You are on page 1of 6

1

Non-Preemptive Queueing-Based Performance


Analysis Of Dynamic Spectrum Access For Vehicular
Communication System Over TV White Space
Mst. Najnin Sultana Kyung Sup Kwak
Graduate School of IT & Telecommunications, Graduate School of IT & Telecommunications,
Inha University, Inha University,
Incheon, South Korea Incheon, South Korea
Naj250@gmail.com kskwak@inha.ac.kr

Abstract--- Recently Television White Spaces are used without Dynamic spectrum access (DSA) was first demonstrated in
license according to FCC rule. It is based on dynamically 2006 by the Defense Advanced Research Projects Agency
selection of available TV spectrum (white spaces). It gives a (DARPA) and Shared Spectrum Company (SSC) of Vienna,
possibility of new sets of applications. In vehicular wireless VA, which enables users of virtually any modern radio device
communication system, these white spaces can easily be used on to utilize dynamic spectrum access techniques and thereby
an opportunistic base. To model a multi-access multiuser dramatically improve spectrum efficiency, communications
architecture and analyze network performances in both wired reliability, and deployment time. A quantitative assessment of
and wireless system queueing theory has been massively used. In DSA was made in [5].
this paper we modeled a vehicular dynamic spectrum access
(VDSA) system by using queueing theory which uses vacant TV Several licensed frequency bands could facilitate DSA-
channels for communication. Dynamically assigned allowed based vehicular communication networks. But the UHF
available channels are modeled as servers. To efficient use of television frequency range has been identified as a primary
these limited number of channels we defined some non- candidate frequently due to its relatively static frequency
preemptive priority classes among users. The availability of channel usage by incumbent TV broadcasters. The band
servers is occurred in a time-location phenomenon. This paper ranges from 470 to 698 MHz (Channel 14 through Channel
examined the feasibility analysis of vehicular dynamic spectrum 51) [6].
access performance through TV white spaces via multi server
multi priority non-preemptive queueing theory. Both M/M/m and The FCC has recently explored the use of unused 700MHz
M/G/m models are employed to evaluate the probability of all spectrum for white space applications. These are unlicensed
channels are busy, the amount of channel utilizations by different applications that actively avoid interference with licensed
priority class users and also to calculate the transmission applications in the same band. On February 17, 2009, the FCC
latencies. released the final rules for “Unlicensed Operation in the TV
Broadcast Bands” [7]. TV Band Devices (TVBDs) are divided
Keywords--- VDSA, Non-Preemptive Queueing, M/M/m, M/G/m, into two categories: fixed and personal/portable.
UHF TV Channels. Personal/portable devices are restricted to channels 21 – 51
(except Channel 37) and are allowed a maximum EIRP of 100
I. INTRODUCTION mW on non-adjacent channels and 40 mW on adjacent
Intelligent transportation promises the convergence of channels. Channel 37 is used for vehicular communication.
modern information technology (IT), communication For portable device, a Cognitive Radio Standard has declared
infrastructure, and sensing technology with standard in these channels [8].
transportation systems to improve safety, reduce transportation In order to utilize vacant UHF TV channels for vehicular
times, minimize vehicle wear, and optimize fuel consumption. communications, all vehicles must follow the rules defining
Applications of ITS range from vehicle-based collision TV band devices (TVBDs) published by FCC on November
avoidance to general traffic monitoring and toll collection. 14, 2008. The vacant channels in this frequency range are
Dedicated short range communication (DSRC) is a robust shared among wireless microphones and TVBDs [9].
communication system to encompass all such vehicle-to-
vehicle, vehicle-to-infrastructure communication [1]. In 1999, To model multiple access schemes or transmission delay in
FCC allocated 75 MHz of spectrum in the 5.9GHz band for a communication system, queueing theory has been
DSRC. The standard for this system is IEEE802.11p. 5.9 extensively used [10, 11]. In Cognitive Radio system, some
DSRC provides data rate up to 27 Mbps and covers the range works have been made based on queueing theory to calculate
from 100 to 1000 meters [2]. In addition, DSRC has a new throughput and delay and dynamic channel allocation [12, 13].
interest in vehicular-based communication in the 700 MHz The study of queueing models with service interruptions is
band. The 700MHz frequency band offers a significant very old fashion. Nowadays, different kinds of traffic models
coverage advantage over current 5.9 GHz DSRC have been employed in queueing system to model the primary
implementations. At an identical transmitter power, a low- users who have the license the spectrum and the secondary
frequency signal will have greater range than a high-frequency users who are unlicensed but authorized users use the vacant
signal due to 2 Journal of Electrical and Computer channels opportunistically. In [14], an M/G/1 queue is used to
Engineering decreased free space attenuation and lower model a system containing one primary user and multiple
absorption by various building materials and obstructions [3]. secondary users, where the secondary users function as
cooperative forwarder for primary users. A multiserver-
multipriority preemptive model is employed in [15] where

