et al : BROADBAND
4. BROADBAND FOR RURAL INDIA
When considering any technology for rural India,
the question of affordability must be addressed first.
Given the income levels, one must work backwards to
determine the cost of an economically sustainable
solution. Approximately 200 households in a typical
village having disposable incomes can spend on an
average Rs 50-100/- per month for telephony and data
services. Assuming an average of two public kiosks
per village, the revenue of a public kiosk can be of the
order of Rs 5000/- per month. Apart from this, a few
wealthy households in each village can afford private
connections. After providing for the cost of the terminals,
it is estimated that, a cost of at most Rs 15000/- is
sustainable for the connection. This includes the user
equipment, as well the per-subscriber cost of the
network equipment and infrastructure (towers) linking
the user up to the optical fiber POP.
4.1. Coverage, system gain, and cost of
towers
It has been mentioned that one needs to cover a
radius of 15-20 km from the fibre POP using wireless
technology. The ‘system gain’ is a measure of the link
budget available for overcoming propagation loss and
penetration losses (through foliage and buildings), while
still guaranteeing satisfactory system performance.
Mobile cellular telephone systems have a system gain
typically of 150-160 dB, and achieve indoor penetration
within a radius of about 5 km. They do this with Base
Station towers of 40 m height, which cost about Rs
500,000/- each. If a roof-top antenna is mounted at the
subscriber-end at a height of 6 m from the ground,
coverage can be extended up to 15-20 km. When the
system gain is lower at around 135 dB, as in any line-
of-sight system, the coverage is limited to approx 10
km and the antenna-height at the subscriber-end has
to be 10 m in order to clear the tree tops. This
increases the cost of the installation by about Rs 1000/
- per connection.
Thus, roof-top antennas in the villages are a must
if one is to obtain the required coverage from the fiber
POP. A broadband wireless system will also need a
system gain of around 150 dB if it is to be deployed
with 6 m poles. This system gain may be difficult to
obtain at the higher bit-rates supported by emerging
technology, and one may have to employ taller poles in
order to support higher bit-rates at distant villages.
There is an important relationship between
coverage and the heights of the towers and poles, and
thus, indirectly, their cost. The Base Station tower
TO EMPOWER RURAL INDIA 197
must usually be at least 40 m high even for line-of-
sight deployment, as trees have a height of 10-12 m
and even in the plains, one can expect a terrain variation
of at least 20-25 m over a 15-20 km radius. Taller
Base Station towers will help, but the cost goes up
exponentially with height. A shorter tower will mean
that the subscriber-end installation will need a 20 m
mast. At around Rs 20,000/-, this is substantially costlier
than a pole, even if the mast is guyed and not self-
standing. The cost of 250-300 masts of this type is
very high compared to the incremental cost of a 40 m
tower over a 30 m one. With 40 m towers, poles are
sufficient at the subscriber-end, and only rarely need
to be more than 12 m high.
In summary, for a cost-effective solution, the system
gain should be of the order of 150 dB (at least for the
lower bit rates), a 40 m tower should be deployed at
the fiber POP, and roof-top antennas with 6-12 m
poles at the subscriber-end. The cost per subscriber of
the tower and pole (assuming a modest 300 subscribers
per tower) is Rs 3000/-. This leaves about Rs 12000/-
per subscriber for the wireless system itself, inclusive
of both the infrastructure and terminal sides.
4.2. What constitutes broadband?
The Telecom Regulatory Authority of India has
defined broadband services [4] as those provided with
a minimum data rate of 256 kbps. At this bit-rate,
browsing is fast, video-conferencing can be supported,
and applications such as telemedicine and distance
education using multi-media are feasible. There is no
doubt that a village kiosk could easily utilize a much
higher bit-rate, and as technology evolves, this too will
become available. However, it is important to note
that even at 256 kbps, since kiosks can be expected to
generate a sustained flow of traffic, 300 kiosks will
generate of the order of 75 Mbps. This is a non-trivial
level of traffic to carry over the air in each Base
Station, even with a spectrum allocation of 20 MHz.
4.3. Suitability of broadband wireless
technologies
One of the pre-requisites for any wireless
technology for it to cost under Rs 12,000/- is that it
must be a mass-market solution. This will ensure that
the cost of the electronics is driven down by volumes
and competition to the lowest possible levels. As an
example, both GSM and CDMA mobile telephone
technologies can today meet the above cost target,
(however, an even lower cost is needed for a non-
broadband technology since the services provided are
limited).
198 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
The third-generation cellular telephone technologies
will probably continue to meet this cost target while
offering higher bit-rate data services. However, they
will not be able to provide broadband services as
defined above (as at most they will provide 64 kbps to
each user).
If one were to turn one’s attention to recently
standardised broadband technologies such as
WiMAX-d (IEEE 802.16d) [5], it is found that at
present, the volumes are low and costs high. Of these,
WiMAX-d has a lower system gain than the others
(which are all around the required 150 dB). All of
them will give a spectral efficiency of around 2 bps/
Hz/cell (after taking spectrum re-use into account),
and thus can potentially deliver 40 Mbps at each Base
Station with a 20 MHz allocation. However, high cost
due to low volumes is the inhibitory factor with these
technologies.
It is likely that one or more OFDMA-based
broadband technologies will become widely accepted
standards soon. One such technology is WiMAX-e
(IEEE 802.16e) [6] that is emerging rapidly. These
will certainly have a higher spectral efficiency, and
more importantly, if they become popular and successful,
the cost will be low. However, it will be several years
before a widely-adopted technology derives the benefits
of market size and the cost drops to affordable levels
for rural India. The obvious question is whether there
are alternatives in the interim that meet our performance
and cost objectives.
4.4. Near-term Technologies
4.4.1. Broadband CorDECT technology
One of the earliest and most widely deployed
examples of such re-engineering is the corDECT
Wireless Access System developed in India [7]. It
provided toll-quality voice service and 35/70 kbps
Internet access to each subscriber without the
bandwidth having to be shared. The next-generation
Broadband corDECT system has also been launched
recently, capable of supporting 70 Mbps per cell with
a 5 MHz bandwidth (supporting 144 full-duplex 256
kbps connections simultaneously). With this system,
each subscriber will get 256 kbps dedicated Internet
access, in addition to toll-quality telephony. These
systems are built around the electronics of the European
DECT standard, which was designed for local area
telephony and data services.
Broadband corDECT incorporates added
proprietary extensions to the DECT physical layer
that increase the bit-rate by three times, while being
backward compatible to the DECT standard. Thus,
the spectral efficiency goes up three times when
compared to conventional DECT. Additionally, dual-
polarization antennas have been used to exploit
polarization isolation while till operating within the
DECT MAC framework, and further double spectral
efficiency. More importantly, all this has been done
while retaining the use of the low-cost DECT chipsets.
The system gain in Broadband corDECT for 256
kbps service is 125 dB. This can be increased by a
few dB, where required, by increasing the antenna
gain at the subscriber-end (which is currently 11 dBi).
This is sufficient for 10 km coverage under line-of-
sight conditions (40 m tower for BS and 10-12 m pole
at subscriber side). A repeater is used, as in the
corDECT system, for extending the coverage to 25
km. The corDECT system, and now the broadband
corDECT system, both meet the rural price-
performance requirement comfortably, but with the
additional encumbrance of 10-12 m poles and one
level of repeaters. The first-generation technology is
proven in the urban and rural Indian environment, and
much is known about how to deploy it successfully.
The Broadband corDECT system works with the
same deployment strategy. It is being deployed in
significant numbers beginning 2007.
4.4.2. WiFi rural extension (WiFiRe) [8]: A
new WiFi-based wide-area rural
broadband technology
In recent years, there have been some sustained
efforts to build a rural broadband technology using
WiFi chipsets. The WiFi bit rates go all the way up
to 54 Mbps. The system gain is about 132 dB for
11 Mbps service, and as in corDECT, one requires a
40 m tower at the fibre POP and 10-12 m poles at the
subscriber-end. The attraction of WiFi technology is
the de-licensing of spectrum for it in many countries,
including India. In rural areas, where the spectrum is
hardly used, WiFi is an attractive option, provided its
limitations when used over a wide-area are overcome.
Various experiments with off-the-shelf equipment
have demonstrated the feasibility of using WiFi for
long-distance rural point-to-point links. The more serious
issue with regard to the 802.11 standard is that the
commonly supported MAC protocol is a Carrier Sense
Multiple Access (CSMA) protocol suited only for a
LAN deployment. When standard WiFi equipment is
used to set up a wide-area network, medium access
efficiency becomes very poor, and spectrum cannot
be re-used efficiently, even in opposite sectors of a
base station.
 ASHOK JHUNJHUNWALA et al : BROADBAND
A solution for this problem is to replace the MAC
protocol of 802.11 with a MAC protocol more suited
to wide-area deployment. Such a new MAC, christened
WiFiRe, has indeed been defined [8] carefully such
that a low-cost WiFi chipset can still be used, and the
in-built WiFi MAC in it can be by-passed. The new
MAC can be implemented on a separate general-
purpose processor with only a modest increase in
cost. With the new WiFiRe MAC, it is estimated that
using a single WiFi carrier, one can support about 25
Mbps (uplink + downlink) per cell. This would be
sufficient for about 100 villages in a 10-15 km radius.
Repeaters, possibly operating on a different frequency,
will be needed for covering more villages over a
greater distance.
4.5. Tomorrow’s Broadband Technologies
In a few years time, we expect to see significantly
enhanced broadband technologies, which could provide
the 150 dB system gain, even while offering data
speeds of 256 kbps or more for each connection. The
three most promising technologies are all standard-
based and are therefore expected to meet the price
targets required for Rural India. These technologies
are: (i) IEEE 802.16 m, (ii) 3GPP-LTE, (iii) 3GPP2-
UMB.
We present each of these efforts in brief.
4.5.1. IEEE 802.16 m [6]
This is an OFDMA based standard emerging out
of efforts of IEEE. The earlier version of the standard
is IEEE 802.16 e which was finalized in 2006 and is
popularly known as WiMAX. This broadband wireless
standard, using state of art modulation, coding, scheduling
and multiplexing would use smart antennas to enable
peak data rates of 100 Mbpsec for mobile users in a
20 MHz spectrum. The working group finalizing the
standard aims to finalize the requirements, channel
model and evaluation methodology by May 2007 and
make a proposal to ITU-R Working Party 8F (WP8F)
for IMT-advanced requirements by March 2008. The
principal stakeholders driving this effort are vendors
developing 802.16 products, licensed carriers using
802.16 products and members of WiMAX ForumTM.
4.5.2. 3GPP LTE [9]
The Third Generation Partnership Project (3GPP)
Long Term Evolution (LTE) was started with feasibility
study on evolution of Universal Terrestrial Radio
Access Network (UTRAN) in 2004 and grew with
the recommendations for delivery of mobile broadband
TO EMPOWER RURAL INDIA 199
services by Next generation Mobile Networks
(NGMN) initiative in 2006. A Technical Report (TR
25.913) provides detailed requirements, which include
downlink peak date rate of 100 Mbps within a downlink
spectrum of 20 MHz using the OFDMA technology.
The uplink peak data rate is expected to be 50 Mbps
with a 20 MHz uplink spectrum using SC-FDMA
technique. It is proposed to support 200 users in active
state in each cell. The users are expected to get high
performance even with mobility as high as 120 Kmph.
MIMO is expected to be used and an enhanced
multimedia service is expected to be a part of the
standard.
4.5.3. 3G-PP2 [10]
The CDMA Development group (CDG) is
collaborating with Third generation Partnership Project
2 (3GPP2) to define Ultra Mobile Broadband (UMB)
standard as an evolution of CDMA 2000. The systems
requirement document was approved in May 2006
and uses scalable bandwidth up to 20 MHz. The
forward direction peak data rate is expected to be as
high as 500 Mbps in fixed and 100 Mbps in mobile
environment using OFDMA. The reverse direction
data rate is to be 150 Mbps in fixed and 50 Mbps in
mobile environment using qusi-orthogonal transmission
based on OFDMA, together with non-orthogonal user
multiplexing with layered superposed OFDMA (LS-
OFDMA). The reverse link also supports CDMA for
control and low-rate, low latency traffic. The advanced
air interface agreement has been reached by Technical
specification group C (TSG-C) based on a consolidated
framework proposal submitted by several operators
and equipment vendors worldwide. The detailed
technical specification of air interface framework is
expected by end of second quarter of 2007 and the
Technological Evolution Framework (TEF) outlines
the evolution strategy beyond the 2010 time frame.
5. IMPLICATIONS FOR INDIA
These next generation broadband wireless
standards are important for India, as it would enable
broadband wireless to reach urban as well as rural
India, just like GSM / CDMA mobiles do today. The
Broadband Wireless Consortium of India (BWCI) has
been formed (see: cewit.org.in) between operators,
equipment manufacturers, component suppliers,
academia and contribute the standardization efforts
based on OFDMA technologies defined above. While
these technologies would be available beginning 2009,
the operators are starting to use Broadband corDECT
in small towns and rural India. The next-generation
wireless system would have the capability required to
200 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
deliver broadband to all villages. The bit-pipes would
be there. The challenge beyond would be use the bit-
pipes to transform the rural economy.
REFERENCES
1. Ashok Jhunjhunwala, Next Generation Wireless For
Rural Areas, IJRSP, Special Issue on Rural Wireless
Communication, vol 36, no 3, June 2007.
2. Bhaskar Ramamurthi, Broadband wireless
technology for rural India, IJRSP, Special Issue on
Rural Wireless Communication, vol 36, no 3, June
2007.
3. G Venkatesh & Ashwin Ramachandra, Can 3G
technologies benefit rural India?, IJRSP, Special
Issue on Rural Wireless Communication, vol 36, no
3, June 2007.
4. http://www.dotindia.com/ntp/broadbandpolicy
2004.html
Authors
Ashok Jhunjhunwala is Professor
of the Department of Electrical
Engineering, Indian Institute of
Technology, Chennai, India and was
department Chair till recently. He
received his BTech degree from IIT,
Kanpur, and his MS and PhD degrees
from the University of Maine. From
1979 to 1981, he was with Washington
State University as Assistant Professor.
Since 1981, he has been teaching at IIT, Madras.
Dr Jhunjhunwala leads the Telecommunications and
Computer Networks group (TeNeT) at IIT Madras. This group
is closely working with industry in the development of a number
of Telecommunications and Computer Network Systems. TeNeT
group has incubated a number of technology companies which
work in partnership with TeNeT group to develop world class
Telecom and Banking products for Rural Markets.
Dr Ashok Jhunjhunwala has been awarded Padma Shri in
the year 2002. He has been awarded Shanti Swarup Bhatnagar
Award in 1998, Dr Vikram Sarabhai Research Award for the
year 1997, Millennium Medal at Indian Science Congress in the
year 2000 and H K Firodia for “Excellence in Science &
Technology” for the year 2002, Shri Om Prakash Bhasin
Foundation Award for Science & Technology for the year 2004,
Awarded Jawaharlal Nehru Birth Centenary Lecture Award by
INSA for the year 2006 and IBM Innovation and Leadership
Forum Award by IBM for the year 2006. He is a Fellow of
INAE, IAS, INSA and NAS.
Dr Jhunjhunwala is a Director in the Board of SBI. He is
also a Board member of several companies in India, including
TTML, BEL, Polaris, 3i Infotech, Sasken, Tejas, NRDC, and
IDRBT. He is member of Prime Minister’s Setup Scientific
5. IEEE Standard for Local and metropolitan area
networks, Part 16: Air Interface for Fixed Broadband
Wireless Access Systems, IEEE Std 802.16™-2004
(Revision of IEEE Std 802.16-2001).
6. Draft IEEE Standard for Local and metropolitan area
networks “Part 16: Air Interface for Fixed and
Mobile Broadband Wireless Access Systems, IEEE
P802.16e/D5, September 2004.
7. corDECT, Wireless access system, Technical report
of Midas Communication Technologies, Chennai,
India, December 2000.
8. Paul Krishna, Varghese Anitha, Iyer Sridhar,
Ramamurthi Bhaskar & Kumar Anurag, WiFiRe:
Rural Area Broadband Access using the WiFi PHY
and a new MAC, IEEE Commun Mag (USA), Jan
2007 (accepted).
9. http//www.3gpp.org/ftp/pcg/Beijing workshop
presentation
10. www.3g-pp2/CDG Press Release (Aug 15, 2006).
Advisory Committee. His research interests are :
Telecommunications and Wireless Systems; Technologies for
Rural Areas.
* * *
Bhaskar Ramamurthi, a founding
member of TeNeT, holds an MS (1982)
and a PhD (1985) degree in Electrical
Engineering from the University of
California, Santa Barbara, USA. In 1986,
he joined the faculty at the Department
of Electrical Engineering, Indian Institute
of Technology Madras (IIT-M), his
alma mater. He had earlier graduated from
IIT-M in 1980 with a BTech in
Electronics Engineering. He is Dean (Planning) of IIT Madras
since 2005, the youngest faculty member ever to become a dean
at IIT Madras.
Image Coding using Vector Quantization, an area in which
he did some of his early work, formed the subject of his doctoral
thesis. Soon after obtaining his doctorate, he joined the AT&T
Bell Laboratories, USA, where he worked on problems in indoor
wireless communications.
Awarded University of California, Regents Fellowship for
1980-81 and 1981-82. The paper titled “Perfect-Capture ALOHA
for Local Radio Communications” selected for reprinting in IEEE
Press book on Key Papers in Multiple Access Communications.
Joined the Fellowship of Indian National Academy of Engineering
in the year 2000, awarded Vasvik Award for Electronic Sciences
and Technology in the year 2000, awarded Tamil Nadu Scientist
Award 2003 for Engineering and Technology.
 ASHOK JHUNJHUNWALA et al : BROADBAND
Member of the Board of Directors in Midas
Communications Technologies, N-Logue Communications,
Valued Epistemics (P) Ltd and Hon Director and Member,
Governing Council, Centre of Excellence in Wireless
Technology (CEWiT), Chennai.
He is a member of National Frequency Allocation Review
Committee 2001, QoS Specifications Committee of TRAI for
ILD VoIP Services and member Sectional Committee of INAE.
His research interests: Modulation and coding for Mobile
Communications; Wireless Communication Networks; Design
and Implementation of Wireless Local Loop Systems.
* * * the areas of technical management and intellectual property
David Koilpillai received the
BTech degree from IIT Madras, and the
MS and PhD degrees (all in Electrical
Engineering) from California Institute
of Technology, Pasadena, USA.
In June 2002, David joined IIT
Madras as a Professor of Electrical
Engineering, and is a member of the
TeNeT group. During January 2007-
July 2007, David is on leave from IITM and is working as the
Chief Scientist, Centre of Excellence in Wireless Technology
(CEWiT), a public-private R&D initiative of the Govt of India.
At CEWiT, David is leading a national project – Broadband
Wireless Consortium of India (BWCI), focusing on emerging
broadband wireless standards. Prior to joining IITM, David was
at Ericsson USA for twelve years, where he held different
technical and managerial positions. In 2000, he became the
Director of the Ericsson’s Advanced Technologies and Research
Department at RTP, North Carolina and a member of the global
management team of Ericsson Mobile Platforms, an Ericsson
TO EMPOWER RURAL INDIA 201
company developing all components of mobile phone technology.
In this role, he was responsible for an R&D team of 75 engineers
developing GPRS/EDGE handset technology, with an annual
budget of US $20 million. David has actively provided technical
inputs to Ericsson efforts in 3G standardization, and has been
presenting extensively on 3G and cellular evolution. David’s
technical areas of expertise include 3G/3.5G cellular systems,
broadband wireless systems, and DSP techniques for wireless
communications.
His technical contributions at Ericsson have resulted in 32
US patents, 10 journal papers and 26 conference publications. In
1999, David received the “Ericsson Inventor of the Year” award,
the highest technical recognition within Ericsson. In addition, he
has received numerous management awards for contributions in
rights. He has served as a technical expert for Ericsson in
technology-related litigations.
In November 2003, David was elected as Fellow of the
Indian National Academy of Engineering.
As part of IITM’s TeNeT group, David’s research and
teaching focus is in the areas of DSP applications in wireless
communications. He is involved in research and consultancy
projects in wireless and cellular systems. Some of David’s projects
are listed below:
• DSP applications in cellular systems (WCDMA, HSPA,
LTE, CDMA 2000)
o Algorithms for improved PHY performance – CDMA
and OFDM
o DSP techniques for software radio applications
o Cost-optimized GSM / EDGE basestation
* * *
 INVITED PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 203-213

MIMO Communications - Motivation and

a Practical Realization
G KALYANA KRISHNAN AND V UMAPATHI REDDY, FIETE

Hellosoft India Pvt. Ltd., 8-2-703, Road No.12, Banjara Hills,

Hyderabad 500 034, India.
email: {kallu, vur}@hyd.hellosoft.com

Under rich scattering environment, Multiple Input Multiple Output (MIMO) systems have
the potential to achieve capacities inconceivable by Single Input Single Output (SISO)
systems. This makes it one of the most exciting developments to have occurred in wireless
communications. Within a short duration, it has matured from a reasearch topic into a
technology to find a place in upcoming wireless communication standards. This paper
focuses on the spatial multiplexing aspects of MIMO. Starting from AWGN channels, we bring
out how fading, which is typically an undesirable phenomenon, can be gainfully exploited to
attain large capacities under MIMO configuration. We first briefly discuss some MIMO
architectures initially designed to realize a large portion of the theoretical MIMO capacity,
namely V-BLAST (Vertical Bell Labs Layered Space-Time) and D-BLAST (Diagonal BLAST),
and then describe briefly the conventional V-BLAST receivers and a receiver that attains
near optimal performance while keeping the complexity low. The combination of Orthogonal
Frequency Division Multiplexing (OFDM) and MIMO is introduced as a simple method to apply
MIMO communication in a delay spread environment. Taking the example of Wireless Local
Area Network (WLAN) standard IEEE 802.11n , we see how the research developments have
been incorporated in a practical system.

1. INTRODUCTION the research developments are reflected in upcoming

standards, for example, IEEE 802.11n [5].

S INCE Shannon laid down the fundamental capacity

limits for single transmit - single receive system in
1948, communications, both wireline and wireless, have
come a long way. Although advanced modulation and
coding techniques have realized the capacity limits,
our hunger for higher data rates has hardly got satiated.
In [1], the authors describe in great detail how the
maximum achievable data rates are severely limited
for a SISO system. This limitation with SISO system
triggered a search for alternatives within the constraints
of power and bandwidth. Seminal papers by Telatar
[2] and Foschini [3] have activated intense research
on MIMO systems (see [4] and [1]). These papers
reveal the massive potential of MIMO system due to
exploitation of the spatial domain. Within a short
duration, it has matured from a research topic into
upcoming wireless communication standards [5, 6].
In this paper, we will focus on spatial multiplexing
aspects of MIMO assuming CSI (Channel State
Information) at the receiver. We will also discuss how
Paper No 125-C; Copyright © 2007 by the IETE.
The remainder of the paper is organized as follows.
Section 2 brings out the huge potential of the MIMO
system. We attempt to convey how fading channels,
considered as unfavourable channels, can yield very
high data rates with multiple transmit and receive
antennae. This section is based on the excellent
treatment provided in [7]. In section 3, we discuss the
architectures that are devised to exploit the promised
capacity of MIMO. A popular architecture known as
V-BLAST is discussed in section 4. In this section,
we also describe a low complexity technique for
VBLAST decoding. In section 5, we see how all the
techniques have been incorporated into the WLAN
draft standard IEEE 802.11n [5]. Finally, the paper is
concluded in section 6.
In this paper, we use C interchangeably for both
spectral efficiency and capacity with corresponding
units bits/s/Hz and bits/s, respectively. Perfect
knowledge of CSI is assumed at the receiver. When
the transmitter also has CSI information, we mention
it explicitly. Bold upper/lower case letters are used
for matrices/vectors.

203
204 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
2. CAPACITY OF WIRELESS CHANNELS
In this section, starting from AWGN capacity, we
investigate the capacity for MIMO channel to fully
reveal its immense potential. The treatment of this
section is heavily based on [7].
The capacity of Additive White Gaussian Noise
(AWGN) channel is well known and is given by
æ P ö
Cawgn = W log ç1 + ¾ ¾ ÷ bits /s (1)
è N0W ø
where W is the channel bandwidth, P is the transmit
signal power and N0 is the one-sided noise power
spectral density. Here, log refers to logarithm to the
base 2. Replacing ¾ ¾ with SNR (Signal to Noise
Ratio), (1) can be expressed as
Cawgn = W log (1 + SNR) bits/s. (2)
Note from (1) that the two resources that control the
capacity are power and bandwidth. For fixed bandwidth
W, as power increases
Cawgn
¾ ¾ » log SNR bits/s/ Hz, (3)
W
which shows that for an increase in spectral
efficiency by 1 bit/s/Hz, the power needs to be increased
by 3 dB. Thus, increasing spectral efficiency by adding
more bits per dimension (i.e., by choosing a higher
constellation size) is a power costly option. On the
P antenna. Now suppose, we have the knowledge of
other hand, as W ® ¥, Cawgn » ¾ ¾ log2 e, i.e.,
N0
capacity is bounded even with infinite bandwidth. This
is due to the fact that as power is finite, the signal
power spectral density becomes vanishingly small as
W ® ¥. Note, however, that the capacity does increase
linearly with W if the SNR is held constant which calls
for an increase in P commensurate with W. This again
is not practical beyond a limit as bandwidth is a scarce
resource and also the power is limited. This explains
our helplessness and limitations in the pursuit of higher
spectral efficiency and higher data rates with SISO
communications. For a more detailed account of these
constraints, see [1].
As a step towards investigating fading channels,
we first examine deterministic channels. Consider first
the Single Input Multiple Output (SIMO) case (SISO
is a special case of this). With Mr receive antennas,
the capacity is [7]
æ P ö
C = W log ç 1 + || h ||2 ¾ ¾ ÷
è N0W ø
= W log (1+ || h ||2 S N R) bits/s, (4)
where h = [h1, h2, ..., hMr]T with hi denoting the fixed
channel gain from the transmit antenna to ith receiving
antenna and the superscript T denoting transpose of a
vector. At high SNR, the spectral efficiency is now
given by
C
¾ ¾ » log (|| h ||2 SNR) bits/s/Hz. (5)
W
Thus, we obtain a boost in the received SNR due to
multiple receive antennas. Though there is a SNR
improvement, increasing spectral efficiency is still a
daunting task. Let us now examine what happens if
we have multiple transmitters and single receiver which
corresponds to Multiple Input Single Output (MISO)
case.
Consider a MISO system with M t transmit
antennas and a single receiving antenna. If we constrain
the total transmitted power to P as before and allot a
transmit power of P/Mt to each antenna, the capacity
of MISO system, extending the SISO AWGN result,
will be
æ || h||2 P ö
C = W log ç 1 + ¾ ¾ ¾ ¾ ÷ bits /s, (6)
è M t N0W ø
where h = [h1, h2, ... , hMt]T with hi denoting the fixed
channel gain from ith transmit antenna to the receiving
CSI at the transmitter. Then, we can direct the
transmission to the receiving antenna using transmit
beamforming which will increase the received power
by Mt times, yielding the capacity of MISO system as
æ P ö
C = W log ç 1 + || h ||2 ¾ ¾ ÷bits /s (7)
è N0W ø
Thus, the capacity of a MISO system with knowledge
of CSI at the transmitter is similar to that of a SIMO
system.
In practical wireless channels, the received signal
strength fluctuates randomly (which is called fading).
Consider a SIMO channel. Assume that channel varies
slowly so that during each transmission, we have only
one realization of the channel. Such a channel is called
a slow fading channel. The maximum data rate of
transmission for each realization is
 KRISHNAN & REDDY : MIMO COMMUNICATIONS
æ P ö
C = W log ç 1 + || h ||2 ¾ ¾ ÷bits /s (8)
è N0W ø
Clearly, for a target transmission rate R, one cannot
guarantee that each realization of channel will give
C > R. Thus, strictly speaking, C = 0. To overcome this
predicament, we introduce the notions of outage and e
– outage capacity. For a target transmission rate R,
the system is said to be in outage with outage probability
P realizations. This is called the ergodic capacity of the
Pout(R) = P [W log (1 + || h ||2 ¾ ¾ ) < R], and the
N0W
e –outage capacity is the maximum rate of
transmission R such that Pout(R) is less than e (see [7]
for more details). For i.i.d Rayleigh frequency flat
fading channels, the outage probability at high SNR is
given by
1 and receive antennas, and a deterministic channel H.
Pout(R) µ ¾ ¾ ¾M ¾ . (9)
SNR r
In fading channels, at high SNR, errors occur
mostly when the channel is in outage, and hence, the
1
error probability also varies as ¾ ¾ ¾M ¾ . When the
SNR r
1 and Mr denoting the number of transmit and receive
error probability falls as ¾ ¾ ¾d ¾ , d is said to be the
SNR
diversity order of the system. Clearly, higher the
diversity, smaller the error rate. Note that in the SIMO
case, the diversity obtained is Mr. This is because,
there are Mr independent fading paths from transmitter
to receiver and outage occurs only if all Mr paths are
in fade. In general, if the data arrives through d
independent fade paths, the outage probability at high
1 æ
SNR will be proportional to ¾ ¾ d ¾ . Thus, in slow
SNR
fading, diversity can be used to counter the ill effects
of fading.
Let us now assume that the transmission extends
over multiple, say L, independent fade realizations.
Under this scenario, with receive CSI only, the channel
capacity is the average of the maximum rates that can
be obtained over the individual fade realizations
1 L æ P ö
C = W ¾ S log ç 1 + || hi ||2 ¾ ¾ ÷ bits /s. (10)
L i=1 è N0W ø
where hi denotes the channel vector in ith fade
realization. In arriving at the above capacity, the same
power allocation over each fade interval is assumed.
When the transmission extends over large number of
fade realizations, we consider it as a fast fading
scenario, and the diversity is obtained from the multiple
independent fade realizations.
205
Note that for finite L, the quantity given by the RHS
of (10) is random, and with non-zero probability it will
drop below any target rate. However, as L ® ¥, the
above expression approaches a finite value given by
æ æ P öö
C = WE ç log ç 1 + || hi ||2 ¾ ¾ ÷ ÷ bits / s, (11)
è è N0W ø ø
where the expectation E(.) is taken over the channel
fast fading SIMO channel.
In the presence of only receive CSI, both slow
and fast fading capacities are less than that of AWGN
capacity [7]. Thus, fading is clearly disadvantageous
in SISO, SIMO and MISO cases.
We now consider a scenario with multiple transmit
The received signal for an Mt ´ Mr MIMO system is
given as
r = Hs + n, (12)
where H is the channel matrix of size Mr ´ Mt with Mt
antennas, respectively. The element Hij corresponds
to the channel tap between ith receive antenna and jth
transmit antenna. r is the received vector of size
Mr ´ 1, s is the transmitted data vector of size Mt ´ 1
and n is an i.i.d circularly symmetric complex Gaussian
noise vector of size Mr ´ 1. The capacity of this
MIMO channel is given by [2]
1 ö
C = log det ç IMr + ¾ ¾ HK s H† ÷ bits / s/ Hz, (13)
è N0W ø
where Ks = E (ss†) with the superscript† denoting
Hermitian transpose. With no channel knowledge at
the transmitter, it is reasonable to transmit equal power
from all transmitters which gives Ks = ¾ I Mt.
Substituting this for Ks in the above expression and
after some manipulations, we get
n min æ SNR ö
C= S log ç 1 + ¾ ¾ li2 ÷ bits /s / Hz, (14)
i=1 è Mt ø
where nmin = min (Mt, Mr) li2 is the ith eigenvalue
P
of HH† and we substituted SNR for ¾ ¾ . Equation
N0W
(14) conveys that the overall capacity of a deterministic
MIMO channel can be viewed as the sum of individual
capacities of nmin parallel channels. This capacity is
maximized when all the eigenvalues are equal. Noting
206 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
that singular values of H are the square roots of the
eigenvalues of HH† and condition number of a matrix
with equal singular values is unity, the result (14)
states that a MIMO channel with lower condition
number will yield larger capacity.
The deterministic case can be extended to the
fading case in a straight forward manner assuming
only receiver CSI. We will also assume that the channel
tap variances are such that the average received
power at each receiving antenna is same as the total
transmit power P. The ergodic capacity of the
frequency flat fading MIMO channel (i.e., the channel
between each transmit-receive antenna pair has no
delay spread) is given by [2]
é æ 1 öù
C = max E êlog det ç IMr + ¾ ¾ ¾ HK sH† ÷ú
Ks:tr{Ks}<P ë è N0W øû
bits/s/Hz, (15)
where tr {Ks} gives total transmit power from all
antennas. It was shown by Telatar [2] and Foschini
[3] that under Rayleigh fading, the capacity of MIMO
channels (ergodic capacity or outage capacity) scales
linearly with minimum of transmit/receive antennas.
Thus, at high SNR, a 3 dB increase in power (and
hence SNR) translates to nmin bits/s/Hz increase in
spectral efficiency. To see this, consider a practical
scenario with Mr >> Mt and assume equal power from
all transmit antennas subject to the total power constraint.
For the Rayleigh flat fading case, for each realization
of H, the columns of H tend to be orthogonal and for
such case all the eigenvalues of HH† are equal to Mr.
Then, (15) reduces to
æ SNR ö the first time interval, only 1A is transmitted from the
C = nmin log ç 1 + ¾ ¾ M r ÷ bits /s / Hz, (16)
è Mt ø
which shows that capacity scales with nmin. Thus,
MIMO offers spatial multiplexing.
To emphasize the significance of fading channels
in the MIMO communications, let us look at the
capacity of an AWGN channel. In this case, for each
realization, all the channel taps will be equal and H
becomes a unit rank matrix, implying that there is only
one non-zero eigenvalue. Hence, from (14),
æ SNR ö The process is repeated. An MMSE (Minimum Mean
C = log ç 1 + ¾ ¾ ¾ l12 ÷ bits /s / Hz, (17)
è Mt ø Square Error) based D-BLAST achieves outage
where l12 is the only non-zero eigenvalue. That is, in
AWGN scenario, the capacity does not scale linearly
with nmin. Thus, in the MIMO case, the presence of
fading, far from being disadvantageous, is essential to
make high data rate communication possible.
3. MIMO CAPACITY ACHIEVING
ARCHITECTURES
We now discuss briefly the transmit/receive
architectures that can achieve the capacity promised
by MIMO system. We assume only receiver CSI as
this is the more practical scenario. Recall that a MIMO
channel decomposes into nmin parallel channels (see
(14)). This means that the total capacity is the sum
capacity of the individual parallel channels, and outage
occurs for a slow fading channel only when
é æ nmin æ SNR öö ù
Pout (R) = P ê ç S log ç1+ ¾ ¾ li2 ÷÷ < R ú
ë è i=1 è Mt øø û
(18)
Thus, the outage depends not on the capacity of a
single parallel channel but on the combined capacity of
all the parallel channels. In the slow fading scenario,
the capacity depends on individual realizations and
some of these channel paths between transmitter and
receiver could be in fade. Then, the performance will
be dictated by these channels unless the data is coded
over other channel paths also. It is with this perspective
that Foschini [8] proposed the D-BLAST algorithm
where each data stream passes through all transmit
antennas so that the data passes through all channel
paths thereby ensuring full diversity. This is explained
below for two transmit antennas case.
Each segment of data stream is coded into two
blocks, A and B, and transmitted as shown in Fig 1. In
first antenna. Soft decoding is performed directly as
no transmission is present on the other antenna. In the
next time interval, 1B is transmitted from second
antenna and 2A is transmitted from first antenna (see
Fig 1). This diagonal alignment of code words gives
rise to the name D-BLAST (Diagonal BLAST). At
the receiver, 1B is soft decoded by treating 2A as an
interference to be suppressed. Finally, 1A and 1B soft
outputs are used to decode segment-1. Once this
decoding is done, 1B is cancelled out from the data
received during second time interval. The situation for
segment-2 is similar to that when segment-1 started.
capacity for any value of outage probability. However,
D-BLAST suffers from error propagation and requires
strong codes to prevent it. Also, for small number of
data segments to be transmitted, there is a wastage of
 KRISHNAN & REDDY : MIMO COMMUNICATIONS
3A 2A 1A
3B 2B 1B
Fig 1 D-BLAST transmission for two transmit antennas
capacity as some of the antennas are not used in the
initial and final stages.
V-BLAST was proposed by Wolniansky et al [9]
as a low complex alternative to D-BLAST. In V-
BLAST, independent data are transmitted in parallel
as shown in Fig 2, and hence, the name Vertical
BLAST. In [9], the authors report that the laboratory
prototypes achieved spectral efficiencies to the tune
of 20-40 bits/s/Hz in indoor propagation environment
at SNRs of the order 20-30 dB. Note, however, that
under slow fading, the transmit diversity is lost in V-
BLAST as each stream passes through only one
transmit path and this impacts performance. In fast
fading, on the other hand, the transmit diversity loss is
not crucial as multiple fade realizations will provide the
required diversity. In fact, V-BLAST with MMSE
decoder and Successive Interference Cancellation
(SIC) (see sections 4.2 and 4.3) achieves ergodic
capacity of MIMO channel. We will now focus on V-
BLAST receivers.
4. V-BLAST RECEIVERS
For notational convenience, we will denote the
noise power N0W by s2. We consider the slow fading
case with H Rayleigh flat fading, i.e., each of the
Encoder
and Mappe
Data Serial to
Parallel
Encoder
and Mappe
Fig 2 V-BLAST transmission for two transmit antennas
207
channel tap values Hij is a complex Gaussian with
zero mean and unit variance. It is assumed that the
transmitted power is distributed equally among all
transmit antennas, and the total transmitted power is
unity. This is realized by selecting the transmit data
points from a constellation A with average power
1
¾ ¾ . For the channel model assumed, the average
Mt
received power at each antenna is then unity and the
1
average received SNR is ¾ 2 ¾ . Denoting the ith
s
column of H by hi, (12) can be expressed as
r = h1s1 + ... + hMt sMt + n (19)
4.1. Maximum Likelihood (ML) Algorithm
The ML algorithm obtains sML the estimate of the
transmitted data vector s, as sML = minaÎ AMt (r – Ha)†
(r–Ha). The ML receiver provides a diversity order
of Mr. The number of searches required to obtain the
ML estimate is g Mt where g is the constellation size.
Thus, complexity of the ML algorithm increases
exponentially with the number of transmit antennas
for a given constellation size.
4.2. Zero Forcing (ZF) and the MMSE
Receivers
The ZF receiver decodes each data point by nulling
out all the other data points. The ZF solution corresponds
to multiplication of the received vector by pseudo
inverse of the matrix. This can potentially lead to noise
enhancement. The MMSE receiver, on the other hand,
takes into account the interfering data as well as the
noise, and determines a solution that minimizes the
mean squared error E (|| s – s ||2). The diversity order
of the receiver is Mr – Mt + 1. Intuitively, while a
maximum receive diversity of Mt – 1 is possible, the
need to null Mt – 1 interfering streams limits the
diversity to Mr – (Mt – 1).
4.3. Successive Interference Cancellation
(SIC)
The motivation for the SIC can be seen from the
ZF receiver. Notice that the diversity order obtained
with a linear ZF receiver is given by Mr – Mt + 1. It
differs from the ideal receiver by a value Mt – 1, and
Mt – 1 is precisely the number of unwanted data
interferers for each data to be decoded. Hence, after
decoding one data point, say the ith, its contribution
hi si can be cancelled out from the received data. The
system will now have one interferer less and the
208 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

diversity order for the next decoded data will improve
by one assuming no error propagation. This procedure
can be repeated iteratively till all the data are decoded.
To minimize error propagation, the data layer with
maximum post-detection SNR (the layer most probable
to be decoded correctly) is decoded at each stage of
the iteration. The method can be applied to the MMSE
receiver also.
4.4. Low Complex Near ML Receiver for V
BLAST
Though ZF-SIC and MMSE-SIC algorithms can
be implemented with low complexity (see [10]), the
performance of these receivers is poor under slow
fading. On the other hand, the ML decoder exploits
the entire receive diversity but its computational
complexity is prohibitive. For example, for a system
with Mt = Mr, the diversity of ZF and MMSE algorithms
is only 1. Though SIC technique does improve the
performance, the overall performance is limited by
error propagation. Hence, a lot of effort has gone into
finding receivers with performance close to ML, but
with complexity comparable to that of ZF-SIC or
MMSE-SIC. We now present one such algorithm
proposed in [11]. Although developed independently,
the algorithm falls under the class of algorithms proposed
in [12] and also it is similar to [13].
Table 1 shows the complexity of the various
decoders for different constellation sizes in terms of
the number of complex multiplications involved. In
[11], the authors also propose a further modification to
reduce complexity further (though at the expense of
performance) by modelling the problem in an equivalent
real domain. These methods are named ZF-SIC-AXIS-
SEARCH and MMSE-SIC-AXIS-SEARCH,
respectively. The performance of these algorithms is
shown in Fig 3. Note that the plots of ZF-SIC-SEARCH,
MMSE-SIC-SEARCH and ML overlap, and therefore,
are not distinguishable.
All the theory we have considered so far has
assumed frequency flat fading channels. The real
world channels, on the contrary, are typically delay
spread channels. For example, see [14] which specifies
TABLE 1. Complexity of ML, ZF-SIC, ZF-SIC
SEARCH for 4- and 16-QAM
constellations
M Complex Multiplications, 4-QAM
(16-QAM), Mt = Mr = M
- ML ZF-SIC ZF-SIC-SEARCH
5 5 3
gM M2 ¾ M3 ¾ M3 + ¾ g M 2
2 2 2

4.4.1. Steps of the Algorithm 2 64(1024) 20 44(116)

3 576 (3e4) 68 122(284)
The algorithm implements the following steps.
4 4e3(1e6) 160 256 (544)
1. Find the diagonal entry of (H†H)–1 ((H†H +
s2 MtI)–1) with maximum value (this entry
corresponds to minimum post-detection SNR).
Let this be ith entry, and this corresponds to
ith data layer. Update H by deleting the ith
column hi. For each mapping of this data
layer to a signal point, say al, from the
constellation A, update the received vector as
rl = r – hial, l = 1, ..., g. Now, for each
updated received vector rl, perform ZF-SIC
(MMSE-SIC) iterations for the remaining data
layers using the updated H. This results in a
set of possible solution vectors of size g, say
bl, l = 1,2, ...g, for the decoded data. Let this
set be denoted as {S}.
2. From the set of possible solution vectors, choose
the one nearest to the received data vector r
by solving s = minbÎ s (r – Hb)† (r – Hb).
We refer to the above algorithm as ZF-SIC-
SEARCH (MMSE-SIC-SEARCH) when ZF-SIC Fig 3 Performance of various decoders of 5´5 MIMO
(MMSE-SIC) is used to decode the remaining layers. system with 16-QAM
 KRISHNAN & REDDY : MIMO COMMUNICATIONS
a set of indoor channel models. A simple way of
overcoming this problem is to combine MIMO with
OFDM which results in a frequency flat channel in
each subcarrier bin. Spatial multiplexing is then applied
separately in each subcarrier bin. Thus, (12) can be
rewritten for each subcarrier index m as
r(m) = H (m) s (m) + n (m) (20)
To simplify the notation, we will drop the subcarrier
indices and use the same notation as in (12). It is
assumed that all the techniques discussed above are
applied in each subcarrier bin.
5. A PRACTICAL IMPLEMENTATION -
IEEE 802.11N
The IEEE 802.11n standard [5] is a MIMO
extension for the WLAN standard [15]. In this section,
we describe briefly the transceiver for 802.11n in
relation to the MIMO architectures discussed in the
previous section. Typical channels, encountered by a
practical 802.11n system, are quasi-static or slow
fading, and as pointed out earlier, diversity becomes
important for good performance in the slow fading
scenario. Fortunately, with OFDM, the fluctuations in
frequency (arising out of the delay spread channel)
can be utilized to bring in additional diversity. An
interleaver can be used to re-arrange the data loading
onto the subcarriers so that adjacent coded data are
transmitted on subcarriers far apart. If the coherence
bandwidth of the channel is small enough, interleaving
Interleaver
Convolutional stream
Encoder parser
Interleaver
Fig 4 802.11n transmitter for two transmit antennas
209
in the frequency domain ensures that adjacent coded
data see independent channel fluctuations. Thus, the
interleaver, in some sense, attempts to realize the
benefit of a fast fading channel by taking advantage of
fading across subcarriers.
In the SISO case, it was shown by Zehavi ([16]
and later by Caire [17]) that under fast fading
conditions, bit interleaved coded modulation (BICM)
is a good choice. The technique involves placing an
interleaver at the bit level rather than at symbol level,
as done in symbol level interleaving. The bit level
interleaving helps to achieve better diversity as each
bit undergoes independent fading, whereas in a symbol
level interleaver, only each symbol (group of bits)
undergoes independent fading. Thus, in BICM , the
encoder is followed by an interleaver and this is
followed by symbol modulation. This method is
employed in WLAN standard IEEE 802.11a [15].
BICM can also be extended to the MIMO case in a
straight forward manner. With these ideas, the
baseband transmitter block diagram of 802.11n for
the case of 2 transmit antennas is as given in Fig 4.
Only blocks relevant to spatial multiplexing are shown
for simplicity. The BICM structure is evident from
the transmit block. The interleaver takes advantage
of the fading across the subcarriers and the single
Forward Error Correction (FEC) coder (a
convolutional encoder) ensures that the data goes
through all transmit paths so that transmit diversity is
not lost. The corresponding baseband receiver is
shown in Fig 5.
QAM
IFFT
Mapper
QAM
IFFT
Mapper
210 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

FFT

DeInterleaver

MIMO
Decoder, Soft
DeMapper Viterbi
Deparsing
FFT and soft Decoder
metric
calculation
DeInterleaver

FFT

Fig 5 802.11n receiver for two transmit and 3 receive antennas

We note from Fig 4 that the transmit structure is || r – Ha || 2 || r – Ha ||2

not exactly V-BLAST since the convolutional coded
data is demultiplexed into two streams and transmitted
from the two antennas so as to take advantage of
transmit diversity. The transmitter can be seen as a
combination of the BICM scheme for the SISO case
and the V-BLAST structure. The receiver has two
stages. The first stage is that of MIMO decoder, and
though the transmission is not V-BLAST, the V-
BLAST receivers can be readily used because of the
two stage decoding. In the first stage (the MIMO
decoder), the receiver does not take into account the
presence of FEC and treats the data as independent.
This is, however, justified in view of the presence of
interleaver prior to the mapping of bits to symbols.
Hence, the first stage is exactly the V-BLAST decoder.
The overall diversity gain obtained is a combination of
the diversity of the MIMO decoder stage and that of
the Viterbi decoder stage. Hence, an improvement in
V-BLAST decoder performance is expected to improve
the diversity gain of the overall receiver which is
important as the channel is slow fading. Similarly, an
important consideration for performance improvement
is the generation of soft metrics for the Viterbi decoder
in the BICM framework. The derivation of soft metrics,
referred to as Log Likelihood Ratio (LLR), is well
known for the SISO case (see [16,18]). For the MIMO
case, efficient ways of generating the LLR is required.
We outline here some of the common approaches.
5.1. ML Metric
The LLR metric for bki (kth bit of ith data layer) is
defined as [19]
LLR = min ¾ ¾ ¾ ¾ – min ¾ ¾ ¾ ¾ ,
a:ai Î Ak0 s2 a:ai Î Ak1 s2
(21)
where Ak0 corresponds to all constellation points in A
that have bit value 0 for the kth bit position and similar
definition applies to Ak1. To compute (21), we need to
evaluate the terms of the form
|| r – Ha ||2
min ¾ ¾ ¾ ¾ ¾ , l Î {0, 1} (22)
a:ai Î Akl s2
This can be done by calculating the distance metric
|| r – Ha ||2 for all values of the vector a. After that, for
each bit, minimum value has to be chosen and inserted
into (21). Thus, the distance metric has to be calculated
gMt (g is the constellation size of A) number of times
which becomes prohibitively complex as constellation
size and antenna size increase.
For the case of Mt = 2 we now explain how the
computations can be reduced. We need to compute
||r–Ha||2
mina:aiÎAlk ¾ ¾ ¾ . This minimization is done according
s2
to lemma 1 given in section 3.2 of [11], which is as
follows. For each constellation point substituted from
Alk for ith data layer, the corresponding element in the
remaining layer is determined using ZF-SIC such that
||r–Ha||2 is minimized. For example, if i = 1 and bit
value is 0, for each a1 Î Ak0, a2 can be determined
using ZF after subtracting out the contribution from
a1. The vector thus obtained a = [a1 a2]T will minimize
||r–Ha||2 for the particular value of a1. For bit value 0,
 KRISHNAN & REDDY : MIMO C OMMUNICATIONS 211

g
¾ such pairs will be identified and the computation of
2
|| r – Ha ||2 will be performed only for these pairs. As
computations are done for bit values {0,1} and for i Î
{1,2}, the total number of distance metric calculations
will be 2g, thus significantly reducing the complexity
compared to the direct approach which would involve
g2 distance metric computations.

