Performance of Hybrid ARQ on Dual-Branch Diversity Receiver in Rayleigh Fading Channel

Transcription

1 Performance of Hyrid ARQ on Dual-Branch Diversity Reiver in Rayleigh Fading Channel Ghaida A. AL-Suhail Computer Engineering Department University of Basrah Basrah, Iraq Astract Error control thniques in wireless links such as Automatic Repeat Request (ARQ) and Forward Error Corrtion (FEC) are still quite efftive approaches in the improving system throughput in particular in time varying channel environments. In this paper, a dual-ranch seltion diversity reiver of comining Seltive Repeat ARQ scheme and BCH channel coding (hyrid ARQ) in moile cellular networks like UMTS is proposed. The aim is to provide imal achievale throughput when a predefined channel code is employed under various Rayleigh fading channel conditions. Both spatially uncorrelated and correlated antenna ranches are investigated for identical and non-identical channels. The numerical results show that significant redundancy coding has a reasonale efft on the overall throughput performance. In contrary, the channel coding influences significantly on the lower range of operating channel SNR in the seltion comining reiver when identical and non-identical correlated antenna ranches are considered. I. INTRODUCTION UMTS (Universal Moile Tele-communication System) network ased on Wideand Code Division Access (WCDMA) provides a clear evolution path to 4G high speed packet access (HSPA). In such packet ased wireless communication system, the downlink has een developed for high data rate transmission up to 4 Mps (and.76 Mps in uplink) []. To achieve a high reliaility in such cellular systems, diversity thniques are popular used at several layers of the network stack to comat variaility in performance and provide reliaility over unreliale timevarying channels. When the channel performance is good, a higher information rate can e sent; and when it is ad, it needs to add more reliaility and sends at a lower information rate. The reliaility thniques may involve two categories () error control schemes such as Automatic Repeat request (ARQ) protocols [2] [3], Forward-Error-Corrtion (FEC) strategies, and hyrid ARQ that adds oth Error-Dettion code (ED) and FEC information to each message prior to transmission like soft-comining in UMTS [4] and incremental comining in HSPA using punctured /3 Turo code []; and (2) diversity transmission and reption schemes with different modes (e.g. time, frequency, space, etc.); that oth categories can provide more sophisticated throughput performance in time-varying channel environments [] [6]. Rodney A. Kennedy College of Engineering and Computer Science The Australian National University Canerra, Australia rodney.kennedy@anu.edu.au ARQ protocols generally can achieve a high reliaility when erroneous packet is detted using error dettion code such as a cyclic redundancy chk (CRC). A retransmission request is sent to the transmitter over a feedack channel until an error-free reption is oserved. More spifically, there are three asic ARQ protocols: stop-and-wait (SW), go-ack-n (GBN), and seltive-repeat (SR). Among all ARQ schemes, SR is reported to show the est throughput performance [7] and thus it is interesting to analyze and investigate its ehavior in order to have an upper limit on the throughput performance that any ARQ protocol can achieve in practice. FEC scheme is also widely used in communication to dett and corrt errors. It is proved to e efficient if the type of errors is known and errors do not exceed the imum corrtion capacity of FEC. For example, (, 493, 2) Bose-Chaudhuri Hochquengh (BCH) FEC can corrt 2 its of random errors per packet. However, the prolem of FEC that it cannot efficiently handle urst errors. To solve this prolem, some systems use interleaving thnique which may introduce a large delay, and such delay is not suitale for real-time video communication systems [8]. On the other hand, the comination of ARQ and FEC reduces the numer of retransmissions and yields three types of Hyrid ARQ schemes (HARQ I, HARQ II, and HARQ III) in which error corrtion followed y error dettion is applied every reived packet [9] [4]. For example, in HARQ I, when erroneous reived packets are discarded NACK is sent to transmitter, and the entire packet is then