International Journal of · Email: [email protected], [email protected]...

I. J. Communications, Network and System Sciences, 2008, 3, 207-283 Published Online August 2008 in SciRes (http://www.SciRP.org/journal/ijcns/).

Copyright © 2008 SciRes. I. J. Communications, Network and System Sciences, 2008, 3, 207-283

TABLE OF CONTENTS

Volume 1 August 2008 OCDMA/OCDMA Overloading Scheme for Cellular DS-CDMA Using

Orthogonal Gold Codes and Complex Scrambling

P. KUMAR, S. CHAKRABARTI………………………………………….………………….……………….. 207

Distributed Relay Diversity Systems for OFDM-Based Networks

A. OSSEIRAN, A. LOGOTHETIS, S. B. SLIMANE……………………….……….………..……………..… 215

Joint Adaptive Modulation and Power Allocation in Cognitive Radio Networks

D. LI, X. H. DAI, H. ZHANG……………………………….………....……………………………………... 228

Hardware/Compiler Memory Protection in Sensor Nodes

L. LOPRIORE……………………………………………..…………..……………………..……………..... 235

On Energy-Efficient Node Deployment in Wireless Sensor Networks

H. WANG, K. Z. LU, X. H. LIN…………………………...………………………………………..…….....… 241

Relations among Mobility Metrics in Wireless Networks

X. SHU, X. N. LI……………………………..………………………..……………………………………...... 246

Stereo Video Transmission Using LDPC Code

R. GUO, L. X. WANG, X. X. JIANG……………..………………………………………………..……...….. 254

An Agent-Based Multimedia Intelligent Platform for Collaborative Design

Q. LIU, X. R. CUI, X. Y. HU…………………………………………………………………..………….…. 260

A Multi-User Cooperative Diversity for Wireless Local Area Networks

J. CHEN, K. DJOUANI……………………..………………………..……………………………………...... 266

Priority-Based Resource Allocation for Downlink OFDMA Systems Supporting RT and NRT Traffics

H. WANG, L. DITTMANN………………………..………………………………………………..……...….. 274

International Journal of Communications, Network and System Sciences

(IJCNS)

Journal Information

SUBSCRIPTIONS

The International Journal of Communications, Network and System Sciences (Online at Scientific Research

Publishing, www.SciRP.org) is published quarterly by Scientific Research Publishing, Inc. 5005 Paseo Segovia,

Irvine, CA 92603-3334, USA.

E-mail: [email protected]

Subscription rates: Volume 1 2008 Print: $50 per copy.

Electronic: free, available on www.SciRP.org.

To subscribe, please contact Journals Subscriptions Department, E-mail: [email protected]

Sample copies: If you are interested in subscribing, you may obtain a free sample copy by contacting Scientific

Research Publishing, Inc at the above address.

SERVICES

Advertisements

Advertisement Sales Department, E-mail: [email protected]

Reprints (minimum quantity 100 copies)

Reprints Co-ordinator, Scientific Research Publishing, Inc. 5005 Paseo Segovia, Irvine, CA 92603-3334, USA.


COPYRIGHT

Copyright© 2008 Scientific Research Publishing, Inc.

All Rights Reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in

any form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise, except as

described below, without the permission in writing of the Publisher.

Copying of articles is not permitted except for personal and internal use, to the extent permitted by national

copyright law, or under the terms of a license issued by the national Reproduction Rights Organization.

Requests for permission for other kinds of copying, such as copying for general distribution, for advertising or

promotional purposes, for creating new collective works or for resale, and other enquiries should be addressed to

the Publisher.

Statements and opinions expressed in the articles and communications are those of the individual contributors and

not the statements and opinion of Scientific Research Publishing, Inc. We assumes no responsibility or liability for

any damage or injury to persons or property arising out of the use of any materials, instructions, methods or ideas

contained herein. We expressly disclaim any implied warranties of merchantability or fitness for a particular

purpose. If expert assistance is required, the services of a competent professional person should be sought.

PRODUCTION INFORMATION

For manuscripts that have been accepted for publication, please contact:




OCDMA/OCDMA Overloading Scheme for Cellular DS-CDMA Using Orthogonal Gold Codes and

Complex Scrambling

Preetam KUMAR, Saswat CHAKRABARTI G S Sanyal School of Telecommunications, IIT Kharagpur, India Email: [email protected], [email protected] Received on May 5, 2008; revised and accepted on June 27, 2008

Abstract Overloading is a method to extend capacity limitation of multiple access techniques. The system becomes overloaded, when the number of users exceeds the signal dimensions. One of the efficient schemes to overload a CDMA system is to use two sets of orthogonal signal waveforms (O/O). In this paper, the BER performance of a new overloading scheme using scrambled orthogonal Gold code (OG/OG) sets is evaluated with soft decision interference cancellation (SDIC) receiver. When complex scrambling is not used, it is shown that OG/OG scheme provides 25% (16 extra users) channel overloading for synchronous DS-CDMA system in an AWGN channel, with an SNR degradation of about 0.35 dB as compared to single user bound at a BER of 1e-5. We have evaluated the overloading performance, when two set are scrambled with set specific deterministic or random complex scrambling sequence. It is shown that the amount of overloading increases significantly from 25% to 63% (40 extra users) by using random complex scrambling for N=64. For deterministic (periodic) scrambling, the overloading percentage increases considerably to 78. On a Rayleigh fading channel, an overloading of 40% is obtained without scrambling at a BER of 5e-4 with near single user performance. With complex scrambling overloading % increases considerably to 100%. Keywords: DS-CDMA, Orthogonal Codes, Overloading, Interfernce Cancellation

1. Introduction

Efficient use of the available radio spectrum is an important requirement for future wireless communication. The number of users supported in a DS-CDMA cellular system is typically less than spreading factor (N), and the system is said to be underloaded. As the demand for cellular CDMA increases, the number of users naturally exceeds the available dimension due to bandwidth limitation. Overloading is a technique to accommodate more number of users than the spreading factor N. This is an efficient scheme to increase users in a fixed bandwidth, which is of practical interest to mobile system operators. But the increase in capacity is obtained with a cost in BER and receiver complexity. In fact this type of channel overloading is provisioned in the 3G standard [1].

Among the approaches described in the literature, the most efficient ones use multiple sets of orthogonal codes [2]. The concept of overloading in a DS-CDMA system

using two sets of orthogonal codes is explained with the help of Figure1. For the first N users, the system allocates orthogonal codes drawn from the first set of N codes. When the number of intending users exceeds ‘N’, the excess users are accommodated in the system by providing suitable codes drawn from a second set of M codes. In this way, we are able to accommodate greater number of users (K) than the spreading length N (K>N), and the cell becomes overloaded.

The number of active users (K) in a conventional synchronous orthogonal CDMA environment is limited by the spreading factor N, which is WT where W is the transmission bandwidth and T is the duration of a symbol. When K exceeds N, the system becomes overloaded and the signatures are no longer orthogonal. This leads to multiple access interference (MAI). In an overloaded system, a conventional matched filter receiver is not optimal, due to the high level of MAI. Multiuser detection (MUD) is required in order to obtain a

208 P. KUMAR ET AL.


satisfactory performance of the users. Linear MUDs, such as the decorrelator, the minimum mean squared error detector or linear decision directed interference cancellation are devised to detect users in an underloaded system. The Maximum Likelihood (ML) detection is not an option because of its complexity that is exponential in the number of users. The nonlinear MUDs such as multistage parallel interference cancellation (PIC) and successive interference cancellation (SIC) [3], have good complexity- performance trade-off as compared to other MUDs. Hence these MUDs are suitable for overloaded systems.

It is interesting to note that several studies have been made in the recent past to understand, analyze and evaluate the detrimental effects of overloading. Almost all studies consider the uplink or reverse link and several studies suggest usage of appropriate multiuser detection (MUD) schemes at the base station receiver. For example, a method of accommodating K = N + M users in an N-dimensional signal space that does not compromise the minimum Euclidean distance of the orthogonal signaling has been presented in [4] for AWGN channel. A tree-like correlation coefficient structure of user signatures suitable for optimal multiuser detection has been proposed in [5]. In another approach, two sets of orthogonal codes which are orthogonal within the sets is introduced in [6]. In [6], the orthogonal sets are generated using Walsh-Hadamard (WH) codes, where the same WH code set is scrambled with set specific scrambling sequence (s-O/O). An iterative multistage detection technique has been proposed to cancel the interference between the two sets of user. In [7], it is shown that for uncoded BPSK modulated CDMA signal with N=64, an overloading of 11% can be achieved in an AWGN channel for s-O/O scheme. Another kind of receiver simplification is presented in [8], where signals are divided into groups that are orthogonal to each other. A new overloading scheme using hybrid techniques has been proposed in [9], where the spreading codes and transmission modes are different for the two sets to increase the overloading performance. The attractive property of the overloading scheme was the incentive to integrate a particular type of O/O, called quasi-synchronous sequences (QOS) [10], into cdma2000 standard [11].

To the best of our knowledge, the usage of orthogonal Gold codes has not been considered in any of the overloading schemes. In [12], a new method for generating different orthogonal sets of same length has been proposed. The new algorithm generates (N-1) distinct, orthogonal sets of N sequences of length N. It has been shown that the peak value of crosscorrelation between different sets of same length is less than half the sequence length for 32N ≥ . Such sequence sets would offer low intracell interference, when used in overloaded environment. Recently, the present authors have proposed a new overloading scheme using a set of Gold codes [13], which provides better performance than s-

Figure 1. Overloading scheme in a DS-CDMA cellular system. O/O scheme [7]. In this paper, we have evaluated the BER performance using orthogonal Gold code (OG/OG) sets with IMSD schemes. An efficient iterative multistage detection with soft decision interference cancellation is used to increase the amount of overloading.

This paper is organized as follows. In the next section, we describe the system model for the O/O overloading scheme. In section 3 we explain the IMSD operation and describe the process of iterative interference cancellation. Simulation results are presented and discussed in Section 4. Finally, we present the conclusion of this paper. 2. System Model for OCDMA/OCDMA

In the section we consider a DS-CDMA system with processing gain N and the number of users K users (=M+N), where M is the number of set-2 users. We assume that the channel is a nondispersive additive white Gaussian noise (AWGN) channel and that the different user signals are in perfect time synchronization. The discrete-time matrix model of the received BPSK modulated CDMA signal after demodulation and chip sampling may now be expressed as:

1 2 2 2= + +1 2 1 1r = r + r b A S b A S n (1)

Here r1 and r2 are the received samples from set-1 and set-2 uers respectively; 1 11 12 1, ,......., Nb b b=b and

2 21 22 2, ,......., Mb b b=b represent information bits of two

sets of users, where 1ijb ∈ ± . 1 11 1,.......TT T

N=S s s is the

set of orthogonal sequences for set-1 users and is of

dimension ( )N N× . 2 21 2,.....,TT T

M=S s s is another set of

orthogonal sequences for set-2 users and is of dimension ( )M N× . Additive noise n is normally distributed with

zero mean and variance equal to 2σ . 1A and 2 A are the

OCDMA/OCDMA OVERLOADING SCHEME FOR CELLULAR DS-CDMA 209 USING ORTHOGONAL GOLD CODES AND COMPLEX SCRAMBLING


Figure 2. Block diagram of iterative multistage detection (IMSD) for overloaded DS-CDMA scheme. diagonal matrices of received signal amplitudes of set-1 and set-2 users respectively.

The received chip sampled and demodulated signal (1) is despread and integrated over a bit duration to get a soft estimate about a transmitted bit. These soft estimates are fed to an iterative multistage detector (IMSD) (Figure 2). As N > M set-1 users matched filter outputs are more reliable (due to less MAI) as compared to the set-2 matched filter outputs (MAI=1). In the first iteration we

will assume 02I =0 and accordingly find 1

1b and

estimate 11I . This estimated interference is removed from

the received signal, before taking decisions on set-2 users.

Subsequently we find 12b and estimate the interference 12I on set-1 users from set-2 users. In the second iteration,

we obtain the refined data estimates of set-1 users, 21b

after removing the estimated interference 12I as shown in.

As we have more reliable data estimates of set-1 users, a

refined estimate of 21I is obtained in second iteration.

This process repeats for required number of iterations, so that near single user performance is obtained.

When scrambling is used, the orthogonal Gold codes of both the sets are overlaid by a set-specific pseudo-noise (PN) sequence which is the same for all users within the set. In other words, we have

1 1 2

1[ ]N

N=S α α ........α and 2 1 2 M

1[ ]

N=S β β ........β .

Let 1 11 22 1( , ,..... )TNp p p=P and 2 12 22 2( , ,..... )TNp p p=P

designate the PN sequences overlaying the orthogonal Gold sequences in the two sets of users. In order to split the interference power evenly over the in-phase and quadrature components of the useful signal (irrespective of the carrier phase), we consider complex valued PN sequences: the chips nup takes their values from the set

exp( / 4), exp( 3 / 4), exp( 5 / 4), exp( 7 / 4)

C

j j j jπ π π π=

The scrambling sequence can be deterministic (periodic) or random. In periodic scrambling, the scrambling sequence randomly takes values form the set C and it is kept constant for all symbols. On the other hand, in random complex scrambling, it takes random complex values from the set C for each transmitted symbol.

In the next section, we explain iterative multistage detection scheme, which reduces the high level of interference due to overloading. 3. Iterative Multistage Detection

The received demodulated and chip sampled signal (1) is despread and we obtain soft outputs of the transmitted bits corrupted by multiple access interference (MAI) from other users and AWGN noise. In conventional matched filter detection, these outputs are fed to the decision device to make hard decisions of the transmitted information bits. In this paper, iterative multistage detection (IMSD) technique is used to remove the MAI between two sets users. The basic principle of this receiver is to iteratively remove the estimated interference from each set due to the users of other set in multiple stages such that near single user performance is achieved. The interference power from set2-user (assuming that the useful signal power is normalized) is 1/N, and therefore the total interference power that affects set1-users is M/N. As long as M remains small compared to N, preliminary decisions can be made on the symbols transmitted by set1-users with some good reliability. But each of the set2-users gets an interference power of N (1/N) =1 from set1-users. Clearly the bit error (BER) performance will be poor for this set of users if detection is made prior to interference cancellation. As set1-users are detected with some good reliability, we can

+ +

Set-1 Matched

Filter Output

Set-2 Matched

Filter Output

r

--

1ˆ ib 2

ˆ ib ( 1)2i −I

( 1)2i−I

Composite signal of set-1 users

Composite signal of set-2 users

Sequence spreader

Sequence spreader

1iy 2

iy

Decision Decision

1iI

210 P. KUMAR ET AL.


estimate the interference created from this set on set2-users. This estimated interference is removed from set2-uers before making the decision. Now in second iteration, interference from set2-users on set-1 are estimated form the first iteration outputs of set-1 and a more reliable set1 bits are obtained. This process continues till we get a near single user performance.

To explain the operation of IMSD the following

notations are used: 1ˆ ib and 2ˆ ib are decisions about set-1

and set-2 user data bits at the i -th iteration; 1iy and 2

iy

are set-1 and set-2 matched filter outputs at i -th iteration respectively.

At each stage of iteration, the decision on an information bit is made according to the following expressions:

( )i ( -1)1 1 1 2

ˆ ( ) ( )i T iφ φ= = −b y S r I (2)

( )2 2 2 1ˆ ( ) ( )i i T iφ φ= = −b y S r I (3)

Here the reconstructed interference for two groups in i -

th iteration are 1 1 1ˆi i= 1I b A S and 2 2 2 2

ˆi i=I b A S . We assume

that the reconstructed interference for the first group of users in the first iteration is zero i.e., 0

2I =0. Matched

filter outputs after interference cancellation form the decision vectors 1

iy and 2iy for set-1 and set-2 user data

bits respectively. We have assumed equal power, equal phase

synchronous users in a single cell environment over an AWGN channel. The set-1 matched filters outputs in matrix form may be expressed as

11 1 2( ) i T (i- )= −y S r I (4)

( 1)

1 1 2 2 2 2 2 2ˆ( )T i−= + + −i

1 1 1y S b A S b A S n b A S ( 1)

1 2 2 2 2 1ˆ( )T i T−= + − +1b S S A b b S n (5)

In the case of AWGN channel, amplitude matrix is an

identity matrix, i.e., A = I . For the l -th user of set-1, the matched filter output during i-th iteration is

M( -1)

1, 1, 2, 2, ,k=1

ˆ= + ( - ) + i il l k k l k ly b b b zρ∑ (6)

where l = 1, 2, 3………, N and lz = 1 [S n]Tl is the noise

sample for l -th user. The following notations have been

used: 1,ls = l -th user (set-1 ) signature; 1,lb = l -th user

transmitted data-bit (set-1 ); 2,kb = k -th user transmitted

data-bit (set-2); ( 1)2,ˆ i

kb − = k -th user tentative decision (set-

2) on 2,kb at (i-1) th iteration; ,l kρ =Normalized cross-

correlation value between set-1 l -th user and set-2 k -th

user signatures. The matched filter output 1,i

ly has three

components:

( -1)

1, 1, 2, 2, ,

ˆ( - )1 - 2

i il l k k l k l

Total Noise

My b b b z

k Noise sampledesiredMAI from set users

data

ρ= + +∑=144424443

14444444244444443

(7)

Considering all N set-1 users and sufficient number of set-2 users, we assume Gaussian approximation of MAI. So, the following expression from (7) is used to indicate the total noise:

Total Noise = ( -1)2, 2, ,

1

ˆ( - ) M

ik k l k l

k

b b zρ=

+∑ (8)

In equation (2) and (3), ( )1,lbφ is the decision

function of l-th user (set-1). According to the decision

function ( )1,lbφ , IMSD can be classified as hard decision

interference cancellation (HDIC) or soft decision interference cancellation (SDIC). For HDIC receiver the decision function is defined as:

( ) 1,

1, 1,1,

1 0( )

1 0 l

l ll

bb sgn b

bφ

− <= = > (9)

For SDIC except for the last iteration, where we take

hard decision, in other iterations several nonlinear decision functions can be used. We have used piecewise linear approximation of hyperbolic tangent and is defined as:

( )1,

1,

1, 1,

1, 1,

( )

sgn( )

ll

l l

l l

bb

b sgn b

b b

θθφ

θ

<= =

≥

(10)

Here θ is selected to minimize the average BER.

4. Simulation Results

This section presents the Monte-Carlo simulation results of the proposed scheme with SDIC technique. The simulation has been carried out in MAT-LAB to evaluate the BER performance of the proposed scheme in an AWGN channel.

Relevant simulation parameters are shown in Table 1. The value of the parameter θ is found through simulation as 0.5 for SDIC and it is fixed for all iterations. For all simulations, the system performance is evaluated by means of critical overload. We define the critical overload as the maximum achievable channel over load

( )max maxβ = K -N /N with interference cancellation, so that

the SNR degradation for an average BER of 1e-5 is less than 0.35 dB as compared to single user performance. It is a measure for the maximum acceptable channel overload,



Table 1. Description of some parameters relevant in simulation.

Parameters Specifications

Transmission mode Synchronous

Modulation/ Spreading BPSK/BPSK

Spreading factor, N

64

Spreading codes Orthogonal Gold codes

Power and phase of users

Equal

Type of Receiver SDIC

Assumptions Perfect chip, symbol and carrier synchronization

so that the system performance is degraded slightly as compared to the single user performance.

To increase the amount of overloading an efficient soft decision interference cancellation receiver is used as described in Section 3. In Figure 3, BER performance of this receiver at different overloading has been shown for N=64 at 28%, 25% and 22% overloadings. It is observed that 28% overloading cannot be achieved, with less than 1.0 dB SNR degradation at an average BER of 1e-5. If we reduce the overloading to 25%, the SNR degradation is about 0.35 dB as shown in Figure 3 and we can ensure a BER of 510− for all users. So, we can obtain a critical overload of 25%, when the spreading factor N is 64. We have observed that the amount of critical load is only 19%, when the spreading length is reduced to 32.

The critical channel overload for s-O/O is 3% and 11% for N=32 and 64 respectively [7] for the same set of parameters. So there is a significant improvement in critical channel overload in OG/OG scheme as compared to s-O/O scheme.

In Figure 4, the BER performance of OG/OG scheme with random complex scrambling is shown for N=32. Here, both the sets are scrambled by a set specific complex random scrambling sequence. We observe from the figure that, with complex scrambling the amount of overloading is 31%, with about 0.35 dB SNR degradation as compared to single user performance. In Figure 5, overloading performance with periodic scrambling is shown. It is interesting to observe that the overloading performance increases to 50% with less than 0.35 dB SNR degradation.

In Figure 6, the BER performance with complex random scrambling is shown for N=64. Here we observe that we can support 40 extra users (63% channel overloading), with less than 0.35 dB SNR degradation as compared to single user bound. In Figure 7, BER performance with periodic complex scrambling is shown. It is shown that with periodic scrambling, critical load increases to 78%. This is a significant amount of channel overloading, which can be obtained with complex

scrambling. Hence, complex scrambling increases the amount of overloading significantly in an overloaded DS-CDMA system as compared to unscrambled OG/OG scheme [13]. 5. BER Performance on a Rayleigh Fading

Channel

We notice that the case of an AWGN channel is obtained by taking the received signal amplitude matrix, kA = I . The Rayleigh fading channel model can be described by fading amplitudes generated according to

( ) ( )jI Qk k ka a a= + , where ( )I

ka and ( )Qka are independent

zero-mean real Gaussian distributed random variables with variance ( ) ( )

2 2 1/ 2I Qk ka a

σ σ= = .

In order to compare the performance of these schemes, we define the critical overload as the maximum achievable channel overload ( )max max maxβ = / K -N /NM N =

with interference cancellation receiver, so that the SNR degradation as compared to a single user system at an average BER of 5e-4 is less than 1 dB in an AWGN channel. It has to be emphasized that the receiver does not require any kind of user sorting to yield the desired overloading performance. As a consequence, this measure guarantees that the mean BER performance remains close to that of the ideal BER curve provided that maxM M< . It is worth noting that the BER performance in case of perfect interference cancellation is identical to the performance of a non-overloaded system where the users are orthogonal, and also to the performance of a single-user system. The BER achieved by a single-user transmitting over a Rayleigh fading channel is given by

0

1 11

2 1 /be

b

PN E

= −

+ (11)

In Figure 8, the BER performance of OG/OG scheme with conventional matched filter and SDIC receiver on a Rayleigh fading channel is shown. We have considered set 1 and set 6 users for simulation. Figure 7 also shows the theoretical single-user BER performance over a Rayleigh fading channel. The channel overloading is fixed at 40% (26 extra users). It can be observed that the SNR degaradation at a BER of 45.10− is about 1 dB. So, we can obatain 40% channel overloading on a Rayleigh fading channel with the SDIC receiver for N = 64.

Figure 9 shows the BER performance of scrambled OG/OG scheme, where a single set of orthogonal Gold code is used. The amount of overloading is fixed at 100% for N = 64. It is interesting to observe that the overloading increases considerably from 40% to 100% (64 extra users). The overloading for s-O/O [6] scheme is only 75% with complex scrambling with same set of simulation parameters. When we choose two different sets of orthogonal codes and complex scrambling, the

212 P. KUMAR ET AL.


achievable overloading is again 100%. This is a significant amount of overloading on a flat Rayleigh fading channel. 6. Conclusions

Overloading is an efficient scheme to increase number of users in a DS-CDMA system. In this paper, overloading performance of OCDMA/OCDMA overloading scheme is evaluated, which uses orthogonal Gold codes. It is shown that this scheme with soft decision interference cancellation (SDIC) can overload the DS-CDMA systems by 25% at BER of 1e-5 for N=64, with an SNR degradation of about 0.35 dB as compared to single user bound. The amount of overloading increases significantly from 25% to 78%, with periodic complex scrambling. On a Rayleigh flat fading channel, we can obtain an overloading of 40% at a BER of 5e-4 without complex scrambling and SDIC receiver. With complex scrambling overloading % increases considerably to 100%. 7. References

[1] H. Sari, F. Vanhaverbeke, and M. Moeneclaey, “Multiple

access using two sets of orthogonal signal waveforms,” IEEE Communications Letters, Vol. 4, No. 1, pp. 4–6, January 2000.

[2] P. Kumar, M. Ramesh, and S. Chakrabarti, “Overloading cellular DS-CDMA: A bandwidth efficient scheme for capacity enhancement,” Springer-Verlag LNCS, Vol. 4904, pp.515–527, January 2008.

0 1 2 3 4 5 6 7 8 9 1010

-6

10-5

10-4

10-3

10-2

10-1

100

Eb/No (dB)

BE

R

SDIC (28% channel overload)



single-user performance

Figure 3. BER performance comparison with Soft decision interference cancellation (SDIC) with N = 64 at different values of overloading.

0 1 2 3 4 5 6 7 8 9 1010

-6

10-5

10-4

10-3

10-2

10-1

100

Eb/No (dB)

BE

R




Figure 4. BER performance comparison of OG/OG scheme with random complex scrambling with SDIC recevier for N = 32 at different overloadings.

0 1 2 3 4 5 6 7 8 9 1010

-6

10-5

10-4

10-3

10-2

10-1

100

Eb/No (dB)

BE

R

Matched filter (50%)

SDIC (50%)

Single user

Figure 5. BER performance comparison of OG/OG scheme with periodic complex scrambling with SDIC receiver for N = 32.



0 1 2 3 4 5 6 7 8 9 1010

-6

10-5

10-4

10-3

10-2

10-1

100

Eb/No (dB)

BE

R




Figure 6. BER performance comparison of OG/OG scheme with random complex scrambling with Soft decision Interference cancellation (SDIC) for N = 64.

0 1 2 3 4 5 6 7 8 9 1010

-6

10-5

10-4

10-3

10-2

10-1

100

Eb/No (dB)

BE

R

SDIC (83% overload)

SDIC (78% overload)


Figure 7. BER performance comparison of OG/OG scheme with periodic complex scrambling with Soft decision Interference cancellation (SDIC) for N = 64.

0 5 10 15 20 25 3010

-4

10-3

10-2

10-1

100

Eb/No (dB)

BE

R

conventional receiver

SDIC ( channel overload 40%)

single user performance

Figure 8. Mean BER performance of OG/OG with 40% overload and SDIC receiver over a Rayleigh fading channel without scrambling.

0 5 10 15 20 25 3010

-4

10-3

10-2

10-1

100

Eb/No (dB)

BE

R

Matched filter (100%)

SDIC (100%)

Single user performance

Figure 9. BER performance of s-OG/OG scheme with 100% overloading with random complex scrambling over a Rayleigh fading channel for N = 64.

214 P. KUMAR ET AL.


[3] S. Verdu, “Multi–user detection,” Cambridge University Press, 1998.

[4] J. A. F. Ross and D. P. Taylor, “Vector assignment scheme for M+N users in N-dimensional global additive channel,” Electronics Letters, Vol. 28, August 1992.

[5] R. E. Learned, A. S. Willisky, and D. M. Boroson, “Low complexity joint detection for oversaturated multiple access communications,” IEEE Transactions Signal Processing, Vol. 45, pp. 113–122, January 1997.

[6] F. Vanhaverbeke, M. Moeneclaey, and H. Sari, “DS/CDMA with two sets of orthogonal sequences and iterative - detection,” IEEE Communications Letters, Vol. 4, pp. 289–291, September 2000.

[7] F. Vanhaverbeke and M. Moeneclaey, “Critical load of

oversaturated systems with multistage successive interference cancellation,” IEEE VTC, Vol. 4, pp. 2663–2666, April 2003.

[8] D. Djonin and V. K. Bhargava, “New results on low complexity detectors for oversaturated CDMA systems,”

in Proceedings of Globecom 2001, pp. 846–850, November 2001.

[9] P. Kumar and S. Chakrabarti, “A new overloading scheme for DS-CDMA system,” National Conference on Communication, IIT Kanpur, pp. 285–288, 26–28 January 2007.

[10] K. Yang, Y. K. Kim, and P. V. Kumar, “Quasi-orthogonal sequences for code-division multiple-access systems,” IEEE Transactions on Information Theory, Vol. 46, pp. 982–993, May 2000.

[11] Physical Layer Standard for cdma2000 Spread Spectrum Systems, Realse B, TIA/EIA 3GPP2 C.S0002-B, January 16, 2001.

[12] H. Donelen and T. O. Farrell, “Methods for generating sets of orthogonal sequences,” Electronics Letters, Vol. 35, pp. 1537–1538, September 1999.

[13] P. Kumar and S. Chakrabarti, “A new overloading scheme for cellular DS-CDMA using orthogonal gold codes,” IEEE Vehicular Technology Conference (VTC), pp. 1042–1046, May 2008.



Distributed Relay Diversity Systems for OFDM-Based Networks

Afif OSSEIRAN1, Andrew LOGOTHETIS2, Slimane Ben SLIMANE3

1Ericsson Research, Stockholm, Sweden 3Royal Institute of Technology (KTH), Stockholm, Sweden

E-mail: [email protected], [email protected] Received on April 3, 2008; revised and accepted on August 21, 2008

Abstract In this paper, distributed relay diversity systems are analyzed, modeled and evaluated in an Orthogonal Frequency Division Multiplexing (OFDM) based networks. The investigated distributed relay diversity schemes extend the ideas of a single hop transmit antenna schemes such as Cyclic Delay Diversity (CDD), Space Time Transmit Diversity (STTD), transmit Coherent Combining (CC) and Selection Diversity (SD) to distributed diversity systems. In contrast to the classical single hop system, the antennas in the distributed systems belongs to distributed relays instead of being co-located at the transmitter.

The distributed relay diversity methods considered in this paper: Relay CDD (RCDD), Relay Alamouti (i.e.STTD), Relay CC (RCC) and Relay SD (RSD) are compared to the traditional 1-hop system. Analytical expressions for the received Signal to Interference Noise Ratio (SINR) are derived and used in a dynamic multi-cell multi-user simulator. Results show considerable SINR gains for both Round Robin and Max-SINR schedulers. The SINR gains translate into substantial cell throughput gains, up to 200%, compared to 1-hop systems. Despite its low complexity, the RCDD scheme has similar performance to that of other more sophisticated 2-hop schemes such as Relay Alamouti and Relay Coherent Combining. Marginally better results are observed for the Relay Selection Diversity scheme. Keywords: Coherent Combining, Cooperative Communication, Cyclic Delay Diversity, Multi-User

Diversity, OFDM, Relay Node, Selection Diversity, Space Time Transmit Diversity, System Performance

1. Introduction The main driving force in the development of wireless communication networks and systems is to provide, among other aspects, increased coverage and/or support for higher data rates. At the same time, the cost of building and maintaining the system is of great importance and is expected to become even more so in the future, as data rates and/or communication distances are increased. The problem of increased battery consumption is another area of concern. Until recently the main topology of wireless communication systems has been fairly unchanged, for the three existing generations of cellular networks. The topology of existing wireless communication systems is characterized by the cellular architecture, which consists of fixed radio Base Stations (BSs) and User Terminals (UTs) as the only

transmitting and receiving entities in the network. Several radio access transmission technologies have been proposed to increase capacity, flexibility and/or coverage in communication systems. Orthogonal Frequency Division Multiplexing (OFDM) is an example of such technologies. The OFDM receiver is relatively simple, since the multiple data streams are transmitted over a number of parallel flat fading channels, and equalization is done in the frequency domain involving a single tap filter per frequency tone. Despite the simplicity of the OFDM receiver, un-coded OFDM lacks diversity which is required to combat the fast fading.