5.2. LLR metric for ZF MIMO Decoder

A simple but effective method was proposed in

[19] to obtain LLR metrics for ZF criterion. The
motivation is as follows. With the ZF criterion, (12)
gets transformed to

(H†H)–1 H†r = s + (H†H)–1 H†n

~
r = s + ñ. (23)

This can now be treated as a set of independent
equations (ignoring the correlation between the noise
elements). Thus, the scenario becomes similar to the
SISO case except that the noise variances should be
accounted carefully. The covariance matrix of the
noise is E(ññ†) = s2 (H†H)–1. The LLR metric for the
kth bit of the ith layer is given by [19]
~
| ri – si|2
LLR = min ¾ ¾ ¾ ¾ ¾ (24)
ai:ai Î Alk s2[(H†H)–1]ii
~ ~
where ri corresponds to the ith element of r. These
calculations are exactly similar to SISO case, and
hence, the low complexity approximations in [18] are
also valid for this case. A similar formulation for
MMSE case is given in [20]. As the LLR metric is
computed on individual streams, unlike in the ML case
(section 5.1) where it is computed using all streams
jointly, the complexity is much lower.
For the ZF-SIC and MMSE-SIC, though superior
in performance to ZF and MMSE, it is not
straightforward to obtain good soft metrics due to the
possibility of error propagation among the decoded
data.
The performance of 802.11n with 20 MHz
bandwidth for data rate of 130 Mbps, with 2 transmit
and 3 receive antennas is shown for ML and ZF based
soft decoders in Fig 6. The spectral efficiency is
approximately 6 bits/s/Hz. The performance is shown
for tgn-b and tgn-e channels ([14]). No doppler is
introduced in the simulations. Training is performed to
obtain symbol lock and carrier frequency offset.
Fig 6 IEEE 802.11n performance at 130 Mbps with ZF and
ML based soft decoders
Channel estimation is done in the frequency domain. It
can be seen that the ZF based soft decoder performs
within 3 dB of the ML based soft decoder, while its
complexity is much lower.
6. CONCLUSION
In this paper, we have attempted to provide the
underlying motivation for MIMO communications.
Specifically, we have brought out how fading (rich
scattering environment) can be gainfully exploited to
achieve spectral efficiencies which are inconceivable
with a SISO communication. We also looked at
architectures proposed to achieve significant portions
of this huge capacity - D-BLAST and its simpler
version V-BLAST. The typical V-BLAST receivers
are described as well as a low complexity receiver
that bridges the gap between linear receivers, like ZF-
SIC and MMSESIC, and the ML decoders for small
number of transmit antennas. Simulation results are
provided to demonstrate the performance of various
decoders.
We have described IEEE 802.11n [5] as a practical
system where MIMO ideas are utilized. In this context,
we have also presented soft Viterbi techniques [15]
for typical MIMO decoders. We have included some
simulation results to illustrate the performance of
802.11n with indoor channel models.
212 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
7. REFERENCES
1. A J Paulraj, D A Gore, R U Nabar & H Bolcskei, An
overview of MIMO Communications - a key to
gigabit wireless, in Proc IEEE, Feb 2004, vol 92(2),
pp 198-218.
2. E Telatar, Capacity of multi-antenna gaussian
channels, European Transactions on Telecommuni-
cations, vol 10, pp 585-595, Nov-Dec 1999.
3. G J Foschini & M J Gans, On limits of wireless
communications in a fading environment when
usingmultiple antennas, Wireless Personal
Communications, vol 6, pp 311-335, March 1998.
4. S N Diggavi, N Al-dhahir, A Stamoulis & A R
Calderbank, Great expectations: the value of spatial
diversity in wireless networks, in Proc IEEE, Feb
2004, vol 92(2), pp 219-270.
5. Wireless LAN medium access control (MAC) and
physical layer (PHY) specification: Enhancements
for higher throughput, IEEE Draft Standard
802.11nD2, 2007.
6. Air interface for fixed and mobile broadband wireless
access systems: Physical and medium access control
layers for combined fixed and mobile operation in
licensed bands, IEEE 802.16e-2005, 2005.
7. D Tse & P Viswanath, Fundamentals of Wireless
Communication, Cambridge University Press, 2005.
8. G J Foschini, Layered space-time architecture for
wireless communication in a fading environment
when using multiple antennas, in Bell Labs
Technical Journal, 1996, vol 1(2), pp 41-59.
9. P W Wolniansky, G J Foschini, G D Golden & R A
Valenzuela, V-BLAST: An architecture for realizing
very high data rates over the rich scattering wireless
channel, in Proc ISSSE, Pisa, Italy, 1998.
10. B Hassibi, An efficient square-root algorithm for
BLAST, in Proc ICASSP, Istanbul, Turkey, June
2000, pp 737-740.
11. G K Krishnan & V U Reddy, High performance low
complexity receiever for V-BLAST, in Proc IEEE
SPAWC ’07, Helsinki, June 2007.
12. D W Waters & J R Barry, The chase family of
detection algorithms for multiple-input multiple-
output channels, in Proc IEEE Globecom ’04, Nov.-
Dec 2004, vol 4, pp 2635-2639.
13. Y Li & Z Q Luo, Parallel detection for V-BLAST
system, in Proc ICC ’02, 2002, vol 1, pp 340-344.
14. Vinko Erceg et al., Tgn channel models, IEEE
802.11-03/940r4, May 2004.
15. Wireless LAN medium access control (MAC) and
physical layer (PHY) specifications high speed
physical layer in 5 GHz band, IEEE 802.11a, 1999.
16. E Zehavi, 8-psk trellis codes for a rayleigh channels,
IEEE Trans on Comm., vol 40, pp 873-884, May
1992.
17. G Caire, G Taricco & E Biglieri, Bit-interleaved coded
modulation, IEEE Trans on Information Theory, vol
44, pp 927-946, May 1998.
18. F Tosato & P Bisaglia, Simplified soft-output
demapper for binary interleaved COFDM with
application to HIPERLAN/2, in Proc of IEEE ICC
’02, 2002, vol 2, pp 664-668.
19. M R G Butler & I B Collings, A zero-forcing
approximate log-likelihood receiver for MIMO
bitinterleaved coded modulation, IEEE Comm
Letters, vol 8, pp 105-107, Feb 2004.
20. D Seethaler, G Matz, & F Hlawatsch, An efficient
MMSE-based demodulator for MIMO bit-
interleaved coded modulation, in Proc of the IEEE
Globecom ’04, Dallas, TX, USA, November 2004, pp
2455-2459.
 KRISHNAN & REDDY : MIMO COMMUNICATIONS
Authors
V U Reddy joined IIT, Madras as
an Assistant Professor in 1972, and
moved to IIT, Kharagpur as a Professor
in 1976. He held a Visiting Professorship
with Electrical Engineering at Stanford
University during 1979-82. On the
invitation from Osmania University, he
joined the University in April 1982 as a
Professor and Project Director of
Research & Training Unit for
Navigational Electronics; he was its Founding Director. In 1988,
he moved to Indian Institute of Science (IISc), Bangalore as a
Professor of Electrical Communication Engineering (he was the
department chair during 1992-1995). Since 1986, he held several
short-term appointments at Stanford University and University
of Iowa. After retiring from the IISc in August 2001, he joined
Hellosoft India Pvt. Ltd as Chief Technology Officer. During his
tenure at Hellosoft, he worked on WLAN projects involving
802.11b and 11a standards. In June 2003, he joined International
Institute of Information Technology (IIIT), Hyderabad, as
Microsoft Chair Professor where he started wireless
communications research center. He returned to Hellosoft as the
Chief Scientist in December 2005. His present work is focused
in synchronization and channel estimation in wireless systems,
and low complexity decoders for MIMIO systems.
His earlier research interests were in adaptive and sensor
array signal processing.
213
He is a Fellow of the IETE, the Indian Academy of Sciences,
the Indian National Academy of Engineering, the Indian National
Science Academy, and the IEEE. He was the Chairman of Indian
National Committee for International Union of Radio Science
(URSI) (1997-2000). He was a Member of the Editorial Boards
of SADHANA (1992-95), a Proceedings of the Indian Academy
of Sciences, the Indian journal of Engineering and Materials
Sciences (1995-98), and the Proceedings of IEEE (2000-2001).
He received S K Mitra Memorial Award 1989 from IETE for the
best research paper.
* * *
Kalyana Krishnan received his
BTech degree from NIT, Kozhikode
(Kerala) in the year 2000 and ME degree
in Telecom from Indian Institute of
Science, Bangalore in the year 2003. Since
2003, he has been working in Hellosoft
Pvt. Ltd, Hyderabad on WLAN
standards. His focus areas include OFDM
and MIMO technologies.
* * *
 INVITED PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 215-224

Role of Satellite Communication and Remote

Sensing in Rural Development
A S MANJUNATH, D S JAIN, S RAJENDRA KUMAR, AND R V G ANJANEYULU
National Remote Sensing Agency, Balanagar, Hyderabad 500 037, India.
email : manjunath-as@nrsa.gov.in; jain_ds@nrsa.gov.in

The advances in Satellite Communication and Remote sensing has led to the
development of solutions for various national development programs like National Natural
Resource Management (NNRMS), disaster management, tele-education and e-Health care
using telemedicine. The advances in sensor technologies and image processing have
resulted in sensors with increased spatial, spectral and radiometric resolutions and provide
images with rich details for remote sensing applications.
India is a vast country with 29 states and 6 union territories with more than 1 billion
population, which is predominantly rural and distributed at distant geographical locations
apart from the high-density urban areas. Ensuring basic minimum health care to masses
living in remote rural areas is a top priority for any government. Tele-medicine using the
satellite communication as well as Fiber Optic communications is playing an important role
to address health care requirements of people living in remote areas, rural areas, soldiers in
battlefields etc. The Image processing techniques and advances in Database management of
images and text made it possible to create 3D virtual images of human organs by using
imaging techniques from MRI, CT & Ultrasound.
The universalisation of education has become the top priority, especially for the
developing countries. But the extension of quality education to remote and rural regions
becomes a Herculean task for a large country like India with multi-lingual and multi-cultural
population separated by vast geographical distances, and, in many instances, inaccessible
terrain. Satellites can establish the connectivity between urban educational institutions with
adequate infrastructure imparting quality education and the large number of rural and semi-
urban educational institutions that lack the necessary infrastructure.
This paper describes utilization of Satellite Communication and Remote sensing with
advanced techniques in providing solutions to areas of primary importance in rural
development such as harnessing of natural resources, education and healthcare.

INTRODUCTION soil and land degradation, mineral exploration,

groundwater targeting, geomorphological mapping,

S PACE technology comprising of satellite based

communication and Remote sensing which acts as
an eye in the space, have been very powerful tools for
various development plans of a nation. The applications
of the space technology are only limited by the human
ability to put them into use. Emerging trends in the role
of space technology in the areas of Rural Development
like natural resource utilization, education and health
care are presented in the sections that follow.
1. SPACE TECHNOLOGY FOR RURAL
DEVELOPMENT
coastal and ocean resources monitoring, environment,
ecology and forest mapping, land use and land cover
mapping, urban studies, utility mapping, etc. The
information from these analyses can be archived in a
central database with suitable storage and retrieval
mechanisms. Mobile information technology, which is
currently dominated by cellular phones, personal digital
assistants (PDAs) and laptops can be used to
disseminate this information with enhanced features
leading to an information technology culture of
‘anywhere’, ‘anytime’ to deliver a wide range of data
and services.

Natural resources can be estimated and monitored
using Remote Sensing Satellite data and this has been
useful in areas such as water resources, agriculture,
Paper No 125-B; Copyright © 2007 by the IETE.
215
1.1. Role of Remote Sensing in Natural
resources utilisation
Any country’s development depends on its ability
to harness the natural resources like land, water,
216 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
forests etc. As India has an agriculture-oriented
economy, the inputs provided in this area are invaluable
to the farmers. The Remote sensing applications are
diverse, and can be broadly classified into the following
areas: (1) Agriculture (2) Land and soils (3) Forestry
and ecosystem (4)Water resources (5) Geosciences
(6) Infrastructure and town planning (7) Disaster
management.
In a country like India, which has diverse spreads
in terms of geography and population, it is essential to
put satellite data to effective use and ensure that the
advantages percolate to the grassroot levels for overall
development. Management of natural resources,
developmental planning, environmental monitoring and
disaster management are some of the key applications
of space borne Remote Sensing.
1.2. Natural Resources Repository (NRR) -
A space based spatial infrastructure
programme under NNRMS
The National Natural Resources Management
System (NNRMS), for which Dept of Space is the
nodal agency, has the objective of ensuring optimal
management/utilisation of natural resources by
integrating information derived from remote sensing
data with conventional techniques. A large number of
remote sensing application projects in the fields of
agriculture, forestry, environment, geology, ground
water, disasters (flood, drought, earthquake & landslide),
etc., are being carried out under the aegis of NNRMS.
With a large amount of spatial information
generated under NNRMS, one of the important
requirements is to archive, retrieve and serve this
information for various applications. Establishing and
maintaining a national Natural Resources Repository
(NRR) - an infrastructure of distributed GIS data
servers, linked through high-speed network and making
multi-scale natural resources information accessible
through a NNRMS gateway Portal thus becomes
critical. The main aim of the NNRMS-NRR would be
to create and maintain a systematic archive of all the
digital spatial data holdings of thematic and base maps
generated using remote sensing images and promote /
encourage its use for government, business and societal
needs.
As a part of the NRR programme as shown in Fig
1, the Natural Resources Census (NRC) project
addresses periodic inventory of land-use / land cover;
bringing out Large-Scale Maps at 1:10,000 scale using
the high-resolution satellite remote sensing data; It
also involves creating a Natural Resources Data Base
(NRDB) architecture taking care of the horizontal and
vertical networking, data formats, standards taken up
to reap the full benefits of these organised data bases
at various levels. NNRMS Portal serves as the front-
end for the NRR, enabling the users to interact and
obtain the needed data for their applications.
Under NRC, the project of periodic inventory of
land use/land cover at 1:250,000 scale has been taken
up on annual basis using IRS AWiFS data. Assessment
has been successfully completed for first cycle and
second cycle. The detailed inventory of 7 natural
resources themes (land use/land cover, soil,
geomorphology, vegetation, snow/glacier, land
degradation, wetlands) at 1:50,000 scale using IRS
LISS – 3 data has also been initiated. The pilot project
of Large Scale Mapping is nearing completion, and the
operational activity using Cartosat data has been initiated
for selected areas.
The NRR databases have been used by district
and state officials for a variety of applications such
as, watershed development, wasteland development,
integrated land and water resources development
planning, road connectivity analysis, rural amenities
(schools, health Centres etc) planning, disease
infestation analysis, etc. It serves as a backbone
database for supporting the developmental needs.
1.3. The Village Resource Centre (VRC)
Programme
The VRC programme is a unique societal application
conceived by Dept of Space, wherein the capabilities
of satellite communication and earth observation are
integrated, to reach the variety of services emanating
from the space systems and other IT tools directly to
the rural communities. The VRCs promote a need
based single window delivery system for the services
in the areas of education, healthcare, agriculture,
weather, environment, disaster resilience, and livelihood
support to the rural population. It is also aimed to
empower them towards improving their quality of
lives. With the support of the local government and
enhanced participation by the communities, during the
course of time, the VRCs are also envisaged to provide
the e-government services, and act as a village level
helpline.
Space-enabled Village Resources Centre
(VRC)
VRCs, will be configured with satellite bandwidth
based communications backbone, which will be multi-
tasked to enable a variety of community-centric
services like tele-medicine, tele-education, spatial
 A S MANJUNATH et al : ROLE OF SATELLITE COMMUNICATION 217

Create and maintain a systematic

archive of all the digital spatial NR Regional
data of thematic and base maps
generated using RS data Database Internet Servers

NRCENSUS LSM CADASTRAL LIS NR Mgmt.Prj

(1:50,000) (1:10,000) (1:2,000+) (Any scale)
7-8 Layers New base map Referenced User Defined
Every 5 Years Phased Manner “SPOT” dbases

23M/5M 2.5M/1M 1M / AP 23M/5M/

Images Images Images 2.5M/1M
Images

Fig 1 Natural resources repository

information for management of natural resources,
disaster management through participatory approach,
and weather based farmers’ advisories. The VRCs
have a interactive VSAT (Very Small Aperture
Terminal) based video-conferencing and information
exchange facility, linking them with selected knowledge
institutions/ centres/ specialty hospitals, etc., for
providing the above services. The services to be
delivered by the VRCs are depicted in Fig 2.
The VRCs are being set up in large villages/
block/ taluk Hqs., in association with selected partner
institutions, who have strong and committed presence
in the grassroots. The satellite connectivity and
bandwidth, minimal telemedicine and tele-education
facilities, and available/ customized spatial information
on natural resources along with indigenously developed
query system are provided by DOS. The task of
housing, managing and operating the VRCs, with all
relevant contents are taken care by the partner
agencies.
Till date, 180 VRCs have been set up in 11 States,
and another 160 VRCs are being set up. The various

Tele-Education, SATCOM / VSATs: Deployable Access to Fax,

Tele-Medicine Terminals, Mobile WLL VSATs Telephone, Internet

Disaster Mgmt support,

Spatial Info. support,
Emergency communication,
Weather info, Farmers
Vulnerability, Risk & early
advisory services
warning

Fig 2 Space based services for community outreach

218 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

areas addressed/ services provided by these VRCs
include: agriculture development, fisheries development,
supplementary teaching, adult & computer literacy,
alternate livelihood related vocational training, marketing
of agro-products, micro-finance/ enterprises, livestock
management, healthcare, etc. Encouraged by the
positive impact being made by the VRCs, plans are
being drawn to set up VRCs in all the rural/ semi-
urban blocks of the country during the 11th Five Year
Plan period.
1.4. Satellite for rural development
(Gramsat Programme)
The Gramsat Programme (GP) intends to
provide a communications network at the state level
connecting the state capital to districts and blocks and
enabling a reach to villages. The networks provide
computer connectivity, data broadcasting and TV
broadcasting facilities having applications like e-
Governance, Natural Resource Information Service
(NRIS), Development Information, Tele-conferencing,
Disaster Management and Rural/ education
Broadcasting. The GP aims to cover/connect the entire
country – its each and every district and taluka in a
phased manner. Central agencies, the concerned state
governments, their user departments, NGOs, etc. are
the partners. The networks will be ultimately owned
and operated by the state governments and utilized by
all their departments. NGOs are visualised as other
major users. Development and Educational
Communication Unit (DECU) is the nodal agency.
Gramsat Pilot Projects are currently operational in
Madhya Pradesh, Gujarat, Karnataka, Orissa and
three districts of Rajasthan. Planning and project
implementation for other states like Meghalaya and
other North-east Region, Goa, Andaman and Nicobar
Islands, Himachal Pradesh, Chattisgarh and Jharkhand
are underway. Presently there are seven teaching
ends (Uplink and Studio) and more than 2000 Direct
Reception Set (DRS) terminals set up by various state
governments, NGOs and other educational institutions.
1.5. Geo Information in Disaster Manage-
ment
The remote sensing inputs have been used for
many disasters including drought, flood, earthquake,
cyclone, landslides, volcanoes, avalanches, forest fire,
crop pest / diseases, etc., Similarly, cyclone and forest
fires can be monitored and assessment of damage can
be made. The information required for decision making
during drought is diverse and spatial/temporal in nature.
But, synergistic coupling of remote sensing inputs with
conventional systems and space communications, in

Major Activities Services

Disaster Watch/Alert Floods Earthquakes

Acquisition to Satellite Data (Indian & foreign Flood Inundation Maps
Damage Assessment Damage Assessment
satellites)
Hazard Zonation Bank
Aerial surveys (SAR, ALTM, DC) Erosion Studies Geological
Assessment
Generation of vulnerability and hazard
zonation maps
Cyclone
Provide Aero-space information for damage
Landslides
assessment, relief & rescue operations during Inundation Maps
disasters Hazard zonation
Recession Maps
Provide information for planning disaster Damage Assessment
Damage Assessment
mitigation measures
Database warehousing & maintenance Drought Forest Fires
Networking with Emergency Control Rooms Monthly
(National, State, DOS) & Knowlede Agricultural Drought
Active Fire Detection
Institutions and Line Departments Report
Damage Assessment
Support the International Charter on Space End-of-the-season
Agri. Drought Report
and Major Disasters

Fig 3 Activities of decision support centre

 A S MANJUNATH et al : ROLE OF SATELLITE COMMUNICATION 219

well-knit multi-energy interface, offers better
operational services and decision support. For
operationally integrating space inputs and services on
a timely basis Department of Space has launched a
major programme called Disaster Management Support
Programme (DMSP).
Under DMSP, Decision Support Centre (DSC) of
NRSA is providing remote sensing based services in
near real time for monitoring, mapping and management
of natural disasters viz. Drought, Forest Fires,
Landslides, Earthquakes, Floods and Cyclones using
satellite based information along with other ancillary
and ground based information. Decision Support Centre
at NRSA as a single-window operational service
provider for natural disaster management and support
the International Charter on Space and Major Disasters,
as a signatory. The major activities and services provided
by DSC are shown in Fig 3.
1.6. Remote Sensing and GIS for e-
Governance
National projects using remote sensing data, like
Drought Monitoring, Crop Acreage and Production
Estimation, Mapping and Damage Assessment of
Disasters like Floods, Cyclones, Landslides, etc.,
Snowmelt Runoff Forecasting, Wasteland Mapping,
Urban Studies, Soil Mapping and Land Degradation
Studies, Geological Surveys, Mineral Targeting, Drinking
Water Potential Zone Mapping, Watershed
Development Planning, Irrigation Command Monitoring,
Forest Resources Assessment, Potential Fishing Zone
Mapping and several others, have been realized by the
Deptt of Space. The databases of these projects can
be maintained and can be made accessible to users of
interest for better planning.
While remote sensing data is being utilised to
prepare thematic maps/information on various resources,
thrust has been on integrating these information with
conventional data sets towards generating action plans
for sustainable development at local level. Integrated
Mission for Sustainable Development has helped in
preparation of location-specific action plan. Besides
these, Environment Impact Assessment Studies like
Forest Fire Mapping and Monitoring, Assessing the
Impact of Power Plants/Mining Areas, Atmospheric
Aerosol Studies, Coastal Zone Studies, all these form
an integral part of the resource planning. The concept
of e-governance with relevant inputs feasible from
remote sensing is depicted in Fig 4.
2. TELE-EDUCATION USING SPACE
TECHNOLOGY
The pivotal role of education as an instrument of
social change by altering the human perspective and
transforming the traditional mindset of society is well
recognised. The universalisation of education has
become the top priority, especially for the developing
countries. But the extension of quality education to

IRS Satelites
Toporgaphics Data Storage
and Existing & Retireval
Maps & Maps
Census Data

Field Survey Data analysis

Aerial Photographs Visual

IRS

Images and Remote Sensing e-Governance

GIS With RAID
Application Theme For Storage & GIS Information
Digital Warehouse
Photo & Information Retirewal
For planners
Digital Analysis

Data from
Census Modelling &
Network
Analysis

Other Sources Aerial Data

Output Forms

Fig 4 Role of remote sensing with GIS for effective e-governance

220 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

remote and rural regions becomes a herculean task services, a need was felt to launch a satellite dedicated
for a large country like India with multi-lingual and for educational service and ISRO conceived the
multi-cultural population separated by vast geographical EDUSAT Project in October 2002.
distances, and, in many instances, inaccessible terrain.
EDUSAT shown in Fig 5 is the first exclusive
Since independence, India has seen substantial increase
satellite for serving the educational sector. It is specially
in the number of educational institutions at primary,
configured for audio-visual medium, employing digital
secondary and higher levels as well as the student
interactive classroom and multimedia multicentric
enrolment. But the lack of adequate rural educational
system. The satellite has multiple regional beams
infrastructure and non-availability of good teachers in
covering different parts of India. The potential uses of
sufficient numbers adversely affect the efforts made
the EDUSAT are depicted in Fig 6.
in education.
EDUSAT is primarily meant for providing
Satellites can establish the connectivity between
connectivity to school, college and higher levels of
urban educational institutions with adequate
education and also to support non-formal education
infrastructure imparting quality education and the large
including developmental communication. ISRO will
number of rural and semi-urban educational institutions
provide technical and managerial support in the
that lack the necessary infrastructure. Besides
replication of EDUSAT ground systems to
supporting formal education, a satellite system can
manufacturers and service providers. The ground
facilitate the dissemination of knowledge to the rural
infrastructure to meet the country’s educational needs
and remote population about important aspects like
will be built and during this period, EDUSAT will be
health, hygiene and personality development and allow
able to support about 25 to 30 uplinks and about 5000
professionals to update their knowledge base as well.
remote terminals per uplink. The indigenous realisation
Thus, in spite of limited trained and skilled teachers,
and launch of EDUSAT has provided a substantial
the aspirations of the growing student population at all
capability for countrywide distance education in India.
levels can be met through the concept of tele-education.
The forerunner for this concept was the SITE (Satellite
3. e-HEALTH CARE USING TELEMEDICINE
Instructional Television Experiment), telecasted
progammes directly to over 2400 villages during 1975-
76, Subsequently in 1983 INSAT was used for similar India is a vast country gifted with ancient historic
tasks. background and geographically nature has provided
with all varieties like the mountain regions, deserts,
In the 1990s, Jhabua Developmental green planes and far-flung areas in the northeast and
Communications Project (JDCP) and Training and the offshore islands of Andaman and Lakshadweep.
Developmental Communication Channel (TDCC) For our population, which is predominantly rural and
further demonstrated the efficacy of tele-education. distributed at distant geographical locations apart from
With the success of the INSAT based educational the high-density urban areas, to provide the basic

Fig 5 EDUSAT : India’s first exclusive educational satellite

 A S MANJUNATH et al : ROLE
Fig 6 Potential uses of EDUSAT
minimum health care has been one of the priorities for
the health administration all along.
In today’s world with several advances made in
the medical field still the benefits are available to the
privileged few, residing mainly in the urban areas.
With the advent of communication technology especially
the SatCom combined with Information Technology,
we have means to benefit from the advance medical
sciences even to the remote and inaccessible areas. It
is known that 75% of the qualified consulting doctors
practice in urban centres and 23% in semi urban
centre (towns) and only 2% operate from rural areas,
where as the vast majority of population live in the
rural areas. Hospital beds/1000 people are 0.19 in
rural as compared to 2.2 in urban areas.
Telemedicine helps patients in distant and remote
areas to avail timely consultations from specialist
doctors without going through the ordeal of travelling
long distances at large expense. The facility caters
normally for transmission of patient’s medical images,
records, output from medical devices, besides live
two-way audio and video conferencing. With the help
of these, a specialist doctor advises the local doctor or
a paramedic at the patient’s end, on line, on medical
care. In the context of distant and rural areas, the
telemedicine-based medical care is also highly cost
effective.
OF SATELLITE COMMUNICATION 221
Telemedicine as shown in Fig 7 is a confluence of
communication technology, the information technology,
medical engineering and medical science. The
Telemedicine system consists of customized hardware
and software at both the Patient & Doctors end with
some of the diagnostic modalities like ECG, X-ray &
pathology provided at the patient end. They are
connected through a communication backbone, which
could be VSAT or ISDN lines or leased lines or even
a Telephone lines.
ISRO’s Telemedicine initiative has been broadly
divided into the following areas:
a) Providing Telemedicine Technology &
connectivity between remote/rural hospital and
Super Speciality Hospital for Teleconsultation,
Treatment & Training of doctors, nurses &
technicians.
b) Providing the Technology & connectivity for
Continuing Medical Education (CME) between
Medical Colleges & Post Graduate Medical
Institutions/Hospitals.
c) Providing Technology & connectivity for
Mobile Telemedicine units for rural health
camps especially in the areas of ophthalmology
and community health.
222 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
Patients End
Referral Hospitals
Radiology
Pathology etc.
Fig 7 Tele-medicine system
d) Providing Technology & connectivity for
service during disasters like Flood, Cyclone,
and Earthquake, etc.
Further, towards reaching speciality healthcare to
larger sections of the society, across the country, and
in various faculties of medical specialities; the need
Referral Hospitals
• GB Pant Hospital, Port Blair
• ISRO’s SHAR hospital, Sriharikota
• Aragonda Apollo Hospital, AP
• District Hospital, Chamrajanagar
• Vivekananda Memorial Hospital,
Sargur
• Tripura Sundari District Hospital,
Udaipur
• Guwahati Medical College &
Hospital
• District Hospital, Leh, Srinagar,
Kahtue
• Indira Gandhi Hospital, Karavati
• Medical College Hospitals, Cuttack,
Behrampur & Butta
Fig 8
III SAT
Export Doctors End
Health Specialist
centres
Panel of doctors
Mobile Service
for demonstrating further ‘technology packaging’ and
implementation efforts - by way of ‘Point-to-Multipoint’
and ‘Multipoint-to-Multipoint’ type of connectives -
has been realized. These efforts would also facilitate
imparting training to local doctors and paramedics
working in rural areas, besides supporting Continuing
Medical Education (CME) efforts.
Speciality Hospitals
• Sri Ramachandra Medical College and
Research Institute, Chennai.
• Apollo Hospitals, Chennai
• Narayana Hrudayalaya, Bangalore
• Rabindranath Tagore International
Institute of Cardiac Sciences, Kolkata
• ANMS, New Delhi
• Amritha Institute of Medical Sciences,
Cochin.
• Sanjay Gandhi Post Graduate Institute
of Medical sciences, Lucknow.
ISTRAC.
Bangalore
ISRO tele-medicine links
 A S MANJUNATH et al : ROLE
ISRO has identified “Health Communication
Support Network” as an important activity. In this
connection, the types of services which are planned to
be initiated are Tele-medical-education, Tele-
consultation, Tele-referral and Tele-health. ISRO has
plans to integrate telemedicine with tele-education
wherever applicable and the Village information Kiosks/
Community Centres to reach out to more rural areas
of the country. The various tele-medicine links
established by ISRO are shown in Fig 8.
3.1. Image processing in e-Healthcare
Applications of telemedicine are found in every
area of specialisation. Teleradiology is the most common
application followed by cardiology, dermatology,
psychiatry, emergency medicine, home health care,
pathology, and oncology, etc. The technological basis
and the practical issues are highly variable from one
clinical application to another. The image processing
methods like contrast enhancement, segmentation,
registration, clustering, mosaic, 3D constructions,
rendering, etc., are applied for various medical images.
Other image processing techniques such as Image
acquisition, digitization and image compression are
used effectively for image transmission.
3.2. Remote sensing and GIS for vector
borne Diseases
Vector-borne diseases have been the most
important health problems worldwide for many years
and still represent a constant and serious risk to a large
part of the world’s population. Satellite Remote Sensing
techniques are used to monitor factors affecting disease
transmission. Both spatial and temporal changes in
environmental conditions are important determinants
of vector-borne disease transmission. The satellite
images are used for identifying these changes, and
able to define and predict areas and periods of high
spreading. Satellite images, digitized land use maps
and Geographic Information Systems (GIS) are used
for predicting changes in habitats of disease vectors,
and high resolution satellite data is used to characterize
immature habitats of vector borne diseases like cholera,
OF SATELLITE COMMUNICATION 223
malaria etc., Remote Sensing provides useful indicators
for identifying the presence or impact of the vectors,
which can be analysed to provide an effective solution.
CONCLUSION
The paper on “Role of Satellite Communicatgion
& Remote Sensing in Rural Development” describes
the efforts by the dept of Space in harnessing Space
Technology and Image processing and GIS in areas
that influence rural development viz, natural resources
utilisation, education and health care etc. As the current
trend is integration of technologies for enhanced utility,
the space technology can be made integral part in
various phases of rural development for the country’s
overall development.
ACKNOWLEDGEMENTS
We are grateful to Dr K Radhakrishnan, Director,
NRSA for his encouragement in bringing out this
publication.
REFERENCES
1. http://www.isro.org
2. http://www.nnrms.gov.in
3. Dr. Rao, K M M, et al, Emerging Technologies and
Remote Sensing with GIS for effective e-
Governance, Proceedings of National Conference
on e-Governance, 10th -12th September, 2004, IETE,
Hyderabad-7.
4. Adam William Darkins, Margaret Ann Cary
Telemedicine and Telehealth – principles, policies,
performance and pitfalls, Springer publishing
company, 2002.
5. Marlene Maheu, Pamela Whitten, Ace Allen, E-
Health, Telehealth, and Telemedicine, Wiley
publishers, 2001.
6. David L Verbyla, Satellite Remote Sensing of Natural
Resources, CRC Publishers,1990
7. ISPRS proceedings – Istambul 2004, Volume XXXV,
Parts B1-B6.
8. Proceedings of the National Workshop on Flood
Disaster Management – space Inputs, Organised by
NRSA, ITC Netherlands, 2004.
224 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
Authors
A S Manjunath obtained his BE
degree from Bangalore University
(1970) and MTech from IIT, Madras
(in 1972). He then joined Electronics
Systems Division of Space
Applications Centre, Ahmedabad in
1972 where he worked for the Satellite
Instructional Television Expt. (SITE).
He joined the National Remote Sensing
Agency in 1975 and since then he is
with NRSA. He has worked in the areas of Image Processing,
Microwave Remote Sensing, Artificial Intelligence, Data
Quality Evaluation and Quality Assurance activities at NRSA.
He developed a real aperture X band Side Looking Airborne
Radar System. Presently as Dy Director (Data Processing
Area), he is handling the data processing activities which
include Product Generation, Data Quality Evaluation, Product
Quality Control, Microwave Remote Sensing, Special Products
generation, Imaging systems, QA activities, data archival,
Software development, hardware development, etc. He has
made significant contributions in these areas of activities at
NRSA. His research interests are in the image processing and
microwave remote sensing areas.
* * * NRSA. He has worked in the areas of
D S Jain was born in December,
1953 in a village Mandalgarh in Rajasthan
State. He has done his BE (Hons)
(Electronics in 1975 and ME
(Electronics) in 1977, both with
distinction, from BITS, Pilani.
During 1977-79 he worked as
Scientist at Central Electronics
Engineering Research Institute (CEERI),
Pilani where he was involved in design and development of
process control instruments for sugar industry which is patented
in India.
Since 1979 he is working at National Remote Sensing Agency
(Department of Sapce, Govt of India), Hyderabad. Initially he
was involved in development of several innovative technologies
used for Remote Sensing applications such as Color Photowrite
System, Satellite Image Processing System (SIPS), Additive
Color Viewer, Dual Densitometer, Image Analyser etc. These
technologies were subsequently transferred to the Industry.
At NRSA he has held several key positions in past on such
as Project Director, ASDF; Deputy Project Director, IRS-1B;
Operations Director, IRS-1B; Deputy Project Director, Cartosat-
1 etc. Presently he is Engineer ‘G’ and Group Director (Imaging
Systems and Special Products) wherein he is responsible for
development and generation of Value Added customized Products
from Remote Sensing data acquired from various Satellites.
He has been honoured with NRDC award from President of
India for meritorious invention of Color Photowrite System;
Best Import Substitution and Best Invention awards for
technological developments from Exhibition Society, Hyderabad;
‘The Outstanding Young Indian (TOYI)’ Award-1991 by the
Indian Junior Chamber Secunderabad Chapter, AP State Chapter
and IJC National body for his contribution and achievement in
scientific & technological development; 2 patents; several design
articles and publications in international technical journals. He
was selected in the final panel for Indian Astronaut to fly on-
board US space shuttle ‘Challenger’ in 1986.
He is a Senior Member of IEEE (for past 19 years) and life
member of ISRS.
* * *
S Rajendra Kumar obtained his
Diploma in Electronics &
Communication Engineering in 1974. He
worked in Electronic Research Laboratory
as a technical assistant in the field of
VHF/UHF receivers for about five years.
He joined the National Remote Sensing
Agency in 1989 and since then he is with
Digital Film Recorders Development.
Trained on Digital film recorders at Canada in 1984 and 1987. He
was also recipient of NRDC award for design and development
of small format digital film recorder. Presently he is Head, Special
product development Section involved in Image Processing,
Special product generation, S/w and H/w development activities
and direct writing large format film recorder operations.
* * *
R V G Anjaneyulu obtained his
BTech degree from JNTU University,
Hyd (1996) and MTech from Osmania
University, Hyd (in 2000). He worked
as a technical officer in ECIL, Hyderabad
for about two years. He joined the
National Remote Sensing Agency in 1998
and since then he is with NRSA. He has
worked in the areas of Digital Film
Recorders maintenance, Image
Processing, Special product generation, S/w and H/w development
activities at NRSA making significant contributions in these
areas. His research interests are in the areas of image processing
namely merging methods improvement / development, Natural
color composite generation for the IRS images etc.
* * *
 INVITED PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 225-242

Design of a TDD Multisector TDM MAC for

the WiFiRe Proposal for Rural
Broadband Access
ANITHA VARGHESE*
General Motors India Science Lab, Creator Building, ITPL, Bangalore 560 066, India.
email: anitha.varghese@gm.com
AND

ANURAG KUMAR
Department of Electrical Communication Engineering,
Indian Institute of Science, Bangalore, 560 012, India.
email: anurag@ece.iisc.ernet.in

1. INTRODUCTION • In order to leverage the price advantage of

The WiFiRe (WiFi Rural Extension) proposal for
rural broadband access is being developed under the
aegis of CEWIT. The system leverages the widely
available, and highly cost-reduced, WiFi chipsets.
However, only the physical layer from these chipsets
is retained. A single base station carries several WiFi
transceivers, each serving one sector of the cell, and
all operating on the same WiFi channel in a time
division duplex (TDD) manner. We replace the
contention based WiFi MAC with a single-channel
TDD multisector TDM MAC similar to the WiMax
MAC. In this paper we discuss in detail the issues in
designing such a MAC for the purpose of carrying
packet voice telephony and for Internet access. The
problem of determining the optimal spatial reuse is
formulated and the optimal spatial reuse and the
corresponding cell size is derived. Then the voice and
data scheduler is designed. It is shown how throughput
fairness can be implemented in the data scheduler. A
capacity assessment of the system is also provided.
1. INTRODUCTION
The WiFiRe standard for rural Internet access
(see [1], and [2]) is being developed under the aegis of
CEWIT, IIT Madras, as a technology for providing
wireless broadband voice and data access for rural
areas. The following are the key features of the
current version of this standard.
Paper No 124-E; Copyright © 2007 by the IETE.
* This work was performed while the author was at ECE
Deptt, IISc, Bangalore.
225
using existing mass produced integrated
circuits, the physical layer has been taken to
be the same as that of IEEE 802.11, the
popular standard for wireless local area
networks (WLANs).
• One access point (AP) (or base station
controller (BSC)), using a single IEEE 802.11
channel, will serve a “cell” with about 80-120
villages spread over a 15 km to 20 km radius.
• The cell will be sectored (typically 60o), with
each sector containing a directional base
station (BS) antenna. There will be one fixed
subscriber terminal (ST) in each village, which
could be connected to voice and data terminals
in the village by a local area network. The ST
antennas will also be directional, thus
permitting reliable communication between
the BS antenna in a sector and all STs in that
sector. However, because of antenna side-
lobes, transmitters in each sector will interfere
with receivers in other sectors. Depending
on the attenuation levels, a scheduled
transmission in one sector may exclude the
simultaneous scheduling of certain transmitter-
receiver pairs in other sectors. Further,
simultaneous transmissions will interfere,
necessitating a limit on the number of
simultaneous transmissions possible. A typical
configuration of a WiFiRE system is shown
in Fig 1.
• There will be one MAC controller for all the
sectors in a cell. The multiple access
mechanism will be time division duplexed
226 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
Fig 1 WiFiRe network configuration. The figure on the left shows a deployment with three sectors, and the figure to
the right shows a tall tower carrying several BSs, with sector antennas, and several STs in a sector, with lower
height directional antennas
multisector TDM (TDD-MSTDM) scheduling
of slots. Time is divided into frames, which
contain traffic slots. The set of slots in a
frame is partitioned into contiguous uplink and
downlink segments. During the downlink
segment, in each slot, one or zero transmissions
can take place in each sector; and similarly in
the uplink segment. Because of site and
installation dependent path loss patterns, and
because of time varying traffic requirements,
the schedule will need to be computed on-line.
The objective of this work is to abstract out the
basic scheduling problem, to develop a mathematical
formulation for the problem, to provide some scheduling
algorithms, and to provide a capacity assessment of
the MAC architecture.
A. Related Literature
Bhagwat et al [3] have discussed issues related
to using 802.11 family of wireless technologies for
long distance transmission in rural environment, such
as the quality of 802.11 PHY performance outdoors,
range extension, spectral vs. cost efficiency. The
authors provide details of the 802.11-based mesh
network deployed in the Digital Gangetic Plains Project
providing voice and data services to villages. Raman
and Chebrolu [4] discuss the issues in using CSMA/
CA in networks including long distance links. CSMA/
CA is designed to resolve contention in the indoor
environment, but is inefficient in long distance point to
point links. The authors provide a new MAC for mesh
networks synTX/synRX, which in the context of our
problem translates to saying that the antennas at the
base station should either all be in transmit mode or all
in receive mode and the transmissions should satisfy
some power relations. These ideas have been made
use in the spatial reuse model that we discuss.
Shetiya [5] considers the joint routing and
scheduling problem in WiMax mesh networks. A
dynamic programming problem is formulated to
maximize the throughput, and is found to be
computationally complex to solve. Heuristics are used
to attain a near optimal solution by considering the
routing and scheduling problems separately.
B. Preview of Contributions
We begin by developing a model for antenna
coverage and spatial reuse in a single channel multi-
sector operation. It is seen that in multi-sector operation,
depending on the path loss, receive sensitivity and the
antenna directivity, the number of simultaneous
transmissions can be 3 or 4. We then set up an
abstraction for the TDD, single channel multi-sector
scheduling problem.We begin by developing capacity
bounds for fair rate allocation, sum rate and sum of log
rate.This analysis also shows how the sectors should
be angularly oriented.We then develop a scheduling
methodology for voice and data traffic.
C. Overview of the Paper
Section 2 sets up the notations used through the
rest of the report and also explains the voice and TCP
traffic models used. Section 3 explains the model used
to characterize the interference in the network by
disallowing transmission in some regions and by limiting
 VARGHESE & KUMAR : DESIGN
the total number of simultaneous transmissions. Section
4 provides bounds on the capacity of the system. The
optimum antenna positioning can be obtained based on
these bounds. Section 5 gives the scheduling problem
in hand. A dynamic programming problem formulation
of the problem is given in Section 5C. In this section,
we also propose a greedy heuristic scheduler for
uplink and downlink. Section 5G gives a scheduling
algorithm to improve the fairness among users.
2. PROBLEM SETUP
A. Some Typical System Parameters
Typically, there will be 80 to 120 subscriber
terminals (STs) in a 15 to 20 km radius covered by a 6
sector system. Each station will be associated with a
base transceiver station (BTS). The TDD-MSTDM
scheduling is done over a frame. A typical frame time
is 10ms with slot time of 32µs, giving rise to 312 slots
per frame. The frame is divided into downlink and
uplink segments in a ratio which is a design parameter.
During downlink transmissions, a significant amount
of power from the transmitting BTS reaches other
BTSs, the distance separating them being very small.
So, when downlink transmissions are scheduled in any
one of the sectors, other BTSs cannot be in the
receive mode. Therefore, downlink and uplink
transmissions must alternate over the entire cell. It
follows that the ratio of number of slots in uplink to
that in downlink must be the same in all sectors. This
ratio is kept constant. A beacon marks the beginning
of the frame and also carries the scheduling MAP. A
total of 24 slots are needed for the beacon in every
frame.
All links are at 11 Mbps. A slot is of 32µs. At 11
Mbps, this is 44 bytes. A VOIP packet is 40 bytes
long. Thus, assuming that the MAC overhead is 4
bytes per packet, the transmission of a VOIP packet
can be done in a slot. Each transport block (TB) has a
96µs PHY overhead, i.e., 3 slots. Hence, the minimum
size of a transport block is 4 slots. A TB should fit into
an integral number of slots. An uplink TB is always for
one ST, but downlink TBs can be for multiple STs.
There is a maximum size of TB (Tmax) which indicates
PHY limitations or may correspond to higher layer
limitations.
Implications for Scheduling: Since each TB
involves a 3 slot overhead, it is advantageous to use
long TBs. However, this would result in starving some
STs while favoring others. Note also that, because of
3 N
Tmax, there is a minimum overhead of ¾ ¾ slots per
Tmax
slot.
OF A TDD MULTISECTOR TDM MAC 227
B. Directional Antennas and Intersector
Interference
The radiation pattern of a typical antenna used in
the deployment is shown in Fig 2. Based on the
antenna pattern, we can divide the region into an
association region, a taboo region and a limited
interference region with respect to each BTS.
The radial zone over which the directional gain of
the antenna is above –3dB is called the association
region. In our analysis, we take the directional gain to
be constant over this region. Any ST which falls in this
region of a BTS antenna j is associated to the BTS j.
The region on either side of the association region
where the directional gain is between –3dB and –15dB
is called the taboo region. Any ST in this region of
BTS j causes significant interference to the transmissions
occurring in Sector j. When a transmission is occurring
in Sector j, no transmission is allowed in this region.
In the limited interference region the directional
gain of the BTS antenna is below –15dB. A single
transmission in this region of BTS j may not cause
sufficient interference to the transmission in Sector j.
But a number of such transmissions may add up
causing the SINR of a transmission in Sector j to fall
below the threshold required for error free
transmission.We take care of this by limiting the total
number of simultaneous transmissions in the system
as explained in section 3.
As an example, for the antenna pattern shown in
Fig 2, the association region is a 60o. sector centered
at the 0o. mark, the taboo region is 30o. on either side
of this association region, and the limited interference
region covers the remaining 240o.
C. Notation and Terminology
n he number of BTSs (e.g., 6); BTSs are indexed
clockwise; for Sector j the interference region
in the previous counter-clockwise sector will
be denoted by j– and the next clockwise
sector will be indexed by j+.
n0 the number of sectors that can have
simultaneous transmissions; see section 3.
m the number of STs (e.g., 40 or 120); the
number of STs in Sector j will be denoted by
mj and the number of taboo STs in the previous
counterclockwise previous sector by mj– ;
similarly we define mj+.
the number of slots in a scheduling frame.
Each slot is used either for downlink
communication or for uplink communication.
228 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

Fig 2 Radiation pattern of a typical BS antenna that could be used in the deployment.
The association region is a 60o. sector centered at the 0o. mark, the taboo region
is 30. on either side of this association region, and the limited interference region
covers the remaining 240o

For example, N = 312 slots, as per the
numerical values provided earlier.
A the association matrix; an m × n matrix, where
each row corresponds to an ST and each
column corresponds to a BTS. The (i, j)th
element of the matrix is a 1 if the ith ST is
associated with BTS j. Otherwise it is 0. We
will sometimes refer to each nonzero element
of A as a link.
B(i) the BTS with which ST i, 1 £ i £ m is associated.
i.e., Ai,B(i) = 1.
I the exclusion matrix; an m × n matrix, where
each row corresponds to an ST and each
column corresponds to a BTS. The (i, j)th
element of the matrix is 1 if ST i is taboo for
BTS j or i is associated with j. Otherwise it is
0. Note that the matrices A and I together
define the scheduling constraints.
u: Activation vector: a 1 × n matrix, where the
ith element denotes which ST in Sector i is
transmitting. If we decide to transmit between
ST j and its BTS, say BTS i, the ith element
of u is j. Evidently an activation vector should
satisfy |{j : uj > 0}| £ n0. Also, u must satisfy
the exclusion constraints given by I.
U: Maximal activation vector: If no more links
in an activation vector can be activated without
causing interference to some other
transmissions scheduled in the same vector
then this activation vector is maximal.
u: Activation set: the set of all maximal activation
vectors.
S: A schedule: A schedule is an N × (n + 1)
matrix, with rows corresponding to slots and
columns (except the last column) corresponding
to sectors, where the (i, j)th entry corresponds
to the link in the jth sector that is scheduled to
transmit in the ith slot. If no ST in the jth
sector can transmit in the ith slot of the frame
(because this will interfere with other scheduled
transmissions in the frame) the corresponding
entry is 0. The last column indicates the number
of consecutive slots for which the activation
vector is used.
 VARGHESE & KUMAR : DESIGN
I(s): This is the set of links that can interfere with
any of the links in s.
D. Traffic Models, QoS Objectives
A possible network architecture for a WiFiRe
deployment is shown in Fig 3. There are a number of
telephones and PCs connected to the WiFiRe BTS
through STs. Several BTSs are controlled by a single
base station controller (BSC). The BSC is connected
to the Internet and the PSTN through switches. All
telephony traffic is carried as VoIP over the WiFiRe
access network. For this purpose, notice that there is a
gateway between the PSTN and the WiFiRe network.
For packet voice telephony, we assume that the voice
coder emits a frame every 20 ms. Assuming a frame
time of 10 ms we need one voice packet to be
transmitted in each direction (uplink and downlink),
every 2 frames. This also implies that if a voice packet
is transmitted in the frame following the one in which it
arrives, then its delay is bounded by 20 ms. We
propose to admit only so many VoIP calls, so that the
probability of a voice packet not getting transmitted in
the slot after the one in which it arrives is small, say
0.02.We note that the slot utilization can be optimized
by performing silence suppression before the periodically
arriving voice packets enter the system. This will give
rise to an on-off packet arrival process for each VoIP
call in each direction. The onoff process will be random
(typically modeled by a Markov process). For calls
between the BTS and ST i, it suffices to allocate Ci £
mv,i (mv,i is the number of voice calls for ST (i) slots
Fig 3 A possible network access architecture for a WiFiRE deployment
OF A TDD MULTISECTOR TDM MAC 229
per frame in the uplink and downlink such that the
desired probability of packet dropping is achieved [9,
Chapter 5].
Assuming a mean call holding time of 3 minutes
(µ–1 = 3 minutes) and a call arrival rate of 3 calls per
hour (l = 3/60 per minute), which are typical values
for a home telephone, r = l/µ = 0.15 erlangs. The
frame time in WiFiRE is 10 ms. Given that the vocoder
emits one packet every 20 ms, CBR traffic requires
one uplink slot per call in every 2 frames, and VBR
traffic requires almost one uplink slot per call in every
4 frames. So, with 2 slots reserved for voice calls per
ST per frame, the number of calls that can be supported
is Nv = 4 for CBR traffic, and Nv = 8 for VBR traffic.
With r = 0.15 and Nv = 4, we can have 7 subscribers
with a probability of blocking as low as 0.02. With
r = 0.15 and Nv = 8, we can have 24 subscribers at
probability of blocking 0.02. With 4 slots reserved per
ST these numbers are 24 for CBR calls and 65 for
VBR traffic. Given the village economics we expect
that just 2-4 slots per ST may be all that are required.
One VOIP packet is 40 bytes. The MAC header
has been taken to be 4 bytes, so that transmission of a
lone voice packet can be completed in 4 slots (3 slots
of PHY overhead + 1 slot voice packet). It is possible
to send more voice packets in a single transport block
without additional PHY overhead. Thus, for a single
call from an ST, we need 4 slots each in uplink and
downlink every 2 frames; for two calls from an ST,
we need 5 slots each in the uplink and downlink every
2 frames and so on.
230 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

For TCP controlled data traffic (the predominant
traffic over the Internet), we assume that this wireless
access system is the bottleneck along the path. As a
first model, we assume that TCP packets are
backlogged in each direction (i.e., at the BTS and the
STs) and the scheduling objective is to pack as many
TCP packets as possible into the schedule, after ensuring
that voice QoS is met. We will also consider the
problem of ensuring some form of fairness between
the TCP users.
3. SPATIAL REUSE MODEL
In [6], the authors prove that maximizing the
cardinality of independent sets used in a schedule
need not necessarily increase the throughput, since as
the cardinality of the set increases, the prevailing
SINR drops, thereby resulting in an increase in the
probability of error, decreasing the throughput. Hence
it is necessary to limit the cardinality of the independent
set used so as to satisfy the SINR requirements. i.e.,
there is a limit to the number of simultaneous
transmissions possible.
In this section we consider the problem of finding
the maximum number of simultaneous transmissions
possible in different sectors in the uplink and the
downlink. We assume that there is no power control in
the downlink. The BTS transmits to all the STs at the
same power. We can have static power control in the
uplink. Each ST transmits to the BTS at a fixed power,
such that the average power received from different
STs at the BTS is the same. The STs near the BTS
transmit at a lower power and the ones farther away
transmit at a higher power.
A. Uplink
In the uplink, we assume that there is static power
control. All STs transmit at a power such that the
average power received at the BTS is P times noise
power. Let the maximum power that can be transmitted
by an ST be Pt times noise power. Let R0 be the
distance such that when Pt is transmitted by an ST at
distance R0, the average power received at the BTS is
P0 times noise power, where P0 is the minimum SNR
required to decode a frame with a given probability of
error. Also, let R be such that when Pt is transmitted
from an ST at distance R, power received at the BTS
P be the power required, so that the receiver decodes
the frame with a given probability of error, in the
presence of interferers. The directional gain of the
BTS antenna is –15dB in other sectors. Hence, the
interference power from a transmission in any other
3
sector would be 10–¾2 P. For decoding a frame with
less than a given probability of error, we need a SNR
of P0 at the receiver. If there are n0–1 simultaneous
transmissions, the path loss factor being h, we need R
such that:
P0 ( ¾R ) –h
R
y rcv = ¾ ¾ ¾ ¾ ¾ ¾0 3¾ ¾ ¾ ¾ ¾ ³ P0
1 + (n0 – 1) 10–¾2 P0 (¾R ) –h
R0
–h
R
( ¾R ) – 1
0
n0 £ 1 + ¾ ¾ ¾ ¾3 ¾ ¾ ¾ ¾ ¾ ¾
10 –¾2 P0 ( ¾RR ) –h
0
To provide a margin for fading, we consider a
reduced range R´ such that 10 log ( ¾R´ )–h ³ 2.3s
R
where s is the shadowing standard deviation. Thus
99% of the STs in a circle of radius R´ around the BTS
can have their transmit power set so that the average
power P is received at the BTS in the uplink.
Notice that, to allow spatial reuse, the coverage of
the system needs to be reduced to R´ < R0. There is
thus a tradeoff between spatial reuse and coverage,
which is captured by the spatial capacity measure C =
n0R´2, which has units slots × km2. We note that this
measure has the same motivation as the bit metres per
second measure introduced in [7].
The variation of the number of transmissions and
system capacity with coverage is as shown in Fig 4.
We can see that, for each h, that there is an optimal n0
and R´ such that the coverage is maximum.
The coverage for which the capacity is maximum
can be obtained from dC
¾ = 0 where r´ = ¾R´ Thus
dr´ R0
we get the optimum value of r´ and h0 as
1
æ 2.3 s 1 + 10– ¾3 P0 ö¾
r´ = ç10–¾1 0¾ ¾ ¾ ¾ 2¾ ¾ ÷h
è 1 + ¾h ø
2

P æ R ö –h (1 + 10– ¾32 P0)h

is P times noise power, i.e., ¾ = ç ¾ ÷ n0 = ¾ ¾ ¾ ¾ ¾ ¾
P0 è R0 ø 3
10– ¾2 p0 (h + 2)
In the presence of interferers, the power required
at the receiver will be greater than P0 times noise. Let The results are shown in Table 1.
 VARGHESE & KUMAR : DESIGN OF A TDD MULTISECTOR TDM MAC 231

Fig 4 Variation of the number of simultaneous transmissions possible (n 0) and system capacity (C) with coverage
relative to a reference distance R0 for n = 2.3, 3, 4 and s = 0, 4, 8. Plots for s = 0, 4, 8 are shown left to right