retransmitted. This version adds oth ED and FEC information to each packet prior to transmission. When the coded data lock is reived, the reiver first dodes the error-corrtion code. If the channel quality is good enough, all transmission errors should e corrtale, and the reiver can otain the corrt data lock. If the channel quality is ad, and not all transmission errors can e corrted, the reiver will dett this situation using the error-dettion code, then the reived coded data lock is discarded and a retransmission is requested y the reiver, similar to standard ARQ. In contrary, HARQ II does not discard erroneous reived packets and the retransmission contains incremental redundant its. The reiver then comines these redundant its with its of the previous transmission resulting in lower code rates //$26. 2 IEEE

2 Fig. A lock diagram of non-switched dual seltion diversity system in the presence of SR-ARQ scheme and BCH channel coding. Finally, in HARQIII, individually transmitted packets are self-dodale and packets are only comined after doding has een attempted on each of the individual packets. If the standard ARQ scheme has to e used over a moile radio channel, the time-varying multipath fading characteristics introduce a certain amount of correlation etween different ARQ packet (re)transmissions; that is, channel errors cannot e assumed to e independent from packet to packet. Therefore, many rent researches [2] [] [3][7] have dealt with such ARQ schemes in different approaches in order to increase and enhance the system throughput (quality of service) in several ways () y applying adaptive modulation format and coding (AMC) [], or (2) y varying the packet size [] (3) y adapting packet size in the dual ranch diversity reiver [2] or (4) y controlling, for example, the throughput efficiency over multipath lock fading channels when having a perft knowledge of the complex path gains [9]. On the other hand, diversity thnique is also a powerful tool in wireless communication systems used to mitigate fading effts [6]. The depth of the fades and/or the fade duration is reduced y supplying the reiver with multiple replicas of transmitted signal that have passed over independently fading channels [3] [4]. In general, there are three common types of linear diversity comining MRC, EGC, and SC normally employed in digital reivers over multipath fading channels. Seltion Comining (SC) is the least complicated one since it processes only one of the diversity ranches. Spifically, the cominer chooses the ranch with the highest signal-to-noise ratio (SNR) or equivalently with strongest signal assuming equal noise power among the ranches, however, in a practical system; the diversity ranches may have unequal average SNR due to different noise figures or feedline lengths. Therefore, it is important to assess the efft of the correlation factor and the unalance of average SNR (non-identical ranches) on the throughput performance of the SC diversity. Many schemes [3] [] [6] have een proposed to provide significant power gains over existing seltion diversity schemes in Rayleigh fading channels. In [], for example, has introduced that in very slow fading channels, the dual-ranch switched diversity scheme at the transmitter also can improve the throughput efficiency of the ARQ protocol significantly. More spifically, the diversity enefit otained from this scheme can reduce the delay in transmitting the data packet and this will e an attractive goal in many applications such as audio, image and real-time video streaming. In this paper, we develop a framework to provide efficient andwidth access (i.e., efficient throughput performance) over the downlink in UMTS network using SR-ARQ scheme (at the radio link) and BCH channel coding at the physical layer in over Rayleigh fading channel. We introduce the dualranch diversity reiver (a classical dual-ranch seltion reiver) which is so-called non-switched diversity (NSD). Both spatially uncorrelated and correlated antenna ranches are investigated for identical and non-identical channels using HARQ Type I. II. SYSTEM MODEL Let us consider Fig. as a UMTS communication system ased on Downlink Dedicated Channel (DL DCH) [6]. The system model consists of one transmit antenna at Node B and Radio Network Controller (RNC) and two reiver antennas at