One way to introduce diversity in the received signal is to utilize multiple antennas at the transmitter and possibly also at the receiver. The use of Multiple Input

2Airspan Networks, previously with Ericsson Research.

216 A. OSSEIRAN ET AL.


Multiple Output (MIMO) channels offers significant diversity and multiplexing gains relative to single antenna systems [1,2]. Spatial diversity offered by MIMO can thus improve the link reliability and the spectral efficiency relative to Single-Input Single-Output (SISO) channels. An alternative approach is to introduce macro-diversity utilizing cooperative relaying. A relaying system is based on a conventional radio network complemented with Relay Nodes (RNs). The RNs communicate wirelessly with other network elements (e.g. BS, another RN or a UT). A cooperative relaying system is a relaying system where the information sent to an intended destination is conveyed through various routes and combined at the destination. Each route can consist of one or more hops utilizing the RNs. In addition, the destination may receive the direct signal from the source. Such systems offer the possibility to reduce path loss between communicating (relay) entities, which may benefit the end users. Cooperative relaying systems are typically limited to only two (or a few) hops. In the literature, several names are in use, such as cooperative diversity [3] cooperative coding [4], and virtual antenna arrays [5].

Cooperative relaying systems are generally divided into two categories. A signal may be decoded, re-modulated and re-transmitted, or alternatively simply amplified and re-transmitted. The former is known as decode-and-forward or regenerative relaying, whereas the latter is known as amplify-and-forward, or non-regenerative relaying.

The introduction of cooperative relaying systems will increase macro-diversity gains. There are at least two well-known schemes that offer diversity gain: Alamouti diversity based cooperative relaying [6] and coherent combining based relaying, which in addition offers a beamforming gain as described in [7]. Alamouti diversity based cooperative relaying requires a receiver to perform signal processing in accordance with the Alamouti code. Besides that, the Alamouti diversity order is limited to two and has shown little system capacity gain in a WCDMA system [8]. This is also true for other space-time codes similar to Alamouti diversity. Higher order space-time codes cannot be constructed without reducing the code rate. Coherent combining based cooperative relaying requires some sort of feedback to estimate channel phase and amplitude. This increases the control signaling in the system, which in turn may reduce the user data rate. Additionally the feedback channel can be erroneous hence may reduce rather increase the system performance as intended. In the amplify-and-forward schemes the desired signal is amplified at the Relay Nodes (RN) at the expense of amplifying the interference. However, in the coherent combining case, the signal of interest is coherently added whereas noise and interferers are added non-coherently. In the decode-and-forward schemes if the signal is erroneously decoded, the error will propagate to the destination.

Inducing multi-path diversity by means of cooperative communications nodes have already been proposed for a single carrier system [6,9–11]. At the destination the signal consists of several delayed copies due to either the introduced delay at the nodes or asynchronous operation, or due to the propagation delay or the processing delay. The diversity is exploited either by increasing the data symbol length [9] in order to avoid ISI or by using a complex equalizer such a generalized decision feedback equalizer [10] or by introduction Cyclic Prefix (CP) in conjunction with Frequency Domain Equalization (FDE) [11].

In this paper several distributed diversity schemes are investigated. One of these schemes is a new cooperative relaying scheme for future wireless communication systems referred to as Relay Cyclic Delay Diversity (RCDD) [12]. The idea is to use a set of distributed RNs each associated with a different cyclic shift. This cyclic shift can be pre-determined or random. Besides the improved link gain usually obtained from a cooperative relaying system, the proposed scheme introduces frequency selectivity and macro-diversity in OFDM based systems. RCDD scheme, relay Alamouti, relay coherent combining and relay selection diversity are investigated and evaluated in a multi-cell (radio network) scenario. Moreover the received SINR at the mobile unit is derived analytically. To the authors best knowledge, no system level (i.e. multi-user in a multi-cell radio network) investigation that tackles distributed diversity schemes has been reported in the open literature. This comprehensive study is the main objective of the contribution.

The paper is organized as follows. In Section 2, the principle of 2-hop distributed RCDD is introduced and its relation to conventional CDD is explained. Relay communication in combination with the Alamouti scheme is discussed in Section 5. Signal representation for different 2-hop relaying schemes are given in Section 6. SINR derivation is exposed in Section 7. The system modeling and simulation assumptions are given in Section 8. Finally simulation results are presented in Section 9. 2. Distributed Relay Cyclic Delay Diversity As an alternative to the conventional diversity methods, an artificial diversity, which mimics the effects of spatial diversity, can be introduced in wireless communication systems by the use of artificial Delays. The idea is to transmit the same information modulated symbols from separate antennas with different time delays, thereby causing multipath replicas even when the channel is flat. With the help of an equalizer the receiver can extract these paths and achieve a Delay Diversity (DD) gain [13]. For communication systems that employ the IFFT/FFT operators such as OFDM and Single Carrier with Frequency Domain Equalization (SC-FDE), DD can be achieved through the use of a cyclic delay. Cyclic Delay

DISTRIBUTED RELAY DIVERSITY SYSTEMS FOR OFDM-BASED NETWORKS 217


Diversity (CDD) has been used in combination with OFDM in [14,15] and has been used with SC-FDE systems with no additional complexity at the receiver in [11]. In this context, cyclic delay increases the selectivity of the fading channel which in the presence of channel coding can increase the diversity gain of the system.

The use of distributed CDD (i.e. located in different BSs) to exploit macro-diversity has been presented in [16]. The drawback of this scheme is that the wireless infrastructure must rely on Radio Network Controler to deliver in a timely fashion the data to multiple BS. Obviously, the control and data signalling in the backhaul will be very high. To avoid this complexity, distributed RNs should be used instead.

In analogy to CDD and in order to provide frequency selectivity and spatial diversity in a cooperative relaying wireless communication system, a set of distributed RNs is treated as a single entity composed of multiple antennas where at each RN transmit antenna a cyclic shift is applied to the OFDM symbol that is forwarded between the BS and the UT. An example of a cooperative relaying network is shown in Figure 1. The figure shows one cell of a wireless network comprising of a BS, several RNs and a UT.

The transmission is made in two phases as illustrated in Figure 2. In the first transmission phase as it is shown in Figure 2(a), the BS transmits the desired data to the scheduled UTs and to the active RNs in a specific cell. In the second transmission phase (see Figure 2(b)), the active RNs or/and the BS retransmit the same data sent by the BS to the scheduled users in the first transmission phase.

According to the proposed method artificial frequency selectivity and spatial diversity is provided in a cooperative relaying wireless communication system. The artificial frequency selectivity is exploited in conjunction with forward error correction coding (FEC) to provide a coding diversity gain. The BS transmits to K RNs and possibly also directly to the UT. The RNs1 forward the information received from the BS to the UT

using cyclic delay diversity. Each RN applies an individual cyclic shift and adds the cyclic prefix (CP) to solve the problem of Inter Symbol Interference (ISI) and preserve the orthogonality between the OFDM sub-carriers in a fading multi-path environment. The signals are then up-converted from base-band into the RF-band and transmitted. The receiver and transmitter of one specific RN is illustrated in Figure 3 where we have assumed that decode-and-forward is employed.

It is interesting to mention that the UT does not need any information about the number of RNs. The UT receives the multiple access signals and decodes the data that may be combined with the signal directly received from the BS. The combined signal as experienced by the UT will in a sense be similar to the signal from a transmitter with multiple antennas, utilizing CDD, but with the added benefits with regards to gain and coverage associated with cooperative relaying. 3. Relay Selection Diversity Relay selection diversity (RSD) is similar to antenna selection diversity. In RSD the BS will select one RN out of a set of RNs belonging to the BS. In the second transmission phase, the selected RN will forward the information from the BS to the UT. Contrary to the classical antenna selection where the selected antenna for transmission is based on the short term statistics (e.g. fast fading), the selection criterion for RSD is done on a slow basis and consists of the distance and shadow fading gain. In particular the BS station will calculate the channel gains between the UT of interest and all the RNs belonging to that BS based on distance and shadow fading maps. Consequently, the BS will select the RN which has the highest channel gain to the UT of interest.

1All nodes are assumed to be OFDM symbol synchronized in the downlink.

Figure 1. The channel impulse responses of a simple 2-hop cooperative system: One BS, K RNs and a UT.



(a) Phase I.

(b) Phase II.

Figure 2. Transmission phases in a 2-hop relay network.

Figure 3. The receiver and transmitter structure of a RN for RCDD. 4. Relay Coherent Combining Relay coherent combining was first proposed in [7]. It consists of multiplying the transmitted signal at each RN by a phase that compensates the one introduced by the channel. In fact the effective channel at the UT will be a constructive summation of all the RN signals transmitting to the desired UT. 5. Relay Alamouti Diversity Relay Alamouti diversity consists of two distributed relays that are used to mimic conventional STTD. For instance, the BS will transmit an even number of OFDM symbols in the first transmission phase then the RN will re-transmit these symbols during the second transmission phase. For simplicity let us assume that the duration of the transmission phases is two OFDM symbols, e.g.

(0)1s and (1)

1s . For the second transmission phase (of equal duration to the first transmission phase), two RNs attached to the BS are selected. Each of these RNs is equipped with a single antenna as shown in Figure 4. The two antennas of the RNs will act jointly as in the case of two transmitting antennas for a conventional STTD. The

only difference is that each of the antenna is attached to a different antenna system instead of being controlled by the same radio unit. After demodulating the two received symbols the RNs will act jointly as a STTD encoder. While the first RN will resent the same symbols i.e

(0) (1)1 1[ , ]s s , the second RN will swap the order of the

received symbols in addition to conjugating them as it is done in the second antenna of a conventional STTD encoder, i.e. *(1) *(0)

1 1[ , ]−s s . 6. Mathematical Models To illustrate the evaluated concepts we consider K RN nodes and one BS as shown in Figure 1, where the BS and each RN is equipped with one transmit and one receive antenna. Let x be the OFDM symbol at a given symbol interval and = %s x its corresponding modulated data vector. Let us assume a slowly varying fading multi-path channel, and let hk denote the channel impulse response between the transmit antenna of RN k and the receive antenna of the UT. It is assumed that the CP is greater than the maximum delay of the channel. 6.1. Relay Cyclic Delay Diversity



Figure 4. The receiver and transmitter structure of Relay Alamouti. Let x denote the [N×1] post IFFT data vector, where N is the number of sub-carriers. At each RN, the data vector x will then undergo a cyclic shift of length lδ , yielding:

l lδ δ=x P x , with )]([)( Nlnxnxl

δδ += (1)

where Pl is a permutation matrix that applies a delay of length l to a vector x, Pl is a right circulant matrix with

1 [1 ]Nl+ −e as the first row i.e. 1 [1 ]Circ( )Nl l+ −=P e , ek is

column vector with all elements equal to zero except the element at position k which is equal to one. [a]b, denoted as a modulo b, is the remainder of the division a by b. It is well known that a cyclic shift of length l in (1), is equivalent to a phase shift in the frequency domain. Hence, the noise free received OFDM symbol with cyclic shift can be written as

l lδ δ=%x Fx , with 2

( ) ( )l

l

njNx n e x nδ

π

δ

−=% % ,

where [ (0), (1),..., ( 1)]Tx x x N= = −% % % %x Fx , and F is the unitary discrete Fourier transform matrix of size N×N, its (n,m)th element for 1,...1,0, −∈ Nmn , is given by

2 /1( , ) j nm Nn m eN

π−=F (2)

Hence, lδ

%x can be expressed as a function of the

original modulated data vector %x as

l lδ δ=(

% %x P x (3) where

l l

Hδ δ=(P FP F

NNjNj ll ee /)1(2/2 ,...,,1diag −−−= πδπδ .

Combined with the transfer function of the channels between the RN transmit antennas and the receive antenna, this permutation matrix will increase the frequency selectivity of the channel seen at the UT receiver.

In the case of relay CDD, a cyclic shift of length

kδ is applied to the signal at the antenna of the thk

RN before transmission. Let (0)1y and (1)

1y be the received symbols at the UT from the BS and RNs for the first and second transmission phases, respectively. They can be expressed as:

0

(0) (0) (1) (1)1 0 1 0 1 1, 1

1

ˆ,k

K

k kk

δ δ=

= + = +∑y H P x n y H P x n , (4)

where (1)1,ˆ kx is the estimate of the symbol (0)

1x at RN k, n0 and n1 represent the thermal noise at the UT from the first and second transmission phases, respectively.

The matrix kH represents the channel matrix from the kth RN transmit antenna to the receive antenna of the UT. Since kH is a circulant matrix, the channel matrix can be diagonalized as follows:

FhDFH )~( kH

k N= , (5) where2

( ) diag=% %k kD h h

diag (0), (1),..., ( 1)k k kh h h N= −% % % .

Since kH and kδ

P are circulant matrices then

1 1 k

Ke kk δ==∑H H P is also a circulant matrix [17] and

can be decomposed as FhDFH )~(11 e

He N=

where 1eh can be called the effective channel impulse

response from the relays, where its FFT is given by:

∑=

=K

kke k

N1

)~~(~1 δehh ⊙ , (6)

where ⊙ denotes the Hadamard product. Taking the FFT of the received signals in (4) yields:

(0) (0)1 1 0( ) n= +0

%% % %y D h x , (1) (0)1 1 1( )= +%% % %

1ey D h x n , (7)

2D(x) is a diagonal matrix with x on its main diagonal.



where it is assumed that the modulated data is correctly decoded at the different RNs, i.e. )0(

1)1(

,1ˆ xx =k , k∀ . The signals can be combined using the MRC method.

Note that the channels from each antenna do not need to be explicitly estimated. The effective channel impulse response and the channel response from the BS can be estimated using a common time-frequency pilot pattern, which is not antenna specific. The same conclusion is reached when multiple transmit antennas are used at the BS and/or the RNs. 6.2. Relay Coherent Combining The expression of received signals at the UT for the coherent combining is similar to the RCDD case (see (7)). The only difference resides in the expression of the effective channel which for the coherent combining case is simply given by:

∑ ∑= =

==K

k

K

kkk

k

ke

1 1

* ~~~~~

1hh

hhh ⊙ , (8)

where ⊙ is the Hadamard product. 6.3. Relay Alamouti Diversity Let )0(

1y and )1(1y denote the received signals at the

UT corresponding to modulated transmitted symbols )0(

1s and )1(1s from the BS, respectively. Let )2(

1y

and )3(1y the received signals from the RNs during the

second transmission phase. The received signals at the UT can be expressed as:

1)1(

10)1(

10)0(

10)0(

1 , nxHynxHy +=+= (9)

∑∑==

+=+=2

13

)3(,1

)3(1

2

12

)2(,1

)2(1 ,

kkk

kkk nxHynxHy , (10)

where (2)1,kx and (3)

1,k%x are the transmitted OFDM

symbols from the thk RN during the second transmission phase. Assuming a perfect detection at the RNs of the transmitted symbols from the BS during the first phase then:

(2) (0) (0)1,1 1 1= =% %x x s , (3) (1) (1)

1,1 1 1= =% %x x s (11)

(2) *(1) *(1)1,2 1 1= − = −% %x x s , (3) *(0) *(0)

1,2 1 1= =% %x x s (12)

Taking the FFT of the received signals in (9) and (10) during the two transmission phases, diagonalizing the channel matrices Hj for 2,1,0∈j , and using (11) and (12), we obtain:

(0) (0)1 1 0= +0

%% %y h s n⊙ (13)

(1) (1)1 1 1= +0

%% %y h s n⊙ (14)

(2) (0) *(1)1 1 1 2+= −1 2

% %% %y h s h s n⊙ ⊙ (15)

(3) (1) *(0)1 1 1 3+ += 1 2

% %% %y h s h s n⊙ ⊙ (16)

At the UT, the symbols (0)1s and (1)

1s are estimated during the first phase by simply equalizing the received signals defined in (9). Further, a second estimate of the transmitted symbol (0)

1s is obtained at the end of the second phase by applying simply the STTD decoding operations as follows:

(0) * (2) (3) *1 1 1ˆ ( )= +1 2

% %% %s h y h y⊙ ⊙ (17)

(1) * (3) (2) *1 1 1ˆ ( )= −1 2

% %% %s h y h y⊙ ⊙ (18)

7. SINR Derivation 7.1. Received Signal & Combining In this section the received signals at the UT for RCDD, Relay Alamouti, Relay coherent combining and Relay Selection Diversity schemes are derived.

1) Relay Cyclic Delay Diversity: In the previous section, the received signal at the UT was derived by ignoring the intra-cell and inter cell interference. Assuming that Nu radio elements (BS or RN), are transmitting on the same frequency band, then the received signal at the UT of interest during the second transmission phase can be expressed as

0 ,0

(0) (0) (0)1 0 1 0

2

u

i

N

i ii

δ δ=

= + +∑y H P x G P x n (19)

1

,

(1) (0) (0)1 1 , 1

1 2 1

u i

k i k

N KK

k i k ik i k

δ δ= = =

= + +∑ ∑∑y H P x G P x n (20)

where Gi is the channel matrix between the interfering ith radio element and the UT of interest (here UT = 1) during the first transmission phase, xi is the transmitted data sequence by the ith radio element, Ki is the number of active RNs in cell i, and Nu is the number of co-channel links. Assuming that the Gi are circulant, hence these matrices can be diagonalized using the FFT decomposition. Otherwise the interference is counted as a part of the noise term. Taking the FFT of the two vectors in (19) and (20) we obtain

(0) (0) (0)1 1 0

2

uN

i ii=

= + +∑0%% % % % %y h x g x n⊙ ⊙ (21)

(1) (0) (0)1 1 1

2

u

i

N

e ii=

= + +∑1

%% % % % %ey h x g x n⊙ ⊙ , (22)



where 0%H and

1e%H are as defined earlier, and

ie%g is the effective frequency channel response between the ith radio element and the UT of interest and is given by:

ie =%g ∑=

iK

kki

1,

~g . (23)

2) Relay Selection Diversity: The expression of the received signals for relay selection diversity method is similar to the RCDD expressions given in (21) and (22). The only difference is that only one RN instead of several is transmitting in the second phase. The RN is selected out of K RNs based on the slow term channel statistics. Then the channel frequency response between radio element 1 and the UT of interest is given by

ke n max)(~1

=h |])(~[| nkh , Kk ,...,1∈ , (24)

where ][⋅ is the expectation function. Further, the effective channel frequency response between the ith interfering element and the UT of interest is given by

| ( ) | maxie k

n =%g |])(~[| nki,g , ,...,1 iKk ∈ . (25)

3) Relay Coherent Combining: The expression of the received signals for relay coherent combining method is identical to the RCDD expressions given in (21) and (22). The effective channel frequency response between radio element 1 and the UT of interest is as given by (8). Further, the effective channel frequency response between the ith interfering element and the UT of interest can be easily shown to be equal to

| ( ) |ie n =%g ∑

=

iK

kki, n

1

|)(~| g (26)

4) Relay Alamouti Diversity: As in subsection 1 of section 7.1., taking into account co-channel interference into the expression of the received signals for relay Alamouti scheme at the UT for the two transmission phases, equations (13) to (16) become

(0) (0) (0)1 1 0

2

uN

i ii=

= + +∑0%% % %y h s g s n⊙ ⊙ (27)

(1) (1) (1)1 1 1

2

uN

i ii=

= + +∑0%% % %y h s g s n⊙ ⊙ (28)

(2) (0) *(1)1 1 1= − +1 2

% %%y h s h s⊙ ⊙

(0) *(1),1 ,2 1 2

2( )

uN

i i ii=

− +∑ % % %g s g s n⊙ ⊙ (29)

(3) (1) *(0)1 1 1= +1 2

% %%y h s + h s⊙ ⊙

(1) *(0),1 ,2 1 3

2( )

uN

i i ii=

+∑ % % %g s + g s n⊙ ⊙ (30)

It is interesting to mention that the received signals of interest (0)

1s and (1)1s will be amplified by the combined

channel from the transmitting RNs, 1e%h . Replacing the

received signals in (17) and (18) by its expression given by (29) and (30), it can be easily shown that the combined signal of the 2-hop signal takes the form given in (22) with

1

2 21 2| ( ) | | ( ) | | ( ) |e n n n= +% % %h h h , (31)

and

| ( ) |ie n =%g ∑

=

2

1

2, |)(~|

kki ng . (32)

7.2. SINR Expression A generic expression of the received SINR for the two communication phases for the evaluated schemes can be derived from section 7.1. The received SINR for RCDD, relay selection diversity and relay coherent combining can be simply derived from (21) and (22). The SINR of relay Alamouti can be derived from (27), (29) and (30). Assuming that the elements of ( )l

i%x , l 0,3∈ are independent variables with zero mean and having pi as variance, the thermal noise is AWGN (with zero mean and variance 2σ ) then the received SINR on the nth tone from the first and second hop can be shown to be given by

21 0

0 2 22

( ) | ( ) |( )( ) | ( ) |uN

i ii

n nnn n σ

=

Γ =+∑

%

%

p hp g

(33)

1

21

1 2 22

( ) | ( ) |( )

( ) | ( ) |u

i

eN

i ei

n nn

n n σ=

Γ =+∑

%

%

p h

p g (34)

Applying maximum ratio combining, the received SINR of the combined signal on the nth tone becomes

0 1( ) ( ) ( )n n nΓ = Γ +Γ . (35)

The expressions of the effective channel IR 1e%h for

RCDD, relay Alamouti (RALA) and relay coherent combining (RCC) are given in equations (6), (31) and (8), respectively. The channel IR of the interfering BS for RCDD, relay Alamouti, and relay coherent combining are given in (23), (32), and (26), respectively.

It can be clearly seen from (34), that the performance of the different schemes is dictated by the effective channel impulse responses of the useful part and interfering part of the received signal (

1( ), ( )

ie en n% %h g ).

An example is shown in Figure 5 where 1~h and 2

~h are the channel IRs seen at the UT from two RNs. The amplitude of the effective channel of the useful signal for the RCDD scheme as given in (6), with random cyclic shifting in the two RNs, is shown and compared to that of



the RALA scheme as given in (31). It is observed that the applied cyclic shift on the two RNs impacts hugely the form of the effective channel. For instance, as illustrated in Figure 5(b), the RCDD effective channel has a greater amplitude for most of the subcarriers as compared to that of the RALA scheme, up to 5 dB greater for some of the subcarriers. In general, the cyclic shifts introduced at the different relays produce more selective channel (big variance in terms of its amplitude, see Figure 5(a)). The effective channel shown in Figure 5(a) exhibits considerable selectivity, thus having substantially lower amplitude for some sub-carrier and higher for others. In fine, the BS may greatly benefit from this large scale variance in terms of the channel amplitude by choosing to transmit to the desired users on those subcarriers with favorable channel conditions that was produced artificially by the cyclic shift. For RALA and RCC schemes, the channels from the different relays are coherently combined at the UT. Hence their effective channels exhibit less frequency selectivity but a higher power average is observed. 8. System Modeling and Assumptions In this section the air interface characteristics, the system assumptions for the evaluation of the relaying schemes are described. The evaluation is made via a dynamic system simulator that focuses on the physical layer, radio network deployment and algorithms (such as link adaptation and scheduling), utilizing the extended spatial channel model proposed in [18]. The relaying schemes will be evaluated in a 2-hop radio network and compared to a 1-hop system. The most relevant parameters that characterizes the air interface are summarized in Table3 1. Note that only the downlink connection (when the UT is receiving) is considered. An average number of 10 UT per sector is generated. A simple data traffic model is used since the interest is in the relative performance of the simulated schemes and not in the traffic modelling

(a) CDD

~h is very selective.

(b) CDD

~h is the best. Figures 5. Effective channel impulse response samples of the useful signal for the different relaying schemes. Comparison

of Channel Impulse responses of two RNs 1~h and 2

~h , the

RCDD RCDD~h and the RALA RALA

~h .

Table 1. Air interface parameters.

Parameter Value Carrier Frequency 5.0GHz Bandwidth 20MHz Modulation Type [4 16 64 128 256 512] QAMNumber of Subcarrier 512 Number of Data Subcarriers 416 Sub Carrier Spacing 39.062kHz Symbol Period 28.8μs Frame Length 12 OFDM symbol Super Frame Length 2 Frames Coding Rate [1/3 1/2 2/3 8/9] Power Control None Handover Hard Scheduling Max-SINR or Round Robin Relaying Transmission Method Decode and Forward

performance per se. Generated users have full buffers, ready to transmit when they are scheduled. 8.1. BS and RN Deployment The simulated cell plan consists of 7 sites, each site with three sectors, with a regular hexagonal deployment topology as shown in Figure 6. The BSs are represented by red circles and are placed at the intersection of three hexagons. The RNs are placed on a circular arc with a BS as an origin. As seen from Figure 6, three RNs per sector, represented by the star signs were assumed. The distance

3The terms LOS and NLOS designate Line Of Sight and none LOS, respectively.



of the RNs to their associated BSs is set to half of the cell radius. No attempt has been made to optimize the locations of the RNs within the network. Clearly, this optimization must be addressed and will be considered in future studies. 8.2. Path Loss and Channel The radio propagation channel and distance path model is an extension of the one proposed in the SCM [19] and developed by the Winner project4 [18,20,21]. Similar radio propagation conditions between the UT the BS and RN are assumed. Further, as indicated in Section 4, the radio communication between the BS and the active RNs is assumed error free. 8.3. Link to System Interface The model used for the link to system interface is based on the mutual-information metric which accounts for the modulation alphabet [22], called the Mutual Information Effective SINR Metric (MIESM). In fact this method has been shown to be an efficient link quality model for predicting the Packet Error Rate (PER) in system simulations involving low complexity. In order to calculate the PER or BLock Error Rate (BLER) the method follows the following steps: The user SINR of a specific resource element

(according to SINR sampling period in time and frequency in an OFDM system) is calculated after receiver processing.

For each resource element, the average mutual information per bit ( MIB ) is calculated (depending on the modulation used).

The MIB is mapped to a PER depending on the channel coding type and rate.

8.4. Antenna Arrangement The BS is equipped with a 3-Sector site antenna with maximum antenna gain of 14 dBi as defined in [19], while an omnidirectional antenna of 10dBi was assumed at the RNs. 8.5. Radio Network Algorithms The most relevant Radio Network Algorithm (RNA) used to evaluate the relaying schemes are described in this section.

Handover: Only hard handover is considered. In order to avoid ping-pong effect a handover margin of 3 dB is used.

Power Control & Link Adaptation: No power control is applied. Modulation and channel coding were adapted to the channel conditions, which can be considered as an alternative and preferred method to power control.

Scheduling: Two types of scheduler are investigated. The first one is round robin where the scheduler assigns time slot for transmission to each user in equal portions and order. The second investigated scheduler is Max-SINR, unfair compared to the round robin. The Max-SINR scheduler allows to fully exploit the RCDD concept. The frequency bandwidth is divided into several sub-bands (consisting of more than one sub-carrier each), the user having the highest (i.e. maximum) SINR on each sub-band is scheduled. For instance, if the bandwidth is divided into 10 sub-bands then up to 10 users can be scheduled (i.e. one user is scheduled in each sub-band) for the whole available bandwidth. Note that the SINR or an equivalent measure per sub-band of the UTs connected to an BS is considered to be available at the BS. When using the Max-SINR scheduler the MAC objective is to maximize the system throughput. The update rate of the scheduler is equal to the super-frame length (i.e. the duration of 24 OFDM symbols).

Radio Node Selection: The BS selection is based on the path loss and shadow fading. The active RNs in a cell are chosen based on the path loss. During the second transmission phase, only two out of six RNs will retransmit data to the scheduled users. 8.6. Environment Description and Evaluations

Criteria The simulations were carried out using a dynamic system level simulator. The radio network cell plan is simulated over a certain number of snapshots. Each snapshot consists of a large number of super-frames which itself is composed of a certain number of radio frames. Each radio frame consists of a number of OFDM symbols. A set of new users is generated at the start of each super-frame, placed in the cell plan and associated to a

Figure 6. Cell Plan of 21 sectors where each sector is equipped with 3 RNs.

4Winner is a part of the European Union’s 6th Framework Programme.



Table 2. Simulation Parameters.

Parameter Value Number Of Sites 7 Number Of Sectors Per Site 3 Wrap Around Modeled BS power 43dBm RN power 37dBm BS Antenna Gain 14dBi RN Antenna Gain 10dBi UT Antenna Gain 0 dBi UT Speed 3 m/s UT Noise Figure 9dB Shadow Fading Variance for LOS 3.5dB Shadow Fading Variance for NLOS 8dB

BS according to their path loss and shadow fading. On the frame level, first the users will move5 and then the propagation channel conditions will be generated (i.e. path loss, slow and fast fading). Thereafter the traffic model will generate the data packets for all active users. Then, the scheduler will decide which users will be scheduled. The users’ modulation and coding schemes will be decided by the link adaption algorithm. The link to system interface derives the block error rate taking into account the interference of all BSs and RNs. This process will be repeated until the desired number of snapshots is reached. The most relevant simulation parameters are described in Table 2.

Cell throughput is used as a measure to compare the performance of the investigated concepts. The cell throughput is the average number of correctly received bits over the entire simulation time divided by the simulation time and the number of cells. 9. Simulation Results In this section we provide numerical results for the performance of distributed relay diversity schemes such as RCDD, relay Alamouti, relay coherent combining and relay selection diversity and compare it to that of the classical 1-hop system. We look at the cumulative distribution function (cdf) of the received SINR at the UT and the cell throughput. The investigated schemes are: The reference case, i.e. a single hop system and is

designated in the figures by “1-hop”. The relay cyclic delay diversity is denoted

“RCDD”. The relay selection diversity is denoted “RSD”. The relay Alamouti method is denoted “RALA”. The relay coherent combining is designated by

“RCC” in the figures label. For all the 2-hop methods, the signal from the first

transmission (i.e. BS to UT) is combined with the signal from the second transmission (RN to UT) at the UT using maximum ratio combining. 9.1. Performance with Round Robin Scheduling

With Round Robin scheduling, all users are allocated the same share of bandwidth. The cdf of the received SINR at the mobile unit of the 1-hop and 2-hop systems for a cell radius of 1 km is illustrated in Figure 7. We notice that the 2-hop schemes outperform the 1-hop scheme by about 10 dB. However, the performance difference between the various 2-hop schemes is quite small (less than 1 dB). The cdf of the received interference is shown in Figure 8 and that of the received useful signal is shown in Figure 9 for the same cell radius of 1 km. While the experienced interference in the 2-hop schemes is larger than that in the 1-hop scheme, the received power of the useful signal improves due to the gain in terms of path and the additional diversity provided by the fading multi-path channel.

One way to reduce interference within the system is to allow only one relay to be active at a time. This is denoted by “RSD” in the figures. It is observed that such a scheme reduces the experienced interference with reduction in received power.

The SINR gain for the 2-hop schemes translates into a good cell throughput gain as shown in Figure 10. For a cell size ranging from 1 km to 3 km, the relay cyclic delay diversity scheme “RCDD” yields 1.7 to 2.6 times cell throughput gain compared to a single hop system. We also notice that Relay Alamouti “RALA” yields no additional cell throughput gain while requiring additional overheard and higher complexity compared to the case “RCDD”. Although the case “RCC” gives the highest cell throughout, its relative gain compared to the case “RCDD” remains marginal in the order of 10% at the expense of higher overhead, feedback signalling and complexity. Furthermore, feedback errors in the channel state information are expected to degrade substantially the performance of the “RCC” scheme. 9.2. Performance with Max-SINR Scheduling With Max-SINR scheduling (see section 8.-8.5.), the system can take full advantage of the channel variation of the different users as only the best subcarriers are allocated to the user. The cdf of the received SINR of the 1-hop and 2-hop systems is illustrated in Figure 11 where we notice a much higher received SINR in comparison with that obtained with Round Robin scheduling. We also notice that the 2-hop schemes outperform the 1-hop scheme and the “RSD” provides the best performance. The “RSD” scheme experiences less interference in comparison with the other 2-hop schemes and hence, can take better advantage of the variation of the fading multipath channel. This improvement in the received SINR will translate in a better system cell throughput. Figure 12 shows the cell throughput gain with respect to the 1-hop case for different cell radii and a demodulating

5The users are uniformly distributed in the cells which are of hexagonal shapes. They are slowly moving in the cell plan with an average speed of 3 m/s and a small acceleration.



loss of 3 dB. It is observed that the gain is moderate, only 1.2-1.35 times. This is a result of the limited modulation levels used where, with a high received SINR as shown in Figure 11, the 2-hop relaying system becomes modulation limited and the SINR improvement provided by the 2-hop relaying schemes cannot be reflected in the system cell throughput.

In the case of more pronounced demodulation loss (10 dB), the cell throughput gain is evident as illustrated in Figure 13. A cell throughput gain of about 2 times is obtained with the different 2-hop relaying schemes in comparison with the 1-hop scheme.

It can be noticed from Figures 12 & 13 that for small to medium cell radii (i.e. smaller or equal to 2km) the relative cell throughput gain of the 2-hop schemes is roughly constant, whereas for larger cell radii (only shown for the round robin case in Figure 10) the cell throughput and by product the spectral efficiency increases further more. 10. Conclusions Distributed relay diversity methods such as Relay Cyclic Delay Diversity (RCDD), Relay Alamouti, Relay Coherent Combining and Relay Selection Diversity that introduce frequency and macro diversity in a cooperative communication OFDM system were proposed, modeled, and evaluated in a dynamic multi-cell multi-user simulator. Further, analytical expressions for the received Signal to Interference Noise Ratio (SINR) were derived. The results were compared to the traditional 1-hop system. Simulation results showed considerable SINR gains, for both Round Robin and Max-SINR schedulers. As expected, SINR gains translated into substantial cell throughput gains. Finally some interesting observations were made: The investigated distributed diversity schemes

outperforms the classical 1-hop scheme for both Round Robin scheduling and Max-SINR scheduling. These schemes yield up to a factor of 3 times performance gain.

RCDD scheme provides up to 2.7 performance gain compared to a 1-hop system.

Relay Alamouti yields no additional cell throughput gain compared to RCDD.

Relay coherent combining method gives marginal gain compared to RCDD, in the order of 10%, at the expense of higher overhead, feedback signaling and complexity.

Relay selection diversity yields the highest system throughput gain, slightly higher than the other investigated distributed diversity schemes. In fact, selecting a single relay to transmit during the second hop, reduces the experienced interference within the system quite considerably.

Figure 7. cdf of the SINR for 1-hop, RCDD, RSD, RCC and RALA for Round Robin Scheduling assuming 3 dB demodulating loss and a cell radius of 1 km.

Figure 8. cdf of the interference for 1-hop, RCDD, RSD and RALA for Round Robin Scheduling assuming 3 dB demodulating loss and a cell radius of 1 km.

Figure 9. cdf of the signal for 1-hop, RCDD, RSD, RCC and RALA for Round Robin Scheduling assuming 3 dB demodulating loss and a cell radius of 1 km.



Figure 10. Normalized cell throughput versus the cell radius for 1-hop, RCDD, RSD, RCC and RALA for Round Robin Scheduling and 3 dB demodulating loss.

Figure 11. cdf of the SINR for 1-hop, RCDD, RSD, RCC and RALA for Max-SINR scheduling and 3 dB demodulating loss and a cell radius of 1 km.

Figure 12. Normalized cell throughput versus the cell radius for 1-hop, RCDD, RSD, RCC and RALA for Max-SINR scheduling and 3 dB demodulating loss.

Figure 13. Normalized cell throughput versus the cell radius for 1-hop, RCDD, RSD, RCC and RALA for Max-SINR scheduling and 10 dB demodulating loss. 11. References [1] S. M. Alamouti, “A simple transmit diversity technique

for wireless communication,” IEEE J. Select. Areas Commun., Vol. 16, pp. 1451–1458, October 1998.

[2] V. Tarokh, N. Seshadri, and A. Calderbank, “Space-time codes for high data rate wireless communication: Performance criterion and code construction,” IEEE Transactions on Information Theory, Vol. 44, No. 2, pp. 744–765, March 1998.

[3] J. N. Laneman, “Cooperative diversity in wireless networks: Algorithms and architectures,” Ph.D. dissertation, Massachusetts Institute of Technology, Cambridge, MA, August 2002.

[4] A. Stefanov and E. Erkip, “Cooperative coding for wireless networks,” IEEE Transactions on Communications, Vol. 52, No. 9, pp. 1470–1476, September 2004.

[5] M. Dohler, E. Lefranc, and H. Aghvami, “Virtual antenna arrays for future wireless mobile communication systems,” in ICT 2002, Beijing, China, June 2002.

[6] P. Anghel and M. Kaveh, “On the performance of distributed space-time coding systems with one and two non-regenerative relays,” IEEE Transactions on Wireless Communications, Vol. 5, No. 2, pp. 682–692, March 2006.

[7] P. Larsson, “Large-scale cooperative relay network with optimal coherent combining under aggregate relay power constraints,” in Proceedings Future Telecommunications Conference, Beijing, China, pp. 166–170, December 2003.

[8] H. Rong, Z. Zhang, and P. Larsson, “Cooperative relaying based on Alamouti diversity under aggregate relay power constraints,” in Proceedings IEEE Vehicular Technology Conference, Spring, Melbroune, Australia, May 2006.

[9] A. Scaglione and Y. Hong, “Opportunistic large arrays: Cooperative transmission in wireless multihop ad hoc networks to reach far distances,” IEEE Transactions SP ’03, Vol. 51, No. 8, pp. 2082–2092, August 2003.



[10] S. Wei, D. Goeckel, and M. Valenti, “Asynchronous cooperative diversity,” in Conference on Information Sciences and Systems, 2004.

[11] S. B. Simane and A. Osseiran, “Relay communication with delay diversity for future communication systems,” in Proceedings IEEE Vehicular Technology Conference, Fall, Montreal, Canada, September 2006.

[12] A. Osseiran, “Advanced antennas in wireless communications: Co-located & distributed,” Ph.D. dissertation, Royal Institute of Technology, Stockholm, Sweden, May 2006.

[13] A. Wittneben, “A new bandwidth efficient transmit antenna modulation diversity scheme for linear digital modulation,” in IEEE International Conference on Communications, pp. 1630–1634, May 1993.

[14] K. Witrisal et al., “Antenna diversity for OFDM using cyclic delays,” in 8th Symposium on Communications and Vehicular Technology, Benelux, pp. 13–17, October 2001.

[15] M. Bossert et al., “On cyclic delay diversity in OFDM based transmission schemes,” in 7th International OFDM-Workshop, Hamburg, Germany, pp. 1–5, September 2002.

[16] O. Hyunseok et al., “Novel transmit diversity techniques

for broadcast services in cellular networks,” in Proceedings IEEE Vehicular Technology Conference, Spring, vol. 2, Stockholm, Sweden, pp. 896–900, 2005.

[17] P. J. Davis, Circulant Matrices, 2nd Edition, New York: Chelsea, 1994.

[18] D. S. Baum, J. Salo, G. D. Galdo, M. Milojevic, P. Kyösti, and J. Hansen, “An interim channel model for beyond-3G systems,” in Proceedings IEEE Vehicular Technology Conference, Spring, Stockholm, Sweden, May 2005.

[19] 3GPP, “Spatial channel model for multiple input multiple output (mimo) simulations,” Technical Report 3GPP TR 25.996 V6.1.0, September 2003, http://www.3gpp.org/ftp/Specs/html-info/25996.htm.

[20] J. Meinilä, Ed., IST-2003-507581 WINNER I, D5.4, Final report on link level and system level channel models, 2005, No. v1.

[21] -----, IST-2003-507581 WINNER I, D5.2, Determination of Propagation Scenario, 2005, No. v1.

[22] K. Brueninghaus et al., “Link performance models for system level simulations of broadband radio access systems,” in IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), Berlin, Germany, September 2005.



Joint Adaptive Modulation and Power Allocation in Cognitive Radio Networks

Dong LI, Xianhua DAI, Han ZHANG School of Information and Science Technology, Sun Yat-Sen University, Guangzhou, China

E-mail: [email protected] Received on May 4, 2008; revised and accepted on August 25, 2008

Abstract Link adaptation is an important issue in the design of cognitive radio networks, which aims at making efficient use of system resources. In this paper, we propose and investigate a joint adaptive modulation and power allocation algorithm in cognitive radio networks. Specifically, the modulation scheme and transmit power are adjusted adaptively according to channel conditions, interference limit and target signal-to-interference-plus-noise ratio (SINR). As such the total power consumption of cognitive users (CUs) is minimized while keeping both the target SINR of CUs and interference to primary user (PU) at an acceptable level. Simulation results are provided to show that the proposed algorithm achieves a significant gain in power saving. Keywords: Adaptive Modulation, Power Allocation, Cognitive Radio

1. Introduction

With the increasing number of various bandwidth-consuming wireless services, spectrum for available bands becomes more and more scarce. Moreover, these bands are not occupied or underutilized by licensed users most of time, which leads to the waste of bandwidth resources and low spectral efficiency. One solution to this problem is that cognitive (unlicensed) users (CUs) are allowed to have opportunistic access to these idle bands or to the active ones without causing harmful interference to the primary (licensed) user (PU), in order to improve the bandwidth utilization. This technology is called cognitive radio [1,2]. The major advantage of cognitive radio technology is its ability to search for available spectrums in its surrounding environment and adjust its transmit parameters accordingly to enhance the system performance. The transmit parameters, for example, include modulation scheme, beamforming vector, center frequency, transmit power and so on. The whole process can be summarized as “sense-cognition-adaptation”.

In wireless network, a fundamental characteristic is the interference introduced by multi-user or co-channel transmission at the same time or over the same frequency radio channel. It is well-known that power allocation [3,4]

is an effective way to mitigate interference by means of updating transmit powers according to the target SINR. Besides, the effective use of transmit power can not only minimize the interference introduced by other transmit nodes to enhance the capacity, but also conserve energy to prolong battery life. In [3], the author proposed a simple distributed power allocation algorithm, in which the power level at next iteration only depends on target and actual SINR as well as current power level. The goal is to minimize the total power consumption subject to the target SINR requirement. Further studies are shown in [5–8]. In [5], the joint optimization of beamforming and power control is studied in the downlink of a cognitive radio network. The objective of the proposed algorithm is to minimize the total transmit power while satisfying the target SINR constraint of CUs and maximum tolerable interference to PU. However, this work can not be extended to the energy-constrained wireless networks, in which there is a constraint of maximum transmit power for each CU. Literature [6] proposes a cross-layer framework for joint scheduling and power control combined with adaptive modulation in ad hoc networks, which can be viewed as the situation where only CUs share the same frequency band with the absence of PU. Therefore, the proposed algorithm can not be applicable to the case of the co-existence of PU and CUs in the same

JOINT ADAPTIVE MODULATION AND POWER ALLOCATION IN COGNITIVE RADIO NETWORKS 229


frequency band, since it does not consider the interference introduced to PU caused by CUs. While literature [7] and [8] only consider the problem of adaptive modulation and power control of a single CU in the presence of PU.

In contrast to previous work in [5–8], we consider the scenario where one PU and multiple CUs share the same frequency band in wireless networks. So far as we know, little attention has been paid to the topic of joint adaptive modulation and power allocation in cognitive radio networks, in which the protection of PU and the quality of service (QoS) of CUs are assured. In this contribution, our goal is, therefore, to jointly optimize the modulation schemes as well as transmit powers in order to minimize the total power consumption while keeping both the interference to PU and target SINR of CUs at an acceptable level. More specifically, we perform a two-stage power allocation processing for the proposed algorithm: First, transmit powers are allocated to all CUs with the same modulation scheme, under the constraint of target SINR of CUs and a given interference limit to PU; Second, each CU with adaptive modulation scheme adjusts its transmit power based on the first allocated power, in order to reduce the total power consumption.

The rest of this paper is organized as follows: Section 2 describes the system model and basic assumptions. In Sections 3, we develop the proposed algorithm for joint adaptive modulation and power allocation in cognitive radio networks. Performance analysis of the proposed algorithm is investigated in Section 4. Section 5 concludes this paper.

Notation: All vectors and matrices are denoted in bold letters. NI stands forN N× identity matrix. ( , )i jA

denotes the ( , )i j th element of the matrix A . The

operators †( )⋅ , 1( )−⋅ and ( )T⋅ represent pseudo inverse,

inverse and transpose, respectively.

Figure 1. System model with one PU in dashed line and N CUs in solid line.

2. System Model

The cognitive radio network under consideration is composed of one PU and N CUs, which are modeled as a collection of separate (N+1) transmit-receive pairs with a single channel, as illustrated in Figure 1. All CUs are allowed to transmit at the same time and share the same frequency band by adopting code division multiplexing access (CDMA). The transmission mode for each CU is half-duplex in order to avoid self-interference [9] caused by one node simultaneously transmitting and receiving. The channel propagation model is characterized by path loss, which is given by [10]

0 100

( ) ( ) 10 log ( )d

PL d PL d dBd

α= + 0d d≥ (1)

where 0d and d are the reference and transmitter-

receiver (T-R) distance, respectively. α denotes path loss exponent, which depends on propagation circumstance.

Then, the actual SINR for ith CU can be expressed as

0 01,

1ii i

i N

ij j ij i j

G PSINR

G P G PK

γη

= ≠

= ≥+ +∑

(2)

where iP and 0P denote power level of ith CU and

primary user, respectively. iiG is the channel gain over

CU i, ijG and 0iG represent the channel gain between CU

j’ transmitter, primary user’s transmitter and CU i’ receiver, respectively. η is the background noise power

and K denotes the spreading gain. γ is the target SINR

for all CUs and the constraint iSINR γ≥ guarantee the

QoS for ith CU. On the other hand, the total interference introduced to PU is given by

0 01

N

i ii

G Pξ ξ=

= ≤∑ (3)

where 0 jG represents the channel gain between CU j’

transmitter and PU’s receiver and ξ denotes the

maximum tolerable interference for PU. Throughout this paper, we make the following

assumptions: 1) The system consists of an access point (AP) [11] for

dialogues with CUs through a dedicated control channel, and the global information of channel gains is assumed to be available at AP.

2) The local information of channel gains and SINR measurements at the receivers of all CUs are sent to their respective transmitters via a dedicated feedback channel.

3) All CUs are well synchronized, and are assumed to be immobile or move slowly so that the corresponding channel gain remains constant during the convergence of transmit power.

230 D. LI ET AL.


3. Joint Adaptive Modulation And Power Allocation

The objective of this algorithm is to assign constrained transmit powers and available modulation schemes to all CUs, in order to minimize the total power consumption while satisfying the target SINR constraint of CUs. Besides, we should also consider maintaining the interference introduced to the primary user within a given interference limit, since CUs coexist with the PU in the same frequency band. Therefore, we can formulate the following constrained optimization problem:

minimize 1

N

ii

P=∑ (4)

subject to iSINR γ≥ (5)

0ξ ξ≤ (6)

where max[0, ]iP P∈ and maxP is the maximum transmit

power. In what follows, transmit powers which satisfy both

the constraint (5) and (6) are calculated by AP and sent to all CUs, in which the modulation scheme is the same and chosen to guarantee the existence of positive powers. Then, the modulation scheme (or equivalently, target SINR) is modified based on the initial SINR for each CU, in order to maintain a certain BER requirement. It will be shown that, based on the modified target SINR for each CU, the total power consumption is greatly reduced by iteratively updating transmit power of each CU, while both the QoS and interference constraint can also be satisfied. 3.1. First Power Allocation The constraint (5) can be expressed in the following way

0 0

1,

( )Nij i

i jj i j ii ii

G G PP P

KG G

γ γ η= ≠

+− ≥∑ (7)

Note that the target γ is the same for all CUs in this stage.

Let 1 2( , , , )TNp p p=P L , rewrite (7) with equality in the

matrix form, we can obtain

( )γ− =N 1 1I F P U (8)

where 1F and 1U are given by in the following

, , 1,2, ,( , )

0,

ij

ii

Gi j i j N

i j KG

i j

≠ =

=

1F =L

10 0 20 0 0 0

11 22

( ) ( ) ( )( , , , )TN

NN

G P G P G P

G G G

γ η γ η γ η+ + +=1U L

Meanwhile, we can also rewrite constraint (6) with equality in the matrix form as

2 2β =F P U (9)

where 2 01 02 0( , , , )NG G GF = L is a 1 N× vector and

2 ξ=U . Note that, in equation (9), 1β ≥ is a constant,

and the interference constraint of PU can be easily satisfied by increasing the value of β , because the

powers allocated to all CUs are kept at a low level. However, it does not mean that large β will satisfy the

target SINR requirement for all CUs. Combine (8) and (9), the constraints (5) and (6) can be finally expressed in the matrix form as follows

2 2

γβ

N 1 1

UF

I - F UP =

F U14243

(10)

where F is a ( 1)N N+ × matrix and U is a ( 1) 1N + ×

column vector. Note that the equation (10) has a feasible solution, i.e., there exists a positive power vector P , if the following condition holds

max

1

( )γ

λ<

1F (11)

where max( )λ 1F stands for the maximum eigenvalue of

matrix 1F . Otherwise, no CUs can be admitted to share

this frequency band with PU. As for the value of γ , we

will give details in the next section. Once F and U are determined, the powers first allocated to CUs are given by

(1)max

†( , )minP = P F U (12)

where max max max max( , , , )Tp p p=P L is a 1N × column

vector and † 1( )T T−=F F F F .

Remarks: 1) Equation (10) is overdetermined, and (12) is its least

square (LS) solution to the optimization problem of (4) with constraints (5) and (6).