B. Downlink transmit at a power Pt times noise, and repeating the

calculations as for uplink, we find that n0 for downlink
In the downlink, the transmit power is kept gives the same expression as for uplink. The plots and
constant. In downlink, assuming that the BTS antennas tables for uplink applies for downlink also.
232 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
TABLE 1: The optimum coverage normalized to R 0
and the optimum number of simultaneous
transmissions in a multi sector system for
different values of h and s
s 0 4 8
h
h 2.3 3 4
2.3 0.77 0.31 0.12
n0 3 3 4
3 0.78 0.39 0.20
4 0.80 0.47 0.28
C. Number of Sectors
Once we get the maximum number of simultaneous
transmissions possible, n0, we get some idea about the
number of sectors required in the system. In an n0
sector system, when a transmission occurs in the
taboo region between Sector j and Sector j +1, no
more transmissions can occur in Sectors j and j + 1.
So, the number of simultaneous transmissions can be
at most n0 –1, one in Sector j and j +1 and at most one
each in each of the other sectors. Thus maximum
system capacity cannot be attained with n0 –1 sectors.
With n0 +1 sectors, we can choose maximal independent
sets such that the sets are of cardinality n0. So, we
need at least n0 + 1 sectors in the system. From the
spatial reuse model we see that we can have up to 4
simultaneous transmissions in the system, so we need
at least 5 sectors in the system.
4. CHARACTERISING THE AVERAGE
RATE REGION
There are m STs. Suppose a scheduling policy
assigns kj(t) slots, out of t slots, to ST j, such that
kj (t)
lim ¾ ¾ exists and is denoted by rj . Let r = (r1,r2,
t r£ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾
... rm) be the rate vector so obtained. Denote by R(n)
the set of achievable rates when the maximum number
of simultaneous transmissions permitted is
n. Notice that for n1 > n2, R1 É R2. This is evident
because any sequence of scheduled slots with n = n2
is also schedulable with n = n1. In the previous section,
we have determined the maximum value of n, i.e., n0.
Denote R0 = R(n0). A scheduling policy will achieve
an nÎ R 0 . In this section, we provide some
understanding of R0 via bounds.
A. An Upper Bound on Capacity
Suppose each ST has to be assigned the same
rate r. In this subsection an upper bound on r is
determined. In general, the rate vector (r, r, ... r) Ï
R0. The upper bound is obtained via simple linear
inequalities. Consider the case n ³ 3. Suppose one
wishes to assign an equal number of slots k to each ST
in the up link. There are NU uplink slots in a frame.
Consider Sector j, which contains mj STs. Thus k · mj
slots need to be allocated to uplink transmission in
Sector j. When STs in the interference region j– or j+
transmit, then no ST in Sector j can transmit. Suppose
kj± slots are occupied by such interference transmission.
Now it is clear that
k · mj + k j ± = N U
because whenever there is no transmission from the
interference region for sector j there can be a
transmission from sector j. Let mj– and mj+ denote the
number of STs in the interference regions adjacent to
Sector j. Since the nodes in j– and j+ can transmit
together, we observe that
kj± ³ max(k · mj– , k · mj+)
with equality if transmission in j– and j+ overlap
wherever possible. Hence one can conclude that for
any feasible scheduler that assigns k slots to each ST
in the uplink
k · mj + max(k · mj–, k · mj+) £ NU
For large frame time N, divide the above inequality by
N and denote the rate of allocation of slots by r. Thus
if out of t slots, each ST is allocated k slots, then r =
k
lim ¾ £ 1
t
r · mj + r · max(mj–, mj+) £ fu
where fu is the fraction of frame time allocated to the
uplink or
fu
mj + max(mj–, mj+)
This is true for each j. So,
fu
r£ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾
max1£j£n (mj + max(mj–, mj+))
For the case n = 2 for j Î ¸ {1, 2} denote the
interfering nodes in the other sector by mj. One easily
sees that
 VARGHESE & KUMAR : DESIGN
fu
r£ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾
max(m1 + m1´, m2 + m2´)
B. An Inner Bound for the Rate Region
In this section a rate set RL is obtained such that
RL Ì R0. i.e., RL is an inner bound to the achievable
rate set.
Reuse constraint graph: The vertices of this
graph represent the links in the WiFiRe cell. In any
slot we consider only uplinks or only downlinks. Two
vertices in the graph are connected if a transmission in
one link can cause interference to a transmission in
the other link. The reuse constraint graph is represented
as (V, e), where V is the set of vertices and e is the set
of edges.
Clique: A fully connected subgraph of the reuse
constraint graph. A transmission occurring from an
ST in a clique can interfere with all other STs in the
clique. At most one transmission can occur in a clique
at a time.
Maximal clique: A maximal clique is a clique
which is not a proper subgraph of another clique.
Clique incidence matrix: Let k be the number of
maximal cliques in (V, e). Consider the k × m matrix Q
with
ì1 if link j is in clique i
Qi,j = í
î0 o.w.
By the definition of r and Q, a necessary condition
for r to be feasible is (denoting by 1, the column vector
of all 1s.)
Q·r£1
since at most one link from a clique can be activated.
In general, Q· r £ 1 is not sufficient to guarantee the
feasibility of r. This condition is sufficient if the graph
is linear. A linear graph is one in which the nodes can
be indexed in such a way that if nodes i, j, i < j, are in
the clique then each node k with i £ k £ j is also in the
clique; i.e., the nodes of each maximal clique are
contiguous in the indexing. Such a graph will have a
clique incidence matrix of the form
é 1 1 1 1 ... ... 0 0 0 ù
ê ú
ê 0 0 1 1 1 1 ... 0 0 ú
ê ú
Q= ê 0 0 0 0 1 1 1 1 ... ú
ê ú
ê :
. :
. ú
ê ú
ë 0 0 0 0 0 ... 1 1 1 û
OF A TDD MULTISECTOR TDM MAC 233
The reuse constraint graph in the multisector
scheduling problem being considered has a ring
structure. In any case, the set of rate vectors satisfying
Q · r £ 1 provides an outer bound to the rate set. A
linear subgraph can be extracted from the reuse
constraint graph by deleting one sector, or equivalently
setting all rates in one sector to 0. Let us index the STs
in such a way that we can write a rate vector r as:
r = (r1, r2, · · · , rm)
where, for 1 £ k £ m, rk is the rate vector for the STs
in Sector k. Since deleting one sector yields a linear
graph, if r is such that rk = 0, and Q · r £ 1 then r is a
feasible rate vector (which assigns 0 rates to all STs in
Sector k). Linear combinations of feasible rate vectors
are also feasible (since time sharing can be done over
the schedules that achieve these vectors). Define
R := {r : Q · r £ 1, for some k, 1 £ k £ m, rk = 0}
Further, let RL be the convex hull of R. By the
above discussion, it follows that RL Ì R0. Thus we
have an inner rate region. We will use this inner bound
in the next section.
C. Optimum Angular Positioning of the
Antennas
As can be seen from the previous section, the
feasible rates set, R0, of the system depends on the
spatial distribution of the STs around the BTS. Thus
the R0 varies as the sector orientation is changed. A
system where the antennas are oriented in such a way
that most STs fall in the association region of the
BTSs rather than in the taboo region will have more
capacity than one in which more STs are in the taboo
regions.
One sector boundary is viewed as a reference.
Let R0(q) denote the feasible rate set, when this
boundary is at an angle q with respect to a reference
360o
direction. Then, for each 0 £ q £ ¾ n ¾ , we have a
rate region R0(q), where n is the number of sectors.
Since R0(q) is not known, the inner bound RL(q)
(obtained earlier) is used in the following analysis. If
each vector r is assigned a utility function U(r), then
one could seek to solve the problem
max max U(r)
0 £ q £ ¾ ¾ rÎ R0(q).
and then orient the sectors corresponding to the optimum
value of q.
234 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
We can examine various forms for the utility
function. The optimization can be done so as to
maximize the average rate allocated to each ST, with
the constraint that each ST gets the same average
rate. This is called max-min fairness. Trying to optimize
the rates such that the rate to each ST is maximized
will adversely affect the sum capacity of the system.
So, take U(r) = Smj=1 (rj ). This maximises the sum
capacity of the system, giving preference to STs that
are in a favourable position, causing less interference
to other STs. This has an impact on the fairness of the
system. To improve the fairness, we can take the log
utility function, U(r) = Smi=1 log(r(i)); this is known to
lead to what is called proportional fairness.
Evaluating the upper and lower bounds derived
above, we find that the bounds are close to each other.
So, we only report the results from the lower bound.
Hence, we have computed maxrÎ RL(q) U(r) for various
values of . For each we obtain a vector r of average
rates. We evaluate the fairness of this vector by using
the the fairness index is given by . part with NU uplink slots, such that ND + NU = N – NB,
If the rates to different STs are equal, then fairness
index is 1, and the index decreases if there is rate
variability between the STs.
In Fig 5 we plot, as a function of q, the total rate
(left panel) over all the STs for each of the three utility
functions, and also the fairness index (right panel) (the
lower bounds are plotted here). It can be seen that
Fig 5 Variation of sum rate and fairness index with antenna orientation for different utility functions
maximizing the sum rate gives high overall capacity,
but poor fairness. On the other hand, maximizing the
average rate to each ST gives good fairness, but low
sum capacity. Maximizing Sm log ri gives a good trade
off between maximizing the system capacity and
providing fairness. It is interesting to note that in
maximum Sm log ri case, the sum capacity is higher
when fairness is lower and vice versa. For example, at
q = 10, we can see that the sum rate is close to 4. The
fairness index is also close to 1. So, we may choose
this orientation to operate the network.
Note that the above computation can be done off-
line once the ST locations are known. Then the sector
orientations can be obtained from this analysis.
5. SCHEDULING: PROBLEM FORMULA-
TION AND SCHEDULER DESIGN
Based on the discussion up to this point, the
scheduling problem we are faced with is the following.
First partition the frame of size N slots into a
contiguous part with ND downlink slots and an uplink
where NB is the number of slots required for the
periodic beacon. Typically we will have ND » NU. This
is because data transfer traffic is highly asymmetric,
as users download a lot more than they upload. During
downloads, long TCP packets (up to 1500 bytes) are
received in the downlink and one 40 byte TCP ACK is
sent in the uplink for every alternate TCP data packet
received.
 VARGHESE & KUMAR : DESIGN
Now, when mv,i, 1 £ i £ m, VoIP calls are admitted
for ST i, we need to determine the number of slots Ci
to be reserved in the uplink and downlink subframes
for ST i, such that the QoS targets are met for all the
voice calls. For doing this, evidently the set of vectors
C = (C1, ...,Cm) that are feasible (i.e., can be scheduled)
needs to be known. For each A and I, there will be an
optimal set of such vectors Copt(A, I), and for any
practical scheduler, there will be an achievable set of
admissible vectors C Î ¸ Copt(A, I).
Once the required vector of voice payload slots
has been scheduled, we need to schedule as many
additional payload slots, so as to maximize the traffic
carrying capacity for TCP while ensuring some fairness
between the flows.
A. Obtaining the Activation Set
Consider a graph with the STs and BTSs as nodes
and the communication links between the STs and
BTSs as edges. An activation vector is a matching on
this graph. [8] proposes randomized algorithms that
can be used for finding nearmaximal matchings, with
complexity O(Number of nodes). But, the inherent
graph in the problem we consider being bipartite in
nature, and the scheduling being centralized, the maximal
matchings can be found without randomized algorithms.
The algorithm for enumerating a maximal activation
vector is as follows.
Algorithm 5.1
1) Choose a link to be included in the activation
vector. This might be based on criteria such as
(i) the link with the longest queue length, (ii) or
the link that received the lowest average rate
over previous window of frames.
2) Eliminate all the links that can cause interference
to transmission on the set of links chosen until
this point.
3) Choose the best link (according to the above
criterion) from the remaining set of links.
4) Repeat Steps 2 and 3 until there are no more
links that can be activated, or the set contains n0
links (the maximum that can be activated at a
time).
Remarks 5.1:
Suppose that we consider only TCP traffic. If we
try to maximize the number of useful slots without any
regard to fairness, then the schedule will be to use the
vector U with maximum cardinality (U Î u) for all
OF A TDD MULTISECTOR TDM MAC 235
slots. Then, a bound on the total number of slots
available for transmission would be
æ Tmax – 3 ö
|U| ç¾ ¾ ¾ ¾ ÷ N
è Tmax ø
if the frame size N is an integral multiple of Tmax. Note
that the maximal activation vector depends on the
antenna used, the distribution of STs, etc. If there are
several activation vectors with with the maximum
cardinality, we can achieve some fairness by cycling
between these vectors every Tmax slots. Still, this
throughput maximising approach may result in some
STs getting starved.
B. The Optimal Scheduling Problem for
Uplink
As an illustration, we focus here on the uplink
scheduling problem. The uplink scheduling problem
avoids the slight complexity in the downlink problem,
that, in the downlink, we can combine transmissions
from a BTS to multiple STs in the same TB.
An instance of a scheduling problem is defined by
an association matrix A, an interference matrix I, and
a vector C Î Copt(A, I) of required uplink voice
capacities for the STs. We formulate the uplink
scheduling problem as a constrained dynamic program
over a finite horizon NU, i.e., over the indices k Î {1, 2,
..., NU}, where NU is the number of uplink slots in a
frame. The state of the system at the beginning of slot
k is denoted by (xk, qk), where
xk: a 1 × m vector with xki denoting the number of
consecutive slots for which ST i has been
transmitting; clearly, x0 = (0, 0, ..., 0)
qk: a 1 × m vector, with qki being the number of
required voice slots yet to be scheduled; clearly
q0 = C = [C1,C2, ..., Cm]
At the beginning of slot k, k Î ¸ {0, 1, ..., NU–1} an
activation vector uk Î ¸ u is selected. Then in the slot,
all links appearing in u are allowed to transmit, with
the voice queues being depleted first. The state evolves
as follows.
xk+1,i = f1(xki, uki)
ì (xk,i + 1)I{uk,B(i)= i} xk,i + I{uk,B(i)=i} < Tmax
=í
î 0 xk,i + I{uk,B(i)=i} = Tmax
i.e., if ST i is scheduled in slot k, (uk,B(i) = 1) and the
maximum TB length has not been reached, then we
236 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
increment the burst length from ST i by 1, else the
burst length is reset to 0. Further, qk, k ³ 0, evolves as
follows.
qk+1,i = f2(qki, xki, uki) = (qk,i – I{uk,B(i)=1} I{xk,i>3})+
i.e., the number of required voice slots reduces by 1
provided the overhead part of the current TB has
elapsed. Note that x+ = max(0, x).
We now define the reward structure. We wish to
satisfy the need for voice slots and having done that,
we wish to maximize the number of slots remaining
for TCP data. Define the reward in slot k, 0 £ k £ N –
1 by
m
gk(xk, uk) = S I{uk,B(i)=i} I{xki>3}
i=1
i.e., this is the total number of payload slots scheduled
in slot k. Clearly, gk(xk, uk) £ n, since there can be at
most n transmissions at a time.
Then, we set a terminal cost
ì 0 qN = 0
gN = í
î –¥ if qNi > 0 for some i
i.e., we incur an infinite cost if we are unable to
schedule all the required voice slots.
A scheduling policy maps the state at slot k to a
vector u Î ¸ u. Let p denote a generic policy. Then
there is a sequence of functions pk : (x, q) ® u that
define the policy. Define
N–1
JNp (0, C) = S gkp (Xk, uk) + gNp
k=0
where Xk evolves as explained as above under the
policy p We wish to solve
max Jnp (0, C)
p
and obtain the optimal policy. Let JN* (0, C) denote the
optimal value and p* be an optimal policy. Since the
number of policies are finite for each C Î Copt(A, I),
JN* (0, C) is finite and there exists a p*.
C. On Solving the DP formulation
In the dynamic programming formulation of the
scheduling problem, the state of the system can be
æ xk ö
written as, Xk = ç ÷ . The system evolves as
è qk ø
æ f (x , u ) ö
Xk+1 = f(Xk, uk) = ç 1 k k ÷
è f2(qk, xk, uk) ø
The single stage reward in using the control uk,
when the system is in state Xk is given by
m
g(Xk, uk) = S I{uk,B(i)=i}, I {xki>3}
i=1
Terminal cost is
ì0 qN = 0
JN(XN) = í
î–¥ o.w.
The control uk, i.e., the activation vector to be
used in slot k, is the one that attains the maximum in
the recursion
Jk(XK) = max {g(Xk, uk) + Jk+1 (f(Xk, uk), uk)}
We can do a backward recursion, with all possible
XN , and proceed to find the controls that maximize the
reward at each stage. The set of controls, {u0, u1,...,
uN–1} that maximizes J0(0, q0) as obtained by the
above recursion is the optimal schedule.
But, this approach is feasible [10, Chapter 1] only
when the number of stations is small. The size of
space occupied by xk is almost (¾ ¾ ¾) n, since one
station could be transmitting in each sector, and xki
may be any integer between 0 to Tmax. So also, the
number of controls that can be applied increases as
O(Nm), since we can choose one station from each
sector, such that it obey the constraints. The exponential
increase in the size of state space and control space
with the number of stations make this approach
infeasible.
D. A Greedy Heuristic Scheduler for
Voice in Uplink
At each slot k, we heuristically build an activation
vector uk Î u starting from an ST in {i : qk,i =
maxj qk,j}. Then we follow the approach in Algorithm
5.2 each time we choose an ST with max residual qk,j
Algorithm 5.2:
1) Modify the voice queue lengths to include the
overhead slots required. i.e., if an ST has a voice
queue of 2 packets, add 3 slots of PHY overhead
to make the queue length 5.
 VARGHESE & KUMAR : DESIGN
2) Initially, slot index k = 0. Let ST i be such that
qki = max {qkl}
l=1...m
i.e., The ST with longest voice queue at the
beginning of slot k is i. Form activation vector u
with link i activated. i.e., u = {i}
3) Let ST j be such that
qkj = max {qkl : l Ï I(u)}
l
j is such that it is the noninterfering ST with
maximum queue length. Augment u with link j.
Now, find I(u) corresponding to the new u.
4) Repeat Step 3 until the activation vector that we
get is a maximal activation vector.
5) Let
n = {qkl : min (qkl, l Î u)}
l=1,...,m
i.e., n is the minimum number of slots required
for the first ST in u to complete its transmission.
Use u in the schedule from kth to (k + n)th slot.
Fig 6 A typical deployment of a system with 3 sectors and 15 STs
OF A TDD MULTISECTOR TDM MAC 237
ì q –n for i Î u´
qk+n,i = í k,i
î qk,i for i Ï u´
and k = k + n i.e., slot index advances by n, and
the queue length for the STs at the beginning of
k + nth slot is n less
6) At the end of the k + nth slot,
u = u – {l : qkl = min(qkl, l Î u)}
i.e., remove from the activation vector, those
STs that have completed their voice slot
requirement.
7) Go back to Step 3 and form maximal activation
vector including u. Continue the above procedure
until q = 0 or n = NU In this step, we form a new
activation vector with the remaining STs in the
activation vector (which need more slots to
complete their requirement).
8) Once the voice packets are transmitted, we serve
the TCP packets in the same way, except that if
in forming a maximal activation set, it is found
that the only schedulable ST has only TCP packets
to send, then TCP packets are scheduled.
If q > 0 when n = NU, the allocation is infeasible.
238 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
E. A Greedy Heuristic Scheduler for Voice
in Downlink
The difference of the downlink scheduling problem
from the uplink scheduling problem is that in the
downlink, a transport block can contain packets to
multiple STs. By combining the voice packets to
different STs to a single TB, we save considerable
PHY overhead. For transmitting a single voice packet
needs 4 slots, where 3 slots are for the PHY header.
Transmitting 2 voice packets need only 5 slots. So, it is
always advantageous to have transmissions in longer
blocks. This can be done by grouping together the STs
to those which are heard only by ith BTS, those heard
by ith and (i – 1)th BTS, but associated to the ith BTS
and those heard by ith and (i–1)th BTS, but associated
to the (i–1)th BTS, for all values of i.
In Fig 6 we show a simple deployment, with 3
BTSs. In each sector the taboo regions are also shown.
STs 3 and 4 are associated with BTS 1 and are not in
either of the taboo regions. So, any ST in the interference
set of 3 will also be in the interference set of 4. Any
transmission to ST 3 can equivalently be replaced by a
transmission to 4. Thus, they form a group for the
down link schedule. Similarly, STs 8 and 9 are associated
with BTS 2 and interfere with BTS 3. They are
associated to the same BTS and cause interference to
the same STs. So, ST 8 and 9 also form a group.
The STs are grouped together based on the above
criterion. The queue length of each group would be
the sum of queue lengths of the STs forming the
group. The greedy heuristic scheduler for the uplink
scheduling problem can then be used over these groups.
F. Round Robin Scheduling
A low complexity scheduler can be designed as
follows. The uplink and downlink parts of the frame
may further be divided into two contiguous parts.
Alternate sectors are served in these two parts. For
example, with 6 Sectors, Sectors 1, 3, 5 are served in
the first part, and Sectors 2, 4, 6 can be served in the
second part of the frame. Interference between
adjacent sectors can be eliminated in this way. Within
the round robin scheduler, the STs can be sceduled
based on queue lengths. With the number of sectors
equal to 2n0, the performance of this scheduler would
be equivalent to that of the scheduler discussed in
Section D, since we can have n0 transmissions going
on in each slot, with this scheduler. But, with n0 = 4,
this would require 8 sectors in the system. With the
number of sectors less than 2n0, the number of
simultaneous transmissions would be less than n0 with
1
2
5 1
2
3
4
4 3
5
Fig 7 Deployment of a system with 4 sectors and 5 STs
the round robin algorithm, whereas we can have up to
n0 transmissions with the greedy algorithm.
The round robin scheduler can achieve maximum
throughput when the distribution of villages and traffic
is uniform. But under admissible traffic it might lead to
instability and unfairness. This can be demonstrated
by a simple example. Consider the deployment of 5
STs in four sectors, as shown in Fig 7. STs 1 and 2 are
in the same sector. The scheduling constraints are that
n0 = 3 and Links 2 and 3 cannot transmit together. The
arrival rate vector is denoted by a vector a, where ai is
the arrival rate at the BTS for ST i ST i. An arrival
rate (1/2, 1/2, 1/2, 1/2, 1)(slots/slot time) is admissible,
but, not schedulable by a round robin scheduler.This is
clear from the examples in Tables 2 and 3. Table 2
shows the a schedule that schedules maximal
independent sets. Table 3 shows the way the round
robin scheduler schedules the STs, where STs 1 and 2
are scheduled only in alternate bursts, so that the two
STs have to share the slots, such that they are scheduled
only in half the slots. We see that the service rates
applied to ST 1 and ST 2 are ¼ and ¼ and to ST 5
is ½.
TABLE 2: Schedule for maximal independent set
scheduler
1 2 1 2 1 2 1
3 3 3 3
4 4 4
5 5 5 5 5 5 5
 VARGHESE & KUMAR : DESIGN OF A TDD MULTISECTOR TDM MAC 239

Let Rk be the vector of average rates allocated to

TABLE 3: Schedule for round robin scheduler
1 2 1
3 3 3 Rk+1 = a Rk + (1 – a )rk
4 4 4 4
5 5 5 independent set as follows
Another observation is that with the increase in
variability of the distribution of STs in sectors, the
round robin scheduler tends to become unfair.
G. Fair Scheduling for Data
To provide fairness to users, we keep track of the
average rates allocated to STs over time. The STs
with low average rate are given the chance to transmit
first. Maximal independent sets are formed starting
from the ST with the lowest average rate. Once the
slots for voice transmission are scheduled, we attempt
to include TCP transmissions in blocks of size Tmax,
so that the PHY overhead per slot is minimized.
STs until the kth slot and rk be the vector of rates
allocated to the STs in the kth slot.
1) Given a rate vector R, obtain a maximal
a) u1 = {i1}
i1 = argmin1£j£n Rj
I(u1) is the set of links interfering with the
links in u1. In this step, we select the ST with
the smallest average rate Rk for transmission.
b) Choose i2 Î ¸ arg min1£j£n,i2 Ï I¸(u1) Rj
u1 = {i1, i2}. In this step, we select one of the
non interfering STs with minimum average
rate for transmission.
c) Repeat the above until a maximal independent
set is obtained. Now, we have a set with STs
which have received low average rates in the
previous slots. So, once all STs transmit their
voice packets, we schedule these STs for
data packets.
2) Let l1 denote the number of nodes in u1 at the
end of step 1. Repeat the above for the remaining

Fig 8 Variation of total rate and fairness index with averaging interval for different
values of a . The upper set of plots are of the total rate, and bottom set are for
the fairness index
240 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

TABLE 5: Simulation results for uplink data rates

TABLE 4: Simulation results for downlink data
and packet drop with 80 STs in 6 sectors, averaged
rates with 80 STs in 6 sectors, averaged
over 30 random deployments. The data througputs
over 30 random deployments. The data
are given in kilo bits per second
througputs are given in kilo bits per
second n 0q Number of voice calls
per station
n 0q Number of voice calls
1 2 3
per station
o
3, 10 min u/l rate 17.1 8.1 0
1 2 3
max u/l rate 85 59 34
3, 10o min d/l rate 164 148 134
sum u/l rate 3570 2286 1229
max d/l rate 178 182 167
packet drop u/l 0 0.0029 0.0229
sum d/l rate 13749 12852 11690
3, 20o min u/l rate 13 5 0
3, 10o min d/l rate 163 151 136
max d/l rate 179 173 177 max u/l rate 88 57 31
sum d/l rate 13545 12798 11799 sum u/l rate 3510 2285 1110
packet drop u/l 0 0.0033 0.0312
3, 10o min d/l rate 167 153 137
max d/l rate 180 173 161 3, 30o min u/l rate 16 5 0
sum d/l rate 13883 13000 11750 max u/l rate 83 62 43
sum u/l rate 3463 2114 1176
3, 10o min d/l rate 224 204 190
max d/l rate 294 278 258 packet drop u/l 0 0.0042 0.0346
sum d/l rate 19807 18377 17007 4, 10o min u/l rate 38 18 0
3, 10o min d/l rate 204 194 177 max u/l rate 106 92 78
max d/l rate 283 255 274 sum u/l rate 5161 3776 2906
sum d/l rate 19312 17919 16430 packet drop u/l 0 0.0029 0.0283

3, 10o min d/l rate 172 165 140 4, 20o min u/l rate 25 9 0
max d/l rate 212 208 190 max u/l rate 157 168 160
sum d/l rate 15573 14078 12499 sum u/l rate 4833 3699 2771
packet drop u/l 0 0.0025 0.0304

4, 30o min u/l rate 15 7 0

max u/l rate 92 70 53
n–l 1 nodes. Now we have a maximal sum u/l rate 3468 2400 1359
independent set from the remaining Nu – l1 nodes. packet drop u/l 0 0.0029 0.0354
If any one of the l1 nodes can be activated along
with the maximal independent set formed from
the Nu–l1 nodes, add that till one get a maximal
independent set. This yields u1, u2 ... uk such that blocks of length equal to Tmax as long as it is
each node is included at least once. Each node is possible.
included at least once since a given number of 4) Update the rate vector as
slots is to be reserved for each ST in every
frame. Rk+1 = a Rk + (1 – a )rk
3) Now, we need to schedule u1 for t1, u2 for t2, etc.
To maximize throughput, we take tj = Tmax or We simulated the algorithm and obtained the
number of voice slots required. The vectors in fairness index of the rates allocated, and the total rate
the initial part of the schedule had low average achieved for various values of a . These are plotted
rate over frames. So, they get priority to send vs. the rate averaging interval in Fig 8. The averaging
data packets. So, starting from j=1, i.e., from the interval on the x axis is the number of frames over
first activation vector, if the sum of number of which the average throughput or fairness index is
slots allocated to STs in the frame is less than calculated. The fairness index is found to be close to
Nu, tj = Tmax. Else, tj = number of voice slots one unless the averaging interval is very small. This
required.Therefore transmission takes place in occurs partly because of the small number of STs
 VARGHESE & KUMAR : DESIGN
considered. A larger a in the rate averaging algorithm
yields a smaller average throughput.
6. VOICE AND DATA CAPACITY:
SIMULATION RESULTS
The scheduling algorithm discussed in Section 5D
was implemented in a MATLAB simulation. The PHY
rate is 11 Mbps. We consider a random distribution of
80 STs in 6 sectors. The spatial reuse n0 of 3 or 4 has
been considered, and the taboo regions in each sector,
on either side of the sector, are q = 10o, 20o, 30o.
Simulation is done with all STs having the same number
of ongoing voice calls: 1, 2 or 3. One VoIP call
requires one slot every alternate frame. A voice packet
that arrives in the system is scheduled within the next
two frames. If the scheduling constraints do not allow
the voice packet to be transmitted within two frame
times of arrival, the packet is dropped. In the simulation,
we have assumed synchronous arrival of voice packets,
i.e., if two voice calls are going on from an ST,
packets for both calls arrive synchronously, in the
same frame. The data traffic model is that all the STs
have packets to be transmitted throughout.
The results are shown in Table 4 and Table 5.
Here, min d/l rate is the average of the minimum rate
over STs in the downlink, averaged over 30 random
deployments; max d/l rate is the average of the
maximum rate over STs in the downlink, and sum d/l
rate is the average of the sum of downlink rates to the
STs. The same measures are also given for the uplink.
The packet drop u/l is the fraction of voice packets
dropped in the uplink, this being the bottleneck direction.
All the rates indicated are in terms of the MAC
payload. The PHY overhead has already been
accounted for in the calculations.
Each voice call requires a payload of 44 Bytes
every 20 ms, and hence 1.41 Mbps are utilised per
voice call, in the uplink and downlink, for 80 STs. With
a PHY rate of 11 Mbps, with n0 = 3 we have an
aggregate nominal rate of 22 Mbps in the downlink
and 11 Mbps in the uplink (assuming that 2/3 of the
frame time is allocated to the downlink). From the
table, it can be seen that with 80 STs in 6 sectors, and
1 voice call, with a taboo region of 10o on either side of
each sector, and n0 = 3, each ST gets an average
minimum data throughput of 164 kbps, and the average
total rate is 13.749 Mbps. Adding to this 1.41 Mbps,
we obtain about 15.16 Mbps, for a nominal downlink
bandwidth of 22 Mbps. The difference is because of
PHY overheads, and the inability to fill up all slots in a
frame. We notice that a second simultaneous call at
each ST reduces the data throughput by less than 1
OF A TDD MULTISECTOR TDM MAC 241
Mbps; this is because the packing can become more
efficient. For this same case, with one voice call, the
average minimum uplink data throughput is 17 kbps,
and the average total downlink data throughput is 3.57
Mbps. Adding to this 1.41 Mbps for voice, we obtain a
total uplink utilisation of 5.18 Mbps over a nominal
bandwidth of 11 Mbps allocated to the uplink. Because
of being smaller, the uplink frame is more inefficiently
packed.
If n0 = 3 and a taboo region of width q = 10o, the
fraction of voice packets dropped is 0.29% when we
support 2 calls per ST and 2.29% when we support 3
calls per ST. With 3 voice calls per station, we can see
that the packet drop is high, and the uplink capacities
to some STs are 0. With n0 = 3, the width of the taboo
region does not have an effect on the system capacity,
since we are always able to schedule in 3 sectors.
With n0 = 4, the system capacity reduces as q
increases. With q = 30., we can usually schedule
transmissions in just 3 sectors in a slot, even though
the SINR constraints allows 4 transmissions in a slot.
7. CONCLUSION
We consider the problem of finding the amount of
spatial reuse possible with a single channel multi sector
WIFiRE system with 802.11 MAC. It has been found
that there is an optimum value for the number of
simultaneous transmissions possible, so as to maximize
the total system capacity. The number of simultaneous
transmissions is found to depend on the path loss
factor and the radiation pattern of the antenna. For the
antenna pattern considered, and for path loss factor
2.3, as is applicable for rural environments, the number
of simultaneous transmissions possible is found to be
3. This can be improved by the user of antennas with
lesser back lobe radiation.
Also, for a given deployment, we find bounds to
the system capacity, assuming full channel reuse.
Based on this we can find the optimum positioning of
the antenna, such that the log utility function of rates
obtained by different STs is maximized. The bounds
obtained here are weak since we do not consider the
constraint imposed by the maximum number of
simultaneous transmissions.
A constrained dynamic programming problem is
found to give the optimum schedule for the system.
But, the problem is intractable due to the explosion of
state and action space. We employ a maximal weight
algorithm were the weights are the queue lengths of
the voice queue, such that each ST transmits in
contiguous slots, so as to minimize the PHY overhead.
242 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
For scheduling data in the system, we follow the
maximal weight algorithm, where the weights are
reciprocal of the average rate obtained by each ST in
previous slots. The average considered is an exponential
weighted average of rates. A simple round robin
scheduler have also been considered. Scheduling
examples are given for the greedy heuristic scheduler.
The different schedulers considered were
implemented in MATLAB and the data throughput in
each case is obtained as the average data throughput
over deployments, in terms of payload slots.
Deployments with different number of sectors and
STs, width of taboo region have been considered for
different voice loads, and the data throughput has
been obtained in each case.
REFERENCES
1. Wi-Fi Rural Extension (Wi-FiRE), Technical report,
www.cewit.org.in/docms/cewit07/review-
comments-draft.pdf.
2. Krishna Paul, Anitha Varghese, Sridhar Iyer,
Bhaskar Ramamurthi & Anurag Kumar, WiFiRe:
Rural area broadband access using the WiFi PHY
and a multisector TDD MAC, IEEE Communica-
tions Magazine, 2007.
Authors
Anurag Kumar received BTech
from IIT Kanpur and PhD from Cornell
University, both in electrical
engineering. He was with Bell
Laboratories, Holmdel, New Jersey, for
over six years. He is now a professor of
the Electrical Communications
Engineering Department at the Indian
Institute of Sciences, Bangalore. His area
of research is Communication
networking, and he has recently focused primarily on wireless
networking. He is a fellow of the IEEE, the Indian National
Science Academy (INSA), and of the Indian National Academy
of Engineering (INAE). He is an associate editor of IEEE
Transactions on Networking and IEEE Communications
Surveys and Tutorials. He is a co-author of the textbook
Communication Networking: An Analytical Approach by
3. Pravin Bhagwat, Bhaskar Raman & Dheeraj Sanghi,
Turning 802.11 Inside-Out, ACM SIGCOMM
Computer Communication Review, 34(1):33–38, Jan
2004.
4. Bhaskar Raman & Kameswari Chebrolu, Revisiting
MAC Design for an 802.11-based Mesh Network,
Third Workshop on Hot Topics in Networks, San
Diego, Nov 2004.
5. Harish V Shetiya, Efficient Routing and Scheduling
for IEEE 802.16 Mesh network, Master’s thesis,
Electrical Communication Enigineering, IISc, 2005.
6. Arash Behzad & Izhak Rubin, On the Performance of
Graph Based Scheduling Algorithms for Packet
Radio Networks, GLOBECOM, 2003.
7. Piyush Gupta & P R Kumar, The Capacity of
Wireless Networks, IEEE Transactions on
Information Theory, IT46(2), pp 388-404, March
2000.
8. Devavrat Shah, Paolo Giaccone & Balaji Prabhakar,
An efficient randomized algorithm for input-queued
switch scheduling, IEEE Micro, 22(1), pp 19-25,
January-February 2002.
9. Anurag Kumar, D Manjunath & Joy Kuri,
Communication networking, an analytical
approach, Norgan, Kaufman Networking Services,
Elsevier 2004.
10. Dimtri P Bertsekas, Dynamic Programming and
Optimal Control, vol 1, Athena Scientific, 2001.
Kumar, Manjunath, and Kuri (Morgan-Kaufman/Elsevier).
* * *
Anitha Varghese obtained BTech
degree from the College of Engineering,
Thiruvananthapuram, Kerala, in 2004
and ME from the Electrical
Communications Engineering
Department, Indian Institute of Science,
Bangalore. She is presently working in
the Vehicular Communications and
Information Management group in
General Motors , ISL, Bangalore.
* * *
 INVITED PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 243-248

Trends in VLSI Technology –

Rural Applications Perspective
K LAL KISHORE
Professor in ECE and Registrar, JNT University, Kakatpally, Hyderabad 500 085, India.
email: lalkishorek@yahoo.com

VLSI Technology has advanced rapidly since late 90s. Low cost, high performance
Chips designed and fabricated made strong impact in systems development and resulted in
applications to diversified fields like computers, communications, entertainment electronics,
medicine, and rural necessities. Spreading of Internet globally, making this world as a ‘global
village” and rapid strides in mobile communications have a direct bearing on the progress
made in VLSI technology. Research work is being done to further develop the technology in
new dimensions like Silicon Photonics, Organic semi-conductors, Flexible displays etc. These
developments take the technology further close to rural applications and can result in
significant improvement in the quality of life in villages. Fibre optic connectivity, hosting of
portals with useful information to rural folk, e-governance, energy conservation, wireless
sensor networks, Telephone-TV can improve the standard of living in rural areas significantly.
VLSI technology directly or indirectly is playing a vital role in this direction.
This paper reviews various developments taking place in the technology, research
work being done in this area, new devices and systems being developed and their
applications. The application of the technology and systems developed for rural areas is also
discussed. The projects implemented and possible developments to improve quality of life in
rural areas is discussed.
With the rapid strides being made in silicon device technology, efficient, low cost
devices with high performance are being manufactured. The chip design methods, making
use of the software tools are producing hardware for various applications like computers,
communications, defence, household appliances, commercial toys etc. These developments
are also making inroads into rural areas and changing the technological applications scenario
there. Hosting of websites / portals with information useful to rural areas is proving to be a
boon, for which the VLSI technology is contributing significantly.

1. TECHNOLOGY DEVELOPMENT
T HE International Technology Road Map for semi-
conductors, which is published periodically by the
semi-conductor Industry Association (SIA, San Jose
Califf) gave projections for 2003 Technology node to
90 nm and to 6 nm by 2019. Technology node refers
to the set of processes needed to print the smallest
feature. According to this road map, high performance
ICs will contain more than 8.8 billion transistors in an
area of 280 mm2 by 2016. This is more than 25 times
the packing density of chips built with 130 nm feature
size. Typical feature size, which is also referred to as
line widths, will shrink to 11 nm by 2014. Power
dissipation on high performance micro-processors will
Paper No 124-C; Copyright © 2007 by the IETE.
double its present value requiring more elaborate heat
sinks. Some of the extra power consumption will
come from gate-to-substrate and source-to-drain
current leakage that will grow larger as channel lengths
scale down to a few tens of nanometers. Approximately
9 billion transistors on a chip may sound like a dream,
but it is already being made a reality. The packing
density of cells in human brain is 107 cells / cm3.
As the channel length (the distance between the
Source and Drain) decreases, the transistor dimensions
also continue to shrink. A short channel means faster
transistor switching, because the charge carriers have
a shorter distance to travel. The voltage on the drain
begins to lower the energy barrier in the channel
reducing the threshold voltage and freeing the carriers
to flow even when there is no voltage on the gate. This
is what in essence called “Short Channel effect” [1].

243
244 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
To avoid short channel effect, transistor
performance is to be sacrificed to some extent, and
tolerate some increase in power consumption. Reducing
the thickness of the depletion region under the gate by
increasing the doping in the channel, maintains gate
control, but it also reduces carrier mobility. If the
thickness of Silicon dioxide gate insulation above the
channel is reduced to give gate, better control over the
channel, the thinner oxide lets more current leakage
between the gate and substrate, driving up power
consumption. This is tackled by having complex doping
profiles but the device engineers are running out of
steam.
The gate length of a MOSFET, which is an indicator
to gauge, the size of CMOS transistor, was 50 nm in
2002, and is projected to be 9 nm by 2015. To improve
the performance, silicon is to be mixed with a semi-
conductor like Germanium to produce strained
crystalline structure to allow electric charge carriers’
move faster. To reduce leakage current which pushes
up power consumption, gate oxide is to be made of
materials with more than eight times the dielectric
constant (k) of SiO2.
Some of the materials are: Nitrided Oxides, Al2O3,
CeO, SrTiO3, Ta2O5, ZrO2 MgAl2O4 .
For better control of transistors’s switching states,
the gate of the MOSFET can be of metal instead of
Polysilicon. If more than one gate is employed, power
consumption will be reduced and there will be better
control over ON-OFF action of the transistor.
In a pure silicon crystal, the distance between a
silicon atom and its nearest neighbours is the same in
all three directions. But in the strained silicon layer, the
atomic separations in the wafer plane are different
from those in the perpendicular direction. This change
in the crystal symmetry changes the energy band
structure in the conduction and valence bands. This
change is to reduce electron – hole collisions with
phonons (vibrations of atoms in crystal), the so called
scattering that slows the carriers down.
Silicon-on-Insulator (SOI) Wafers are being widely
used in semi-conductor manufacturing for the past
several years. SOI Wafers have a layer of Silicon
dioxide insulator buried under the device layer, to
reduce junction capacitance and so speedup transistor
switching. SOI gets rid of lot of junction area to
reduce junction capacitance. To reduce leakage and
power consumption, the present gate insulation of
SiO2 is to be replaced with a material having a larger
dielectric constant (k), ‘A materials’ dielectric constant
“k” value is a measure of the extent to which it
concentrates electric field lines. It is found that when
hafnium dioxide gate insulation is combined with a
strained – silicon substrate, the mobility is found to be
60% higher. A high dielectric material replacement to
SiO2 in the MOSFET structure reduces gate leakage
while strained silicon increases the performance of a
transistor by material innovation. If these two
advantages could be combined together without
inferring, it is advantageous.
The use of high – k dielectrics plus metal gates
plus strained silicon plus increasingly complex doping
profiles will extend life of the planar CMOS transistor
for atleast another decade. The researchers are looking
at double gate MOSFETS. In double-gate devices, the
gate is on both sides of the channel giving much tighter
control of the transistor’s ON and OFF states. Fin
FET structure is becoming popular and the advantage
is, channel is undoped. So the channel length shrinks.
Another advantage is Fin can be made extremely thin.
Power consumption is lower, because there is no
leakage path for charge carriers to flow along, between
Source and Drain, when the device is OFF. With
narrow line width, and small source and drain junctions,
the resistance in series with the transistor’s channel
will increase, increasing power consumption and
degrading the performance. Engineers have to solve
these problems in future. When the dimensions of
CMOS transistor can’t be scaled down further, new
devices like nanotubes, single electron transistors, super
conducting transistors and molecular transistors will
have to emerge [2].
2. MODERN FABS
The 32 bit / 64 bit Microprocessors are premier
markers of technological developments. A postage
stamp size chip will be having a billion transistors.
Hundreds of such chips are manufactured on a silicon
Wafer of a dinner plate size. In a single day, state-of-
the-art fab can make nearly 100 trillion transistors,
roughly 250 times the number of stars in the Milky
Way Galaxy. The device engineering is to be
appreciated for this technology marvel of the 20th
century. The fabs have a material handling system
that uses software – controlled robots and monorails
to transport Wafers, etching chambers, Wafer polishers,
photolithographic steppers etc. Manufacturing software
tools are used that track Wafers to the right tools, at
the right time and process control systems to manage
the chip making recipes the mix of gases, chemicals,
metals and semi-conductors that constitute chip. The
Automated Manufacturing Technology or AMT,
monitors, and controls the hundreds of steps, Wafers
must pass through, on their way to become Pentium,
Itanium and core 2 Duo processors etc. The Wafer
 K LAL KISHORE : TRENDS
used for manufacturing chips is a mirror – polished
disk of 99.9999999 percent pure silicon, 300 nm
diameter and less than 1 mm thick.
The AMT suite of programmes have been improved
tremendously over the last couple of decades. Software
called ‘THE GRID’ developed by Intel’s Logic
Technology Development Unit allows various machines
in the fab and number of applications that make up the
AMT suite to communicate with each other to fully
automate fab operations. The GRID is basically a
giant electronic bulletin board, where machines and
Wafers inside the fab announce their respective states
and locations. AMT programs are written in a variety
of programming languages including C, C++ etc and
the common language is called extensible Markup
Language or XML. Each program posts messages
encoded in XML on the grid for other programs to see
and, if necessary act upon.
Software has automated all routine tasks needed
to run a fab. Software being used in fabs has freed our
engineers to investigate ways to work with silicon
through the 22-nm generation of chips. In the near
future, Carbon nanotubes function as transistors and
light – emitting semi-conductors for optical outer
connects [3].
The Semi-conductor Industry became a global
industry in the 1990s. Chip manufacturing units and
assembly facilities were established in different parts
of the world. So International Technology Road map
for semi-conductors (ITRS) is created in the late 90s.
The Semi-conductor Industry Association (SIA)
extended invitation to Europe, Japan, Korea and Taiwan
to co-operate on ITRS. This was done at the world
semi-conductor Council in April 1998. Full revisions of
ITRS were produced in 1999, 2001, 2003 and 2005.
ITRS updates were produced in even-numbered years,
2000, 2002, 2004 and 2006.
The boundary between Silicon based and III - V
semi-conductors continues to move to higher
frequencies with time. Parameters such as noise figure,
output power, power-added efficiency; linearity and
ultimately cost will become more important. For CMOS,
long term prediction of device RF and noise
performance becomes more uncertain with the
introduction of metal gate electrodes, high permittivity
(high - k) gate dielectrics and new device structures
such as fully depleted or double gated Silicon - on -
Insulator (SOI).
For millimeter wave applications, InP- based RF
transistors have demonstrated very high frequencies
and GaN transistors have demonstrated record power
IN VLSI TECHNOLOGY 245
densities at 40 GHzs of 10W / mm with 40 V drain
bias.
Future wireless challenges include signal isolation
and software defined radio (SDR) SDR presents many
issues such as analog-to-digital (ADC) performance;
transmitter solutions and cost. Implementation of high
- k (dielectric constant) gate stack materials in low
standby power applications should be achievable. But
implementation of these materials in high performance
(HP) Logic and Low Operating applications is still
considered a difficult challenge. Introduction of 450
nm wafers in 2012 is still considered a difficult challenge
facing numerous issues [4].
3. SILICON PHOTONICS
With the emerging Nano Technology in chips, the
transit time of electrons between transistors in chips
will get reduced but, the speed between chips in a
system is not that fast. This bottleneck is being tackled
by Luxtera Inc. researchers. They have sculpted a
laser waveguide and modulator into a tightly coupled
silicon package that can be hooked up easily. They
claim that their chips handle 10 gigabits per second
and even higher speeds are in the offing. Then
broadband can become cheap and a feature film can
be downloaded in minutes [5].
4. PHOTONS INSIDE THE COMPUTER
The movement of data in a computer is in the
congested core of the microprocessor, the bits ‘fly’ at
a very high speed. But the copper that links one
processor to another and one circuit board to the next
slows down the movement. In a Pentium 4 processor
operating at 2-4 GHz, the data travels on a bus operating
only at 400 MHzs. Many researchers say that soon,
many of the copper connections in computers will
yield to high-speed optical inter connects, in which
photons rather than electrons will pass signals from
board to board or chip to chip or even from one part of
a chip to another. An electrical signal from the processor
would modulate a miniature laser beam which would
shine through the air or a wave guide to a photo
conductor, which would in turn pass the signal on to
the electronics. Though at the moment it is expensive
to communicate with light than with electric current, in
the near future only optical technologies will be able to
keep up with the elements of even more powerful
microprocessors. Compact optical I/O devices will be
used within the computer. At high frequencies, the
wire, series inductance becomes more important than
its resistance as an impeding factor, which also puts a
limit on the rate at which the trace can transmit pulses.
246 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
These parasitic factors depend heavily on the geometry
of the wire, its length and cross sectional area (c.s.a).
So a wires’ ultimate bit rate turns out to be proportional
to its cross section, but falls with the square of its
length. So a thinner and longer wire means - a lower
bit rate. Transition time limitations can be fought by
driving the line harder, but it is not a good solution. It
adds noise, increases power requirements and
aggravates already serious thermal management
problems. If the wires are made thicker, they occupy
more space. Photons don’t suffer from these limitations.
The exact point which economics of moving data
dictate optical inter connects is debatable. When
Microprocessors reach speeds of 10 GHzs, their inability
to communicate quickly with computer memory or
with other chips in a multiprocessor system will start
to stifle their usefulness [6].
5. ORGANIC SEMI-CONDUCTORS – evaporates pushing the conductive polymers onto the
FLEXIBLE DISPLAYS
Organic electronic materials can help in the
development of ICs with plastics. These use semi-
conducting and sometimes conducting materials that
are made of molecules containing carbon, mostly in
combination with hydrogen and oxygen. Such devices
are slower than silicon devices, but much cheaper.
The technology is developed to produce circuits with
hundreds of transistors printed on plastics, sensors,
memories and displays that can be bent like paper.
The present R&D work is aiming at cheap, flexible,
flat panel displays. The conventional LCD displays
are made of glass and are heavy. If the organic semi-
conductors technology is fully developed, it is capable
of creating arbitrarily large displays on light weight
flexible plastic substrates, the use of flat panel displays
could be revolutionized. Researchers are aiming at
displays printed onto rolls of plastic which could be
unfurled, processed, and cut up into devices of any
size. Such devices will have more adaptability for rural
applications and electronics and information technology
can penetrate faster into rural areas. Two types of
devices are under development:
1) Organic Field effect transistors (OFETs)
2) Organic Light Emitting Diodes (OLEDs)
Because polymers can be solution printed like ink,
researchers are looking to the printing industry for
technology. This research work requires a different
set of lab skills from silicon or III - V compounds. To
carryout the research works expertise in organic
chemical synthesis novel manufacturing techniques
and device physics is needed. Bell labs have produced
useful n-type organic semiconductors, complementary
circuits, and plastic backed active-matrix organic display
back planes. Plastic is a better substrate material
because it is light, bendable and rugged. Fortunately,
many organic semiconductors are like amorphous
silicon.
Micro contact printing, laser and inkjet printing
techniques are being used for organic semi-conducting
materials. Thermal printing technology of Dupont is
also being adapted to organic electronics manufacture.
First a large plastic sheet is coated with an organic
compound sensitive to light from a laser used in the
printing process. Then a layer of conductive polymer
that will form the circuit’s gate electrodes and inter
connects is made. The coated plastic sheet is then
processed to a second piece of plastic that will act as
the circuit substrate, and the two are run through an
apparatus that acts like an industrial laser printer.
Wherever the laser strikes, the sensitive compound
circuit substrate, where it strikes. OLE D displays
have light weight, superior image quality and consume
low power. Still quite a lot of fundamental research
remains to be done, including work on organic
semiconductors, having greater charge carrier mobilities
and operating lifetimes. The mobility of charge carriers
helps define a transistor switching speed. But it is far
lower in organic semiconductors than in the crystalline
silicon used in ICs. Researchers could overcome the
problems and limitations invocated with amorphous
silicon devices. The same is also expected to be true
of organic transistor [7].
6. RURAL PERSPECTIVE
Fibre optic connectivity to rural areas can bring
the impact of ‘e-revolution’ to villages. Websites can
be created keeping in mind the needs of rural people in
the areas of Agriculture, Health, e-governance, primary
education, Rural Energy etc.
The farming community must be kept informed
about the changes and new developments in the
agricultural sector. A website with a platform for
farmers, co-operatives, form a machinery vendors,
fertilizer and chemical companies’ agronomists and
farm advisors will be useful to the rural people. Thus a
knowledge society in agriculture can be created.
Medical facilities in rural areas are very poor.
Child mortality rate is high. A website which gives
information on different stages in the life of a child
beginning with pregnancy to adolescence, nutrition
food to be taken, vaccinations to be given, information
about public Health Centers, schemes related to rural
health and sanitation, will be useful in improving the
quality of life of rural people.
 K LAL KISHORE : TRENDS
A website on e-governance providing link to
Governments’ schemes and services for the people
can help villagers with easy access to the government.
A portal on providing information products on
children going to schools, girl child education, IT literacy
will also be very much useful. On Line discussions in
regional languages on current topics and education
will help the rural children in a big way.
Energy conservation is universally important. The
importance of this is to be explained to the rural people
in their own language to educate them, to bring
awareness, and to make them participate in energy
conservation. A portal to motivate villagers to take up
clean energy production through bio-gas, usage of
solar energy for home and village lighting, pump water
technologies can help rural people. This can help
families in a remote village find ways to improve
access and use energy for livelihood improvement in
their place of living [8].
6.1. Wireless Sensor Networks
Wireless networks are large number of small
sensing self-powered nodes which gather information
or detect special events and communicate in a wireless
fashion. The processed data is handed over to a base
station. Sensing of data, processing of data and
communicating the data or information, the combination
of all is done in a timing device. This gives scope for
many applications. Recent Advances in LOW POWER
VLSI, embedded systems, computing, communication
hardware and in general the convergence of computing
and communications are making this technology a
reality.
Applications
Sensor networks can be applied to:
1. Agriculture
2. Environmental Monitoring
3. Micro-Surgery, Medicine
4. Child education
5. Surveillance
6. Warfare.
VLSI Chips play an important role in the hardware
to be provided for the networks. The nodes send the
data to a base station. If habitat monitoring is to be
done in a forest area, as they are sensitive to human
presence, using sensor network provides a non-invasive
approach. Similarly data pertaining to air, temperature,
IN VLSI TECHNOLOGY 247
light, wind, relative humidity, and rainfall can be collected
by a network of weather sensors, embedded in
communication units for Environmental Monitoring. In
Princeton’s’ Zebra net Project, a dynamic (B9) sensor
network has been created by attaching special collars
equipped with a low power GPS system to the necks
of Zebras to monitor their moves and their behaviour.
More or less on similar lines, cattle health, and veterinary
medical operations can be carried out. Usage of water
for agricultural purposes, humidity level in the soil,
usage of pesticides, information about weather
conditions, agricultural product – all such applications
for rural areas can be implemented through such
networks, in which VLSI chips and embedded systems
play vital role. Wireless – Networked, sensor –
enhanced toys and other class room objects supervise
the learning process of children and allow un-obstructive
monitoring by the teacher [9].
In rural areas where medical facilities are not
available, sensors can be used to capture vital signs
from patients in real-time and relay data to hand held
computers carried by medical personnel and wearable
sensor nodes can store patients’ data. Such an
infrastructure designed to support wireless medical
sensors, PDAs, PCs and other devices that may be
used to monitor and treat patients in various medical
situations (B 13). Smart houses can also be established
wherein wireless sensor and activator networks
integrated within buildings could allow distributed
monitoring and control, improving living conditions and
reducing the energy consumption, by controlling
temperature, air flow etc. Fire accidents and loss due
to such mishaps can be minimized, where facilities are
minimum in rural areas.
Earth Sensing satellites are used to predict weather,
cyclones and hurricanes. Loss of lives, property and
other calamities can be minimized. The degree and
precision of weather prediction increases with VLSI
chip, of efficient and power computing capabilities.
6.2. Broad Band Over Power Lines
The idea is simple and the consumer gets access
to a new broadband vendor and has the house wired
for data with no additional effort. For rural applications,
this technology is particularly suitable and economical.
The cost of infrastructure required will be very much
reduced and high technology can be taken to rural
areas quickly. Police departments, amateur radio
operators are opposing this idea saying that this will
broadcast into their own bands. Google, Intel, Cisco
are some giants working in this area.
248 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
6.3. Telephone TV
Watching TV over cellphone is useful technology
for watching news, sports, information, and also light
entertainment to suit the requirements of village
population. Using new third generation (3 G) mobile
broad band networks, many cellphone operators are
streaming mobile TV services [10].
6.4. Wiring Villages – Chinese Model
The fibre optic connectivity for Internet is changing
the way the citizens of a coal-choked central China’s
Henam province live. Almost 40% of the population of
middle class standard go on-line. Even outlying mud-
wall villages have 8 megabit-per-second connection.
Liu Zhaingno a peasant launched website for his
Pigform and was selling a third of his production at
premium prices via the Net to buyers in neighbouring
provinces. Not only Liu, farmers in a neighbouring
village, the poorest in the region consult a newly
installed internet run by the agricultural ministry to
decide, what to plant and where to sell.
Since mid-2004, all interrogations in the state
inspection Bureau which investigates corrupt officials
are now carried out and recorded in a state-of-the-art
facility with six hidden video cameras manipulated
from a control room. A social science researcher of
China says he is convinced that the internet is helping
to make China a more open society. Tens of thousands
of Internet Cafes are used in which people who can’t
afford a computer and Internet subscription pay
affordable price per hour to go on-line. Corrupt
Government servants take illiterate villagers for a ride
in the implementation of welfare schemes of the
Government. With the Internet and mobile technology
such corrupt employees are scared about being
exposed.
Author
Lal Kishore obtained MTech, and
PhD, degrees from Indian Institute of
Science Bangalore. He has more than 30
years of teaching experience, and has
more than 60 publications to his credit.
He is guiding number of Research
Scholars in the area of VLSI Design,
Technology and Micro Electronics. He
had won the First Bapu Seetharam
Memorial Award for Research work from
Institution of Electronics and Telecommunication Engineers
(IETE), New Delhi in the year 1986. He had also received Best
Teacher Award from the Government of Andhra Pradesh in the
year 2004. He wrote 2 text books on Electronics Devices &
Thus the development in VLSI technology to make
cheaper, high performance, miniature devices and
processors is helping in taking new technology and
systems to rural areas to provide better quality of rural
living and ‘Knowledge Society’ in villages [10].
CONCLUSIONS
Rapid strides in VLSI technology have helped in
the development of low cost, high performance devices
and systems. This is resulting in applications to new
areas and realizing the concept of ‘Global Village’.
Internet Connectivity and Computer Communications
are bringing significant changes in the standard of life.
The same technology can also be taken and moulded
for rural applications. Wiring villages, wireless sensor
networks, websites on-e-governance, literacy, energy
conservation, medical facilities, and agriculture can
help the rural population and will result in improvement
in the quality of life, in rural areas also.
REFERENCES
1. IEEE /Spectrum, 2002.
2. IEEE Spectrum, October, 2002.
3. IEEE Spectrum, March, 2007
4. ITRS Updates, 2006.
5. IEEE Spectrum, January, 2006
6. IEEE Spectrum, August, 2002
7. IEEE Spectrum, 2002
8. C-DAC News letter, no 1, April, 2007
9. IEEE Circuits and Systems, vol 5, no 3, p. 19-27,
2005.
10. IEEE Spectrum, June, 2005.
Circuits and Electronic Circuit Analysis. Two more text books
on I C Applications and Electronic Measurements are in print.
He is a Fellow of IETE, Member IEEE, ISTE and ISHM. He has
conducted refresher course on VLSI at the UGC Academic Staff
College, Hyderabad 8 times. He has implemented number of
Research Laboratory Development Projects of DST, MHRD
and AICTE. He is closely associated with various activities of
IETE, Hyderabad Centre. He held number of administrative
posts in Jawaharlal Nehru Technological University, Hyderabad,
including Principal of University Engineering College, and
presently Registrar of the University.
* * *
 INVITED PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 249-256

Temporally Adaptive, Partially Unsupervised

Classifiers for Remote Sensing Images
SHILPA INAMDAR AND SUBHASIS CHAUDHURI
Department of Electrical Engineering, Indian Institute of
Technology Bombay, Powai, Mumbai 400 076, India.
email: {shilpa,sc}@ee.iitb.ac.in

Remote sensing is being increasingly used over the last few decades as a powerful tool
for monitoring, study and analysis of the surface of the earth as well as the atmosphere. In
this paper we shall consider temporally adaptive pattern recognition techniques for land-
cover classification in multitemporal and multispectral remote sensing images. The technique
comprises of pre-processing using global and classwise probability density function (PDF)
matching for temporally adapting the statistics before classification. We focus on the utility
of these techniques in generating improved partially unsupervised land-cover classifiers and
their comparative study.

