THE transmission of video over the wireless channel represents

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 56, NO. 3, MAY 2007 1223 On the Performance of Space Time Block-Coded MIMO Video Communications Shunan Lin, Member, IEEE, Andrej Stefanov, Member, IEEE, and Yao Wang, Fellow, IEEE Abstract The problem of efficient video communications over multiple-input multiple-output (MIMO) wireless systems is of great significance due to the high capacity of the multiple antenna system. The high data rates provided by the MIMO system can be traded off with diversity gain by using different channel-coding schemes. Also, by using different video source-coding methods, high compression gain can be traded off with the error resilience gain. One should jointly consider source coding and channel coding when designing a MIMO wireless video system. However, little is known so far about what combinations of channel-coding and source-coding methods have the best overall performance in a MIMO system. In this paper, by comparing the performances of several different typical combinations through both theoretical and simulation studies, we show that no single combination is the best for the entire range of channel conditions, but rather, different combinations may be best for a subrange. Index Terms Diversity, error resilience, joint source channel coding, layered coding (LC), multiple description coding (MDC), multiple input multiple output (MIMO), space time coding, video coding. I. INTRODUCTION THE transmission of video over the wireless channel represents a challenging issue since current wireless systems have strict bandwidth limitation and high transmission error rates due to the harsh transmission environment, including fading, multipath, shadowing, and noise. Next-generation wireless networks, such as wireless local area networks (LANs) and cellular systems, are designed to have higher data rates than current systems such that wireless data and multimedia (including video) services can be provided. An efficient way to achieve this goal is using multiple-input multiple-output (MIMO) systems. Information-theoretic results [1], [2] have demonstrated that the capacity of the MIMO system in the presence of Rayleigh fading improves linearly with the number of transmit antennas, as long as the number of receive antennas is greater than or equal to the number of transmit antennas. Consequently, MIMO systems have been widely accepted as the right direction in the design of future LANs and cellular Manuscript received May 26, 2004; revised February 20, 2005 and February 2, 2006. The review of this paper was coordinated by Dr. R. Heath. S. Lin was with the Department of Electrical and Computer Engineering, Polytechnic University, Brooklyn, NY 11201 USA. He is now with the Video Algorithm Group, Harmonic Inc., White Plains, NY 10601 USA (e-mail: slin@ vision.poly.edu). A. Stefanov and Y. Wang are with the Department of Electrical and Computer Engineering, Polytechnic University, Brooklyn, NY 11201 USA (e-mail: yao@vision.poly.edu; stefanov@duke.poly.edu). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TVT.2007.895473 systems. Therefore, video transmission over MIMO systems emerges as an important research topic in wireless systems design. In a classic wireless video transmission system, a video signal is first coded by a video encoder, whose primary function is to remove the redundancy within the video signal. Then, the compressed video bit stream is coded again by a channel coder to add some controlled amounts of redundancy to combat the transmission errors. The design of a MIMO video transmission system involves two components: 1) video source coder and 2) space time channel coder. Recently, several researchers have noted the necessity of adjusting the MIMO transmission scheme for the transmission of video. Zheng and Liu [3] studied how to combine layered source coding and the Bell Laboratories layered space time (BLAST) system. By employing the horizontal layered space time approach and zero-forcing detector, different receive diversities can be provided to different source layers, thus achieving unequal error protection (UEP). The authors also examined power allocation among the layers to minimize the total distortion. Another work by Zheng and Samardzija [4] considered the transport of H.263 coded video streams over a BLAST test bed. It showed that remarkable performance improvement can be achieved by adapting the number of transmit antennas and channel-coding rates. However, it did not investigate the optimal way to perform such an adaptation. Also, it did not investigate the influence of different sourcecoding options. On the other hand, several other researchers investigated the design of new source coders for MIMO systems with fixed channel-coding schemes. In [5], the authors considered the application of multiple description coding (MDC) using correlating transforms in MIMO systems and investigated the design of optimal transforms based on the Rayleigh fading channel model. More recently, Bahceci et al. investigated the performance of joint MDC and