978-1-4577-1177-0/11/$26.00 ©2011 IEEE 43 ICUFN 2011


2

when among secondary users there are also some priorities.


When the higher priority secondary users enter the systems,
they treated as same as like primary users, the lower priorities
in service should be quit the channel. So, it increases the
transmission latencies for lower priority ones.
In [13] they used a dynamic learning algorithm (DSL) to
assign channel among users based on different utility
functions, delay deadlines, traffic priorities in Cognitive Radio
networks. A G/M/K/0 queueing model is employed in [16]
which multiplexed the arrival processes formed by primary
and secondary users. However, the total bandwidth is assumed Fig.1. Opportunistic vehicular spectrum access in vacant TV channels.
to be fixed here.
In this paper we applied multi-server, multi-class non-
preemptive queueing model to evaluate the quantitative
measure of the available resources through TV vacant
channels in the case of VDSA system. In [15], the authors
generated a spectral map of these white channels along I-90 in
the states of Massachusetts in America based on a geo-
location database approach. We use the same geo-location
database approach from [15] and set two priorities based on
arrival rate and service rate among secondary users but higher
priority users don’t preempt the lower ones.
The rest of the paper organized as follows. In section II we
model the VDSA system at the system level using queueing
theory. After that in III we give the exact and approximate
solutions for the different performance measurement
parameters. Section IV shows the corresponding results and Fig.2. TV Channel availability at different locations along I-90 in the
state of Massachusetts, USA [6].
analysis of VDSA system in TV white bands along I-90.
Finally in section V we conclude our paper. MATLAB is used After some assumption and abstractions a vehicular
as a source code and to analyze the system performance. communication system can be modeled as a virtual queueing
system. The queue is only used to approximate the process,
II. SYSTEM MODEL not representing a real entity. We don’t consider each vehicle
For traveling vehicles across location-varying vacant TV as a customer. Rather we consider a pair of communication
channels, to assign vacant spectrums dynamically is an entities as a customer. A communication set is defined as the
opportunistic multiple access scenarios shown in Fig.1. set of transmitters and receivers performing wireless
transmission regardless the transmission is one-to-one, one-to-
To use UHF TV channels, instead of the entire UHF TV many, or possibly multiple-to-multiple if cooperative
range, the authors in [6] selected 600 - 750 MHz and captured transmission is employed. The bandwidth request from each
4 sweeps per minute on average along the length of I-90 • The communication set is considered as an entity that enters the
sweep index increases from 1 on the top left corner of Fig. 2 queue for service. Fig. 3 demonstrates the virtual queueing
and indicates the traveling along 1-90 in the state of MA from system.
west to east. The western most point in their study was West
Stockbridge, MA and the eastern most point was Boston, MA. Then the available bandwidth resources within the
Fig. 2 shows the spectral temporal database of available transmission range of the vehicular communication set are
channels. modeled as server. There are 30 TV channels available in total
for access of portable TVBD. So, we consider 30 servers here.
To predict the performance measure based on this spectral- Each server is assumed to handle one communication link at a
temporal database and the estimated number of cars in the time.
proximity we consider a time-location snapshot when the
vehicles are travelling along the highways instead of the total Due to very short transmission time of a packet in compare
hours of driving. Because depending on location as well as with snapshot; we assumed a static channel here. Within one
time in each direction the number of cars per kilometer will be snapshot there is no service interruption is also considered
varied. Each time-location snapshot is about one minute in during the travelling location change with different TV
time and one kilometer in distance. Here, performance channel availability.
measures are made on a transmission of a typical packet. The transmission time of each communication link is
We consider the transmission range is as same as defined modeled as the service time of each communication set. The
in DSRC, 300 meters. And it is assumed that at each instance response time of delivering a packet includes the transmission
the bandwidth requirements for all types of communications time as well as waiting time when all channels are busy. Few
including both point-to-point and broadcast and also vehicle- times are also spent during channel sensing and searching the
to-vehicle and vehicle-to-infrastructure communications are channel availability list.
same.