1. INTRODUCTION histogram or probability density function (PDF)

S ATELLITE remote sensing technologies provide
large amount of data which is acquired from
various sensors at a periodical or regular basis. Images
that are acquired on the same geographical area at
different instants of time are termed as multi-temporal
images. Multiple sensors (fitted on the satellite) which
operate on different spectral bands are employed to
capture images. This results in multi-spectral images.
These images are a primary source of information
about the temporal behavior of land-cover in the area
of interest. Land-cover classification is associated
with change detection process too which helps in
studying the changes or trends in land-cover over a
period of time, say due to urban expansion, deforestation,
floods, disasters, etc. This kind of study is particularly
useful to environmentalists, civil engineers, urban and
transportation planners, etc to define policies and also
to reduce risks of natural disasters [1]. Since the data
is generated by the satellites periodically, there is a
need to develop methods which are temporally adaptive.
The work presented is an attempt towards the design
of classifiers with improved accuracies for
multitemporal remote sensing images.
The paper is organized into five sections. Section
2 describes an overview of the partially unsupervised
classification schemes with respect to temporally
adaptive analysis of remote sensing images. Section 3
highlights the need and then explains the concept of
Paper No 124-D; Copyright © 2007 by the IETE.
249
matching. Section 4 presents some experimental results.
Section 5 discusses the conclusions derived from this
work.
2. REVIEW OF PARTIALLY UNSUPER-
VISED CLASSIFICATION
There are two main approaches to any classification
problem; the supervised approach which requires the
availability of ground-truth to design a suitable training
set, and the unsupervised approach which does not
require any such information. The ground truth is in
the form of labeled pixels and the training set is
derived from this. However, collecting this data is
very expensive in terms of time and money. Also, the
process needs the services of a domain expert for the
region under consideration who can classify the pixels
from the captured image. Therefore, it may not be
possible to have the ground truth for images acquired
at every instant of time. Hence, supervised classification
cannot always be used for images at different time
instants due to logistic problems. On the other hand,
the limitation of unsupervised techniques is the low
accuracy of classification and nonavailability of labels.
The accuracy of classification may be a critical factor
since incorrect identification of classes or confusion
among them may lead to wrong interpretations and
possibly severely affecting in applications like natural
disaster management or environmental studies, etc.
The above mentioned reasons have led to the
development of partially unsupervised classification
methods, which emerge as a golden mean between
250 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
the supervised and unsupervised approaches [1].
With reference to multi-temporal analysis, partially
unsupervised classification schemes carry a lot of
importance. The term “partially unsupervised” is used
to indicate that although no ground truth is available
for the specific image to be classified, some training
data that exists for another image of the same
geographical area (but captured at some different
instant of time) and this is utilized for generating the
classifier. In this case, classifiers are first trained on
the existing training set, whose parameters are then
updated using an assumed distribution of the second
image [1]. The popular technique used is updating the
parameters of the previously trained classifier using
the distribution of the test image by Expectation-
Maximization (EM) algorithm [2]. It shows promising
improvement in classification accuracies after retraining
but faces the limitation that it suits mostly to image
data following Gaussian distributions. Another method
to improve classification accuracies in partially
unsupervised classifiers is the use of Markov Random
Field based approach which also considers the
contextual information [3]. In general, these methods
aim at adapting the parameters of an existing classifier
to the statistics of the image to be classified. The
concept of matching of the statistics is elaborated in
the subsequent section.
3. MATCHING OF STATISTICS
The main issue in case of multi-temporal images is
that the images (even of same geographical area)
acquired at different times are characterized by
differences in statistics owing to the change in
atmospheric or ground conditions or differences in
sensor calibration at the acquisition time [1]. These
issues become even more prominent in case of images
captured during different seasons. For this reason, the
classifiers developed for the image of the same
geographical area tend to exhibit unacceptable
classification accuracies if applied directly to the image
of interest. Hence is the need for matching of statistics
among the training and test images as an additional
pre-processing step.
A. Histogram matching technique
One way to adapt the statistics is the use of image
processing techniques for matching of statistical
properties of the multi-temporal images at a pre-
processing stage. The simplest technique that can be
employed to match the probability density function in
each spectral band of the two images is explained
here. This method uses histogram matching (or
histogram specification) which performs a matching
of shapes of the histograms by means of the use of
their cumulative histograms [4]. This method is applied
to multispectral images band-by-band.
Let X1 and X2 denote two multispectral images
with N channels and captured at time instants and t1
and t2 respectively. One of these images is considered
as the source and the other as the target in terms of
their histograms. Each image can be represented as a
collection of pixels each being represented by a vector
with N components corresponding to the intensity
levels in N spectral bands. Hence, for each image, we
can get N marginal distributions p1 (Xb) and p2 (Xb),
one for every spectral band (b = 1, 2, ... N). For a
single dimension or band, the problem can be
represented as finding a monotone mapping function
T(Xb) [4,5] such that:
T(Xb) = C2–1 [C1 (Xb)] (1)
where C1 and C2 are the cumulative PDFs of the
source and target images, respectively. This is the
standard band-by-band histogram matching (or PDF
matching) problem which can be solved using discrete
lookup tables. Note that the terms PDF matching,
histogram matching, PDF transfer are used
interchangeably in this paper. The above procedure
can be applied N times independently for N bands.
This results in bringing the individual distributions of
the bands as closer as possible to each other.
B. Classwise PDF matching
It is worth noting that till this point we are discussing
about the histograms in one band of the entire image.
But the entire image comprises of some “M” different
land-cover classes. Depending on soil and vegetation
condition, these classes exhibit different spectral
properties, i.e., they respond with different reflectance
for different part of the spectrum. It is precisely this
property that is exploited in distinguishing one class
from the other and hence in classification. Thus, the
problem of classification demands an analysis of both
the spectral and the statistical properties of various
classes. This issue can be handled more effectively if
we consider the histogram in each band as comprising
of “M” individual histograms, each corresponding to a
class, i.e., modeling the distribution in each band as a
mixture density. In practice, it is assumed in many
applications that every class follows a Gaussian
distribution. This assumption of model has been
observed to suit well to satellite image data. Moving
beyond this assumption, in the presented approach of
partially unsupervised classifier, there is no need of
 INAMDAR & CHAUDHURI : R EMOTE SENSING IMAGES
any assumption on the nature of the distributions at the
PDF matching stage, though assumed so in the later
stages of Maximum-Likelihood classification. The
histogram matching method in single dimension will
match the shape of any source histogram to any other
shape of the target. In multi-temporal images, the
difference in the shapes of the source and the target
histograms totally depends on the temporal changes
that occur in the spatial distribution, vegetation,
atmospheric conditions, air and soil moisture content.
One point that should be again noted here is that every
class may react differently to the temporal changes.
Hence, it may be more advantageous if we can derive
the PDF matching transfer function in eqn (1) differently
for each class rather than a global or single transfer
function for the entire image in a particular band.
In the framework of the presented classification
problem, the aforementioned discussion can be neatly
represented as two experiments, viz. the global
bandwise histogram matching (GB) and the classwise
bandwise PDF matching (CB) as preprocessing
techniques. The subsequent section will provide an
elaborate discussion on the related issues and
implementation of these two methods.
4. EXPERIMENTATION
This section describes the data set used, the various
experiments carried out, the measures used for assessing
the performance of the method presented and the
results obtained.
A. Description of the data set
Experiments were carried out on a dataset
comprising of three sets of multi-spectral images.
These images have been acquired by the Wide Field
Sensor (WiFS) mounted on Indian Remote Sensing
(IRS-1C) satellite. The images (550´900 pixels) cover
the geographical area that includes the town of Varanasi
in the state of Uttar Pradesh, India and the agricultural
area around it. The images have been captured in
November 1999 (T1), January 2000 (T2) and March
2000 (T3). These three time instants partly fall in
different periods of the year and hence somewhat
different seasons. The images are perfectly co-
registered and hence the same co-ordinates in the
three sets of images correspond to the same area on
earth’s surface. Each data set comprises of images
captured by sensors in two spectral bands, one visible
(red) (620-680 nm) and the other near-infrared (770-
860 nm). The spatial resolution of the sensor is 188 m
´ 188 m. Band 2 of the images at timeT1, T2 and T3 are
shown in Fig 1.
251
The ground truth has been made available to us
for all the three images. But for a given experiment of
partially unsupervised classification, we assume that
the training data are available only for one image, to be
used as training image. The test data set corresponding
to the other two dates is just used for validation of
classification results.
The accompanying ground truth information was
used to derive training and the test sets. The area
under consideration is characterized by five land-cover
classes viz wheat field, urban area, water body, rice
field and forest. Please note that the dark background
in Fig 1 is not a part of the image to be considered for
classification purposes. The corpus of training and test
data is as shown in Table 1.
TABLE 1: Training and test sets corresponding to
images acquired at T1, T2 and T3
Land-cover No. of samples
Class
Training Test Set
Set
1. Wheat Crop 979 1664
2. Urban area 159 281
3. Water body 40 72
4. Rice crop 221 652
5. Forest area 422 437
Total 1821 3106
B. Design of experiments
The experiments in classification were carried out
by training the classifier using the training set
corresponding to the time T1 and then tested on the
images acquired at time T2 and T3. The ability of the
classifier to handle bi-directional data was confirmed
by training the classifier using the training set
corresponding to the time T3 and then tested on the
images acquired at time T2 and T1. The next subsection
presents the experimental details of the PDF matching
phase, following which the design of the classifier,
without and with three variations of the pre-processing
technique, is presented in order to study the comparative
performances of classifiers. It must be noted here that
though there is no assumption on the distribution of
classes at the preprocessing or histogram matching
stage, the Gaussian distribution for classes is assumed
252 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
at the classification stage. This assumption has been
found to be reasonable. The final classification is done
using maximum likelihood (ML) rule in which every
class is modeled by Gaussian parameters.
C. Histogram matching
In the first set of experiments, the image at time
T1 was considered as the source (and the training
image) and the test image, either T2 or T3, was
considered as the target. Similarly, in the next set of
experiments the image at time T3 was chosen as
source and the test image, either T2 or T1, was
considered as the target. The aim was matching the
statistics of the source image to that of the target.
Initially bandwise histogram matching was performed
globally (GB method) using the histogram of the entire
image in a particular spectral band. Probability density
functions or the histograms were constructed from
band 1 and 2 of the source and targets and the
cumulative distribution functions were found out. Then,
for every pair of source and target image, two histogram
matching problems were solved corresponding to each
spectral band according to eqn (1). Let the modified
source image be represented by X1GB.
As discussed in the previous section, the bandwise
histogram matching was also performed locally or in
other words on each class separately, i.e. classwise
bandwise matching (CB method). The critical issue
here was defining source and target clusters or classes.
In case of the source image, which is also the training
image, forming clusters is trivial. The pixels labeled as
belonging to the same class were grouped together to
form a cluster. But the target clusters were needed to
be formed from the test images which have no
information available for them. This problem was solved
as follows:
1) Choose source (training) and target (test)
images.
2) Form the source clusters by grouping together
the pixels labeled as belonging to the same
class in the training set.
3) Train the ML classifier using training data
from X1 .
4) Perform classification on the test set from
target image.
5) Group all the pixels from the test set classified
into same class and form “M” target clusters.
Now, we have “M” source and “M” target classes
with the correspondence known between them. The
problem of bandwise histogram matching now becomes
that of “M” independent problems. Bandwise histogram
matching is performed individually on each pair of
source and target classes. Let the modified source
image in this case be represented as X1CB-ML. It is
expected that the CB method should bring the
distribution of each class closer to each other than the
GB method. However, the performance of ML
classifier is expected to be poor without retraining and
the method is not expected to yield accurate results. In
the next variation for CB method, instead of forming
the target classes using the direct ML classification
result, they were formed using the results of ML
classification after GB method of pre-processing. The
steps are as follows:
1) Choose source (training) and target (test)
images.
2) Form the source clusters by grouping together
the pixels labeled as belonging to the same
class in the training set.
3) Perform global bandwise histogram matching
and get modified source X1GB .
4) Train the ML classifier using training data on
X1GB.
5) Perform classification on the test set from
target image.
6) Group all the pixels from the test set classified
into same class and form “M” target clusters.
As before, we have “M” source and “M” target
classes with the correspondence known between them.
Bandwise histogram matching is performed individually
on each pair of source and target classes similar to the
previous case. Let the modified source image in this
case be represented as X1CB-GB.
In order to evaluate the effectiveness of this
bandwise matching technique in bringing close the
distributions of source and target clusters, we calculated
the Bhattacharya distances between the source and
the target clusters before and after the modification
using the presented technique. The Bhattacharya
distance between the distributions of any class w i from
the source and the target (say, p1 (X | w i) and p2( X | w i
)) is calculated as follows:
Bi = – ln òx p1 (X | w i) p2 (X | w i) dX (2)
As already mentioned, it is reasonable to assume
a Gaussian distribution for the classes at the
classification stage. Hence, the distance calculation
can be simplified as
1 æ Si,1 + Si,2 ö –1
Bi = ¾ (µi,1 – µi,2)t ç ¾ ¾ ¾ ¾ ÷ (µi,1 – µi,2) +
8 è 2 ø
 INAMDAR & CHAUDHURI : REMOTE SENSING IMAGES 253

1 éê Si,1 + Si,2 ê ù D. Classification results

¾ log êê ¾ ¾ ¾ ¾ ¾ ê êSi,1 êêSi,2 ê ú (3)
2 ëê 2 ê û
where µi,1 , µi,2 and Si,1, Si,2 are the mean vectors
and the covariance matrices for the distributions of w i
at t1 and t2 respectively. The corresponding results
obtained are shown in Table 2 and Table 3 respectively.
From both the tables, we can observe that the estimated
classes at other time instants have fairly large
Bhattacharya distances with the corresponding
supervisory data over which the classes were trained.
This is due to the temporal change in the statistics of
individual classes. If one performs classwise histogram
matching, the corresponding Bhattacharya distances
reduce drastically, making the statistics of various
land-cover classes invariant to temporal changes. When
one performs a single global histogram matching, the
reduction in the distance measure is not that significant.
This indicates that each land-cover class undergoes a
different type of change in statistics and this cannot be
captured by defining a single PDF transfer for all the
classes.
For every combination of training and test set, in
all four types of classifiers can be trained. They differ
in the image used for training the classifier. Though
there is no assumption of any distribution model in the
pre-processing stage, for classification we assume a
parametric (Gaussian) model for the distribution of
land-cover classes. The most basic type of classifier is
the direct maximum-likelihood (ML) classifier without
any kind of pre-processing. It is designed using the
training set from X1 and applied to the test images.
The second type of classifier was trained on
X1CB-ML. The third type of classifier was trained on
X1GB and the fourth one was trained on X1CB-GB. It is
observed that the performance of the second and the
fourth classifier improves over the first and third
classifier, respectively, in all cases. Also, in general
the CB-GB classifier performs better than the CB-
ML classifier. This can be attributed to the fact that
the target classes used in CB-GB method are more
accurate than those used in CB-ML method, as they
are derived from the result of GB classifier which is

(a) November 1999 (T1) (b) January 2000 (T2) (c) March 2000 (T3)

Fig 1 Band 2 of the multitemporal image data set used for experiments

TABLE 2: Bhattacharya distances for various classifiers trained on November 1999 dataset (T1)

Test Data Classifier Land-cover class

Wheat crop Urban area Water Body Rice crop Forest

T2 Direct ML 0.1392 0.2235 1.2538 1.3226 0.2433

CB-ML 0.0325 0.1010 0.1552 0.3211 0.0647
GB 0.1982 0.1435 0.4973 0.4009 0.2250
CB-GB 0.0257 0.1181 0.2804 0.0595 0.0605

T3 Direct ML 0.2808 0.2735 1.1237 1.5053 0.3136

CB-ML 0.0422 0.0691 0.4997 0.9906 0.0256
GB 0.1536 0.2056 0.5463 0.2206 0.2004
CB-GB 0.0138 0.0897 0.1753 0.0751 0.0332
254 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

TABLE 3: Bhattacharya distances for various classifiers trained on March 2000 dataset (T3)

Test Data Classifier Land-cover class

Wheat crop Urban area Water Body Rice crop Forest

T2 Direct ML 0.3165 0.2615 0.4681 0.8929 0.6526

CB-ML 0.0324 0.0771 0.1545 0.2847 0.1294
GB 0.1740 0.1358 0.3062 0.4899 0.4245
CB-GB 0.0252 0.1567 0.0921 0.1893 0.0253

T1 Direct ML 0.3018 0.6214 1.3092 0.7876 0.9470

CB-ML 0.0692 0.3481 0.1312 0.1181 0.3053
GB 0.1091 0.1434 0.4391 0.4077 0.3507
CB-GB 0.0129 0.0343 0.0367 0.1705 0.030

TABLE 4: Classification accuracies for classifiers trained on November 1999 dataset (T1)

Test Type of Classification Accuracies (%)

Data Classifier
Average Wheat crop Urban area Water Body Rice crop Forest

T2 Direct ML 83.35 91.22 78.29 93.05 69.93 75.05

CB-ML 89.79 97.11 56.22 90.27 85.73 89.47
GB 92.27 98.19 72.9 93.05 92.63 81.46
CB-GB 93.04 98.19 66.9 88.80 92.79 91.30

T3 Direct ML 76.62 97.65 75.08 97.22 11.65 91.07

CB-ML 79.23 97.89 72.59 95.83 20.39 97.48
GB 88.86 97.35 87.90 95.83 68.55 86.27
CB-GB 93.30 96.75 88.96 95.83 84.96 94.96

TABLE 5: Classification accuracies for classifiers trained on March 2000 dataset (T3)

Test Type of Classification Accuracies (%)

Data Classifier
Average Wheat crop Urban area Water Body Rice crop Forest

T2 Direct ML 82.58 89.60 81.85 88.88 86.34 49.65

CB-ML 89.21 94.89 83.27 88.88 87.73 73.68
GB 90.95 97.47 63.34 91.66 92.33 81.69
CB-GB 92.85 97.83 59.43 91.66 98.31 87.41

T1 Direct ML 76.46 99.87 70.81 66.66 38.19 35.46

CB-ML 88.98 99.69 72.59 86.11 88.65 59.72
GB 92.69 97.71 85.40 87.50 90.03 83.06
CB-GB 95.04 97.83 88.25 87.50 94.01 91.53
 INAMDAR & CHAUDHURI : REMOTE SENSING IMAGES
more accurate than the ML classifier used in CB-ML
method. Overall results indicate that the modified training
images which are closer in statistics to the target
images to be classified result in improved classifiers
which are “adapted” to the test images from the
temporal series. In other words, the methods used are
“temporally adaptive”.
The results of classification are shown in Table 4
and Table 5. The average accuracy of classification
and the classwise accuracies are presented. It is
worth noting that the classwise transfers improve the
classification accuracy of each class significantly and
hence also the average accuracy of classification.
Further, the use of this procedure on multitemporal
images yields results of classification which are at an
acceptable level of accuracy and hence can eliminate
the need for retraining using a procedure like the EM
algorithm which can be computationally exhaustive.
For sake of verification, the January 2000 dataset was
also used as training set and the above mentioned four
classifiers were tested on November 1999 and March
2000 datasets. Results obtained were similar with
classifiers designed with classwise transfers
outperforming those designed with global transfer.
5. CONCLUSIONS
In this work, the suitability of global and classwise
PDF matching techniques has been described with
their application to partially unsupervised classification.
In multitemporal framework, these techniques have
been used to generate modified training images which
are closer in statistics to the images to be classified. In
general, for every test image to be classified, a modified
training image can be generated which matches the
statistics of the test image. Thus, the training of the
classifier is adaptive to the test image. Experimental
results confirm the effectiveness of the methods. The
255
classwise PDF matching more closely matches the
statistics of the classes in the source and target images
than the global PDF matching. Hence it also turns out
to be more effective for generating modified training
images which help generate more accurate classifiers.
The advantages of this method include elimination of
the need of retraining, and the low computational
complexity. The main drawback of this method is that
it considers each spectral band separately, neglecting
the correlation among them. Our current work focuses
on it.
ACKNOWLEDGEMENTS
The authors are thankful to ISRO, India and Dr B K
Mohan, IIT Bombay for making available the image data
set with ground truth used for this experiment. Funding
support from ISRO Cell is gratefully acknowledged.
REFERENCES
1. L R Bruzzone & D F Prieto, Unsupervised retraining
of a maximum likelihood classifier for the analysis of
multitemporal remote-sensing images, IEEE Trans.
Geosci. Remote Sensing, vol 39, no 2, pp 456-460,
Feb 2001.
2. T K Moon, The Expectation-Maximization
Algorithm, IEEE Signal Processing Magazine, Nov
1996.
3. R Cossu, S Chaudhuri & L Bruzzone, A context-
sensitive Bayesian technique for the partially
supervised classification of multitemporal images”,
IEEE Geosci. Remote Sensing Letters,vol 2, issue 3,
pp 352-356, July 2005
4. R Gonzalez & R Woods, Digital image processing,
2nd edition, Prentice Hall, 2002.
5. F Pitie, A Kokaram & R Dahyot, N-Dimensional
probability density function transfer and its
application to color transfer, Proc IEEE ICCV,
Beijing 2006.
256 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
Authors
Subhasis Chaudhuri was born
in Bahutali, India. He received his BTech
degree in Electronics and Electrical
Communication Engineering from the
Indian Institute of Technology,
Kharagpur in 1985. He received MS and
PhD degrees, both in Electrical
Engineering, respectively, from the
University of Calgary, Canada and the
University of California, San Diego. He
joined the IIT, Bombay in 1990 as an assistant professor and is
currently serving as the professor and head of the Dept.
He has also served as a visiting professor at the University
of Erlangen-Nuremberg, Germany and the University of Paris
XI. He is a fellow of the Alexander von Humboldt Foundation,
Germany, the Indian National Academy of Engineering, and the
Indian Academy of Sciences. He is the recipient of Dr Vikram
Sarabhai Research Award for the year 2001 , and the Prof SVC
Aiya Memorial Award and the Swarnajayanti Fellowship both
in 2003. He received the S S Bhatnagar Prize in engineering
sciences for the year 2004.
He is the co-author of the books ‘Depth from defocus: a
real aperture imaging approach’, and ‘motion-free super-
resolution’ both published by Springer, NY. He has also edited
a book on ‘Super-resolution imaging’ published by Kluwer
Academic in 2001. His research interests include image processing,
computer vision and multimedia.
* * *
Shilpa Inamdar received her
Bachelor of Engineering degree in
Electronics and Telecommunications
from University of Pune in 2003 and
the Master of Technology degree in
Electrical Engineering from I IT Bombay
in 2007. She has served as a Research
Assistant for the Indo-Italian project on
“Advanced Techniques for Remote
Sensing Image Processing” from July
2004 - June 2007. Her areas of interest include pattern recognition
and remote sensing image processing.
* * *
 INVITED PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 257-269

Texture Feature Matching Methods for

Content based Image Retrieval
IVY MAJUMDAR, B N CHATTERJI
B P Poddar Institute of Management and Technology, 137 VIP Road, Kolkata 700 052, India.
AND

AVIJIT KAR
Jadavpur University, Kolkata 700 032, India.
email: ivy_majumdar@yahoo.co.in; bnchatterji@gmail.com; dr.avijit.kar@gmail.com

Texture features are widely used for matching in content based image retrieval. Since
mid nineteen nineties we find a lot of R & D activities in this area. In this paper an attempt has
been made to review these works. Seventy one papers were reviewed and these were
classified into spatial domain, wavelet based and miscellaneous methods. These methods
were further sub-classified on the basis of mathematical techniques used or the algorithm
applied. A very brief description of these methods is given. Finally the paper discusses
about the need for comparison of the methods and on the future directions in this area.

1. INTRODUCTION Texture is one of the important features for

F OR more than one decade we find a lot of research
efforts for searching image database with
applications in areas like digital libraries, advertisement,
entertainment, medical diagnosis, communication and
several other areas. This is due to the tremendous use
of Internet and the development of world wide web
(www). Content based image retrieval (CBIR) method
is the most popular method used for searching image
databases. Kokare et al [1] have given a detailed
literature survey on CBIR. The block diagram of a
CBIR system is given in Fig 1. It has a stage known as
Database Generation where the known input images
are digitized and features are extracted from the digitized
images. In the database we store the features of the
image. The feature can be colour, texture, shape,
image semantics or other features in other domains.
These features are such that these can describe the
contents of the image. In the retrieval stage the query
image is digitized and the features are extracted. Then
we have a matching and indexing step where we
determine the similarity between the features of the
query image with the features stored in the image
database. The retrieved image is the one having
maximum similarity with the query image. Retrieval
results are also given by ranking a few images on the
basis of the similarity index.
Paper No 124-B; Copyright © 2007 by the IETE.
257
matching in CBIR. Texture features can be the only
feature or it can be used along with the other features
for matching. According to Sklansky [2] “A region in
an image has a constant texture if a set of local
statistics or other local properties of the picture are
constant, slowly varying or approximately periodic”.
For the last 15 years we find that there was a tremendous
interest in the use of texture features for matching in
CBIR. In this paper we have tried to provide a survey
review of some of these methods. We can classify
these methods broadly into three categories.
(1) Spatial Domain based Methods in which the
processing is done in the spatial domain
(2) Wavelet based Methods in which the image is
transformed into the wavelet domain first and
then the processing is done.
(3) Miscellaneous Methods where we use some
special techniques in which processing can be
in spatial or wavelet domain or in both.
The various methods are given in the classification
tree given in Fig 2. A brief description of these methods
will be given in the subsequent sections.
2. SPATIAL DOMAIN BASED METHODS
A large number of texture image matching methods
use the image in spatial domain. These methods can
be further sub classified into the following depending
on the texture modeling and the method of analysis.
258 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
Images for
Digitizer Feature
Database
Extraction
Query Image Digitizer Feature
Extraction
Fig 1 Block diagram of content based image retrieval system
2.1. Histogram based Method
Here we use the image histograms. Qin and Gao
[3], Quin et al [4] gave a matching method for texture
images with the help of a descriptor based on a two
dimensional histogram ARPIH (Angular Radial
Partitioning Intensity Histogram ). This is determined
from local invariant regions of the image and has the
property of distinctiveness and robustness to
deformations. Grauman and Darell [5] described an
efficient and accurately matching method using an
approximation to Earth Mover’s Distance (EMD)
which does not require clustering descriptors.
2.2. Co-occurrence Matrix based method
Co-occurrence matrix is in use for texture image
analysis for more than three decades. CBIR using co-
occurrence also received a good attention. We can
further sub classify these methods into the following
two categories.
2.2.1. Using Gray Level
Aksoy and Haralick [6-9] used the variance of
gray levels spatial dependencies as texture features
and determined the gray level co-occurrence matrices
as five distances and four orientations to measure
textures. The matching was done using likelihood ratio
classifier and nearest neighborhood (NN) classifier.
Lazebnik et al [10] gave a method in which a generative
model of the local descriptors is first determined using
EM algorithm. Then they determined the gray level
co-occurrence matrices of the neighboring descriptors.
For matching a relaxation algorithm was used using
the above models. Qin and Yang [11] used Gray
Image Database
Database Generation
Retrieval stage
Image Retrieved
Matching and
image
Indexing
Level Aura Matrices (GLAM) to model texture images
and matched the images using the similarity measure
Support Vector Machine (SVM) to achieve excellent
retrieval accuracy. Guo et al [12] used a metric for
texture image retrieval which is the signed distances
of the image from the boundary obtained using the
SVM learning algorithm to obtain better accuracy
compared to that obtained using Euclidian distance.
The retrieval was insensitive to sample distribution
and same result was obtained using different but visually
similar queries.
2.2.2. Using Ordinal Measure
Partio et al [13-15] used a combination of co-
occurrence matrices and ordinal measures which they
called Ordinal Co-occurrence Matrix. The matching /
retrieval was achieved using city block distance
functions giving better performance compared to that
using gray level co-occurrence matrices.
2.3. Autoregressive and Perceptual
Models based Methods
Abbadini [16, 17] used autoregressive model and
perceptual model for texture image content
representation. The similarity model was the
determination of Gower’s similarity coefficient. He
obtained interesting results.
2.4. Modal Analysis based Method
Carcassoni et al [18, 19] determined the modal
structure of the positions of the peaks of the power
spectrum and performed the matching by (i) comparing
 IVY MAJUMDAR et al : TEXTURE FEATURE MATCHING METHODS 259

Texture Image Matching Methods

Spatial Domain Wavelet Based Miscellaneous Methods

Based Based on

Daubechies Semantics
Wavelet
Fuzzy Set
Similarity Matching theoratic
Using Distance
Three Step
GGD K-L Distance Hierarchic

Steerable Pyramid Texture and

colour
Gabor Wavelet
Meta Search
Spline Wavelet Engines

Polar Wavelet Hypo Graph

Complex Wavelet Neuro Science

M Band Wavelet Modified Zer-

nike Moments
Cosine Modulated wavelet

Histogram Co-occurrence Modal Active appearance Markov Random Other

Matrix Analysis Model Field Based Methods
Learning
Auto regressive
Model Neural
Net

Gray level Ordinal Using Four Structural Survey of

Wold Model
Measure Classes of Information CBIR
Texture Feature Descriptor Texture Features
For Visual
Perception

Fig 2 Classification tree of texture image matching methods for CBIR

260 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
the modal structure (ii) latent semantic indexing to
obtain good results with database fabric and wrapping
paper images.
2.5. Active Appearance Model based
Method
Cootes et al [20] described a statistical appearance
model which is obtained by combining the models of
shape and texture variations. They obtained an iterative
matching algorithm in which perturbations of the model
parameters are related with minimization of texture
error to provide improved texture match.
2.6. Neural Network based Method
Long et al [21] described a neural network based
method which optimizes the invariant / perceptual
mappings of texture images. Gabor features of texture
image are mapped to an invariant space in the first
three layers. Texture is mapped into the perceptual
space in the last three layers. They obtained improved
retrieval performance.
2.7. Markov Random Field based Method
Gimel’farb and Jain [22] used learning scheme for
a Markov random field image model with Gibbs
probability distribution. Their distance measure is based
on the gray level histogram collected in accordance
with the structure of multiple pairwise pixel interactions
in the sub-images to be matched. The method can
tolerate texture rotation and scale to a certain extent.
2.8. Other Spatial Domain Methods
In addition to the above seven types we find a
few more spatial domain models which does not fall
into any of the above. Ohanian and Dubes [23] used
four classes of texture features namely random field
modeling, fractal geometry, co-occurrence matrices
and Gabor filtering separately and found that the co-
occurrence matrices perform the best amongst these.
Liu and Picard [24] used a new Wold model for
texture images and demonstrated that this model
perform better compared to (i) shift invariant principal
component analysis and (ii) multiresolution
simultaneous autoregressive model. Carkacioglu and
Vural [25] gave a statistical analysis of structural
information (SASI) descriptor for texture image
retrieval and demonstrated its effectiveness. Xu and
Liao [26] used clustering technique for texture image
retrieval and used this method for medical images.
Tamura et al [27] proposed several features of
texture images namely coarseness, directionality,
contrast etc which are used in the IBM QBIC System
of CBIR.
3. WAVELET BASED METHODS
A large number of works can be seen in image
matching for CBIR with texture features using
wavelets. Efforts were made to use different wavelets
and the effectiveness of these for the retrieval
performance was found out. A brief description of
these methods will be discussed in the following
subsections.
3.1 Daubechies Wavelet based Method
Wang et al [28] used Daubechies wavelet and
determined the wavelet coefficients as features and
used a two level multiresolution matching for retrieval
which they called image indexing and searching to
obtain good results.
3.2. Method based on Similarity Matching
using Distance
Laine and Fan [29] used the conventional discrete
wavelet transform to determine the features and used
distances as similarity measure for matching. Chang
and Kuo [30] used tree structured wavelet transform
and distance functions for texture analysis. Unser [31]
used wavelet frame for texture classification and
segmentation. June and Scharcanski [32] used discrete
dyadic wavelet transform (DDWT) to represent the
images in multiple resolutions to determine the
orientation and gray level distributions. They determined
a multi-resolution distance measure to match similarity.
Fauzi and Lewis [33] used sub-images at various
resolutions of wavelet transform and determined the
texture features in each sub-image. In their matching
algorithm they compared the distance between the
features of the query image with the features in each
sub-image thereby obtaining good accuracy. Kokare
et al [34] used Daubechies orthogonal wavelet 8-tap
filter coefficients as feature and gave an excellent
comparison for retrieval performance using Euclidian,
Manhattan, Chebychev, Mahalanobis, Square Chord,
Squared Chi-Squared, Canberra, Bray-Curtis and
weighted mean variance distance functions. Shang et
al [35] also used DDWT and a novel technique to
describe the textures with their orientation and gray
level distribution using global spatial relationship of
colour to achieve better performance in comparison to
that of [32].
 IVY MAJUMDAR et al : TEXTURE FEATURE MATCHING METHODS
3.3. Generalized Gaussian Distribution and
Kullback-Leibler Distance based
Method
Tzagkarakis and Tsakalides [36] used statistical
feature and similarity measurements for CBIR. They
used generalized Gaussian distribution (GGD) for
modeling wavelet coefficient and matched the textures
using Kullback-Leibler (K-L) distance between alpha-
stable distributions. Do and Vetterli [37, 38] also used
the GGD for modeling marginal distribution of wavelet
coefficients and a closed form K-L distance between
GGD’s for matching thereby achieving significant
improvement in retrieval rates.
3.4. Steerable Pyramid based Method
Patrice and Konik [39] derived the texture features
on a multiresolution paradigm called steerable pyramid
in wavelet based decomposition. They used several
similarity criteria like weighted L2 norm between two
feature sets and symmetrized version of K-L divergence
for matching. Finally they introduced a relevance
feedback for improvement in the retrieval performance.
3.5. Gabor Wavelet based Method
Wu et al [40,41] and Manjunath and Ma [42] used
Gaussian function modulated by complex sinusoid called
Gabor functions as features and using normalized
Euclidean distance obtained approximately 74%
retrieval accuracy. Ma and Manjunath [43] used the
Gabor features in a hybrid neural network for learning
the similarity as feature space clustering thereby
improving the retrieval performance.
3.6. Spline Wavelet based Method
Qiao et al [44] used spline wavelets to determine
features like mean and standard deviation of the sub-
band coefficients. Their database consisted of 1,792
images. It was found that linear Battle-Lemarié spline
wavelet, cubic bi-directional orthogonal spline wavelets
and quadratic spline dyadic wavelet give better
performance compared to cosine modulated and Gabor
wavelets.
3.7. Polar Wavelet based Method
Pun [45] used polar wavelet for retrieval of texture
images. The feature extraction is the polar transform
followed by an adaptive row shift invariant wavelet
packet transform. During similarity matching k-nearest
neighbor search is used.
261
3.8. Complex Wavelet based Method
Rivaz and Kingsbury [46] used complex wavelet
transform for texture image retrieval. They used the
complex wavelet transform to obtain texture features
and using a similarity matrix they obtained accuracy
approximately same as that obtained using Gabor
wavelets. Hatipoglu et al [47] used dual tree complex
wavelet transform (DT-CWT) for texture image
features and obtained retrieval performance better
than that obtained using real discrete wavelet transform
(DWT). Kokare et al [48-50] used (i) rotated complex
wavelet filter(RCWF), (ii) dual tree complex wavelet
transform (DT-CWT) and (iii) RCWF and DT-CWT
as three different feature sets and obtained improved
retrieval performance compared to that obtained using
real DWT and using Gabor wavelet. They could use
the method for both rotated and non-rotated database.
The method is effective for small size rotated, medium
size rotated and large size rotated image databases.
3.9. M-Band Wavelet based Method
Acharya and Kundu [51] used M-Band wavelet
transform to give an adaptive and unsupervised
segmentation method for texture images. Kokare et al
[52] used the M-Channel wavelets for texture image
retrieval. They could achieve good retrieval accuracy
with less computational effort than that using Gabor
wavelets.
3.10 Cosine Modulated Wavelet based
Method
Koilpillai and Vaidyanathan [53] gave the concept
of cosine-modulated FIR filter banks which is a special
class of unitary filter banks. Kokare et al [54,55]
used this as (i) cosine modulated wavelet based and
(ii) cosine modulated wavelet packet based texture
features for CBIR. They obtained better performance
both in terms of accuracy and retrieval time as
compared to that obtained using Gabor wavelets.
4. MISCELLANEOUS METHODS
In this section we will review some of the methods
which do not fall in the category mentioned in the
previous sections. These methods either use a new
concept like fuzzy sets, neuroscience, semantics etc
or use a combination of concepts described in the
previous section. These methods are given in the
following subsections.
262 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
4.1. Semantic based Method
Li et al [56] and Wang et al [57] gave a semantic
sensitive integrated matching method which they called
SIMPLIcity. They used integrated region matching
(IRM) approach in which the features of one region of
the first image is compared with the features of several
regions of the second image for matching and a
similarity measure is then determined. Their method is
robust for variation in intensity and sharpness, rotation,
colour distortion, cropping and shifting.
4.2. Fuzzy Set Theoretic Method
Fuzzy set theoretic concept is very well established
now and has been used in many engineering applications.
Santini and Jain [58] gave a new model called Fuzzy
Feature Contrast (FFC) by extending the Tversky’s
[59] feature contrast model. They used this concept
as a similarity measure and used it for texture image
matching for CBIR. Chen and Wang [60] gave fuzzy
set theoretic method namely unified feature matching
(UFM) for region based segmentation. They used this
concept in their experimental SIMPLIcity content based
image retrieval system for image matching. Chen et al
[61] gave an efficient and robust measure of similarity
which they called Fuzzy Integrated Region Matching
and used it for region-based image retrieval. Their
measure for similarity was between two families of
fuzzy sets. They tested their method on COREL
database consisting of 20,000 images and could achieve
better retrieval accuracy, speed and robustness
compared to IRM. Huang and Dai [62,63] gave a two
stage CBIR system for texture images. The first stage
was a fuzzy matching process based on energy
distribution pattern string (EDP). In the second stage
the composite sub-band gradient (CSG) vectors of
images and query image are compared after they pass
through a filter to provide good matching performance.
4.3. Three Step Hierarchic Method
Vujovic and Brzakovic [64] described a three step
hierarchic method to match random texture pattern in
a large image database. They could take into account
the misregistration of random pattern relative to the
representation in the database. They could determine
the probabilities of (i) two images having same pattern
(ii) wrong match and (iii) performance in terms of
processing time.
4.4. Method based on Texture and Colour
Mojsilovic et al [65] used a new distance function
using texture and colour features which correlates
with human performance. They could extract the
perceptual features from the vocabulary and their
match process use grammatical rules to compare the
patterns. They considered several examples in different
applications to illustrate the method. Nilufar and Chen
[66] gave a method with visual features colour and
textures using natural language. Their retrieval is using
a fuzzy description of these features. The use of
linguistic terms in description helps in efficient retrieval
from large databases.
4.5. Meta Search Engine based Method
Beigi et al [67] gave the concept of a content
based meta-search engine which they called
MetaSEEK, in which they used visual information to
find images. Their method has provision for interactions
with several on-line image search engines and for
intelligent selection and ranking according to
performance.
4.6. Hypo graph based Method
Chastel and Paulus [68] described a method in
which a digital image is matched with the help of hypo
graph. They discussed the problem of segmentation,
edge-detection, and noise detection. In their texture
classification they considered number of possible
configurations, determined one-to-one maps and
equivalent classes up to which mapping is possible
based on thresholds in the different grids.
4.7. Neuroscience based Method
Bileschi and Wolf [69] considered neuroscience
inspired set of visual features and used it for shape
based object detection, texture understanding and
classification. Their method has good potential for
texture image in CBIR.
4.8. Modified Zernike Moments
Sim et al [70] gave a method for texture image
retrieval which is rotation, translation and scale invariant.
They converted the image in Fourier domain and the
descriptor consisted of (i) the power spectrum for
translation invariance, (ii) normalized power spectrum
for scale invariance and (iii) modified Zernike moments
for rotation invariance. They tested the method on
several databases and found that it gives better accuracy
compared to Gabor, Radon and wavelet based methods
and it requires very low computational effort.
5. RESULTS
In the previous sections brief descriptions of a
large number of methods for texture image matching
 IVY MAJUMDAR et al : TEXTURE FEATURE MATCHING METHODS 263

were presented. In this section we are going to discuss Brodatz texture images of size 512 ´ 512 were taken
the results of a few of these important methods. from University of Southern California (USC) [74]
database. Then each image was divided into 16
5.1. Gray Level Co-occurance matrix based nonoverlapping subimages of size 128 ´ 128 to produce
spatial method nonrotated database D1. Small rotated database D2
was created by rotating the 13 images from Brodatz
Qin and Yang [11] in their method used Brodatz
database at 0°, 30°, 60° and 120° and then partitioning
[71] and the VisTex [72] texture databases for the
each of them into 128 ´ 128 nonoverlapping images.
experimentation. They compared their method with
Medium rotated database D3 were created by the
Ma and Manjunath’s [42,43] and Guo et al’s [73]
same method as that of the D2 by taking 40 images
method. Brodatz texture database contains 112 texture
from VisTex database [72]. Large rotated database
classes each of which is a 512 ´ 512 image. These
D4 were created by the same method as that of the
112 texture image classes were grouped into 32 clusters.
D2, D3 by taking 109 textures from the Brodatz
Each image was divided into 49 subimages of 128 ´
texture photographic album [71] and seven textures
128 pixels each with overlapping. The first 33 subimages
from the USC database of size 512 ´ 512 [74]. Do and
were used as the training set and rest of them were
Vetterli [75] used steerable wavelet domain- hidden
used for retrieval. Guo et al’s algorithm performs
Markov model (SWD-HMM) and gave the retrieval
image retrieval only with learning. On the basis of
result. Kokare et al compared their method with
retrieving images of same class as that of the query
standard DWT, dual tree complex wavelet transform
image results of Ma and Manjunath’s paper [42] tells
(DT - CWT) and SWD- HMM. Kokare
that their learning algorithm has about 90% retrieval
et al’s method combination of DT-RCWF and DT-
performance where as without learning average
CWT is the best for both rotated and nonrotated
retrieval accuracy of GLAM method is about 95%
databases. In precisely characterizing texture features
and with learning it has got 100% accuracy. For a
based on proposed scheme are more expressive than
given query image few retrieved images may not
the DWT features. In comparison with a standard real
belong to the same class as the query image but
DWT-based approach Kokare et al’s method improves
belong to the same cluster. Still they are taken as
the retrieval performance from 83.17% to 93.75%
similar to the query image because all subimages of
(database D1), 82.21% to 90.86% (database D2),
the same cluster are taken to be similar. GLAM
72.81 to 76.09% (database D3) and 64.17 to 78.93%
algorithm show best performance in terms of
(database D4). For CWT the retrieval performance
percentage of retrieving similar patterns also. GLAM
are 89.42%, 88.46%, 73.82%, 76.83% for D1, D2, D3
shows the best performance on Vistex database also.
and D4 respectively and for SWD-HMM 86.41%,
With learning it shows 100% retrieval accuracy and
86.77% for D1 and D2 respectively. As the texture
without learning the accuracy is about 91% for these
features of proposed method are robust and
textures. Samples of query and retrieved images of
efficient it shows much better performance for large
Qin and Yang’s method are shown in Fig 3.
database. Kokare et al followed the method stated by
5.2. Complex Wavelet based method B S Manjunath and W Y Ma [42] and evaluate the
performance in terms of the average rate of retrieving
Four different sets of experiment were carried out relevant images as a function of the number of top
by Kokare et al [50] using different database. 13 retrieval images. For top 116 retrieval of images DT-

(a) (b) (c) (d) (e) (f)

Fig 3 Samples of query and retrieved images for GLAM method. (a) query image, (b) first of the forty similar
image retrieved, (c) first of the fifteen perceptually similar images of first category, (d) first of the ten
perceptually similar images of second category, (e) first of the eight perceptually similar images of
third category, (f) first of the eleven perceptually similar images of fourth category
264 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

(a) (b) (c) (d) (e) (f)

Fig 4 Samples of query and retrieved images for combination of DT-RCWF and DT-CWT method.
(a) query image, (b) first similar image retrieved, (c) and (d) two of the sixteen similar image
retrieved, (e) and (f) two of the four perceptually similar images retrieved

CWT + DT-RCWF method (96.82%) is always better
than DWT (94.55%). Computational complexity is
approximately same for all the methods. In Fig 4 the
query and retrieved image samples of Kokare et al’s
method are given.
In the database there are 16 ground truth images
for each class. DWT retrieved 20 perceptually similar
images out of which only 8 were ground truth images
of that class in the top. DT-CWT retrieved 14 ground
truth images in the top but other six images were
perceptually very different. On the other hand Kokare
et al’s method retrieved16 ground truth images of that
class and other 4 images were perceptually very similar
to that of the query image.
5.3. Fuzzy Set Theoretic Method based
miscellaneous method
P W Huang and S K Dai [62] carried out their
experiment by using 150 images of size 512 ´ 512
taken from Brodatz album [71]. Each image was
partitioned into 16 nonoverlapping 128 ´ 128 images.
They compared their methods namely composite
subband gradient (CSG) method and hybrid method,
combination of CSG and energy distribution pattern
string (EDP) with gradient vector method [76]. CSG
shows better retrieval efficacy (93.20%) than the
gradient method (87.75%). CSG method has more
powerful discriminating ability (88.51%) than the
gradient vector method (87.75%). Hybrid method
shows almost the same efficacy as that of the CSG
method. For texture discrimination gradient vector
generated from the four subimages LL, LH, HL, HH
by wavelet decomposition for an image, LL has got
the maximum information. Retrieval efficacy of
different combination of gradient vector was calculated
and it was observed that the combination of
LL+LH+HL is the best in term of retrieval efficacy
and computation efficiency. In their system EDP strings
were used as filtering signatures to improve the speed
by keeping the retrieval efficacy same. The image
retrieval speed may be increased by two to five times.
As some of the qualified images may be filtered out
the retrieval efficacy may be reduced. However by
choosing proper fuzzy set we can have high speed as
well as high efficacy by using this method. Proposed
method has better performance (92% retrieval
accuracy) than Do and Vetterli’s [38] method (78%
retrieval accuracy). But in terms of representation of
texture features Do and Vetterli’s method is more
economic.Figure 5 shows the samples of query and
retrieved images of PW Huang and SK Dai’s method.
5.4. Modified Zernike Moments based
miscellaneous method
Dong-Gyu Sim et al [70] carried out their
experiment using Brodatz album [71], Corel texture
dataset CDs (size of each image is 768 ´ 512) and
some real texture of outdoor object taken from

(a) (b) (c) (d) (e) (f)