turbo coding for multiple antenna systems [6]. The performance of MDC over multiple antenna systems has also recently been investigated from an information-theoretical perspective by Effros et al. [7]. In this paper, we investigate the influence of two important characteristics on the overall performance: 1) the diversity level of the space time code and 2) the error-resilient ability of the video coder. For a MIMO video system, the selection of diversity levels and the selection of an error-resilient video coder with appropriate redundancy are not independent of each other. In general, the video data need to be lossy coded before transmission. The distortion caused by the lossy source encoder directly depends on the available bit rates. Although a space time coder with full diversity can provide the strongest 0018-9545/$25.00 2007 IEEE

1224 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 56, NO. 3, MAY 2007 Fig. 1. General MIMO video system architecture. error protection of the source data, it usually supports a lower bit rate, which means that the video encoder can only output low-quality pictures. On the other hand, a space time coder with maximum rate (minimum diversity) allows the highest bit rate for video coding, but its weak error correction ability may cause devastating effects in the received video quality. Hence, for video transmission, an intermediate tradeoff between rate and diversity may be more desirable. This paper is organized as follows: In Section II, we describe the MIMO video system architecture and channel model. In Section III, we study the performance of different sourcecoding and channel-coding combinations. In Section IV, we investigate the performance of practical MIMO video transmission by simulating the transmission using real video coders. Section V concludes this paper. II. SYSTEM ARCHITECTURE AND CHANNEL MODEL In this section, we present the architecture of a MIMO video system, including different source-coding and channel-coding options of interest. A. System Architecture and Channel Model A general MIMO video transmission system architecture is shown in Fig. 1, which employs L t antennas at the transmitter and L r antennas at the receiver. In such a system, a video signal 1 X(n), n =0, 1,...,N is first coded by a video encoder. The basic function of the video encoder is to remove redundancy from the input video signal. Usually, for transmission over error-prone channels, the video encoder is also an errorresilient encoder that inserts some redundancy intentionally 1 A discrete video signal consists of a series of pictures X(n), n =0, 1,...,N,wheren is the time order of the pictures. Each picture at time n further consists of I J pixels X(i, j, n), i =0, 1,...,I 1 and j =0, 1,...,J 1, where{i, j} is the spatial position of this pixel in the picture. Since almost all video coding processes are for picture-based encoding, to simplify the notation, here, we use picture level expression X(n) to represent a video signal. into the bit stream. Then, the compressed bit stream is fed into a space time encoder. The space time coder encodes the bit streams to protect them and exploit the high capacity of the MIMO system. The coded bit streams are passed through a serial to parallel converter, mapped to a particular signal constellation, and transmitted through multiple transmit antennas. At the receiver, the transmitted signals are received by multiple receive antennas and decoded by a space time decoder. The decoded bit stream is recovered into a video signal ˆX(n), n = 0, 1,... by a video decoder. The transmission is block (frame) based. One block in the channel coder corresponds to one (data) block from the source encoder. Usually, the block size is much smaller than the bit stream size of one video picture; thus, the bits from one picture have to be divided into several blocks. At the receiver, if there are residual channel errors in one block after channel decoding, the whole block is claimed as lost. 2 The video decoder needs to have the ability to deal with lost packets through both the redundancy inserted at the compressed bit stream and some postprocessing approaches. In the aforementioned MIMO system, at each time slot t,the output of the modulator is a signal c t,i that is transmitted using transmit antenna i, 1 i L t. All signals are transmitted simultaneously (each from a different transmit antenna), and all signals have the same transmission period T.At time t, the received signal by antenna j, which is denoted by r t,j, is given by L t r t,j = α i,j c t,i + η t,j (1) i=1 where the noise samples η t,j are modeled as independent samples of a zero-mean complex Gaussian random variable with variance N 0 /2 per dimension. The coefficient α i,j is the path 2 In this paper, we only consider block-based transmission, which divides the video bit streams into equal-sized data blocks and channel codes each data block into a transmission block. Because the term block is also used in video coding to refer to a group of pixels in a square area of one picture, we refer to a channel-coded data block as a packet (frame).

