Professional Documents
Culture Documents
AbstractHeterogeneous wireless networks where several sys- as energy consumption [3]. That is, a system operator has to
tems with different bands coexist for multimedia service are consider the allocation of transmission data to subsystems and
currently in service and will be widely adopted to support spectrum band for its operational cost savings. Also, optimal
various traffic demand. Under heterogeneous networks, a mobile
station can transmit over multiple and simultaneous radio access bandwidth (BW) and power allocation of the MRA system
technologies (RATs) such as WLAN, HSPA, and WCDMA LTE. should be determined at the same time. In this work, an
Also, cognitive radio for the efficient use of underutilized/unused optimal solution for how these resources should be allocated is
frequency band is successfully implemented in some networks. considered. We intend to provide the solution for the systems
In this letter, we address such operational issues as air interface of CR over SDR, where lots of work have described that BW
and band selection for a mobile and power allocation to the
chosen links. An optimal solution is sought and analyzed and of each RAT can be dynamically assigned rather than fixed.
a distributed joint allocation algorithm is proposed to maximize Some promising MRA concepts for increasing system ef-
total system capacity. We investigate the benefit of multiple trans- ficiency may be found in literature [2] [4] [5] related with
missions by multiple RATs over a single transmission by a single Ambient Networks (AN) [6]. Feasibility study on MRA [2],
RAT at a time, which can be interpreted as network diversity. an implementable MRA [4], and system architecture and
Numerical results validate the performance enhancement of our
proposed algorithm. necessary functions for the realization [5] are introduced,
each as a subproject under the umbrella of AN project in
Index TermsAccess network selection, joint allocation, multi- Europe. Especially in [5], they provide universal link layer
radio access, optimization, radio resource management.
processing over different RATs for the purpose of multi-radio
cooperation and improving connectivity at the radio access
I. I NTRODUCTION level. Also, in [2] and [5], MRA can be accomplished in
two ways: a switched MRA scheme among available RATs,
N ETWORKS with multiple radio access technologies
(RATs) would become one of the most prevalent features
in next generation mobile networks. These networks where
where only one RAT is used for the transmission of data at a
time, and a parallel MRA scheme, where multiple RATs are
utilized for data transmission (with or without data duplication
a user equipment can transmit its data over multiple RATs
over different RATs). In [7], authors address joint multi-
simultaneously are named multi-radio access (MRA) system
radio resource management amongst RATs. Investigations
[1], which accommodates RATs such as 3GPP Long Term
of MRAs are also carried out in [8], where packet-based
Evolution (LTE), Worldwide Interoperability for Microwave
radio channel diversity is adopted with emphasis on physical
Access (WiMAX), and Wireless LAN (WLAN) as subsystems
layer. In addition, [9] shows another resource allocation with
[2]. For such a mobile one can consider a user equipment
QoS supporting where multi-radio networks have also been
implementing cognitive radio (CR) over software defined
investigated from theoretical standpoints. However, studies are
radio (SDR). For such an MRA system, many issues are yet
rare on the optimal solution of joint resource (i.e., RAT, BW,
to be resolved on implementation and network optimization
and power) allocation for a parallel MRA scheme from a
including radio resource management problems. And, opti-
viewpoint of a scheduler, partly due to the fact that these are
mal operation issues have taken lots of attention recently to
multidimensional optimization problems with complexity. We
increase system efficiency and improve connectivity as well
investigate the optimal resource allocation issues to support
Manuscript received January 12, 2010; revised April 30, 2010 and July 27, parallel MRA for much higher system capacity in this context.
2010; accepted July 27, 2010. The associate editor coordinating the review The rest of this letter is organized as follows. Section II
of this letter and approving it for publication was E. Hossain. begins with introducing the MRA transmission and a system
This research was supported by the MKE (The Ministry of Knowledge
Economy), Korea, under the ITRC (Information Technology Research Center) model. In Section III, an optimal solution for parallel MRA
support program supervised by the NIPA (National IT Industry Promotion is provided, and Section IV proposes a joint resource alloca-
Agency) [NIPA-2010-(C1090-1011-0013)]. This paper was presented in part tion algorithm for maximum capacity. Numerical results are
at the IEEE Vehicular Technology Conference (VTC)09, Fall, Anchorage,
Alaska, USA, September 2009. presented in Section V, followed by conclusions in Section VI.
Y. Choi is with the Department of Electrical Engineering, Stanford Univer-
sity, Stanford, CA 94305 (e-mail: yonghoon@stanford.edu).
H. Kim is with the Department of Electronics Engineering, University of II. S YSTEM M ODEL
Incheon, Incheon, Korea (e-mail: hoon@incheon.ac.kr).