User Equipment (UE). In order to predict the link ehaviour, the transmitter has to e informed of past link errors in the forward link via a CSI feedack of the error status of the reived data. In case of the UMTS this is accomplished y the radio link control (RLC) acknowledged mode (AM). From the link error characteristics of the UMTS model, the estimation for future error proaility can e predicted. We also assume that perft CSI is availale at the reiver and the channel is time-varying and frequency flat fading. CSI information are sent ack to the BCH controller and modulator using an error free CSI feedack, with a CSI sensing delay D (round-trip delay of positive (ACK) or negative (NACK) acknowledgements at the radio-data link layer); whereas the sensing delay can e artificially reduced

3 y using prediction filters [7]. Fig. presents schematic lock diagram of the proposed dual seltion diversity system in the presence of the comination of SR-ARQ scheme and BCH channel coding. B. RLC/MAC/PHY Models In Fig. 2, since the PHY layer passes the transport lock to the MAC layer together with the error indication from the Cyclic Redundancy Chk (CRC) and/or channel coding, the output of the Transmission Time Interval (TTI) in PHY layer can e characterized y the overall proaility of transport lock error also called transport lock error rate (BLER) in this paper. The smallest TTI is ms which is the same as the radio frame length [6]. For example, PHY model generates a radio frame at a fixed period of ms (for 384 kps and spread factor (SF) of 8) and 2 ms (for 64 kps with SF=32) [8]. The RLC entity reives a PDCP PDU which comprises an IP packet of 2 ytes or an ACK of 4 ytes, and additionally PDCP header of yte. This PDCP PDU is segmented into multiple RLC PDUs of fixed sizes. Each of these PDUs fits into a transport lock in which a CRC is attached. A typical size of each RLC PDU is 4 ytes, ut in our simulation, we choose a typical imum transport lock (TB) size at the PHY layer equals RLC payload plus RLC and CRC headers and overhead parity its of BCH channel code is no more than its. III. PROPOSED APPROACH Fig. 2 Packetization for proposed HARQ over UMTS radio interface: - PDCP, 2- RLC segmentation, 3- RLC paload and MAC header addition, 4- CRC addition for pure ARQ, -Transport Block* concatenation and channel coding (FEC + CRC addition) for HARQ Type I 6- after rate matching and interleaving over one TTI. *Assume a minimum transport lock size equals RLC payload of 32 its and a imum equals its. A. SC Diversity At the transmitter, SR-ARQ controller and BCH controller are oth considered to provide error prottion on the transmitted data depending on the two feedacks of CSI and ACK/NACK. BCH controller is responsile for adding channel coding to transport lock (TB) at the physical layer; and if the BCH doder at the reiver fails in corrting the it errors in TB then SR-ARQ controller retransmits the entire transport lock (TB) once NACK signal is reived. At the reiver, a post-seltion comining, i.e., two individual demodulators re associated with each antenna, is considered to e followed y BCH doder which is used to notify the transmitter with NACK signal if BCH doding is failed to corrt the it errors. The dision process after the reivers (demodulators) considers a reption as corrt using firstly BCH doding then followed y the error dettion using the usual cyclic redundancy chk (CRC) its (as in HARQ I scheme), if it is corrtly reived over any of the diversity ranches. A corrt transport lock (TB) hence is chosen to the output. A. RLC SR- ARQ For standard ARQ or hyrid Type-I SR-ARQ with fixed modulation, the throughput efficiency can e defined as [9] RC LC SR p () T L r where (L-C)/L denotes the ratio of information it (RLC payload) to the total its in a transport lock (L), C is eing the CRC its, and RC can express the channel code rate. T r is the average numer of transmission attempts per lock. Assuming that the ARQ scheme retransmits a packet (lock) until the ACK of a successful reption is achieved y ( p), where p is eing a lock or packet error rate (BLER or PER). Hence SR of () can e defined as a function of the modulation mode in terms of p ( ) the channel it error rate (BER), packet length (L) in its including the numer of over head its (e.g., CRC), and reived channel SNR per it ( )[7] LC L SR p ( ) (2) L B. Adaptive RLC Wireless link throughput is severely affted y channel impairments such as shadowing, multi-path fading, and interference. To solve this prolem, two solutions of adaptive link layer thniques can e employed either y adapting the packet length (size) for given average SNR and link layer