LIN et al.: PERFORMANCE OF SPACE TIME BLOCK-CODED MIMO VIDEO COMMUNICATIONS 1225 gain from transmit antenna i, 1 i L t to receive antenna j, 1 j L r. We assume frequency-nonselective quasi-static Rayleigh fading; hence, the path gains α i,j are modeled as samples of independent zero-mean complex Gaussian random variable with variance 0.5 per dimension. The path gains are constant during the transmission of one packet and vary independently from one packet to another. The signal constellation at each transmit antenna is normalized such that the average energy of the constellation is 1/L t ; hence, the total transmitted energy at each transmission interval is normalized to unity. We define the signal-to-noise ratio (SNR) as 1/N 0. III. PERFORMANCE ANALYSIS OF DIFFERENT SOURCE-CODING AND CHANNEL-CODING COMBINATIONS FOR INDEPENDENT IDENTICALLY DISTRIBUTED (I.I.D.) GAUSSIAN SOURCES In this section, we consider the performance of a MIMO system with different combinations of source-coding and channelcoding schemes. To investigate the end-to-end performance of the system, we use the union bound on the frame error probability to estimate the performance of the space time code and the rate-distortion function of the three source coders. We consider the case of an i.i.d. Gaussian sequence. Although some other sources with memory (e.g., Markov model) are perhaps a better representation for video signals, there are no solid endto-end rate-distortion models for such sources when they are compressed and transmitted over a channel with random errors. Therefore, we focus on an i.i.d. Gaussian source, which makes the theoretical performance analysis possible, since its ratedistortion functions for different source coders are known. Furthermore, although this is a simple model, it provides us with an insight into the performance of various source-coding and channel-coding options. The gained insight can then be utilized as a motivation for a more comprehensive study employing practical video coders. For practical video encoders, we consider two kinds of popular multistream encoders, i.e., layered coding (LC) and MDC, as special cases. We also consider a conventional singlelayer coder since this encoder is the most widely used one. For LC and MDC, we focus on the case of two substreams. A. Rate-Distortion Functions To study the rate-distortion functions, hereinafter, we use D i,j, i =0, 1 and j =0, 1 to denote the distortions for three different scenarios: when both substreams are received (D 11 ), when only one substream is received (D 10 or D 01 ), and when neither substream is received (D 00 ). For layer coding, since the importance of the two layers are different, we use D B+E to denote the scenario when both the base layer (BL) and the enhancement layer (EL) are received, and we use D B and D E to denote the cases when either layer is received. It is well known that when a sequence of i.i.d. Gaussian random variables with variance σ 2 are encoded at bit rate R using an optimal source coder, the distortion is [8] D 11 (R) =σ 2 2 2R. (2) We will refer to the coder achieving the aforementioned ratedistortion bound as the optimal single description coder (SDC), which aims at minimizing the distortion when there are no transmission errors given a fixed rate R. On the other hand, an MDC is designed to jointly reduce both D 11 and D 10 (or D 01 ). To achieve this goal, an MDC coder needs more bits than an SDC coder to obtain the same errorfree distortion D 11. We define those extra bits as redundancy ρ = R R, where R is the bit rate needed to achieve some D 11 by an MDC encoder, and R is that needed to achieve the same error-free distortion by an optimal SDC, i.e., D e (R )= D 11. When the aforementioned Gaussian source is coded into two balanced descriptions at bit rate R/2 for either one, the distortion when both descriptions are received is D 11 = σ 2 2 2(R ρ). (3) When only one description is received, the distortion can be obtained from Ozarow s bound [9], [10] ( 1 2 σ 2 (1 1 2 2ρ )+D 11 (1+ 1 2 2ρ ) ) ( if ρ R ρ 1 + log ) D 10 =D 01 = 2 1+2 2(R ρ) σ 2 (1 1 2 2ρ ) if ρ>r ρ 1 + log 2 (1+2 2(R ρ) ). (4) It is easy to see that D 11 and D 10 cannot be reduced at the same time. When the sequence is coded by an optimal layer coder into two layers, i.e., a BL with bit rate R B and an EL with bit rate R E, the distortions when both layers are received (D 11 ), when only the BL is received (D 10 ), and when only the EL is received (D 01 ) can be expressed as [11] D 11 = D B+E = σ 2 2 2(R B+R E ) D 10 = D B = σ 2 2 2R B D 01 = D E = σ 2. (5) B. Performance of Different Source-Coding and Channel-Coding Combinations To evaluate the performance of different source-coding and channel-coding combinations in a MIMO system, we assume that a sequence of i.i.d. Gaussian random variables with variance σ 2 =1.0 are coded using different source-coding methods and then transmitted through a MIMO system. We consider a MIMO system with three transmit and three receive antennas utilizing Khatri Rao space time block codes [12], which can achieve diversity levels d =3, d =2, and d =1. For sourcecoding options, we consider SDC, MDC, and LC. Three source-coding options and three space time codes with three different diversity levels are combined in seven ways: 1) SDC with diversity d =3; 2) MDC with diversity d =3; 3) SDC with diversity d =2; 4) MDC with diversity d =2; 5) SDC with diversity d =1; 6) MDC with diversity d =1; and 7) LC with UEP. The last case means that the BL and the EL are protected by different diversity levels. These combinations are based on the following consideration: The substreams of

1226 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 56, NO. 3, MAY 2007 SDC and MDC are equally important; therefore, we should use the same diversity to protect them. On the other hand, the importance of the substreams for LC is different; therefore, it is suitable to provide unequal protection for different substreams. Specifically, the diversity that is used for the BL is not lower than that used for the EL. Assume that the space time block code with diversity d =3 can support source rate R s. Then, d =2can support 1.5R s, and d =1can support rate 3R s. For LC with UEP case, the bit rates for the BL and the EL are adjusted according to the channel coders they use. For example, if space time code with diversity 3 is used for the BL and that with diversity 1 is used for the EL, the bit rate of the BL and the EL could be ar s and 3(1 a)r s, respectively, where a is any number between 0 and 1. The output bit streams from the source encoder are packetized. The packets are channel coded and sent out sequentially. The packet error rate is given by P l =1 (1 P f ) L (6) where L is the packet size, and P f is the union bound on the frame error probability for a specific space time block code [13]. The end-to-end distortion when using SDC is D =(1 P l )D 11 + P l σ 2. (7) For MDC, the end-to-end distortion is D =(1 P l ) 2 D 11 +(1 P l )P l D 10 +(1 P l )P l D 01 + P 2 l σ 2 and for LC, it is D =(1 P lb )(1 P le )D 11 +(1 P lb )P le D 10 + P lb σ 2 where D 11 and D 10 (D 01 ) are given in (2) (5) for different source-coding schemes, respectively. 3 Fig. 2 shows the performance of different source-coding and channel-coding combinations in a binary phase-shift keying transmission by using the aforementioned rate-distortion functions and the packet loss rate function (6). The packet length is equal to the block fading length, which is assumed here as 100 transmission time slots. The bit rate R s for diversity d =3 code is equal to 1 bit/sample. For each given channel s SNR, the optimal bit allocation, i.e., the one leading to the minimal total distortion between the BL and the EL for LC coding, and the optimal redundancy ρ for MDC are determined through exhaustive searches. 