2) The first allocated transmit powers (1)P and fixed modulation scheme (or γ ) are obtained at AP, and this

information is sent from AP to the transmitter of corresponding CU through the dedicated control channel.

Table 1. Constellation size and corresponding minimum

required SINR for target 30 10BER −= [6].

Constellation size (M) SINR

64 179.85

32 113.90

16 45.11

8 27.65

4 9.55

2 4.77



It should be noted that the protection of PU and QoS of CUs are both met without the maximum power constraint. However, due to the fact of power-constrained CUs, the target SINR constraint of all CUs may not be satisfied. In the following, we will address this issue in detail. 3.2. Adaptive Modulation Adaptive modulation enables the system to support high data rate by varying the number of bits per symbol in accordance with the instantaneous SINR, while keeping a target BER requirement. The transmission rate of ith CU for M-ary quadrature amplitude modulation (M-QAM) is given by [6]

2log (1 )i iR k SINR= + ⋅ (13)

where 0

1.5

ln(5 )k

BER= − and 0BER is the target BER

requirement. Note that transmission rate obtained in equation (13) is contiguous and should be quantized to a finite number of integer values in practical situation. The minimum required SINR corresponding to the target

0BER for M-QAM can be obtained by [6]

22 21 0 2

2

1 0 2

log2( ) ,

3 2 ( )

log2( 1)( ) ,

3 1

BER M MI Jerfc rectangular QAM

M I JSINR

BER M MMerfc square QAM

M

−

−

⋅ ⋅+ −− +

=⋅−

−

(14)

where MI 2= and 2

MJ = . As a result, the minimum

required SINRs corresponding to 30 10BER −= for M = 2,

4, 6, 8, 16, 32 and 64 can be calculated and the results are shown in Table 1.

According to Table 1, γ is determined in such a way

that the maximum SINR is chosen from available ones which satisfy the condition (11), in order to achieve high data rate. However, the actual SINR based on the initial allocated power for each CU can not satisfy the minimum required SINR for certain 0BER at the same time.

Therefore, the modulation scheme, i.e. target SINR γ ,

should be modified for each CU. To be specific, assuming that the set Ω is composed of SINRs in Table 1 which are no more than γ , then the modified target

SINR for each CU denoted as 'iγ , is chosen in such a way

that the corresponding required SINR in Ω is no more than its actual SINR. In the worst case, if the actual SINR falls below 4.77, the corresponding CU will abort transmission. For instance, if 9.55γ = and the CUs’

actual SINRs based on the first allocated powers are 10, 5 and 3, respectively, then [4.77, 9.55]=Ω , and therefore

'1 9.55γ = , '

2 4.77γ = and '3 0γ = (no transmission),

respectively. Note that, in this case, the constraint (5) is

changed into 'i iSINR γ≥ . Therefore, the modified target

SINR constraint of all CUs can be satisfied.

3.3. Second Power Allocation Since both the QoS and interference constraint are satisfied as discussed before, we consider reducing the total power consumption for all CUs. Let

' ' ' '1 2( , , , )Ndiag γ γ γ=γ L be a diagonal matrix of size

N N× , then the equation (8) can be rewritten in the following form as

'( 1) ( )k k+ +1 1P = F γ P U (15)

where ( 1)k +P and ( )kP denote power level at next and

current iteration, respectively. Then, the optimal transmit power can be obtained by iteration of [3]

'

max( 1) ( , ( ))( )

ii i

i

P k min P P kSINR k

γ+ = (16)

Note that the above algorithm terminate with convergent power if | ( 1) ( ) |i iP k P k ζ+ − ≤ , where 0ζ > is a

negligibly small error. Based on (1)P and 'γ , it can be

known that the total power consumption will be reduced after second power allocation using equation (16), in

which the initialized power is (1)P . The following theorem supports our conclusion.

Theorem: Given (1)iP and corresponding (1)

iSINR

which satisfy (1) 'i iSINR γ≥ , 1,2, ,i N= L , then there

exists a steady-state (2)iP to achieve '

iγ such

that (2) (1)

1 1

N N

i ii i

P P= =

≤∑ ∑ while satisfying both the QoS and

interference constraint. Proof: It can be known from equation (16) that, if

SINR for each CU satisfies the condition '( )i iSINR k γ≥ , then we have

Table 2. Simulation parameters.

Number of CUs (N) 9

Noise variance (η ) 410−

Negligible error (ζ ) 410−

Reference distance (0d ) 1 m

Path-loss factor (α ) 4

Processing gain (K) 128

Maximum power ( maxP ) 1 W

Constant (β ) 2

Maximum tolerable interference (ξ ) 1 W

Maximum number of iteration 20

232 D. LI ET AL.


' '(1) (1) (1) (1)

max( 1) min( , ( )) ( ) ( )( ) ( )

i ii i i i

i i

P k P P k P k P kSINR k SINR k

γ γ+ = = <

The iteration will terminate if (2) (1) '( )i i iSINR SINR k γ= = ,

so the QoS constraint of CUs is satisfied. That is to say

that there is no negligible change in (1) ( )iP k such that (1) (1)| ( 1) ( ) |i iP Pk k ζ+ − ≤ . Therefore, the convergent

power (2) (1) (1) (1)( ) (1)i i i iP P P Pk= ≤ = . As a result, we have

(2) (1)

1 1i i

N N

i i

P P= =

≤∑ ∑ , and the interference constraint of PU is

also satisfied with ( 2 ) (1)(2 ) (1)

0 0 0 01 1

i i

N N

i ii i

P PG Gξ ξ ξ= =

= ≤ = ≤∑ ∑ .

4. Simulation Results In this section, we provide numerical results to demonstrate the effectiveness of the proposed algorithm

0 5 10 15 20−40

−35

−30

−25

−20

−15

−10

−5

0

Iteration

Pow

er(d

B)

(a)

0 5 10 15 206

8

10

12

14

16

18

20

22

24

26

Iteration

SIN

R(d

B)

(b)

Figure 2. Convergence of transmit power and actual SINR for each CU in the second power allocation stage, where 0P

= 5W. (a) Transmit Power (dB). (b) Actual SINR (dB).

in reduction of the total power consumption while satisfying both the QoS constraint of CUs and interference constraint of PU. Besides, we analyze the effect of different 0P on the proposed algorithm.

In our simulation, we consider the cognitive radio network placed in a 10 10m m× square area, in which transmit nodes are located uniformly and the corresponding receive nodes are random placed within 6 6m m× square area centered around them. The specific parameters used in this simulation are listed in Table 2, in which the channel gain can be expressed as 4

ij ijG d −= ,

where ijd is the distance between jth CU’s transmitter

and ith CU’s receiver.

4.1. Performance Analysis First, we examine the performance of proposed algorithm with respect to power saving. Figure 2(a) shows the convergence property of transmit power for each CU in the second power allocation stage with 0P = 5W, in

0 5 10 15 20−40

−35

−30

−25

−20

−15

−10

−5

0

Iteration

Pow

er(d

B)

(a)

0 5 10 15 206

8

10

12

14

16

18

20

22

24

26

Iteration

SIN

R(d

B)

(b)

Figure 3. Convergence of transmit power and actual SINR for each CU in the second power allocation stage, where 0P

= 10W. (a) Transmit power (dB). (b) Actual SINR (dB).



which the initial powers are (1) =P (1, 1, 0.006, 0.0773, 1, 0.9601, 1, 1, 0.0065) in W according to equation (10),

and the convergent powers are (2) =P (0.8079, 0.2538, 0.0002, 0.0086, 0, 0.1626, 0.7089, 0, 0.0025) in W. From Figure 2(a), we can find that the total power consumption is greatly reduced after second power allocation with adaptive target SINR.

Then, we investigate the SINR performance of the proposed algorithm. Based on (1)P , the actual SINRs for all CUs are (1) =SINR (5.6153, 8.5909, 260.7868, 15.5663, 0.1612, 45.7656, 6.6999, 0.0398, 24.4918). γ is set to be 9.55 so that the constraint (11) is satisfied, and

=Ω (4.77, 9.55). Therefore, the adjusted target SINRs for all CUs in the second power allocation stage are set to be ='γ (4.77, 4.77, 9.55, 9.55, 0, 9.55, 4.77, 0, 9.55). We notice that the target SINRs of two CUs are equal to 0, which means that the corresponding CU will not transmit. Figure 2(b) shows the convergence property of the actual SINR for each CU, in which the actual SINRs for the admitted CUs converge to 4.77 and 9.55, respectively. In other words, the constellation size M is chosen to be 2 and 4, accordingly.

Finally, we examine the performance of the proposed algorithm in terms of the interference constraint of PU. According to equation (3), the actual interferences introduced to PU in the first and second power allocation stage are 8.8 mW and 4.8 mW, respectively, which are far less than the interference limit ξ = 1 W. This is mainly due to the maximum transmit power limit for each CU. 4.2. Impact of Different P0 In our previous study, a constant 0P is chosen for all

simulations. To be more practical, we study the impact of different 0P on the system performance.

5 10 15 20 253

3.5

4

4.5

5

5.5

6

6.5

7

P0(W)

Num

ber

of A

dmitt

ed C

Us

Figure 4. Number of admitted CUs versus P0.

As expected, with the increasing value of 0P , the

interference introduced to all CUs will increase

accordingly. This means the first allocated power (1)P will be increased for each CU in order to satisfy the target

SINR constraint. For instance in Figure 3(a), (1) =P (1, 1, 0.0018, 0.1003, 1, 1, 1, 1, 0.0082) in W. Besides, the

convergent powers are (2) =P (0, 0.3813, 0.0003, 0.0153, 0, 0.2909, 0, 0, 0.0048) in W.

Meanwhile, the increased interference caused by PU will certainly affect the SINR performance for each CU. In this case, the SINR constraint of some CUs will be violated. Therefore, the initial SINRs of these CUs decrease and corresponding target SINRs should be adjusted accordingly, which can be seen from Figure 3(b). In Figure 3(b), we can find that two CUs’ target SINRs are reduced from 4.77 to 0 compared with Figure 2(b). That is to say these CUs abort transmission. Specifically in Figure 3(b), the actual SINRs for all CUs are

(1) =SINR (2.9760, 6.7423, 50.5086, 19.2122, 0.1538, 29.2723, 3.3640, 0.0204, 16.3146), γ = 9.55 and

=Ω (4.77, 9.55). Therefore, the adjusted target SINRs for all CUs in the second power allocation stage are

='γ (0, 4.77, 9.55, 9.55, 0, 9.55, 0, 0, 9.55).

Furthermore, A close observation of Figure 2 and Figure 3 shows that the convergence of transmit power and actual SINR for each CU requires only several iterations with different 0P , which is quite acceptable.

As mentioned before, some CUs are turned off in the second power allocation stage, in which their QoS can not be guaranteed due to the interference from PU. It would be also interesting to study the relationship between the number of admitted CUs and 0P . As can be seen from

Figure 4, the number of accepted CUs decreases with the increasing value of 0P in general. This is because more

CUs’ QoS requirements can not be satisfied and switched off accordingly. In this case, the total power consumption after second power allocation decreases roughly with the decreasing number of admitted CUs. However, we also notice that the number of admitted CUs remains the same for 0P = 15 and 20 W, which means more power

consumption is needed when 0P = 20W than the case of

0P = 15W. It can be known that, when 0P is large enough,

there is no CUs that can be admitted by the system. 5. Conclusions In this paper, we have proposed and investigated a joint adaptive modulation and power control algorithm in cognitive radio networks. Our goal is to minimize the total power consumption while keeping the interference introduced to PU below a given limit and satisfying the SINR constraint of CUs. Specifically, the proposed algorithm is implemented in a two-stage power allocation

234 D. LI ET AL.


processing with fixed and adaptive modulation, respectively, which has been proved to greatly improve the power efficiency. Simulation results are shown to confirm the effectiveness of the proposed algorithm. 6. Acknowledgement This work is supported by the National Science Foundation of China (NSFC), Grant 60772132, and the joint foundation of National Science Foundation of China (NSFC) and Guangdong Province U0635003, and also supported by the Science & Technology Project of Guangdong Province, Grant 2007B010200055. 7. References

[1] S. Haykin, “Cognitive radio: brain-empowered wireless

communications,” IEEE Journal on Selected Areas in Communications, Vol. 23, No. 2, pp. 201–220, February 2005.

[2] J. Mitola, “Cognitive radio: an integrated agent architecture for software defined radio,” Tekn. Dr. dissertation, Royal Institute of Technology (KTH), Stockholm, Sweden, 2000.

[3] G. J. Foschini and Z. Miljanic, “A simple distributed autonomous power control algorithm and its convergence,” IEEE Transactions on Vehicular Technology, Vol. 42, No. 4, pp. 641–646, November 1993.

[4] J. Huang, R. A. Berry, and M. L. Honig, “Distributed interference compensation for wireless networks,” IEEE Journal on Selected Areas in Communications, Vol. 24, No. 5, pp. 1074–1084, May 2006.

[5] M. H. Islam, Y. C. Liangand, and A. T. Hoang, “Joint beamforming and power control in the downlink of cognitive radio networks,” in Proceedings IEEE Wireless Communications and Network Conference, pp. 21–26, March 2007.

[6] W. L. Huang and K. B. Letaief, “Cross-layer scheduling and power control combined with adaptive modulation for wireless ad hoc networks,” IEEE Transactions on Communications, Vol. 55, No. 4, pp. 726–739, April 2007.

[7] M. Hong, J. Kim, H. Kim, and Y. Shin, “An adaptive transmission scheme for cognitive radio systems based on interference temperature model,” in Proceedings IEEE Consumer Communications and Networking Conference, pp. 69–73, January 2008.

[8] L. Guo, P. Wu, and S. Cui, “Power and rate control with dynamic programming for cognitive radios,” in Proceedings IEEE Global Telecommunications Conference, pp. 1699–1703, November 2007.

[9] T. ElBatt and A. Ephremides, “Joint scheduling and power control for wireless ad hoc networks,” IEEE Transactions on Wireless. Communications, Vol. 3, No. 1, pp. 74–85, Januray 2004.

[10] T. S. Rappaport, “Wireless communications principles and practice,” Prentice Hall Inc. 1996.

[11] IEEE 802.22 Working Group on Wireless Regional Area Networks, http://www.ieee802.org/22.



Hardware/Compiler Memory Protection in Sensor Nodes

Lanfranco LOPRIORE

Dipartimento di Ingegneria dell’Informazione: Elettronica, Informatica, Telecomunicazioni Università di Pisa, via G. Caruso 16, 56122 Pisa, Italy

E-mail: [email protected] Received on June 9, 2008; revised and accepted on August 25, 2008

Abstract With reference to sensor node architectures, we consider the problem of supporting forms of memory protection through a hardware/compiler approach that takes advantage of a low-cost protection circuitry inside the microcontroller, interposed between the processor and the storage devices. Our design effort complies with the stringent limitations existing in these architectures in terms of hardware complexity, available storage and energy consumption. Rather that precluding deliberately harmful programs from producing their effects, our solution is aimed at limiting the spread of programming errors outside the memory scope of the running program. The discussion evaluates the resulting protection environment from a number of salient viewpoints that include the implementation of common protection paradigms, efficiency in the distribution and revocation of access privileges, and the lack of a privileged (kernel) mode. Keywords: Access Control, Protection Domain, Protection System, Sensor Node

1. Introduction

In sensor node architectures, stringent limitations in terms of hardware complexity and energy consumption [1] prevent utilization an intrinsically complex device such as a memory management unit for virtual to physical address translation [2]. This situation is not likely to change in the near future. Rather than incrementing the hardware power of the single sensor node, system designers are likely to take advantage of progress in integration technologies to reduce the node size and cost, so as to support new applications using sensor networks connecting an always increasing number of nodes [3,4].

In the absence of a memory mapping device, a single address space is shared by all processes, and the form of protection enforced by address space separation between processes is lacking. The code and data areas of all applications are exposed to the risk of corruption by an erroneous process that can even crash the system kernel [5]. This problem is exacerbated by the fact that the writing of application software for sensor nodes is an especially challenging activity, owing to the limitations in terms of available memory and processing power, event-driven concurrency, requirements of real-time response, dynamic application update, and the need to comply with

a variety of different sensors. Even worse, programmers may usually rely on very limited support for debugging [2]. These considerations suggest that the presence of protection mechanisms between processes, which is a common feature in general-purpose systems, is highly desirable even in sensor node environments.

By taking the salient characteristics of an environment of this type into considerations, we shall propose a form of fine-grained memory protection [6] as a solution to the protection problem, outlined above. Our solution takes advantage of a form of synergy between the hardware and the compiler. The interface of the protection hardware consists of a set of primitives, the protection operations. The compiler inserts the calls to these operations at appropriate points of the object code to enforce separation of memory privileges between tasks while preventing the application programmer from calling these operations explicitly. These are easy compiler tasks, which can be made largely transparent to the programmer. Placing new burdens on the compiler is a tendency now exploited in the solution of several architectural problems, e.g., data prefetching, cache control, translation lookaside buffer management, and instruction scheduling at compile time.

The rest of this paper is organized as follows. Section 2 introduces a simple, low-cost addition to the hardware

236 L. LOPRIORE


Figure 1. Hardware configuration featuring a memory protection unit interposed between the processor and the memory devices. inside the microcontroller, and the software primitives to control this hardware and enforce memory protection. Section 3 evaluates the resulting hardware/compiler memory protection scheme from a number of salient viewpoints that include the implementation of common protection paradigms, efficiency in the distribution and revocation of access permissions, and the lack of a privileged (kernel) mode. For each protection problem, we devise a solution that demonstrates the flexibility of the proposed approach to memory protection. 2. The Protection System We shall refer to a classical sensor node configuration in which a microcontroller includes a processor that interfaces both volatile (RAM) and non-volatile reprogrammable (Flash/ROM) storage devices. The memory space is logically partitioned into 2n blocks of a fixed size. Blocks are the passive entities to be protected from tasks. By the term task we mean any active entity capable of generating memory accesses; thus, a task may be a scheduled computation [5], or, in an event-driven paradigm, the activity produced by a function activated by a hardware interrupt [3,4].

A protection domain is a collection of access permissions for memory blocks which can be randomly scattered throughout the whole memory. When a task is running, it is associated with a domain, called the active domain. When the task performs an access attempt to a given information item in memory, the access terminates

successfully only if the active domain includes access permission for the block storing that information item. 2.1. Hardware Support for Memory Protection At the hardware level, protection is supported by a circuitry inside the microcontroller, the memory protection unit (MPU), interposed between the processor and the memory devices (Figure 1). For each given memory block βi, i = 0, 1, …, n – 1, MPU contains a block protection register BPRi associated with this block. The size of BPRi is d bits, where d is the number of the basic domains ∆0, ∆1, …, ∆d-1 which are supported by the protection system (as will be made clear later, more domains can be defined in terms of unions of the basic domains). Let BPRi,j denotes the j–th bit of BPRi. If set, this bit specifies that domain ∆j holds access permission for block βi. This means that a task running in ∆j can successfully access the memory locations in βi for both read and write.

At any given time, a d-bit register of MPU, the active domain register (ADR), contains a quantity with a single bit set, the j-th bit corresponding to the name ∆j of the domain that is active at that time. An address generated by the processor is partitioned into the index i of a block βi and an offset within this block. Quantity i is sent to the array of block protection registers to select the register BPRi associated with βi. If the result of the bitwise AND of the contents of ADR and BPRi is 0, then domain ∆j has no access permission for βi, and an exception of

HARDWARE/COMPILER MEMORY PROTECTION IN SENSOR NODES 237


protection violation is raised to the processor. If this is not the case, the memory address is delivered to the storage devices and the access to memory is accomplished successfully.

When a given task is running, ADR contains the bit configuration corresponding to the name ∆j of the domain of this task. As a result, the task can freely access (and even corrupt) all the blocks in this domain; whereas it cannot read or modify the contents of the other blocks. In this way, we implement a form of error confinement. Corrupting the memory areas in the task’s own domain is less serious than corrupting information items outside the boundaries of this domain.

Of course, more bits of BPRi can be set at the same given time, to indicate block sharing between domains. If both bits BPRi,j and BPRi,k are set, then block βi is shared by domains ∆j and ∆k, for instance.

Table 1. Protection operations.

Operation Effect

setDomain(∆) Activates domain ∆.

grantAP(β, ∆) Grants access permission for block β to domain ∆. Fails if the active domain does not include this access permission.

revokeAP(β, ∆) Revokes access permission for block β from domain ∆. Fails if the active domain does not include this access permission.

2.2. Protection Operations A set of primitives, the protection operations, makes it possible to access the active domain register and the block protection registers and modify their contents (Table 1). Let bi, dj and dk denote bit configurations featuring a single bit set, i.e. the i-th, the j-th and the k-th bit, respectively. Operation setDomain(dj) writes quantity dj into ADR, thereby activating domain ∆j. Operation grantAP(bi, dk) sets bit BPRi,k, thereby granting access permission for block βi to domain ∆k. Execution terminates successfully only if the active domain ∆j, as specified by the contents of ADR, includes the access privilege to be granted, i.e. bit BPRi,j is asserted. Finally, operation revokeAP(bi, dk) clears bit BPRi,k, thereby revoking the access permission for block βi from domain ∆k. Execution terminates successfully only if the active domain ∆j includes the access privilege to be revoked, i.e. bit BPRi,j is asserted. The protection operations are idempotent; each of these operations yields the same result after applying it multiple times.

It should be clear that a harmful task could well use the protection operations unfairly, to change the active domain and gain control of the blocks in a different domain, for instance. We rely on the compiler to prevent the programmer from inserting explicit calls to the

protection operations into application programs; whereas these calls will be inserted by the compiler at appropriate points of the program code. Of course, it would be easy for the programmer to circumvent a loose protection of this type. Rather that precluding deliberately harmful programs from producing their effects, our protection environment is aimed at confining the consequences of programming errors within the memory scope of the running program. 3. Discussion 3.1. The Protection Model In a traditional paradigm, a protection system is modeled by using an access matrix with one row for each protected object β0, β1, …, βn-1 and one column for each protection domain ∆0, ∆1, …, ∆d-1 (Figure 2). The matrix element corresponding to a given object and a given domain specifies the access rights held by this domain on the object. In a representation by rows, the access matrix takes the form of a set of access control lists, one list for each protected object; the access control list of a given object specifies the access rights held by each domain on this object. In a representation by columns, the access matrix takes the form of a set of capability lists, one list for each domain; the capability list of a given domain specifies the access rights held by this domain on each protected object [7].

Access control lists make it easy to manage the access rights held by all domains for a given object. However, determining the access rights that form a given domain is a costly action that implies inspection of all access control lists. Capability lists allow straightforward administration of the access rights in a given domain and facilitate actions of access right transmission between domains. However, access rights tend to propagate throughout the system. This makes it hard if not impossible to determine the domains that hold access rights for a given object, as is required to revoke access permissions, for instance [8].

In our protection environment, by reserving a bit for each domain, register BPRi implements the concept of an access control list for block βi (Figure 3). Furthermore, the bits in position j of all block protection registers, considered as a whole, form the capability list of domain ∆j. In facts, the array of block protection registers is our hardware implementation of the access matrix, which allows us to take advantage of both methods of access right representation, access control lists and capability lists.

For instance, in the capability list approach, access right transmission between domains corresponds to execution of the grantAP() operation that copies an access permission from the active domain into a given domain. This can be useful to pass ownership of a buffer

238 L. LOPRIORE


Figure 2. Configuration of the access matrix. The ♦ symbol in a given element of the matrix indicates that the domain of the corresponding column holds access permission for the object of the corresponding row.

Figure 3. Hardware implementation of the access matrix.

between tasks, for instance. In the access control list approach, access privilege revocation is obtained by executing the revokeAP() operation and eliminating the access permission for a given object from a given protection domain. Revocation is important when the sharing of a data item is done on a temporary basis, for instance. In spite of its simplicity, this technique allows selective revocation of an access privilege from any subset of the domains that hold this privilege [9].

3.2. Memory Protection and the Privileged Mode At the hardware level, the classical concept of a privileged (kernel) mode corresponds to both a set of privileged instructions and unlimited device access. Increased hardware complexity follows in the implementation of the instruction set as well as in processor interfacing. At the software level, system efficiency is negatively affected by the need to save and then restore the context of the running task at each system call [5]. Furthermore, the privileged mode prevents in-line expansion of the system calls [10]. All these sources of processor time wastage give raise to additional energy costs.

We give no special privilege to the kernel. As a result, we may well expand the protection operations in-line into the object code. In-line expansion will be straightforward and very effective. In fact, at the assembly language level, the protection operations translate into few instructions or even a single instruction (as may be the case for setDomain(), for instance). We never disable protection. Instead, we limit the scope of each application and even of the kernel to the smallest extent necessary to carry out its job.

Of course, our protection hardware may well emulate the unrestricted memory access of a traditional privileged mode. A result of this type will be obtained by writing

the all-one bit configuration into the active domain register. The negative effects of an approach of this type on overall system stability are well known [11]. A better solution is to have the kernel run in its own, separate domain. In this way, a stable kernel can always guarantee a form of cold restart after a system crash due to application memory corruption [12]. 3.3. Protection Domain Switching Let us first refer to an event-driven environment featuring non-blocking functions activated by hardware interrupts. In an environment of this type, when execution of a function is started up, the active domain must change to reflect the memory scope of the new function. A result of this type will be obtained by reserving a specific domain to that function or to a set of correlated functions sharing a common memory scope. The compiler will use the setDomain() operation to produce the necessary domain switch, by inserting a call to this operation at the beginning of the code of each of these functions.

A problem connected with domain switching is that of restoring the previous domain on termination of execution of the activities in the new domain. A common approach relies on a protection stack where to save the name of the old domain. Given the memory restraints of sensor node environments, the cost of a separate stack for each task is usually considered prohibitive [12]. We shall take advantage of the idempotence property of the protection operations. On returning from the new domain, the caller will use setDomain() to restore the original domain, independently of possible situations of coincidence of the new domain with the old.

Of course, the approach outlined above to treat asynchronous hardware interrupts can as well be used to deal with synchronous system calls issued by tasks explicitly. In a system featuring no memory protection, system calls may well take the simple form of a library of

HARDWARE/COMPILER MEMORY PROTECTION IN SENSOR NODES 239


software routines [10]; whereas protection usually forces these calls to be implemented as traps. In our environment, we are in a position to take advantage of linked system primitives or even in-line expansion of the system calls in the application code while preserving separation of access privileges between applications and the system kernel.

Let us now consider a task starting up execution of an operation on a given encapsulated object. In a situation of this type, the memory scope of the task should be enlarged to include the memory area reserved for the internal representation of this object. We can implement this form of amplification of access rights at little effort by relaxing the constraint that, at any given time, the active domain register must contain only one bit set. So doing, we can define the active domain in terms of the union of two or more basic domains, by setting the bits of ADR that correspond to these domains. In our example, let ∆j be the active domain and ∆k be the domain including the internal representation of the encapsulated object. We shall use setDomain() to replace the contents of ADR with the result of the bitwise OR of these contents and a quantity having a single bit set, the k-th bit corresponding to ∆k. So doing, we expand the active domain to be the logic union of the original domain and the domain of the object. 3.4. Hardware Costs As seen in the Introduction, the overall design of a sensor node must carefully comply with stringent requirements in terms of energy consumption and space efficiency [13,14]. As far as protection is concerned, the processor time required to manage the protection information has an energy cost that must be kept low [5], and the memory space reserved for storage of the protection information should be kept to a minimum.

In our design, the protection hardware is a small fraction of the total. Its overall complexity is much lower than that of a memory management unit supporting address translation besides protection. For eight basic protection domains (not counting the domains defined in terms of unions of the basic domains), the cost of the protection circuitry in term of the memory resources for the block protection registers and the active domain register is n + 1 bytes, n being the number of memory blocks. Thus, the memory requirements of the protection information are kept low. Furthermore, the memory protection strategy neither is an inherent source of memory space waste (by implying a separate stack for each process [3,5], for instance) nor produces processor inefficiencies (e.g., by hampering in-line expansion of the calls to the protection operations and the system kernel). 4. Concluding Remarks

A widely-used approach to the construction of sensor node software is to compile and link all applications and the kernel, and then load the resulting system image into the sensor node; the software is now operational as a whole [4]. An alternative is to permit forms of dynamic linking of application programs, to bring a new application into the system or to upgrade an existing application, for instance [15,16]. In both cases, in the absence of a privileged mode and of address space separation between applications, no protection mechanism inhibits application software from corrupting code and data in memory, even within the scope of the kernel.

On the other hand, the ever increasing complexity of sensor node software deserves special attention from the system architect, especially given the possible effects of programming errors, which may spread even outside the node onto the whole sensor network [12].

The costs in terms of both hardware and energy requirements connected with classical forms of memory management and protection are usually considered prohibitive for a sensor node. This is certainly true for a memory management unit supporting address translation and address space separation between processes. On the other hand, we have shown how to take advantage of a synergy between the hardware and the compiler and implement a form of memory protection between application programs and the kernel, at low costs in terms of additional hardware inside the microcontroller, processing time, memory space requirements and energy consumption.

We have been aimed at safety rather than security [10]. The software installed on a sensor node is usually considered as reliable, but, of course, not bug free. Our purpose has been to limit the spread of programming errors outside the memory scope of the running program [17]. This means that we hypothesize a set of protection mechanisms at the network level, preventing delivery to nodes of deliberately harmful programs.

In our opinion, in a sensor node environment, the advantages that ensue from a hardware/compiler approach to protection may well compensate the disadvantages connected with the lack of a privilege mode and of address space separation between the processes. We hope that our design effort will be a significant contribution in this direction.

5. References [1] T. Liu, C. M. Sadler, P. Zhang, and M. Martonosi,

“Implementing software on resource-constrained mobile sensors: experiences with Impala and ZebraNet,” Proceedings of the 2nd International Conference on Mobile Systems, Applications, and Services, Boston, Massachusetts, USA, pp. 256–269, June 2004.

[2] R. Kumar, A. Singhania, A. Castner, E. Kohler, and M. Srivastava, “A system for coarse grained memory

240 L. LOPRIORE


protection in tiny embedded processors,” Proceedings of the 44th Annual Conference on Design Automation, San Diego, California, USA, pp. 218–223, June 2007.

[3] A. Dunkels, B. Grönvall, and T. Voigt, “Contiki - a lightweight and flexible operating system for tiny networked sensors,” Proceedings of the First IEEE Workshop on Embedded Networked Sensors, Tampa, Florida, USA, pp. 455–462, November 2004.

[4] P. Levis, S. Madden, J. Polastre, R. Szewczyk, K. Whitehouse, A. Woo, D. Gay, J. Hill, M. Welsh, E. Brewer, and D. Culler, “TinyOS: an operating system for wireless sensor networks,” in Ambient Intelligence, New York: Springer-Verlag, pp. 115–148, 2005.

[5] H. Cha, S. Choi, I. Jung, H. Kim, H. Shin, J. Yoo, and C. Yoon, “RETOS: resilient, expandable, and threaded operating system for wireless sensor networks,” Proceedings of the 6th International Conference on Information Processing in Sensor Networks, Cambridge, Massachusetts, USA, pp. 148–157, April 2007.

[6] J. Shen, G. Venkataramani, and M. Prvulovic, “Tradeoffs in fine-grained heap memory protection,” Proceedings of the 1st Workshop on Architectural and System Support for Improving Software Dependability, San Jose, California, USA, pp. 52–57, October 2006.

[7] L. Lopriore, “Access control mechanisms in a distributed, persistent memory system,” IEEE Transactions on Parallel and Distributed Systems, Vol. 13, No. 10, pp. 1066–1083, October 2002.

[8] L. Lopriore, “Access privilege management in protection systems,” Information and Software Technology, Vol. 44, No. 9, pp. 541–549, June 2002.

[9] V. D. Gligor, “Review and revocation of access privileges distributed through capabilities,” IEEE Transactions on Software Engineering, Vol. SE-5, No. 6, pp. 575–586, November 1979.

[10] D. Lohmann, J. Streicher, W. Hofer, O. Spinczyk, and W. Schröder-Preikschat, “Configurable memory protection by aspects,” Proceedings of the 4th Workshop on

Programming Languages and Operating Systems, Stevenson, Washington, USA, October 2007.

[11] A. S. Tanenbaum, J. N. Herder, and H. Bos, “Can we make operating systems reliable and secure,” Computer, Vol. 39, No. 5, pp. 44–51, May 2006.

[12] R. Kumar, E. Kohler, and M. Srivastava, “Harbor: software-based memory protection for sensor nodes,” Proceedings of the 6th International Conference on Information Processing in Sensor Networks, Cambridge, Massachusetts, USA, pp. 340–349, April 2007.

[13] A. Eswaran, A. Rowe, and R. Rajkumar, “Nano-RK: an energy-aware resource-centric RTOS for sensor networks,” Proceedings of the 26th IEEE International Real-Time Systems Symposium, Miami, Florida, USA, pp. 256–265, December 2005.