S. Han is with the Department of Electrical and Computer Engineering, As shown in Fig. 1, we consider an MRA system model
UBC, Vancouver, BC, Canada V6T 1Z4 (e-mail: swhan@ece.ubc.ca). based on AN [6], where subsystems (i.e., radio interfaces)
Y. Han (corresponding author) is with the Department of Electrical Engi- are available for each user terminal. It is assumed that a
neering, Korea Advanced Institute of Science and Technology, Daejeon, Korea
(e-mail: ynhan@kaist.ac.kr). multi-mode, multi-band user terminal (MMT) implemented by
Digital Object Identifier 10.1109/TWC.2010.11.100045 the reconfigurable SDR technology [10] has ability to access
1536-1276/10$25.00
c 2010 IEEE
CHOI et al.: JOINT RESOURCE ALLOCATION FOR PARALLEL MULTI-RADIO ACCESS IN HETEROGENEOUS WIRELESS NETWORKS 3325
as ( )
= log 1 + (2)
=1
B. Optimality Conditions for the Parallel MRA Property 1: (Convergence) The proposed technique based
For the optimal solution of problem ( ), the Lagrangian on Newtons method converges to the optimum.
can be obtained as below. Proof: For the optimal solution, the algorithm starts the
( ) process off with some arbitrary positive initial values 0 and
( , , , ) = log 1 + 0 . This method will always converge because the objective
=1 =1
function (3) is concave (refer to Appendix) and ( ) in (15)
is decreasing in since (3) is concave and ( ) < 0.
( )
Also, since (3) is concave and increasing in ( , ) for
+ + (7) a fixed , it can be verified that the objective function is
=1 =1 =1 =1 increasing in for fixed and . Therefore, one can get
the converged optimal value of , regardless of initial values
where shadow prices and are nonnegative Lagrange 0 and 0 , where ( ) satisfies (4) - (6).
multipliers. By taking derivatives with respect to and Property 2: (Speed of convergence) The process for itera-
respectively, we can have the Karush-Kuhn-Tucker (KKT) tive calculating the optimum converges quadratically.
conditions as, Proof: Let us use the abbreviation for simplic-
( )
ity. Assume that is a root of ( ) and that as
= log 1 + 0, (8)
+ , where is given by (17). Let the error at the
iteration to be = , so = + . We know ( )
is continuous near and use a Taylor approximation about
= 0, (9)
+ [17].
( ( ) ) 0 = () = ( + )
log 1 + = 0, (10) = ( ) + ( ) + 2 ( )/2! + (3 ). (18)
+
( ) If ( ) = 0, one writes ( ) 2 ( ) 3
( ) = + 2 ( ) + ( ).
Then,
= 0, (11) ( )
+ ( )
+1 = +1 =
( ) ( )
( )
= 0, (12) ( ) 2 3
= + + ( )
=1 2 ( )
( ) 2
= + (3 ). (19)
2 ( )
= 0. (13)
Therefore, we say that the process converges quadratically in
=1
the sense that the error at a given stage is the square of the
Using (9) and (11), the relation between BW and power error at the preceding stage.
allocation can be obtained After is obtained, can be determined by using (14).
[ ]+ In addition, from the calculated values (i.e., , ) and KKT
1
= (14) conditions, the usual waterfilling level ( ) of each MMT
also can be represented as below.
where []+ = {, 0}. From (14), to get the optimal
and value, we need to have one of them. For the optimal + 1 = , if > 0
value firstly, the well-known Newtons method [17] can be
(20)
1 , if = 0.
applied because it converges faster towards a local maximum
than other algorithms based upon gradient descent or non- To update the
and
values for the optimal solution,
gradient methods. Now, let us take function ( ) as (15) we consider the continuously differentiable dual function as
and its derivative ( ) as in (16).
( ) ( , ) = max ( , , , ). (21)
,
( ) = log 1 + , (15) Using a gradient-based search [17], the updated +1 value
+
for power allocation is given by
( ) [ ]+ +
1 (
,
)
( ) = (16) +1 =
= + (22)
+ +
=1
where superscript represents the iteration. Then, by where > 0 is a constant step size. Likewise, the updated
simple algebra, a better approximation +1 value can be +1 value for the BW allocation can also be determined as
obtained as [ ]+ [ ( )]+
( ) (
,
)
+1 =
. (17) +1
= = + (23)
( ) =1
CHOI et al.: JOINT RESOURCE ALLOCATION FOR PARALLEL MULTI-RADIO ACCESS IN HETEROGENEOUS WIRELESS NETWORKS 3327
Algorithm 1 at MMT 3
1: if = 0, then
(0) (0) (0) (0) 2.5
2: Initialize , , , and .