3 Although the aforementioned rate-distortion bound can only be achieved without constraint on coding length, the operational rate-distortion function of a practical coder usually shows the same exponential rate decay 2R at high bit rate. For example, the operational rate-distortion function of scalar quantizer is ( 3π/2)σ 2 2 2R [14]. Hence, the observations in this paper, which are based on relative comparison, are still valid for practical coders with limited coding length. (8) (9) Fig. 2. Performance of different source-coding and channel-coding combinations in a three transmit and three receive MIMO system, with different combinations of source coders SDC, MDMC, and LC and Khatri Rao space time block codes. In Fig. 2, we can have the following observations: 1) Channel coding has a dominant influence on the end-to-end performance of the system. When the channel s SNR is low, channel coding with full diversity provides the best performance. As the channel becomes better, channel coding with lower diversity becomes better, leading to SNRs at which the diversity level and the rate should be traded off. 2) Source coding also plays an important role in determining the end-to-end performance. For different source-coding methods, the SNRs at which a diversity-rate tradeoff is necessary are different. Finally, no single combination is the best over the whole range. Specifically, in the cases under study, MDC with a full-diversity space time block code is the best choice for SNRs below 8 db. In the SNR range between 8 and 10 db, LC with UEP offers a slight advantage over the other two cases. MDC with diversity 2 and LC with UEP achieve almost the same performance between 10 and 14 db, and LC with UEP is the best choice between 14 and 16 db. In the range between 16 and 20 db, full-rate MDC with d =1achieves the best performance. Full-rate MDC and SDC yield the same performance as LC with UEP beyond 20 db. IV. PERFORMANCE EVALUATION OF PRACTICAL VIDEO CODERS In the previous section, we studied the performance of several source-coding and channel-coding combinations, assuming that the source signal is i.i.d. Gaussian. In this section, we will focus on real video signals. As rate-distortion models for actual video coders are difficult to obtain, especially in the nonerror-free case, we present a comprehensive study based on simulation results when a video signal is encoded by real video encoders and transmitted over a simulated MIMO system. Again, we select a three transmit and three receive antenna system with Khatri Rao space time codes [12] to achieve diversity levels d =3, d =2, and d =1. The block fading length is set to 100 transmission slots. Three video coding techniques are considered, namely 1) the multiple description

LIN et al.: PERFORMANCE OF SPACE TIME BLOCK-CODED MIMO VIDEO COMMUNICATIONS 1227 TABLE I SETUP OF VIDEO ENCODERS FOR DIFFERENT DIVERSITY LEVELS. (a) BIT RATE (in KILOBITS PER SECOND) OF MDC AND SDC FOR DIFFERENT DIVERSITY LEVELS.(b)BIT RATE OF THE BL AND THE EL FOR DIFFERENT UEP APPROACHES motion compensation (MDMC) video coding [15], 2) the H.263 LC with UEP video coding, and 3) the H.263 baseline video coding [16], simulating SDC. We compare the performance of the aforementioned three video coders combined with three different diversity levels. The total bit rate, after source coding and channel coding, is 192 kb/s for all cases of interest. As presented in Table I, when MDC and H.263 baseline are used, the bit rates allocated to source coding are 192, 96, and 64 kb/s for diversity levels d =1, d =2, and d =3, respectively. For LC with UEP, the bit rates for the BL and the EL are adjusted according to the channel coding they use. For example, the BL is coded at 96a kb/s for d =2 code, and the EL is coded at 192(1 a) kb/s for d =1code, where 0 a 1. For MDMC coding, the sourcecoding redundancy is fixed, rather than optimized, based on the channel s SNR. 4 On the other hand, for LC with UEP, we use the optimal bit allocation between the BL and the EL for LC with UEP by searching through a set of a s. As the computational complexities of the channel coders for the three different diversity levels are similar, the computational complexities of the different combinations are determined by those of the video encoders. As compared with SDC, the major complexity increase of MDC comes from motion estimation for one more frame, which means that its complexity is around 30% more than SDC [17]. For LC, the