4 parameters, to optimize throughput, or via optimizing link layer parameters such as symol rate and constellation size for fixed packet (lock) length for imal throughput [9]. Indeed, when the packet (lock) length is too large, packet error rate (PER) or (BLER) increases, and throughput in consequence is limited y frequent retransmissions. On the other hand, the smaller the packet means the larger the * overhead. Hence, the optimal packet length L which imizes the throughput can e derived from (2) as * C 2 4C L ( ) C (3) 2 2 ln( p ( )) Using (3) it is noticed that a much smaller lock size is more efficient under higher channel BER espially in real-time applications ause small locks have low BLER mean while larger locks make efficient use of the channel when the channel BER is low. C. BCH Channel Coding A channel coding is required for wireless communications to prott data from the errors which may result from noise, fading, and interference. For low SNR region of spread sptrum modulation, where each it is multiplied y a chip sequence and spread into L it times, or time division multiplexing, a common idea of non-extending a period time of the lock can e applied. This does not increase the energy per information it, and such variation is called namely an adaptive FEC. We now consider a lock code BCH FEC with redundancy of parity its adding to the lock, ut without extending the total lock length (L) (in its) to exceed a imum length ( L ). The lock error rate, BLER, then, with imum error capacityt can e expressed as [], t L i L i BLER p p i i (4) In our hyrid ARQ scheme, since the error capacity of BCH code is considered to e nine parity its per error it for a - its Transport Block (TB), then the imum throughput (i.e., transmission efficiency) can e calculated as, L HARQ PLRBCH () L where L L L C denotes the length of ARQ encoded lock, and C 9 t is the length of inclusive period of total parity its per lock. Note that L does not exceed its in RLC/PHY layer. For simplicity, we can rewrite (4) as, PLR BCH t i i p p i i (6) p is the it error rate defined also in (2) and PLR BCH is the residual lock error rate after error-corrtion at radio link layer. The goal is to otain t under predetermined p for imal throughput. IV. THE CHANNELS PERFORMANCE A. Identical Channels In SC-NSD cominer ased on a BCH SR-ARQ postreption scheme, the proaility of successfully reiving a lock after error corrtion is that of a successful reption over either of the diversity ranches, and this can e expressed y P BLER 2 (7), And consequently the efftive throughput can e computed as, L SC NSD, P, L (8) If the coherent BPSK modulation is used the average it error proaility of correlated Rayleigh fading channel with the average SNR equal to can e expressed in closed form as [4, Eq ], ( ) p (9),SC ( ) 2 2 2( ) where represents the average SNR of the first ranch for equal average ranch SNRs ( 2 ), and is the power correlation coefficient etween the estimated and actual fadings, and the values of denotes a measure of the quality of channel estimation in terms of time delay and the imum Doppler frequency shift (e.g., in land-moile communication systems). Spifically, the statistical properties of fading signals depend on the field component used y the antenna, the vehicular speed, and the carrier frequency as follows o d s J 2 f T () J o is a Bessel function of the zero-th order first kind. T denotes the symol time which is less than the coherence s time Tc of fading period, i.e. correlation etween two symols over channel in case of slow fading, and the Doppler spread f of the signal represents a function of the moile speed d and the carrier frequency (/). Note that the correlation properties of the fading process depend only on f. When f d T s T d s is small (e.g., <.), the process is much correlated ( slow fading); on the other hand, for large values of