[14] S. Yi, H. Min, J. Heo, B. Gu, Y. Cho, J. Hong, H. Oh, and B. Song, “XMAS: An eXtraordinary Memory Allocation Scheme for resource-constrained sensor operating systems,” in: Mobile Ad-hoc and Sensor Networks, Lecture Notes in Computer Science, Vol. 4325, pp. 760–769, 2006.

[15] A. Dunkels, N. Finne, J. Eriksson, and T. Voigt, “Run-time dynamic linking for reprogramming wireless sensor networks,” Proceedings of the 4th International Conference on Embedded Networked Sensor Systems, Boulder, Colorado, USA, pp. 15–28, October 2006.

[16] C. Han, R. Kumar, R. Shea, E. Kohler, and M. Srivastava, “A dynamic operating system for sensor nodes,” Proceedings of the 3rd International Conference on Mobile Systems, Applications, and Services, Seattle, Washington, USA, pp. 163–176, June 2005.

[17] N. K. Jha, S. Ravi, A. Raghunathan, and D. Arora, “Architectural support for safe software execution on embedded processors,” Proceedings of the 4th International Conference on Hardware/Software Codesign and System Synthesis, Seoul, Korea, pp. 106–111, 2006.



On Energy-Efficient Node Deployment in Wireless Sesnor Networks

Hui WANG1, KeZhong LU2, XiaoHui LIN1

1 Department of Electronic Engineering, Shenzhen University, Shenzhen, China 2 Supercomputing Center, Shenzhen University, Shenzhen, China

Email: wanghsz, kzlu, xhlin @szu.edu.cn Received on July 15, 2008; revised and accepted on August 26, 2008

Abstract In wireless sensor networks, sensor nodes collect local data and transfer to the base station often relayed by other nodes. If deploying sensor nodes evenly, sensor nodes nearer to the base station will consume more energy and use up their energy faster that reduces system lifetime. By analyzing energy consumption, a density formula of deploying nodes is proposed. The ratio of whole energy of sensor nodes to energy consumption speed of sensor nodes in every area can get consistent if deploying nodes by the density formula, therefore system lifetime is prolonged. Analysis and simulation results show that when communication dominates whole energy consumption and the monitored region is big compared with radio range of sensor node, system lifetime under this scheme can be 3R/(2t) times of that under deploying nodes evenly, where R is radius of the monitored region and t is radio range of sensor node. Keywords: Wireless Sensor Networks, Sensor Node, Deploying Node

1. Introduction Recently, wireless sensor networks composed of large sensor nodes come to realize [1–4]. Sensor nodes have very limited processing and communication capabilities. Sensor networks are multi-hop and sensor nodes play a dual role as both data generators and routers [5]. Energy is identified as one of the most crucial resources in sensor networks dual to the difficulty of recharging batteries of thousands of devices in remote or hostile environments [6–8]. Researches show that energy consumption of sensor node is dominated by communication [4,9].

In a typical sensor network, there is a base station in the network. Sensor nodes sense environment, collect sensor reading, process the data and then forward the information to the base station. An example of such applications is habitat monitored [10]. Sensor nodes are deployed in the habitat and the base station collects data from sensor nodes. So information of the habitat can be achieved from the base station.

Because remote sensor nodes transferring data to the base station needs some close nodes to relay, data stream density nearer to the base station is bigger. If sensor

nodes are deployed evenly, nodes nearer to the base station will consume more energy. If close nodes use up their energy, other nodes can’t transfer their data to the base station and system lifetime of sensor network is over.

Presently, there’s few research about deploying nodes with the purpose of prolonging system lifetime. Most researches assumed that sensor nodes are deployed evenly scattered by airplane or other tools in the monitored region [11,12]. So system lifetime of the sensor network can’t be long. [13] researched data stream in wireless sensor network and showed that data streams of different areas aren’t balancing. Data stream of middle area is dense while data stream of boundary area is sparse. This paper determinates rational scheme of deploying nodes by researching energy consumption in different areas to prolong system lifetime.

The remainder of this paper is organized as follows. In section 2, assumptions and base model are given. Section 3 describes how to deploy nodes in wireless sensor networks. Section 4 presents some theoretic analysis and section 5 presents a comparative performance evaluation using simulation. This paper concludes with section 6.

242 H. WANG ET AL.


2. Assumptions and Base Model This section presents the basic model of the sensor network that this paper targets. The network model makes the following assumption:

− Wireless sensor network is large-scale. There are many sensor nodes and a base station in the network. Monitored region of the sensor network is a circle which radius is R. The base station is located in the center of the circle.

− Sensor nodes are homology. Initial energy of sensor node is e.

− Each sensor node senses environment and transfers local information to the base station. Events occur evenly in the monitored region. Data generating speed is λ per unit area.

− Sensor nodes communicate with the base station by delivering data across multiple hops. Radio range of sensor node is t.

− Energy consumption of sensor nodes transmitting unit data and receiving unit data is T and E respectively.

− Energy consumption not caused by communication is evenly distributed in the monitored region. Other energy consumption speed per unit area is W.

In above, we assume that the monitored region is a circle for the sake of analyzing easily. System lifetime is mostly optimized when the base station is located in the center.

3. Scheme of Deploying Nodes In this section, we will deduce a density formula of deploying nodes that will prolong system lifetime of wireless sensor networks.

In wireless sensor networks, the ratio of whole energy of sensor nodes to energy consumption speed of sensor nodes in every area should be consistent. Thus more sensor nodes will be deployed in area where energy is consumed sooner. As a whole, sensor nodes in different areas will tend to use up energy at the same time. Therefore system lifetime gets prolonged furthest.

Because we have assumed that the monitored region is center symmetry, node density should be only relational with distance to the base station. We denote ρ(r) as node density of point that is r far from the base station.

Next we analyze energy consumption in area B in Figure 1. Area B is a ring centered at the base station that inner radius is r-t/2 and outer radius is r+t/2. Because radio range of sensor node is t, sensor nodes that are farther from the base station than r+t/2 (i.e. sensor nodes outside area B) transferring data to the base station needs some node in area B to relay.

Data generating speed outside area B is∫ + xdxλπ Rtr 2/2 .

Energy consumption speed of sensor nodes in area B

Figure 1. Energy consumption in area B.

relaying data generated outside area B is )(2 ETxdxλπR

t/2r+×∫ +

.

Data generating speed in area B is∫+

−

2/

2/2

tr

trxdxλπ .

Energy consumption speed of sensor nodes in area B

transmitting data generated in area B is Txdxλπtr

tr×∫

+

−

2/

2/2 .

Other energy consumption speed in area A is /2

/22

r t

r txdx Wπ

+

−×∫ . So energy consumption of sensor

nodes in area B is

/ 2 /2

/2 /22 ( ) 2 2

R r t r t

r t / 2 r t r txdx T E xdx T xdx Wπ λ π λ π

+ +

+ − −× + + × + ×∫ ∫ ∫ .

The number of sensor nodes in area B is /2

/22 ( )

r t

r tx xdxπ ρ

+

−∫ . So the whole energy of sensor nodes

in area B is /2

/22 ( )

r t

r tx xdx eπ ρ

+

−×∫ .

The ratio of whole energy of sensor nodes to energy consumption speed of sensor nodes in area B should be a constant. So following condition is satisfied:

/2

/ 2/ 2 / 2

/ 2 / 2

2 ( )

2 2 2

r t

r tR r t r t

r t / 2 r t r t

x xdx eb

xdx (T E ) xdx T xdx W

π ρ

π λ π λ π

+

−+ +

+ − −

×=

× + + × + ×

∫

∫ ∫ ∫

where b is a constant. Because node density of each point in area B is almost

equal. So we approximately choose ρ(r) as average node

density of area B. i.e. 2

/22 ( )

r t

r tx xdxπ ρ

+

−≈∫

/ 2

/22 ( )

r t

r tr xdxπ ρ

+

−∫ .

So we can get: /2

/2/2 / 2

/2 /2

2 ( )

2 2 2

r t

r tR r t r t

r t / 2 r t r t

r xdx eb

xdx (T E ) xdx T xdx W

π ρ

π λ π λ π

+

−+ +

+ − −

×=

× + + × + ×

∫

∫ ∫ ∫

From the above formula, we can deduce:

2 2 2[ ( ) ( / 2) ( / 2) ] 2( )

2

b R T E r t T r t E brtWr

rte

λρ + − − − + += .

AN ENERGY-EFFICIENT SCHEME OF DEPLOYING NODES IN 243 WIRELESS SENSOR NETWORKS


When r/t large, we have r-t/2≈r, r+t/2≈r. So we get:

2 2

( ) ( )R r

r c dr

ρ −= + , where 2

( )

tWc

T Eλ=

+, ( )

2

T E bd

te

λ+= .

Assume that the number of all sensor nodes is n, then

nrdrrR

=∫0 )(2 ρπ . We can deduce )34(

32 cRR

nd

+=

π. So we

can obtain node density of point that is r far from the base station:

)34(

3)()(

2

22

cRR

nc

r

rRr

+×+−=

πρ , where

)(

2

ET

tWc

+=

λ.

From above density formula, node density of close area is bigger than that of remote area. It is consistent with our expectation.

ρ(r) consists of two parts: r

rR 22 − and c, where r

rR 22 −

is reverse with distance to the base station and c is a constant independent of distance to the base station. In

the formula of )(

2

ET

tWc

+=

λ, λ(T+E) is energy

consumption speed of sensor nodes transmitting and receiving data and W is other energy consumption speed. If the proportion of other energy consumption speed is bigger, c is bigger and node densities of the monitored region tend to be more even. Otherwise node densities of the monitored region tend to be more uneven. This is because other energy consumption speeds in different areas are close while energy consumption speeds of sensor nodes transmitting and receiving data in different areas are different.

4. Analysis of System Lifetime In this section, we will analyze system lifetime under deploying nodes by the density formula deduced in last section compared with deploying nodes evenly in wireless sensor network.

First analyze system lifetime under deploying nodes

by the density formula )()(22

cr

rRr +−=ρ ×

)34(

32 cRR

n

+π.

Because in this case the ratio of whole energy of sensor nodes to energy consumption speed of sensor nodes in every area is consistent, sensor nodes in every area tend to use up their energy at the same time. Then system lifetime approximates the ratio of whole energy of sensor nodes to energy consumption speed of sensor nodes in some area. Therefore we can only analyze system lifetime of sensor nodes in area C which is less than t far from the base station.

Energy consumption speed of sensor nodes in area C

is 0 0

2 ( ) 2 2R t t

txdx T E xdx T xdx Wπ λ π λ π× + + × + ×∫ ∫ ∫ .

Whole energy of sensor nodes in area C is ∫t

xdxr0

)(2 ρπ =

)34(

)326(2

232

cRR

ctttRne

++−

. So system lifetime is:

2 3 2

2

0 0

(6 2 3 )

2 (4 3 )[ ]R t t

t

ne R t t ct

R R c xdx (T E ) xdx T xdx Wπ λ λ

− +

+ × + + × + ×∫ ∫ ∫.

Next let’s analyze system lifetime under deploying nodes evenly. Sensor nodes outside area C transferring data to the base station needs some node in area C to relay. Because sensor nodes are deployed evenly, energy consumption speed of sensor nodes in area C is fastest. Therefore sensor nodes in area C will use up their energy earliest. When sensor nodes in area C use up their energy, the base station can’t receive any data from the sensor network and system lifetime is over. So system lifetime approximates the ratio of whole energy of sensor nodes to energy consumption speed of sensor nodes in area C also.

Energy consumption speed of sensor nodes in area C

is 0 0

2 ( ) 2 2R t t

txdx T E xdx T xdx Wπ λ π λ π× + + × + ×∫ ∫ ∫ . Whole

energy of sensor nodes in area C is neR

t2

2

. So system

lifetime is: 2

2

0 02 [ ( ) ]

R t t

t

net

R xdx T E xdx T xdx Wπ λ λ× + + × + ×∫ ∫ ∫.

The ratio of system lifetime under deploying nodes by

the density formula )34(

3)()(

2

22

cRR

nc

r

rRr

+×+−=

πρ to

that under deploying nodes evenly is:

Rc

tc

tcR

R

tcR

cttR

43

23

)3(4

6

)3(4

326 222

+−+

+=

++−

,

where t and c are fixed values. From the above ratio, we can see system lifetime

under deploying nodes by the density formula is long than that under deploying nodes evenly. When other energy consumption is little and the monitored region is big, i.e. c≈0 and R»t , the ratio approximates to 3R/(2t). Because t is a fixed value, the ratio is bigger when the monitored region is larger. 5. Simulation In this section, we compare system lifetime under deploying nodes by the density formula proposed by this paper with deploying nodes evenly by simulation. We adopt ns-2.28 simulator [14] as experiment platform. We use the following model for our simulation study: − MAC protocol is 802.11 DCF. − Radio bandwidth is 1 Mbps. − Radio range is 50 m. − Initial power of sensor node is 10000 J. − Sensor node’s sending and receiving power are 0.660

244 H. WANG ET AL.


W and 0.395 W respectively. − The size of packet is 64 B. − Other energy consumption of sensor node is 0. − The occurring of event in the monitored region

satisfies Poisson distribution. The speed of event occurring in area is 0.01 m-2s-1.

− Sensing data of one event has 10 packets averagely. We simulate under various sizes of the monitored

region. Choose radiuses of the monitored region be 200 m, 400 m, 600 m, 800 m, and 1000 m. Fix average node density of the whole monitored area be 1/(400π) m-2. So the numbers of sensor nodes in various monitored regions are 100, 400, 900, 1600 and 2500 respectively. We choose system lifetime as the time from beginning to when average ratio of event being successfully monitored by the base station is under a threshold of 90%. Observe system lifetime under various simulation conditions. In order to make results more precise, we simulate 10 times for each simulation condition and choose average value.

Table 1 shows system lifetime under various conditions. R denotes radius of the monitored region. t denotes radio range. n denotes the number of sensor nodes. α denotes system lifetime under deploying nodes by the density formula proposed by this paper. β denotes system lifetime under deploying nodes evenly. The ratio of system lifetimes under these two schemes is near to 3R/2t. It’s consistent with analysis result in section 4. Figure 2 shows comparison of system lifetimes under these two schemes. System lifetime under deploying nodes by the density formula proposed by this paper is much longer than deploying nodes evenly.

Table 1. System lifetime under various conditions (t is 50 m).

R(m) n α(104s) β(103s) α/β 3R/2t

200 100 4.61 8.21 5.61 6

400 400 2.26 1.87 11.40 12

600 900 1.47 0.86 17.15 18

800 1600 1.08 0.47 22.89 24

1000 2500 0.85 0.30 28.62 30

Figure 2. System lifetime: deploying nodes by the density formula vs. deploying nodes evenly.

6. Conclusions This paper deduces the density formula

)34(

3)()(

2

22

cRR

nc

r

rRr

+×+−=

πρ of deploying nodes in

wireless sensor networks by analyzing energy consumption speeds in different areas. The ratio of whole energy of sensor nodes to energy consumption speed of sensor nodes in every area can get consistent if deploying nodes by the density formula. Then we analyze system lifetime under this scheme of deploying nodes. When communication dominates whole energy consumption and the monitored region is big compared with radio range of sensor node, system lifetime under this scheme can be 3R/(2t) times of that under deploying nodes evenly, where R is radius of the monitored region and t is radio range of sensor node. Finally simulation results validate this conclusion.

7. Acknowledgement The research was jointly supported by research grants from Natural Science Foundation of China under project number 60602066 and 60773203, grant from Guangdong Natural Science Foundation under project number 5010494. The work has also got support from Foundation of Shenzhen City under project number QK200601. Corresponding author: Ke-Zhong Lu ([email protected]).

8. References [1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E.

Cayirci, “Wireless sensor networks: a survey,” Computer Networks, 38(4), pp. 393–422, 2002.

[2] S. Brown and C.J. Sreenan, “A new model for updating software in wireless sensor networks,” IEEE Network, 20(6), pp. 42–47, 2006.

[3] L. Cui, H. L. Ju, Y. Miao, T. P. Li, W. Liu, and Z. Zhao, “Overview of wireless sensor networks,” Journal of Computer Research and Development, 42(1), pp. 163–174, 2005.

[4] J. M. Kahn, R. H. Katz, and K. S. J. Pister, “Next century challenges: mobile networking for smart dust,” in Proceedings of the 5th Annual ACM/IEEE International Conference on Mobile Computing and Networking, pp. 263–270, 1999.

[5] H. S. Kim, T. F. Abdelzaher, and W. H. Kwon, “Minimum-energy asynchronous dissemination to mobile sinks in wireless sensor networks,” in Proceedings of the 1st International Conference on Embedded Networked Sensor Systems, pp. 193–204, 2003.

[6] Y. Yang, V. K. Prasanna, and B. Krishnamachari, “Energy minimization for real-time data gathering in wireless sensor networks,” IEEE Transactions on Wireless Communications, 5(11), pp. 3087–3096, 2006.

syst

em li

fetim

e (1

03 s)

AN ENERGY-EFFICIENT SCHEME OF DEPLOYING NODES IN 245 WIRELESS SENSOR NETWORKS


[7] H. Kwon, T. H. Kim, S. Choi, and B. G. Lee, “A cross-layer strategy for energy-efficient reliable delivery in wireless sensor networks,” IEEE Transactions on Wireless Communications, 5(12), pp. 3689–3699, 2006.

[8] Y. W. Hong and A. Scaglione, “Energy-efficient broadcasting with cooperative transmissions in wireless sensor networks,” IEEE Transactions on Wireless Communications, 5(10), pp. 2844–2855, 2006.

[9] Y. Yu, V. K Prasanna, and B. Krishnamachari, “Energy minimization for real-time data gathering in wireless sensor networks,” IEEE Transactions on Wireless Communications, 5(11), pp. 3087–3096, 2006.

[10] A. Mainwaring, J. Polastre, R. Szewczyk, D. Culler, and J. Anderson, “Wireless sensor networks for habitat monitored,” in First ACM Workshop on Wireless Sensor

Networks and Applications, pp. 88–97, 2002.

[11] J. Chen, Y. Guan, and U. Pooch, “A spatial-based multi-resolution data dissemination scheme for wireless sensor networks,” in the 5th IEEE International Workshop on Algorithms for Wireless, Mobile, Ad Hoc and Sensor Networks, 2005.

[12] J. Chen, Y. Guan, and U. Pooch, “An efficient data dissemination method in wireless sensor networks,” in the IEEE Global Telecommunications Conference, 2004.

[13] U. Bilstrup, K Sjoberg, B. Svensson, and P. A. Wiberg, “Capacity limitations in wireless sensor networks,” in Proceedings of IEEE International Conference on Emerging Technologies and Factory Automation, pp. 529–536, 2003.

[14] “Network Simulator”, http://www.isi.edu/nsnam/ns.



Relations among Mobility Metrics in Wireless Networks

Xiao SHU, Xining LI

Computing and Information Science, University of Guelph, Canada E-mail: xshu, xli @uoguelph.ca

Received on June 1, 2008; revised and accepted on August 29, 2008 Abstract In wireless network simulation analysis, researchers tweak mobility metrics, such as the speed or the pause time of the nodes, to get different stability levels of the network. Meanwhile, in theoretical analysis, link failure rate is widely used to model the stability of a wireless network. This paper presents an analysis of a simplified mobility model and shows that the link failure rate is positively correlated with the average speed of nodes in this model. Though this result is based on a mobility model with many restrictions, a simulation evaluation suggests that the result still holds in the popular random waypoint model and random direction model. Based on this observation, this paper also encourages the use of link failure rate as mobility metric instead of the problematic pause time in future practice. Keywords: Mobility Metrics, Link Failure Rate, Simulation

1. Introduction

In simulation analysis of wireless networks, the movement of virtual nodes follows certain patterns called mobility models. The best known and widely used mobility model is the random waypoint model in which each host alternately pauses for a random length of time and moves to a new location at a random speed [1]. Recent research has focused a lot of attention on other models, such as the random direction model [2], random trip model [3] and empirically-based models [4]. Whatever the mobility model used in a simulation, in general, if nodes pause shorter or move faster, two nearby nodes will have a higher probability of moving out of the radio range of each other and the topology of wireless networks will be more unstable. The pause time and the speed of nodes are widely used as mobility metrics to measure the stability of a wireless network in simulation analysis.

Differing from simulation studies, theoretical analysis jumps to modeling the stability of wireless links directly, as they are easier to handle mathematically. It is common in theoretical analysis to make assumptions that the lifetime of links follows a probability distribution, such as the exponential distribution [5]. Generally, the shorter the average link lifetime is, or the more links fail in a unit of time, the more unstable a wireless network will be.

Both simulation and theoretical analysis can provide valuable information about the characteristics of a wireless network. However, the mobility models and mobility metrics used in these two methods are different; thus, the results produced by them are not directly comparable. This paper presents an analysis of a simplified mobility model and shows that the link failure rate is positively correlated with the average speed of nodes in this model. This result creates a mathematical bridge between these two analysis methods, and suggests that average speed is a better mobility metric than the widely used pause time as it is linear with link failure rate. Although we obtain this conclusion under a mobility model with strict restrictions, a simulation evaluation suggests that the analysis result still holds in the popular random waypoint model and random direction model.

The remainder of this paper is organized as follows. Section 2 outlines related work. Section 3 presents the mathematical relation between link failure rate and node speed. Section 4 analyzes the properties of mobility metrics. In Section 5, we conclude this paper and address some ideas about future research. 2. Related Work The most popular mobility model, the random waypoint model, was first used by Johnson and Maltz in the

RELATIONS AMONG MOBILITY METRICS IN WIRELESS NETWORKS 247


evaluation of the Dynamic Source Routing (DSR) protocol [1,6]. It is implemented in the simulation tools ns-2 [7] and GloMoSim [8] and widely used in evaluations of network algorithms and protocols. In a typical simulation with random waypoint model, N nodes are placed at random initial locations over a rectangular area of size Xmax × Ymin. Each node is then assigned a destination which is uniformly distributed over the two-dimensional area with a speed v, which is either in the form of a constant value or in the form of a certain distribution [1,9,10], such as uniform distribution over (0, Vmax]. A node will then start travelling toward the destination on a straight line, at the chosen speed v. Upon reaching the destination, the node stays there for some constant or random pause time. When pause time expires, it chooses the next destination and speed in the same way, and the process repeats until the simulation ends.

The original random waypoint model is problematic. It has been observed in [11,12] that the spatial distribution of nodes tends to be denser at the center of the rectangular simulation area as the simulation runs. Bettstetter et al. presented a solution to the spatial distribution changing problem by setting up a proper initial distribution of nodes [13]. Another problem is that average speed of nodes decreases gradually as more and more nodes become “stuck” travelling long distances at low speeds [14]. In particular, if the random speed is uniformly chosen from (0, Vmax], the average speed will decrease to 0 over time rather than the desired speed of Vmax/2. Thus, a “pre-run” to stabilize the random waypoint model is necessary. This paper applied such method to ensure the accuracy of the simulation.

Random direction model is another widely studied mobility model [11]. Its difference with the random waypoint model is that each node chooses a direction rather than a position as the next target. There are two variations, called random direction with wrap around [15] and random direction with reflection [16], representing two different strategies when a node hits boundary of simulation area. Random direction model is less physically appealing than random waypoint model. However, it exhibits some nice properties, especially useful in theoretical studies; because, at any time of simulation, users are uniformly distributed within the space and distributions of speeds are easily calculated and understood with respect to the model inputs [2].

Theoretical analysis usually does not discuss the movement of nodes, but abstracts wireless networks to link level. For example, Nasipuri and Das assumed that the lifetime of a wireless link between a pair of nodes is a random variable independent from other links, and it is

exponentially distributed with mean λ1 , which means that

the probability distribution function is te λλ − [5]. The

advantage of this assumption is its simplicity, that is, in any given time unit, the same proportion of nodes will depart from the radio range of current node. Tsirigos et

al. [17] proposed an analytic model which is based on the assumption of the lifetime of routes rather than links, but their analysis methods are similar.

To see how a protocol performs in wireless networks with different stability levels, researchers evaluate metrics under several simulations with different node mobility setups. In the paper presenting DSR, which is also the first paper to use the random way point model [1], Johnson simulated several situations with different node pause times, and compared how metrics, such as the packet delivery ratio and routing overhead, change over increasing pause time. Later, many researchers followed Johnson’s method and used pause time rather than speed as a control variable of mobility in their protocol performance comparison researches [9,10].

Johansson et al. used the overall average speed of nodes as the mobility metric in a simulation [18]. They suggested that the pause time metric is ill-defined when node motion is continuous or when nodes use different pause times; however, the speed is more relevant for how often links break down and form. A paper by Perkins et al. also supports using speed as a metric by showing that: although both node speed and pause time can affect the performance of routing protocols, node speed is shown to be a significant factor, while pause time is not [19]. Camp and Boleng showed that the relation between node speed and link breakage is linear with simulation results [20,21]. This observation is confirmed by later researches [22,23]. 3. Link Failure Rate and Speed For simplicity and clarity of our illustration, we will analyze a simplified version of the random waypoint model with the following assumptions: Nodes move in an arbitrarily large area without

obstacles. Pause time is zero—nodes are always moving toward

their destination. All nodes move at the same speed v0.

Such simplifying assumptions help to isolate and emphasize how the motion of nodes affects the lifetime of links. Moreover, simulation evaluation in the latter part of this section shows that our conclusion based on this model remains true for the original random waypoint model or random direction model. 3.1. Analysis of Link Failure Rate In the simplified mobility model, all nodes have same circ le radio range with radius R. Therefore, a bidirectional link is created when the distance of two nodes is less than R, and it is broken when the distance is larger than R. The time interval, T, between the creation and the loss of the link is a random variable called the lifetime of the link. From another aspect, an established

248 X. SHU ET AL.


Figure 1. Illustration of Theorem 2. link may have a probability λt to fail at time t, which can be defined formally as:

Definition 1: Link lifetime T is a random variable. For any time t and a very small time interval ∆t, if the conditional probability P[T < t + ∆t|T ≥ t] = λt∆t + o(∆t), we say that λt is the link failure rate at time t.

By this definition, it is not hard to see that if the link failure rate λ is constant, the probability distribution function of link lifetime is exponential. This implies that the constant link failure rate model and exponentially distributed model are identical in theoretical analysis. Meanwhile, if λ is constant, how long the link already lasts does not affect the future status. This stateless property is very nice for theoretical analysis.

Now, with the formal definition of link failure rate, we can prove following theorem:

Theorem 2: Suppose at time t, the relative speed of two nodes forming a link is random variable V. Then the

failure rate of this link is [ ]R

VE

πλ 2= .

Proof: Let A and B be two mobile nodes. Assume that at time t, the distance of A and B is random variable S, and the angle between the directions of relative speed and the straight line connecting the two nodes is random variable Θ. Since node B is uniformly distributed in the circle radio range of node A as indicated in Figure 1, the probability distribution function of S is:

( )

≤≤=

other

RsRsf

s

s

0

02

2 (1)

And Θ is uniformly distributed on [0,2π), so its probability function is:

( )

≤≤=Θ

otherf

0

202

1 πθπθ (2)

By the law of cosine, after a small interval ∆t, the

distance of the two nodes will be as follows at time t+∆t:

( )( )

( )( )totVS

SStSVtVS

tSVtV

tSVtVSS

∆+Θ∆−=

++Θ∆−∆+

Θ∆−∆=

Θ∆−∆+=

cos

cos2

cos2

cos2

22

2

22'

Let ( )totV ∆+Θ∆−=Φ cos , and let its probability distribution function (PDF) be fФ. As ∆t is sufficiently small and the relative speed is a limited number, there exists a constant C, such that φM = C∆t and |Ф| ≤ φM.

Since S' = S + Ф, by the convolution law, the PDF of S' is:

( ) ( ) ( ) ϕϕϕ dfxfxf SS Φ−∞−

+∞= ∫'

If the lifetime random variable of the link is T, and the link has existed for time t0, we will have:

( ) ( )

'0 0[

S

P T t t T t P S R

f x f d dxR

ϕ ϕ ϕΦ

≤ + ∆ ≥ = >

+∞ +∞= −

−∞∫ ∫ (3)

By Equation 1, ( ) 0≠−ϕxfS when Rx ≤−≤ ϕ0 and

MM ϕϕ ≤Φ≤− ; thus, ( ) ( )ϕϕ Φ− fxfS is not zero if

MRx ϕϕ ≤≤− and MRxR ϕ+≤≤ . Therefore,

Equation 3 can be written as:

( ) ( )

( ) ( )

( ) ( )

( ) ( )

0 0

2

2

2

[

0

2

0

2

0 0

M MS

MS

M

M M

P T t t T t

Rf x f d dx

R x R

Rf x f dxd

R

xRf dxd

R R

f d f dR R

ϕ ϕϕ ϕ ϕ

ϕ ϕϕ ϕ ϕ

ϕϕ ϕϕ ϕ

ϕ ϕϕ ϕϕ ϕ ϕ ϕ

Φ

Φ

Φ

Φ Φ

≤ + ∆ ≥+

= −−

+= −

−+=

= −

∫ ∫

∫ ∫

∫ ∫

∫ ∫

(4)

In above equation, 0 ≤ φ ≤ φM = C∆t, thus:

( ) ( )

( )

22

2 2

2 2 2

2 2

00 0M MM

M

f d f dR R

C to t

R R

ϕ ϕϕϕ ϕ ϕ ϕ ϕ

ϕ

Φ Φ≤ ≤

∆≤ = = ∆

∫ ∫ (5)

With this result, Equation 3 can be simplified as:

( ) ( )

0 0

20M

P T t t T t

f d o tR

ϕ ϕ ϕ ϕΦ

≤ + ∆ ≥

= + ∆∫ (6)



As Ф = – V∆t cosΘ + o(∆t), where V and Θ are two independent random variables, so the conditional mathematical expectation of Ф is:

[ ] ( )

[ ] ( )totVE

totEVE

E

∆+∆=

∆+∆

∈ΘΘ−=

∈ΘΦ

π

ππ

ππ

2

2

3,

2cos

23

,2

Furthermore, by the definition of mathematical expectation,

( )

( )

3,

2 2

3, ,

2 2

3,

2 2

0 3,

2 2

20

M

M

E

P

d

P

fd

P

f d

π π

π πϕϕ ϕ

π π

ϕϕϕ ϕ

π π

ϕϕ ϕ ϕ

Φ

Φ

Φ Θ ∈

Φ = Θ ∈ +∞ = ⋅−∞ Θ ∈

= ⋅ Θ ∈

=

∫

∫

∫

Therefore:

( ) [ ] ( )22

0M f d E V t o t

ϕϕ ϕ ϕ

πΦ = ∆ + ∆∫ (7)

Combining Equation 6 and Equation 7, we get:

[ ] [ ] ( )totVER

tTttTP ∆+∆=≥∆+≤π2

00

Thus, by Definition 1, the failure rate of this link is

[ ]R

VE

πλ 2= .

From Theorem 2 we can see that link failure rate is positively correlated with the mathematical expectation of relative speed. This conclusion coincides with the intuition that if nodes move faster, links will have higher probability to break. Also, Theorem 2 shows that the link failure rate is inversely correlated with the radius of the radio circle. Larger radio range makes link more stable, that is, the probability of a node moving out of the current node’s radio range is smaller.

Until now, we have not used the third assumption of simplified mobility model, that is, all nodes move at a constant speed v0. With this assumption, we can prove a simplified version of Theorem 2:

Corollary 3: Suppose that all nodes move at a constant speed, v0, and г is the random variable of angle between moving directions of two nodes, which are

uniformly distributed on [0, 2π). Then R

v2

08

πλ = .

Proof: By Theorem 2 and the law of cosine,

[ ]

R

v

dR

vE

R

v

vvvvERR

VE

20

00

0020

20

8

2sin

21

0

24

2sin

4

cos222

π

βγπ

πππ

ππλ

=

⋅=

Γ=

Γ⋅⋅−+==

∫

3.2. Simulation By Corollary 3, assuming that all nodes in a simulation

move at the overall average speed v , we can estimate the average link failure rate as:

R

v2

8

πλ =

So, if a simulation has N nodes moving in a D×D rectangular area, the number of link failures during time t is approximately equal to:

( )

( )tv

D

RNN

tD

RNN

R

v

tLinksofNumberesLinkFailur

⋅−=

⋅⋅−⋅=

⋅⋅=

2

2

2

2

142

18

π

ππ

λ

Ten separate simulations were conducted to study the accuracy of the above prediction for each mobility model. Each simulation contains 100 nodes moving in a 1000m×1000m rectangular area at uniformly distributed speeds over [Vm, 2Vm], where Vm is selected from 1m/s to 10m/s.

In the random waypoint model, to avoid the spatial distribution change problem [13] and speed dropping problem [14], every simulation is given a 1000-second pre-run period to warm-up the mobility model to a stable state, and only the data collected from after the 1000-second period is used. Since the speed of nodes is uniformly distributed, the expected average speed of

random waypoint model is ( )2

22

m mm

m m

V Vv v ln

ln V V

−= = .

Figure 2(a) illustrates that the predicted number of link failures is close to the actual number in the random waypoint model simulation. Although the random waypoint model has boundaries, which is different from the simplified model, the density of nodes is higher

250 X. SHU ET AL.


(a) Link failures vs. average speed.

(b) Relative error of link failure rate prediction. Figure 2. Relation of link failure rate with average speed in random waypoint model (speed= [v,2v], pause=0s, area= 1000m * 1000m).


(b) Relative error of link failure rate prediction. Figure 3. Relation of link failure rate with average speed in random direction model with reflection (speed= [v,2v], pause= 0s, area=1000m * 1000m).

around the center, so the nodes near the borders will not significantly impact the result. Figure 2(b) shows that the relative error of our prediction is 5% to 8%, which is a small range and it approximates to a horizontal line. It means that our prediction is increasing at the same ratio with the actual number of link failures, or in other words, the actual number is linear with our prediction and the average speed. Therefore, to get a more accurate result, we can simply multiply an empirical constant to the predicted number.

Different from the random waypoint model, nodes in the random direction model are always distributed uniformly in the simulation area. Figures 3 and 4 show the data collected from the random direction model with reflection [16] and random direction model with wrap around [15] with the same simulation settings as random waypoint model. Similar to the previous result, the linear property holds in these two models as well; and the relative error is small and can be fixed with an empirical constant.

The reason why predictions overestimate the results in the random direction model simulation with reflection while underestimate in the other two lays behind radio range distortion problem caused by the borders of

simulation area. Figure 5 illustrates the shape of the radio range of node A when it is not far away from the border. By the nature of random direction model with reflection, if node B, which is in the radio range of A, is right next to the border of the simulation area, it could not possibly escape from A in the next unit of time. However, this is possible in the random direction model with wrap around.

In general, the ratio of the girth of radio covered area to the size of the area is a decisive factor of the frequency of link breakages, since with the same size of radio area, the longer the girth is, the more nodes could possibly move in and out from the area. It is not hard to see from Figure 5 that the girth/area ratio of the random direction model with reflection is smaller than the ratio of simplified model whose radio range is always a circle. The random direction model with wrap around has larger girth/area ratio than both of them. This characteristic causes the predictions, which are based on simplified model, overestimate the results in one model and underestimate in the other. The random way point model has similar radio range distortion problem with the random direction model with reflection. However, since there are few nodes close to the border in this model [14], the problem will not affect the prediction results as much.




(b) Relative error of link failure rate prediction. Figure 4. Relation of link failure rate with average speed in random waypoint model with wrap around (speed= [v,2v], pause=0s, area= 1000m * 1000m).

Figure 5. Radio range of a node near the border in the random direction model. The gray area is radio range of the node and the dark gray area is where a node in the radio range could possibly escape in the next unit of time.

4. Mobility Metrics Mobility metrics is a measure of how actively nodes move in a simulation. With different mobility metrics setups, researchers can produce different scenarios to evaluate the performance of their wireless network protocols. One of the most popular mobility metrics is pause time. In the paper presenting DSR [1], Johnson simulated several situations with different node pause times, and compared how performance metrics, such as