44
3

Fig. 4. The Basic model of priority queue (two classes-two servers).

 = 1 −   +   
 (1)
   
Fig.3. Packet-based queueing model for VDSA in vacant UHF TV bands. where, γ be the length of time from an instant when all servers
are busy, an arbitrary customer enters service and there are no
CSMA/CA with moderate traffic load is assumed to be class-a customers in queue, until the first moment there after
employed by the transceiver on each vehicle to prevent that the number of busy servers decreases to m-1 or a
collision. In the case of heavy load the performance of customer from one of the classes k, k+1,…,n enters service.
CSMA/CA is not so good so it will not be give an accurate The Laplace-Stieltjes transform (LST) [21] of γ is given by,
result for VDSA multiple access system and may give an
overly optimistic prediction of performance measure. In the
case of light traffic load of data packets, the packet collision
  =   
can be avoided in a well manner in CSMA/CA, thus VDSA 
multiple access system can be readily consider as a virtual =  +  +  −  +  +   − 4   2 
queue following a FCFS or based on different criteria priority
(2)
queue is also handled.
and PQ is the probability that all servers are busy,
In this paper, in the queueing model we consider different
priority classes of users using the vacant bands. There is no
service interruption of lower priority when higher priority 
  
customer enters the system and lower one is in service. Higher  = ∑
 +  (3)
! ! !
priority customer should wait in particular queue assigned for
higher priority until the service completion of lower priority
user. After completion the service higher one enters in service and
even though some lower priority customers are in their
assigned queue from earlier the higher one.
 = / ,  = ∑  ,  = ∑ 
The services of secondary users are only stopped when
primary users demand that channels in which the services are Wagner applied matrix-analytic methods to calculate the
going on. But we avoid this topic here. We also not consider Laplace-Stieltjes Transform of the actual waiting time.
any kind of interference, extensively long transmission delay, Leemans, Venkataramani et al. [22] considered the stationary
and channel jamming. distribution of queue lengths and waiting times. They also
used three dimensional state spaces and applied a matrix-
That means we consider a multi server multi priority non- geometric method to analyze the queue [23, 24]. Federgruen et
preemptive model whose performance measures will be done al [25] characterized the performance space of M/G/m non-
based on the average number of cars as well as the available preemptive queueing system.
vacant UHF TV channels within the transmission range.
For M/M/m system, the average waiting time for different
III. SOLUTIONS FOR THE QUEUEING MODEL priority classes customer [26],
The basic model of multi server multi priority queue is 
depicted in Fig. 4. It consists of two servers and there are only  = (4)
  
two classes of customers. The heterogeneous priority structure
is constructed by giving class A non-preemptive priority over ∑
  
class B on Si, while class B receives priority over class A on where  = .