5 f d T s (e.g., >.), successive samples of the channel are almost independent ( fast fading ). For high data rates (i.e., small T ), the fading process can typically e considered as s slowly varying, at least for the usual values of the carrier frequency (9-8 MHz) or (2 GHz) and for typical moile speeds. B. Non- Identical Channels The unalance in the average ranch SNR in the SC-NSD scheme modeled in Fig. a efftively leads to lower the overall throughput performance. The average BER of the coherent BPSK can e written as [4, Eq. 9.26, Fig 9.34] 3 p, (4) i, Non SCl 2 i where 2 and 3. For, p can also e, Non SC numerically computed using (Eq. 9.66) in [4] to evaluate the efft of correlation coefficient and the unequal average SNR in the dual ranch SC-NSD cominer (Fig. a). V. NUMERICAL RESULTS In this stion, we focus on average SNR of identical and non-identical ranches, and the redundant amount of BCH channel coding for the achievale imal throughput. The channel model used in the simulations of the throughput performance is a flat Rayleigh fading channel. As a reference, we conducted the numerical results using a typical set of parameters as: the width of the Doppler sptrum, which is determined y the carrier frequency 2GHz and the vehicular speed v, the numer of the CRC parity its is 6 at the datalink layer, and the (raw) or reference it rate at the physical layer is 64kit/s as in Dedicated channel DCH UMTS standards. We set the imum packet size L to e no greater than it for oth the optimal throughput without BCH coding and the imal throughput under various values of BCH code that can e added to the RLC PDU (payload). BLER is assumed independent for the BPSK coherent demodulation scheme. A. Performance Evaluation In this study, we evaluate the throughput of HARQ Type I in the two types of diversity schemes descried in Stion II. The results are conducted using Matla programming to verify the system performance for correlated and uncorrelated ranches when alance and unalance of average channel SNR are considered in SC scheme. We first consider a fixed small packet size of it including 6 CRC it. To provide highest performance, the optimal throughput versus the corresponding optimal average SNR is illustrated in Fig. 3. * Achieving L ( ) in (3) and setting the upper ound of lock length to its in the region of higher values of SNRs, the Optimal throughput Identical SC Cominer =. =. =.7 =. 2 3 Average channel SNR/it (db) Fig. 3 Comparison of optimal system performance over Rayleigh fading channel. Assume L = its, CRC-6, and no channel coding. optimal performance in terms of the corresponding it error rate can e simply evaluated. It is found that there exists a significant channel gain which leads to the imum upper throughput in particular when correlation factor of the later scheme tends to e. or less.. In this case, throughput outcomes nearly 2%-4% channel gain compared to the case of equals.. It means that the power correlation coefficient will efftively degrade the system efficiency as far as this coefficient tends to e close to. Fig. 4 reveals the efft of BCH channel coding on the throughput performance in the identical ranches of SC diversity reiver for uncorrelated and correlated channels. HARQ Type I scheme illustrates various error-corrtion its required for three values of correlation coefficient ( ) of,. and.. The results compare the optimal throughput (at optimal lock size and no BCH) and the non-optimal throughputs otained at various values of error-corrtion conditions of BCH code. It is shown two findings: (i) optimal throughput outperforms non-optimal throughput of the various error-corrtion its in the lower range of channel SNR nearly elow (db); and this throughout can gradually achieve the higher level up to.968 at the higher values of SNR. On the other hand, for uncorrelated channel (Fig. 4a) the imal throughput of a Hyrid ARQ scheme can e oserved when error-corrtion its equal 9 in order to improve the achievale throughput at the lower range of SNR values (3dB- 6dB), i.e., higher it error rates. Then throughput eyond this range will e constant at 8%. Moreover, the correlation coefficient (Fig. 4 and Fig.4 c) degrades the throughput compared to the uncorrelated channel in Fig. 4a. It is also noticed in all three figures that there is a clear degradation in the imal throughput of HARQ compared to the optimal standard ARQ at high SNR region. This is due to the redundancy of parity its which are added to the RLC payload