Fig 5 Samples of query and retrieved images for combination of CSG and EDP string method. (a) query
image, (b)-(f) sample of retrieved images of five different categories with ranks 1, 2, 3, 4 and 5.
 IVY MAJUMDAR et al : TEXTURE FEATURE MATCHING METHODS
(a) (b) (c)
Fig 6 Samples of query and retrieved images for modified Zernike moments based method. (a) Query image, (b) first
retrieved similar images, (c)-(f) perceptually similar retrieved images
Information and Communications University (ICU).
Each of 109 Brodatz texture images (512 ´ 512) was
partitioned into 16 nonoverlapping images (128 ´ 128).
The rotated texture database was obtained using 30
Brodatz textures and 25 ICU textures. The images
were rotated with 16 arbitrary angles and textures
rotated by 30° were taken as queries. Each image of
Corel set was divided into four subimages of size
384 ´ 286. Proposed algorithm is based on invariant
features so its performance is better (88.6%) than
conventional method namely Gabor and wavelet based
technique (89.58%, 70.79% respectively) for translated
and rotated textures. Zernike moment based method
is scale invariant. So it shows better retrieval accuracy
for scaled images (86.2%) than the Gabor, Radon and
wavelet based method (82.33%, 84.91%, 79.31%). It
can show good result for even 50% scaled images.
Descriptor size of Zernike moments method is small
(22 byte) in comparison with Gabor (50) and Radon
(56) method. Computational complexity is also less for
the proposed method. For feature extraction it takes
approximately 0.1 s on the Pentium-II 450 MHz.
which is less than the time taken by Gabor (1.7s) and
Radon (0.2s) but greater than wavelet based method
(0.05s). Figure 6 shows the samples of query and
retrieved images of Dong-Gyu Sim et al According to
the human perception the retrieved textures are very
similar to the query image.
6. CONCLUSION
In this paper an attempt has been made to review
the research and development work for matching
methods in content based image retrieval (CBIR)
systems using texture features. The two important
steps in matching namely the feature extraction and
similarity determination were considered and a set of
seventy one available research works were reviewed.
The performance of the system is usually specified in
terms of retrieval accuracy and retrieval time. In the
opinion of the authors there are two broad classification
namely (i) spatial domain methods and (ii) wavelet
based methods. Each of these classes were further
265
(d) (e) (f)
classified into a number of subclasses depending on
the method used in spatial domain and the type of
wavelet used in the wavelet domain. We find that
there are few other methods which could not be
classified into the above two classes which we called
miscellaneous methods. We have tried to give a very
brief idea to compare the effectiveness of some of the
methods. Kokare [77] has given a numerical comparison
of some of the wavelet based methods. A recent trend
is the combination of more than one method to improve
the accuracy. Although this will lead to increase in
retrieval time but with the development of VLSI
technology, increase in the processor speed and due to
the development of efficient algorithms/software there
can be considerable improvement in the retrieval time.
If we look into the works on CBIR using texture
images one interesting observation is that except for
wavelets, the other image transformations were not
tried or not reported even if some work was done in
some places. This may be because wavelet transform
became standard for JPEG, MPEG and several other
image processing algorithms. Fourier transform and
Fourier Mellin transform have excellent properties like
rotation, translation and scale invariance. Karhunen-
Loeve transform is optimum in some sense. Similarly
the transform like Hadamard, Haar, Walsh, Slant etc.
have several interesting properties. In our opinion
these can be tried for CBIR using texture patterns and
the advantages / limitations can be found out. Also
there can be R & D work to determine (i) application
specific methods, (ii) combination of more than one
method for texture matching and (iii) the computational
complexities of the methods.
REFERENCES
1. Manesh Kokare, B N Chatterji & P K Biswas, A
survey on current content based image retrieval
methods, IETE Journal of Research, vol 48, nos 3 &
4, pp 261-271, May-Aug 2002.
2. J Sklansky, Image segmentation and feature
extraction, IEEE Trans on Systems, Man and
Cybernetics, vol 8, pp 237-247, 1978.
266 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
3. Lei Qin & Wen Gao, Image Matching Based on A
Local Invariant Descriptor, IEEE International
Conference on Image Processing, pp 766-769, 2005.
4. Lei Qin, Wei Zeng, Wen Gao & Weiqiang Wang,
Local Invariant Descriptor for Image Matching, IEEE
International Conference on Acoustic, Speech &
Signal Processing (ICASSP 2005), pp II-1025 - II-
1028, 2005.
5. Kristen Grauman, Trevor Darrell, Efficient Image
Matching with Distributions of Local Invarient
Features, IEEE International Conference on
Computer Vision and Pattern Recognition, San
Diego, CA, June 2005.
6. Selim Aksoy & Robert M Haralick, Textural Features
for Image Database Retrieval, IEEE Workshop on
Content-Based Access of Image and Video
Libraries, Santa Barbara, CA, pp 45-49, June 21,1998.
7. Selim Aksoy & Robert M Haralick, Content-based
Image Database Retrieval Using Variances of Gray
Level Spatial Dependencies, IARP International
Workshop on Multimedia Information Analysis and
Retrieval, Hong Kong, pp 3-19, August 13-14,1998.
8. Selim Aksoy & Robert M Haralick, Using Texture in
Image Similarity & Retrieval, International
Workshop on Texture Analysis in Mashin Vision,
Oulu, Finland, pp 111-117, 1999.
9. Selim Aksoy & Robert M Haralick, Using Texture in
Image Similarity & Retrieval, M.Pietikainen, ed.,
Texture analysis in Machine Perception and
Artificial Intelligence, World Scientific, pp 129-149,
2000.
10. Svetlana Lazebnik, Cordelia Schmid & Jean Ponce,
Affine Invariant Local Descriptors & Neighborhood
Statistics for Texture Recognition, 9th IEEE
International Conference on Computer Vision
(ICCV’03), 2003.
11. Xuejie Qin & Yee-Hong Yang, Similarity Measure &
Learning with Gray Level Aura Matrices (GLAM) for
Texture Image Retrieval, IEEE Computer Society
Conference on Computer Vision and Pattern
Recognition (CVPR’04), 2004.
12. Guodong Guo, Hong-Jiang Zhang & Stan Z Li,
Distance-from-boundary as a Metric for Texture
Image Retrieval, IEEE International Conference on
Acoustics, Speech and Signal Processing
(ICASSP’01), vol 3, 2001.
13. Mari Partio, Bogdan Cramariuc & Moncef Gobbouj,
Texture Similarity Evaluation using Ordinal Co-
occurrence, IEEE International Conference on
Image Processing (ICIP), vol 3, pp 1537- 1540, 2004.
14. Mari Partio, Bogdan Cramariuc & Moncef Gobbouj,
Texture Retrieval using Ordinal Co-occurrence
Features, Proceedings of the 6th Nordic Signal
Processing Symposium - NORSIG 2004, June 9-11,
2004.
15. Mari Partio Bogdan Cramariuc & Moncef Gobbouj,
Block Based Ordinal Co-occurrence Matrices for
Texture Similarity Evaluation, IEEE International
Conference on Image Processing, vol 1, pp 517-520,
2005.
16. Noureddine Abbadeni, Content Representation &
Similarity Matching for Texture-based Image
Retrieval, ACM, Proceedings of MIR’03, Berkeley,
pp 63-70, 2003.
17. Noureddine Abbedeni, A New Similarity Matching
Measure Application to Texture Based Image
Retrieval, Proceedings of the 3rd International
Workshop on Texture Analysis and Synthesis
(TEXTURE 2003), 2003.
18. Marco Carcassoni, Eraldo Ribeiro & Edwin R
Hancock, Texture recognition through Modal
Analysis of Spectral Peak Patterns, IEEE,
International Conference on Pattern Recognition,
pp 243-246, 2002.
19. Marco Carcassoni, Eraldo Ribeiro, Edwin R
Hancock, Eigenvector Method for Texture
Recognition, IEEE, International Conference on
Image Processing, pp III-321-III-324, 2002.
20. Timothy F.Cootes, Gareth J Edwards & Christopher
J Taylor, Active appearance models, IEEE
Transaction on PAMI, vol 23, no 6, pp 681-685, June
2001.
21. Huizhong Long, Chee Wee Tan, Wee Kheng Leow,
Invariant & Perceptually Consistent Texture
Mapping for Content Based Image Retrieval, IEEE
International Conference on Image Processing, vol
2, pp 117-120, 2001.
22. Georgy L Gimel’farb & Anil K Jain, On Retrieving
Textured Images from an Image Database, Pattern
recognition, vol 29, no 9, pp 1461-1483, 1996.
23. P.P Ohanian & R C Dubes, Performance evaluation
for four classes of textural features, Pattern
Recognition, 25(8), 1992.
24. F Liu & R W Picard, Periodicity, directionality &
randomness: Wold features for image modeling &
retrieval, IEEE Trans on Pattern Analysis and
Machine Intelligence, vol 18, pp 722-733, 1996.
25. A Carkacioglu & F Y Vural SASI: a generic texture
descriptor for image retrieval, Pattern Recognition,
vol 36, pp 2615-2633, 2003.
26. Hui Xu & Mengyang Liao, Cluster-Based Texture
matching for Image Retrieval, IEEE International
Conference on Image Processing, pp 766-769, 1998.
27. H Tamura, S Mori & T Yamawaki, Textural features
corresponding to visual perception, IEEE Trans on
System, Man, Cybernetics, 8(16), pp 460-473, 1978.
28. James Ze Wang, Gio Wiederhold, Oscar Firschein &
Sha Xin Wei, Content-based image indexing &
searching using Daubechies Wavelets, Inter-
national Journal on Digital Libraries, pp 311-328,
1997.
29. A Laine & J Fan, Texture classification by wavelet
packet signature, IEEE Trans on Pattern Analysis
and Machine Intelligence, vol 15, no 11, pp 1186-
1191, Nov 1993.
 IVY MAJUMDAR et al : TEXTURE FEATURE MATCHING METHODS
30. T Chang & C C Kuo, Texture analysis &
classification with tree structured wavelet transform,
IEEE Trans on Image Processing, vol 2, no 4, pp
429-441, 1993.
31. M Unser, Texture Classification & Segmentation
using Wavelet Frames, IEEE Trans Image
Processing, vol 4, no 11, pp 1549-1560, 1995.
32. Junc C R, Scharcanski J, Stochastic Texture Image
Retrieval & Similarity Matching, IEEE, xiii brazilian
symposium on computer graphics and image
processing, pp 178-184, 2000.
33. Mohammad F A Fauzi & Paul H Lewis, Texture-
based Image Retrieval Using Multiscale Sub-image
Matching, IS and T/SPIE’s 15th Annual Symposium
Electronic Imaging : Image and Video
Communications and Processing, Santa Clara,
California, pp 407-416, 2003.
34. Manesh Kokare, B N Chatterji & P K Biswas,
Comparision of Similarity Metrics for Texture Image
Retrieval, IEEE, conference on convergent
technologies for Asia pacific region, TENCON 03,
pp 571-575, 2003.
35. Zhao -Wei Shang, Gui-Zhong Liu & Ya- Tong Zhou,
Texture Image Retrieval & Similarity Matching,
Proceedings of the third International Conference
on Machine Learning and Cybernetics, Shanghai,
pp 4081- 4084, August 2004.
36. George Tzagkarakis & Panagiotis Tsakalides, A
Statistical Approach To Texture Image Retrieval
via Alpha-stable Modeling of Wavelet
Decompositions, WIAMIS Lisbon, Portugal, April
21-23, 2004.
37. Minh N Do & Martin Vetterli, Texture Similarity
Measurement using Kullback-Leibler Distance on
Wavelet Subbands, IEEE International Conference
on Image Processing, vol 3, pp 730-733, 2000.
38. Minh N Do & Martin Vetterli, Wavelet-Based Texture
Retrieval Using Generalized Gaussian Density &
KullBack-Leiblar Distance, IEEE Transaction on
Image Processing, vol 11 no 2, pp 146-158, February
2002.
39. Blancho Patrice, Hubert Konik, Texture Similarity
Queries & Relevance Feedback for Image Retrieval,
IEEE International Conference on Pattern
Recognition, vol 4, pp 55-58, 2000.
40. P Wu, B S Manjunath, S D Newsam, & H D Shin, A
Texture Descriptor for Image Retrieval & Browsing,
IEEE International Conference on Computer
Vision and Pattern Recognition, pp 3-7, June 1994.
41. P Wu, B S Manjunath, S D Newsam, & H D Shin, A
Texture Descriptor for Browsing & Similarity
Retrieval, Journal of Signal Processing: Image
Communication, vol 16, Issue 1-2, pp 33-43, Sept
2000.
42. B S Manjunath & W Y Ma, Texture Features for
Browsing & Retrieval of Image Data, IEEE
Transaction on PAMI, vol 18, no 8, pp 837-842, Aug
1996.
267
43. W Y Ma & B S Manjunath, Texture Features &
Learning Similarity, IEEE International Conference
on Computer Vision and Pattern Recognition, pp
425-430, 1996.
44. Yu-Long Qiao, Zhe-Ming Lu, Jeng - Shyang Pan, &
Sheng - He Sun, Spline Wavelets Based Texture
Features For Image Retrieval, International Journal
of Innovative Computing, Information and Control,
vol 2, no 3, pp 653-658, 2006.
45. Chi Man Pun, Rotation-invariant texture feature for
image retrieval, Computer Vision and Image
Understanding, vol 89, pp 24 - 43, 2003.
46. P D Rivaz & N Kingsbury, Complex wavelet features
for fast texture image retrieval, Proceedings of IEEE
International Conference on Image Processing
(ICIP 1999), vol 1, pp 109-113, 1999.
47. S Hatipoglu, S K Mitra & N Kingsbury, Image texture
description using complex wavelet transform,
Proceedings of IEEE International Conference on
Image Processing (ICIP 2000), vol II, pp 530-533,
2000.
48. Manesh Kokare, P K Biswas & B N Chatterji, Rotated
Complex Wavelet based Texture Features for
Content Based Image Retrieval, Proceedings of
IEEE International Conference on Pattern
Recognition (ICPR 2004), 23rd to 26th Aug, 2004,
Cambridge (UK).
49. Manesh Kokare, P K Biswas & B N Chatterji, Texture
image retrieval using new rotated complex wavelet
filters, IEEE Trans on Systems, Man and
Cybernetics- part B, vol 35, no 6, pp 1168-1178,
2005.
50. Manesh Kokare, P K Biswas & B N Chatterji,
Rotation invariant texture image retrieval using
rotated complex wavelet filters, IEEE Trans on
Systems, Man and Cybernetics- part B, vol 36, no 6,
pp 1273-1282, 2006.
51. M Acharyya & M K Kundu, An adaptive approach
to unsupervised texture segmentation using M-Band
wavelet transform, Signal Processing, vol 81, pp
1337-1356, 2001.
52. Manesh Kokare & B N Chatterji & P K Biswas, M-
Band Wavelet based Texture Features for Content
Based Image Retrieval, Proceedings of Indian
Conference on Vision, Graphics and Image
Processing (ICVGIP 2002), pp 173-178, Dec 2002,
Ahemadabad (INDIA).
53. R D Koilpillai & P P Vaidyanath, Cosine-modulated
FIR filter bank satisfying perfect reconstruction,
IEEE Trans Signal Processing, pp 770-783, April
1992.
54. Manesh Kokare, B N Chatterji & P K Biswas, Cosine-
modulated wavelet based texture features for
content based image retrieval, Pattern Recognition
Letter, vol 25, no 4, pp 391-398, March 2004.
55. Manesh Kokare, P K Biswas & B N Chatterji, Cosine-
modulated wavelet packet based texture features for
content based image retrieval, IETE Journal of
Research, vol 51, no 6, pp 477-483, Nov-Dec 2005.
268 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
56. Jia Li, James Z Wang & Gio Wiederhold, IRM:
Integrated Region Matching for Image Retrieval, 8th
ACM international Conference on Multimedia
2000, Los Angeles, USA, Oct 30-Nov 3, pp 147-155,
2000.
57. Jams Z Wang, Jia Li & Gio Wiederhold, SIMPLIcity:
Semantics-Sensitive Integrated Matching for Picture
Libraries, IEEE Transaction on PAMI, vol 23, no 9,
pp 947-963, September 2001.
58. Simone Santini & Ramesh Jain, Similarity Measure,
IEEE Transaction on PAMI, vol 21, no 9, pp 871-
883, September 1999.
59. A Tversky, Features of similarity, Psychological
Review, vol 84, no.4, pp 327-352, July 1977
60. Yixin Chen, James Z Wang, A Region Based Fuzzy
Feature Matching Approach to Content Based Image
Retrieval, IEEE Transaction on PAMI, vol 24, no 9,
pp 1252-1267, September 2002.
61. Yixin Chen, James Z Wang, Jia Li, FIRM: Fuzzily
Integrated Region Matching for Content-based
Image Retrieval, 9th ACM international Conference
on Multimedia 2001, Ottowa, Canada, Sept 30-Oct
5, pp 543-545, 2001.
62. P W Huang & S K Dai, Image retrieval by Texture
similarity, Pattern Recognition, vol 36, pp 665-679,
2003.
63. P W Huang & S K Dai, Design of a two-stage content
based image retrieval system using Texture
similarity, Information Processing and
Management, vol 40, pp 81-96, 2004.
64. N.Vujovic & D.Brzakovic, Evaluation of an
Algorithm for Finding a Match of a Distorted Texture
Pattern in a Large Image Database, ACM
Transaction on Information Systems, vol 16, no 1,
pp 31- 60, January 1998.
65. Aleksandra Mojsilovic, Jelena Kovacevic, Jianying
Hu, Robert J.Safranek, Kicha Ganapathy, Matching
& Retrieval Based on the Vocabulary & Grammar of
Color Patterns, IEEE Transaction on Image
Processing, vol 9, no 1, pp 38-54, January 2000.
66. Sharmin Nilufar & Liang Chen, H K Kwan, A Fuzzy
Query Based Image Retrieval System, Proceedings
of IEEE International Conference on
Communication, Circuits and Systems, vol 2, pp
1017-1021, 2005.
67. Mandis Beigi, Ana B Benitez & Shih-Fu Chang,
MetaSEEK: A Content-Based Meta-Search Engine
for Images, SPIE, January 1998.
68. S Chastel & D Paulus, Texture Identification Using
Image Neighborhood Hypergraphs, SPIE, San Jose,
2004.
69. Stanley Bileschi & Lior Wolf, AUnified System For
Object Detection, Texture Recognition, & Context
Analysis Based on the Standard Model Feature Set,
Proceedings of British Vision Conference (BMVC),
2006.
70. Dong Gyu Sim, Hae Kwang Kim, Rae-Hong Park,
Invariant texture retrieval using modified Zernike
moments, Image and Vision Computing, vol 22, pp
331-342, 2004.
71. P Brodatz, Textures: A Photographic Album for
Artists & Designers, New York: Dover, 1966.
72. MIT Vision & Modeling Group, Vision Texture
[Online] Available: http://vismod.www.media
mit.edu
73. Guo, G, S Z Li, & K L Chan, Learning Similarity for
Texture Image Retrieval, ECCV, pp 178-190, 2000.
74. University of Southern California, Signal & Image
Processing Institute, Rotated Textures [Online]
Available: http://sipi.usc.edu/services/database/
Database.html
75. M N Do & M Vetterli, Rotation invariant texture
characterization & retrieval using steerable wavelet-
domain hidden Markov models, IEEE Trans
Multimedia, vol 4, no 4, pp 517-527, Dec 2002.
76. S.R Fountain, T N Tan, Efficient rotation invariant
texture features for content-based image retrieval,
Pattern Recognition, vol 31, pp 1725-1732, 1998.
77. Manish Kokare, Wavelet Based Features Texture
Image Retrieval, PhD Thesis, IIT Kharagpur, 2004.
 IVY MAJUMDAR et al : TEXTURE FEATURE MATCHING METHODS
Authors
Ivy Majumdar born on 12th May
1972, obtained BTech degree in the year
of 1997 in Optics and Optoelectronics
and MTech degree in the year of 1999 in
Radio Physics and Electronics both from
Calcutta University. She worked as JRF
in the Optics and Optoelectronics
Department of Calcutta University in
the year 1999.Since January 2000 she is
working as faculty member at B P Poddar
Institute of Management and Technology in the Electronics and
Communication Engineering Department where she became
Assistant Professor in the month of February 2006. She has
registered herself for PhD degree in Jadavpur University in the
area of Content Based Image Retrieval. Her research interests
include Pattern Recognition, Image Processing. Image Retrieval,
Soft Computing, Optoelectronics and Communication
Engineering.
* * * Governors of IIT Kharagpur during 1998-2000. He has published
Avijit Kar did his MSc and PhD
from IIT Kharagpur in 1980 and 1984
respectively. He is a Professor of
Computer Engg. in Jadavpur University,
Kolkata. He has supervised several PhD
theses and is actively involved in many
R&D activities and IT related consultancy
for the Govt and the private sector. He
mainly works in computer vision with
varied application areas, engg. systems
269
reliability and e-security modelling. Blended with his profession
he is a photography buff. Western classical music relaxes his
nerves.
* * *
B N Chatterji born on 10 Nov
1942, obtained BTech (Hons) (1965)
and PhD (1970) in Electronics and
Electrical Communication Engineering
of IIT Kharagpur. He did Post Doctoral
work at University of Erlangen-
Nurenberg, Germany during 1972-73.
Worked with Telerad Pvt Ltd Bombay
(1965), Central Electronics Research
Institute Pilani (1966) and IIT Kharagpur
as faculty member during 1967-2005. He was Professor during
1980-2005. Head of the Department during 1987-1991, Dean
Academic Affairs during 1994-1997 and Member of Board of
about 150 journal papers. 200 conference papers and four
books. He was Chairman of four International Conferences and
ten National Conferences. He has co-ordinated 25 short term
courses and was the chief investigator of 24 Sponsored Projects.
He is the Fellow/ Life Member/ Member of eight Professional
Societies. He has received ten National Awards on the basis of
his Academic/ Research contributions. His areas of interests are
Pattern Recognition, Image Processing, Signal Processing, Parallel
Processing and Control Systems.
* * *
 INVITED PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 271-276

Building a Strong Nation - The ECT way

R SREEHARI RAO, FIETE

Director, DLRL, Hyderabad

email: drrsreeharirao@yahoo.com

The strength of any nation depends on the availability of resources, its industrial
strength and infrastructure base, with specific emphasis on the communication
infrastructure. India, on one hand has abundant natural resources and excellent industrial &
infrastructure base, which is mostly confined to urban and civilized society. On the other
hand, it has also a vast majority of population residing in villages, which is deprived of the
basic amenities and other growth opportunities. In this backdrop, the goal of achieving the
government’s vision to make India a ‘developed’ nation by the year 2020 is not a simple task.
This vision of transformation to a ‘developed’ India can only be realized by empowering the
rural people with the Knowledge Powered PURA (Providing Urban amenities in Rural Areas).
This is only possible through the creation of a Knowledge Society. This paper discusses the
ways and means of leveraging the ‘Electronics and Communication Technology’ towards
achieving this goal.

INTRODUCTION Integration of above three components results in

‘Economic Connectivity’ that will lead to self-actuating

D URING the last century, the world civilization

has undergone a change from the manual labour
based ‘Agriculture Society’ to the technology/capital/
labour based ‘Industrial Society’ and then to
‘Information Society’, where connectivity and software
products were the critical factors that drove the
economy of many nations to greater heights. A new
concept of ‘Knowledge Society’ is now emerging
where knowledge is the primary production resource
instead of capital and labour. Effective utilisation and
dissemination of knowledge can create comprehensive
wealth for the nations and can improve the quality of
life - in the form of better health, education,
infrastructure and other social indicators. Ability to
create and maintain the knowledge infrastructure,
develop knowledge workers and enhance their
productivity through creation, growth and exploitation
of new knowledge will be the key factors in deciding
the prosperity of this Knowledge Society.
Rural empowerment depends on the extent of
available connectivity, which has four components.
‘Physical Connectivity’ can be improved by providing
more roads in rural areas, ‘Electronic Connectivity’ by
providing reliable communication network and
‘Knowledge Connectivity’ by establishing more
professional institutions and vocational training centers.
Paper No 126-D; Copyright © 2007 by the IETE.
people and economy. Electronics and Communication
Technology plays a vital role in the last three
connectivity mechanisms.
The realm of ‘Electronics and Communication
Technology’ (ECT) encompasses Telecommunication,
Broadcasting and Information Technology, resulting
in a unified Information Infrastructure - ‘Information
& Communication Technology’ (ICT) capable of
carrying any type of information, be it text, data, voice
or video. Thus ECT and ICT can mostly be used
interchangeably. The Internet, with its huge quantities
and variety of content, is an effective delivery and
exchange system for information and knowledge,
continuing education and learning.
ECT TOWARDS ‘GLOBAL-VILLAGE’,
INDIA PERSPECTIVE
There are about 637,000 villages in India, with an
average 250 - 300 households per village. Like in any
country, the rural per capita income is distinctly lower
than the national average, and this rural income
distribution is also more skewed. A typical village
may have only 100 households with sufficient income,
while the rest struggling to earn just enough to meet
the essential needs. Two-thirds of the Indian
households are dependent on agriculture for income,
and even this is often seasonal and dependent on
rainfall.

271
272 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
In order to improve rural economy, agriculture has
to be run as an agri-business rather than for mere
subsistence. This needs re-educating the farmer on
crop cultural practices for high yield with minimum
expenditure. Hence, innovative improvements are the
need of the hour. The rural economy can be boosted if
villagers have access to the following Facilities/Services:
Capacity Building
– Education
– Health
– Land distribution, Title and other Grievance
Redressal etc.
Income Generation
– Agriculture
– Entrepreneurship
– Outsourcing
Enabling Services
– Finance
– Markets (exchanges /trading)
– e-Governance
– Water Management
– Energy
– Communications and Transportation
Since most of these services can be extended
through ECT, it can be effectively utilized to leverage
rural income. At the basic level, Communication
Facilities reduce the need for physical movement of
people, which in turn results in the saving of cost, time,
effort and energy. ECT itself will not work wonders,
but it will open up people, especially the young, to new
ideas and new worlds. Computers connected through
ECT will make people learn new skills, which could be
harnessed in a myriad of different areas.
For example, farmers could use ECT to get
commodity prices faster, or get information on new
agricultural techniques. The youth would get details on
job opportunities across the state. The district
administration could get details of problems in near
real-time. The eligible could search for matrimonial
matches across adjacent villages. The voters would
communicate their concerns to the politicians and
bureaucrats electronically. The village officials could
share governance best practices faster among their
counterparts elsewhere.
If access to the following facilities is made available
through ECT, preferably with local language interfaces,
it will certainly improve the quality of life of the poor
rural communities:
1. Information on Market Prices, Cropping
Pattern, Weather Forecast, Agriculture
Marketing (getting better prices for produce),
2. e-Governance: Access to all Forms, Copies
of Land Records, Applications, Certificates,
Grievance Redressal etc.
3. Collection of bills: Electricity, Telephone, and
House-Tax etc.
4. e-Commerce: Insurance, e-Banking etc.
5. Telemedicine, e-Learning.
The farmer needs a complete end-to-end solution
from pre-plantation consulting to post harvest storage
and sales. He needs timely information on various
inputs like seeds, fertilisers, pesticides, farm
machinery and financing opportunities. He also needs
timely solutions to irrigation risks like production risk
(crop disease), rainfall risk and market risk (price
risk). In addition, he also needs complete information
on market trends, harvest storage procedures and
facilities, and transportation means. ECT can be
effectively used in addressing all these aspects. For
example, risk of crop disease can be managed by use
of video conferencing to connect the farmer to an
agricultural expert and obtaining the answers before it
is too late.
e-Governance mechanism ensures speedy service
to the people under one roof for all their government
related service requirements. It is addressed with
electronic connectivity between the various state and
central Government Departments and their respective
databases with real time updated data exchange for
fully committed and transparent administration. With
this, all forms and applications are made available for
online submission for speedier processing. Access to
land records can also be made available and copies
of required records can be obtained online. In addition,
any required certificates from Govt Departments can
also be obtained online. Similarly, these services can
also be extended for online bill payments in respect of
electricity, telephone, and house-tax, insurance etc.
Access to e-Commerce simplifies the banking
transactions including loan sanctions and repayments.
With telemedicine, rural PHCs equipped with ECT
facilities may obtain timely treatment and remedial
advices for any advanced or abnormal deceases. A
moderately learned technician or a doctor can interact
with the specialist by forwarding the symptoms of the
decease and the condition of the patient to get
appropriate medical advice for proper treatment. Expert
advice for moving the patient to appropriate high level
medical centers and suggestions on the vacancy position
in various nearby hospitals may also be obtained
 R SREEHARI RAO : B UILDING
interactively online. This will be more helpful for timely
action under emergency conditions such as epidemic
outbreaks or natural disasters.
With e-Learning, normal schools can be turned
into hi-tech ones with online e-lessons, to cope up with
the prevailing low teacher/student ratio of the order of
1:58 at primary level in rural regions of India. It can
also be used for imparting high-level computer education
to rural students. e-Learning ensures equal opportunity
to students as it presents the same content and standard
to all, irrespective of their physical location.
The brighter side of India
India is the world’s largest democracy and the
10th largest industrial power, with solid and consistent
economic growth. It has the third largest scientific and
technical workforce with its engineers holding top
positions in computer corporations all over the world.
In agriculture, India is the highest producer of sugar,
groundnuts, tea & fruits and is the number two in
production of rice, wheat, vegetables and milk.
India has also recorded tremendous growth in
communication infrastructure. It presently holds a mobile
phone customer base of about 100 million and landline
customer base of about 40 million. The explosion of
Internet has also had its impact over India, which is
evident from the whooping demand for Broadband &
Dial-up Internet connections. The current Internet
customer base of about 10 million is expected to touch
20 million by the year 2010. All this advancement has
been possible with the right use of Electronics and
Communication Technology.
The Darker side of India
Like in many other countries, unfortunately in
India too, ECT has been concentrated in the urban
pockets, leading to a clear digital-division between the
urban and rural population. Since timely access to
news and information makes remarkable difference in
promoting trade, education, employment, health and
wealth, this “digital divide” threatens to exacerbate
wide gaps between the rich and the poor.
With about 70 per cent of Indian population residing
in rural areas characterized by grinding poverty and
social injustice, it is not possible to build a strong nation
without extending the fruits of the advances in
technology, in particular the ‘Information &
Communication Technology’, to the rural masses.
Unfortunately, the rural populations mostly comprising
of uneducated peasants and artisans are not able to
make use of these technologies to their advantage.
A STRONG NATION 273
The literacy rate in rural India is only about 50 per cent
as against 85 per cent in urban areas. India has 192
million illiterate women (mainly concentrated in rural
areas) that represent nearly one-third of all illiterate
women in the world.
With regard to child education, even though the
Eighty-sixth Constitution Amendment Act makes free
and compulsory education for children between the
age group of 6 to 14 years as a fundamental right, their
desperate economic conditions force the poor rural
children to toil often in subhuman conditions, depriving
them of their most basic rights as children, i.e., education
and a joyful childhood. It is pitiable to note that India
has the largest number of child workers in the world.
As per a case filed in Supreme Court last year, there
are 100 million working children in India, nearly half of
India’s child population.
Social impact of the ‘Digital Divide’
The following scenario illustrates how rural youth
with little or no ICT skills can be marginalized in the
job market:
There are kids in village schools who do not
even receive their textbooks on time while
their counterparts in urban schools have all
the necessary resources to send e-mails and
surf the Internet for supplementary information
for school projects. Without any doubt, all
these children will be queuing up for
employment in the near future and most
probably the ones who lack familiarity and
skills in ICT will be sidelined. Children
everywhere have the same levels of curiosity.
They can learn at the same quick pace of
their city brethren. For these children, the
availability of computer and Internet services
made all the difference.
This type of marginalisation from the
mainstream and frustration of the vernacular
educated youth due to unequal opportunities
often lead to violent youth insurrections and
civil wars. Thus, for the overall prosperity of
any nation, creation of equal opportunities is
of prime importance.
These circumstances lend to large-scale
exploitation of rural poor even by moderately wise
businessman. Rural illiteracy often makes them fall
prey for the dominant landlords and other financiers in
villages. Since village administrations (Panchayats)
are normally in the hands of influential few, it is likely
that the benefits of the government sponsored programs
274 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
and schemes aimed at economic development and
social justice will not reach the innocent poor. They
remain powerless, locked in an oppressive system of
economic exploitation, class division, caste prejudice
and pervasive corruption. While the economy of modern
India grows, the rural people lack access to education,
nutrition and health care, sanitation, land and other
assets that could otherwise enable them to escape
from the trap of poverty. As a result the poor remain
poor and the rich grow richer.
It is said that if a society cannot help the many
who are poor, it cannot save the few who are rich.
This means, the nation as a whole cannot flourish
without involving the larger rural population. On the
other hand, the rural populations do not have the
necessary awareness, skills and facilities of ICT to
contribute to their own development. Schools with
best infrastructure, teachers who love teaching, primary
health centres, silos for storage of products and markets
for promoting cottage industries and business,
employment opportunities for artisans are some of the
elements of PURA. Therefore, we should plan to
promote rural prosperity by focusing on the information
and knowledge-led rural economy.
THE SOLUTION
One sustainable solution to this situation is to
reduce poverty by bridging the digital divide through
the process of ‘e-Learning’ among rural masses, aimed
at simultaneously imparting education and information.
With the current PC density of only about 4.6 per
1,000 people and a telephone density of about 32 per
1,000 people, India needs to do a lot (to increase
penetration of PCs and communication lines) for any
e-Learning project to be successful. The high cost of
ownership and higher tariff levels are preventing
proliferation of this technology in rural areas. Though
wireless connectivity could be considered as a cost
effective method to bridge the “digital divide”, the fact
that the bids for licenses under different telecom circles
failed to attract investment in rural circles clearly
indicates the lack of interest of the telecom providers
to cater to low volume and low paying users.
Under these circumstances, the state will have to
step in and either provide additional incentives to the
market or cater to the needs as a publicly funded
initiative. The government may also involve corporate
sector to come forward to fund such initiatives by
luring them with tax concessions.
Internet kiosks and community information centers
are to be planned with the involvement of NGOs that
have a strong presence in Rural India and which can
play an intermediary role with regard to empowerment
and capacity building. The governments also need to
promote access to education and health care through
distance learning and telemedicine, preferably with
local language interface. The rural ICT requires special
efforts to create appropriate models for those who
can neither afford the Internet access nor have the
language capacity to understand the content. One
example of such model is the UNESCO’s ‘Kothmale
Community Radio and Internet programme’ running in
central Sri Lanka.
Radio Web Browsing
The Kothmale Community Radio station
broadcasts a daily ‘Radio Browsing the Internet’
programmes, and in this programme, the
broadcasters, supported by resource personnel,
browse the Internet on-air together with their
listeners and discuss and contextualise information
in local language.
The radio programme thus contributes to raise
awareness about the Internet in a participatory
manner - the listeners request the broadcasters
to surf the WEB on their behalf and the
programme transmits information in response to
their requests. This information is explained and
contextualised with the help of the studio guests,
for example: a local doctor may explain data on a
health website.
A kiosk is a multi-point service delivery channel in
a village for delivering the government and other
services to villagers through ICT. The kiosk is equipped
with equipment like a computer, a digital camera and a
photo printer. The kiosk operator provides different
kinds of services to villagers like computer education,
computer lending, printing, copying, fax, telephone,
digital photos, insurance and other Internet-based
services. In addition, the kiosk also offers electronic
delivery of all government services to the people at
nominal charges with convenience of local availability.
Involvement of Self Help Groups (SHGs) may
also be considered as the answer for the transparent
delivery of a host of government schemes, education
and credit extension. In such a case the ICT based
village knowledge centres (internet kiosks) will be
organized and managed by the SHGs. The SHGs may
in turn identify some rural women and impart them
special training to run the kiosks.
THE PRESENT STATUS IN INDIA
The Indian government, on its part, has taken
 R SREEHARI RAO : BUILDING
significant steps towards dissemination of information
through a number of e-Learning projects, not only for
rural students but also for the community at large.
The ‘social development projects’ provide informal
IT training to the rural illiterate masses. They focus on
generating network of information centres, employment
and investment opportunities in rural areas. Examples
are akshaya, aqua choupal, gramin gyan kendra
(village knowledge center), etc.
The ‘community information services projects’
are aimed at the people who have the minimum required
knowledge. They have been designed to disseminate
information on rural farm-gate price realization, cut
transaction costs, supply of high quality farm inputs
etc. Examples are: e-Choupal and PlantersNet, etc.
Finally the ‘school based curriculum projects’ impart
computer education to rural students. Examples are
SchoolNet India, Uttaranchal’s Aarohi etc.
In addition to government, a number of ICT projects
are being sponsored by corporate sector. The most
notable among them are e-Choupal, i-Shakti, and e-
sagu etc. Today, they lead a silent revolution that
empowers farmers with relevant information to make
their lives better.
ITC’s e-Choupal web portal brings real-time
information on weather forecasts and customized
knowledge on better farming practices to the farmers’
doorstep to improve his crop management. ITC’s e-
Choupal supply chain brings good quality farm inputs
at competitive prices to increase his farm yields.
HLL’s i-Shakti provides information and services
to the farmers through a portal, which has contents
pertaining to a variety of rural issues. It enables farmers
to have solutions to pest problems and it also enables
rural women by providing a sustainable micro enterprise
opportunity to improve their living standards through
health and hygiene awareness.
e-sagu has a three-tier system consisting of farmers
as end users. The agricultural scientists with knowledge
system prepare farm advices. Coordinators as
intermediaries forward crop status through digital
photographs and text to Scientists and communicate
their advice to the farmers.
Indian Institute of Technology (IIT)-Bombay and
its partners sponsor the web site called aAqua.org to
provide farmers with relevant demand-driven farming
knowledge.
Govt of Andhra Pradesh has pioneered in
introducing e-Governance under the name ‘Rajeev
A STRONG NATION 275
Internet Village’ by use of Information and
Communication Technology. Under this program short
named ‘RajIV’, it is committed to provide good
governance and revitalize the rural economy for
integrated and sustained growth, to empower common
man in the rural parts of the State to access all
government, local-body and e-commerce services in
an integrated manner in order to create better means
of livelihood and improve the quality of life. Currently
more than 150 services are provided under this scheme.
Contributions of Defence Research and
Development Organisation (DRDO) - in
Electronics and Communications
Based on technologies from DLRL (Defence
Electronics Research Laboratory), Hyderabad, Convoy
Jammers for VIP protection are being developed.
Any explosive devices planted nearby the roads on
which vehicles travel can be deactivated by these
Jammer Systems. Technologies are also available with
regard to Low Power and High Power Cell phone
Jammers. These can be used to thwart the efforts of
terrorists who operate their destruction methodology
by using cell phones. Muting systems are developed
for Safety of Convoys of VIPs against Remotely
Controlled Improvised Explosive Devices.
Antennas Developed by the laboratory can be
used for different civilian communication systems. A
variant of Pulse Doppler Radar, developed can be
used for monitoring over speeding vehicles.
DEAL (Defence Electronics Applications
Laboratory), Dehra Dun, contributes to Development
of Satellite Communication Systems. In the last decade
by making use of Mobile S Band ‘Sat Com’ terminal,
during floods in Orissa, the disaster management was
successfully undertaken.
MM (Millimeter wave) imaging technology
developed in DEAL will be useful for identification of
any armaments, if any, concealed in the clothes of
persons.
Based on the technologies developed in CAIR
(Center of Artificial Intelligence and Robotics),
Bangalore, Robotics realized can be used for handling
Nuclear Materials.
“Light Weight Battery Powered Radar (BFSR -
Battle field surveillance Radar) can be used for
surveillance .of large industrial and other civil
installations to detect intruders up to the low distance
of around 2 km. Even a crawling man can be detected
by utilizing this system. Targets such as a crawling
276 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
man, walking man, group of walking men, and moving
light vehicles can be properly distinguished by this
BFSR.
‘Radar Data Processing Technology’ utilizing the
state-of-the-art processors (similar to power PCs)
and very efficient algorithm finds its applications in
automatic tracking of Multiple Targets in Air Traffic
Control Scenario.
Technologies Developed by LRDE (Electronic
Research and Development Establishment), Bangalore,
resulted in “Terminal Based ATC Radars”. In this
case, 3D Medium Surveillance Radar System can be
upgraded with specific Radar Data Processing for
improved and efficient handling of larger Air Traffic.
The Radar Networking Station can be utilized to inter
link various ATC Radars to extend the Air Surveillance
coverage to regulate Civil Air Traffic. Transfer of
Radar Air Pictures through appropriate advanced
Digital Data Link will enable melting shore based and
Author
R Sreehari Rao took over as
Director,Defence Electronics Research
Laboratory (DLRL), Hyderabad with
effect from 01 Nov 2005. He obtained
his BE (Electronics and Communication
Engineering) from College of Engineering,
Kakinada in 1972, MTech (Microwave
& Radar Engineering) from IIT,
Kharagpur in 1974, and PhD from IIT,
Madras in 1995. Dr Rao joined DLRL as
SSO-II in 1974 and served the organization in various levels.
In the last three decades, Dr R Sreehari Rao has made
significant contributions to the indigenous Design and
Development of Broad band Antennas, Microwave components
and systems for all three services i.e. Navy, Army and Air force.
off-shore Radar Installations.
CONCLUSION
As we are crossing the information society and
leading to knowledge society, irrespective of rural or
urban area, distance will be shortened using ECT.
Removal of poverty calls for improving the quality of
life in rural places to the extent of ‘Providing Urban
amenities in Rural Areas’ (PURA). Once this
proposition comes true, it may prompt for reverse
migration, i.e., instead of village population coming to
urban area, the urban population may come back to
villages. Such instances are becoming more common
in the developed countries, mainly due to the increasing
‘work from home’ culture. It helps to reduce the
urban congestion prevalent today in most of the metro
cities. Such an ideal situation will lead to more uniform
distribution of people, resources, facilities and
opportunities.
These systems were productionized and inducted into services.
In view of indigenous R&D efforts of DLRL, many production
agencies were established in the country to cater the needs of
services. A strong indigenous technology base has been established
in this field. He has successfully planned, steered and executed a
number of Airborne, Ship-borne and Ground based projects and
programmes. Because of indigenous efforts DLRL is in a strong
position to meet present as well as future requirements of all
three services avoiding perpetual dependence on foreign sources.
He has been honoured with many commendation and awards
towards his praiseworthy contributions in various programmes.
He has authored 66 technical reports and published several
articles in National and International Journals. He is a fellow of
IETE.
* * *
 INVITED PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 277-286

Satellite Technology Utilization for

Rural and Urban India
S PAL, DFIETE

Out Standing Scientist/ Program Director, SATNAV, Deputy Director,

ISRO Satellite Centre, Airport Road, Bangalore-560 017, India.
email: pal_surendra@hotmail.com
AND

V S RAO, FIETE

Group Head, Communication System Group, ISRO Satellite Centre,

Airport Road, Bangalore 560 017, India.
email: vsrao@isac.gov.in

Communication capabilities in various forms provided by the modern spacecraft are

bridging the gap between modernized cities and backward rural villages. The remote sensing
satellites are equally helping rural India by providing useful information to increase farm
yield, fish catching and also helping to save lot of lives by disaster warning and weather
forecasting. This paper describes the utilization of the satellite technology in various forms
for Indian masses particularly rural India.

INTRODUCTION cripple very often. Satellites come to the rescue in

such events for disaster management, which requires

T HE society having a better and quicker means of

communications is considered to be an advanced
and forward looking society. Most of our country’s
population in the past was not served by the telephone
and television networks that so greatly influence cities
and western societies. There is a great influence of
present day fast evolving technologies particularly the
digital techniques on the development of the society. It
includes present day computers and Satellite
Communication. Present day technological tools support
and complement each other to complete the total
communication scenario. Satellites are perceived largely
as a means to reach isolated places. Because of the
broadcasting nature of the satellites, a signal sent up to
the satellite comes down everywhere over a wide
area, thus providing connectivity to the inaccessible
parts of the country. Satellites bring the television to
the homes even in remote villages. Television is an
extremely powerful medium for education, literacy
rate in rural areas can be improved through tele-
education services based on satellite technology. Being
a tropical country with long coast line, natural disasters
often knock at the country and the rural people are
often the victims. Terrestrial links and transport systems
Paper No 124-A; Copyright © 2007 by the IETE.
277
real time decision making and action.
The introduction of space communication and the
present era of information technology has changed
the current scenario and we have seen the third
technological revolution (information + communication
= Information Technology) in the last decade of the
last century. The present era is called the IT era. The
technological evolutions which have taken place in
the last century are simply beyond one’s imagination.
Many technological reversals have been seen, like
the telephone which should have been on the wired
network has become wireless while the TV which
was wireless, works on cable.
Many individual spheres of working have become
almost universal, like education has come to the drawing
room from school and colleges, sectors like banking,
medicines, hospitals etc which were location specific
are available on net.
With the development of space technology, time
and distance have lost their conventional meaning,
thereby permitting men and women all over the world
to share their experiences, frustrations and successes
with great ease. Sophisticated and expensive medical
treatment has been made reachable to those
underprivileged society living in inaccessible parts of
278 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
India through telemedicine project conceived with
satellite technology. In a nutshell, the present day
world has become a global village and thereby we are
in a shrinking globe and expanding universe. There is
tremendous convergence and fusion of
communications, computers and associated technology
in the present era. Space communication technology
takes all this technological advancement to the rural
areas which are backbone of Indian democracy.
DESCRIPTION
As we become increasingly networked, our world
will grow smaller and bigger simultaneously.
Communication is at the root of the progress of every
society. Society is often described as essentially people
in communication.
The conventional communication tools are:
¨ Telegraphy using copper wire line
¨ Telephony using copper wire line
¨ Television
¨ News papers, books, etc.
¨ Means of Transport
All these have undergone tremendous changes and
the new technologies emerged are
¨ Mobile Satellite Telephone - Personal
Communication Systems
¨ INFOSAT (information satellite)
¨ A major shift from analogue to digital domain
¨ An increase in the value of software as
opposed to hardware content
¨ Extension of optical technology towards the
local loop coupled with increasing use of optical
switches and optical processing
¨ Use of wireless in access technology - New
access & Modulation codes
¨ Ultra broad band services
¨ Extensive use of spread spectrum and code
division multiple access techniques.
Services planned / in existence are:
¨ Virtual Private Networks
¨ PCS - Personal Communication Services
¨ Call Collect Services
¨ Desk top Video conferencing
¨ Card Phone Services
¨ Tele-shopping
¨ Video Telephone
¨ Tele-health
¨ Tele-education
¨ Interactive Video & Video on demand
¨ Multimedia transmission / reception
¨ e-mail
¨ e-Governance
¨ e-commerce
¨ Global positioning/timing system along with
SATNAV
¨ Digital TV, Direct TV to Home.
Most Important - The Internet & World Wide
Web
All these need either terrestrial or satellite channels
to serve the user. The present day scenario can be
best summarized by Fig 1. Here space communication
becomes an important tool irrespective of the large
growth in terrestrial communication networks
particularly when the technology is to be taken to rural
India.
Space Communication is provided by satellites in
various orbits. Satellites have been put in to all these
orbits and successful communication links have been
established. It will be a matter of large discussion if
one starts explaining the pros & cons of the orbits &
their satellites. Satellite communication started with
fixed services and expanded to greatest potential
applications in the mobile and broad cast services.
Satellites have innate advantage that make them an
attractive alternative or complement to terrestrial
broadband circuits. Satellite communication provides
reliable means of providing information or monitoring
inaccessible areas even in the case of severe cyclones
and disasters and when the entire terrestrial network
fails. The conventional services provided by satellites
are: Telephony/TV Broadcasting/Data reception and
distribution/Direct Television broadcasting/Disaster
warning/Continuous weather monitoring/Spacecraft
Vehicle Tracking and Commanding/ Inter satellite links/
Mail /Internet/Data mining, Position (GPS) and time
determination / Moving motor vehicle tracking etc.
Mobile Personal Communications
Most of the time technological advancements have
taken place due to defense requirements. However in
the area of communications where more than the
military requirements, it is the business requirements
which have given a big flip to the overall scenario of
communication and ushered us from fixed copper
wire communication era to the era of mobile
communication - meaning an individual (stationary or
 PAL & RAO : SATELLITE TECHNOLOGY UTILIZATION 279

Television Internet TV
Cable TV
Video on
AUDIO SYSTEM
Demand
DVD Space
Telephone Internet
Cellular
Cellular
Pager Internet
Telephony World
Fax Wide
Home Banking Web

Network
Printer Computing
Video Camera Terrestrial
Video
Desktop PC Conferencing
PDA
Credit Card Smart Card

Fig 1 A future communication scenario

on move) carrying small inexpensive hand held
communicator and being reached by voice, fax or data
with a single telephone number independent of location.
Basically it means to provide communication to and
from users located anywhere on the globe and
possessing a portable light weight hand held mobile
telephone. This could be termed as Personal
Communication Services (PCS).
Today, a hand held mobile phone has become a
personal hold and has become affordable by every
citizen across societies. This service is available only
in densely populated cities and towns. Even today, the
cellular mobile service is a dream for people in remote
areas. This can be fulfilled with mobile Satellite
Communication though it is yet to take off commercially.
The advantage of space communications is that one
does not have to be at a particular place to derive its
benefits. The Communication through space is
independent of place and time and geo political limit.
Though the Irridium and Global Star mobile satellite
communication projects have failed due to the
astronomically high terminal & service cost as
compared to terrestrial based systems, on engineering
account the two projects cannot be considered as
failures since the projects were engineering led -
rather than marketing led. In spite of cellular telephones
and other mobile services, the two systems have shown
their usefulness over others in Afghanistan operations
and also communications to inaccessible places.
In the foreseeable future new equipments and
techniques may be used in mobile “satellite technology
to bring down the costs and make the system affordable.
The day is not far off for providing the benefits of
technological advancement being enjoyed by urban
people to the people in rural inaccessible India.
Internet over satellite
It will be quite important and relevant to talk about
latest arrival on the communication technology scene:
the internet and the web. Today, the internet provides
a vast array of services with high bandwidth links that
can simultaneously carry telephone, video (Television)
and data and is accessible from anywhere through a
variety of information appliances ranging from personal
computers and hand-held digital assistants to screen
phones and televisions, not to mention the computers
embedded in everything from automobiles to vending
machines. The Net is a phenomenon that cannot be
ignored. It is an agent of change in all sectors of
society. In turn, this will lead to truly profound
280 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