3: else 2
4: Calculate +1
using Newtons method. 1.5 parallel MRA
( )
+1
= (
switched MRA
) 1
Determine +1 +1
5 10 15 20 25 30 35 40 45 50
5: using the obtained value. Number of MMTs (N
)
[ ]+
+1 +1 1
=
Fig. 2. Total system capacity of the parallel MRA which implemented the
6: if the equilibrium value of and is obtained, then proposed joint allocation algorithm.
7: Transmit data unit to the RAT(s) using +1 and
+1
.
8: else efficiency (i.e., = 1 for = 1, 2), and 100 separation
9: Update +1 using +1 information. between subsystems access points. MMTs are assumed to
[ ( )]+ have maximum power of 20mW and randomly distributed in
+1
+1
= + multi-RAT region, where any MMT can access to multiple
=1 RATs, with a speed of 3km/h. We consider path loss and
10: Feedback the +1
information to each RAT. shadowing as a channel propagation model. For path loss, we
11: end if adopt the modified Hata urban propagation model [18]:
12: end if {
122 + 38 log(d), if d 0.05 km
(24)
122 + 38 log(0.05), if d < 0.05 km
Algorithm 2 at access point of RAT where represents the distance between a subsystem and
1: Compute +1
using +1
information. an MMT in kilometers. The shadowing component follows a
[ ( )]+ lognormal distribution with a mean value of 0dB and standard
+1
+1
= + deviation of 8dB. Thermal noise power is assumed to be -
=1 174dBm/Hz.
2: Broadcast the new +1
value to all MMTs.
3: + 1
B. Simulation Results
Fig. 2 compares the system capacity of the parallel MRA
Based on the optimality conditions for the parallel MRA
scheme and a switched MRA method with respect to the num-
in Section III-B, we propose a distributed joint allocation
bers of MMTs. It can be seen that the parallel MRA achieves
method over multiple subsystems as shown in Algorithms 1
more capacity increase than a switched MRA method. The
and 2. Note that the value is determined not by each
reason is that the parallel MRA makes use of multi-RAT
subsystem but by each MMT. That is, through the proposed
diversity if MMTs can use parallel MRA method for data
algorithm, the optimal allocation value of and for
transmission, whereas the switched MRA use only one RAT
the parallel MRA can be determined in a distributed manner.
at any given time slot. In general, the co-existence of different
Although (3) could be solved via centralized optimization
RATs causes the enhanced entire system throughput and it
techniques where convergence is guaranteed without signaling
can be considered as RAT-diversity gain. Also, in Fig. 3,
overhead, it is preferred to use a distributed optimization
it is shown that the occurrence rate of the parallel MRA
method in order to make a decision at each MMT under
increases and approaches to about 55% as the number of
perfect synchronized updates. Hence, this algorithm shows the
MMTs increases. Hence, from Figs. 2 and 3, it can be asserted
possibility and feasibility of a distributed decision making for
that by applying the parallel MRA scheme which involves the
the parallel MRA when MMTs sharing multiple heterogeneous
proposed joint allocation algorithm, one can achieve increased
subsystems act in a selfish manner.
system capacity.
An example for finding an optimal solution for the parallel
V. P ERFORMANCE E VALUATION
MRA and the corresponding shadow prices is provided in
A. Simulation Environment Figs. 4 and 5. Particularly, in an MRA system with = 2,
In order to evaluate the system capacity, we consider a we apply the proposed algorithm of Section IV for evaluation.
simplified MRA system consisting of two OFDMA-based In Fig. 4, it can be seen that MMT 2 selects parallel MRA
subsystems. It is assumed that each subsystem has total system to both RATs 1 and 2, and others choose only one subsystem
bandwidth of 5 and 20MHz, respectively, with the same after iterative calculations for the maximum capacity. Fig. 5
3328 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 9, NO. 11, NOVEMBER 2010
5
x 10
55 2.85 4.5
2.5
50
4
Percentage of the parallel MRA ][%
40
1.5
3
35
1
2.5
30
0.5
25 2
Percentage of the parallel MRA
RAT1
Number of the parallel MRA
RAT2
20 0 1.5
5 10 15 20 25 30 35 40 45 50 0 50 100 150 200 250 300
Number of MMTs (N
) Iteration (k)
Fig. 3. Occurrence rate of the parallel MRA when the proposed joint Fig. 5. An illustration of the corresponding shadow price ( ) when the
allocation algorithm is applied at = 2. proposed algorithm is applied.