complexity could range from around 35% to 100% more than SDC, depending on how motion estimation for EL frames is performed [17]. For all three video encoders, to improve the error resilience, a sync word is inserted in the header of each group of video macroblocks (GOB), which is a row of macroblocks (MB) in a video picture. Also, since more error resilience should be applied to the channel that is worse, 20% of the video MBs of each video picture are coded using predictive coding (intracoded) for d =1, 10% are intracoded for d =2, and 5% are intracoded for d =3. To observe the system performance for video signals with different characteristics, we selected two quarter common intermediate format, QCIF, (176 144 pixels/picture) video sequences: one sequence Foreman, which has medium object motions in the pictures, and a low-motion sequence Mother and Daughter. The first 100 pictures of both sequences are source coded at a picture rate of 10 pictures/s. The coded bit streams are then channel coded and transmitted. The data packet size is 100. At the receiver, the space time decoder performs maximum-likelihood decoding of the received signal. Selected by a channel simulator, a packet can either be decoded successfully or still have residual errors after decoding. For the former case, the packet is passed to the video decoder. Otherwise, for unsuccessful decoding, an all-zero packet is passed to the video decoder. The video decoder decodes the received packets and constructs the decoded video pictures. The distortion between the decoded video and the original video signal is measured as peak SNR (PSNR), 5 which is defined as ( ) 255 255 PSNR (in decibels) = 10 log 10. MSE Fig. 3 shows the average decoded video quality in terms of PSNR. The results lead to the following observations: 1) Channel coding has dominant influence on the overall performance for medium and high frame error probabilities. 2) The choice of the source-coding method also has significant influence on the overall system performance. For example, for the d =3 space time coded cases in both sequences, MDC has better performance than SDC when the channel s SNR is lower than 6 db. 3) No single option is the best for the entire range of channel conditions. In the cases under study, when the channel s SNR is low, MDC with full-diversity channel coding provides the best performance. As the SNR increases, the SDC coupled with d =2space time block code provides the best performance. Then, LC with UEP and optimal bit allocation outperforms the other two when the channel s SNR becomes higher. At last, when the channel quality becomes very good, SDC with the lowest diversity (or the highest rate) becomes the best choice. 4) Most observations from this set of simulation results are consistent with those from the theoretical performance study, although the influence of different sourcecoding methods is not exactly the same in the two studies, which is mainly due to the simplicity of the Gaussian ratedistortion model. For example, the H.263 layered coder incurs nonnegligible redundancy over the H.263 baseline coder, which is one of the reasons that the LC with UEP approach does not perform as well as that in the theoretical study presented in the previous section. Also, the redundancy in the MDMC coder is fixed to reduce computation cost in our implementation. With optimal allocation of redundancy based on the channel s SNR, the performance for MDC is likely to be better. In this figure, we can observe that the performance also depends on the properties 4 Although it is possible to use exhaustive search to determine the optimal encoder operating parameters and associated redundancy for a given channel s SNR, we did not pursue this direction because of the extensive search time that is involved. 5 Although there are some other metrics that consider human visual system, and therefore match human s subjective quality measurement better than PSNR, here, we still use PSNR since it is most widely used. Furthermore, we believe that changing to other metrics would not influence the paper s conclusions.