packet delivery ratio and routing overhead, change over the increasing pause time. Johansson used average speed of nodes as mobility metric in his simulation study [18].

Link failure rate, which reflects the wireless network topology change rate, can be used for mobility metrics as well. Usually, we cannot set link failure rate directly in wireless network simulations; however, as we presented in Section 3, link failure rate is positively correlated with the average speed of nodes. Therefore, we can create any level of link failure rate as we want with proper setup of the speed of nodes and the radio range. The remainder of this section presents an analysis of the mathematical relationship between the link failure rate and the other two mobility metrics.

Suppose in a random waypoint model simulation, a node moves at constant speed v and stays at each destination for a constant pause time, p. Let δ be the average distance between two waypoints. If the node passed through n destination points, and n is large enough, which implies that this simulation has been run long enough, it can be estimated that the overall average speed of the nodes as follows:

vp

v

npv

nn

v+

=+

=δ

δδ

δ

This formula shows that the pause time p is not linear with the average speed, and obviously, it is not linear with the link failure rate as well, since average speed is linear with link failure rate. The derivative of the average speed is:

( )22

vp

v

dp

vd

+−=

δδ

This implies that the change of p has a larger impact on the expectation of average speed and the link failure rate when pause time, p, is comparatively small. However, when p is large, its impact is not that distinct. If one uses pause time as the X-axis of the graph showing how other network performance metrics change, the amplitude of these metrics will become smaller as pause time increases. This phenomenon, which has been observed in previous research [19], suggests that graph analysis with pause time as mobility metrics may be inaccurate and not intuitive.

Johansson [18] defined another mobility metric as follows:

( )∑ ∑= +=−

=n

x

n

xyxyM

nnM

1 11

2

where n is the number of nodes, and Mxy is defined as the average relative speed between nodes x, y during the simulation. This mobility metric has a nice property—it is linear with the link failure rate.

Theorem 4: Let M be the mobility metric defined in

252 X. SHU ET AL.


[18] and λ be the average link failure rate, then

[ ]2λπR

ME = .

Proof: By Theorem 2, the failure rate of the link

between x and y is ( ) ( )[ ]R

tyxVEtyx

πλ ,,2

,, = at time t, so

( )[ ] ( )tyxλRπ

tyxVE ,,2

,, = .

By the definition of M,

( ) ( ) ( )1

01 11 0

2, ,

1

n n t

tx y x

M V x y t dtt t n n = = +

=− ⋅ − ∑ ∑ ∫

Therefore, its mathematical expectation is:

[ ]

( ) ( ) ( )[ ]

( ) ( ) ( )

λRπ

dttyxλnntt

Rπ

dttyxVEnntt

ME

n

x

n

xy

t

t

n

x

n

xy

t

t

⋅=

⋅−⋅−

⋅=

⋅−⋅−

=

∑ ∑ ∫

∑ ∑ ∫

= +=

= +=

2

,,1

2

2

,,1

2

1 101

1 101

1

0

1

0

Since M is linear with link failure rate, it is more

reasonable to use it as a mobility metric than pause time. However, Theorem 4 also shows a problem of this mobility metric. Consider two similar simulations, whereby nodes move at exactly the same speed and path. Suppose the first simulation has a larger radius of radio range. Although M shows no difference of the mobility levels of these two simulations, it is obvious that links in the second simulation are not as stable as in the first one since their radio range is smaller, and these two simulations will produce different analysis results. Therefore, the mobility metric should consider the radius R of radio range as well. By Theorem 4, if the mobility

metric is defined as R

M , the simulation analysis could

avoid said situation. Or, the average link failure rate λ can be used as the mobility metric directly, since it has the same linear property as M and will not affect by the radius problem. 5. Conclusions and Future Work By analyzing a simplified mobility model, which is similar to random waypoint model but has no boundary and no pause time, we find that the link failure rate is linear with the speed of nodes in simulation analysis. That is, if node increases speed, the number of links failing during a unit of time increases accordingly. This property makes average speed a better mobility metric

than pause time, since it reflects the topology change rate better. Although this analysis is based on simplifying assumptions, simulation analysis suggests that the result can also be applied on the random waypoint and random direction models.

For future research, we believe that the study of link lifetime will be a great help for optimizing routing protocols, because a node can choose the appropriate link that has highest probability of living if it knows how link lifetime distributed. Bettstetter et al. have done a valuable job of analyzing the link lifetime of the random waypoint model [24]. We expect further study of link lifetime in real scenario based mobility model.

Based on this paper, we also conjecture that if two mobility models create the same probability distribution of link lifetime, a non-geographic-based wireless network algorithm should produce similar results on both mobility models. This implies that the link failure rate has great weight on the topological change of a wireless network.

6. References [1] D. B. Johnson and D. A. Maltz, “Dynamic source routing

in ad hoc wireless networks,” Mobile Computing, Chapter 5, pp. 153–181, 1996.

[2] P. Nain, D. Towsley, B. Liu, and Z. Liu, “Properties of random direction models,” INRIA technical report RR-5284, July 2004.

[3] J. Y. Le Boudec and M. Vojnovic, “Perfect simulation and stationarity of a class of mobility models,” IEEE Infocom 2005, Miami, FL, 2005.

[4] A. K. Saha and D. B. Johnson, “Modeling mobility of vehicular ad hoc networks,” ACM VANET 2004, October 2004.

[5] A. Nasipuri and S. R. Das, “On-demand multipath routing for mobile ad hoc networks,” Proceedings of the 9th International Conference on Computer Communi- cations and Networks (IC3N), Boston, October 1999.

[6] D. B. Johnson, D. A. Maltz, and J. Broch, “DSR: The dynamic source routing protocol for multi-hop wireless ad hoc networks,” Ad Hoc Networking, edited by Charles E. Perkins, Addison-Wesley, Chapter 5, pp. 139–172, 2001.

[7] The Network Simulator - ns-2, Online: http:// www.isi.edu/ nsnam/ns/, November 2006.

[8] Global Mobile Information System Simulation Library – GloMoSim, ,November 2006, Online: http://pcl.cs.ucla.edu/projects/glomosim/.

[9] J. Broch, D. A. Maltz, D. B. Johnson, Y. C. Hu, and J. Jetcheva, “A performance comparison of multi-hop wireless ad hoc network routing protocols,” Mobile Computing and Networking (MobiCom), pp. 85–97, 1998.

[10] C. E. Perkins, E. M. Royer, S. R. Das, and M. K. Marina, “Performance comparison of two on-demand routing protocols for ad hoc Networks,” IEEE personal Communications, Vol. 9, No. 1, pp. 16–28, February 2001.



[11] C. Bettstetter, “Mobility modeling in wireless networks: Categorization, smooth movement, and border effects,” ACM Mobile Computing and Communications Review, Vol. 5, No. 3, 2001.

[12] E. M. Royer, P. M. Melliar-Smith, and L. E. Moser, “An analysis of the optimum node density for ad hoc mobile networks,” Proceedings of IEEE International Conference on Communications (ICC), June 2001.

[13] C. Bettstetter, G. Resta, and P. Santi, “The node distribution of the random waypoint mobility model for wireless ad hoc networks,” IEEE Transactions on Mobile Computing, Vol. 02, No. 3, pp. 257–269, July–September 2003.

[14] J. Yoon, M. Liu, and B. Noble, “Random waypoint considered harmful,” IEEE Infocom 2003, San Francisco, CA.

[15] Z. J. Haas, “The routing algorithm for the reconfigurable wireless networks,” Proceedings of ICUPC’97, San Diego, CA, October 1997.

[16] N. Bansal and Z. Liu, “Capacity, delay, and mobility in wireless ad-hoc networks,” Proceedings of INFOCOM 2003, San Francisco, CA, March 2003.

[17] A. Tsirigos, Z. J. Haas, and S. Tabrizi, “Multipath routing in mobile ad hoc networks or how to route in the presence of topological changes,” Proceedings of the MILCOM 2001 IEEE Military Communications Conference, October 2001.

[18] P. Johansson, T. Larsson, N. Hedman, B. Mielczarek, and M. Degermark, “Scenario-based performance analysis of routing protocols for mobile ad hoc networks,” Annual International Conference on Mobile Computing and

Networking (MOBICOM), pp. 195–206, August 1999.

[19] D. D. Perkins, H. D. Hughes, and C. B. Owen, “Factors affecting the performance of ad hoc networks,” Proceedings of the IEEE International Conference on Communications (ICC), 2002.

[20] J. Boleng, “Normalizing mobility characteristics and enabling adaptive protocols for ad hoc networks,” LANMAN 2001: 11th, IEEE Workshop on Local and Metropolitan Area Networks, pp. 9–12, March 2001.

[21] T. Camp, J. Boleng, and V. Davies, “A survey of mobility models for ad hoc network research,” Wireless Communications F4 Mobile Computing (WCMC): Special Issue on Mobile Ad Hoc Networking: Research, Trends and Applications, 2002.

[22] L. Qin and T. Kunz, “Mobility metrics to enable adaptive routing in MANET,” Proceedings of the 2nd IEEE International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob 2006), Montreal, Canada, pp. 1–8, June 2006.

[23] C. Yawut, B. Paillassa, and R. Dhaou, “Mobility metrics evaluation for self-adaptive protocols,” Journal of Networks, Academy Publisher, Finland, Vol. 3, No. 1, ISSN : 1796–2056, pp. 5364, January 2008.

[24] C. Bettstetter, H. Hartenstein, and X. Pérez-Costa, “Stochastic properties of the random waypoint mobility model,” ACM/Kluwer Wireless Networks: Special Issue on Modeling and Analysis of Mobile Networks, 10(5), September 2004.



Stereo Video Transmission Using LDPC Code

Rui GUO, Lixin WANG, Xiaoxia JIANG Communication College, Hangzhou Dianzi University, Hangzhou, China

Email: [email protected] Received on April 16, 2008; revised and accepted on August 18, 2008

Abstract Stereo video is widely used because it can provide depth information. However, it is difficult to store and transmit stereo video due to the huge data amount. So, high efficient channel encoding algorithm and proper transmission strategy is needed to deal with the video transmission over limited bandwidth channel. In this paper, unequal error protection (UEP) based on low density parity check (LDPC) code was used to transmit stereo video over wireless channel with limited bandwidth. Different correction level LDPC code was used according to the importance of video stream to reconstruction at the receiver. Simulation result shows that the proposed transmission scheme increases the PSNR of reconstructed image, and improves the subjective effect.

Keywords: Stereo Video, LDPC Code, UEP, Data Partion, EEP 1. Introduction Stereo video is widely used because it can provide depth information in stereoscopic television, video conference, remote control, telemedicine and other fields [1,2]. However, it is difficult to store and transmit stereo video due to the huge data amount. So, high efficient channel encoding algorithm and proper transmission strategy is needed to deal with the video transmission. In practical channel, especially in wireless channel with limited bandwidth, signal errors will inevitably appear at the receiver as a consequence of channel fading, multipath, noise and so on. Stereo video is highly compressed data stream, and is very sensitive to error, which will decrease the quality of reconstructed image at the receiver. Therefore, high efficient error correction technology and transmission control strategy is needed [3]. Recently, with the deep research in LDPC, more and more people focus on the video communication based on LDPC code [4,5].

In this paper, we mainly focus on the stereo video communication on AWGN channel with limited bandwidth using UEP scheme based on LDPC encoding. According to the different contribution to image reconstruction at the receiver, the video stream is divided into different parts [6], then UEP is used to protect the different bits stream in different level. The structure is shown in Figure 1.

2. Bits Stream Structure and Data Partition of Stereo Video

Conceptually, the structure of H.264 encoder is divided into two layers: Video Coding Layer (VCL) and Network Abstraction Layer (NAL). VCL provides high performance function in video compression, including common definitions of video compression, block, macroblock, sub-graph layer and so on. NAL is responsible for network abstraction, which provides different adaptive capacity for different networks and transmitting package with proper mode. NAL works in two kinds of mode: Single Slice mode and Data Partition mode. When using Data Partition mode, H.264 puts all variable length codes with the same data type together in each frame [6].

Head information includes head information, macroblock type, frame type, predicted residual of motion vectors, frame flag etc. In H.264, this part is called A segmentation, which is the most important part. Intra-frame segmentation is called B segmentation. It loads the coding mode and the correlation coefficient in intra frame blocks. B segmentation works under the effect of A segmentation. Compared with information of inter-frame information segmentation, intra-frame information can prevent further drift, thus it is more effective than inter-frame segmentation. Inter-frame segmentation is called C segmentation. It only includes the coding mode and the correlation coefficient in

STEREO VIDEO TRANSMISSION USING LDPC CODE 255


Figure 1. The diagram of stereo video communication based on UEP.

inter-frame blocks, it is the biggest segmentation in the video stream. Inter-frame segmentation is relatively subordinate, because it won’t provide synchronous information in encoding or decoding.

Here, we use stereo video compression encoding based on H.264, the output bit stream has the same structure as that of H.264 encoder. Compared with the bit stream of H.264 encoder, the stereo video adds the disparity information of different video channel, such as disparity vector, disparity prediction model and predictive residual etc. Disparity vector and motion vector have the same important level. Once disparity vector goes wrong, the data of assisted video will be affected. So in this paper, we regard the slice head information, types of encoding frame, types of disparity prediction macroblock and disparity vector among different video channel as the same importance as the motion vector in one channel. Here, we divide the bit stream of stereo video into two parts: A segmentation and B segmentation .

A segmentation includes head information, MB-Type, Reference frame, motion vector prediction difference, disparity vector prediction difference, frame ending flag, the encoding mode and the correlation coefficient in intra frame, etc.

B segmentation includes the encoding mode and correlation coefficient in inter-frame.

We realize stereo video encoder based on H.264/AVC encoder platform. The parameters are listed in Table 1. The outputs of the encoder are two binary data

Table 1. The parameter of video encoder.

Temporal direct conference frames 5

Prediction mode Modes of all blocks size

Entropy coding method CABAC

Motion estimation scope of the search

±32

Disparity estimation scope of the search

Level±64, Vertical: ±16

Structure of Group of picture IBBP…, N=15, M=3

file—test.264 and coding pursuit file—trace.txt [7,8]. We divide the bit stream according to these files.

The divided stereo video bit stream will be stored separately in file A and B. We adopt high-level protection to A bit stream because of its higher importance. For the B bit stream, we use low-level protection for the sake of encoding efficiency. At the receiver, if A segmentation data is lost, B will be abandoned. If part of B segmentation is lost, the head information still can be used to improve the effect of error concealment [9,10]. 3. Realization UEP Based on Different

Rates Irregular LDPC Codes 3.1. The Principles of Realizing UEP Based on

Different Rates Irregular LDPC Codes The key point of unequal error protection is: under the condition of limited bandwidth and the premise of prior protection to the important parts. We properly allocate the redundancy to the source and channel so as to reduce end-to-end distortion [11].

Assuming the total bit rate (the total bandwidth) is

totalR , the parts of source and channel isSR and CR

respectively. Suppose, the source is divided into A segmentation and B segmentation, so:

a a b btotal s c s cR R R R R= + + + (1)

In formula (1): asR , b

sR represent the bits of A and B segmentation used in source encoding respectively. a

cR , bcR represent the bits of A and B bit stream used in the

channel encoding respectively. Assuming the bit rate of channel encoding corresponding to A segmentation and B segmentation are ar , br . Then, we can get the following formula:

as

a a as c

Rr

R R=

+,

bs

b b bs c

Rr

R R=

+ (2)

So, the total transform is:

256 G. RUI ET AL.


( ) ( , ) ( , )a b a bS S S C C CD R D R R D R R= + (3)

Thus the question converts to: with the limited bandwidth and supposed data priority, the total distortion of the video at the receiver is minimized:

min( ( )) min( ( , ) ( , ))a b a bs s s s c cD R D R R D R R= + (4)

Subject to the following constraint conditions:

total T

a b

R R

r r

= ≤

(5)

where, TR represents the total channel bandwidth,

and ( )D R represents the total distortion of source and channel, which is calculated by MSE. The relationship between PSNR and MSE is given by:

225510lgPSNR

MSE=

By adjusting the parameters of channel encoding and bit rate allocating between source encoding and channel reasonably, we can get the optimal reconstructed image at receiver.

UEP needs to allocate bit-rate between source and channel, under the condition that the total bandwidth is decided. According to adjusting the parameters of source encoding or the bit rate of channel encoding , the situation will be different.

3.2. The Performance of UEP When the Source

Rate is Decided When the bit rate of source encoding is pre-assigned, UEP can be realized by just adjusting the bit rate of channel .In this paper, we compared the properties of UEP scheme and equal error protection (EEP) scheme (a br r< or a br r= ) in the same bandwidth. When the total bandwidth is decided, and the bit rate of source is pre-assigned (QP is unchangeable), that is, in the formula (1), a b

s sR R+ and

totalR are determined. From a a b btotal s c s cR R R R R= + + + , we can

know that the bits allocated to channel is also determined. UEP can be realized by adjusting the bit rate of channel encoding with different grade importance, that is by choosing ar , br to determine a

cR and bcR , so as to add

different redundancy to information with different importance. After analyzing the segmentations of stereo video bit

stream, we found that the ratio of A segmentation and B segmentation is about 1:3 in Race 1 sequence. In order to keep the total bandwidth unchanged, the parameters of error correcting code are shown in the Table 2.

We use irregular LDPC code based on IeIRA permutation matrix with three different encoding rate in the experiment. They are 1/4, 1/2 and 3/4, respectively. The code length of LDPC is 4064 bits, using BP decoding, the maximum iteration number is 80.The length of Turbo code is 3568 bits. When using UEP scheme, the LDPC

code with 1/4 rate is used to protect the important data (the bit stream from A group), the LDPC code with 3/4 rate is used to protect the less important data(the bit stream from B group). If A segmentation is lost, B segmentation will be abandoned at the receiver.

In the experiment, we test the properties of UEP, EEP based on LDPC code and Turbo code in the AWGN channel with limited bandwidth by using BPSK modulation. Test sequences–Race1 sequences from Japanese KDDI lab are used. The image size is 320×240.

At first, we give the BER performance of 3 kinds of code in AWGN channel in Figure 2. Clearly, not only the properties of irregular LDPC code based on IeIRA permutation matrix are better than Turbo code with the same length, but also error floors is lower than Turbo code. Figure 3 gives the differences of PSNR properties of image reconstructed by different protection scheme in the case of fixed source rate. (QP is 30 in stereo video compression encoding). Figure 4 shows the reconstructed 7th frame main and assistant video by different encoding strategy at the receiver when SNR is 2.1dB.

Figure 3 and Figure 4 show that the UEP scheme based on LDPC code is always better than the other two schemes based on EEP. This is because we can reduce the error probability of important information by adding more redundancy in the important part of video stream. With the improvement of channel condition(increase of signal-to-noise ratio), this advantage is decreased (Eb/No from 1.8 to 2.4, the advantage of main video stream based on LDPC UEP relative to LDPC EEP is reduced from 2.1 dB to 0.8 dB). What’s more, because of the excellent performance of LDPC code, the performance of EEP scheme based on LDPC code is better than that based on Turbo code.

Table 2. The parameters of various channel coding.

Channel Encoding Rate Encoding

Scheme Segmentation A Segmentation B LDPC UEP 1/4 3/4 LDPC EEP 1/2 Turbo EEP 1/2

0.5 1 1.5 210

-9

10-8

10-7

10-6

10-5

10-4

10-3

10-2

10-1

Eb/No(dB)

Bit

Err

or R

ate(

BE

R)

LDPC,r=1/4

LDPC,r=1/2

LDPC,r=3/4

Turbo,r=1/2

Figure 2. Performance of different channel code.



1 1.2 1.4 1.6 1.8 2 2.2 2.415

20

25

30

35

40

Eb/No(dB)

PS

NR

(dB

)

LDPC UEP

LDPC EEP

Turbo EEP

1 1.2 1.4 1.6 1.8 2 2.2 2.410

15

20

25

30

35

40

Eb/No(dB)

PS

NR

(dB

)

LDPC UEP

LDPC EEP

Turbo EEP

Figure 3. PSNR Compare of main, assistant video stream in different transmission scheme.

Figure 4 shows the reconstructed 7th frame at the receiver. It is clear that Figure 4(a) is better than Figure 4(b) and Figure 4(c), and Figure 4(b) is a little bit better

than Figure 4(c). On condition that a br r< , it ensure the

important data of A segmentation is fully protected, so we can get better subjective effect by using UEP scheme.

3.3. The Performance of UEP in Case of Fixed

Channel Encoding Rate When the total bandwidth is given, we also can pre-assigned the channel encoding rate, change the bit rate of the source (change QP) to adjust redundancy allocation between source and channel to get joint optimization .We will give the performance of different encoding strategies with fixed channel rate. The rate of LDPC code used in each encoding strategy is shown in Table 3.

The total bandwidth is fixed (for example 1.5Mbps). When we use above mentioned three encoding strategies, the QP is 31, 26, and 28 in turn. The PSNR of reconstructed image by using different encoding strategies at the receiver are given as follows.

(a) LDPC UEP

(b) LDPC EEP

258 G. RUI ET AL.


(c) Turbo EEP

Figure 4. The reconstructed main and assistant image of different transmission scheme at receiver.

Table 3. The channel coding RATE.

Channel Encoding Rate Encoding Scheme A Segmrntation B Segmentation

EEP1 1/4 1/4 EEP2 1/2 1/2 UEP 1/4 1/2

Figure 5 shows that EEP1 scheme can obtain the

optimal performance for both main video and assistant video in low SNR condition. As the channel condition is going better, the performance of UEP gets over that of EEP1 at 1.8 dB, EEP2 is exceeding the other two schemes when SNR is 2.1dB. The reason lies in that, in low SNR, channel error is the main factor affecting reconstruction quality in the receiver, so EEP1 scheme can get the best performance by using lower rate channel encoding strategy. When the channel is good, the bit error rate will reduce, and channel error will not be the main factor affecting PSNR of reconstructed image. Compared to EEP2, EEP1 and UEP have over protection, so they are worse than EEP2 in high SNR condition. 4. Conclusions In this paper, we study the characteristics of stereo video bit stream based on H.264. and stereo video communication

1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 315

20

25

30

35

40

Eb/No(dB)

PS

NR

(dB

)

UEP

EEP2

EEP1

1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 310

15

20

25

30

35

40

Eb/No(dB)

PS

NR

(dB

)

UEP

EEP2

EEP1

Figure 5. PSNR compare of main, assistant video stream in different transmission scheme (channel encoding is fixed, QP is changeable). based on UEP scheme of LDPC code. According to the different importance to video reconstruction, we divide the stereo video into A and B segmentation. Then, according to the importance, UEP is used to get the best reconstructed image at the receiver.

According to source rate is fixed or channel rate is fixed when realizing UEP, we put forward two implementation algorithm. After analyzed the properties in the AWGN channel with fixed bandwidth, we obtain the following conclusions:

1) When the source rate is fixed (QP is fixed), the performance of UEP scheme is always better than that of EEP. UEP can correct error with different abilities for different data, which gives more protection on more important data. Especially in the case of poor channel condition, the UEP scheme has advantage obviously.

2) When the channel code rate is fixed, we need to change the bit rate of source encoding to adapt to the requirement of bandwidth. After simulation, we find the scheme using low channel code rate have better performance than the scheme using high channel code



rate in low SNR condition. As the channel condition is improving, channel error will not be the main factor affecting bit error, the scheme using low channel code rate is worse than that using high channel code rate on the contrary. 5. References [1] J. Konrad, “Visual communication of tomorrow: natural,”

IEEE Communication Magazine, Vol. 39, No. 1, pp. 126–133, 2001.

[2] H. Kalva, L. Christodoulou, L. Mayron, O. Marques, and B. Furht, “Challenges and opportunities in video coding for 3D TV,” ICME 2006, International conference on multimedia & Expo.

[3] G. Gagnon, S. Subramaniam, and A. Vincent, “3-D MPEG-2 video transmission over broadband network and broadcast channels,” Stereoscopic Displays and Virtual Reality Systems VIII, Proceedings SPIE, Vol. 4297, pp. 290–298, 2001.

[4] L. Wang, “Application of LDPC for digital TV terrestrial broadcasting system,” Nanjing, China, A Dissertation Submitted to Southeast University for the Academic Degree of Master, March 2006.

[5] Q. Liu, Y. Lu, W. S. Wang, H. J. Cui, K. Tang, “Robust video transmission scheme using dynamic rate selection LDPC and RS codes,” IMACS Multiconference on

Computational Engineering in Systems Applications (CESA), Beijing, China, October 4–6, 2006.

[6] T. Stockhammer, M. Hannuksela, and T. Wiegand, “H. 264/AVC in wireless environments,” IEEE Transactions on Circuits and Systems for Video Technology, Vol. 13, pp. 657–673, July 2003.

[7] ITU-T Rec. H. 264/AVC /ISO/IEC 11496-10, “Advanced video coding,” Final Committee Draft, Document JVT-G050, 2003.

[8] J. Ribas-Corbera, P. A. Chou, and S. Regunathan, “A generalized hypothetical reference decoder for H.264/ AVC,” IEEE Transactions on Circuits and Systems for Video Technology, Vol. 13, pp. 674–687, July 2003.

[9] M. Yin and H. Y. Wang, “A scheme for H.264-based data partition and unequal error protection,” Journal of Huazhong University of Science and Technology (Nature Science Edition), Vol. 34, July 2006.

[10] T. Wiegand, H. Schwarz, A. Joch, F. Kossentini, and G. J. Sullivan, “Rate-constrained coder control and comparison of video coding standards,” IEEE Transactions on Circuits and Systems for Video Technology, Vol. 13, No. 7, pp. 688–703, July 2003.

[11] P. Y. Yip, J. A. Malcolm, W. A. C. Fernando, K. K. Loo, and H. K. Arachchi, “Joint source and channel coding for H.264 compliant stereoscopic video transmission,” in Canadian Conference on Electrical and Computer Engineering, Saskatoon, Canada, May 2005.



An Agent-Based Multimedia Intelligent Platform for Collaborative Design

Quan LIU, Xingran CUI, Xiuyin HU

School of Information Engineering, Wuhan University of Technology, Wuhan, P.R. China

E-mail: quanliu, xingran, huxiuyin @whut.edu.cn Received on April 14, 2008; revised and accepted on August 26, 2008

Abstract Collaborative design can create added value in the design and production process by bringing the benefit of team work and cooperation in a concurrent and coordinated manner. However, distributed design knowledge and product data make the design process cumbersome. To facilitate collaborative design, an agent-based intelligent CAD platform is implemented. Intelligent agents are applied to the collaborative design. Adopting the JADE platform as framework, an intelligent collaborative design software (Co-Cad platform for short) is designed. In this platform, every man, design software, management software, equipment and resource is regarded as a single agent, the legacy design can be abstracted to be interaction between agents. Multimedia technology is integrated into Co-Cad platform, communication and identity authentication among collaborative designers from different areas are more convenient. Finally, an instance of collaborative design using Co-Cad platform is presented. Keywords: Agent, Co-Cad Platform, Collaborative Design, JADE, Multimedia Technology

1. Introduction Product design is becoming a collaborative task among designers or design teams that are physically, geographically, and temporally distributed. Plenty of product modeling tools and engineering knowledge from various disciplines spread around different design phases, making effective capture, retrieval, reuse, sharing and exchange of these heterogeneous design knowledge a critical issue [1]. An ideal product design environment which is both collaborative and intelligent must enable designers and manufacturers to respond quickly to commercial market pressures [2].

Compared with current standalone CAD, the collaborative CAD is “not generally accepted” because of both technical and non-technical problems [3]. As an emergent approach to developing distributed systems, agent technology has been employed to develop collaborative design systems and to handle the problems and overcome the limitations [4]. The way in which intelligent software agents residing in a multi-agent system interact and cooperate with one another to

achieve a common goal is similar to the way that human designers collaborate with each other to carry out a product design project. Thus, we believe that a collaborative design environment implemented by taking an agent-based approach will be capable of assisting human designers or design teams effectively and efficiently in product design [5].

In order to make collaborative design more convenient, multimedia information is absolutely necessary. Integrating multimedia technology into collaborative design platform can express information more quickly and naturally than letters.

This paper presents an ongoing project on the application of intelligent agents to collaborative design. In the project, an intelligent CAD platform (Co-Cad platform) is implemented, adopting multi-agent technology, collaborative design technology and multimedia technology. The platform uses the proposed JADE as middle-ware and integrates multimedia system. Based on the platform, intelligent agents can interact with each other via network. Thus, the ideal digital collaborative design is implemented.

AN AGENT-BASED MULTIMEDIA INTELLIGENT PLATFORM FOR 261 COLLABORATIVE DESIGN


The rest of the paper is organized as follows: Section 2 gives an overview of the work related to our research. The basic principle and key method are described in Section 3. Section 4 introduces the implementation of Co-Cad platform. A case study is presented in Section 5. Finally, a number of concluding remarks are made in Section 6.

2. Related Works Web-and agent-based approaches have been dominant during the past decade for the implementation of collaborative product environments. This section provides a brief updated review of the applications of agents and web-based technologies to collaborative design engineering.

An earlier review of multi-agent collaborative design systems can be found in reference [6]. Shen et al. [7] provide a detailed discussion on issues in developing agent-oriented collaborative design systems and a review of significant, related projects or systems. The interesting aspects of PACT include its federation architecture using facilitators and wrappers for legacy system integration. SHARE [8] was concerned with developing open, heterogeneous, network-oriented environments for concurrent engineering, particularly for design information and data capturingand sharing through asynchronous communication. SiFAs [9] was intended to address the issues of patterns of interaction, communication, and conflict resolution using single function agents. DIDE [10] was a typical autonomous multi-agent system and was developed to study system openness, legacy systems integration, and geographically distributed collaboration.

Co-Designer [11] was a system that can support localized design agents in the generation and management of conceptual design variants. A-Design [12] presented a new design generation methodology, which combines aspects of multiobjective optimization, multi-agent systems, and automated design synthesis. It provides designers with a new search strategy for the conceptual stages of product design, which incorporates agent collaboration with an adaptive selection of designs. 3. The Basic Principle and Key Method 3.1. Basic Principle Usually, each agent is regarded to be a physical or abstract entity. Distributed in the network environment, each agent is independent and can act on itself and the environment, manipulated part of the environment reflection and react to the changes in the environment. More importantly, through communication and cooperation with other agents, it can perform mutual

work to complete the entire task. Man, design software, management software, as well as equipment and resources can be viewed as agents. Legacy design activities can be abstracted to be informational communication between agents, which includes not only communication between homogeneous agents but also the communication between heterogeneous agents through technology-aided design software. We can further abstract them to informational exchange. Theoretically, all of the information can be digital. In other words, if we provide a suitable platform for interaction, digital collaborative design can be realized.

This paper seeks to build a multi-agent middleware, through which, man, software, manufacturing equipment within this collaborative organization can carry out informational communication. Different types of agents can communicate with their agent middleware through their respective forms of communication and achieve information interaction with agents in other organizations. Thus, it enables man to relieve from various types of software and achieve collaborative design efficiently and effectively.

In light of the above principle, this paper seeks to construct a collaborative design software platform for agents—Co-Cad platform, and integrates multimedia technology into this platform. Co-Cad platform, using pure Java language, is independent and flexible. In the network it realizes the collaboration between designers—“You see what I see”. Each designer’s operation will be reflected in others’ platform. Designers in different places can exchange their ideas on the interactive design in the form of video chat. Video chat can be the most direct way to confirm the identity of the others, which guarantees the safety of the collaborative design, it is also the fastest and most natural form of information expression. 3.2. The Multi-agent System Based on JADE JADE is a middleware that facilitates the development of multi-agent systems in compliance with the FIPA (Foundation for Intelligent Physical Agents) specifications. Each running instance of the JADE runtime environment is called a container. The set of active containers is called a platform. A single special main container must always be active in a platform and all other containers register with it as soon as they start. Figure 1 illustrates the above concepts through a sample scenario showing two JADE platforms composed of 3 and 1 container respectively. JADE agents are identified by a unique name and, provided they know each other’s name, they can communicate regardless of their actual location. A main container differs from normal containers as it holds two special agents: AMS and DF. The AMS (Agent Management System) provides the naming service and represents the authority in the platform. The

262 Q. LIU ET AL.


Figure 1. Containers and Platforms.

Figure 2. Communication model.

DF (Directory Facilitator) provides a yellow pages service by means of which an agent providing the services he requires in order to achieve his goals.

One of the most important features that JADE provides for agents is the ability to communicate. The communication paradigm adopted is the asynchronous message passing. Each agent has a sort of mailbox (the agent message queue) where the JADE runtime posts messages sent by other agents. Whenever a message is posted in the message queue the receiving agent is notified. If and when the agent actually picks up the message from the message queue to process it is completely up to the programmer. 4. Implementation of Co-Cad Platform 4.1. Communication Model Under the currently popular mode of communication services, if A is going to discuss about certain parts of the design with B, A firstly produces his own design using computer-aided design software, such as Auto CAD, and then uploads it to the FTP server, while B using the same design software uploads his own design to the FTP server. Then they download each other’s program. Through traditional telephone or E-Mail and even more network communication tools, they exchange their opinions and finally reach a consensus. When the design is completed, it is uploaded onto the FTP server, and then the WWW server will issue the note that the components design is completed for other people to use. Under this model, the exchanging and sharing of productive data don’t fit the requirements of network and it is at a low level of intelligence. It cannot meet the needs of increasingly complex product design.

