Robust Wireless Video Transmission Employing Byte-aligned Variable-length Turbo Code

Robust Wireless Video Transmission Employing Byte-aligned Variable-length Turbo Code ChangWoo Lee* and JongWon Kim** * Department of Computer and Electronic Engineering, The Catholic University of Korea 43-1 Yoo 2-dong, Wonmi-gu, Puchon City, Kyunggi-do 420-743, Korea **Department of Information & Communication, Kwang-Ju Institute of Science & Technology 1 Oryong-Dong, Bu-Gu, KwangJu, 500-712, Korea. ABSTRACT Video transmission over the multi-path fading wireless channel has to overcome the inherent vulnerability of compressed video to the channel errors. To effectively prevent the corruption of video stream and its propagation in spatial and temporal domain, proactive error controls are widely being deployed. Among possible candidates, turbo code is nown to exhibit superior error correction performance over fading channel. Ordinary turbo codes, however, are not suitable to support the variable-size segment of the video stream. A version of turbo code, byte-aligned variablelength turbo code, is thus proposed and applied for the robust video transmission system. Protection performance of the proposed turbo code is evaluated by applying it to GOB-based variable-size ITU-T H.263 video pacets, where the protection level is controlled based on the joint source-channel criteria. The resulting performance comparison with the conventional RCPC code clearly demonstrates the possibility of the proposed approach for the time-varying correlated Rayleigh-fading channel. Keywords: Wireless video transmission, time-varying wireless channel, joint source-channel rate control, proactive error control, byte-aligned variable-length turbo code 1. INTRODUCTION The explosive growth of 3G wireless communication system has spared the increased demand for new and exciting information services based on the reliable transmission of continuous media. However, because of the fundamental limitations due to power, available spectrum, mobility, and fading, the quality of continuous media transmitted over time-varying wireless channel suffers heavily from the unstable and error-prone wireless channel [1]. Especially, wireless video transmission over the error-prone multi-path fading channel has to overcome the inherent sensitivity of compressed video to the channel errors and their propagation in spatial and temporal domain. Source-side schemes such as error-resilient encoding and error concealment are adopted in the ITU-T H.263 and ISO/IEC MPEG- 4 to address this weaness. From the channel side, the fluctuating wireless channel error on the continuous media streams is controlled at a suitable level with bandwidth, delay, and power-consumption constraint. As one of the promising error-control solutions, proactive FEC techniques have been widely proposed despite of their overhead channel rates. In case of wireless video, the mobile version of ITU-T H.324, H.324M, defines two inds of FEC codes, RCPC (rate-compatible punctured convolutional) code and shortened RS (Reed-Solomon) code in the H.223 multiplexing Annex C and Annex D, respectively [2,3]. The resulting FEC-based protection has the RC (ratecompatible) characteristic so that the error correction capability is increased gradually. Another potential candidate for FEC error control, an iterative decoding based turbo code, has attracted lots of interests due to its superior performance [4]. Turbo code, composed of encoder and interleaver, depends its error correction performance on the type and the size of interleaver. In addition, a rate-compatible version of turbo, namely RCPT (rate-compatible punctured turbo) code, has been proposed [5]. However, since the bloc size of turbo code depends on the adopted interleaver, its usage is * Correspondence: Email: lcw@www.cu.ac.r

restricted to the fixed size blocs (i.e., fixed size segment of video stream) and, as a result, it is hard to support the variable size segment of the video stream (e.g., GOB- or slice-based) without fragmentation or stuffing. Prior wors on turbo-coded wireless video have adopted this fixed bloc turbo code, albeit in tie with the burst-by-burst modulation adaptation [6]. In order to overcome this limitation of turbo code while preserving its superior performance, it is necessary to modify the interleaver to allow the bloc size variation per each bloc. The widely utilized random interleaver, nown to be the most efficient, is however difficult to be modified since it does not exhibit any ind of regularity. Thus, in this paper, a byte-aligned variable size turbo code is developed utilizing a different type of interleaver denoted as JPL interleaver. The JPL interleaver is nown to sustain the interleaving performance with a special regularity suitable for variable size realization [7]. With the additional modification to the JPL interleaving, we designed a byte-aligned variable size turbo code with the desired RC (rate-compatible) characteristic. The adaptive protection capability of the proposed FEC scheme is evaluated with a special attention to the dynamic and coordinated control of both channel and source rates. With the proposed RCPT, the source video stream, represented in series of compressed segments of GOB or slice, may be prioritized according to the layer it belongs and the strength of pacet loss/delay impact. Then, each priority group can be differently protected to jointly match the layered priority and channel condition. For these cases, we analyze and compare the proactive protection performance of turbo codes for the wireless ITU-T H.263 video. The adopted FEC codes, such as the RCPC code and the proposed RCPT code, are evaluated under both AWGN (additive white Gaussian noise) and time-varying correlated Rayleighfading wireless channels. That is, a simulation-based direct wireless lin is established and H.223-multiplexing transmission scenario is emulated for evaluation. The proactive protection is then performed for each H.223 pacet based on the joint source-channel criteria, where the source and channel coding rates are jointly optimized. The rest of paper is organized as follows. Section 2 describes the outline of the proposed video transmission system. Both RCPC and RCPT codes are introduced in Section 3, where detailed explanation on the proposed bytealigned variable length turbo is covered. Simulation results are shown in Section 4, where concluding remars are also given. 2. VIDEO TRANSMISSION SYSTEM ITU-T Recommendation H.324 describes terminal systems for low bit-rate multimedia communication, which may transmit real-time voice, video, and data [8]. Fig. 1 illustrates the major components of an H.324 terminal. For H.324M, a wireless interface with a wide range of bit rates can be used instead of V.34 to transmit the multimedia streams. The H.223 multiplexer interleaves video, audio, data and control streams into a single bit stream, and allows highly dynamic allocation of bandwidth to different channels. It consists of a lower multiplex layer, which actually mixes the different media streams, and a set of adaptation layers above. However, base mode of H.223 is not robust against errors since it is designed to wor in the low error-rate channel. To provide an error resiliency for the error- H.324 multimedia terminal Video I/O equipment Video codec H.263 Audio I/O equipment User data applications Audio codec G.723 Data protocols V.14, LAPM, etc. Receive path delay Multiplex/ demultiplex H.223 Modem V.34/V.8 System control Control protocol H.245 SRP/LAPM procedures Modem control V.25ter Fig. 1: ITU-T H.324 multimedia terminal.