changes in society & the present technological communication network using the Ku-band capacity
paradigm. of INSAT system. The Ku-band capacity of INSAT
will be used for promoting satellite based communication,
To help all these & to spread the net at a faster
specifically, in the areas of distance education, tele-
pace even to inaccessible & remote places satellite
medicine, agricultural extension, e-governance, self-
communication plays a major role, besides the
help groups, marketing and HRD, community internet
conventional terrestrial links, optical links etc., which
centres, etc. While ISRO will provide the transponder
cater to cities and larger population bases owning to
capacity on board INSAT and the technical support of
the economics.
the state governments will operate the satellite
It is expected that the space based high speed communication network for the various objectives afore-
internet service will reflect triple digit growth rate. As mentioned.
of today internet is the greatest thing to happen in the
SRO has already helped the Orissa State
satellite industry. For many space segment operators
Government in the implementation of Vidya Vahini
INTERNET is the biggest earner. As of today internet
Yojana. Besides, it has established the Jhabua
is going towards zero tolerance for failure. The goal is
Developmental Communication project in Madhya
more & more critical services to NET. The availability
Pradesh that is in progress since 1996. ISRO has also
of internet services through space is a big boon to rural
started the INSAT training and developmental
folk to get all the information required for increasing
communication channel in 1995 that is being used by
their knowledge including agricultural productivity.
several private and Government agencies for industrial
training, training of panchayat workers, agricultural
Services Provided by ISRO
extension workers, etc. The MOU signed with Andhra
Pradesh marks yet another milestone in the
ISRO had mainly three types of satellites viz.,
implementation of a countrywide Gramsat network
Communication satellites mainly for communication
for grassroots level development.
operating in GEO, Remote sensing satellites deployed
in LEO and scientific satellites; for various applications
Video-conferencing facility
driving the growth and development of entire India.
Many applications are mainly focused towards
Provided by INSATs is a great achievement in
upliftment of rural areas by bringing education, health
bringing various segments of administration together.
etc. to the door steps of rural homes/community centres.
To give an example, the facility enabled then Prime
Minister, Mr Atal Behari Vajpayee, on August 01,
INSAT Utilization
2000 to have a virtual “tete-a-tete” with the chiefs of
Panchayat Raj Institutions from 14 video/studio link
Satellite Television broadcast was first introduced
centres spread across the states of Uttar Pradesh,
by ISRO in India 32 years back when the Satellite
Gujarat and Karnataka. In an hour-long conversation,
Instructional Television Experiment (SITE), targeted
the Prime Minister had a first hand assessment of the
for rural areas, was commenced on August 1, 1975.
nitty-gritty of the local problems being faced at the
The video conference involved the use of INSAT-2C
grassroots level in the country. The video conferencing
and INSAT-3B satellites. Over the years, ISRO has
centres had been established at Bakshi ka Talab,
also conducted interactive training and distance
Gaurabaug, Sarojini Nagar, Nishatganj and Cantonment
education programmes through the INSAT satellite
Park Road in Lucknow district (UP), Himmat Nagar,
system.
Mehsana, Nadiad, Palanpur and Gandhinagar in Gujarat
INSAT series satellites deployed in GEO orbit are and at Mangalore, Dharwar, Raichur and Tumkur in
mainly providing Karnataka. Lauding the role of science in the rapid
strides being made by the nation, the Prime Minister
• Telecommunication
congratulated the scientists and engineers for their
• Television endeavor to make the video conference a reality. Mr
• Search and Rescue Vajpayee said that this would bridge gaps and open up
• Meteorology new vistas of growth and communication.
• Radio Networking.
Video conference provided a platform for the
The Indian Space Research Organisation (ISRO) Prime Minister to give a patient hearing to the
signed Memorandum of Understanding with various Panchayat representatives and, at the same time,
state governments for establishing a satellite-based apprised them of the limitations and practical problems
 PAL & RAO : SATELLITE TECHNOLOGY UTILIZATION
faced by the government at times and to discuss vast
array of issues viz., specific problems faced by
panchayat institutions like education, water shortage,
and irrigation.
Tele-medicine
The advances and convergence of IT and
telecommunication can bring the entire health care
services to the patient’s doorstep. Tele-medicine is
delivery of health care information across distances
using telecom technology. This includes transfer of
images like X-rays, CT, MRI, ECG, etc from patient
to expert doctors seamlessly, apart from the live video
conferencing between the patient at remote hospital
with the specialists at the super speciality hospital for
tele-consultation and treatment.
ISRO has been spearheading satellite based tele-
medicine programme in the country with remote district
hospitals connected to Super Specialty Hospitals in
major cities using INSAT satellite (Fig 2).
The advantage of tele-medicine in reaching out to
the rural and remote population has been well
established through the experience of this present
telemedicine network. However, there is a need to
Fig 2 Telemedicine via satellite
281
provide a common platform to all concerned agencies
like communication systems/software and medical
equipment providers, super-specialty hospitals,
healthcare administrators, various departments of
Government, trust and private hospitals, NGOs and
corporate hospitals dealing with healthcare both in
India and abroad. It will also help to exchange ideas,
practices and methodology to effectively implement
and use the emerging telemedicine technology
In the field of tele-medicine, the first pilot project
was started in Andhra Pradesh in 2000 connecting
Apollo Hospitals at Chennai and Aragonda village and
ISRO’s hospital at Satish Dhawan Space Centre,
SHAR, Sriharikota, in Nellore District. The tele-
medicine network is extended to more hospitals.
DTH-TV Broadcasting
Common entertainment today is through Television.
Television broadcasted by satellites enabled every
citizen in all parts of the country to have the benefits.
DTH - TV broadcasting comes under BSS
(Broadcasting services) as well as under FSS (Fixed
Satellite Services). Under FSS services it is from point
to multipoint but whose location is known while under
BSS it is always in universal broadcast mode. With
282 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
digital TV in 36 MHz standard transponder, one can
transmit almost 12 TV channels using latest
compression, error correcting codes etc, where even
at low SNR better TV quality can be received. This
will give rise to the possibility of reducing orbital
separations between DTH Satellites. This is only the
means to provide entertainment communication to rural
inaccessible areas.
Village Resource Centres (VRC)
Communication Satellites are effectively used to
provide all most all the useful information needed for
the village folks by establishing village resource centres/
information kiosks (Fig 3). This is another highly
successful project to bring technology to the benefit of
rural areas.
VRC project strives to promote a need based
single window delivery system for providing services
in the areas of education, health, nutrition, weather,
environment, agriculture and livelihoods to the rural
population to empower them to face challenges. This
satellite based project, aims for digital connectivity to
remote villages for providing multiple services such as
tele-medicine, tele-education and remote sensing
applications through a single window. The setting up
of VRC is yet another saga of adventure that has been
taken up by ISRO to benefit the rural society.
The VRC concept has been evolved and
implemented by integrating ISRO’s capabilities in
satellite communications and satellite based earth
Fig 3 Village community information kiosks
observation to disseminate a variety of services
emanating from the space systems and other
Information Technology tools to address the changing
and critical needs of rural communities. The VRC is a
totally interactive VSAT (Very Small Aperture
Terminal) based network.
Users located at one node of this network can
fully interact with others located at another node through
video and audio links. Each of the nodes can be
further extended using other technologies like Wi-Fi,
Wireless and Optical Fibre. These extensions may
serve as local clusters around the areas where the
VRC is located.
The information provided will be in the form of
geo-referenced land record, natural resources, suitable
sites for drinking water as well as sites for recharging
to replenish ground water, water harvesting, wastelands
that can be reclaimed, rural employment creation,
watershed, environment, infrastructure, alternate
cropping pattern, and so on. By suitably blending the
information derived from earth observation satellites
with ground derived and weather related information,
locale-specific community advisory services can be
provided. Community based vulnerability and risk
related information, provision of timely early warning
and dissemination of severe weather related information
can lead to reliable disaster management support at
the village level.
VRCs can also provide a variety of services like
tele-education, tele-medicine, online decision support,
 PAL & RAO : SATELLITE TECHNOLOGY UTILIZATION
interactive farmers’ advisory services, tele-fishery, e-
governance services, weather services and water
management. By providing tele-education services,
the VRCs act as virtual community centric learning
centres. At the same time, VRCs will provide
connectivity to speciality hospitals thus bringing the
services of expert doctors closer to the villages.
In addition, VRCs will facilitate access to spatial
information on important subjects like land use/land
cover, soil and ground water prospects which can
enable the farmers to get support in taking important
decisions based on their query. Besides, VRCs will
enable online interaction between the local farmers
and agricultural scientists. Fishermen can obtain
information on sea state and wave heights. Provision
of information on many governmental schemes,
location and farming system specific action plans
based on weather, community specific advice on soil
and water conservation are the other services rendered
by VRCs.
Search & Rescue
The search and rescue facility provided by satellites
is a breakthrough in technology in ensuring safety and
providing support in case of disasters/accidents to the
tourists, fishermen etc.
Meteorology
All the first generation INSATs and INSAT-2A,
2B and 2E in the second generation in the operational
INSAT system were configured for multi purpose
mission with transponders for telecommunication &
broadcasting and Meteorological payloads. To meet
the demanding requirements of Indian Meteorological
Department’s, ISRO launched a satellite with dedicated
meteorological payloads - Very High Resolution
Radiometer (VHRR) and Data Relay Transponder
(DRT), and is named as Kalpana-1 after launch in the
memory of Late (Ms) Kalpana Chawla, Indian born
American Astronaut.
Very High Resolution Radiometer (VHRR) is a
three band VHRR capable of imaging the earth in
three spectral bands namely Visible (VIS), Water
Vapor (WV) and Thermal Infra Red (TIR), to provide
both day and night coverage.
The weather data relay transponder (DRT) receives
402.75 MHz signals from unattended weather data
collection platforms, translates them to 4506.05 MHz
and retransmits it to a central facility - Meteorological
Data Utilization Centre, IMD, Delhi. This data is being
283
used along with the VHRR imageries for weather
forecasting.
Satellite derived products data are increasingly
used in conjunction with conventional meteorological
observations in the synoptic analysis and conventional
weather forecasts to extract information of relevance
to various sectors in India. The impact of satellite data
is phenomenal in certain areas of meteorological
applications such as thunderstorm forecasts, Tropical
Cyclone monitoring, aviation forecasts etc. The major
application of satellite data has been the monitoring of
Synoptic weather systems ranging from thunderstorms,
fog detection to cyclones and planetary scale
phenomena such as monsoon.
Synoptic applications of satellite imagery are in
use at Indian Meteorological Department. Following
satellite data products are derived from Kalpana-1
and the same are also archived and displayed on IMD
website daily.
• Earth cloud imagery in visible, IR bands and
water vapor every hour on black & white and
color.
• Clouds Motion vectors over Bay of Bengal,
Arbian Sea and Indian Ocean at 00, 07, 12
and 1 8 UTC and being disseminated over
GTS.
• Outgoing long wave radiation (OLR) on daily/
weekly/monthly basis at 0600UTC.
• Sea surface Temperature (SSTs) at 0600 UTC
is being derived over Bay of Bengal, Arabian
Sea and Indian Ocean.
• Quantitative Precipitation Estimates (QPE)
on daily/weekly/monthly basis.
Satellite bulletins based on 3 hourly / 1 hourly
Kalpana-1 cloud imageries are prepared and transmitted
to all the forecasting offices on Global Telecom Service
(GTS) through RTH, New Delhi. Satellite imageries
and animations are also put on the IMD website on
regular basis for the users and public. Heavy rainfall
advisory, bulletins are also transmitted regularly for
IMD forecasting offices and other centres daily. During
the tropical cyclones in Arabian Sea and Bay of
Bengal hourly special satellite bulletins with intensity
and position are also issued for all the users by the
satellite Div., IMD.
During winter season, fog formation is one of the
most important weather events over northern parts of
India which affects the aviation badly. The Kalpana-1
imageries are regularly utilized for capturing such
events.
284 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
IRS utilization
Remote Sensing satellites launched and operated
by ISRO are highly successful in providing vital
information regarding earth resources and also in
providing cartographic maps for town planning and
development.
Data from Indian Remote Sensing Satellite is used
for various applications of resources survey and
management under the National Natural Resources
Management System (NNRMS).
• Preharvest crop acreage and production
estimation of major crops.
• Drought monitoring and assessment based on
vegetation condition.
• Flood risk zone mapping and flood damage
assessment.
• Hydro-geomorphological maps for locating
underground water resources for drilling well.
• Irrigation command area status monitoring
• Snow-melt run-off estimates for planning water
use in down stream projects
• Land use and land cover mapping
• Urban planning
• Forest survey
• Wetland mapping
• Environmental impact analysis
• Mineral Prospecting
• Coastal studies
• Integrated Mission for Sustainable
Development for generating locale-specific
prescriptions for integrated land and water
resources development in 174 districts.
Many state governments are utilizing the remote
sensing data for various applications including town
planning. For example, Andhra Pradesh has been
extensively using ISRO satellites for societal
applications. Andhra Pradesh is one of the foremost in
the utilisation of remote sensing technology for
management of land and water resources and disaster
management. Remote sensing technology is used for
watershed management in drought prone districts of
Adilabad, Ananthapur, Kurnool, Mahaboobnagar,
Nizamabad and Ranga Reddy. Management plans for
major irrigation projects like Srirama Sagara and
Nagarjuna Sagara have been planned using remote
sensing data. More than 35,000 bore wells have been
drilled using remote sensing data with better than 90
percent success rate.
Oceansat
The 1050 kg satellite placed in a Polar
sunsynchronous orbit of 720 km height. IRS-P4 carried
on board an Ocean Colour Monitor (OCM) and a
Multifrequency Scanning Microwave Radiometer
(MSMR).
OCM is a solid state camera operating in eight
narrow spectral bands. The camera is used to collect
data on chlorophyll concentration, detect and monitor
phytoplankton blooms and obtain data on atmospheric
aerosols and suspended sediments in the water.
The Microwave Scanning Radiometer (MSMR)
which operates in four microwave frequencies both in
vertical and horizontal polarisation with its all weather
capaility is useful for measuring sea surface temperature
and meteorological parameters like Sea Surface
Temperature (SST), Atmospheric water vapor and
Sea surface winds.
The Oceansat data helped fisherman to a great
extent in locating fish concentration in the sea and in
increasing the fish produce. The satellite data also
helped to study the ocean behaviour and to provide
safety guidelines and precautions to fisherman, divers
and navy to a great extent.
INFOSAT
The future Telecommunication spacecrafts will
be developed from transmission in to Information
Satellites (INFOSAT). They will be given many of the
properties of terrestrial telephone exchanges and signal
processing equipments and it will be possible to integrate
them directly into future global networks. They will
thus permit immediate applications of many existing
and future services. Because of their inherent built up
flexibility, these satellites will be able to support and
speed up the initial experimental phase of many new
services before their trial on terrestrial networks. These
type of satellites will enable new services to be tried
out over a large area before being put in to the market
and optimally adapted to suit the most appropriate
transmission medium.
The future INFOSATS will be of three types:
National / Regional, International & Relay.
Future technologies will enable the construction of
an INFOSAT network in which the above three will
be connected to each other. The onboard processors
will ensure that the signals to be exchanged between
the satellite and the terrestrial subscribers are combined
using Time Division and Space Division multiplexing
 PAL & RAO : SATELLITE TECHNOLOGY UTILIZATION
techniques and distributed in accordance with the user
requirements. Special coding techniques will ensure
the security of transmissions.
Configuration of INFOSAT
The satellite platform will have multiple
reconfigurable antennas/transmitters with dynamic
power sharing / Receivers in X various frequency
ranges / large reconfigurable switching matrices at
baseband and at RF level / intersatellite links permitting
signals to be exchanged between satellites according
to changing requirements / complex and efficient analog
/ optical / digital signal processors / New modulation
technique and multiple access techniques. It will be
sufficiently broadband systems & may have even
optical space communication components.
The satellites besides their autonomous control
and power generation equipment may have sensors to
observe the earth’s atmosphere and pass the data to
an appropriate station after processing. This will help
to deal with the situation like Orissa cyclone. Such a
system will obviously have Geo synchronous
spacecrafts with some orbiting satellites to take care
of North/South pole regions.
The INFOSAT Network (Fig 4) is to be an integral
component of the planned worldwide broadband
telecommunication network. Therefore it will be
Fig 4 INFOSAT configuration
285
necessary for planning of the satellite network and the
terrestrial network to be closely coordinated. Until
now satellites have connected the terrestrial networks
of various countries and organizations.
Services to be provided by INFOSAT
Radio and data distribution services to many users
which are spread over a large area. Dara collection
services for large areas with many data transmitter
stations (multi point-to-point operation : weather, oil,
electricity & water meter reading etc.).
• Telecommunication services for thin routes
• Telecommunication services with ships, aeroplanes,
space vehicles, etc).
• In the area of business communications demand is
growing for broad band internet & multimedia
communication facilities which can be applied
flexibly using satellites.
• Worldwide Radio paging
• Video conferencing and high resolution TV
broadcast trials.
• Mobile radio services can be combined with location
finding services and be used for automatically
locating subscribers.
• Earth observation with special warning mechanism
and environmental protection services in close
coordination with terrestrial sensors.
286 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
By integrating INFOSAT into terrestrial network,
it will be possible to supply all the subscribers of a
future global network with all essential information.
This will make the location of the user quite irrelevant.
CONCLUSION
There is a great influence of present day fast
evolving technologies particularly the digital techniques
on the development of the society. Satellite technology
bridges the gap between urban and rural areas in
utilizing the technologies. Inspite of advancement of
technology to a great extent, many parts of India are
still backward and inaccessible by terrestrial means.
Authors
S Pal an alumnus of Birla Institute
of Technology & Science (BITS), Pilani
and Indian Institute of Science (IISc),
Bangalore is presently Outstanding
Scientist & Deputy Director (Digital &
Communication Area) and also Program
Director, Satellite Navigation Program
at ISRO Satellite Centre, Bangalore.
He joined Indian Space Research
Organization (ISRO) in 1971, after a brief tenure at TIFR. Dr
Pal is responsible for starting Antenna & Microwave activities
in the ISRO Satellite Centre at the inception. He has been
responsible for the development & fabrication of all spacecraft
related telecommunication systems for India’s all satellites starting
with Aryabhata to present day IRS & INSAT series of spacecraft.
He has been consultant to: (i) INMARSAT/ICO (UK) for
LEOSAT definitions, responsible for the development of Satellite
Hand Held Telephone Antenna System for ICO (UK)
(ii) International Telecommunication Union (ITU) for defining
Regional African Satellite Communication System and
(iii) Nanyang Technolgical University, Singapore on X-Sat. He
was also Chairman of Information Infrastructure Working Group
of Ministry of Information Technology. Presently he is spear
heading the Indian Satellite Based Wide Area Augmentation
System – GAGAN (GPS Aided Geo Augmented Navigation
System) and Indian Satellite Navigation activities for civil aviation
applications.
He is also on the United Nations panel of experts on Global
Navigation Satellite Systems (GNSS).
Dr Pal a Distinguished Fellow of IETE, Fellow of Indian
National Academy of Engineering, Indian National Academy of
Sciences, Fellow, IEEE(USA), MAMTA(USA), MASI and has
received more than a dozen awards, including, Performance
Excellence Award by ISRO, Distinguished Achievement Award
for launching Aryabhatta, Hari Om Ashram-Vikram Sarabhai
Award, IETE Ramlal Wadhawa & Hari Ramji Toshniwal Awards,
Om Prakash Bhasin Award, four NRDC awards and IEEE Third
Satellite technology connects total country irrespective
of location. This paper briefed about the services
provided by satellites to bring the cutting edge
technology and benefits to the rural and inaccessible
area.
REFERENCES
1. Surendra Pal, Advanced Satellite Communication
Technology-A Perspective, Journal of Spacecraft
Technology, vol 12, no 2, July 2002.
2. Lectures by the author Surendra Pal on Satellite
Communication and various studies carried out in
the area.
Millennium Medal for various developments, inventions and
overall contributions towards the growth of communication and
space technology in India. Besides these he holds Indian,
European & International patents for his various inventions.
Dr Pal has published more than 170 papers in international
and national journals of repute , one book on communications &
guided PhD students. He has been on the editorial board of
various technical journals. He is a Distinguished Visiting Professor
of Indian National Academy of Engineering. His fields of interest
are Space Communications, Information Technology,
Electromagnetics, Antennas and Satellite Navigation.
Dr Pal has been closely associated with IETE and served
IETE in various capacities. He was also a Council Member for
seven years. He was instrumental in starting IETE research
board activities. He has also been guest editor of many IETE
Technical Review issues on Antenna and Microwaves.
* * *
V Sambasiva Rao obtained BE
(Electronics and Communication
Engineering) from College of Engineering,
Kakinada, (Andhra University) in 1973.
He joined ISRO Satellite Centre,
Bangalore in 1974 and presently, heading
Communication Systems Group. He is
responsible for the development of high
bit rate data transmitters for all IRS
satellites and various RF and microwave
systems for IRS and INSAT missions.
Shri Rao is a fellow of IETE and received Distinguished
Achievement Award for launching Aryabhatta, NRDC award
1994 for the development of X-band high bit rate QPSK Modulator
and IETE-IRSI (83) award 2006. He has published more than 30
technical papers in national and international journals.
* * *
 INVITED PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 287-311

A Programmable Built-in Self-Test for

Embedded Memory Cores
SHIBAJI BANERJEE, DIPANWITA ROY CHOWDHURY
Department of Computer Science and Engineering, Indian Institute of Technology,
Kharagpur 721 302, India.
AND

BHARGAB B BHATTACHARYA
Advanced Computing and Microelectronics Unit, Indian Statistical Institute,
203, BT Road, Kolkata 700 108, India.
email: shibaji@vlsi.iitkgp.ernet.in; drc@cse.iitkgp.ernet.in; bhargab@isical.ac.in

A memory test algorithm for detecting neighborhood pattern sensitive faults (NPSFs),
including static NPSF (SNPSF), passive NPSF (PNPSF) and active NPSF (ANPSF), is proposed
in this paper. The patterns can also detect all the traditional faults present in the memory
array such as stuck-at faults (SAFs), transition faults (TFs), coupling faults (CFs) and address
decoder faults. Next, a programmable BIST architecture is designed. The BIST circuit allows
the users to select a vast variety of test algorithms based on their choice. The single BIST
circuit is capable of testing different types of memory cores embedded in SOC. The proposed
BIST circuit is shared among the different memory cores in an SOC. For this purpose, test
wrappers for the shared BIST circuits and the memory cores are designed. Finally, a test
scheduling algorithm is developed to reduce the overall test time.

1. INTRODUCTION issue in managing the BIST circuit, along with test

activation, response collection, and analysis. It provides

T HE continuous decline in cost per gate makes

embedded memory cores more and more popular
for system-on-chip (SOC) designs. Memory on SOC,
is growing rapidly in terms of size and density [1].
However, embedded memories are subject to complex
design rules and they are more prone to manufacturing
defects than any other embedded core in an SOC [2].
Hence, testing of embedded memory is a real challenge
for SOC design and production. For an SOC, the
inability to have direct access to a core is one of the
major problems in testing and diagnosis. Further, the
available bandwidth between the primary I/O pins of
the system chip and the embedded core is usually very
limited. Thus, application of test patterns and
observation of corresponding test responses are
constrained by the bandwidth.
Built-in-self-test (BIST) is considered to be one of
the most cost-effective solutions for embedded memory
testing [3, 4]. In a typical SOC, the embedded memory
may be a single-port SRAM, dual-port SRAM, n-
read-m-write register files, DRAM, or flash memory.
Therefore, test integration has become an important
Paper No 125-D; Copyright © 2007 by the IETE.
287
at-speed and high bandwidth access to the embedded
memory cores. It needs only a low cost ATE to
initialize the tests and to analyze final results. Most of
the previous works on memory BIST employed the
March-based algorithms. Traditional March algorithms
[5] have been widely used in memory testing because
of their linear time complexity, high fault coverage,
and ease in BIST implementation. However, for better
fault modeling, the Neighborhood Pattern Sensitive
Faults (NPSFs) are preferred. The NPSFs are now
becoming more and more important, especially for
DRAMs. However, testing based on such a fault
modeling needs a large number of patterns. In addition,
a complex architecture is required to deliver the test
patterns to the memory cells and to compare the
corresponding test responses. An efficient VLSI im-
plementation is therefore, extremely important. Earlier
approaches to detect NPSFs include the tiling method
[5, 6] and the two-group method [7]. These methods
can detect and locate all NPSFs, but suffer from
having a low fault coverage of stuck-at faults (SAFs),
transition faults (TFs), coupling faults (CFs), and
address decoder faults (AFs). Similarly, March
algorithms can detect and locate all SAFs, CFs, AFs,
but fail to cover all NPSFs.
288 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
Today, an SOC may contain more than 100
memory cores. In order to reduce test cost and shorten
time-to-market, testing time for these cores should be
minimized by adopting an appropriate test access
architecture. However, power dissipation is becoming
a key challenge for the SOC. Therefore, power
consumption of the embedded memory cores has to
be considered during SOC testing. In SOC test
scheduling, multiple cores have been tested concurrently
while satisfying the power constraints during test.
To lower the BIST area overhead by implementing
an efficient architecture, the main challenge lies in
finding a solution, which also reduces the overall testing
time. The main contributions of this paper can be
summarized as follows.
• Introduction of a memory test algorithm, which
can detect all NPSFs, SFs, TFs, and CFs
present in the memory array.
• Implementation of a new programmable
memory BIST. The proposed BIST module
allows testing of memories for various fault
models. A simple test pattern generator for
NPSFs is implemented by employing the regular
structure of Cellular Automata (CA). Hence,
the proposed BIST architecture provides a
very simple and cost effective memory test
solution for handling various fault models. The
area overhead due to proposed BIST is similar
to that of the existing memory BISTs.
• Design of wrapper cells for the embedded
memories, so that they can share a common
BIST circuit. Sharing of BIST by several
memory cores reduces area overhead.
• Power-constrained test scheduling to reduce
the overall testing time.
The rest of the paper is organized as follows. In
section 2, we discuss the related work on memory
BIST, different March algorithms, and their
implementation. Section 3 presents the fault model
used to develop the proposed test algorithm. The
memory test algorithm is proposed in section 4. The
proposed programmable BIST circuit design and the
shared BIST approach are described in section 5 and
section 6 respectively. Section 7 presents the
experimental results and finally section 8 concludes
the paper.
2. RELATED WORK
Recently, several approaches have been proposed
for designing of the BIST circuits. We can classify
these approaches into two categories, the dedicated
BIST module and shared BIST module. The dedicated
BIST module does not reuse the exiting on-chip
resources for testing of embedded memories. The
dedicated BIST can be classified into categories: FSM-
based and micro-code based. The FSM based BIST
circuit [8-12] can generate a single simple pattern or
complex suite of patterns. These types of BISTs are
mainly used to generate a single pattern by adopting a
suitable March algorithm. However, a series of patterns
is required to obtain a better fault coverage. This
makes the design of FSM complex. The FSM based
approach requires large number of flip-flops to
implement the FSM. So, modifying the test algorithms
or memory architecture requires changing of the BIST
design. Micro-code based BIST is a programmable
BIST that also does not reuse the on-chip resources
but, it overcomes the previous limitations. These types
of BISTs are highly flexible for different test structures.
They use a storage element to store the test algorithms.
The micro-based BIST architecture can be found in
[10,13,14]. In [13], authors used two operations per
March element. However, this algorithm was not able
to detect NPSFs and CFs. In [14], authors proposed a
diagnosis process in a much more efficient way.
However, implementations of these architectures
require a certain amount of storage elements, which
leads to increase in the area overhead. In summary,
an FSM-based programmable BIST circuit requires
less area overhead but provides relatively less flex-
ibility, while a micro-code based BIST circuit provides
more flexibility but requires high area overhead.
On the other hand, the shared BIST approach
reuses the on-chip resources, which result low area
overhead. Since, a complex SOC usually contains one
or more processors (or microprocessors), reusing these
existing on-chip processors for testing the embedded
memories can lead to lower area overhead. But, reusing
a processor as a BIST needs much more sophisticated
wrapper cells for both the embedded memories and
processors. In [15], an on-chip microprocessor was
reused for testing the other cores in an SOC. It
eliminated the area overhead caused by dedicated
BIST hardware. In [16], authors developed an SOC
test solution, which includes both software and
hardware based approaches. They introduced a new
BIST approach, called hardware/software co-testing
for embedded memories. An embedded processor-
based built-in self-repair (BISR) design for embedded
memories was proposed in [17]. In [17], authors
removed the controller a typical BISR by reusing the
embedded processors. They also developed a memory
wrapper for this purpose. In [18], a processor-based
BIST design for embedded DRAMs was proposed. In
[18], the authors used two separate instruction storage
 SHIBAJI BANERJEE et al : A PROGRAMMABLE BUILT-IN SELF-TEST FOR EMBEDDED MEMORY CORES 289

memories, combined with flexible address, data, and scheduling problem. While in [27], the authors used
clock generators to perform the test using a minimum network transportation approach to solve the above
of dedicated test pins. The processor-based BIST problem, in [29,33], the authors used a bin packing
design can also be found in [19,20]. However, shared approach, which provides an efficient and effective
BIST techniques require significantly longer testing solution for test scheduling. In [33], the authors pro-
time, which increases the test cost. Again, reusing a posed a restricted 3-D bin packing approach to optimize
processor for BIST may also introduce delay in the the test time under pin and power constraints. A
functional path of the chip. This is not desirable for power-constrained test compatibility graph (P - TCG)
high speed SOC. was created in [24]. Using the P - TCG, the authors
had constructed a set of power-constrained concurrent
Most of the existing BIST techniques have used
test set (PCTSs) to facilitate concurrent testing. They
the March test algorithms. However, these BIST
scheduled the test sets by dynamically partitioning and
techniques are not capable of detecting NPSFs. In
allocating the tests. The memory test scheduling
[21], the authors proposed a modified March algorithm
problem can be found in [16,34]. In [34], it is assumed
that can also detect NPSFs along with other
that the March elements of a March algorithm can be
conventional faults. They used complex data
performed by distinct test resources. The disadvantage
background to detect the NPSFs. The BIST
of this method is that it needs a special control environ-
implementation of this algorithm requires complex
ment, which adds complexity to the test architecture.
control circuits, which increase the area overhead. In
In the present work, we have designed a programmable
[22], the authors have used auxiliary memory to create
BIST. This BIST is capable of testing different memory
the periodic data backgrounds in the memory cell
cores with various sizes and types. The proposed
array for NPSFs. They loaded the desired data
BIST can be shared among the different memory
background by means of a scan path. Thus, gener-
cores in the SOC to reduce the area overhead. The
ation of the periodic data background needs an auxiliary
test wrappers for these shared memory cores are
memory. In addition, complex control circuits are
presented next. Finally, a test scheduling algorithm is
required to deliver the data from the auxiliary memory
proposed to reduce the overall testing time.
to the memory under test. In the present work, we
have proposed a memory test algorithm which removes
3. FAULT MODEL
the requirement of the periodic data background. The
proposed test algorithm can detect the NPSFs, SFs,
TFs, and CFs present in the memory array. A
programmable BIST circuit is then designed for the
embedded memories. The BIST technique supports B E N
the proposed algorithm along with the other existing S W B
March algorithms and allows the users to select a
N S
specific algorithm based on their choice.
W B E
Embedded memories are usually much smaller
than stand-alone memories [23]. This leads to high S
BIST overhead. So, sharing a BIST with different
memory cores in an SOC leads to a cost effective N
solution. However an SOC contains several memories
W B
of different sizes, with different access protocol and
timing. The shared BIST circuit should be able to test
them. Thus, the BIST circuit should be independent of
the type and size of the memory cores and should be Fig 1 The five-cell NPSF
programmed for different types of memory cores. In
other words, a single BIST circuit may test different
types and size of memory cores. Next, an appropriate A pattern sensitive fault (PSF) is a conditional
scheduling algorithm is required to schedule the memory coupling fault in which the content of a memory cell or
cores for the BIST. Several approaches of test the ability to change its content is influenced by a
scheduling for the digital cores in the SOC can be certain bit pattern in other cells in the memory. Here
found in [24-32]. In [26], the authors presented an the data retention and transition of the victim cell are
integer linear programming (ILP) model to formulate affected by a set of aggressor cells. A neighborhood
the integrated wrapper/TAM co-optimization and test pattern sensitive fault (NPSF) is a special case of
290 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
pattern sensitive faults, wherein the influencing
(coupling) cells are in the neighborhood of the influenced
(coupled) cell. The coupled cell is called the base (or
victim) cell and the coupling cells are called the deleted
neighborhood cell. The neighborhood includes all the
cells in the deleted neighborhood as well as the base
cell. The five-cell NPSF as shown in Fig 1 is considered
in the paper. The neighborhood includes the base cell
(B) as well as the deleted neighborhood cells (including
N, E, W and S cells that are physically adjacent to B).
The boundary cells on the memory cell array are
special cases that have only 2 or 3 deleted neighborhood
cells. This fault model can further be categorized into
three classes as follows.
• A static NPSF (SNPSF) occurs if the base
cell is forced to a certain state because of the
appearance of a certain pattern in the deleted
neighborhood. To detect SNPSF, all the 32
static neighborhood patterns shown in Table
1a should be applied.
• A passive NPSF (PNPSF) occurs if the base
cell can not change its state from 0 to 1, or
from 1 to 0 because of the appearance of a
certain pattern in the deleted neighborhood.
To detect PNPSF, all 32 passive neighbor-
hood patterns shown in Table 1b should be
applied.
• An active NPSF (ANPSF) occurs if the base
cell is forced to a certain state when a transi-
tion occurs in a deleted neighborhood cells,
while other deleted neighborhood cells as-
sume a certain pattern. To detect ANPSF, all
the 128 active neighborhood patterns shown
in Table 1c must be applied.
Traditional NPSF test patterns (as shown in Table
1 are used to detect all the NPSFs. In addition to
NPSFs, other faults can also be present in the memory
array. These faults are stuck-at faults, transition faults,
coupling faults, decoder faults and Read/Write logic
faults as described below [35]:
• Stuck-at fault (SAF): A permanent stuck-at-0
fault (SAF0) or stuck-at-1 fault (SAF1) that
may occur in any memory cell.
• Transition fault (TF): A memory cell in state s
(s Î {0,1}) fails to undergo an s to s transition
when s is to be written in the cell. A TF can
be either a rising-transition fault (TF1 or TF-)
or a falling-transition fault (TF0 or TF¯).
• Coupling Fault (CF): During write operation a
transition in one (coupling) memory cell may
affect the value of another (coupled) memory
cell. The coupled cell can have a lower address
than the coupling one (it is indicated as Ù), or
a higher address (Ú). There are three types of
CFs:
a. Idempotent CF (CFid): A transition in
coupling cell forces the coupled cell to
have a certain value. There are eight
subtypes of CFid: Ù(-,0), Ù(-,1), Ù(¯,0),
Ù(¯,1) Ú(-,0), Ú(-,1), Ú(¯,0) and Ú(¯,1).
b. Inversion CF (CFin): A - or ¯ transition
in coupling cell inverts the value of coupled
cell. The following four CFin sub-types
may occur: Ù(-, ), Ù(¯, ), Ú(-, ) and
«
«
«
Ú(¯, ).
«
TABLE 1: Neighbourhood pattern
B 00000000000000001111111111111111
N 00000000111111110000000011111111
E 00001111000011110000111100001111
W 00110011001100110011001100110011
S 01010101010101010101010101010101
(a) Static neighborhood patterns (SNPs)
B ----------------¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
N 00000000111111110000000011111111
E 00001111000011110000111100001111
W 00110011001100110011001100110011
S 01010101010101010101010101010101
(b) Passive neighborhood patterns (PNPs)
B 00000000000000001111111111111111
N --------¯¯¯¯¯¯¯¯--------¯¯¯¯¯¯¯¯
E 00001111000011110000111100001111
W 00110011001100110011001100110011
S 01010101010101010101010101010101
B 00000000000000001111111111111111
N 00000000111111110000000011111111
E ----¯¯¯¯----¯¯¯¯----¯¯¯¯----¯¯¯¯
W 00110011001100110011001100110011
S 01010101010101010101010101010101
B 00000000000000001111111111111111
N 00000000111111110000000011111111
E 00001111000011110000111100001111
W --¯¯--¯¯--¯¯--¯¯--¯¯--¯¯--¯¯--¯¯
S 01010101010101010101010101010101
B 00000000000000001111111111111111
N 00000000111111110000000011111111
E 00001111000011110000111100001111
W 00110011001100110011001100110011
S -¯-¯-¯-¯-¯-¯-¯-¯-¯-¯-¯-¯-¯-¯-¯-¯-¯-¯
(c) Active neighborhood patterns (ANPs)
 SHIBAJI BANERJEE et al : A PROGRAMMABLE BUILT-IN SELF-TEST FOR EMBEDDED MEMORY CORES 291

c. State CF (CFst): The 0 or 1 state of a after 9 independent test operations. The

coupling cell forces the coupled cell to 0 rationate behind selecting 9 cells for each
or 1 state: Ù(0,0), Ù(0,1), Ú(1,0), Ú(1, 1). block lies in the fact that the base cell and its
• Decoder and Read/Write Logic faults can be deleted neighborhood of a block are
modeled as memory array faults and do not independent of the base cell and its deleted
need to be considered explicitly. neighborhood of the other blocks for a particular
test operation. So, the patterns can be delivered
Traditional March algorithms are widely used to in parallel without any modification during each
detect all SAFs, TFs and CFs, but they cover only a test operation.
small percentage of NPSFs [21]. Similarly, test patterns
for NPSFs cover only a small percentage of SAFs, Figure 3 shows the position of the base cells in
TFs, and CFs. So a unified test algorithm is needed each block during the 1st testing operation. Here the
that can detect all the above faults. cells marked by B are the base cells and the subscript
associated with them indicates the block number, to
4. PROPOSED TEST ALGORITHM which they belong. The cells marked by N, E, W and S
are the deleted neighborhood cells. Similarly, cells
The proposed test algorithm will detect all the marked by B´ represent the cells that are neither base
NPSFs, SAFs, TFs, and CFs in an embedded DRAM. cells nor the deleted neighborhood of any base cell.
Before describing the algorithm, the concept of memory
test-bed is introduced first.
Block 1 Block 2 Block 3
• Let us divide the memory into different blocks
as shown in Fig 2 Each block consists of 9 B1 E1 W2 B2 E2 W3 B3 E3
cells if available, otherwise the remaining cells.
S 1 B’ B’ S 2 B’ B’ S 3 B’
As shown in Fig 2. Each block 1 consists of 9
cells, while block 3 consists of 6 cells. During N4 B’ B’ N5 B’ B’ N6 B’
each test, one cell will be selected from each B4 E4 W5 B5 E5 W6 B6 E6
Block 4 Block 6
block as a base cell. Thus, during the 1st S 4 B’ B’ S 5 B’ B’ S 6 B’
testing operation, cell 1 of the each block will
be selected as the base cell and the N7 B’ B’ N8 B’ B’ N9 B’
corresponding deleted neighborhood will be B7 E7 W8 B8 E8 W9 B9 E9
chosen accordingly. Similarly, during the 2nd S 7 B’ B’ S 8 B’ B’ S 9 B’
test all the cells marked by 2 will be selected
as a base cell and so on. But during the 3rd Block 7 Block 8 Block 9
test operation, block 3 has no cell marked by
3, so no base cell will be selected from the
block 3. Thus the entire memory can be tested
Fig 3 Position of the base cells during 1st test operation

Block 1 Block 2 Block 3

1 2 3 1 2 3 1 2 • The test patterns for NPSF are generated by

4 5 6 4 5 6 4 5 using the Gray code as shown in Table 4 (for
deleted neighborhood).
7 8 9 7 8 9 7 8
1 2 3 1 2 3 1 2 The proposed algorithm can generate tests for all
Block 4 Block 6 NPSFs, SAFs, TFs and CFs. It also allows the user to
4 5 6 4 5 6 4 5
choose a specific fault model depending on test
7 8 9 7 8 9 7 8 requirement.
1 2 3 1 2 3 1 2 Option A: It covers all NPSFs, SAFs, CFs and
4 5 6 4 5 6 4 5 TFs.
Block 7 Block 8 Block 9 Ý{WB0}; Ý {WNp1, WEp2, WSp3, WWp4, WB´ p1};

Fig 2 Divide 8´8 memory into 9 blocks Ý{RB0, WB1}; ß {RB1, WB0}; ß {RB0, WB1};
292 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
ß{WNp1}, {WE p2, WS p3, WW p4, WB´ p1};
ß{RB1, WB0}; Ý {RB0}.
Here WN p1 means that a write operation using p1
value as data is performed in the north neighborhood
of the selected base cell. Similarly WB0 (RB0) means
that a write (read) operation using the 0 value as the
data is performed in the selected base cell. The notation
Ý indicates that all the base cells are processed in up
address order, while ß. indicates that all the base cells
are processed in down address order. The total test
operation is divided into 9 sub-operations. During each
sub-operation, one base cell from each block is selected,
and the patterns according to the proposed test algorithm
are applied.
Option B: If the user wants to detect only NPSFs
the test algorithm can be described as
Ý{WB0}; Ý {WN p1, WE p2, WS p3, WW p4}; Ý{RB0,
WB1}; ß {WN p1, WE p2, WS p3, WW p4}; ß {RBl}.
Option C: For detecting all NPSFs, SAFs and
TFs but not CFs, one may use:
Ý {WB0}; Ý {WN p1, WE p2, WS p3, WW p4}; Ý{RB0,
WB1}; ß {WNp1, WEp2, WSp3, WWp4}; ß {RBl}.
Fig 4 Block diagram of the proposed BIST
Option D: To detect only SAFs, TFs and CFs but
not NPSFs, the following test may be used:
Ý {WB0}; Ý {RB0, WB1}; RB1, WB0}; ß{RB0,
WB1}; ß {RB1, WB0}; Ý {RB0}.
5. THE BIST ARCHITECTURE
The present programmable BIST circuit is ca-
pable of testing different types of memories using
various test algorithms. For this purpose, BIST module
uses an instruction register (IR)- The IR can be loaded
serially from the primary input of the chip. Thus one
primary test input pin is required to load the data into
the IR. If a primary test input pin is not available, one
can multiplex a functional primary input pin to load the
IR. If we have more than one BIST circuits in the
SOC, they can share one primary test input pin. So, no
additional primary test input pins are required for these
BISTs. Once a pattern is loaded into the IR, the
control circuit of the BIST decodes and generates the
different control signals based on the pattern. The
pattern loaded into the IR will determine a particular
March algorithm, size, and the type of the memory
under test. The block diagram of the proposed BIST
circuit is shown in Fig 4. It consists of following four
blocks.
 SHIBAJI BANERJEE et al : A PROGRAMMABLE BUILT-IN SELF-TEST
1) An instruction register (IR) which holds the
instruction that allows the user to enter specific
types of test operation for the memory under
test based on their choice.
2) A controller block that generates the different
types of control signals based on the pattern
loaded in the IR.
3) A test pattern generator that supplies the test
patterns for the NPSF model.
4) An output response verifier to analyze the
output response.
As shown in Fig 4 when the pin “Test mode”
becomes high, the memory unit will be isolated and
connected to the BIST circuit. A hierarchical approach
is employed to implement the BIST architecture, the
details of which are described next.
1. Instruction Register
The Instruction Register consists of the following
fields:
• A march type field or seed (S). It determines
which March algorithm is being used during
testing. In the proposed work, we have taken
16 different March algorithms (including the
proposed algorithm) to test the memory. So,
S consists of 4 bits in the present case. In
general, this field has log2n bits, where n is
the total number of algorithms chosen. Thus,
the proposed BIST module allows the user to
enter a specific fault model of their choice.
Some March algorithms and the
corresponding value in the S field are listed
in Table 2.
Algorithms listed in Table 2 are used for bit-
oriented memories. For word-oriented
memories, the same algorithms can be applied
by modifying the data field of the algorithms.
For example, the MAT++ for a bit-oriented
memory is
Ý {W0}; Ý {R0, W1}; ß {R1, W0, R0};
The same MAT + + algorithm for a 4-bit
word-oriented memory can be modified as,
Ý {W0000}; Ý {R0000, W1111}; ß {R1111,
W0000, R0000};
However, in the case of word-oriented memory
some additional faults may appear [38]. One
can also use the test algorithms proposed in
[38] for a word-oriented memory.
FOR EMBEDDED MEMORY CORES 293
TABLE 2: Value of S and the corresponding test
algorithms used in the present work
S Test Algorithms
0000 Option-A of the Proposed Algorithm
0001 Option-B of the Proposed Algorithm
0010 Option-C of the Proposed Algorithm
0011 Option-D of the Proposed Algorithm
0100 March RAW I [36]
0100 March RAW [36]
0100 MAT + + [37]
0101 MARCH A [37]
----- -------------
----- -------------
• A memory size field (MS). It determines the
size of the memory under test. Usually,
memories contain both row and column
decoders. An address multiplexing technique
is used to select the memory cell. The ad-
vantage of the technique is that it needs fewer
address lines to address a memory compared
to the previous technique. A memory with
only one decoder and 8 bit address line is
capable of addressing 28 memory cells, while
a memory with the same size of address line
and two decoders (i.e., row and column
decoders) can address 216 memory cells. As
an SOC may contain both types of memories,
the MS field is divided into two parts. A one
bit field called flag (F) is used to indicate
whether the memory under test has one or
two decoders. When flag is 0, it indicates only
one decoder is present in the memory under
test and a 1 indicates the other option. The
second part of the MS is called size filed
(SO). The SO indicates size of the memory.
The SO is log2n bits long, where n is the size
of the largest memory embedded in the SOC.
So, during the testing of the memories of
smaller sizes the bits from the MSB side of
this SO field will be set to zero.
• A memory type field (MT). It determines the
type of memory under test. It again consists
of two parts. The first part (MT1) determines
whether the memory is SRAM or DRAM
and the second part (MT2) determines whether
294 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

the memory is bit-oriented or word-oriented. 2. Controller

When different types of SRAMs or DRAMs
with different architectures are used in SOC, It consists of the following blocks:
we need to increase the number of bits of
• An address generator that counts the address
MTl. SRAMs with different architectures need
of the cell under test. The proposed controller
different control signals during read and write
consists of two separate address generators,
cycles. To identify the proper control signals
one to generate the row address and other to
and the corresponding timing diagram for a
generate the column address of the cells.
particular SRAM, each combination of MTl
These generators generate all the cell
will be reserved for a particular type of SRAM.
addresses in increasing (Ý) and decreasing
This is also applicable for the DRAMs. In the
order (ß). These generators can be implemented
present work, we have considered only one
by up/down binary counters. During testing,
type of both SRAM and DRAM for SOC. So,
the sequence generator (described later) issues
MT1 consists of one bit. When MT1 is 0, it
clock cycles to the input of the address
indicates the memory is SRAM and for DRAM,
generators. After each clock cycle, the address
MT1 will be set to 1. Similarly, the MT2
generators provide the address of the current
consists of one bit. For bit-oriented memory
memory cell by increasing, decreasing, or
MT2 will be 0 and for word-oriented memory
without changing their previous contents. Two
it will be 1. Table 3 shows the different
up/down signals (one for each address
combinations of MT1 bits when two different
generator) from the sequence controller
types of both SRAM and DRAM are
determine the address order in which these
available.
counters generate the memory addresses.
• The data field (DT). This field will supply the When the memory contains only one decoder
data to be written into the memory cell. It the sequence controller does not issue any
consists of two parts. The first part DT1 clock signal to the column address generator.
determines the total number of bits in word So, column address generator will be ideal.
line of the memory under test. It is log2n bits
• The sequence generator decodes the content
long, where n is the maximum size of the
of the IR and generates different signals to
word line in the SOC. The second part DT2
activate the other blocks of the BIST. It is
determines the value of the word to be used
implemented through a FSM. Based on the
during March sequences. The size of the DT2
pattern of the IR, it selects the current March
field is 2n bits. For a bit-oriented memory,
test algorithm, and generates different control
DT1 and DT2 will be set to 000 ... 00 and
signals to access the memory. It issues clocks
000... 00 respectively. For a word-oriented
to the address generator, selects the TPG for
memory with n1 bit word line (where n1 < n),
NPSFs, and generates control signals for
the n1 bits of the DT2 from the LSB side will
current March test sequence. It generates the
be set to the desired data word. The remaining
flip (FLIP) signal to the XOR gates (Fig 4).
(n – n1) bits from the MSB side will be set to
Based on the FLIP signal, the contents of the
zero.
data register can be complemented. When
FLIP is 1, the complement of the DT2 field of
the IR is loaded into the data register. But
when FLIP is 0, the content of DT2 is loaded
TABLE 3: MT1 Values and corresponding into data register without any modification.
memory cores The FLIP signal is generated based on the
MT1 Memory Type March sequence.
• The data register supplies the data value to be
00 DRAM (Type - 1)
written (read) into (from) the memory. The
01 DRAM (Type - 2) contents of the data register come from the
10 SR AM (Type - 1) DT2 field of the IR. When FLIP is 1, the DT2
is inverted by the set of XOR gates and loaded
11 SRAM (Type - 2)
into the data register. During each write
operation of the test sequences, the data
register provides the data to be written into
 SHIBAJI BANERJEE et al : A PROGRAMMABLE BUILT-IN SELF-TEST FOR EMBEDDED MEMORY CORES 295

the memory. Similarly, during each read
operation it provides the data value expected
to be read from memory. This expected value
is supplied to the comparator of the output
response verifier. For bit-oriented memories,
the LSB of the data register is used to supply
the data during read and write sequences.
3. Test Pattern Generator (TPG)
binary sequence of length (n + 0.5)2n using an n-bit
shift register, an (n – l)-bit counter, (n – 1) two-input
XOR gates, a FF (flip flop) and a two-input AND
gate. In [43] a recursive technique of generating all
SIC pairs is proposed using a structure called
complementation sequence. An iterative design of the
TPG is proposed in [43], which produces exactly n.2n
SIC pairs in optimal time (i.e., n.2n + 1) with less area
overhead.

A TPG is used to generate the patterns for NPSFs.
The total number of 2-pattern tests needed to sensitize
all NPSFs of a neighborhood of size k is
(k – l).2 k–1 . A sequence of all such tests for
k = 5 is needed to sensitize all NPSFs of size five
(Table 4). It can be observed that the test patterns
form a unit distance code, where the two successive
test vectors differ only in one bit position. If the
vectors of a two-pattern test differ in single (multiple)
bit (s), such a pair is known as single (multiple)-input-
change, i.e., SIC (MIC) pair.
For an n-input CUT, there are n.2n SIC two
pattern tests. In pseudo-exhaustive adjacency testing
(PEAT), all such test pairs are generated in (n + l).2n
clock cycles [39]. The TPG proposed in [40] generates
all SIC test pairs in n.2n+1 clock cycles using a
(2n + l)-bit scan shift register, an n-bit LFSR and n
2-input XOR gates. Design-for-testability techniques
suited to BIST applications for robust testing of delay
faults with SIC pairs, appeared in [41], where the
TPG is driven by a Johnson counter. The TPG
proposed in [42], generates all SIC test pairs in a
In SIC test scheme of NPSF testing, a two-
pattern test consists of a vector X1(xn , xn–1, ..., xi,...
,x2,x1) applied to the CUT, followed by another vector
X2(xn,xn–1,...,xi,... ,x2,x1) that differs from X1 by unit
hamming distance in a single bit xi, 1 £ i £ n. To
include all SIC pairs, X must appear at least n times in
S, and exactly n times if S has to be of minimum
length. Furthermore, if (X1, X2) is a SIC test pair, (X2,
X1) must also be a SIC test pair. The sequence S can
thus be described in terms of graph traversal in a
directed graph.
For an n input CUT, all the n.2n SIC pairs can be
represented by a directed graph G(V, E). The graph G
can be obtained by an undirected graph G´(V,E/2),
whose 2n nodes are labeled with n-bit binary vectors
and two nodes v i and v j are adjacent if their
corresponding binary labels lie at unit hamming
distance. The directed graph G can be obtained by
replacing each undirected edge (vi, vj) in G´ with two
directed edges (vi,vj) and (vj,vi). Such a graph for 3
input CUT (i.e., n=3) is shown in Fig 5.

TABLE 4 : Test vector sequence to detect NPSFs

0 0 0 0 0 0 0 0 1 0 1 0 1 0 1 0
0 0 0 1 0 1 0 0 1 0 1 1 1 1 1 0
0 0 1 1 1 1 0 0 1 0 0 1 0 1 1 0
0 1 1 1 1 1 0 1 1 1 0 1 0 1 1 1
1 1 1 1 1 1 1 1 0 1 0 1 0 1 0 1
1 1 1 0 1 0 1 1 0 1 0 0 1 0 0 1
1 1 0 0 0 0 1 1 0 1 1 0 1 0 0 1
1 0 0 0 0 0 1 0 0 0 1 0 1 0 0 0
0 0 0 0 0 0 0 0 1 0 1 0 1 0 1 0
0 0 1 0 1 0 0 0 1 0 0 0 0 0 1 0
0 1 1 0 1 0 0 1 1 1 0 0 0 0 1 1
1 1 1 0 1 0 1 1 0 1 0 0 0 0 0 1
1 1 1 1 1 1 1 1 0 1 0 1 0 1 0 1
1 1 0 1 0 1 1 1 0 1 1 1 1 1 0 1
- - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - -
296 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

In G, there atleast one path between every pair of

vertices, so G is a connected graph. Also G is an Euler
graph. As the graph G is an Euler graph, there will be 0000
some closed walk in G that contains all the edges of
G. The walk is called an Euler line. An Euler line for a
graph of Fig 5 can be obtained by starting from the 0001 0010 0100
node 000, and traversing the edges guided by the
increasing sequence of labels {1,2,3,..., 24}, as shown
in Fig 5. The generation of SIC pairs for a 4 input CUT
and the test patterns for the four deleted neighbourhood
1001 1100 1010 0011 0101 0110
cells in five-cell NPSF model are described below.

1101 1011 1110 0111

1111

Fig 6 Euler graph for 4 input CUT

25 0000
32
8 116 9 17
24
1000 0001 0010 0100
7
18 10 31
26 15 23
2

1001 1100 0011 0110

Fig 5 Euler graph for a 3 input CUT
6 3
18 11 30
14
27 12
The SIC pairs for a four-input CUT is shown in
Table 4. A graph G, representing the unit distance 1101 1011 1110 0111
code of Table 4 is shown in Fig 6. 13 21 12 4
28 5
The original Euler graph G is divided into two 20 29
Euler subgraphs G1 and G2 as shown in Fig 7 and 1111
Fig 8 respectively. Both G1 and G2 contain 32 edges
and they do not have any common edge. The problem
is to find out an Euler line from each sub graph. A
Cellular Automata (CA) based structure is proposed Fig 7 Subgraph G1 of the Euler graph G
to generate the patterns in proper sequence. The CA
with one seed is allowed to run through 32 cycles to
generate the 32 patterns (first 8 columns of Table 4
which is nothing but the Euler line of G1. Next the 4, the Euler line of G2. For the sake of completeness
seed is changed and the same CA structure generates and proper understanding, a brief introduction of CA is
the remaining 32 patterns (the last 8 columns of Table given below.
 SHIBAJI BANERJEE et al : A PROGRAMMABLE BUILT-IN SELF-TEST FOR EMBEDDED MEMORY CORES 297

D Q D Q D Q D Q
1000 0001 0010 0100
7 Cleck
26
10 27 5
23 9 6 CL CL CL CL
24 8
21 12
11 25 28 CL-Combinational Logic
1001 1100 1010 0010 0101 0110
31 20
16 13
2 1 Fig 9 An one-dimensional CA
14
30 19
3 17 32
18 cells are arranged in a two-dimensional plane.
4 15
1100 1011 1110 0111 Additionally, to achieve dynamic behavior (discussed
later) the connection order of the top row depends on
a signal named MODE.

When MODE = 0, the top row follows rule 170

Fig 8 Subgraph G2 of the Euler graph G with periodic boundary condition, i.e., it implements
left shift function. Only one cell of the top row contains
a 1 at any particular instant. This row basically contains
A Cellular Automation (CA) consists of a number the information regarding which bit position is changed
of cells arranged in a regular manner, where the state in two successive patterns begin generated.
transitions of each cell depends on the states of its
The lower bit plane is driven by the outputs of the
neighbors. The next state of a particular cell is assumed
top row cells. A particular cell in the bottom row
to depend only on itself and on its two neighbors (3-
changes state when the corresponding neighboring
neighborhood dependency). The state x of the ith cell
cell of the top row contains 1. Thus the characteristic
at time (t + 1) is denoted as xit+1 = f(xti–1, xit, xti+1),
matrix of the 8-cell CA is,
where xit denotes the state of the ith cell at time t and
f is the next state function called the rule of the
automata [44]. Since f is a function of 3 variables, æ0 1 0 0 0 0 0 0 ö
there are 223 or 256 possible next state functions. The ç ÷
ç0 0 1 0 0 0 0 0 ÷
decimal equivalent of the output column in the truth ç ÷
table of the function is denoted as the rule number. ç0 0 0 1 0 0 0 0 ÷
The next state function for different rules are stated ç ÷
ç1 0 0 0 0 0 0 0 ÷
below as examples: T1 = ç ÷
ç1 0 0 0 1 0 0 0 ÷
Rule 90 : xit+1 = xti–1 + xti+1 ç ÷
ç0 1 0 0 0 1 0 0 ÷
ç ÷
ç0 0 1 0 0 0 1 0 ÷
Rule 150 : xit+1 = xti–1 + xti + xti+1 ç ÷
è0 0 0 1 0 0 0 1 ø
Rule 51 : xit+1 = xti
It can be shown that T18 = I, which implies that this
Rule 153 : xit+1 = xti + xti+1 CA is a group CA of length 8.

When MODE = 1, We can consider another

Rule 195 : xit+1 = xti–1 + xti+1
In the next state functions + denotes bitwise XOR.
An example of a four cell 1-D CA is shown in Fig 9.
An 8-cell CA, as shown in Fig 10 is used to
generate the code sequence depicted in Table 4. Here,
characteristic matrix T2. In this case the top row
implements a circular double left shift register, i.e.,
each bit in the top row is shifted circularly in two left
bit positions. The connection order in the bottom row
remains unchanged. T2 is shown below (i.e., when
MODE = 1)
298 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

CL CL
D Q a7 D Q a6
D Q a5 D Q a4
clk clk clk
clk
clk clk clk

a3 a2 a1 a0
D Q D Q D Q D Q
clk clk clk

Fig 10 CA to Generate the test patterns

æ0 0 1 0 0 0 0 0 ö < a7,a6,a5,a4,a3,a2,a1,a0 > = < 00011010 > is chosen

ç ÷
ç0 0 0 1 0 0 0 0 ÷ as initial seed. The two cycles of length 32 generated
ç ÷ by the 8-cell CA are shown in Table 5, where the test
ç1 0 0 0 0 0 0 0 ÷
ç ÷ pattern < a3,a2,a1,a0 > is taken out from the bottom
ç0 1 0 0 0 0 0 0 ÷ row of the CA. In G1, Euler line starts from the vertex
T2 = ç ÷ 0000 and after traversing 32 different edges it finally
ç1 0 0 0 1 0 0 0 ÷
ç ÷ terminates at 0000. The order of traversal of each
ç0 1 0 0 0 1 0 0 ÷ edge in G1 is represented by an integer associated
ç ÷
ç0 0 1 0 0 0 1 0 ÷ with them (shown in Fig 7). Similarly for G2, Euler line
ç ÷ starts from the vertex 1010 and after traversing 32
è0 0 0 1 0 0 0 1 ø
different edges it finally terminates at 1010.