6 lock without extending the total lock length to exceed a imum length ( L ), explained in Stion III. As a result, we conclude that the proposed BCH scheme eventually does not introduce any extra improvement in the throughput performance due to the channel coding overhead (redundancy code) espially in the high SNR regions. Moreover, for low SNR regions we need to provide BCH parity its to e not exceed 89 its to maintain the validity of our proposed BCH scheme (i.e., to not exceed a minimum RLC payload size of 32 its in Fig. 2). On the other hand, Fig. illustrates the efft of two factors: () the unalance of the average SNR in a dual-ranch of SC- NSD cominer, and (2) the correlation coefficient ( ) on the throughput performance in pure SR-ARQ and in proposed HARQ I ased BCH coding. When 2 and the correlation increases, a significant degradation in the system performance is oserved. In oth ARQs, uncorrelated channels outperform the correlated ones of equals.. Spifically, throughput of a HARQ I ased BCH attains to e nearly constant (nearly 8%) for the higher range of SNR compared to optimal throughput of pure SR-ARQ scheme. Fig. 6 explains a performance comparison etween a pure (standard) SR-ARQ scheme and a proposed HARQ Type I in identical ranches of SC for a predefined average channel SNR of db. It is clearly shown a significant efft of BCH channel coding on the system throughput versus the client moile speed. The improvement in throughput performance increases rapidly to achieve 72% once BCH code is increased using proposed HARQ scheme compared to optimal throughput of a pure ARQ, and then it can corrt 4 error its of the original RLC payload. Note that the original size of RLC payload must e no less than 32 its in our simulations. In contrary, a constant throughput is also oserved over the range of user moility (km/h-2km/h) or the normalized Doppler frequency ( f T d s is no greater than.8 or fd 37 Hz) over Rayleigh fading channel. The reason is that omes very close to.997 or. according to our system settings. Fig. 7 displays a comparison example of imal throughput performance versus moile speed for two different values of average channel SNR of db and 2 db in identical ranches of SC cominer. The imal throughput (8%- 97%) can e achieved at different conditions of system settings. A higher SNR means a lower BCH coding required, and consequently, no need for channel coding when average SNR achieves 3 db. However, a high BLER (BER) over a fading channel leads a few parity its redundancy of channel coding required to e added to the original RLC payload to corrt only 2 error its when SNR equals 2 dbs (Fig. 8 ); and when SNR is db the BCH parity equals six times of nine PBs to achieve a imal throughput of 92% (Fig. 8a) =. Identical SC Diversity Cominer Optimal no BCH t= no BCH t= BCH t=3 BCH t=4 BCH. t=8 BCH 2 3 Average channel SNR/it =2 (db) =. (a) Identical SC Diversity Cominer Optimal no BCH t= no BCH t= BCH t=3 BCH t=4 BCH. t=8 BCH 2 3 Average channel SNR/it =2 (db) =. () Identical SC Diversity Cominer Optimal no BCH t= no BCH t= BCH t=3 BCH t=4 BCH. t=8 BCH 2 3 Average channel SNR/it =2 (db) (c) Fig. 4 vs. average SNR for various values of correlation coefficient and error-corrtion its of BCH coding. Assume L = its, CRC-6 in identical ranches of SC scheme.

7 Non-identical, =2 L= its (fixed), No BCH Non-identical, = 2 BCH (,484,3) =. 2 3 (a) =. 2 3 (db) =. 2 3 (c) =. (db) 2 3 Non-identical, = 2 Optimal L, No BCH =. =. Non-identical, = 2 BCH (,43,9) =. =. 2 3 () 2 (db) (d) (db) 2 3 Fig. vs. the average channel SNR of uncorrelated and correlated ranch of SC scheme ased HARQ Type I with non-identical ranches 2. (a), () Pure SR-ARQ with fixed lock and with optimal lock size and (c) (d) HARQ with BCH (,484,3) and BCH (,43,9)..9.7 Av. SNR= db Identical SC Diversity Cominer Optimal, No BCH t=4 BCH (Highest) t=8 BCH.9.7 Av. SNR= db Identical SC Diversity Cominer Optimal, No BCH t=4 BCH (Highest) t=8 BCH Moile Speed (km/h) (a) Normalized Doppler Frequency () Fig. 6 of SC scheme ased pure ARQ and HARQ Type I with identical ranches over Rayleigh fading channel. (a) Moile speed and () Normalized Doppler frequency.