S2. The service discipline within each class is FCFS. Both
arrival and service processes are stochastic processes with Then the response time,
class dependent parameters. We assume that the arrival
process is Poisson, with arrival rates λa and λb for class A and 
 = +  (5)
Class B respectively. Consequently, the interarrival times are 
exponentially distributed. It is also assumed that the service
times are exponentially distributed, with parameters 1/a and For M/G/1 multi priority system, the waiting time and the
1/b. response time are as follows:
There are so many approaches to calculate the queue
length and waiting time for each class. Gail, Kao, Wagner et al ∑ 

  
have studied multiserver non-preemptive model with two  = (6)
 ⋯  ⋯ 
priority classes [17, 18, 19]. Kao implemented a power–series
method for the two priority queues. Kella et al [20] calculated 
the waiting time for non-preemptive priority M/M/m queue as  = +  (7)

follows:

45
4

For multiple servers, the calculation of the mean residual time of 100 ms. In this analysis, both M/M/m and M/G/m


time  = ∑   
 is not so easy. But if the service times of
models are used.

all priority classes are identically and exponentially We considering about 20% customers have higher priority.
distributed, then we calculate R in a convenient way. If all For the arrival rate of 20 message/second for priority 1
priority classes have exponentially distributed service times customers experience comparatively less blocking probability.
with common mean 1/µ, then (6) have a closed form solution With arrival rate of 60 message/second priority 2 customers

with  = , where PQ is the probability of all channels are experience higher probability to get the channels are busy. In
the area of Boston both classes have higher probability.
busy.
Fig.7 shows the maximum mean response time of 12 ms
Again, if we set higher priority to customers of short for priority 2 customers at Boston in both M/M/m and M/G/m
service times, the average delay per customer tends to be FCFS models where both priority classes have same data rate
reduced. Then for a non-preemptive system with two customer of 800 kbps. The maximum response time of 11 ms is again
classes A and B, with respective arrival and service rates λA, obtained for the priority 2 customers where lower data rate of
A and λB, µ B, and if A > µ B, then the average delay per 400 kbps are assigned to priority 2 customers and higher data
customer, rate of 800 kbps are assigned to priority 1 customers. For
priority 1 customers we also have higher mean response time
    at Boston of about 10 ms but still lower than that of priority 2
= (8)
  customers. The second largest values for mean response time
are obtained at Auburn.
is smaller when A is given priority over B than when B is
given priority over A. It is noticeable that for the same priority class of customer
we have equal mean response time for both M/M/m and
IV. ANALYSIS OF VDSA IN VACANT UHF TV M/G/m model. It has been happened due to different waiting
CHANNELS times in both model are less comparable with transmission
time in non-preemptive queueing model as there is no service
In this work, we use the data collected from Interstate I-90 interruption due to priority. We also see that when we use two
in the state of Massachusetts shown in Fig.3 of [15]. The different service rates for two classes and given highest
information about the available bandwidth in vacant UHF TV priority to higher service rate reduces the average amount of
channels are extracted from Fig.5 of [6]. We reproduce the mean response time. In addition, if we compare these results
average number of cars per kilometer along I-90 using Fig. with [15] where preemptive queueing model is used, we see
3(b) of [15] and again interpolate the samples which shown in that in our model we have too much reasonable mean response
Fig. 5. time which meets the requirement of DSRC very well for all
From Fig. 1 we can understand that most white channels cases and for all kinds of customers.
are adjacent to TV channels being used for broadcast. So, the
output power of portable devices in vacant channels adjacent V. CONCLUSION
to TV broadcasting channels should keep under 40 mW. In this paper, we analyze the feasibility of VDSA in vacant
UHF TV white space. We use multi-server multi-priority non-
It is assumed that the transmitting vehicle and the preemptive queueing model to evaluate the system
receiving vehicle of a communication pair are separated by the performance. The models are applied to the vehicles traveling
transmission range, 300 meters. So, any car within the 500 along I-90 in the state of Massachusetts. The results show that
meters sensing the communication pair cannot use the same in suburban area, vacant TV band is a feasible resource for
channel. So, in a snapshot the number of cars increases to 1.3 vehicle communication which meets the requirements of
times of the average number of cars per kilometer. As a DSRC. Because we have much lower mean response time for
consequence the number of customers that is number of each class of customers in the suburban area, in a consequence
communication set considered is about 65% of the total we can say that in area of much lower traffic load such as rural
number of cars per kilometer on I-90 in a snapshot. area, this resource also be useful.
Considering data rate for each higher priority customer is
800 kbps and for each lower priority customer is 400 kbps. ACKNOWLEDGEMENT
Average packet length is assumed of 500 bytes. We consider This research was supported by the MKE (The Ministry of
moderate traffic load that means sub-urban area. So, Knowledge Economy), Korea, under the ITRC (Information
Okumura-Hata path loss model is used. Here, we use two Technology Research Center) support program supervised by
types of arrival rates of 20 messages/second and 60 the NIPA (National IT Industry Promotion Agency) (NIPA-
messages/second for two priority classes with meaninterarrival 2011-C1090-1121-0001).