It can be shown that T28 = 1, i.e., it is a group CA
of length 8.
Here, the CA is made to change dynamically with
time by changing its characteristic matrix. The
composite characteristic matrix is chosen as T17T2, so
that the CA is run for 4 subcycles, where in each
subcycles the characteristic matrix is T1 for 7 time
steps and T2 for 1 time step. So the CA forms a group
CA of length (7 + 1) 4 = 32 and generates 32 different
patterns.
Thus the CA generates 32 patterns in a single
cycle. If the cycle is run with a new initial state then
another set of 32 patterns can be generated. By
selecting suitable starting seed, the required 64 patterns
indicated in Table 4 can be generated. An Euler line
can be obtained in G1 with < a7,a6,a5,a4,a3,a2,a1,a0 >
= < 00010000 > is chosen as initial seed, while for G2
Table 5 indicates that when MODE = 0, the top
row implements a single left shift function and when
MODE = 1, it implements a double left shift function.
Initially MODE signal is made to 0. After 7 clock
cycles it becomes 1 and in the next clock cycle it
again resets to 0. The MODE signal can be
generated by AND-ing the output bits of a 3 bit up-
counter. The technique of generating MODE signal
is shown in Fig 11. The counter is initially set to 000
state and after 8 clock cycles it again returns to 000
state. The counter is fed by the same clock as used
by the CA.
Moreover, for the loading of the next seed 00011010
we see from the Table 5 that the end of the first cycle
of length (7 + 1) 4 = 32 can be detected by observing
the pattern < a7,a6,a5,a4,a3,a2,a1,a0 > = < 00010000 >.
Thus for 4 bit, the present technique requires 24.2 = 64
clock cycles to generate all the SIC pairs.
 SHIBAJI BANERJEE et al : A PROGRAMMABLE BUILT-IN SELF-TEST FOR EMBEDDED MEMORY CORES 299

TABLE 5: The generated test patterns

Initial Loading : 00010000

Mode a 7 a6 a5 a4 a3 a2 a1 a0 Mode a7 a6 a5 a4 a3 a2 a1 a0

0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 1 0 0
0 0 0 1 0 0 0 0 1 0 0 0 0 1 1 1 0 0
0 0 1 0 0 0 0 1 1 0 0 0 1 0 1 1 0 1
0 1 0 0 0 0 1 1 1 0 0 1 0 0 1 1 1 1
0 0 0 0 1 1 1 1 1 0 1 0 0 0 1 0 1 1
0 0 0 1 0 1 1 1 0 0 0 0 0 1 0 0 1 1
0 0 1 0 0 1 1 0 0 1 0 0 1 0 0 0 1 0
1 1 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 1 1 0 0 0
0 0 1 0 0 0 0 1 0 0 0 0 1 0 1 0 0 1
0 1 0 0 0 0 1 1 0 0 0 1 0 0 1 0 1 1
0 0 0 0 1 1 1 1 0 0 1 0 0 0 1 1 1 1
0 0 0 1 0 1 1 1 1 0 0 0 0 1 0 1 1 1
0 0 1 0 0 1 1 0 1 0 0 0 1 0 0 1 1 0
0 1 0 0 0 1 0 0 1 1 0 1 0 0 0 1 0 0
1 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 0
0 0 1 0 0 0 0 0 0

Initial Loading : 00011010

Mode a 7 a6 a5 a4 a3 a2 a1 a0 Mode a7 a6 a5 a4 a3 a2 a1 a0

0 0 0 0 1 1 0 1 0 0 1 0 0 0 1 1 1 0
0 0 0 1 0 0 0 1 1 0 0 0 0 1 0 1 1 0
0 0 1 0 0 1 0 0 1 0 0 0 1 0 0 1 1 1
- - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - - - - - -

In [45], an 8-cell CA structure was used to

reset generate the test patterns for NPSF faults. However,
72 clock cycles were necessary to generate 64 patterns.
To obtain the dynamic behavior, two distinct clock
S2 signals were also used. In contrast, the proposed
3-bit Up
Counter S1
MODE
approach requires exactly 64 clock cycles to generate
S0 all the 64 2-pattern tests, and only one common clock
is used.
clk
4. Output Response Verifier
S2 S1 S0 MODE
0 0 0 0
The verifier consists of two blocks: (i) the
0 0 1 0 comparator, and (ii) the fault indicator (FI).
0 1 1 0
0 1 1 0 (i) Comparator: The comparator is used to verify
1 0 0 0 the output responses of the memory under test. It does
1 0 1 0
1 1 0 0 not introduce any aliasing and is simpler than a signal
1 1 1 1 analyzer. It checks the output of the memory cell with
0 0 0 0 the expected value supplied by the data register of the
• • • • controller. If any mismatch is observed, an error signal
• • • •
is sent to FI. The comparator consists of a data
control register and a set of XOR, AND, and OR
Fig 11 The technique to generate MODE signal gates. The comparator is capable of comparing the
300 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

data independent of the size of the data line coming logic circuit used to generate the content of the control
(read) out from the memory i.e., the comparator is register is shown in Fig 13 where the DT1 field is
common for both bit-oriented and word-oriented assumed to be 2-bit long.
memories. Figure 12 shows the block diagram of the
(ii) Fault Indicator (FI): On getting an error
comparator. The data control register indicates the
signal from the comparator, the FI sends the controller
number of bits of the data (read out from the memory
and the other units into the wait state by issuing a
under test) to be compared. The content of the data
BIST error signal (BE). It also sets the ERROR port
control register is determined by the DT1field of the
of the BIST. At the same time, it sets the status of the
IR. The size of the data control register is equal to the
current operation in the EOP (Error Operation) register.
DT2 field of the IR. For bit-oriented memories, the
The error operation includes the failed operation i.e.,
content of the data control register is 000... 01, which
the position of the failed read operation in the read-
indicates that only the last bit of the data register is to
write sequence during which the fault is detected and
be considered during comparison. For a bit-oriented
the address of the faulty cell. For example, the proposed
memories, as shown in Fig 12, the outputs of the AND
test algorithm has 5 read operations (maximum) in the
gates An–1 to A1 and A´n–1 to A´1 zero while the outputs
read-write sequence; so 3 bits (indicated as “session”)
of the AND gates A´0 and A´0 are depend on the data
in Fig 14 are needed to indicate the failure. The FI
read out from the memory and D0 (LSB of the data
then sends the content of EOP register to the user
register) respectively. Thus, the outputs of the all XOR
through ERROR_ADDR port of the BIST. When the
gates are zero except X0 gate whose output will be
content of EOP is completely shifted out, the FI waits
determined by D0 and the data read out from the
for the user input. When the user sends a pulse
memory. For a word-oriented memory, the number of
through CONT_U port, it issues a continue signal
bits from the LSB side of the data control register set
(CONT) and resets the BE signal to allow other units
to one indicates the size of the word line. Thus, for a
to continue the execution. This has an advantage that
word-oriented memory of word size n1 (n1 < n), the
the user can stop the test on every fail.
n1 bits of the data register from the LSB will be
considered during comparison.
6. SHARED BIST APPROACHES
The present comparator is independent of the size
of the word line of the memory under test. The content In current scenario, an SOC may contain more
of the data control register is determined from the than 100 memory cores. In order to reduce the test
DT1 field of the IR by using some logic circuit. The cost and shorten time-to-market, the testing time for

Data Control Register Data Register

MSB LSB Dn–1 D1 D0

Data coming
from the
Memory under
A0 A n–1 test
A n–1 A1 A0
A1
X n–1

Output X1

X0

Comparator

Fig 12 Logic circuit to generate the content of the data control register
 SHIBAJI BANERJEE et al : A PROGRAMMABLE BUILT-IN SELF-TEST FOR EMBEDDED MEMORY CORES 301

DT1 field of IR
1 DT1 DT0 DC3 DC2 DC1 DC0 Decision
DT1 DT0
0 0 0 0 0 1 Bit-oriented Memory
0 1 0 0 1 1 Word-oriented Memory
(2 bits word line)
1 0 0 1 1 1 Word-oriented Memory
(3 bits word line)
DC3 DC2 DC1 DC0 1 1 1 1 1 1 Word-oriented Memory
(3 bits word line)
Data Control Register

Fig 13 Logic circuit to generate the content of the data control register

subject to: Spi £ Pmax, where i are the number of

3 bits n bits a bits
Session Row Address Column Address
Fig 14 The EOP register format
these cores should be minimized by adopting an
appropriate test scheduling algorithm. However, power
dissipation is becoming a key challenge for the SOC.
Therefore, the testing power consumption of the
embedded memory cores should be considered during
SOC testing. In case of SOC test scheduling with
power constraint, multiple cores have been tested
concurrently to reduce the overall testing time. In the
present work, we develop a greedy algorithm for
optimally assigning memory cores to the available test
resources for a given power constraint.
The parameter used in the algorithm are given
below,
Pmax power constraint during test;
n total number of memories;
m total number of test resources (i.e., total
number of available BIST);
pi maximum power dissipation for memory i;
ti testing time of ith memory;
The constraint power test scheduling problem can
be stated as follows.
objective: Minimize T = max Sni=1 ti,
memories currently under test.
The test time ti of a memory core is proportional
to the size of the memory and the number of March
elements in the March algorithm. In the present case,
we have used different March algorithms (listed in
Table 2) for the memory cores under test based on the
user requirement. The test time ti for ith core can be
written as,
ti = miai / bi
where ai is the size and bi is the number of bits per
word of the ith memory, mi depends on the number of
March elements of the March algorithm used to test
the ith rnemory. In order to share the BIST circuit
among the different memory cores, wrapper circuits
are needed to provide the interface between the
memory and the BIST circuit. In the present work, we
have used TAM switches and multiplexers to design
the memory and BIST wrapper circuits. The wrapper
circuits and the proposed test scheduling algorithm are
described in the following two subsections respectively.
A. Designing of the wrapper circuits for
memory and BIST circuits
In order to deliver the test patterns to the em-
bedded memory, each memory is surrounded by a
memory wrapper. It can be used to transport test
stimuli from the test pattern source to the embedded
memory and to transport test responses from the
memory to the test pattern sink. In the present work,
memory wrapper is implemented by using a special
Test Access Mechanism (TAM) switch as shown in
Fig 15 A TAM switch is a synthesizable RTL core
that can be instantiated in a design to provide test data
access to embedded cores in the SOC [46]. The test
can be either deterministic vectors (generated by an
302 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

ATPG) for a scan-based core, or can be pseudo-
random vectors generated by a BIST or can be
functional test vectors.
The TAM switch has two useful modes of operation
namely, cross and pass modes as shown in the Fig 16
and Fig 17. In the cross configuration, the switch is
configured to send the test patterns to the CUT. In the
pass configuration, the switch simply passes the test
patterns to the next switch.

Fig 15 Test Access Mechanism

Fig 16 Cross configured TAM switch

As shown Fig 15, the TAM switch is a
programmable crossbar switch that allows efficient
delivery of test vectors to embedded cores at varying
bandwidth. It consists of an N ´ N switch matrix
where N is the number of input and output lines. The
switch configuration can be programmed by serially
loading the input-to-output of test data that assigns
program into a configuration register. The register
decides which input will be connected to which output
under control of the mode signal (mode = high for
shifting in). There is an update register that loads the
configuration when the mode is low. The value of the
update register decides the connection between input
and output.
The switch canbe programmed by serially loading
the N * (log2 (N))-bit long configuration register. The
configuration register can be logically partitioned into Fig 17 Pass configured TAM switch
N segments, where each segment is log2 (N)-bit wide.
Each segment i corresponds to an output data line Y(i)
with the least significant segment corresponding to In the present work the memory wrapper is de-
Y(0) signal and the most significant segment (N – 1) signed by using a TAM switch. The schematic diagram
corresponding to Y(N – 1). The content k of a segment of a memory wrapper is shown in Fig 18. The size of
i implies that the input data signal A(i) should be the TAM switch depends on the width of the address
connected to output signal Y(k). The test input pins of input (addr) and control input of the memory core. The
the SOC are connected to some input pins of the test mode of the memory is selected by the “Test
TAM switch. These input pins are then connected to Mode” signal. In the test mode, the addr, control, and
some output pins of the TAM switch internally according Din inputs of Ml receive signal from the BIST circuit
to the value loaded in configuration register. These through the TAM switch, which is configured in the
output pins are then connected to the scan pins of the cross mode. As the TAM switch of M2 is in the pass
core to be tested. The scan output pins of the core are mode (Fig 18), it passes the test patterns from BIST to
then routed to the remaining input pins of the switch. Ml and responses from Ml to BIST. The corresponding
 SHIBAJI BANERJEE et al : A PROGRAMMABLE BUILT-IN SELF-TEST
responses are sent to BIST circuit through the Dout pin
and the TAM switch. The BIST then compares the
observed responses with the expected responses. When
a fault is detected, the BIST sends the faulty cell
address and the order of the read operation in the
read-write sequence of the March algorithm. The
BIST core is also encapsulated by a wrapper cell,
which is similar to the memory wrapper cell. The size
of TAM switch for the BIST wrapper cell, depends on
the number of input and output pins of the BIST.
B. Proposed test scheduling algorithm
It is observed that sharing a BIST with large
number of memories, located at different regions in
the chip results in a certain routing overhead. Sharing
a BIST with two memory cores separated by a large
distance leads to high routing overhead. So, in this
case sharing is not desirable. The long routing wire
length gives rise to many unwanted effects in deep
sub-micron technology which degrades the
performance of the chip. In the present work, we
divide the memory cores into different partitions based
on their location in the chip. This location is determined
by the functional proximity between the memory cores
and the other cores in the SOC. Thus an approximate
location of each memory core can be found before
layout [47]. Based on these approximate locations, we
divide the memory cores into different partitions so
Fig 18 Memory wrapper cell
FOR EMBEDDED MEMORY CORES 303
that, memory cores nearer to each other will fall in the
same partition. Next, we assign a certain number of
BIST circuits to each partition based on the total
number of available BIST circuits. In the present
work, we have allotted two BIST circuits to each
partition. For example, if the total number of available
BIST is 10, then we divide the memory cores into 5
partitions and assign two BIST circuits to each partition.
The test structure thus obtained is guaranteed to offer
a good trade-off between the minimum test routing
overhead and minimum test time. The scheduling
algorithm is then used to schedule the memory cores
for the BIST circuits available in a partition. The
scheduling algorithm is described below.
We first explain the scheduling algorithm with two
BIST circuits. Later, we will extend the same algorithm
with some modification when more than two BIST
circuits are available. In order to illustrate the proposed
test scheduling algorithm, we consider an SOC with
six memory cores as an example. Each core is provided
with total test cycles needed to test it and maximum
power dissipation during testing. These parameters
are listed in Table 6. In addition, in the present case,
we assume that the Pmax is 100 mw.
The power-constraint test-scheduling problem can
be solved by finding compatible memory cores based
on test power dissipation. To determine the compatible
304 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
TABLE 6: Data for an example SOC containing
memory cores
core Test maximum power
cycles dissipation (mw)
Ml 800 20
M2 600 80
M3 2000 60
M4 1500 50
M5 1200 50
M6 1000 20
memory cores based on power dissipation, a
compatibility graph G(V, E) is constructed from the
test set (which consists of the time required to test the
core and the test power dissipation). In the compatibility
graph, a vertex corresponds to a core and an undirected
edge between two vertices exists if and only if the
total power dissipation of the two vertices does not
exceed the maximum power dissipation limit. Or in
other words, two cores connected by an edge can be
tested in parallel without violating the maximum power
dissipation constraint. In additional to that, we assign a
weight to each vertex, which represents the time
needed to test the corresponding vertex. This type of
graph with the consideration of power dissipation as a
constraint is called power-constrained test compatible
graph (P – TCG) [24]. The P–TGG for the SOC of
Table 6 is shown in Fig 19a.
In P – TCG, we first schedule the nodes having
larger test time. This will reduce the overall test time
of the circuit. For this purpose, we first consider the
core with maximum test time (assume core C1). Next,
consider the nodes which are adjacent to it. Two
nodes are adjacent if and only if they are connected by
an edge. Among the adjacent cores, we consider the
node having largest test time (assume core C2). After
finding C2, we modify the weight of C1 by subtracting
its weight from the weight of C2 and assign this
modified weight to C1. We then delete the node C2
and all the edges incident on it from the P–TCG. If in
the modified P–TCG, the weight of C1 is non zero and
edge (s) still exists from C1 to the other core C1 will
be considered again for scheduling. But if
after deleting C2, C1 becomes an isolated node in the
P – TCG, C1 can not be scheduled further and will be
deleted from P – TCG. After deleting C1, the node
with maximum degree in P – TCG will be considered
next for scheduling. This process will continue until
the P – TCG contains only isolated nodes. The proposed
algorithm can be explained with the SOC listed in
Table 6. The P – TCG of the SOC is shown in Fig 19a
and designated as G(V,E). How the algorithm works
is described below.
In G(V, E), after examining the weights of the
each node it can be found that M 3 be the node having
maximum weight. So, M 3 will be considered first. Let
B1 and B2 be the two BIST circuits that are available.
So, M 3 can be assigned to any of them. Let M 3 be
assigned to B1. During this time (time to test M 3) the
other BIST circuit B2 is ready to test other cores. But,
among the M 1, M 2, M 4, M 5, and M 6, B2 can be
allocated to a core if and only if the total power
dissipation of the allocated core and the M 3 does not
exceed the maximum power dissipation limit. Thus the
nodes which are adjacent to M 3 can only be assigned
to B2. In Fig 19a both M 1 and M 6 are adjacent to
M 3. Between M 1 and M 6, M 6 has a larger test
time. So, M 6 is assigned to B2. The test time for the
M 6 is 1000. Thus during this time limit (i.e., 0 to 1000
test time) M 3 and M 6 will be tested concurrently by
B1 and B2 respectively. After 1000 time unit, B2 will
finish its job while B1 will be still busy with M 3. Thus
after 1000 time unit, we can assign another core to B2
under the maximum power dissipation limit. This leads
to minimize the total test time. For that we delete M 6
and all the edges incident on M 6 form G to obtain a
new subgraph G1 of G. In G1, a new weight is
assigned to M 3 by subtracting the weight of the M3
from weight of M 6 G. The resultant subgraph G1 thus
obtained is shown in Fig 19b. As the weight of the M 3
in G1 is not zero (i.e., B1 is still operating), we again
consider the node M 3 and find the adjacent node
of M 3 having maximum weight. In Fig 19b node M 1
is only connected to M 3. So, M 1 is now assigned to
B2. After that, we delete the node M 1 and all the
edges incident on M 1 from G1 to obtain the subgraph
G2 of G1. In G2, we subtract the weight of the M 3
form the weight of M 1 to obtain a new weight for
M 3. The subgraph G2 thus generated is shown in
Fig 19c.
In G2, it is observed that both the node M 2 and
M 3 become isolated nodes and so, they cannot be
tested concurrently with the other cores. So, after
completing the testing of M 3, B1 (or B2) can be
allocated to M 2. The other BIST circuit B2 (or B1)
will be ideal during the testing time of M 2. Both the
nodes are then deleted from G2. The subgrah G3 of G2
thus obtained is shown in Fig 19d. In G3, so far no
nodes are scheduled. So, the node M 5 (having
maximum weight in G3) is considered first and assigned
 SHIBAJI BANERJEE et al : A PROGRAMMABLE BUILT-IN SELF-TEST
Fig 19
to B1. M 5 is connected to M 4 by an edge in G3. So,
M4 is assigned to B2. The node M 5 and edges
incident on it are deleted from G3 to obtain subgraph
G4. In G4 the weight of M5 is changed by subtracting
its weight from the weight of M4. The subgraph G4 is
shown in Fig 19e. The subgraph G4 contains only one
node and so scheduling process is ended here. The
final allotment of the cores to B1 and B2 is shown in
Fig 20.
We consider another SOC as an example whose
P – TCG (G´ (V´,E´)) is shown in Fig 21a. The number
of cores and the parameters of each core are kept the
same as in the previous one except the test time of the
core M 1 and M 6 are now changed to 1000 and 1200
respectively. The change in scheduling process due to
these is described below.
As in the previous case, the node M 3 has the
maximum test time in G´ and between M 1 and M 6,
M 6 has larger test time. So, M 3 and M 6 are assigned
to B1 and B2 respectively. The node M6 and edges
4300
B2 M6 M1 M5
B1 M3 M2 M4
1000 2000 3000 4000
Time
Fig 20 Allotment of the memory cores of Fig 19a to
B1 and B2
FOR EMBEDDED MEMORY CORES 305
incident on M 6 are deleted from G´ to obtain G´1 (Fig
21b). In G1´ the new weight of the M 3 is obtained by
subtracting its weight from the weight of M6. In G´1,
M 1 is connected to M 3. So M 1 is assigned to B2.
But, in G1´ the weight of the M 1 is greater than the
weight of the M 3. So, in this case M 3 will be deleted
from G´1. The node M 3 and the edges incident on it
are deleted from G´1 to obtain G´2. In G´2, the weight
of M 1 is obtained by subtracting the weights of M 1
and M 3. The subgraph G1´ is shown in Fig 21c. The
remaining scheduling processes are same as in the
previous one and are shown in Fig 21d, and Fig 21e.
Finally, the allotment of the cores to B1 and B2 is
shown in Fig 22. The outline of the Algorithm_l is
presented in Fig 23.
Now we generalize the algorithm for any number
of available BIST circuits. Here instead of one weight,
two sets of weights W and P are assigned to each
node in P – TCG. The sets of weights are determined
from the test time and the power dissipation of each
node. Let n and m be the number of nodes in P – TCG
and the number of available BIST circuits respec-
tively. The weights assign to ith node will be {w1i,w2i
,...,wmi} and {p1i,p2i,....,pmi}, where w1i and p1i are
the test time and power dissipation of node i
respectively. Initially, other elements of Wi and Pi are
set to zero (i.e., w2i = 0, w3i = 0,..., wmi = 0, p2i = 0, p3i
= 0,..., pmi = 0).
In P – TCG, consider the node having maximum
w1 value. Let Mi be the node having maximum w1.
Next, find an adjacent node of Mi with maximum w1.
We assume it is Mj. After getting Mj, modify Wi =
{w1i, w2i ,..., wmi} of Mi by replacing w2i with w1j and
Pi = {p 1i ,p 2i,...,pmi } by replacing p2i with p 1j
respectively. After assigning Mj to Mi (i.e., to B2),
306 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
Fig 21
4300
B2 M6 M1 M5
B1 M3 M4 M2
1000 2000 3000 4000
Time
Fig 22 Allotment of the memory cores of Fig 21a to B1
and B2
power dissipation of Mi is increased to p1i + p2i + ... +
pmi = p1i + p2i = p1i + p1j . We delete the node Mj and
the edges incident on Mj from P – TCG. The edges of
Mi are redrawn based on the modified power dissipation
of Mi (which is now p1i +p1j). The edges of Mi, which
do not satisfy the power-dissipation limit, will be deleted
from P – TCG. After scheduling of Mi and Mj, another
adjacent core of Mi having maximum w1 is taken and
assigned to the next zero position of Wi. The process
will continue until all the elements of Wi become non
zero or Mi becomes isolate. In the latter case, we will
delete Mi from P – TCG, assign the nodes to B1, B2,
..., Bm based on Wi, and start a new scheduling
process for the remaining cores.
In the former case i.e., when Wi = {w1i,w2i ,...,wmi}
be a set of all non zero elements, we need to find the
smallest element in Wi and subtract the number from
all the elements of Wi. Let the smallest number be wki.
The new value of Wi will be {(w1i – wki),..., (wki –
wki),...., (wmi – wki)}. After modifying Wi, we modify
the set Pi by subtracting pik (corresponding to wik in
Wi) from other elements of Pi. The modified set Pi will
be {(p1i –pki) ,... (pki – pki) ,..., (pmi – pki)}. After
modifying Wi and Pi, if Wi contains any non zero
element, Mi will be considered again in the scheduling
process. Otherwise, Mi will be deleted from the P –
TCG. In the previous case, we redraw the edges
incident on Mi (as Pi is modified) based on the new
values of Pi. Among the adjacent nodes of Mi, the
node (say Ml) with maximum w1l is assigned in the
place of kth element of Wi and Pi. Thus, wki and pki of
Mi will be replaced by w1l and p1l respectively. If w1l
is greater than all the elements of Wi, replace the
weights Wl and Pl by Wi and Pi respectively and delete
the node Mi and the edges incident on it from P –
TCG. Now in Wl , we need to find the smallest
element and continue until all the nodes in P – TCG
are scheduled. In other case, delete the node Ml the
edges incident on it from P – TCG, and find the
smallest element of Wi. The process continues until all
the nodes in P – TCG are scheduled.
It is clear that as the number of available BIST
circuits increases, the time complexity of the algorithm
increases. In the present work, we divide memory
cores into different partitions based on their location in
the SOC (as described before) and assign two BIST
circuits to each partition. The test structure thus obtained
provides less routing overhead, reduced overall testing
time and computation time.
7. EXPERIMENTAL RESULTS
The BIST design has been verified on a SUN
SPARC ULTRA - 60 workstation in SOLARIS 5.8
environment. Five different BIST circuits have been
designed for five different memory cores. The BIST
circuits are synthesized using Design Compiler tool of
Synopsys in 0.18µm CMOS library (provided by the
National Semiconductor, USA). Data obtained after
the synthesis process is shown in Table 7. Column 1 in
Table 7 represents the size of the memory cores for
which BIST circuits have been designed. Column 2
 SHIBAJI BANERJEE et al : A PROGRAMMABLE BUILT-IN SELF-TEST FOR EMBEDDED MEMORY CORES 307

Algorithm_1: power-constraint test-scheduling

Input to the Algorithm_1: (a) Total number of memory cores. Let they be M1, M2,... Mn.
(b) Total test time and test power dissipation of each core.
(c) Pmax of the SOC.
(d) B1 and B2 are two available BIST circuits.
1. Generate the P – TCG G(V,E) using the information given in (a), (b), and (c).
2. Find the vertex with maximum weight. Let it be Mi
3. Assign Mi to B1.
4. Find the cores adjacent to Mi. Among them assign the core Mj to B2 which has maximum
weight.
5. If the weight of Mi is greater than Mj, remove Mj and the edges incident on it from G. In new
G thus obtained, modify the weight of Mi by subtracting its weight from the weight of Mj. Go
to step – 6. On the other hand, if weight of Mi is less than weight of Mj, remove Mi and the
edges incident on it from G. In new G, modify the weight of Mj by subtracting its weight from
the weight of Mi and make i = j.
6. Check the weight of Mi. If it is zero, remove Mi and the edges incident on it (if any) from G.
Check whether Mi is an isolated node or not. If Mi is an isolated node remove Mi from G. If
any of these two conditions is satisfied go to step – 7, otherwise go to step – 4.
7. Is there any isolated node in G? If so, assign it to any BIST circuit and remove it from G.
8. Check whether G is a null graph or not. If no node present in G then the scheduling process
is over, otherwise go to step – 2.

Fig 23 Algorithm_1 to schedule the memory cores

and column 3 represent the number of gates and area To verify the correctness of the proposed test
overhead due to BIST for each of these memory scheduling algorithm, several experiments have been
cores respectively. The results indicate that the area carried out. The scheduling algorithm is applied to 50,
overhead due to BIST is low and it decreases when 100, and 150 embedded memory cores. Table 8 to
the size of the memory core increases. Table 10 show the results of testing time due to the
scheduling algorithm with the number of available
BIST circuits for different number of memory cores.
Each table corresponds to a particular maximum power
TABLE 7: Experimental results for different constraint. For Table 8 maximum power limit is
BIST circuits
considered as 250 mW and the power consumption of
memory BIST gate BIST area each core is randomly assigned from 50 mW to 150
size count overhead mW. Similarly, for Table 9 (Table 10) maximum
64K ´ 1 5890 8.24% power limit is taken as 500 mW (750 mW) and the
power consumption of each core is randomly assigned
128K ´ 1 6958 5.04%
from 100 mW to 300 mW (200 mW to 500 mW). The
256K ´ 1 8235 3.05%
memory cores included in this experiment are of
64K ´ 4 7894 2.92% different sizes with the number of address lines randomly
128K ´ 4 9885 1.85% assigned from 7 to 20 and bits per word are also taken
randomly between 1, 2, 4, 8, 16, and 32.
308 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

TABLE 8:

Maximum Power Constraint 250mW

Number Number of Number of Number of

of cores=50 cores=100 cores=150
available
BIST circuits (m) test time R% test time R% test time R%

1 1335722 — 2502328 — 3388296 —

2 825246 38.22 1572016 37.18 2005864 40.80
4 507838 61.98 1135208 54.63 1234312 63.57
6 425338 68.12 914374 63.46 842416 75.17

TABLE 9:

Maximum Power Constraint 500mW

Number Number of Number of Number of

of cores=50 cores=100 cores=150
available
BIST circuits (m) test time R% test time R% test time R%

1 1335722 — 2502328 — 3388296 —

2 896512 32.89 1602632 35.95 2016354 40.49

4 512286 61.65 1142534 54.34 1276134 62.34

6 420864 68.49 930276 62.82 854472 74.78

TABLE 10:

Maximum Power Constraint 750mW

Number Number of Number of Number of

of cores=50 cores=100 cores=150
available
BIST circuits (m) test time R% test time R% test time R%

1 1335722 — 2502328 — 3388296 —

2 802322 39.93 1534864 38.66 1994578 41.13
4 501824 62.43 1115554 55.42 1224188 63.87
6 406638 69.56 889762 64.44 811832 76.04
 SHIBAJI BANERJEE et al : A PROGRAMMABLE BUILT-IN SELF-TEST
Before applying the scheduling algorithm, we first
divide memory cores into different partitions based on
the number of available BIST circuits. Then we allot
two BIST circuits to each partition. Memory cores are
grouped randomly such that the difference in number
of cores between any two partitions can be at most
one. For example, when the number of available BIST
circuits and total number of memory cores are 6 and
100 respectively, the memory cores will be divided
into 3 partitions. The number of cores in each partition
will be 33, 33, and 34. The partitions thus created are
balanced with respect to each other. Among the 6
available BIST circuits, we assign 2 of them to each
partition and schedule them by using the proposed
algorithm.
When the number of available BIST circuit is one,
cores will be tested sequentially one after another and
the test time can be represented by Tmax. We can
define the test time reduction factor (R) as,
Tmax – test-time-with m BIST circuits
R = ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ¾ ´ 100%
Tmax
8. CONCLUSIONS
The paper demonstrates that for SOCs compris-
ing of SRAMs and DRAMs the proposed, algorithm
can be used to test them. For DRAMs it can detect all
NPSFs, SAFs, TFs, and CFs. For SRAMs it can
detect all SAFs, TFs, and CFs. A programmable BIST
circuit is proposed, which provides good flexibility on
both test algorithms and memory types. It supports
different types of test algorithms and memory cores
according to the user choice. The user needs to program
the BIST before testing of a particular memory core
with a particular test algorithm. Today, an SOC may
consist of tens to hundreds of embedded memories.
Using a dedicated BIST for each of them is not a
feasible solution due to large area overhead. So, sharing
of BIST is needed for these types of memories. A
power-constraint test scheduling algorithm is also
proposed to reduce overall testing time. The
experimental results verify the above claims.
REFERENCES
1. Semiconductor Industry Association. Int.
Technology Roadmap for Semicconductors, http://
public.itrs.net/files/1999_51A. Roadmap/Home.htm.
2. Shankar Hemmady, Tom Anderson, & Yervant
Zorian, Verification and Testing of Embedded Cores,
In Proc of Design SuperCon, January 1997, pp s122-
1-19.
FOR EMBEDDED MEMORY CORES 309
3. K J Lin & C W Wu, PMBC: a programmable BIST
compiler for memory cores, In Proc of IEEE Int,
Workshop on Testing Embedded Core-based
System Chips, April 1999, pp P2.1-P2.6.
4. Kuo-Liang Cheng, Chia-Ming Hsueh, Jing-Reng
Huang, Jen-Chieh Yeh, Chih-Tsun Huang & Cheng-
Wen Wu, Automatic generation of memory built-in
self-test cores for system-on-chip, In Proc of Asian
Test Symposium, 19-21 Nov 2001, pp 91-96.
5 A J van de Goor, Testing Semiconductor Memories:
Theory and Practice, John Wiley and Sons,
CHicEester, England, 1998.
6. J P Hayes, Testing memories for single-cell pattern
sensitive fault, IEEE Transactions on Computers,
1980, vol 29, pp 249-254.
7. K K Saluja & K Kinoshita, Test pattern generation
for API faults in RAM, IEEE Transactions on
Computers. March 1985, vol 34, p 284-287.
8. Y Zorian, S Dey & M J Rodgers, Test of future
system-on-chips, In Proc of IETE/ACM
International Conference on Computer Aided
Design, 5-9 Nov 2000, pp 392-398.
9. P Camurati, P Prinetto, M S Reorda, S Barbagallo, A
Burri & D Medina, Industrial BIST of embedded
RAMs, IEEE Design and Test of Computers, 1995,
vol 12, Issue 3, p. 86.
10. K Zarrineh & S J Upadhyaya, On programmable
memory built-in self test architectures, In Proc of
Design, Automation and Test in Europe Conference
and Exhibition, 9-12 March 1990, pp 708-713.
11. Po-Chang Tsai, Sying-Jyan Wang & Feng-Ming
Charig, TSM-based programmable memory BIST
with macro command, In Proc of IEEE International
Workshop on Memory Technology, Design, and
Testing, 3-5 Aug. 2005, pp. 72-77.
12. Chih-Tsun Huang, Jing-Reng Huang, Chi-Feng Wu,
Cheng-Wen Wu & Tsin-Yuan Chang, A
programmable BIST core for embedded DRAM,
IEEE Design and Test of Computers, Jan-March
1999, vol 16, Issue 1, pp 59-70.
13. H Koike, T Takeshima & M Takada, A BIST scheme
using microprogram ROM for large capacity
memories, In Proc of International Test Conference,
10-14 Sept 1990, pp 815-822.
14. I Schanstra, D Lukita, A J van de Goor, K Veelenturf,
& P J van Wijnen, Semiconductor manufacturing
process monitoring using built-in self-test for
embedded memories, In Proc of International Test
Conference, 18-23 Oct 1998, pp 872-881.
15. R Rajsuman, Testing a system-on-a-chip with em-
bedded microprocessor, In Proc of International
Test Conference, 28-30 Sept 1997, pp 499-508.
16. Bai Hong Fang, Qiang Xu, & N Nicolici, Hardware/
software co-testing of embedded memories in
complex SOCs, In Proc of International Conference
on Computer Aided Design, 9-13 Nov. 2003, pp 599-
605.
310 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
17. Chin-Lung Su, Rei-Fu Huang, & Cheng-Wen Wu, A
processor-based built-in self-repair design for
embedded memories, In Proc of Asian Test
Symposium, 16-19 Nov 2003, pp 366-371.
18. J Dreibelbis, J Barth, H Kalter & R Kho, Processor-
based built-in self-test for embedded DRAM, IEEE
Journal of Solid-State Circuits, Nov 1998, vol
33(11), p 1731.
19. Ching-Hong Tsai & Cheng-Wen Wu, Processor-
programmable memory BIST for bus-connected
embedded memories, In Proc of Asia and South
Pacific Design Automation Conference, 30 Jan-2
Feb 2001, pp 325-330.
20. C A Papachristou, F. Martin & M Nourani,
Microprocessor based testing for core-based
system on chip, In Proc of Design Automation
Conference, 21-25 June 1999, pp 586-591.
21. Kuo-Liang Chen, Ming-Fu Tsai & Cheng-Wen Wu,
Efficient Neighborhood Pattern-Sensitive Fault Test
Algorithm for Semiconductor Memories, In Proc of
IEEE VLSI Test Symgosium, 29 May 2001, pp 225-
230.
22. S Boutobza, M Nicolaidis, K M Lamara & A Costa,
Programmable memory BIST, In Proc of
International Test Conference, 2005, pp 1155-1164.
23. Said Hamdioui & Georgi Gaydadjiev, Future
challenges in memory testing, citeseer.ist.psu.edu/
649203.html.
24. D Zhao & S Upadhyaya, Dynamically partitioned
test scheduling with adaptive TAM configuration
for power-constrained SoC testing, IEEE
Transactions on Computer-Aided Design of
Integrated Circuits and Systems, June 2005, vol 24,
Issue 6, pp 956-965.
25. M Nourani & C Papachristou, An ILP formulation to
optimize test access mechanism in system-on-chip
testing, In Proc of International Test Conference, 3-
5 Oct 2000, pp 902-910.
26. Vikram Iyengar, Krishnendhu Chakrabarty, E J
Marinissen, Test Wrapper and Test Access
Mechanism Co-optimization for System-on-chip, in
Journal of Electronic Testing: Theory and
Applications, April 2002, vol 18, pp 213-230.
27. Sandeep Koranne & Vishal Suhas Choudhary,
Formulation of SOC Test Scheduling as a Network
Transportation Problem, In Proc of Design,
Automation and Test in Europe Conference and
Exhibition, 4-8 March 2002, p 1125.
28. Yu Huang, Wu-Tung Cheng, Chien-Chung Tsai,
Omer Samman, Yahya Zaidan, & Sudhakar M Reddy,
Resource Allocation and Test Scheduling for
Concurrent Test of Core-Based SOC Design, Nov
2001, pp 361-366.
29. V Iyengar, K Chakrabarty, & E J Marinissen, On
using rectangle packing for SOC wrapper/TAM co-
optimization, In Proc of IEEE VLSI Test Symposium,
28 April-2 May 2002, pp 253-258.
30. S K Goel & E J Marinissen, Effective and efficient
test architecture design for SOCs, In Proc of
International Test Conference, 7-10 Oct. 2002, pp
529 538.
31. K Chakrabarty, Design of Systein-on-a-Chip Test
Access Architectures using Integer Linear
Programming, In Proceedings IEEE VLSI Test
Symposium (VTS), 2000, pp 127-134.
32. K Chakrabarty, Test Scheduling for Core-Based
Systems Using Mixed-Integer Linear Programming,
In Transactions on Computer-Aided Design, Oct
2000, vol 19, pp 1163-1174.
33. Yu Huang, Sudhakar M. Reddy, Wu-Tung Cheng,
Paul Reuter, Nilanjan Mukherjee, Chein-Chung Tsai,
Omer Samman & Yahya Zaidan, Optimal Core
Wrapper Width Selection and SOC Test Scheduling
Based on 3-D Bin Packaging Algorithm, In Proc of
International Test Conference, Oct 2002, pp 74-82.
34. Wei-Lun Wang, March based memory core test
scheduling for SOC, In Proc of Asian Test
Symposium, 15-17 Nov 2004, pp. 248-253.
35. M L Bushnell & V D Agrawal, Essentials of
Electronic Testing for Digital, Memory, and Mixed-
Signal VLSI Circuits, Kluwer Academic Publishers,
Boston/Dordrecht/London, 2000.
36. S Hamdioui, Z Al-Ars & A J van de Goor, Testing
static and dynamic faults in random access
memories, In Proc of IEEE VLSI Test Symposium, 28
April-2 May 2002 pp 395-400.
37. M L Bushnell & V D Agrawal, Essentials of
Electronic Testing for Digital, Memory and Mixed-
Signal VLSI Circuits, Springer, Berline, Germany,
2000.
38. A J van de Goor & I B S Tlili, A systematic method
for modifying inarch tests for bit-oriented memories
into tests for word-oriented memories, IEEE
Transactions on Computers, Oct 2003, vol 52(10), p
1320.
39. G L Craig & C R Kime, Pseudo-exhaustive adjacency
test: A BIST approach for stuck-open faults, In Proc
of International Test Conference (ITC), 1985, pp
126-137.
40. W Wang & S K Gupta, Weighted random robust
path delay testing of synthesized multilevel circuits,
In Proc of VLSI Test Symposium (VTS), 1994, pp 291-
297.
41. I P Shaik & M L Bushnell, A graph approach to DFT
hardware placement for robust delay fault BIST, In
Proc of International Conference on VLSI Design,
1995, pp 177-182.
42. I Voyiatzis, A Paschalis, D Nikolos & C Halatsis,
An efficient built-in self test method for robust
path delay fault testing, Journal of Electronic
Testing : Theory and Applications (JETTA), 1996
pp 219-222.
 SHIBAJI BANERJEE et al : A PROGRAMMABLE BUILT-IN SELF-TEST
43. D K Das, I Chaudhuri & B B Bhattacharya, Design of
an optimal test pattern generator for built-in self
testing of path delay faults, In Proc of International
Conference on VLSI Design, January 1998, vol 139,
pp 205-210, IEEE CS Press, USA.
44. S Wolfram, Statistical mechanics of cellular
automata, Rev Mod Phys, vol 55, no 3, pp 601-644,
July 1983.
45. D Roy Chowdhury, I Sengupta & P Pal Chaudhuri,
Cellular automata based pattern generator for
testing RAM, IEE Proceeding – Part E, November
1992, vol 139, pp 469-476.
Authors
Shibaji Banerjee received BTech.
and MTech degrees from the Institute
of Radio Physics and Electronics,
Kolkata, in 2001 and 2003, respectively.
He is currently working toward the PhD
degree in Computer Science and
Engineering at the IIT, Kharagpur. His
research interests are in the area of scan
architecture, system-on-chip testing,
and memory testing.
* * * 2001-2002, and at the Fault-Tolerant Computing Group, Institute
Dipanwita Roy Chowdhury is
Professor of Computer Science and
Engineering, Indian Institute of
Technology, Kharagpur, India. Her
current research interests are in the field
of VLSI design and testing, cryptography
and cellular automata. Dr Roy
Chowdhury received her BTech. and
MTech degrees in Computer Science
from the University of Calcutta in 1987
and in 1989 respectively, and obtained the PhD degree in
Computer Science and Engineering from the Indian Institute of
Technology, Kharagpur, in 1994. She has been teaching for more
than 13 years.
* * * *
FOR EMBEDDED MEMORY CORES 311
46. Subhayu Basu, I Sengupta, D Roy Chowdhury & S
Bhowmik, An Integrated Approach to Testing
Embedded Core and Interconnects Using Test
Access Mechanism (TAM) Switch, in Journal of
Electronic Testing: Theory and Applications
(JETTA), Special Issue on Plug-and-Play Test
Automation for System-on-a-Chip, Aug 2002, vol
18, pp 475 485.
47. A Sehgal, F Liu, S Ozev & K Chakrabarty, Test
planning for mixed-signal SOCs with wrapped analog
cores, In Proc of Design, Automation and Test in
Europe, 2005, pp 50-55.
Bhargab B Bhattacharya received
BSc degree in physics from the
Presidency College, Calcutta, the BTech.
and MTech. degrees in radiophysics
and electronics, and the PhD degree in
computer science all from the University
of Calcutta, India. Since 1982, he has
been on the faculty of the Indian
Statistical Institute, Calcutta, where he
is full professor. He held visiting
positions at the Department of Computer Science and Engineering,
University of Nebraska-Lincoln, USA, during 1985-1987, and
of Informatics, the University of Potsdam, Germany during
1998-2000. In 2005, he visited the Indian Institute of Technology
Kharagpur as VSNL Chair Professor. His research interest includes
logic synthesis and testing of VLSI circuits, nanotechnology,
digital geometry, and image processing architecture. He has
published more than 200 papers in archival journals and refereed
conference proceedings, and holds 9 United States Patents.
Dr Bhattacharya is a Fellow of the Indian National Academy
of Engineering, a Fellow of the National Academy of Sciences,
India, and a Fellow of the IEEE. He is on the editorial board of
the Journal of Circuits, Systems, and Computers (World Scientific,
Singapore), and the Journal of Electronic Testing - Theory and
Applications (Springer).
* *
 INVITED PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 313-321

Space Enabled ICT Applications for Rural

Upliftment – Experience of Participatory
Watershed Development
P G DIWAKAR
Head, RRSSC, ISRO, Banashankari, 40th Main, Eashwar Nagar
(Behind PSTI), Bangalore 560 070, India.
AND

V JAYARAMAN, FIETE

Director, EOS and NNRMS-RRSSC, ISRO Hq, Anthariksha Bhavan,

New BEL Road, Bangalore 560 094, India.

Earth Observation (EO) inputs on various natural resources parameters are effectively
used for integrated land and water resources development at grass roots helping in
enhancing the productivity levels in dryland areas. EO inputs and a judicious mix
Geographical Information System (GIS) and Management Information System (MIS) has
helped in characterization, prioritization, developmental plan preparation, generation of
implementation strategies (with peoples’ participation), concurrent project monitoring and
impact assessment at various stages of implementation. The innovativeness of this study
lies in the infusion of modern technologies in the form of providing EO-based inputs on
natural resources for development, simple-to-use information system with MIS/GIS in local
language, reporting and monitoring through simple client-server solutions, participation of
people, experts, Local Government, NGOs, project monitoring group and other stakeholders
with one clear objective of achieving sustainable development. This innovative method of
project implementation and monitoring has brought about significant impacts on the natural
resources conservation and positive trends in livelihood condition of the people.

INTRODUCTION fed areas experience high degree of land degradation

and suffer from low and uncertain rainfall, poor soil

N ATURAL resources conservation through

watershed development programs with active
participation by communities has been practiced through
many government-sponsored schemes. These programs
have been executed by adopting conventional methods
across the country. Such participatory approaches are
focused on mobilizing the farming community to
collectively take up measures to conserve the soil and
water resources in their respective watersheds. The
idea here is to adopt control mechanism to limit the
runoff potential of the soil, based on terrain conditions,
that ultimately drains to a common point by designing
locale specific approaches through community
participation. The green revolution that transformed
agriculture in India did not significantly impact on rain
fed agriculture in the arid and semi-arid tropical regions,
which experience low agricultural productivity,
degraded natural resources and poor people. The rain
Paper No 126-E; Copyright © 2007 by the IETE.
fertility, sparse vegetation cover, low productivity, lack
of infrastructure and so on. Majority of the people of
these areas have limited access to primary education,
basic health care, clean drinking water, food and decent
livelihood conditions. Present day world has evolved a
special focus on optimal management of land and
water resources with an approach of integrated mission
of economic development, equity and environmental
soundness, evolving multi-pronged strategy of
sustainable development with special focus on poorer
sections of the society and regenerating the eroded
natural resource base. Watershed development is
gradually evolving into a comprehensive program with
simultaneous pursuit of biophysical and rural
development objectives that promote rural livelihoods.
In order to make such programs more acceptable
and simple, a set of innovative methods have been
evolved and field-tested with significant success. Use
of EO data with respect to prespecified time frame,

313
314 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
image analysis, new software tools and geospatial
databases for participatory watershed development
program is well adapted in the ‘Sujala’ project being
implemented in Karnataka state. Attempt has been
made to use an optimal combination of earth observation
inputs, field observations and Information Technology
for planning in implementation, monitoring and impact
assessment.
COMMUNITY BASED WATERSHED
DEVELOPMENT
Organization and societal issues have recently
become important subjects of GIS research
(Obermeyer, 1998). Watershed development is a well
accepted method of treating land through scientific
means for sustainable development of natural resources.
Watershed itself is a natural boundary connecting
similar elevation points in the form of ridge tops which
is self contained polygon of natural resources which is
amenable for conservation and protection. Different
methods, suitable to locale specific conditions, have
been practiced for ages across the world, that involves
treating land on a ridge-to-valley basis to restore
equitable use of soil and water resources in a given
terrain. It is a well known fact that any watershed
development program poses varieties of challenges as
there are many stakeholders and players who are
important for achieving success in implementation. A
few typical parameters considered while watershed
development are: large extent of private land (belonging
to farming community), common land (sharable by the
village community), livestock, fuel and fodder areas,
tanks and waterbodies, waterholding structures,
farming community with land, landless labourers, infants
and so on. The newer techniques of image processing,
extraction of natural resources information, using GIS
for data integration and modeling, providing extract of
such processing in the form of land and water resources
plans, IT tools for monitoring and impact assessment
have made significant impacts in successful
implementation of watershed development.
Watershed Development program
Sujala is a participatory watershed development
programme, which is being implemented by Government
of Karnataka with World Bank assistance in 5 districts,
viz., Kolar, Tumkur, Chitradurga, Dharwad and Haveri
of Karnataka State. The major objective of the project
is to improve the productive potential of degraded
watersheds in dry land areas and poverty alleviation of
rural community. The project is spread across 77 sub-
watersheds covering an area of about 0.51 million
hectares and benefiting about 400,000 households.
This is a community driven program wherein village
people involve in participatory planning, implementation
and maintenance of assets.
Geospatial inputs and IT solutions
Satellite Remote sensing inputs, GIS, a near real-
time MIS solution and GIS solution for participatory
planning are the core areas of technology utilisation
for resource mapping, database generation, analysis
and information extraction for watershed planning,
implementation and monitoring. In Sujala project, high
resolution satellite data with a spatial resolution of 6
meter has been utilised to generate maps on 1:12,500
scale. Typical areas of technology utilization under the
project are
• Watershed prioritization
• Resources Inventory and Mapping
• Land and Water resources action plan
• Site selection for implementation
• Web based Management information system
(MIS) solution
• GIS based action plan development
• Implementation of concurrent Monitoring
• Impact Assessment
• Post project Evaluation
• Run off estimation by imperial methods.
EO DATA FOR DEVELOPMENTAL
PLANNING
EO data from Indian Remote Sensing satellite has
proved to be very effective in similar such programs
earlier. However, with the availability of high resolution
images from space platforms, RESOURCESAT and
CARTOSAT, there has been a significant shift in the
way the technology is used for monitoring. LISS 4
multispectral images (5.8 m) and high spatial resolution
CARTOSAT 1 PAN (2.5 m) could be effectively
used to carry out an on-line tracking of the various
developmental works effectively and accurately. It is
rare that such high resolution image data is used, not
only for facilitating the farming community to take up
technically sound decisions for watershed development,
but also for monitoring such implementations. The
usage of EO data has been facilitated by adopting
special product generation techniques, like, image fusion,
natural color composites, landuse classification, change
detection and other thematic layers creation to aid in
locale specific action plan preparation. Geospatial layers
at high resolution is created and integrated with
cadastral/parcel data to facilitate proper planning of
 DIWAKAR & JAYARAMAN : SPACE ENABLED ICT APPLICATIONS 315

private and common land developments. In addition to

this, such high-resolution data from EO satellites is
also used for bringing out short term and long term
impacts at local level. Availability of such high spatial
resolution data from IRS satellites has brought about
required transparency and unbiased field-level
assessment possibilities in a cost effective manner.
The unique combination of sensors available from IRS
P6 satellite brings out unique possibilities of providing
total solution for such projects. A combination of
AWIFS and LISS 3 with 55 m and 23 m spatial
resolution respectively enables natural resources
assessment at coarser scale and also allows for
watershed prioritization. LISS 3 could further be used
for thematic mapping at 1:50,000 scale for establishing
medium scale database for study area assessment at
Figure 1: Strategy for watershed prioritization
district level. LISS 4 and Cartosat 1 PAN is further
used in combination at 1:10,000 or better scale for
providing specific decision making capabilities at locale Spatial Data modeling
level.
1. Weights assignment for layers
Watershed Prioritization 2. Use of spatial & Non-spatial data
The prioritization of watersheds on the basis of 3. Multiple Unions and Math Algebra
natural resources status, socio-economic, biophysical 4. Analysis of the output for priority
and other criteria has been carried out to select 77 sub 5. Watershed ranking & Prioritization
watersheds for implementation. Geospatial data and 6. GIS output of Prioritization
multi criteria based prioritization of watersheds helps
in making unbiased choice of target areas for
development.
evaluations and other monitoring works. Simple GIS
WP = Uwi Xj j = 1,2,……..nj and IT tools at microwatershed level has been
Where, U represents union of n GIS layers successfully used to establish a strong baseline
information system and database elements under the
wi represents weights assigned to each layer project. Also, the sampling procedure and strategy for
Xj represents elements of a GIS layer obtaining regular data from the field has also been well
thought of. A well defined field data collection procedure,
The multi-layer geospatial analysis results in the which is critical to establish a strong baseline for the
generation of composite mapping units which could study area, has been adopted through scientific means
further be processed through multi criteria analysis to by adopting multistage sampling approach that uses
arrive at the end result. Some of the important GIS criteria which is an optimum mix of natural resources
layers considered in such an exercise are: Natural and socio-economic parameters.
resources: wastelands, irrigated areas, forests, rainfall
distribution, silt yield index; Socio-Economic Indicators: Benchmark survey is carried out to collect data
Agricultural laborers, distribution of BPL families and and information on the pre-project status of the
vulnerable groups (Fig 1). community and natural resources. This information
helps in monitoring the project at various stages and
Baseline and Sampling Strategy: Establishment of assess the changes in the project area [7]. A
systematic Baseline information is most crucial for combination of conventional and Remote Sensing
effective monitoring, evaluation and measuring the approaches are utilized to generate benchmark data.
impacts. That is, it is essential to identify the indicators A typical Geostatistical approach with Multi-stage
and create baseline information system which involves sampling is adopted for establishing baseline. One of
both natural resources information and parameters the important factor considered while designing such a
related to socioeconomic situation at grass roots. This strategy is the adaptation of multistage sample design
proves to be a good reference for subsequent approach (involving Stratification criteria,
316 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
Randomization, Area based sampling with probability
proportion to size) and use of the same for such a big
project and delivering the same on the field effectively.
The sub-watersheds are randomly selected based on
their agroclimatic conditions, status of land use and
soil. Within each sub watershed, 3 micro watersheds
are selected at random, representing Ridge, Middle
and Valley portions. Households are selected and
randomly sampled based on land holding class, i.e.,
Marginal, Small, Big and Landless using Probability
Proportion to Size (PPS) criteria. To achieve acceptable
accuracies of estimates and as per the project
guidelines, 10% sampling intensity is considered for
data collection and analysis (Fig 2).
Geospatial databases on Natural
Resources
Very few community based developmental projects
adopt newer technologies for resource mapping. The
technique based on EO inputs has tremendous potential
to not only establish a strong baseline on natural
Fig 2 Geostatistical approach for baseline establishment
resources but also provide critical inputs for on-line
monitoring of the implementation works. This type of
possibility has given a big boost in the area of project
monitoring by bringing about transparency in project
implementation and peoples’ participation at grass roots.
High-resolution satellite images have been
processed and provided at large scales to the village
communities involved in watershed development to
help them understand the locale specific nature of the
terrain for better planning and management of natural
resources. Image fusion and natural color composite
based product generation techniques were successfully
employed to make the basic satellite images community-
friendly. Image fusion is extensively used to take full
advantage of high-resolution geometry of the pixel and
the optimum spectral mixing of the multispectral
component to produce best possible output. It was
discovered that for preserving spectral characteristics,
one needs a high level of similarity between the
panchromatic image and the respective multispectral
intensity [1]. Varieties of image fusion methods have
been used in the present study amongst some of the
popularly used techniques, such as, IHS, PCA,
Broovey’s method, Discrete Wavelet Transforms,
adaptive linear band combination and so on. However,
these techniques do depend on the inter-band correlation
existing amongst the multispectral channels to
appreaciate the results.Multispectral transformation
techniques are adopted to generate natural color
composite images by establishing a relationship between
the spectral bands producing false color composite
with that of true color composite (Fig 3a).
These images and other ancillary data have been
analysed and integrated under GIS to generate various
types of resource maps, viz., land use / land cover,
soils, slope, hydrogeomorphology, drainage, transport
network, settlements, land parcels, etc., at the micro-
watershed level. These maps play an important role in
understanding the spatial nature and interrelationship
that exists between different resources. From a practical
point of view, a high-resolution satellite image depicting
terrain in true color could be the most comfortable one
for conventional interpretation and visualization [2].
Locale specific action plans for sustainable development
of land and water resources are generated on micro
watershed basis by integrating thematic information
from the resource maps, peoples’ aspirations and
socioeconomic inputs with special emphasis on
community needs (Fig 3b).
Such action plans, prepared through community
participation, basically address private lands (the land
belonging to individuals in the villages) and common
lands (the Government land utilized by the village
 DIWAKAR & JAYARAMAN : SPACE ENABLED ICT APPLICATIONS 317

Keeping these points in view it was decided to

Fig 3a Fusion and natural color
customize specific GIS and IT tools in a simple-to-use
form for the community to adopt at local level. A
simple JAVA based GIS tool “Sukriya Nakshe” with
an option to prepare parcel-wise watershed
development plan through a Participatory Rural
Appraisal (PRA) process (private land and common
land development plans) was developed (Fig 4a) and
at the same time a VB based database engine was
customized to capture details at beneficiary level on all
aspects of watershed development in local language
(Fig 4b).
The impact of such a tool has been quite significant
at the village level and also for the project implementing
agency. It has brought about transparency in the project
as these databases serve as the basic information for
wall-paintings in the villages, the entire community

Fig 3b Geospatial data layers for participatory action

plan preparation

community). These plans are basically the

Fig 4a Information technology solution for community
recommendations towards improved soil and water
level action plan preparation
conservations for ensuring enhanced productivity, while
maintaining ecological / environmental integrity of the
micro-watershed. The action plans also address the
identification of sites / areas for surface water
harvesting, ground water recharge, soil conservation
measures through check dams, vegetative bunding,
sites etc. It also specifies sites for improved / diversified
farming systems with fodder, fuel wood, agro-forestry,
agro-horticulture etc. These action plans are generated
jointly by watershed department experts (agriculture,
horticulture, animal husbandry & forestry) field NGOs
and beneficiaries. While all the land based activities
under such plans address the marginal, small and big
farmers, the landless people get the benefit of the
Income Generation Activities (IGA) mostly focused
on Self Help Groups (SHGs). District Resources Group
(DRG) scrutinizes the action plans technically before
the approval of Zilla Panchayat (local body for decision Fig 4b Community level GIS technology for PRA and
making) for implementation. action plan preparation
318 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

know the contribution of the project and their share for

implementation, NGOs facilitating such an activity at
the community level finds it easy to monitor all activities
with ease, the project authorities are able to quickly
prepare the yearly logical action plans for monitoring,
package enables many technical evaluations like, Equity,
Inclusiveness, Gender sensitivity, Environmental and
social assessment of action plans etc., before approving
the action plan for implementation.