1228 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 56, NO. 3, MAY 2007 Fig. 4. Decoded images of the 69th frame in the sequence Mother and Daughter. (a) SDC. (b) MDC. (c) LC. V. C ONCLUSION In this paper, we investigated joint source channel coding for MIMO video transmission. Specifically, we studied the necessity to jointly select the diversity level of the space time code and the error resilience of the video coder. We compared the performance obtainable with different combinations of video source coding and channel coding through both theoretical and simulation studies. The video coding options considered include MDC, SDC, and LC. In terms of channel coding, we considered Khatri Rao space time block codes [12], which achieve diversity levels d =3, d =2, and d =1. We illustrated that to optimize the overall system performance, the selection of the diversity level of the space time code depends on the selection of the error-resilient source-coding scheme, and vice versa. Furthermore, our results show that the appropriate sourcecoding and channel-coding schemes depend on the channel environment. No single option is the best for the entire range of channel conditions. Hence, to obtain the best overall system performance, the source-coding and channel-coding parameters need to be chosen jointly and with regard to the channel quality. Fig. 3. Received video quality in PSNR in a three transmit and three receive MIMO system, with different combinations of video coders MDMC, LC, and H.263 baseline and Khatri Rao space time block codes. (a) Video sequence Foreman. (b) Video sequence Mother and Daughter. of the source signal. For example, for a video sequence with lower motion, it is easier to recover from errors because there is more similarity between adjacent frames; hence, the encoder with lower redundancy performs better than the high-motion video signal. Consequently, the SDC has relatively better performance in Fig. 3(b) than in Fig. 3(a). Fig. 4 shows one decoded image from three different sourcecoding and channel-coding combinations when the channel s SNR is 12 db: (a) SDC with diversity 2, (b) MDC with diversity 2, and (c) LC with UEP. In this figure, although all three images have some artifacts caused by transmission errors on the mother s hand, we can see that MDC has fewer artifacts than SDC and LC because it has more (source coder) error resilience. On the other hand, we can observe that in the parts without transmission errors, the SDC image has better visual quality than the other two. The reason is that it has minimum sourcecoding redundancy and, hence, minimum encoder distortion. REFERENCES [1] E. Telatar, Capacity of multi-antenna Gaussian channels, AT&T-Bell Labs., Murray Hill, NJ, Internal Tech. Memo., Jun. 1995. [2] G. J. Foschini, Jr. and M. J. Gans, On limits of wireless communication in a fading environment when using multiple antennas, Wirel. Pers. Commun., vol. 6, no. 3, pp. 311 335, Mar. 1998. [3] H. Zheng and K. J. R. Liu, Space time diversity for multimedia delivery over wireless channels, in Proc IEEE ISCAS, Geneva, Switzerland, May 2000, pp. 285 288. [4] H. Zheng and D. Samardzija, Performance evaluation of indoor wireless systems using BLAST testbed, in Proc. IEEE VTC, Rhodes Island, Greece, May 2001, pp. 905 909. [5] N. At and Y. Altunbasak, Multiple description coding for wireless channels with multiple antennas, in Proc. IEEE Globecom, San Antonio, TX, Nov. 2001, pp. 2040 2044. [6] I. Bahceci, Y. Altunbasak, and T. M. Duman, A turbo coded multiple description system for multiple antennas, in Proc. IEEE GLOBECOM, 2003, pp. 4011 4015. [7] M. Effros, R. Koetter, A. Goldsmith, and M. Medard, On source and channel codes for multiple inputs and outputs: Does multiple description beat space time? in Proc. IEEE Inf. Theory Workshop, Oct. 2004, pp. 324 329. [8] T. M. Cover and J. A. Thomas, Elements ofinformation Theory. Hoboken, NJ: Wiley, 1991. [9] V. K. Goyal, Multiple description coding: Compression meets the network, IEEE Signal Process. Mag., vol. 18, no. 5, pp. 74 93, Sep. 2001. [10] L. H. Ozarow, On a source coding problem with two channels and three receivers, Bell Syst. Tech. J., vol. 59, no. 10, pp. 1909 1921, Dec. 1980. [11] W. H. R. Equitz and T. Cover, Successive refinement of information, IEEE Trans. Inf. Theory, vol. 37, no. 2, pp. 269 275, Mar. 1991. [12] N. D. Sidiropoulos and R. S. Budampati, Khatri Rao space time codes, IEEE Trans. Signal Process., vol. 50, no. 10, pp. 2396 2407, Oct. 2002.