In order to overcome the above shortcomings, the

platform’s interactive process design is based on Multi-Agent communication model (Figure 2).

First, A starts his agent middleware and through agent middleware he can know that currently B is also online. Of course, they may have already reached the agreement that software communication through certain middleware. A starts his software, such as Auto CAD, and notifies B. Both of them use the voice and video program and Auto CAD to make the real-time interactive design through agent middleware. After that, the agent middleware will submit the design to the FTP server automatically for others to use, and automatically issues the news that components of the design is finished in the WWW server. In this paper, the middleware is to provide a platform for interaction, but the middleware itself is an agent, they must first interact with other middleware to complete their task. The agent communication protocol adopted in Co-Cad is within JADE platform. The own protocol of JADE facilitates the development of multi-agent systems in compliance with the FIPA specifications.

4.2. Design and Integration of Multimedia

System

Web-based video conferencing system has been widely used in remote collaborative design system, but these systems are using ready video conferencing system to bring about long-distance transmission. Video conferencing systems are separated from remote collaborative design system, which is inconvenient to integrate with the collaborative design system seamlessly. Due to some restrictions of video conferencing system, the remote collaborative design system can not stick together very well.

As Co-Cad platform is completed developed in Java language, in order to make the seamless integration with



Figure 3. The video properties.

Figure 4. The audio properties.

Figure 5. PlugIn viewer of video.

Figure 6. PlugIn viewer of audio.

multi-media audio and video system possible, this paper applies multimedia technology to achieve video chat function and adopts the JMF (Java Media Framework) as development environment. To handle multimedia files and equipment in Java language, it’s a must to install JMF installation package.

4.2.1. Media Capture Time-based media can be captured from a live source for processing and playback. For example, audio can be

captured from a microphone or a video capture card can be used to obtain video from a camera. Capturing can be thought as the input phase of the standard media processing model.

A capture device might deliver multiple media streams. For example, a video camera might deliver both audio and video. These streams might be captured and manipulated separately or combined into a single, multiplexed stream that contains both an audio track and a video track. 4.2.2. Media Processing In most instances, the data in a media stream are manipulated before they are presented to the user. Generally, a series of processing operations occur before presentation: 1) If the stream is multiplexed, the individual tracks are

extracted. 2) If the individual tracks are compressed, they are

decoded. 3) If necessary, the tracks are converted to a different

format. 4) Effect filters are applied to the decoded tracks.

The tracks are then delivered to the appropriate output device. If the media stream is to be stored instead of rendered to an output device, the processing stages might differ slightly. To capture audio and video from a video camera, process the data, and save it to a file: 5) The audio and video tracks would be captured. 6) Effect filters would be applied to the raw tracks. 7) The individual tracks would be encoded. 8) The compressed tracks would be multiplexed into a

single media stream. 9) The multiplexed media stream would be saved to a

file. It includes a viewer that displays a graphical overview

of a processor’s tracks and plug-ins. This graph enables you to monitor the media flow during playback, capturing, or transcoding. Figure 3, Figure 4, Figure 5 and Figure 6 display the media properties and plug-ins currently. 5. Case Study The only software requirement to execute the platform is the Java Run Time Environment version 1.4. All the software is distributed under the LGPL license limitations and it can be downloaded from the JADE web site http://jade.tilab.com/. Having uncompressed the archive file, a directory tree is generated whose root is jade and with a lib subdirectory. This subdirectory contains some JAR files that have to be added to the classpath environment variable.

The process of collaborative design using Co-Cad platform to carry out a product design project is performed

264 Q. LIU ET AL.


Figure 7. Single designer’s interface.

Figure 8. Main interface of the agent manager.

Figure 9. Interface of Selene’s Co-Cad platform.

Figure 10. Interface of Jujumao’s Co-Cad platform.

as follows: Step 1: The host computer used as server launches

JADE, and then Co-Cad platform, starts the test container in JADE. Other collaborative designers only need to run Co-Cad.

Step 2: While JADE platform is running, all collaborative designers can start Co-Cad platform and see Co-Cad software interface. Click on the “CodesignStart” submenu, the moment of collaborative design is coming. Figure 7 shows single designer’s interface on Co-Cad platform before collaborative design with others.

Step 3: Each designer can launch codesign request and select codesign partner. Only when both designers agreed to other’s request, they can go on collaborative design and show the audio and video window. The main user interface of the agent manager in JADE platform is as Figure 8.

Step 4: Collaborative designers can communicate with each other in form of letters or video chat spontaneously and freely. If one of the designers edits or modifies the design, the same operation will show on the other’s

platform. As presented in Figure 9 and Figure 10, Selene and Jujumao are designing a mechanism accessory using Co-Cad platform. Figure 9 presents interface of Selene’s platform, and Figure 10 is Jujumao’s interface. They interact with each other in the form of video and chat. Selene modifies the plan, then the same modification shows in Jujumao’s platform. There is a small window used to input letters for chat, where speaking loudly is not politely.

6. Conclusions In this paper, on the basis of the problem identification and the analysis of the requirements for a collaborative design platform, an agent-based platform supporting collaborative design via the cooperation of a network of intelligent agents is presented. On the platform, all agents are written in the Java language using the JADE platform and work together to perform flexible, adaptive and dynamic design tasks in an autonomous and collaborative way. As the platform is still being fully implemented,



more experiments are required to be carried out in order to test and improve our platform. In the future testing of our approach and software, we intend to involve designers in the process with real design examples since we believe that a design-oriented approach needs to be taken in order to identify those key tasks that need collaboration and support by software agents.

However, some challenging problems, such as task assignment, conflict detection and conflict solution, need to be carefully addressed and further development efforts are required before the technology can be widely deployed. In our project, on-going efforts are being made to refine the coordination agent and its underlying methodology into detail.

7. Acknowledgments This project is supported by International science and technology cooperation project (NO.2006DFA73180) from China’s Ministry of Science and Technology.

8. References [1] J. X. Wang and M. X. Tang, “A multi-agent framework

for collaborative product design,” in: Z. Shi and R. Sadananda (Eds.): PRIMA 2006, LNAI, Vol. 4088, Springer, Heidelberg, pp. 514–519, 2006.

[2] J. X. Wang and M. X. Tang, “An agent-based system supporting collaborative product design,” in: B. Gabrys, R.J. Howlett, and L.C. Jain (Eds.): KES 2006, Part II, LNAI, Vol. 4252, Springer, Heidelberg, pp. 670–677, 2006.

[3] J. X. Wang and M. X. Tang, “Knowledge representation in an agent-based collaborative product design environment,” Proceedings of the 9th International Conference on Computer Supported Cooperative Work in

Design (CSCWD 2005), Vol. 1, IEEE Computer Society Press, pp. 423–428, 2005.

[4] M. A. Rosenman and F. Wang, “A component agent

based open CAD system for collaborative design,” Automation in Construction, Vol. 10(4), pp. 383–397, 2001.

[5] S. M. Wu, H. Ghenniwa, Y. Zhang, and W. M. Shen, “Personal assistant agents for collaborative design environments,” Published by Elsevier B.V. Computers in Industry 57, pp. 732–739, 2006.

[6] S. E. Lander, “Issues in multi-agent design systems,” IEEE Expert 12 (2), pp. 18–26, 1997.

[7] W. Shen, D. H. Norrie, and J. P. Barthès, “Multi-agent systems for concurrent intelligent design and manufacturing,” Taylor and Francis, London, UK, 2001.

[8] G. Toye, M. R. Cutkosky, L. Leifer, J. Tenenbaum, and J. Glicksman, “SHARE: A methodology and environment for collaborative product development,” in: Proceedings of Second Workshop on Enabling Technologies: Infrastructure for Collaborative Enterprises, pp. 33–47, 1993.

[9] D. C. Brown, B. Dunskus, D. L. Grecu, and I. Berker, “SINE: support for single function agents,” in: Proceedings of Applications of Artificial Intelligence in Engineering, Udine, Italy, 1995.

[10] W. Shen and J. P. Barthès, “An experimental environment for exchanging engineering design knowledge by cognitive agents,” in: M. Mantyla, S. Finger, T. Tomiyama (Eds.), Knowledge Intensive CAD-2, Chapman &Hall, pp. 19–38, 1997.

[11] M. J. Hague and A. Taleb-Bendiab, “Tool for management of concurrent conceptual engineering design,” Concurrent Engineering: Research and Applications 6 (2), pp. 111–129, 1998.

[12] M. I. Campbell, J. Cagan, and K. Kotovsky, “A-Design: an agent-based approach to conceptual design in a dynamic environment,” Research in Engineering Design 11, pp. 172–192, 1999.



A Multi-User Cooperative Diversity for Wireless Local Area Networks

Jun CHEN1, Karim DJOUANI 2 1 LISSI Lab., University Paris 12 and CEDRIC-CNAM Paris, France

2 Member, IEEE, F’SATIE/TUT Pretoria, South Africa and LISSI Lab., University Paris 12, France E-mail: [email protected], [email protected]

Received on November 12, 2007; revised and accepted on May 3, 2008 Abstract In this paper, an idea of using space-time block coding (STBC) in multi-user cooperative diversity has been exploited to improve the performance of the transmission in wireless local area networks. The theoretical and simulation results show that, using STBC approaches can always achieve the better performance than existing techniques without introducing the space-time coding. By analyzing the throughput and frame error ratio (FER) of the two different STBC cooperative schemes, we find the trade-off between throughput and reliability. The location of the relay is crucial to the performance, which supposes a rule for future cross-layer design. Keywords: Multiple-input-multiple-output (MIMO), Cooperation, Space-time Block Coding (STBC).

1. Introduction

Diversity is a powerful technique to mitigate fading and improve robustness to interference [1], which refers to the method by conveying the signal to the receiver over multiple independently signal fading channels. The conventional view of transmit diversity is that a single wireless terminal transmits using an array of multiple-antennas so that the paths from each antenna to the destination with independently fading. The recent research work in this area is the space-time coding (STC) techniques that have been developed for multi-antenna arrays. STC is an effective coding technique that uses transmit diversity to combat the detrimental effects in wireless fading channels [7]. Unfortunately, transmit diversity methods based on multiple-input-multiple-output (MIMO) approach are not applicable to many wireless systems because of the size, complexity, power or other constraints, as for instance, ad-hoc networks and sensor networks. On account of these reasons, cooperation between wireless terminals has been recently proposed as a means to provide transmit diversity as which shown in Figure 1, where S, R and D represent source, relay and destination terminal, respectively. A new method introduced in [2] and [3] to realize space

diversity gain has been studied under the name of cooperative diversity. Traditional cooperative diversity transmits the same signals through two different channels as Figure 2. In the first time slot, the source communicates to the relay and to the destination at the same time; in the second time slot, just the relay retransmits the signal received at the first time-slot to the destination. The relay may simply forward the signal received from the source terminal or retransmit the estimates of the received symbols, obtained by detection. We call it as repeat cooperation. In this paper, we present a paradigm for cooperative diversity, which we term space-time block coding (STBC) cooperation [21], integrating user cooperation with STBC.

We summarize here the relevant contributions in the area of the cooperative diversity. Relay channels and space-time code form the basis for our study. The classical three-terminal communication channels originally examined by van der Meulen [5]. For the channels with multiple information sources, Kramer and van Wijngaarden [6] consider a multiple access channel in which the sources communicate to one destination and share one relay.

Laneman et al. examines the mode of user cooperation diversity [2,3] and analyzes space time coding cooperative diversity in nonergodic settings using outage probability as a performance measure [4]. They

A MULTI-USER COOPERATIVE DIVERSITY FOR WIRELESS LOCAL AREA NETWORKS 267


Figure 1. Single-relay cooperative diversity model.

Figure 2. Time sequence of 2 time slots repeat cooperative diversity.

Figure 3. Time sequence of 2 time slots multi-user cooperative diversity. demonstrated the extent to which space-time coding cooperative diversity achieves higher diversity order than repetition-based schemes for larger spectral efficiencies in theorem. The model they analyzed is a selective orthogonal amplify and forward (OAF) protocol, where source transmits the vector of encoded data in the first time slot and relay retransmits the received vector by adjusting the power. The non-orthogonal amplify-and-forward (NAF) scheme was proposed by Nabar et al. [8,9] for the single-relay channel, where source transmits all the time but the relay only transmits on even time slots. They consider three different time-division multiple-access-based cooperative protocols that vary the degree of broadcasting and receive collision in either the amplify-and-forward (AF) or decode-and-forward (DF) modes. And the results indicate that optimal space-time code design in the single relay case consists of satisfying the classical rank and determinant criteria for co-located antennas. These academic works sustain the possibility, existence and benefits for deploying space-time coding cooperative diversity protocols in practice.

This paper examines a new 2×2 full-rate space-time code (Golden-Code) [12] in the single-relay cooperative

NAF model. For source transmits in both two time slots, this protocol can achieve a higher throughput than that of the OAF protocol. And we here consider these two types of cooperative protocols and compare the performance between Golden-Code and the classical Alamouti code [10]. Besides the distinct benefits of the space-time code, we can see the trade-off between throughput and reliability during the transmission by analyzing the results of throughput and frame error rate. At the last part, we give a basic idea about the selection of relay.

Organization of the paper. This paper continues as follows: Section 2 outlines the multi-user cooperative diversity model. Section 3 explains STBC cooperative diversity. Section 4 shows the performance analysis by the simulation results. Section 5 summarizes our conclusions.

2. Multi-User Cooperative Diversity Model We consider wireless network in which two terminals are communicating with a base station. The channel between each terminal and the base station are independent of each other, and independent of the channel between the terminals. All channels are subject to flat (frequency non-selective) fading in order to isolate the benefits of spatial diversity. Considering the multi-user cooperative diversity model, signal is to be transmitted from the source terminal S to the destination terminal D with the assistance of the relay terminal R. All the terminals are equipped with single antenna. Throughout the paper we assume that a terminal cannot transmit and receive simultaneously. the channels S→D, S→R and R→D are known to the destination terminal.

The signal transmits procession is like following: During the first time slot, the source communicates with the relay and destination. In the second time slot, both the relay and source communicate with the destination. Figure 3 shows the detail of the time sequence.

In the AF relaying method [1], the relay simply amplifies and retransmits the signal received from the source (the signal received at the relay is distorted by fading and additive noise). No demodulation or decoding of the received signal is performed at relay in this case.

The signals received by the destination and relay in the first time slot can be defined as

sdsdsdsd nxhwy += 1 (1)

and

srsrsrsr nxhwy += 1 (2)

respectively, where w2

sd and w2sr are the average signal

energies received by destination over channel S→ D and S→ R, respectively [9]. hsd and hsr are the random, complex-valued and unit-power channel gains between S→ D and S→ R. nsd ∽ CN(0, Nsd , nsr∽CN(0,Nsr) is the

268 J. CHEN ET AL.


additive noises, and in general w2sd ≠ w2

sr . The energy of received signal (3) is given by

( ) ( ) ( ) srsrsrsrsrsrsr NhwnExhwEyE +=+= 2222

1

2 (3)

In order to retransmit the signal with the same power as the sender did, the gain β for the amplification is

srsrsr Nhw += 22

1β (4)

Then, the destination receives a superposition of relay and source during the second time slot:

rdsrrdrdsdsd nyhvxhwy ++= β22 (5)

where v2rd is the average signal energy received at the

destination through channel R→ D, the definition of hrd and nrd are the similar to hsr and nsr.

So the equation (5) can be rewritten as:

nxhwvxhwy rdsrrdsdsd~

122 ++= β (6)

where ( )NCNn 0,0~∽ with NNhvN rdsrrdrdβ += 222

0

As the summary, the transmission function of this cooperative diversity is

y = Hx + n (7) where

=

=

2

1

2

1 ,x

xx

y

yy

is the received signal vector and transmitted signal vector, respectively;

+=

rdsrrdrd

sd

nnhv

nn

β (8)

is the noise vector; and H is the 2×2 channel matrix given by

=

sdsdrdsrrdsr

sd

hwhhvw

wH

β0

(9)

Assuming that the channel coefficient matrix H is known or can be estimated, Maximum Likelihood (ML) decoding can be used at receiver to fully explore the diversity advantage of the scheme. In equation (9), the noise of first time slot and second time slot do not have the same powers, the ML estimation can not be used directly. One solution is normalizing the received noise by a parameter ρ as follows:

+

=

n

n

x

x

hwhhvw

hw

y

y sd

sdsdrdsrrdsr

sdsd

~0

2

1

2

1

ρρβρρ (10)

where

rdsrrdrd

rd

NNhv

N

+= 22β

ρ (11)

Then, equation (10) can be noted as y~ = H~

x+ n~ .

Assuming that the channel coefficient matrix H~

is known or can be estimated, the ML estimate of the transmitted packets is presented as follows:

2~~minargˆFx

xHnx −= (12)

where ‖·‖F represents the Frobenius-2 norm, and x takes all possible finite values depending on the signal constellation. 3. STBC Cooperation Model STC is a method employed to improve the reliability of data transmission in wireless systems by using multiple transmit antennas. It relies on redundant copies of a signal to the receiver in the hope that at least some of them may survive the physical path between transmission and reception. Space time codes may be split into two main types: Space-time trellis coding (STTC) [16] and STBC [17]. We are only concerned here with STBC which acts on a block of data at once (similarly to block coding) and provide only diversity gain, but are much less complex in implementation terms than STTC. Alamouti coding [10] and Golden-Code [12] are typical examples of STBC. 3.1. Repeat Cooperation Firstly, we present the model shown in Figure 2, repeat cooperation transmits the same signals through two different channels. In the first time slot, the source communicates to the relay and to the destination at the same time; in the second time slot, just the relay retransmits the signal received at the first time slot to the destination. Then, the transmission function can be noted as follows:

rdsrrdrdrdsrsrrd

sdsdsd

nnhvxhhwvy

nxhwy

++=+=

ββ 12

11 (13)

The cooperative transmission function can be written as

nhxy += 1 (14)

where

+=

=

=

rdsrrdrd

sd

rdsrsrrd

sdsd

nnhv

nn

hhwv

hwh

y

yy

β

β,

2

1

3.2. Alamouti Coding Cooperation Alamouti proposed a simple MIMO scheme that achieves a full diversity gain [17] with a simple ML



decoding algorithm. The transmit signals are modulated using an M-ary modulation scheme, then the encoder takes a block of two modulated signals s1 and s2 in each encoding operation and sends it to the transmit antennas according to the code matrix:

−=

=

*12

*21

43

21

ss

ss

xx

xxC (15)

where * denotes complex conjugate. In this code matrix, the first column represents the first time slot (transmission period) in a 2×2 MIMO system [11] and the second column represents the second time slot. The first row corresponds to the signals transmitted from the first antenna and the second row corresponds to the signals transmitted from the second one. This implies that the signals are transmitting both in space (across two antennas) and time (two transmission intervals), that is to say, space-time coding.

The traditional Alamouti coding is designed for a two-transmit antenna system. Assuming the cooperative method using one-relay AF channel, we define d1 = (x1, x2) and d2 = (x3, x4). Thus, in the first time slot, the source sends d1, the relay and destination receive the signal; in the second time slot, the source and relay send d2 and xr to destination respectively. Then the Alamouti coding cooperative transmission function can be written as

Y = HX +N (16)

where

=

=

43

21

43

21 ,yy

yyY

xx

xxX

are the transmitted and received signal matrix, respectively; channel matrix H and noise N are given by

=

sdsrrdsrsdrd

sdsd

hwhhwv

hwH

β0

1 2

1 1 3 2 2 4

sd sd

rd rd sr rd sd rd rd sr rd sd

n nN

v h n n n v h n n nβ β

= + + + +

3.3. Golden-Code Cooperation The Golden-Code is a STBC for 2× 2 MIMO system as Figure 5, the coding matrix for the model is:

Figure 4. Alamouti coding in 2 × 2 MIMO model.

Figure 5. Golden-code in 2 × 2 MIMO model.

( ) ( )( ) ( )

1 2 3 41 2

3 4 3 4 1 2

1

5

s s s sx xC

x x i s s s s

α θ α θα θ α θ

+ + = = + +

(17)

where s1, s2, s3, s4∈ Z[i] are the information signals,

θαθαθθ iiii −−=−−=−=+= 1,1,2

51,

2

51 and the

factor 5

1 is necessary for energy normalizing purposes

[12]. The Golden-Code achieves the diversity multiplexing

frontier [13], and in [12] the Golden-Code was proposed as a full rate and full diversity code for 2× 2 MIMO systems.

To the cooperative method using one-relay AF channel, we define d1 = (x1,x2) and d2 = (x3, x4) which are transmitted in first time slot and second time slot, respectively. The transmission function is similar to equation (16).

4. Numeral Results In this section, some simulations are presented to show the performances of the presented approaches. In the following simulations, Rayleigh model is used for the fading channel [20], each channel multi-path is a zero mean complex Gaussian random variable, and the distance between all the terminals is assumed to be same. Transmission energies follow the hypothesis as Table 1.

Table 1. Transmission energies in simulations. Protocol 1st time slot 2nd time slot

Cooperation 2sdw =1.0 2

srw =0.5 2rdv =0.5

MIMO 211w = 2

12w =0.5 211v = 2

12v =0.5

The throughput was defined as the average number of

available frames that were transmitted in a specific time slot. We performed a random experiment consisting of 10,000 repeated independent trials. The length of each frame was fixed to N = 600 bits. Considering the multi-pack reception, the throughput can more than 1. 4.1. The Throughput Comparison between the

Repeat Cooperation and STBC Cooperation We conducted comparisons between the STBC

270 J. CHEN ET AL.


cooperation and repeat cooperation scheme. Figure 6 and Figure 7 show the results of throughput versus SNR, for 6Mbps and 12Mbps transmit rates, respectively. We observe that all the three schemes can achieve the maximum throughput with a high SNR (> 25dB). With a special coding method, Golden-Code cooperation scheme achieves a much higher throughput than the other two. Considering the coding matrix of Golden-Code, each row contains all the 4 original signals, which means the full-rate of the transmission. The cooperative method transmits 4 available signals (s1; s2; s3; s4) during 2 time slots, which means the maximum value of throughput is 2.

As to amamouti coding scheme, each row of the coding matrix contains 2 original signals (s1 and s2). In every time slot, the system transmits one signal and the conjugated signal of the other one, where s*

1 and s*2 are

surly the redundancy copies of the original signals. That is why only 2 available signals (s1 and s2) can be obtained at the destination in this scheme while 4 available signals ( 1; s2; s3; s4) can be obtained by using Golden-Code scheme. Thus, by using two pair of conjugate signals, Alamouti coding scheme transmits 2 available signals during 2 time slots of the cooperative period, which means the maximum value of throughput can no more than 1 with the increasing of SNR.

Furthermore, as shown in Figure 2, repeat cooperation transmits one signal during the first time slot and retransmits the same one in the second. Clearly, repeat cooperation can just transmit 1 signal during the two time slots. Thus, its throughput is less than 0.5.

From the simulations, we see that with the help of STBC gains, the STBC cooperation is outperform repeat cooperation. And as a reasonable result of analysis and simulation, the Golden-Code cooperation can clearly achieve the best throughput among all the three schemes. This also proves that the design of the space-time code could impact the performance of the transmission.

Figure 6. Throughput of STBC cooperation and repeat cooperation schemes (6Mbps).

Figure 7. Throughput of STBC cooperation and repeat cooperation schemes (12Mbps). 4.2. The FER Comparison between Non-

ooperation, Repeat Cooperation and STBC Cooperation

The simulation results of FER versus SNR between Non-cooperation and cooperation schemes demonstrate again that the use of relay-assisted communication is not always beneficial when compared to direct transmission (Non-cooperation scheme) [8]. Figure 8 and Figure 9 reveal that the frame error rate of Non-cooperation communication is better than that of the simple repeat cooperation for a high SNR (>35dB).

Further, as expected, cooperation with STBC is always preferred over Non-cooperation scheme. Thus from our simulations, we see that, performance using STBC cooperation improves significantly over Non-cooperation demonstrating the advantage of using STBC cooperation. Between the two STBC cooperation schemes (Alamouti coding and Golden-Code), Alamouti coding method shows a better performance. As we discussed, Alamouti coding transmits the redundance

Figure 8. FER versus SNR for Non-cooperation and cooperation schemes (6Mbps).



Figure 9. FER versus SNR for Non-cooperation and cooperation schemes (12Mbps).

Figure 10. FER versus SNR for STBC in cooperation and MIMO schemes (6Mbps). signals, a original and a conjugate. This is the reason that it has a lower error rate in the destination while Golden-Code just intersperses original signal among all parts of the transmit signals.

Comparing with the simulation results about the throughput of these two STBC cooperation schemes, we see that, Alamouti coding have a lower throughput but a higher reliability than that of Golden-Code. As a summary, there is always a trade-off between the throughput and the reliability. 4.3. The FER Comparison between STB

Cooperation and MIMO Schemes Figure 10 and Figure 11 show us the FER of cooperation and MIMO system referring to the different SNRs. According to the simulation results, the MIMO systems achieve lower FER than the corresponding cooperative schemes. This supports that MIMO channels allowing multiplexing gain [14,15] which is absent in cooperative relaying channel since time is expended in the latter. Thus, using MIMO system always obtains the gain of spatial diversity. And as expected, the Alamouti coding method has a better performance than the corresponding

Golden-Code method. The simulation result demonstrates again that there is a trade-off between the throughput and the reliability. 4.4. Effect via the Movements of the Relay

The main building blocks of a wireless network design are rate control, power control, medium access (scheduling) and routing. These building blocks are divided in layers. Typically, routing is considered in a routing layer and medium access in a MAC-layer, whereas power control and rate control are sometimes considered in a PHY-layer and sometimes in a MAC-layer.

So far, the three stations (S, R, D) were positioned equidistantly and therefore all the three channels had the fixed distance. Let us denote the distance between source and destination as dsd; distance between source and relay as dsr and distance between relay and destination as drd. Denote SNRsd, SNRsr, SNRrd as SNR between the source and destination during the 2 time slots. We have

1

1

1

v

srsr

v

rdrd

v

sdsd

SNRd

SNRd

SNRd

∝

∝

∝

(18)

where v is the path loss exponent. In the following analysis, we assume that v = 4 for urban environment [18].

In this section, the relay is moved, so the distance between the relay and source, the relay and destination will change at the same time. The effects on the signal quality when moving the relay between the source and destination using Golden-Code cooperation with 6Mbps and 12Mbps transmission rate are shown in Figure 12 and Figure 13, respectively. In the simulations, the distance between the sender and the destination is set to one, and therefore the SNRs shown in the X-axis is only valid for the direct link S→ D.

Figure 11. FER versus SNR for STBC in cooperation and MIMO schemes (12Mbps).

272 J. CHEN ET AL.


Figure 12. Benefit results when the relay is located between the source and the destination (6Mbps).

Figure 13. Benefit results when the relay is located between the source and the destination (12Mbps).

The best performance is achieved when the relay is situated in the middle of the source and destination, which means the better channel quality at S→ R and R→ D. And this can be a rule for a relay-selection method at MAC-layer using the information of PHY-layer. 5. Conclusions This paper describes STBC cooperation in wireless communication, a technique that allows single-antenna mobiles to share their antennas for obtaining some benefits of multiple-antenna systems. The diversity is realized by using a third station as a relay and the STBC methods for information coding. We analyze the performance of two different types of STBC cooperative methods (Alamouti coding and Golden- Code) through the theoretical study and simulations, there is the trade-off between throughput and reliability during the transmission. The results show that using the STBC cooperative diversity can always increase the performance. Through the analysis of the two methods with the corresponding MIMO systems, we know that the

performance of MIMOs is always better than that of cooperation with allowing multiplexing gain. The location of the relay is crucial to the performance.

The best performance was achieved when the relay is in the middle of source and destination. And in general the relay should not be to far from the line between the two terminals.

We believe several areas of future research on cooperative communication will be fruitful. Firstly, the generalization of the one hop space-time coded cooperation to multi-hop case. Most of the research work about cooperative communication concerns the single-hop (single-relay or multi-relay) transmission. Nowadays, multi-hop ad-hoc network can be found in everywhere, and the protocol adapted to multi-hop environment always derives from that of the single-hop. Secondly, the integration and interaction with higher layer network protocols can be explored. Recently, the need for protocol adaptation and code cooperation of wireless communication system suggested a new concept of protocol architecture, named cross-layering architecture. Different protocols implemented at different protocol layers may be designed to have mutually cooperative reactions, based on sharing the information between the different layers. Obviously, a cross-layer approach that is based on metrics computed at physical layer as SNR and minimal distance in the decoding process is under investigation. Such approach will be of a certain interest for MAC and Network levels, taking advantage of the information measured or estimated at the physical layer. Our contribution will concern, mainly, link adaptation and frames scheduling at MAC level. Lastly, generalization of the STBC approach to meshed network while considering multi-channel cooperation, radio resources management and link adaptation will be our crucial objective in perspective.

6. Acknowledgement This work comes within the framework of a project supported by the Agence Nationale de la Recherche/ R’eseau National de Recherche en T’el’ecommunications under name RNRT/RADIC-SF/COMSIS and reference ANR-05-RNRT- 014-01.

7. References [1] A. Nosratinia, T. Hunter, and A. Hedayat, “Cooperative

communication in wireless networks,” IEEE Communications Magazine, Vol. 42, No. 10, pp. 68–73, October 2004.

[2] J. Laneman, G. Wornell, and D. Tse, “An efficient protocol for realizing cooperative diversity in wireless networks,” in Proceedings IEEE ISIT, Washington, DC, pp. 294, June 2001.



[3] J. Laneman, D. Tse, and G. Wornell, “Cooperative diversity in wireless networks: efficient protocols and outage behavior,” IEEE Transactions on Information Theory, Vol. 50, No. 12, pp. 3062–3080, December 2004.

[4] J. Laneman and G. Wornell, “Distributed space-time coded protocols for exploiting cooperative diversity in wireless networks,” IEEE Transactions on Information Theory, Vol. 49, No. 10, pp. 2415–2425, October 2003.

[5] E. van der Meulen, “Three-terminal communication channels,” Advanced Applications Probability, Vol. 3, pp. 120–154, 1971.

[6] G. Kramer and A. van Wijngaarden, “On the white Gaussian multipleaccess relay channel,” in Proceedings IEEE International Symposium Information Theory (ISIT), Sorrento, Italy, p. 40, June 2000.

[7] J. Q. Li, K. Letaief, and Z. G. Cao, “Co-Channel interference cancellation for space-time coded OFDM systems,” IEEE Transactions on Wireless Communications, Vol. 2, No. 1, pp. 41–49, January 2003.

[8] R. Nabar and H. Bolcskei, “Space-time signal design for fading relay channels,” Proceedings IEEE Globecom, San Francisco, CA, Vol. 4, pp. 1952–1956, December 2003.

[9] R. Nabar, H. Bolcskei, and F. Kneubuhler, “Fading relay channels: performance limits and space-time signal design,” IEEE Journal on Selected Areas in Communications, Vol. 22, No. 6, pp. 1099–1109, August 2004.

[10] S. Alamouti, “A simple transmit diversity technique for wireless communications,” IEEE Journal on Selected Areas in Communications, Vol. 16, No. 8, pp. 1451–1458, October 1998.

[11] J. Proakis, “Digital communications,” Fourth edition. McGraw-Hill, 2001.

[12] J. Belfiore, G. Rekaya, and E. Viterbo, “The Golden Code: A 2x2 full-rate space-time code with non-vanishing determinants,” IEEE Transactions on

Information Theory, Vol. 51, No. 4, pp. 1432–1436, April 2005.

[13] H. Yao and G. Wornell, “Achieving the full MIMO diversity-multiplexing frontier with rotation-based space-time codes,” Proceedings of Allerton Conference on Communication, Control and Computing, October 2003.

[14] L. Zheng and D. Tse, “Diversity and multiplexing: A fundamental tradeoff in multiple-antenna channels,” IEEE Transactions on Information Theory, Vol. 49, No. 5, pp. 1073–1096, May 2003.

[15] D. Tse, P. Viswanath, and L. Zheng, “Diversity-Multiplexing tradeoff in multiple access channels,” IEEE Transactions on Information Theory, Vol. 50, No. 9, pp. 1859–1874, September 2004.

[16] Vahid Tarokh, Nambi Seshadri, and A. Calderbank, “Space-time codes for high data rate wireless communication: Performance analysis and code construction,” IEEE Transactions on Information Theory, Vol. 44, No. 2, pp. 744–765, March 1998.

[17] V. Tarokh, H. Jafarkhani, and A. Calderbank, “Space-time block codes from orthogonal designs,” IEEE Transactions on Information Theory, Vol. 45, No. 5, pp. 1456–1467, July 1999.

[18] T. Rappaport, “Wireless communications: priciples and practice,” New Jersey: Prentice Hall, 1996.

[19] R. Lin and A. Petropulu, “A new wireless medium access protocol based on cooperation,” IEEE Transactions on Signal Processing, December 2005.

[20] Y. Zheng and C. Xiao, “Improved models for the generation of multiple uncorrelated Rayleigh fading waveforms,” IEEE Communications Letters, Vol. 6, No. 6, June 2002.

[21] J. Chen and K. Djouani. “Space time coding in amplify-and-forward cooperative channel,” In Proceedings IEEE International Conference on Wireless Communications, Networking and Mobile Computing, Shanghai, CN, pp. 267–270, September 2007.