IT tools for on-line project monitoring

The project is implemented at about 750

microwatersheds across 5 Districts in Karnataka.
Concurrent information on the status of implementation
at each microwatershed is required on a continuous
basis for monitoring and management of various
resources under the project. This calls for a systematic
compilation of information at each microwatershed
and the same to be made available at sub-watershed,
District and State level for management at different
levels of hierarchy. Such a requirement is facilitated
by adopting simple web-based tools at various levels
to enable systematic data flow across the different Fig 5a Package architecture & database
stages of decision making. “Sujala Mahiti”, a web- synthesis - web-based model
based MIS-GIS tool enables data capture at the lower
most level of project implementation with respect to
Social aspects, IGA and action plan implementation.
However, the package with different component as
above are used by Field NGOs, Specialist IGA
consultant and District Watershed development office
respectively to facilitate a data synthesis at the District
and State level. This model of data compilation and
synthesis has been proved to be highly successful and
facilitated a smooth weekly monitoring of the project.
The package is a browser based customization on the
client side with server component taking all the
computation load. The server component includes an
Apache server, PHP scripting language interface,
PostGresql RDBMS configured around CygWin
environment. Following illustrations provide broad
architecture of the working MISGIS model (Fig 5a).
Very few projects, implemented at community
level, are able to have access to such a wealth of data
from the field for effective monitoring and carrying
out specific analysis with respect to the key performance
indicators of the project. The data flow from the field
level is regularized with respect to weekly inputs and
the same is used at all levels for online interactions
through Audio conferences. Thus a typical example of
how proper data flow could empower planners and
managers to effectively carry out monitoring of such a
massive project. The package enables wide varieties
Fig 5b Hierarchical organization of query engine and
of report generation on all aspects of the processes data flow
 DIWAKAR & JAYARAMAN : SPACE ENABLED ICT APPLICATIONS
undertaken in the project and also provides varieties of
graphics and analysis functions as value added tool.
Other unique feature of the package is the possibility
to have access to GIS maps through WebGIS
technology with MIS reports and graphs to provide a
spatial dimension for monitoring and management of
the project more efficiently (Fig 6).
Fig 6 Graphic analysis & WebGIS for online monitoring
EO data processing and GIS modeling for
Impact Assessment:
The potential of space technology in generating
the base line information on land and water resources
and in monitoring the progress and success of watershed
development programme has been substantiated from
various studies carried out so far [3]. Impacts due to
various interventions are monitored through a
combination of remote sensing data, GIS, MIS data,
process monitoring data and farmers / household
surveys. Through a scientifically designed mechanism,
impact assessments are done at pre-determined time
319
intervals to establish the net contribution of the project
to poverty alleviation and natural resource regeneration
(Fig 7). Impact is evaluated using a variety of qualitative
and quantitative indicators with respect to baseline,
midterm and post-project status. Impacts are also
analysed based on observations made in the project
and control areas. A comprehensive benchmark data
has been established through judicious combination of
conventional and remote sensing data to facilitate
monitoring and impact assessment. Participatory
Observations, Focus Group Discussions, Transect
Walks are some of the other methods of data collection
at community level. The data available from MIS/GIS
system, thematic reviews and specific case studies
are also utilised for the impact assessment. One of the
most crucial points to be noted in effectively carrying
out such impact studies is the establishment of a
strong baseline database, both from satellite remote
sensing, GIS and field based observations.
Fig 7 Typical example of EO use in measuring impacts
The application of cutting edge technology including
remote sensing, GIS and Computer based monitoring
system in conjunction with ground observations has
provided robust baseline and change data and wealth
of information for in-depth analysis (Grant, 2006). The
mid-term assessment of the project has indicated that
project investments are having a measurable impact
on the indicators representing the project development
objectives. Some of the visible changes observed are:
increase in average crop yields, crop diversity has
increased from an average of 2-5 crops in the baseline
to 4-9 crops, increase in annual household income due
to employment, income generating activities and
320 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
improvements in agricultural productivity has increased
by about 30 %, the average water level has increased
by 3 to 5 feet, shift to agro-forestry and horticulture
and reduction in non-arable lands, increased employment
and income has resulted in changed seasonal migration
pattern and intensity. Greater transparency and capacity
building has resulted improved awareness, participation,
particularly amongst women, and social response. Door
delivery of livestock services has resulted in improved
vet services and livestock health [5].
CONCLUSION
Simple adaptation of Earth observation inputs and
the Information Technology elements of MIS, GIS
and ground-based observations have helped in micro
level plan preparation, concurrent project monitoring
and impact assessment at various stages of project
planning and implementation. The integrated approach
for systematic planning of Monitoring and Evaluation
with application of cutting edge technology including
remote sensing and locale specific information system
based management practices has provided required
transparency and success at community level. Proper
adaptation of technologies with respect to natural
resource regeneration and strengthening of local
institutions has lead the project towards greater
sustainability. EO inputs have provided the state-of-
the-art information enhancements for tracking and
self-assessment of the communities and helped the
project to achieve set goals against development
indicators and milestones. It has also enabled
appropriate policy formulation, implementation of
suitable strategies / action plans, assessing the impacts,
resulting in mid course corrections and better impacts
in the field. It has also significantly brought about
transparency and accountability amongst all
stakeholders in the project.
Thus, the project has ushered a new era of hope
and confidence into the hearts of the rural inhabitants
of 1270 villages across 5 districts in Karnataka by
rejuvenating natural resources and institutional
strengthening through participatory watershed
development. This program is an attempt to bring
about a better balance amongst the basic pre-requisites
of sustainable development, social equity, environmental
quality and economic efficiency. The transformation
in the livelihoods and living standards of poor farming
community is an interesting outcome indicating
reduction in poverty. It has become a milestone on the
path of sustainable development as the plan for the
people, by the people with external facilitators has
provided sophisticated science and technology tools
for decision-making. Sujala, as a project, has become
a blue print for sustainability and a role model, which is
worth emulating in other developmental projects.
REFERENCES
1. Andreja Svab & Kiristof Ostir, High-resolution
Image Fusion: Methods to Preserve Spectral and
Spatial Resolution, PE&RS, vol 72, no. 5, May 2006,
pp 565-572.
2. Chen C F & Tsai H T, A spectral transformation
technique for generating SPOT natural colour image.
In poster session 3, GIS development Proceedings,
ACRS, 1998.
3. Diwakar P G, B K Ranganath & V Jayaraman,
Participatory watershed development - methods of
monitoring and evaluation. Proc of 24th ISRS
Symposium, Jaipur Rajasthan, 2004.
4. Grant Milne et al, Managing Watershed Externalities
in India, Agriculture and Development Sector Unit,
South Asia Region, Report no 1, World Bank, 2006.
5. Muniyappa, N C, B K Ranganath & P G Diwakar,
Improving the quality of life in rural populace: Role
of Remote Sensing and GIS in Participatory
Watershed development in Karnataka, Indo-US
conference, Bangalore, 2004.
6. Obermeyer N J, The evolution of public participation
in GIS, Cartography and Geographic Information
Systems, 25(2), pp 65-66, 1998.
7. Sander C, Planning, Monitoring and Evaluation of
programme performance - a Resource book,
International Development Research Centre
(IDRC) Canada, 1997.
 DIWAKAR & JAYARAMAN : SPACE ENABLED ICT APPLICATIONS
Authors
P G Diwakar, is a Post Graduate in
Statistics, specialization in Mathematical
Modelling; Specialisation in Software
Development - IISc & IIM-B and
Specialisation in Digital Image Processing
(VIPS-32), SEP, France (1986).
He is working in the area of Digital
Image Processing, GIS and Modeling
solutions. He has more than 22 years of
service in ISRO, since joining in 1984.
He is responsible for setting up and managing high-end
Image Processing facilities at Regional Remote Sensing Service
Centres (RRSSCs) of ISRO, across the country and a National
Facility for Remote Sensing Applications at Mauritius.
He has held positions as System Manager and Software
team leader. As a Team leader his responsibilities are Application
software development for 3 major national mission projects.
Project manager, NRIS: A major national mission for setting up
National Natural Resources Information System at district level.
His current responsibilities are Additional Project Director,
Monitoring and Evaluation of watershed development (A World
Bank project for sustainable rural development in Karnataka);
Head, RRSSC, ISRO, Bangalore – A National center for remote
sensing applications in southern region; Program Manager &
Regional Coordinator –Village Resource Centres: A new initiative
to provide space technology based solutions to rural community.
He has More than 50 Publications in national / international
Journals and Conferences. He is Life member of Indian Society
of Remote Sensing.
* * * Agency-Governing Society & NRSA - Governing Body; Member,
V Jayaraman received Bachelor
degree in Electronics and Communication
Engineering from University of Madras;
Master of Science in Electrical engineering
(by Research) from Indian Institute of
Technology, Madras; Diploma in
Management (specialized in Information
Technology) from All India Management
Association, New Delhi and Doctorate
in Physics from Bangalore University,
India.
321
He is Fellow of the Institution of Electronics and
Telecommunications Engineers (FIETE); Fellow, Indian Geo-
physical Union (FIGU); Member, Indian Society of Remote
Sensing (MISRS); Member, Indian Society of Geomatics (MISG);
Member, Astronautical Society of India (MASI).
He has more than 35 years of continuous service in ISRO,
since joining it in 1971 as Design Engineer in Aryabhata Project
for X-ray Astronomy Payload (1971-76); Systems Engineer in
Bhaskara I & II Projects (1977-81); Principal Systems Engineer
(Payload, Mission, and Ground Segment) for Indian Remote
Sensing Satellite, IRS-1A (1981-88) and Deputy Director, Earth
Observations System in ISRO Hq (1989-96).
He is currently, concurrently holding three positions as:
Director, Earth Observations System (from 1997 till date);
Programme Director, ISRO Geosphere Biosphere Programme
(from Sept 2002); Director, NNRMS-Regional Remote Sensing
Service Centres (from Sept 2003).
Besides the above, also heading the National Natural
Resources Management System (NNRMS) Secretariat at ISRO
Headquarters. Member of various State Natural Resources
Management System (SNRMS) management Councils/ Boards.
Currently, serving as the Member Secretary of the Planning
Committee of the NNRMS (PC-NNRMS), the apex body under
the aegis of the Planning Commission, Govt. of India, looking
after the overall remote sensing activities in the country.
He is Member of Management Councils of various ISRO/
DOS Centres/Programmes: Member, National Remote Sensing
Governing Council for the North Eastern- Space Applications
Centre; Member, ISRO Radar Development Unit - Programme
Management Council (I & Member, Indian Remote Sensing
Satellite Programme Management Council (IRS- PMC) and
Member, Earth Observation Applications Mission – Management
Council (EOAM-MC).
He has published more than 230 Technical/ Scientific/
Management papers in national and international journals of
repute; Edited a book entitled ‘ Space & Agenda-21 - Caring for
Planet Earth’ in 1994.
* * *
 CONTRIBUTED PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 323-330

Performance Analysis of SC DS-CDMA and

MC DS-CDMA Systems over Nakagami-m
Fading Channel
S ANURADHA, S SRIGOWRI, K S RAMA KRISHNA
Department of ECE, VRS Engineering College, Vijayawada 520 007, India.
AND

K V V S REDDY
Department of ECE, Andhra University College of Engineering,
Vishakhapatnam 530 003, India.
email: anu_praise2004@yahoo.co.in

This paper deals with simulation of Nakagami-m fading in wireless channels using
generalized SC DS-CDMA and MC DS-CDMA schemes using binary phase shift keying
modulation scheme. The paper proposes for deriving Bit Error Rate (BER) in BPSK-SC DS-
CDMA and MC DS-CDMA systems over Nakagami-m fading channels. The numerical results
are plotted as BER vs SNR for various values of Nakagami factor m, number of users K and
jamming interference (JSR) using MAT LAB software. The performance of Nakagami-m
channel using BPSK-SC DS-CDMA and MC DS-CDMA techniques are compared. It is
observed that there is a decrease in Bit Error Rate for increase in m for SC DS-CDMA and
MC DS-CDMA. The number of users k, JSR increases as Bit Error Rate increases for MC DS-
CDMA. For the same values of m BER for MC DS-CDMA is smaller when compared with SC
DS-CDMA using Maximal Ratio Combining (MRC) diversity technique.

INTRODUCTION impact on system performance can be found in [2].

SYSTEM MODELS
F ADING is observed in wireless communication
channels [1] due to multi path propagation. The
Nakagami-m distribution accurately models the fading
(a) SC DS-CDMA
effect for short distance communications. This paper
Introduction
deals with the derivation of BER in BPSK- SC DS-
CDMA and MC DS-CDMA systems over Nakagami-
In this section we apply the MGF-based approach
m channel. In the generic form, DS-CDMA access is
presented to derive the average BER performance of
a spread spectrum technique for simultaneously
binary DS-CDMA systems over these channels [3].
transmitting a number of signals representing
The results presented in this section are applicable to
information messages from a multitude of users over a
systems that employ RAKE reception with coherent
channel employing a common carrier. The method by
Maximum Ratio Combining (MRC).
which the various users share the channel is the
assignment of a unique pseudo noise (PN) type code
1. Transmitter
to each user (which accompanies the transmission of
information) with orthogonal like properties that allows
We consider a binary DS-CDMA system with
the composite received signal to be separated into its
Ku independent users sharing a channel simultaneously,
individual user components, each of which can then be
each transmitting with power P at a common carrier
demodulated and decoded. A complete discussion of
frequency fc=wc/2P .using a data rate Rb=1/Tb and a
techniques for accomplishing these functions and their
chip rate Rc=1/Tc. The kth user, k=1, 2…Ku is assigned
a unique code sequence {ak, j} of chip elements
Paper No 126-A; Copyright © 2007 by the IETE. (–1, +1). So that its chip waveform is given by
323
324 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
¥ user lth resolved path a k,l which is a random variable
ak (t) = S ak,j PTc (t–jTc)
j=–¥
Where the function PT(.) denotes the chip pulse of
duration T. In the single carrier case we assume that
PT(.) is a unit rectangular pulse, where in the multi
carrier case we will consider Nyquist pulses. The
code sequence {ak,j}is assumed to be periodic, with
period equal to the processing gain PG=Tb/Tc. The
data signal wave form bk(t) given by
¥
bk (t) = S bk,j PTb (t–jTb)
j=–¥
is the binary phase-shift-keyed (BPSK) on to the
carrier at fc, which is then spread by that user’s code
sequence and transmitted over the channel. The
resulting kth user’s transmitted signal sk(t) is given by
sk (t) = 2Pak (t)bk (t) cos (wct)
The composite transmitted signal s(t) at the input
of the channel can then be expressed as
Ku
s (t) = S 2Pak (t)bk (t) cos (wct)
k=1
2. CHANNEL MODEL
DS-CDMA systems involve a spreading process,
which results in a transmitted signal whose bandwidth
is much wider than the channel coherence bandwidth,
and therefore undergoes frequency-selective fading.
This type of fading is typically modeled by a linear
filter, which for a kth user is characterized by a
complex valued low loss equivalent impulse response
[4,5].
Lp
hk (t) = S a k,l e–jqk,l d(t–tk,l)
l=1
Where d(.) is the Dirac delta function. lth
propagation path index, and {a k,l}l=1Lp,{qk,l} l=1Lp and
{tk,l} l=1Lp the random path amplitudes, phases, delays
respectively. We assume that the sets {a k,l}l=1Lp,
{q k,l} l=1Lp and {tk,l} l=1Lp are mutually independent.
Lp is the number of resolvable paths (the first path
being the reference path whose delay tl = 0) and is
related to the ratio of the maximum delay spread tmax
to the chip time duration Tc. If the different paths given
impulse response are different scatter they tend to
exhibit negligible correlations [6,7]. We denote kth-
whose mean square is a 2k,l is assumed to be
independent of k and is denoted by W l.
After passing through the fading channel, the signal
is perturbed Additive White Gaussian Noise (AWGN)
with a one-sided power spectral design which is
denoted by N0(W/Hz). The AWGN is [10] assumed
to be independent of the fading amplitudes. Hence the
instantaneous SNR per bit of the lth channel is given
by gk,l = a 2k,l Eb / N0. Where Eb(j) is the energy per bit
and the average SNR per bit of the lth channel is given
by gl= W l Eb / N0.
3. RECEIVER
With multipath propagation, the received signal
r(t) whose signal component is the time convolution of
s(t) and h(t),may be written as
Ku Lp
ak,l ak (t–tk,l) ´
r(t) = 2P S S
b (t–tk,l) cos [wc (t–tk,l) + qk,l] + n(t)
k=1 l=1 k,
Where n(t) is the receiver AWGN random process.
Here we consider L-branch RAKE receiver. The
optimum value for L is Lp. but L may be chosen less
than Lp due to receiver complexity constraints. Each
of the L paths to be combined is first coherently
modulated through multiplication by the demodulated
carrier cos[wc(t–tk,l) + qk,l], then low pass filtered to
remove the second harmonics of the carrier. All the
operations are assumed that the receiver is coherently
timed and phase synchronized at every branch and
assuming perfect knowledge of fading amplitude on
each finger, the L low pass filter outputs are individually
weighted by their fading amplitudes and then combined
by a linear combiner yielding the decision variable
L
rk = S a k,l rk,l k=1,2...Ku
l=1
4. PERFORMANCE ANALYSIS
The decision variable rk can be written as sum of
desired signal component and three interference
components [8].
æL ö L
rk = ± çS a 2k,l ÷ Eb + S a k,l (IS + IM + N)
èl=1 ø l=1
Where IS is the self-interference component induced
by autocorrelation function of kth user’s spreading
 S ANURADHA et al : SC DS-CDMA AND MC DS-CDMA SYSTEMS 325

code. The multiple access interference (MAI) 1 p/2 L æ gee–(l–1)d ö –ml

component induced by the other Ku–1 users on the Pb (E) = ¾ ò X ç 1+ ¾ ¾ ¾ ¾÷ df
p 0 l=1 è ml sin 2 f ø
desired user and N is the zero mean AWGN component
with variance s2N = N0/2. It can be considered to be a Where
zero mean Gaussian RV with variance
æ (2 k + 1)W –3 1 ö –1
WT – 1 ge = ç ¾ ¾ u¾ ¾ ¾ T¾ ¾ + ¾ ÷
s2 s = ¾ ¾ ¾ W l Eb è 3PG g1 ø
2PG Lp
WT = S e–(l–1)d
Where W T = S W l/W 1 can be interpreted as normalized
l=1
total average fading power and PG is Processing
Gain. IM can be modeled as zero mean Gaussian (b) MC DS-CDMA
Random variable with variance
Introduction
2(Ku – 1)W T
s2 M = ¾ ¾ ¾ ¾ ¾ ¾ W l Eb
6PG MC DS-CDMA Constitutes a trade off between
SC DS-CDMA and MC DS-CDMA, MC DS-CDMA
rk can be considered to be a conditional Gaussian RV typically requires low chip rate spreading codes than
with mean and a conditional variance given by DS-CDMA due to employing multiple sub-carriers. It
necessitates a lower number of sub-carriers than MC
æL ö DS-CDMA due to imposing DS spreading [2] on each
E[rk | {a k,l}Ll=1 ] = ± ç S a 2k,l ÷ Eb
èl=1 ø sub-carrier signal. MC DS-CDMA requires low rate
signal processing than DS-CDMA and has lower worst-
L case peak to average power than MC DS-CDMA.
var (rk | {a k,l }L = ± S a2 k,l (s2N + s2S + s2M) MC DS-CDMA has highest of freedom in family of
l=1)
l=1
CDMA schemes. A MC DS-CDMA exhibit a no of
advantageous properties, this technique is employed in
Assuming that the data bits +1 or –1 are equally wireless communication, irrespective of presence of
probable the Kth user conditional SNR, is given by other techniques [9].
SYSTEM MODEL OF MC-DS CDMA
(E [ri | a i])2 æ L ö Eb
SNR ({a k,l}L = ¾ ¾ ¾ ¾ ¾ = ç S a 2k,l ÷ ¾ ¾ In this model a generalized MC DS-CDMA system
l=1 2 var (ri | ai) è l=1 ø Ne with K users is shown in Fig 1. BPSK modulation is
assumed. At the receiver, the desired user’s signal
Equivalent two-sided interference plus noise power will be decoded and the other K–1 user will contribute
spectral density is defined as to multi-user interference.

Ne 1. Transmitter
¾ ¾ º s2N + s2S + s2M
2
(2Ku +1) W T – 3 N0
= ¾ ¾ ¾ ¾ ¾ ¾ ¾ W lEb + ¾ ¾ s0(i)
6PG 2

Channel r(t)
With gl = W 1Eb / N0 the average received SNR per bit s1(i) Trans- Rake s (n)
corresponding to the first path. mitter h(l) Receiver k
Pilot
Since we are assuming BPSK modulation, the Channel
n(t)
average SNR per bit of lth path Channel
sN–1(i)
Estimation
–1
W l Eb é (2Ku + 1) W T – 3 ù
gl,e = ¾ ¾ ¾ = gl ê 1 + ¾ ¾ ¾ ¾ ¾¾ gl ú Multi-user Interference
Ne ë 3PG û

The expression for average BER is given by: Fig 1 System model for the MC DS-CDMA system
326 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

The block diagram for the transmitter for the kth
user is shown in Fig 2.
Consider a BPSK multi carrier coherent DS-
CDMA system with Ku independent users each
transmitting with power P. The users are simultaneously
sharing an available bandwidth BW=(1+a )/Tc, where
T c is the chip duration of a corresponding
single-carrier wideband DS-CDMA system, and
a (0 £ a £ 1) is the roll off factor of the chip wave-
shaping Nyquist filter. The available spectrum BW is
divided into (not necessarily contiguous) Mf equal
bandwidth subbands each of width BWMf approxi-
mately equal to the coherence bandwidth of the channel.
Each subband is assigned a carrier which is DS-
CDMA modulated with the same user information at
the bit rate 1/Tb and chip rate 1/(MfTc) (see Fig 2).
Each user is effectively assigned a specific periodic
code sequence of chip elements (+1, –1) and of
processing gain per subband PG´ = PG/Mf. We assume
deterministic subband PN codes with ideal auto
correlation function. The use of band limited (Nyquist
shaped spreading) waveforms with wave-shaping filter
transfer function denoted by H(f) guarantees that the
DS waveforms do not overlap.
The spectrum of the MC DS-CDMA signal is
shown in Fig 3.
2. Channel Model
Following the system design and modeling
assumptions of kando and well as their bandwidths are
chosen so that the separate subbands fade slowly and
nonselectively. Under these assumptions, the channel
transfer function of the lth subband for the kth user is
a ´k,l exp (jqk,l) where the {(a ´k,l}l=1Mf are the fading
amplitude RV’s and {q´k,l}l=1Mf are independent
uniformly distributed RV’s over [0,2P ].The average
fading power of lth subband is denoted by
W l´ = (a ´k,l)2 and is assumed to be independent of k.
3. Receiver
The receiver consists of a bank of Mf matched
filters followed by MRC (see Fig 4). Each of the
received modulated sub band carriers is first passed
through a band pass chip-modulated filter H*(f), then
coherently demodulated, sampled, despread and
summed, all these operations assume that the receiver
is correctly phase and time synchronized at every
branch. We denote by X(f) = H(f)H*(f) = |H(f)|2 the
overall frequency response of the chip waveshaping
Nyquist filter and assume that X(f) is a root-cosine
frequency response given by

Chip
pulse
Shaping

2cos(2P f1t+qk,l)
Data Transmitted
Sequence Multi carrier
Dn{k} DS signal
Impulse
modul H(f)
ator

Cn{k}
PN code
sequence
¥
S dh{k} Cn{k} d(t–MfTc)
–¥

¥
S dh{k} Cn{k} d(t–MfTc)
–¥
2cos(2P f1t+qk,l)

Fig 2 Transmitter for the Kth user

 S ANURADHA et al : SC DS-CDMA AND MC DS-CDMA SYSTEMS 327

With W=1/T'c=1/(M f T c) for multicarrier and

W = 1/Tc for single carrier

4. Performance Analysis
f1 f2 f3 fU
U
¾ Conditional SNR
D Te

(U –1)D Case 1: No partial band Interference:

The decision variable of kth user may be written
2
(U–1)D+ ¾ as the sum of a desired signal component and two
Te
interference/noise components [9]

Mf Mf
Fig 3 Spectrum of the MC DS-CDMA signal Eb Mf
rk = S wk,l rk,l = ± S wk,l a ´k,l ¾ + S wk,l (IMl+N)
l=1 l=1 M f l=1

1 W
¾ , 0 £ | f | £ ¾ (1–a ) Gaussian RV with Conditional mean and conditional
W 2

ì1 é
2W ë
æ 1 æ2p|f|
è 2a è W
ö öù
¾ ê1– sin ç ¾ ç¾ ¾ – p ÷ ÷ú ,
ø øû
variance are given by

Mf æ Mf ö Eb
X(f) =
í W W
¾ (1–a ) £ | f | £ ¾ (1+a )
2 2
E[rk| {a ´k,l}l=1 = ± ç S wk,l a´k,l ÷
è l=1 ø
¾ ¾
Mf

î W
0, ¾ (1+a ) £ | f |
2
Mf Mf
N0
var (rk | (a ´k,l}l=1 = S (wk,l)2 ¾
l=1 2

Sample at
H*(f–f1) T=nMfTc
+H*(f+f1) 1 PG´–1
LP ¾ ¾ S() Wk,l
F PG´ n=0

{Cn {k}}

Removes R{k}
second harmonic
of carrier

Sample at
T=nMfTc
rk
Wk,Mf
1 PG´–1
H*(f–fMf)
+H*(f+fMf)
LP ¾ ¾ S()
F PG´ n=0

{Cn{k}}
2cos(2P f1t+qk,l Mf)

Fig 4 Block diagram of reciever

328 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

Maximum conditional SNR of single-carrier system is 1 p/2 Mf bl

given by Pb(E) = ¾ ò P Mg´ (– ¾ ¾ ¾2 ¾ ) df
p 0 l=1 Mf sin f
Eb L é (K u–1)W T Where Mgl´ denotes the MGF of the lth subband SNR/
SNRmax ({a k,l)Ll=1) = ¾ S ( a k,l)2 ê1 +¾ ¾ ¾ ¾
N0 l=1 ë PG bit .
–1
æ a ö ù
ç1– ¾ ÷g1 ú æ bl ö æ bl gl ö –m
è 4 ø û Mgl´ ç – ¾ ¾ ¾ ¾ ÷ = ç1+ ¾ ¾ ¾ ¾ ¾÷
è Mf sin2f ø è mMf sin2f ø
Case 2: Partial band interference. m
1 p/2 l–1 æ mMf sin2f ö
Consider the presence of PBI jammer model as a Pb(E) = ¾ ò P ç ¾ ¾ ¾ ¾ ¾ ¾2 ¾ ÷ d f
p 0 L=0 è gl bl +mMf sin f ø
band limited white Gaussian noise with bandwidthWJ =
BWMf is given by
RESULTS

ì ¾h2 , f – ¾W2
j
j
j W
£ | f | £ fj + ¾ j
2 SC DS-CDMA (for different values of m)
Snj (f) = í 100
î 0, elsewhere
10-1
Where fj denotes jammer carrier frequency. Average BER (dB)
10-2
The decision variable of Kth user may now be
written as sum of desired signal component and 3 10-3
intereference /noise components [9].
10-4
Mf Mf Mf
æ ö Eb
rk = S wk,l rk,l = ± ç S wk,la ´k,l ÷ ¾ + S wk,l
l=1 èl=1 ø Mf l=1 10-5
0 5 10 15 20 25 80
(IJl + IMl+N) Average SNR (dB)

where Ijl is the Gaussian PBI present in lth subband Fig 5 Ber versus SNR of SC DS-CDMA for different m
with variance values

NJ 1 ¥
s2Jl º ¾ l = ¾ ò [Snj (f–fl) + Snj (f+fl)] X (f) df
2 2 –¥ MC DS-CDMA (for different m values)
100
The Kth user maximum conditional SNR is given by
–1 10-5
é K –1 a Nj ù
bl = ê1 + ¾ u¾ ¾ (1 + ¾ ) gl´ + ¾ i¾ ú
ë Mf PG 4 N0 û
Average BER (dB)

10-10
Nj JSRvgv
Where ¾ = ¾ ¾ ¾ ¾ ¾ v = 1,2,…..or Mf
N0 PG´ (1+a )
10-15

Where JSR v = n J W J /W vE b / T b represents the

interference (jamming) to SNR in the vth subbands 10-20
0 5 10 15 20 25 80
Average BER Average SNR (dB)

The Pb(E) of MC DS-CDMA over the MRC Fig 6 Ber versus SNR of MC DS-CDMA for different m
combiner is given by values
 S ANURADHA et al : SC DS-CDMA AND MC DS-CDMA SYSTEMS 329

MC DS-CDMA (for different m values) CONCLUSIONS

100
In this paper, the concept of generalized SC DS-
10-5 CDMA and MC DS-CDMA was understood and the
BER expressions for the system were derived and
Average BER (dB)

10-10 verified. Due to the presence of multiple sub-carriers,

the analysis of the generalized MC DS-CDMA system
is much more complicated than a DS-CDMA system
10-15
although similar methods of analysis apply to both
schemes. The performance analysis for a system
10-20 using MRC at the receiver was also carried out.
Following is a summary of the observations and results
10-25 that were obtained.
0 5 10 15 20 25 80
Average SNR (dB) 1) The BER expressions for the generalized SC
DS-CDMA and MC DS-CDMA scheme
Fig 7 Ber versus SNR of MC DS-CDMA for different K using Maximum Ratio Combining on a
values Nakagami-m fading channel was derived. Since
the analysis was not tractable for a channel
MC DS-CDMA (for different m values) with exponential MIP, a flat MIP was assumed
100 for the channel.
2) As per the graph between BER vs SNR for
different values of ‘m’ for SC-DS CDMA
10-5
and MC-DS CDMA, as value of m increases
Average BER (dB)

bit error rate decreases.

10-10 3) As the value of JSR (Jamming Interference)
increases Bit Error Rate increases too for
MC-DS CDMA.
10-15 4) As the number of users k increases the bit
error rate increases for MC-DS CDMA
10-20 5) For the same values of m, BER for MC DS-
0 5 10 15 20 25 80 CDMA is smaller when compared with SC
Average SNR (dB) DS-CDMA
Fig 8 Ber versus SNR of MC DS-CDMA for different JSR REFERENCES
values
1. B Sklar, Digital communications, 3rd edition,
Pearson Education series, 2001.
Comparisions B/W SC and MC DS-CDMA
100 2. M K Simon, J K Omura, R A Scholtz & B K Levitt,
Spread Spectrum Communications Handbook, 2nd
ed New York: McGraw-Hill, 1994 Originally
published in three parts as Spread Spectrum
10-5
Communications Computer Science Press, New York,
Average BER (dB)

1984.
10-10 3. M S Alouini, M K Simon, & A J Goldsmith, A unified
performance analysis of DS-CDMA systems over
generalized frequency-selective fading channels,
Proc IEEE Int Symp Inf Theory (ISIT '98),
10-15
Cambridge, MA, August 1998, p 8.
4. G L Turin, Communication through noisy, random-
multipath channels, IRE Natl Conv Rec, March 1956,
10-20
0 5 10 15 20 25 80 pp 154-166.
Average SNR (dB) 5. H Suzuki, A statistical model for urban multipath
propagation, IEEE Trans Commun, vol COM-25,
Fig 9 Comparisions B/W SC & MC DS-CDMA for
July 1977, pp 673-680.
different m values
330 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
6. H Hashemi, Impulse response modeling of indoor
radio propagation channels, IEEE J Sel Areas
Commun, vol SAC-11, September 1993, pp 967-978.
7. S A Abbas & A U Sheikh, A geometric theory of
Nakagami mobile radio channel with physical
interpretations, Proc IEEE Veh Technol Conf
(VTC'96), Atlanta, GA, April 1996, pp 637-641.
Authors
Anuradha received the BTech
degree from the Acharya Nagarjuna
University, Guntur, in 1998 and the
MTech degree from the University of
Sri Venkateswara, Tirupati, in 2001, both
in Electronics and Communication
Engineering. She is currently pursuing
the PhD degree at the department of
Electronics and communication
Engineering, Andhra University,
Visakhapatnam.
Her research interest is in Wireless and Mobile
Communications. Currently she is working as a lecturer in V R
Siddhartha Engineering College, Vijayawada.
* * * Engineering at Birla Institute of
Sri Gowri received the BE degree
from Andhra University, Visakhapatnam
and the MTech degree from the JNT
University, Kakinada, both in Electronics
and Communication Engineering. She
received the PhD degree from the
Department of Electronics and
Communication Engineering JNT
University Kakinada. Her research
interest is in Wireless and Mobile
Communications.Currently she is working as a Professor in V R
Siddhartha Engineering College, Vijayawada.
* * * *
8. T Eng & L B Milstein, Coherent DS-CDMA
performance in Nakagami multipath finding, IEEE
Trans Commun, vol COM-43, February-March-April
1995, pp 1134-1143.
9. S Kondo & L B Milstein, Performance of Multi carrier
DS CDMA Systems, IEEE Transactions on
Communications, vol 44, pp 238-246, Feb 1996.
K V V S Reddy is a professor in
Department of Electronics &
Communication, Andhra University,
Visakhapatnam. His research interest is
in Wireless & Mobile Communications,
ANN for RF & Microwaves. He
published several papers in reputed
International & National Journals
* * *
K Sri Rama Krishna received the
BTech degree from JNTU, College of
Engineering, Kakinada, Andhra Pradesh
and MS in Electronics & Control
Technology and Science, Pilani, Rajasthan,
India both in Electronics and
Communication Engineering. He received
the PhD degree from the department of
Electronics and Communication
Engineering Andhra University, Visakhapatnam. His research
interest is in Microprocessors, Digital system Design, Artificial
Neural Networks, ANN for RF & Microwaves. Currently he is
working as Professor and Head of the Department of Electronics
& Communication Engineering, V R Siddhartha Engineering
College, Vijayawada, A P.
* *
 SHORT PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 331-334

Design of a FIR Filtering Core for

High Speed Application
M ARIF
Department of Electronics and Communication Engineering, National Institute of
Technology (Deemed University), Kurukshetra 136 119, India.
e-mail address: arif_nitk@yahoo.co.in

The paper presents high speed realization of transposed direct form FIR filter core by
using sampler switch with its low resistive and capacitive components value that is to be
done in the core itself. It enables computationally efficient implementation of some of the
type of FIR filters by saving the time of the individual processing blocks in the path of the
final response due to these smaller value of components and also the computational time
can be saved to enhance speed by placing few blocks in parallel. The significant reduction of
resistance and capacitance of the sampler switch with their typical values enable to increase
speed by a large multiplying factor. The designed core will be suitable for the real time digital
signal processing applications.

1. INTRODUCTION The VHDL supports a digital system and it can

be modeled as a set of interconnect components and

T HE advent of portable computing has led to a

significant increase in the development of the
various cores for the implementation of any digital
filtering operation for the digital signal processing
applications or it can be used as a digital signal
processor in a chip form for the dedicated application
[1]. The power dissipation is becoming a limiting factor
in the realization of the core [2]. It can be minimized
by reducing capacitance that can be traded to high
speed [3]. A core may require a low and high
throughput [4]. The propagation delay is becoming a
limiting factor in the realization for the designing of
core. In many DSP filtering algorithms, multiplication
is a major function for implementing the operation.
The core consists of the following components
and integrated into one unit i.e. in chip form and they
are connected together into the chip in parallel by
providing same clock to enhance the speed.
• Constant coefficient memory
each component can be modeled as a set of
interconnect subcomponents [5]. Several forms of
FIR filter architectures for hearing aid application is
implemented and the development of the core are
characterized by low power dissipation, low area
consumption and flexible with the use of VHDL [6].
The designing of the core with VHDL is guaranteed
[7].
2. ANALYSIS
The transposed form of the FIR filter structure
is shown in the Fig 1 with its applied input, branch
transmittance, delay element, adder, and finally with
the obtained output [8]. The transposed structure is
concerned with the general form of an IIR system,
and it is expressed by the following difference
equation.
N M

• Input data memory y (n) = – S ak y(n–k) + S bk x(n–k) (1)

k=1 k=0
• Arithmatic and logic unit
• Timing and control unit FIR system is obtained by setting ak = 0, k =
• Register x 1,2,3,………..,N, and also has a transposed direct
• Register b form as shown in the Fig 1.
• Accumulator This transposed form realization may be described
by the set of following difference equations and shown
Paper No 126-C; Copyright © 2007 by the IETE. in the Fig 1.
331
332 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

x(n)

b0 b1 b2 bM–1 bM

y(n)
wM(n) w2(n) w1(n)

Fig 1 Transposed FIR filter structure

wM (n) = bM x (n) (2) in the form of the high speed core are connected by
the same timing and control unit as shown in Fig 2.
wk (n) = wk+1 (n–1) + bk x(n) k = M–1, M–2,....1 (3) First of all, the input data and various coefficients
corresponding to the particular input are stored in the
input data and constant coefficient memories, as shown
y (n) = w1 (n–1) + b0 x(n) (4)
in the Fig 2 along with the other individual components.
The timing and control unit will provide the simultaneous
3. ARCHITECTURE clock to these memories and both the data’s will be
available to the arithmetic logic unit, which will perform
The system architecture of FIR filter is shown in the multiply-shift-add operation as well as accumulator
Fig 2 for the designing of high speed core. Most of the to store the intermediate result for the further processing
individual blocks in the hardware of the designed filter through the next set of data’s.

x(t) X
Sample- x(n) Register x
and-hold ADC Input Data
circuit Memory Arithmatic Y y(n)
and Logic DAC LPF
Unit

Register b
h(t)
Sample-
b B Constant
and-hold ADC Coefficient Accumulator
circuit Memory

Timing and control unit

Fig 2 System hardware of the FIR filter

 M ARIF : DESIGN
4. SIMULATION RESULTS
The final response of the system is expressed
mathematically in the following form.
M structure.
y (n) = S bk x (n–k)
k=0
= b0 x(n) + b1 x(n–1) + ... + bM x (n–M)
First of all, all the data’s are stored in the two
memories i.e. input data memory contains the binary
information after conversion by sample-and-hold circuit
and ADC corresponding to the input signal in analog
form represented by x(t), and, in the similar fashion,
the constant coefficient memory also contains the
same type of the information corresponding to the
coefficient values that are present in the h(t) [9].
Now, the various data’s are suitable for arithmetic and
logic circuit in the binary form. This circuit will perform
all the computations i.e. multiplications, additions, and
shiftings operation. The intermediate results are stored
in the accumulator, which will be required to obtain the
final digital output values. After getting it, this calculated
value will pass through a DAC to convert into discrete
analog form, which will be the desired output value
[10].
Let the input data x(n) has the following discrete
values after converting x(t) by sample-and-hold circuit.
x(n) = {0.1, 0.2, 0.3, 0.4}
and the coefficient corresponding to impulse response
has the following values as per the input values
accordingly.
b = {0.01, 0.02, 0.03, 0.04}
These above data’s are taken arbitrarily for getting
time output signal and the input data may be taken for
any real time signal processing application i.e. speech
signal, audio signal, vedio signal, any other
communication signal, and any biomedical signal etc.
From equation (5) for M = 3, the output equation will
be as below.
y(n) = b0 x(n) + b1x (n–1) + b2 x (n–2)+b3x(n–3) (6)
The final discrete output obtained after the
processing is {0.001, 0.004, 0.010, 0.020}, which is
determined by the MATLAB Software using
convolution to the maximum length of the one of the
signal between input signal and impulse response in
the discrete domain and the analog output y(t) will be
reconstructed by the LPF.
OF A FIR FILTERING CORE 333
Compared with the direct form structure, the direct
form structure will also obtain the same value of the
output as obtained for transposed form structure. As
far as, the analysis contains the similar steps with the
same equation and difference observed only in the
(5)
5. CONCLUSION
In this paper, a high speed FIR filtering core for
transposed form structure is presented. This design
approach can also be applied for the implementation
of other configuration of digital FIR filter and that will
equally work well as far as high speed implementation
is concerned. The digital portion of design can be
easily simulated by VHDL software and also it can be
downloaded on the FPGA package to allocate the
various functioning of the pins for making core in the
IC industry where micro chips are being fabricated.
REFERENCES
1. Ahmet Teyfik Erdogan & Tughrul Arslan, On the
Low-Power Implementation of FIR Filtering
Structures on Single Multiplier DSPs, IEEE
Transactions on Circuits and Systems-II: Analog
and Digital Signal Processing, vol 49, no 3, March
2002.
2. A T Erdogan & T Arslan, Low power multiplication
scheme for FIR filter implementation on single
multiplier CMOS DSP processors, Electronics
Letters, vol 32, no 21, 10th October, 1996.
3. V A Bartlett & E Grass, A Low power Asynchronous
VLSI FIR Filter, IEEE 2001.
4. A Chandrakasan, S Shang & R W Broderson , Low-
Power CMOS Digital Design, IEEE Journal of Solid-
State Circuits, vol SC-27, pp 473-484, 1992.
5. J Bhasker, VHDL Primer, Third Edition, Pearson
Education Asia, India.
6. E P Zwyssig, A T Erdogan & T Arslan, Low power
system on chip implementation scheme of digital
filtering cores, Low Power IC Design Seminar, 19
January 2001, London, UK.
7. Artur Krukowski & Izzet Kale, Simulink/Matlab-to-
VHDL Route for Full-Custom/FPGA Rapid
Prototyping of DSP Algorithms, Matlab DSP
Conference (DSP'99), Tampere, Finland, 16-17
November 1999.
8. John G Proakis & Dimitris G, Manolakis., Digital
Signal Processing Principles, Algorithms, and
Applications, Third Edition, Prentice Hall of India
Private Limited, New Delhi- 110001, India, 2004.
9. Willis J Tompkins, Biomedical Digital Signal
Processing, Prentice-Hall of India Private Limited,
New Delhi, India, 2006.
10. Simon Haykin, Communication System, Third
Edition, Printed by Thomson Press (India) Ltd., John
Wiley & Sons.
334 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007

Author

M Arif born on 11th January, 1965,
at Ghazipur, UP, India. He received BE
in (Electronics Engineering) from
Gorakhpur University and MTech in
(Electronics and Communication
Engineering) from Kurukshetra
University, India in 1987 and 1997
respectively. He joined in Electronics
and Communication Engineering
Department, Regional Engineering
College Kurukshetra, Haryana, India as Lecturer in 1988, and
presently serving in Electronics and Communication Engineering
Department. National Institute of Technology (Formerly Regional
Engineering College) Kurukshetra, Haryana, as Lecturer in Senior
Scale. His research interest is Digital Signal Processing. He is life
member of ‘The Institution of Society for Technical Education’.
* * *
 SHORT PAPER
IETE Technical Review
Vol 24, No 4, July-August 2007, pp 335-337

Interactive Education through ECT —

A Focus on Rural India
S ARUMUGA PERUMAL
Head, Department of Computer Science, S T Hindu College, Nagercoil 629 002, India.
email: visvenk@yahoo.co.in

Science and technology education have enjoyed a meaningful partnership across most
of this century. The work of scientists embraces an array of technologies, and major
happenings in science are often accompanied by complicated applications of technology. As
a result, a complete science education has involved a commitment to the inclusion of
technology, both as a tool for learning science content and processes and as a topic of
instruction in itself. Due to the fast development of information highway there is a remarkable
raise in the quality of education throughout the world. With the invention of new technologies
like high speed computer networks and multimedia computers, there is an increasing
awareness that direct face to face teaching is not the only possible mode of teaching in a
school system. There is a demand for preparing high quality multimedia course materials
across all faculties, which can be used by learners of the course who either cannot attend
the live lectures or prefer to study in an off-line mode. This paper discusses and suggests
some ideas and necessary changes in implementation of independent learning of Science
and Technology education through e-learning concepts in education which leads to improve
the education culture in rural area.

INTRODUCTION Structured approach is an approach that tells us

“What to teach?” and not “How to teach?”. This kind

T HE explosion of digital technology has created a

revolution in development of information highway.
The flexibility, speed, and storage capacity of
contemporary desktop computers is causing science
educators to redefine the meaning of hands-on
experience and rethink the traditional process of
teaching. The challenge facing both science educators
and science teacher educators is to evaluate relevant
applications for information technologies in the science
curriculum. At the same time, instruction utilizing
information technologies must reflect what is known
about the effectiveness of student-centered teaching
and learning. We choose the Core Knowledge Sequence
as the framework for much of the curriculum that
we’ve developed in science and Technology.
Components for Education program
The three main elements of this Program are
• Course structure
• CDROM materials
• Web-based education support system.
Paper No 126-B; Copyright © 2007 by the IETE.
of approach has the following characteristics
• Usefulness
• Simplicity
• Teaching ability
The structured approach has the following
approaches
• Structured approach makes use of carefully
selected and graded language items
• Structured approach lays emphasis on
presentation
• Structured approach stresses habit formation
• Structured approach makes the learner as
active participant.
The course structure is prepared effectively so
that it will coincide with the recent trends in their
respective fields. The course structure is innovative
so that they can be motivated for research in their
fields. The curriculum is designed to make sure that
the students get the core knowledge they need—that
they learn what they need to know to succeed and
thrive. The training material will improve the creativity
of the new fields. So a well advanced planning is

335
336 IETE TECHNICAL REVIEW, Vol 24, No 4, July-August 2007
needed in developing course structure. Before
preparing course structure it must be discussed with
subject experts for standards. Good courseware must
give guidance to candidate and also provide a rich
variety of background material when ever such is
needed. The advent of digital libraries provides not
only the most significant chances for computer-based
training but also the extensive background reservoir of
material needed in many situations.
When designing the course structure in teaching
science and Technology, the following three stages of
teaching structure must be adopted for better
understanding of the subject
• Presentation
• Practice
• Application
Presentation is a one-way traffic. The main aim
of presentation is to make clear-cut meaning of the
structures to the students. The presentation involves
oral teaching and situation teaching. Structures are
first presented orally and then through reading and
writing. Structures are taught in situations so that they
become real and meaningful for the students. Showing
objects, pictures and charts, matchstick figures, actions
and gestures and verbal situations creates different
situations.
The importance of the second stage practicing is
to reinforce the structure in the minds of the students.
The importance of third stage application is to ask
the students to apply the structure in new situations.
The CD-ROM materials are prepared, with the
help of multi-media technology by subject experts so
that it will motivate their learning interest. The materials
are prepared by combining short phases of presence
with interleaved media-based self-study blocks; they
are well suited for web-based education. An innovative
approach is taken by integrating all course elements-
acoustic and written information, diagrams and figures,
animations, simulations, video clips, and laboratory
exercises by presenting in a CDROM materials. The
material should include animation system. Animation
is a best tool for teaching dynamic phenomena.
Animations are used so that they will promote interest
in further studying.
The web based education support system and
CD-ROM materials are very useful to improve
candidate’s knowledge and skills, regardless of prior
knowledge and skills. CD-ROM materials significantly
enhance the effectiveness of teachers creating materials,
especially when the CD-ROM was used for self-
study. The web based support system helps for
interaction and evaluation. The interaction is achieved
in the web by chatting or by mailing. Some times
advanced technology of video conferencing may also
be utilized.
Evaluation and certification can be done by
evaluating individual students by giving assignments.
Assignments can be submitted from home and discussed
with the subject experts or with peers in an interactive
distant mode.
Cultural change
Though the students learn a subject through
e-education the culture of a country should never be
affected. Every country esteems its own culture and
so the fast development in education should go along
in a balanced way without neglecting its own culture.
A country will look more gorgeous if it only shines
with blooming flowers of cultural values mingled with
developed education. To succeed in modern society,
you have to know a lot. And you have to have this
knowledge instantly accessible at your mental fingertips.
In the eyes of sociologists culture is the sum total
of human experiences. It comprises of all human
traits, determined by a cluster of values. These cultural
values are manifested in our traditions, customs,
ideologies, and even in day-to-day life. Each group of
people beholds different but certain values and attitudes.
A change in attitude alone can bring about a change in
the acquisition of cultural values. Languages on its
part are more helpful for the development of culture.
Every language is a product of its own culture. All the
chunks and expressions we come across help us to
understand varied concepts. It is none other than a
language, which assists each and every one to think
and speak in terms of concepts. It is the language that
has a great control over the thought process of an
individual. The younger generation is blessed with the
cultural elements through the decorative pipes of the
language.
The disadvantages of class room teaching are
• The classroom pace is moving too slowly for
the gifted child.
• The classroom pace is moving too fast—the
child needs 55 minutes to master a concept
but the bell always rings at 50.
• The child feels left out or bullied.
• The child can’t sit still in class or has a learning
challenge.
• The child falls outside of the middle range in
the classroom.
 S A PERUMAL : INTERACTIVE EDUCATION THROUGH ECT
• The class is overcrowded.
• The child is homebound or traveling for sports
Factors to be concentrated on to improve
the quality of e-learning
1. Technology should be introduced in the context
of science content.
2. Technology should address worthwhile science
with appropriate pedagogy.
3. Technology instruction in science should take
advantage of the unique features of technology.
4. Technology should make scientific views more
accessible.
5. Technology instruction should develop
students’ understanding of the relationship
between technology and science.
CONCLUSION
The impact of digital technologies on science
teacher education is more pervasive than any curricular
or instructional innovation in the past. The impact can
be felt on three fronts.
Author
S Arumuga Perumal is working
as Reader and Head of the Department
of Computer science in South
Travancore Hindu college, Nagercoil,
Tamilnadu, India for the last 19 years.
He has completed his MS (Software
systems) in BITS, Pilani, Rajasthan,
MPhil Computer Science degree in
Alagappa University, Karaikudi and he
did his PhD in Computer Science in
Manonmanium Sundaranar University. He is a senior member
337
First, digital technologies are changing the ways
teachers interact with students in the classroom.
Second, teacher education courses are not only
influenced by new K-12 curricula, they are also
influenced by instructional approaches that
incorporate a variety of digital technologies.
Technological applications go beyond K-12 curriculum
to the delivery of college level content. For instance,
faculty and students explore web resources for
educational statistics or education-related reports and
course resources.
Third, faculty and students alike are interacting in
new ways afforded by digital technologies. Faculty
and students have virtual discussions related to course
content, advice, and counseling in a wide variety of
times and places via email, cell phones, pagers, and
features of the web. Faculty and students now produce
documents with more information and in far more
diverse formats as a result of desktop publishing,
online libraries and databases, and file transfer
capabilities. The pervasiveness of digital technologies
motivates a thorough review of technological impacts
on curriculum and instruction in science teacher
education.
of Computer society of India and member of IEEE. He is
involved in various academic activities. He has attended number
of national and international seminars, conferences and presented
number of papers. He has also published number of research
articles in national and international journals. Dr Perumal is a
member of curriculum development in universities and
autonomous colleges. He is guiding research scholars from
various Universities. His area of research is Digital Image
compression, Data mining and Biometrics.
* * *
 ECIL
The only multi- technology company that provides
Total Systems Solutions
Serving the needs of Sectors of National Importance like
Atomic Energy, Defence, Space, Security, Power, Telecommunications,
Oil/Gas, Steel, Process, Education and E-Governance towards achieving
Self-Reliance

Offers: Control & Automation, Computers, Communication Systems, Antenna

Systems, Strategic Electronics, Telecommunication Products, Encryption Systems,
Instrumentation Systems, Integrated Security Solutions, Electronic Voting Machines
and more.

Electronics Corporation of India Limited

Corporate Business Development Group
PO ECIL, Hyderabad - 500 062, India
E-mail: cbdg@ecil.co.in web: www.ecil.co.in
Registered with the Registrar of Newspapers for India, Delhi under Regn. No. 42814/84

CALL FOR PAPERS

Special Issue of IETE Journal of Research

on
‘Microwave Circuits and Systems’

Microwave technology has rich history dating back to the 19th Century and fundamental discoveries in
electromagnetic with a very strong and elegant foundation established by Maxwell with his famous equations. The
thrust of the research supporting microwave technology has been in establishing the basic elements to support the
growth and development of the application and also to extend the usable spectrum to higher and higher frequencies.
Although major thrust and applications of Microwave Circuits are for the Defense field, the growth of Satellite and
Wireless Communication in last few decades has proved the utility of this band for civilian applications also. The
purpose of this special issue on ‘Microwave Circuits and Systems’ is to exhibit the state-of-the-art research in our
country in this area and also to give future direction for research to young scientists and engineers.

Topics of interest for the Special issue are:

• RF Circuit Design
• CAD of Microwave Circuits
• EMI/EMC
• Microwave Integrated Circuits
• Satellite Communication Systems
• Passive Microwave Devices
• Radar Components
• Super Components for EW systems
• Feed Networks for Phased Arrays
• TR/ATR Modules
• Air Borne/Space Borne Microwave Circuits and Systems
• MEMS
• Circuit Synthesis and Simulation
• Microwave Measurement techniques

Conceptual review papers covering the theme with less mathematics are invited. Research papers with original contribution
in any of the above area with more stress on ‘Results and Discussion’ are most welcome. Articles of about 4000 to 5000
words including figures, graphs, tables and references typed neatly in A4 size paper in double space as per IETE
guidelines to author (given in website www.iete.org under the heading publication) should be sent to the Guest Editor at
the address given below. Two hard copies along with a soft copy in MS word should be sent to:

Prof V M Pandharipande,
Guest Editor, IETE Journal of Research,
Director,
Centre for Excellence in Microwave Engineering,
Department of Electronics & Communication Engineering,
Osmania University, Hyderabad - 500 007 (AP).

Email: ceme_uceou@yahoo.com;
Ph:040-27071273,27682234

Last Date for receiving the papers: 30th October 2007; Papers to be reviewed: 31st December 2007;
Issue to be published: March 2008.

Printed and Published by Brig (Retd) V K Panday for the Institution of Electronics and Telecommunication Engineers,
2, Institutional Area, Lodi Road, New Delhi 110 003 (India) and printed at the Shivam Offset Press, A-12/1, Naraina Indl. Area,
Phase I, New Delhi-110 028 (India). Dy Managing Editor : CDR A P Sharma, IN (Retd) • Copyright © 2007 by the IETE