Fig. 5. Average number of cars per kilometer along I-90.

46
5

Fig. 6. The Probability of all channel being busy observed by different classes customers.

Fig.7. The predicted response times for customers of two priority classes, for both
M/M/m, M/G/m (FCFS, with two different service rates for two priority classes) model.
[12] H. Li, “Socially optimal queuing control in cognitive radio systems:
REFERENCES Pricing and learning,” in IEEE Wireless Communications and Networking
Conference (WCNC), 2010.
[1] D. Jiang, V. Taliwal, A. Meier, W. Holfelder, and R. Herrtwich, “Design
[13] H. Shiang and M. van der Schaar, “Queuing-based dynamic channel
of 5.9 GHz DSRC-based vehicular safety communication,” IEEEWireless
selection for heterogeneous multimedia applications over cognitive radio
Communications, vol. 13, no. 5, pp. 36–43, 2006.
networks,” IEEE Transactions on Multimedia, vol. 10, no. 5, pp. 896–909,
[2] “Standard Specification for Telecommunications and Information
2008.
Exchange Between Roadside and Vehicle Systems–5GHz Band Dedicated
[14] C. Zhang, X. Wang, and J. Li, “Cooperative Cognitive Radio with
Short Range Communications (DSRC) Medium Access Control (MAC) and
Priority Queueing Analysis,” IEEE International Conference on
Physical Layer (PHY) Specifications,” ASTM Std. E2213-03, 2003.
Communications (ICC), 2009.
Available: http://www.astm.org/.
[15] S. Chen, A. M. Wyglinski, R. Vuyyuru and O. Altintas, “Feasibility
[3] R. Sevlian, C. Chun, I. Tan, A. Bahai, and K. Laberteaux, “Channel
Analysis of Vehicular Dynamic Spectrum Access via Queueing Theory
Characterization for 700MHz DSRC Vehicular Communication”, Journal of
Model,” IEEE Vehicular Networking Conference, p – 223 – 230, 2010.
Electrical and Computer Engineering Received 19 March 2010; Accepted 5
[16] S. Keshavamurthy and K. Chandra, “Multiplexing Analysis for Dynamic
July 2010, vol. 2010, Article ID 840895, 9 pages.
Spectrum Access,” in Military Communications Conference (MILCOM),
[4] I. Mitola, J., “Cognitive radio for flexible mobile multimedia
2006, pp. 1–7.
communications,” in Mobile Multimedia Communications, 1999. (MoMuC [17] H.R. Gail, S.L. Hantler, B.A. Taylor, “Analysis of a nonpreemptive
’99) 1999 IEEE International Workshop on, 1999, pp. 3–10.
priority multiserver queue,” Advances in Applied Probability 20 (1988) 852-
[5] S. Pagadarai and A. Wyglinski, “A quantitative assessment of wireless
879.
spectrum measurements for dynamic spectrum access,” in 4th International
[18] Edward P.C. Kao, Sandra D. Wilson, “Analysis of nonpreemptive
Conference on Cognitive Radio Oriented Wireless Networks and
priority queues with multiple servers and and two priority classes,” European
Communications (CROWNCOM), Jun. 2009.