12
allocated to MMT 1 by RAT
1 A PPENDIX
allocated to MMT 1 by RAT 2
10 allocated to MMT 2 by RAT
1 A. Proof of the Concavity
allocated to MMT 2 by RAT
2
allocated to MMT 3 by RAT
1 To show the concavity of the objective function in (3), let
allocated to MMT 3 by RAT
2
8
us observe the following general form.
Allocated BW [MHz
]
( )
6 (, ) = ln 1 + . (25)
MMT 3 selects RAT
2
4
MMT 1 selects RAT
1
Finding whether a function is concave or not can be started
MMT 2 selects the parallel MRA to both RATs from the first partial derivatives of (, ).
2 ( )
ln 1 + +
= = . (26)
0
0 50 100 150 200 250 300
Iteration (k) +
According to [16], it can be said that the function (, ) [6] WWI Ambient Networks. [Online]. Available: http://www.ambient-
is concave function if the determinants have the following networks.org/
[7] P. Magnusson, F. Berggren, I. Karla, R. Litjens, et al., Multi-radio
pattern with not all zeros. resource management for communication networks beyond 3G, in Proc.
IEEE Veh. Technol. Conf., Sep. 2005, pp. 16531657.
1 0; 2 0. [8] L. Badia, C. Taddia, G. Mazzini, and M. Zorzi, Multiradio resource
allocation strategies for heterogenous wireless networks, in Proc.
From the above calculations, the function (, ) has this Wireless Personal Multimedia Commun. Conf. (WPMC), Sep. 2005.
pattern, so (, ) is a concave function in {, }. [9] S. Merlin, N. Vaidya, and M. Zorzi, Resource allocation in multi-radio
multi-channel multi-hop wireless networks, in Proc. IEEE INFOCOM,
Our objective function is a positive linear combination of 2008, pp. 610618.
the (, ) type functions, where 0 and 0. Since a [10] M. Dillinger, K. Madani, and N. Alonistioti, Software Defined Radio:
positive linear combination of concave functions is concave, it Architectures, Systems, and Functions. John Wiley & Sons Ltd., 2003.
[11] H. Holma and A. Toskala, HSDPA/HSUPA for UMTS. John Wiley &
can be concluded that the objective function of our formulation Sons Ltd., 2006.
is also a concave function. [12] J. Acharya and R. D. Yates, Dynamic spectrum allocation for uplink
users with heterogeneous utilities, IEEE Trans. Wireless Commun., vol.
8, pp. 14051413, Mar. 2009.
R EFERENCES [13] 3G Americas white paper (2008, Sep.). EDGE, HSPA, LTE - Broadband
Innovation. [Online]. Available: http://www.3gamericas.org.
[1] A. Furuskr, Allocation of multiple services in multi-access wireless [14] W. Yu and J. M. Cioffi, FDMA capacity of Gaussian multiple-access
systems, in Proc. International Workshop Mobile Wireless Commun. channels with ISI, IEEE Trans. Commun., vol. 50, no. 1, pp. 102111,
Network, Sep. 2002, pp. 261265. Jan. 2002.
[2] G. P. Koudouridis, R. Agero, E. Alexandri, M. Berg, et al., Feasibility [15] R. Cohen and L. Katzir, Computational analysis and efficient al-
studies and architecture for multi-radio access in ambient networks, gorithms for micro and macro OFDMA scheduling, in Proc. IEEE
in Proc. 15th Wireless World Research Forum (WWRF) Meeting, Dec. INFOCOM, Apr. 2008, pp. 511519.
2005. [16] S. Boyd and L. Vandenberghe, Convex Optimization. Cambrige Univer-
[3] E. Hossain, D. Niyato, and Z. Han, Dynamic Spectrum Access and sity Press, 2004.
Management in Cognitive Radio Networks. Cambrige University Press, [17] J. Stoer, R. Bulirsch, R. Bartels, W. Gautshi, and C. Witzgall, in
2009. Introduction to Numerical Analysis, 3rd edition. Springer, 2002, pp.
[4] E. Gustafsson and A. Jonsson, Always best connected, IEEE Wireless 289363.
Commun. Lett., vol. 10, no. 1, pp. 4955, 2003. [18] T. K. Sarkar, Z. Ji, K. Kim, A. Medouri, and M. Salazar-Palma, A
[5] K. Dimou, R. Agero, M. Bortnik, et al., Generic link layer: a solu- survey of various propagation models for mobile communication, IEEE
tion for multi-radio transmission diversity in communication networks Antennas Propagat. Mag., vol. 45, pp. 5182, June 2003.
beyond 3G, in Proc. IEEE Veh. Technol. Conf., Sep. 2005, pp. 1672
1676.