8 Av. SNR= db Nonoptimal, no BCH Optimal t= BCH t=3 BCH (highest) Moile Speed (km/h).9.7. Av. SNR=2 db (a) Optimal, No BCH Fixed L= it, No BCH t= BCH (Higest) t=8 BCH Moile Speed (km/h) () Fig.7 of pure ARQ and HARQ Type I in identical ranches over Rayleigh fading channel. (a) Average SNR= (db), and () Average SNR=2 (db). As a result, it is noticed that an increase in predefined average SNR values (Fig.6 and Fig. 7) will provide a significant improvement in the throughput performance to achieve a imal value using standard SR-ARQ scheme or under different BCH channel coding using HARQ Type I scheme. Each average SNR gives a certain throughput under a spific channel coding condition (Fig.4c). VI. CONCLUSION In this paper, we have studied the performance of standard SR-ARQ and a proposed HARQ Type I ased BCH channel coding to evaluate throughput efficiency of dual-ranch noswitched seltion diversity reivers with oth identical and non-identical ranches when Rayleigh fading channels are present. The results show that the imal throughput of HARQ using a comination of SR-ARQ and BCH channel coding at the radio link layer outperforms the optimal throughput of pure SR-ARQ for low channel SNRs. Moreover, correlation coefficient and unalance of average SNR introduce a significant degradation in throughput performance compared to the alance of SNR in identical ranches. REFERENCES [] H. Holma and A. Toskala, WCDMA for UMTS, Radio Access For Third Generation Moile Communications, John Wiley & Sons, Ltd., 24. [2] J. Xiao, J. Qiu, and S. Cheno, A Joint Adaptive Packet Size and Modulation Scheme Comined with SR-ARQ over Correlated Fading Channels, IEEE, pp ,. [3] J. Yun, Analysis of Seltive Repeat ARQ Comined with Adaptive Modulation for Fading Channels, IEEE MILCOM 22 Proc., vol., pp. 7-74, 22. [4] J. Roman, F. Berens, M. Kirsch, S. Tanrikulu, Hyrid ARQ Schemes for Future Wireless Systems ased on MC-CDMA, 23. [] S. S. Chakraorty, M. Liinaharja and K. Ruttik, Diversity and packet comining in Rayleigh fading channels, IEE Proc. Commun, vol. 2, no.3, pp , June. [6] X. Cai and G. B. Giannakis, Performance Analysis of Comined Transmit Seltion Diversity and Reive Generalized Seltion Comining in Rayleigh Fading Channels, IEEE Trans on Wireless Commun., 24. [7] E. Cianca, A. D. Luise, M Ruggieri, and R. Prasad, Channel-Adaptive Thniques in Wireless Communications: An Overview. Wireless Commun. Mo. Compu., 2:799-83, 22. [8] T. Wang, H. Fang, and L. Chen, Low-Delay and Error-Roust Wireless Video Transmission for Video Communications, IEEE Trans. on Circuits and Systems for Video Thnology, vol. 2, no.2, pp. 49-8, D. 22. [9] M. W. El Bahri, H. Boujemaa, and M. Siala, Performance of Hyrid ARQ Schemes over Multipath Block Fading Channels, Proc. 3 rd Inter. Conf. : Sciences of Eltronic, Th. of Infor. And Tele., SETIT, March 27-3,. [] W. Junli, H. Xiaolin, Y. Changchum and Y. Guangxin, Variale Packet Size Adaptive Modulation SR-ARQ Scheme for Rayleigh Fading Channels, IEEE 24, pp , 24. [] B. Holter and G. E. Oien, Performance Analysis of a Rate-Adaptive Dual-Branch Switched Diversity System, IEEE on Trans. on Commun., vol. 6, no. 2, pp , D. 28. [2] G. AL-Suhail, T. Lamahewa, and R. Kennedy, Performance of Dual- Branch Diversity Reiver ased SR-ARQ in Rayleigh Fading Channel, IEEE 3 rd ICSPCS 29, Omaha, Neraska, USA, 29. [3] Y. G. Kim and S. Wu Kim, Optimal Seltion Comining for M-ary Signals in Frequency-Nonseltive Fading Channels, IEEE Trans. on Commun., vol. 3, no., pp , Jan.. [4] M. K. Simon and M.-S. Alouini, Digital Communication over Fading Channels, 2 nd edition, J. Wiley, New Jersey,. [] K. Weerackody and W. Zeng, ARQ Scheme with Switched Antenna Diversity and their Applications in JPEG Image Transmission, Proc. Glocom 9, pp. 9-99, Nov. 99. [6] A. Lo, G. Heijenk, and C. Bruma, Performance of TCP over UMTS Common and Dedicated Channels, Proc. IST Moile & Wireless Comm. Summit 23, -8Jun 23, Aveiro, Portugal, pp [7] G. Femenias, SR-ARQ for Adaptive Modulation Systems Comined with Seltion Transmit Diversity, IEEE Trans. On Commun., vol. 3, no. 6, pp , June. [8] W. Karner, O. Nemethova, M. Rupp, Link Error Analysis and Modeling for Video Streaming Cross-Layer Design in Moile Communication Networks, ETRI Journal, Vol. 29, No., pp. 69-9, Oct. 27. [9] T. Yoo, R. J. Lavery, A. Goldsmith and D. J. Goodman, Optimization Using Adaptive Thniques, Th Report, 24.