Journal of Operational Research ,Volume 102, Issue 1, 1 October 1997,
[6] S. Pagadarai, A. Wyglinski, and R. Vuyyuru, “Characterization of vacant
Pages 227-241.
UHF TV channels for vehicular dynamic spectrum access,” in IEEE Vehicular
[19] D. Wagner, “Waiting times of a finite-capacity multi-server model with
Networking Conference (VNC), 2009, pp. 1–8.
non-preemptive priorities,” European Journal of Operational Research 118
[7] Unlicensed Operation in the TV Broadcast Bsnds, Final Rules. Available:
(1999) 181-193.
http://edocket.access.gpo.gov/2009/pdf/E9-3279.pdf
[20] O. Kella, and U. Yechiali, “Waiting times in the non-preemptive priority
[8] J. Wang, M. Song, S. Santhiveeran, K. Lim, G. Ko, K. Kim, S. Hwang, M.
M/M/c queue”, Communications in Statistics - Stochastic Models, 1, 257-262,
Ghosh, V. Gaddam, and K. Challapali, “First Cognitive Radio Networking
1985.
Standard for Personal/Portable Devices in TV White Spaces,” in IEEE
[21] L. Kleinrock, Queueing Systems: Volume 2: Computer Applications. John
Symposia on New Frontiers in Dynamic Spectrum Access Networks(DySPAN
Wiley & Sons New York, 1976.
2010), Singapore, 2010.
[22] H. Leemans, G. Dedene , “Queue Lengths and Waiting Times in the
[9] Federal Communications Commission, “In the Matter of Unlicensed
Two-Class Two-Server Queue with Nonpreemptive Heterogeneous Priority
Operation in the TV Broadcast Bands, Additional Spectrum for Unlicensed
Structures”, Belgium. Available:
Devices Below 900 MHz and in the 3 GHz Band.”
https://lirias.kuleuven.be/handle/123456789/118550.
[10] G. Li and H. Liu, “Dynamic resource allocation with finite buffer
[23] H. Leemans, K.U.Leuven, “A matrix-geometric solution for the
constraint in broadband OFDMA networks,” IEEE Wireless Comm. And
multiserver nonpreemptive priority queueing model with mixed priorities”,
Networking, vol. 2, pp. 1037–1042, 2003.
Belgium.
[11] G. Song, Y. Li, and M. Inc, “Utility-based resource allocation and
Available: https://lirias.kuleuven.be/handle/123456789/7074//items-by-
scheduling in OFDM-based wireless broadband networks,” IEEE
author?author=Leemans%2C+Herlinde%3B+U0003217
Communications Magazine, vol. 43, no. 12, pp. 127–134, 2005

47
6

[24] B. Venkataramani, S. K. Bose and K. R. Srivathsan, “Queuing analysis Performance under Nonpreemptive Priority Rules”, MANAGEMENT
of a non-pre-emptive MMPP/D/1 priority system”, Computer SCIENCE, Vol. 34, No. 9, September 1988, pp. 1121-1138.
Communications, Volume 20, Issue 11, 15 October 1997, Pages 999-1018. [26] D. Bertsekas, R. Gallager, Data Networks, 2nd Edition, Prentice-Hall,
[25] A. Federgruen and H. Groenevelt, “M/G/c Queueing Systems with New Jersey, 1992.
Multiple Customer Classes: Characterization and Control of Achievable

48

You might also like