LIN et al.: PERFORMANCE OF SPACE TIME BLOCK-CODED MIMO VIDEO COMMUNICATIONS 1229 [13] E. G. Larsson and P. Stoica, Space Time Block Coding for Wireless Communications. Cambridge, U.K.: Cambridge Univ. Press, 2003. [14] R. M. Gray and D. L. Neuhoff, Quantization, IEEE Trans. Inf. Theory, vol. 44, no. 6, pp. 2325 2383, Oct. 1998. [15] Y. Wang and S. Lin, Error resilient video coding using multiple description motion compensation, IEEE Trans. Circuits Syst. Video Technol., vol. 12, no. 6, pp. 438 453, Jun. 2002. [16] ITU-T Recommendation H.263, Video Coding for Low Bit Rate Communication, 1998. [17] I. E. G. Richardson and Y. Zhao, Adaptive management of video encoder complexity, J. Real-Time Imaging, vol. 8, no. 4, pp. 291 301, 2002. Andrej Stefanov (M 01) received the B.S. degree in electrical engineering from Cyril and Methodius University, Skopje, Macedonia, in 1996 and the M.S. and Ph.D. degrees in electrical engineering from Arizona State University, Tempe, in 1998 and 2001, respectively. During the summer of 2000, he was with the Advanced Development Group, Hughes Network Systems, Germantown, MD. Since October 2001, he has been with the Department of Electrical and Computer Engineering, Polytechnic University, Brooklyn, NY, as an Assistant Professor. His current research interests are in communication theory, wireless and mobile communications, channel coding, and joint source-channel coding. Dr. Stefanov currently serves as an Editor for the IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS. He was the Technical Cochair of the 2006 IEEE International Workshop on Wireless ad hoc and Sensor Networks. He is a member of the IEEE Communications Society, IEEE Information Theory Society, and IEEE Vehicular Technology Society. He is the recipient of the IEEE Benelux Joint Chapter Best Paper Award at the IEEE Vehicular Technology Conference, Fall 1999, Amsterdam, The Netherlands. Shunan Lin (M 04) received the B.S. degree from the University of Science and Technology of China, Hefei, China, the M.S. degree in electrical engineering from the Institute of Automation, Chinese Academy of Sciences, Beijing, China, and the Ph.D. degree in electrical engineering from Polytechnic University, Brooklyn, NY. In 2000 and 2002, respectively, he was an Internship Student at Microsoft Research, Beijing, and Mitsubishi Electric Research Laboratory, Murray Hill, NJ. Since 2004, he has been with the Video Algorithm Group, Harmonic Inc., White Plains, NY. His research interests are in the fields of video coding, processing, and transmission. Dr. Lin is a corecipient of the 2004 IEEE Communications Society Leonard G. Abraham Prize in the field of communication systems. Yao Wang (M 90 SM 98 F 04) received the B.S. and M.S. degrees in electronic engineering from Tsinghua University, Beijing, China, in 1983 and 1985, respectively, and the Ph.D. degree in electrical and computer engineering from the University of California at Santa Barbara, in 1990. Since 1990, she has been with the Department of Electrical and Computer Engineering, Polytechnic University, Brooklyn, NY, where she is presently a Professor of Electrical and Computer Engineering. She was on sabbatical leave at Princeton University, Princeton, NJ, in 1998 and at Thomson Corporate Research, Princeton, in 2004 2005. She was a Consultant at AT&T Labs-Research, formerly AT&T Bell Laboratories, from 1992 to 2000. Her research areas include video communications, multimedia signal processing, and medical imaging. She is the lead author of a textbook entitled Video Processing and Communications and has published more than 150 papers in journals and conference proceedings. Dr. Wang has served as an Associate Editor for the IEEE TRANSACTIONS ON MULTIMEDIA and the IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY. She is the recipient of the New York City Mayor s Award for Excellence in Science and Technology in the Young Investigator Category in 2000. She was an elected Fellow of the IEEE in 2004 for her contributions to video processing and communications. She was a corecipient of the IEEE Communications Society Leonard G. Abraham Prize Paper Award in the field of communication systems in 2004.