Priority-Based Resource Allocation for Downlink OFDMA Systems Supporting RT and NRT Traffics

Hua WANG, Lars DITTMANN

Department of Communications, Optics & Materials Technical University of Denmark, Lyngby, Denmark

E-mail: huw, ld @com.dtu.dk Received on March 11, 2008; revised and accepted on May 22, 2008

Abstract Efficient radio resource management is essential in Quality-of-Service (QoS) provisioning for wireless communication networks. In this paper, we propose a novel priority-based packet scheduling algorithm for downlink OFDMA systems. The proposed algorithm is designed to support heterogeneous applications consisting of both real-time (RT) and non-real-time (NRT) traffics with the objective to increase the spectrum efficiency while satisfying diverse QoS requirements. It tightly couples the subchannel allocation and packet scheduling together through an integrated cross-layer approach in which each packet is assigned a priority value based on both the instantaneous channel conditions as well as the QoS constraints. An efficient suboptimal heuristic algorithm is proposed to reduce the computational complexity with marginal performance degradation compared to the optimal solution. Simulation results show that the proposed algorithm can significantly improve the system performance in terms of high spectral efficiency and low outage probability compared to conventional packet scheduling algorithms, thus is very suitable for the downlink of current OFDMA systems. Keywords: OFDMA, Radio Resource Management, Quality of Service, Real-time and Non-real-time

Traffics

1. Introduction

Orthogonal Frequency Division Multiple Access (OFDMA) is an attractive multiple access scheme for future wireless and mobile communication systems, which has been developed to support a variety of multimedia applications with different Quality-of-Service (QoS) requirements. OFDMA builds on Orthogonal Frequency Division Multiplexing (OFDM), which is immune to intersymbol interference and frequency selective fading, as it divides the frequency band into a group of mutually orthogonal subcarriers, each having a much lower bandwidth than the coherence bandwidth of the channel. In multi-user environment, each user is dynamically assigned to a subset of subcarriers in each frame, to take advantage of the fact that at any time instance, channel responses are different for different users and at different subcarriers [1]. This capability of OFDMA systems enables the network to perform a

flexible radio resource management, such as dynamic subcarrier assignment (DSA), adaptive power allocation (APA), and adaptive modulation and coding (AMC) scheme to improve the system performance significantly under different traffic loads and time-varying channel conditions.

Recently, radio resource management for OFDMA systems has attracted enormous research interests in both academia and industry. Many scheduling algorithms have been proposed which can adapt to changes in users’ channel conditions and QoS requirements. In the literature, the resource allocation problem can be divided into two categories with different design objectives. The objective of the first category is to minimize the total transmit power subject to individual data rate constraints, see [2–4]. The objective of the second category aims at maximizing the overall (weighted) transmission rate subject to power constraints, see [5–7]. In either case, the optimal resource allocation solutions are difficult to get due to high computational complexity of non-linear

PRIORITY-BASED RESOURCE ALLOCATION FOR DOWNLINK OFDMA 275 SYSTEMS SUPPORTING RT AND NRT TRAFFICS


Figure 1. Adjacent and distributed subcarrier allocation.

optimization with integer variables. Instead, suboptimal solutions based on relaxation, problem splitting, or heuristic algorithms are proposed to reduce computational complexity [8]. Such algorithms are often refereed to as loading algorithms.

In most loading algorithms, the QoS requirement of each user is usually defined in terms of a fixed number of transmission bits per frame. However, in practical communication systems, it is neither sufficient nor efficient to represent different QoS requirements solely by a fixed data rate per frame. The resource allocation problem for systems supporting both realtime (RT) and non-real-time (NRT) multimedia traffic becomes much more complicated when diverse QoS requirements have to be considered. The transmission of RT packets can be delayed as long as the delay constraint is not violated, and the transmission of NRT packets can be more elastic. Furthermore, most loading algorithms assume that users always have data to transmit, which is not the case in real systems. Instead, appropriate traffic models should be taken into account in the design of scheduling algorithms. Therefore, efficient packetbased scheduling algorithms are of interest. Many packet scheduling algorithms with different design objectives have been proposed in the literature [9–11].

In this paper, we propose a novel resource allocation algorithm for downlink OFDMA systems supporting both RT and NRT multimedia traffic. Unlike the conventional approaches, which decompose the resource allocation

into two steps: packet scheduling and subcarrier-and-power allocation [4,11], the proposed algorithm tightly couples these two steps together through an integrated cross-layer approach to take advantage of the inter-dependencies between PHY and MAC layers. The basic idea is that if a packet is scheduled for transmission on a specific subchannel, it will get a priority value based on both the instantaneous channel conditions as well as the QoS requirements. Then we can formulate the resource allocation problem into an optimization problem with the objective to maximize the total achievable priority values. A suboptimal heuristic algorithm is also proposed to reduce the computational complexity. Simulation results show that the proposed algorithms can achieve high spectral efficiency with satisfying QoS performance in each service class.

The rest of the paper is organized as follows. We first give a brief introduction of the system model in Section 2. The resource allocation problem is formulated in Section 3. Section 4 presents a suboptimal heuristic algorithm with low computation complexity. Simulation environments and results are outlined and discussed in Section 5. Finally, conclusions and future work are drawn in Section 6. 2. System Model OFDMA is a multiple access scheme based on OFDM. While OFDM employs fast Fourier transform (FFT) of

276 H. WANG ET AL.


size 256 (subcarriers) in fixed WiMAX, OFDMA employs a larger FFT space (2048 and 4096 subcarriers) which are further grouped into subchannels. The subchannels are assigned to different users and may employ different modulation and coding schemes to exploit frequency diversity as well as time diversity [12]. There are two approaches of allocating subcarriers to form a subchannel in OFDMA: distributed subcarrier permutation and adjacent subcarrier permutation. The two approaches are shown in Figure 1. In distributed subcarrier permutation, a subchannel is formed with different subcarriers randomly distributed across the channel spectrum. This approach maximizes the frequency diversity and averages inter-cell interference. It is suitable for mobile environment where channel characteristics change fast. Both partial usage of subchannels (PUSC) and full usage of subchannels (FUSC) schemes employ distributed subcarrier permutation. In adjacent subcarrier permutation, a subchannel is formed by grouping adjacent subcarriers. This approach creates a ‘loading gain’ and is easy to use with beam-forming adaptive antenna system (AAS). It is suitable for stationary or nomadic environment where channel characteristics change slowly. The AMC scheme employs adjacent subcarrier permutation.

In this paper, we assume that subscriber stations are stationary or nomadic users with slowly varying channel conditions. Therefore, adjacent subcarrier permutation strategy is employed to support AMC. In OFDMA, radio resource is partitioned in both frequency domain and time domain, which results in a hybrid frequency-time domain resource allocation. It provides an added dimension of flexibility in terms of higher granularity compared to OFDM/TDM systems.

We consider the downlink scenario of an infrastructure- based OFDMA system with Us subcarriers and K users. At the physical layer, the time axis is divided into frames with fixed length, each of which consists of a downlink (DL) and an uplink (UL) subframe to support TDD operation. In each DL subframe, there are Ut time slots available for downlink transmissions, each of which may contain one or several OFDM symbols. To reduce the resource addressing space, channel coherence in frequency and time is exploited by grouping Is adjacent subcarriers and I t time slots to form a basic resource unit (BRU) for resource allocation. A BRU is the minimum resource allocation unit as shown in Figure 2. The size of a BRU is adjusted so that the channel experiences flat fading in both frequency and time domain. Thus in each DL subframe, there are S = Us/Is subchannels in frequency domain and N = Ut/I t slots in time domain, which corresponds to a total of S * N BRUs available in frequency-time domain for DL transmissions. Each BRU can be assigned to different users and be independently bit and power loaded. In principle, adaptive power allocation in each BRU can improve the system performance. However, some studies show that performance improvements are only marginal over a wide

Figure 2. Frequency-time domain redio resource allocation in OFDMA systems. range of SNRs due to the statistical effects [1]. Therefore, we assume that the total transmission power is equally distributed among all subchannels.

We further assume that in each frame the base station (BS) has perfect knowledge of channel state information (CSI) for each subchannel of each user. This can be obtained by piggybacking such information in each uplink packet, which is suitable for slowly varying channels. Based on CSI, adaptive modulation and coding scheme is employed to adjust the transmission mode dynamically according to the time-varying channel conditions. Multiple transmission modes are available, with each mode representing a pair of specific modulation format and a forward error correcting code. The transmission mode is determined by the instantaneous signal-to-noise ratio (SNR). To utilize the PHY layer resources more efficiently, fragmentation at the MAC layer is enabled. A separate queue with a finite queue length of L MAC protocol data units (PDUs) is maintained for each connection at the base station. We assume that the MAC PDUs are of fixed size, each of which contains d information bits.

3. Resource Allocation Model The resource allocation at the BS involves the allocation of subchannels, time slots, and modulation order and coding rate assignment. It is executed at the beginning of every frame to properly allocate radio resources to the demanding users according to their queue status, CSI, and QoS requirements.

The real-time traffic is delay-sensitive and has strict delay requirement. The non-real-time traffic can tolerate longer delays, but requires a minimum throughput. We propose a novel priority-based packet scheduling algorithm to support both RT and NRT multimedia traffic with high spectral efficiency and good QoS satisfaction. The basic idea behind the proposed algorithm is that the transmission is scheduled on a



packet-bypacket basis. Specifically, at each scheduling interval, if a PDU was scheduled for transmission on a specific subchannel, it is assigned a priority value based on the instantaneous channel condition (PHY layer issue), as well as the QoS constraint (MAC layer issue). Then we can formulate the scheduling problem into a mathematical optimization problem with the objective to maximize the total achievable priority values.

We apply an extended EXP algorithm as our priority function for both RT and NRT traffics. The EXP rule was proposed to provide QoS guarantees over a shared wireless link in terms of the average packet delay for RT traffic and a minimum throughput for NRT traffic [15].

For RT traffic, if the i th PDU from the kth connection is scheduled for transmission on subchannel n, its priority value is calculated as:

( ) ( )( )

( )

+

−⋅⋅=

aW

aWtWa

tµ

tµank,i, ik,k

k

nk,k

1expP (1)

where ( )∑=k kk tWa

kaW 1,

1 ,and ( )tWTa ikkkk ,max,log ⋅−= δ is

the i th PDU delay of connection k at time t, Tk,max is the maximum allowable delay of connection k, δk is the maximum outage probability of connection k, µk,n(t) is the instantaneous channel rate with respect to the signal-to-noise ratio and a predetermined target error probability if subchannel n is assigned to connection k at time t, and

( )tkµ is the exponential moving average (EMA) channel

rate of connection k with a smoothing factor tc, calculated as:

( ) ( ) ( )tt

tt

t kc

kc

k µµµ 11

11 +−

−= (2)

where ∑ ⋅= = )()( ,,1 tµctµ nknkNnk is the total channel rate

of connection k at time t. If subchannel n is assigned to connection k, ck,n =1, otherwise ck,n =0.

For NRT traffic, the extended EXP algorithm is used in conjunction with a token bucket control to guarantee a minimum throughput [15]. We associate each NRT queue with a virtual token bucket. Tokens in each bucket arrive at a constant rate rk,req, which is the required minimum throughput of connection k. Let us define Vk,i(t) to be the virtual waiting time of the i th PDU from connection k:

( ) ( ) ( ) NRTk

r

ditQtV

reqk

kik ∈⋅−−=

,,

1,0max (3)

where Qk(t) is the number of tokens associated with connection k at time t, and d is the fixed size of a MAC PDU. Note that we do not need to actually maintain the virtual waiting time, as the arrival rates of tokens are

constant. Then, the calculation of the priority for a NRT PDU is similar to Exp.(1), with Wk,i(t) being replaced by Vk,i(t). After a PDU is scheduled for transmission, the number of tokens in the corresponding token queue is reduced by the actual amount of data transmitted.

Let u(k,i,n) be defined as a binary random variable indicating subchannel allocation. That is, u(k,i,n)=1 means that the i th PDU from connection k is allocated for transmission on subchannel n, and u(k,i,n)=0 otherwise. Also let us define m(k,i,n) be the number of time slots occupied on subchannel n if the i th PDU from connection k is scheduled for transmission on subchannel n, calculated as:

( ) ( )

=

tµ

dni,k,

nk,

m (4)

where [x] denotes the smallest integer larger than x. Then, the scheduling problem can be mathematically

formulated as follows:

( )( ) ( )ni,k,ni,k,

K

k

L

i

S

nni,k,

Puu

⋅∑∑∑= = =1 1 1

maxarg (5)

Subject to:

( ) ( ) nNni,k,ni,k,K

k

L

i

∀≤⋅∑∑= =

mu1 1

(6)

( ) ik,ni,k,S

n

∀≤∑=

11

u (7)

( ) ni,k,ni,k, ∀∈ 1,0u (8)

where S denotes the total number of subchannels, N denotes the total number of time slots, K denotes the total number of connections, and L denotes the maximum queue size.

The first constraint ensures that the allocated bandwidth does not exceed the total available bandwidth in terms of time slots on each subchannel. The second constraint says that a PDU can only be transmitted via one subchannel. The instantaneous channel conditions and the QoS related parameters are embodied into the priority function P(k,i,n) with the objective of maximizing the total achievable priority values, thus improving the spectral efficiency while maintaining QoS guarantees.

The above optimization problem can be solved by determining the values of binary variable u(k,i,n) through standard linear integer programming (LIP)1. The solution to the problem provides an optimal resource allocation.

1The optimal solution of the LIP problem formulated in this paper is obtained by using the General Algebraic Modeling System(GAMS).

278 H. WANG ET AL.


However, the computation complexity of the optimal solution is too high to be applied in practical systems. To reduce the computational complexity, we propose a suboptimal heuristic algorithm with low complexity in the next section. 4. Proposed Suboptimal Scheme In the suboptimal algorithm, we allocate radio resources on a packet-by-packet basis. The general idea is that, at each scheduling interval, the packet with the highest priority value from all queues is scheduled for transmission, and this procedure continues until either there is no radio resource left or there is no packet remaining unscheduled in the queue. A detailed description of the proposed scheduling algorithm is listed in pseudocode 1, where k

sΩ is the set of subchannels that are available for data transmission of connection k, tn is the number of residual time slots on subchannel n, qk is the current queue size of connection k, and ik is a pointer to the next PDU to be scheduled of connection k.

It works as follows: If connection k has pending traffic in the queue, the proposed algorithm first pre-allocates the best subchannel n in terms of the instantaneous channel quality to connection k from its available subchannel set k

sΩ (see Step 14). If there is not enough capacity left on the best subchannel n to accommodate one PDU from connection k’s queue, subchannel n will be removed from connection k’s available subchannel set, and the second best subchannel n' will be selected. This procedure continues until a best possible subchannel is pre-allocated to connection k (see Step 13-22). Otherwise, connection k is removed from the scheduling list. After the subchannel pre-allocation process for all connections is complete, the algorithm calculates the priority value of the head-of-line (HOL) PDU in each nonempty queue, and schedule the PDU with the highest priority value for transmission on subchannel n* (see Step 16 & 24). The scheduled PDU is removed from the corresponding queue and the consumed radio resources in terms of time slots are subtracted on subchannel n* (see Step 25 & 26-30). Then it starts from the beginning and continues until either there is no radio resource left or there is no PDU pending in the queue. A detailed flowchart of the proposed suboptimal algorithm is given in Appendix I. 5. Simulation Results and Discussions To evaluate the performance of the proposed resource allocation algorithm for downlink OFDMA systems supporting both RT and NRT multimedia traffic, a system-level simulation is performed in OPNET.

Algorithm 1 Suboptimal Packet Scheduling Algorithm for Downlink OFDMA Systems

1: Set Ntn ← for n∀ ｛initialize nt ｝

2: Set 1←ki for k∀ ｛initialize ki ｝

3: Get kq for k∀ ｛get the queue size of connection k ｝

4: for k =1 to K do 5: if qk＞0 then

6: Set ←Ωks

｛1,…,S｝｛initialize ksΩ ｝

7: else 8: Set φ←Ωk

s｛set k

sΩ to be null｝

9: end if 10: end for 11: while φ≠Ω∃ x

sx, do

12: for k =1 to K do

13: while φ≠Ωks do

14: Select n ← arg )(max , tnkn k

sµΩ∈

｛assign the best

subchannel from the available subchannel set｝

15: if nt ≥

)(, t

d

nkµ then

16: Calculate P(k,ik,n) in Exp. (1) 17: BREAK 18: else

19: −Ω←Ω k

sks ｛n｝｛remove n from the available

subchannel set if there is not enough capacity left｝

20: CONTINUE 21: end if 22: end while 23: end for

24: Schedule the *ki th PDU of connection *k on subchannel

*n , where ( *,, ** nik k

) ← arg max P ( nik k ,, )

25:

−←

)(**,** t

dtt

nknn µ

｛update the residual time slots｝

26: if kk qi =* then

27: φ←Ω *k

s ｛set *ksΩ to be null when all pending PDUs

of connection k* have been scheduled for transmission｝ 28: else 29: 1** +← kk ii ｛point to the next pending PDU｝

30: end if 31: end while

5.1. System Model

We consider the downlink of a single-cell OFDMA system with TDD operation. The cell radius is 2 km, where subscriber stations are randomly placed in the cell with uniform distribution. The total bandwidth is set to be 5 MHz, which is divided into 10 subchannels. The BS transmit power is set to 20W (43 dBm) which is evenly distributed among all subchannels. The duration of a frame is set to be 1 ms so that the channel quality of each connection remains almost constant within a frame, but may vary from frame to frame. The propagation model is derived from IEEE 802.16 SUI channel model [20]. Path loss is modeled according to terrain Type A suburban macro-cell. Large-scale shadowing is modeled by log-normal distribution with zero mean and standard deviation of 8 dB. The rms delay spread is 0.5µs, typical of an urban environment. The effect of small scale



Table 1. A summary of system parameters.

Parameters Value System OFDMA/TDD Central frequency 3500 MHz Channel bandwidth 5 MHz Number of subchannels 10 Length of OFDM symbol 156.25 µs User distribution Uniform Beam pattern Omni-directional Cell radius 2 km Frame duration 1 ms BS transmit power 20 W Thermal noise density –174 dBm/Hz Propagation model 802.16 SUI-5 Channel model Maximum MAC PDU size 256 bytes

Table 2. Modulation and Coding Schemes for 802.16 [16].

Modulation scheme

Coding rate

Bits/ symbol

Target SNR for 1% PER (dB)

BPSK 1/2 0.5 1.5 QPSK 1/2 1 6.4 QPSK 3/4 1.5 8.2

16QAM 1/2 2 13.4 16QAM 3/4 3 16.2 64QAM 1/2 4 21.7 64QAM 3/4 4.5 24.4

Table 3. A summary of traffic paramenters.

Type Characteristics Distribution Parameters

VoIP ON period Exponential Mean=1.34 sec VoIP OFF period Exponential Mean=1.67 sec VoIP Packet size Constant 66 bytes

VoIP Inter-arrival time between packets Constant 20 ms

Video Packet size Log-normal Mean=4.9 bytes Std.dev.=10 ms

Video Inter-arrival time between packets Normal Mean=33 ms

Std.dev.=10 ms

Web Reading time between sessions Exponential Mean=5 sec

Web Number of packets within a packet call Geometric Mean=25

packets

Web Inter-arrival time between packets Geometric Mean=0.0277

sec

Web Packet size Truncated Pareto

k=81.5 bytes α=1.1 m=2 M bytes

multipath fading is modeled by a tapped delay line (TDL) with exponential power delay profile as follows:

( ) ( ) ( )( )ttth i

N

ii ττδβτ −=∑

=0

, (9)

where N is the total number of paths, δ(·) is the Dirac impulse, βi(t) and τi(t) are the time-variant gain and delay of the i th path, respectively. The channel gains βi(t) are zero mean mutually independent Gaussian stationary processes with an exponentially decaying power profile

and a classical Jake’s spectrum. The thermal noise density is assumed to be –174 dBm/Hz.

Table 1 summarizes the system parameters used in the simulation. We assume that all MAC PDUs are transmitted and received without errors and the transmission delay is negligible. The modulation order and coding rate in the AMC scheme is determined by the instantaneous SNR of each user on each subchannel. We follow the AMC table shown in Table 2, which specifies the minimum SNR required to meet a target packet error rate, e.g., 1%. 5.2. Traffic Model In the simulation, three types of traffic sources are generated: Real-time (RT) voice: RT voice traffic is assumed to

be VoIP that periodically generates packets of fixed size. Assuming that silence suppression is used, VoIP traffic can be modeled as a two-state Markov ON/OFF source [17].

Real-time (RT) video: RT video traffic is assumed to be the videoconference which consists of a VoIP source and a video source [17]. A video source periodically generates packets of variable size.

Non-real-time (NRT) data: NRT data traffic is assumed to be Internet traffic such as web browsing that requires large bandwidth and generates bursty data of variable size. We apply the Web browsing model for the Internet traffic [18]. It is assumed that each user has a connection pair

consisting of a RT connection and a NRT connection. VoIP and video traffic is served in RT connection while data traffic is served in NRT connection. Each connection alternates between the states of idle and busy, which are both exponentially distributed, and is loaded with corresponding traffic source when the connection is in busy state. A summary of traffic parameters of different traffic types are listed in Table 3.

5.3. Performance Evaluation

We evaluate and compare the performance of the proposed priority-based scheduling algorithm with other conventional algorithms in terms of the average packet delay, the throughput, the outage probabilities, and the modulation efficiency via extensive computer simulations.

For delay-sensitive RT traffic, the average packet delay and the delay outage probability are the main performance metrics. The delay constraint for RT traffic is set to be 50ms. For loss-sensitive NRT traffic, the average throughput and the throughput outage probability are the main performance metrics. The minimum throughput constraint for NRT traffic is set to be 100 Kbits/sec. The outage probabilities for both RT

280 H. WANG ET AL.


and NRT traffic should be less than 3%. In order to evaluate the spectral efficiency, the modulation efficiency is also considered in the performance evaluation.

For comparisons, we include the simulation results of two conventional scheduling algorithms proposed for OFDMA systems. The first one is maximum SNR, where users are selected for transmission over each subchannel according to their CSI. The second one is proportional fair (PF) [14], where users are selected for transmission over subchannel n according to the following criteria:

( )( )tt

ini

ni

in

,

,* maxargµµ

= (10)

where ( )tni ,µ is the average data rate of the nth

subchannel of user i. To compare the performance between OFDM/TDM and OFDMA based systems, simulation results of the EXP rule applied in OFDM/TDM systems are also included. The EXP rule is considered to be one of the best scheduling algorithms in OFDM/TDM based systems [15], of which each user transmits in the assigned time slots over all subchannels.

Figure 3 shows the average packet delay of RT traffic versus the number of users for different scheduling algorithms. When the number of users is below 48, the average packet delay of the proposed scheme increases marginally and it is well kept below the maximum allowable delay, which is 50 ms in our scenario. After that point, the system is overloaded and the average packet delay increases sharply. Similar phenomenon of the proposed scheme can be observed for the delay outage probability shown in Figure 4. However, the average packet delay of the PF scheme and the MAX- SNR scheme is much larger compared to our proposed scheme, which consequently results in a higher delay outage probability when the number of users is below 48. Furthermore, it can be seen from Figure 4 that when the number of users is above 48, the delay outage probability

Figure 3. Average packet delay in RT.

Figure 4. Delay outage probability in RT.

of the proposed scheme increases rapidly to one, which means that the system is overloaded and almost no RT connections can maintain the required delay constraint. On the other hand, some RT connections in the PF and MAX-SNR schemes can still maintain the required delay constraint as the delay outage probabilities in these two schemes increase steadily with respect to the number of users. This is because in the proposed scheme, it not only takes the instantaneous channel conditions, but also the delay requirement into consideration when scheduling packets. RT connections with larger packet delay are assigned higher priorities in an effort to average out the packet delay among all RT connections. As a result, each RT connection will have similar average packet delay regardless of its channel conditions. When the system is overloaded, congestion occurs and all RT connections will experience bandwidth starvation, which results in a sharp increase of the average packet delay and the delay outage probability. However, in the PF and MAX-SNR schemes, the scheduler selects a connection for transmission only based on instantaneous channel conditions. As a consequence, connections with good channel conditions will always experience very short delay at the cost of bandwidth starvation for connections with poor channel conditions. Therefore, the delay outage probability in the PF and MAX-SNR schemes increases much more smoothly compared to the proposed scheme when the number of users is above 48. As for the EXP rule applied in OFDM/TDM systems, the dotted line in Figure 3 & 4 indicates that the performance of OFDM-based system is much worse than OFDMA-based system.

Figure 4 shows the delay outage probability of RT traffic versus the number of users for different scheduling algorithms. It is obvious that the proposed scheme outperforms over the other conventional schemes. The maximum number of supportable RT users under a predefined 3% outage probability in PF, MAX-SNR and the proposed scheme are 38, 38, and 48 respectively.



Figure 5 shows the throughput of NRT traffic versus the number of users for different scheduling algorithms. The throughput of the MAX-SNR scheme achieves the highest value among all schemes. It increases proportional to the number of users. In the proposed scheme, the throughput increases proportional to the number of users when there are less than 50 users. After that point, the throughput remains on a steady level regardless of the number of users. While in the PF scheme, the throughput is significantly lower than the other schemes. It can be explained as follows: In the MAXSNR scheme, the scheduler simply selects the connection with the best CSI for transmission. When the number of users increases, the scheduler has more chance to serve a user in good channel conditions (multi-user diversity gain) which results in a high throughput. That’s why the spectral efficiency of the MAX-SNR scheme increases with respect to the number of users shown in Figure 7. In the proposed scheme, both the CSI and the QoS constraints are taken into account to guarantee the required QoS performance (i.e., a minimum throughput of 100Kbpbs for each NRT connection). When the system is underloaded (the number of users is less than 50), the bandwidth is large enough to satisfy the QoS requirements of all connections and the scheduling criterion mainly concerns with the CSI of each connection. As a result, the throughput as well as the spectral efficiency increases proportional to the number of users. However, when the system is overloaded (the number of users is above 50), the bandwidth is not sufficient to satisfy the QoS requirements of all connections. Thus congestion occurs and the throughput reaches at a steady level. When congestion occurs, the proposed algorithm tends to put more weight on the QoS constraint than the CSI in an effort to provide equal opportunities of QoS satisfaction among all NRT connections. In other words, the throughput of each NRT connection in the proposed scheme decreases proportionally to the number of users when the system is overloaded. That explains a sharp increase of the throughput outage probability shown in Figure 6. In the PF scheme, the throughput is relatively low due to the reason that the spectral efficiency is significantly lower than the MAX-SNR and the proposed schemes. Again, from Figure 5 & 6, we can see that OFDMA based scheduling algorithms have better performance than OFDM/TDM based scheduling algorithm.

Figure 6 shows the throughput outage probability of NRT traffic versus the number of users for different scheduling algorithms. It is obvious that the proposed scheme outperforms over the other conventional schemes. The maximum number of supportable NRT users under a predefined 3% outage probability in PF, MAX-SNR and the proposed scheme are 34, 34, and 50 respectively.

Figure 7 depicts the normalized spectral efficiency, which is defined as the ratio between the achieved

Figure 5. Average throughput in NRT.

Figure 6. Throughput outage probability in NRT.

Figure 7. Normalized spectral efficiency in RT and NRT.

282 H. WANG ET AL.


modulation and the highest modulation, under different schemes. It can be seen that the MAX-SNR scheme achieves the highest spectral efficiency due to the reason that in the the MAX-SNR, the connection with the best CSI is selected for transmission. The proposed scheme can also achieve a relatively high spectral efficiency as it takes both the channel condition as well as the QoS constraints into account when scheduling packets. While the spectral efficiency in the PF scheme is relatively low compared to the MAX-SNR and the proposed schemes.

From the above figures, we can see that the performance of the proposed suboptimal scheme is close to the optimal scheme, but with considerably low computation complexity. We can also see that the OFDMA based scheduling algorithms outperform the OFDM/TDM based scheduling algorithm as expected.

This is because in OFDMA systems, we can not only exploit multiuser diversity in the time domain, but also in the frequency domain. 6. Conclusions and Future Work This paper addresses the problem of QoS scheduling and resource allocation for downlink OFDMA systems supporting both real-time (RT) and non-real-time (NRT) multimedia traffic. The proposed algorithm assigns a priority to each packet based on the extended EXP rule which tightly couples the PHY layer issue (instantaneous channel conditions) and MAC layer issue (QoS requirements) together. To reduce the computational complexity of a linear integer optimization problem, a suboptimal heuristic algorithm is proposed. Through systemlevel simulation, it is shown that the performance of the suboptimal algorithm is slightly different from the optimal algorithm, and both the optimal and suboptimal algorithms outperform the conventional OFDMA scheduling algorithms in terms of high spectral efficiency and better QoS satisfaction. It is also shown that OFDMA based scheduling algorithms outperform the OFDM/TDM based scheduling algorithm due to an added dimension of multiuser diversity in frequency domain in OFDMA systems.

Base stations are usually equipped with multiple transmit antennas. Hence, space-division multiple access in the form of linear beam-forming provides additional degrees of freedom for user scheduling. Regarding future work, the proposed algorithm could be extended to this more general setup, wherein the radio resource is partitioned in both time-frequency domain and space domain.

Appendix I: Flowchart of the Suboptimal Algorithm.

The diagram of the proposed suboptimal heuristic algorithm is shown in Figure 8.

Figure 8. Flowchart of the proposed suboptimal scheduling algorithm.

7. References [1] S. H. Ali, K. D. Lee, and V. C. M. Leung, “Dynamic

resource allocation in OFDMA wireless metropolitan area networks,” IEEE Wireless Communications, Vol. 14, No. 1, pp. 6–13, 2007.

[2] Y. J. Zhang and K. B. Letaief, “Energy-efficient MAC-PHY resource management with guaranteed QoS in wireless OFDM networks,” ICC 2005, Vol. 5, pp. 3127–3131, May 2005.

[3] Y. J. Zhang and K. B. Letaief, “Adaptive resource allocation and scheduling for multiuser packet-based OFDM networks,” ICC 2004, Vol. 5, pp. 2949–2953, June 2004.

[4] A. Todini, M. Moretti, A. Valletta, and A. Baiocchi, “A modular cross-layer scheduling and resource allocation architecture for OFDMA systems,” GLOBECOM 2006.

[5] X. Zhang, E. Zhou, R. S. Zhu, S. M. Liu, and W. B. Wang, “Adaptive multiuser radio resource allocation for OFDMA systems,” GLOBECOM 2005, Vol. 6, December 2005.

[6] X. Zhang and W. B. Wang, “Multiuser frequency-time domain radio resource allocation in downlink OFDM systems: Capacity analysis and scheduling methods,” Computers and Electrical Engineering, Vol. 32, No. 1–3, pp. 118–134, 2006.



[7] S. S. Jeong, D. G. Jeong, and W. S. Jeon, “Cross-layer design of packet scheduling and resource allocation in OFDMA wireless multimedia networks,” VTC 2006, Vol. 1, pp. 309–313, 2006.

[8] M. Bohge, J. Gross, M. Meyer, A. Wolisz, and T. U. Berlin, “Dynamic resource allocation in OFDM systems: an overview of cross-layer optimization principles and techniques,” IEEE Network, Vol. 21, No. 1, pp. 53–59, 2007.

[9] S. Ryu, B. Ryu, H. Seo, and M. Shin, “Urgency and efficiency based packet scheduling algorithm for OFDMA wireless system,” ICC 2005, Vol. 5, pp. 2779–2785, May 2005.

[10] C. F. Tsai, C. J. Chang, F. C. Ren, and C. M. Yen, “Adaptive radio resource allocation for downlink OFDMA/SDMA systems,” ICC 2007, pp. 5683–5688, June 2007.

[11] K. F. Ahmed and M. F. Khaled, “Opportunistic scheduling of delay sensitive traffic in OFDMA-based wireless networks,” WoWMoM 2006, Vol. 2006, pp. 279–288, June 2006.

[12] B. Rong, Y. Qian, and K. J. Lu, “Integrated downlink resource management for multiservice WiMAX networks,” IEEE Transactions on Mobile Computing, Vol. 6, Issue. 6, pp. 621–632, 2007.

[13] Y. M. Ki, E. S. Kim, S. I. Woo, and D. K. Kim, “Downlink packet scheduling with minimum throughput guarantee in TDD-OFDMA cellular network,” Lecture Notes in Computer Science, Vol. 3462, pp. 623–633, 2005.

[14] Y. M. Ki and D. K. Kim, “Packet scheduling algorithms for throughput fairness and coverage enhancement in

TDD-OFDMA downlink network,” IEICE - Transactions on Communications, Vol. E88-B, No. 11, pp. 4402–4405, 2005.

[15] S. Shakkottai and A. L. Stolyar, “Scheduling algorithms for a mixture of real-time and non-real-time data in HDR,” Proceedings of International Teletraffic Congress (ITC), 2001.

[16] C. Hoymann, “Analysis and performance evaluation of the OFDM-based metropolitan area networks IEEE 802.16,” Computer Networks, Vol. 49, No. 3, pp. 341–363, 2005.

[17] C. Cicconetti, L. Lenzini, E. Mingozzi, and C. Eklund, “Quality of service support in IEEE 802.16 networks,” IEEE Network, Vol. 20, No. 2, pp. 50–55, 2006.

[18] D. H. Kim, H. R. Byung, and C. G. Kang, “Packet scheduling algorithm considering a minimum bit rate for non-real-time traffic in an OFDMA/FDD-based mobile internet access system,” ETRI Journal, Vol. 26, No. 1, pp. 48–52, 2004

[19] H. Wang, “Priority-based resource allocation for RT and NRT traffics in OFDMA systems,” The 3rd IEEE International Conference on Wireless Communications, Networking and Mobile Computing (IEEE WiCOM), Vol. 1, pp. 791–794, 2007.

[20] IEEE 802.16.3c-01/29r4, “Channel models for fixed wireless applications,” IEEE 802.16 Broadband Wireless Access Working Group, July 2001.

[21] T. K. Sarkar, Z. Ji, K. Kim, A. Medouri, and M. Salazar-Palma, “A survey of various propagation models for mobile communication,” IEEE Antennas and Propagation Magazine, Vol. 45, No. 3, pp. 51–82, 2003.

International Journal of · Email: [email protected], [email protected]...

Documents

Transcript of International Journal of · Email: [email protected], [email protected]...