prone mobile environment, additional logical framing, sequence numbering, error detection, and error correction have been defined in H.223 Annexes. Especially in H.223-Annex C and Annex D, RCPC and shortened-rs error correction code are defined [2,3]. Fig. 2 illustrates the procedure of applying RCPC code [2]. AL-SDU (adaptation layer-service data unit) provided by AL user is RCPC-encoded with a CRC (cyclic redundancy code) and a TB (tail bits). Since the code rate of the RCPC code is 1/4, we should puncture the parity bits to meet the system bandwidth. After adding the optional CF (control field), AL-PDU (adaptation layer-protocol data unit) is conveyed to the lower multiplex layer, forming MUX- SDU. At the decoder side, the boundaries of MUX-SDU s are detected using flags. Then, taing the channel code-rate into consideration, the length of each AL-SDU is determined. If we want to replace RCPC code with RCPT code, we just need to encode AL-SDU with RCPT code in the similar way (except for the variable-size issue). Shortly in Section 3, we will discuss the detailed characteristics of RCPC and RCPT codes. The resulting wireless video delivery framewor simulating ITU-T H.324M environment is depicted in Fig. 3. At the sender, H.263 video with simple error resilience and compression efficiency options is currently adopted. A Bitstream Application Layer AL-SDU AL-SDU AL-SDU AL-SDU* CRC TB convolutional encoding convolutional decoding AL-SDU* CRC TB Adaptation Layer transmitting punctured data CF AL-PDU Payload inserting punctured data from (re)-transmission AL-PDU MUX-SDU Multiplex Layer Fig. 2: Applying RCPC code in the H.223 layer. Video Encoder (ITU-T H.263) H.223 MUX for Mobile Channel Modulation Wireless Transmission over Fading Channel Channel Demodulation H.223 DE- MUX Error Resilient Video Decoder Layered Priority (optional) Feedbac with delay for update retransmission (optional) Clean Pacet Corrupted Pacet Fatal Pacet Fig. 3: Proposed wireless video transmission system framewor.

variable-size pacet (i.e., segment) is generated for every GOB that contains a fixed number of macroblocs (MB s). Each pacet is then multiplexed based on the simplified H.223 Annex C. In order to support the variable size nature of the video stream (e.g., GOB or slice) without fragmentation or stuffing, the size of AL-SDU should be variable. Fig. 4 shows the respective average size of GOB segments per each frame for Foreman, Carphone, and Claire sequences at bit rate of 64bps. The cyclic redundancy code (CRC) is calculated to chec its payload and appended for error detection. Based on the layered priority and the pacet size along with the channel state information, unequal error protection (UEP) can be conducted with the chosen RCPC or RCPT FEC codes. However, the layered priority has not been incorporated for the UEP at current stage, since we are mainly comparing the performance between RCPC and byte-aligned variable-length RCPT. Finally, the resulting channel pacet is modulated and transmitted to the underlying wireless channel. At the receiver, the received signal is decoded by the ML (maximum lielihood)-based scheme. The partial update re-transmission capability of H.223 Annex C RCPC is not utilized at current stage. Decoded pacets after the de-multiplexing stage are classified into three types: clean, corrupted (CRC chec failure) and fatal (unrecoverable error in the header, the synchronization or the control fields of multiplexing pacet). Finally, to provide the end-to-end performance in both subjective and objective measures, a decoder capable of handling the corrupted video stream is used for video decoding. 300 250 Foreman Carphone Claire Average pacet size (# of byte) 200 150 100 50 0 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94 97 100 Decoded frame # Fig. 4: Average sizes of GOB. 3. RCPC CODE AND BYTE-ALIGNED VARIABLE-LENGTH RCPT CODE 3.1 RCPC code and RCPT code The RCPC channel encoder defined in H.223-Annex C is based on a systematic recursive convolutional (SRC) encoder with rate R=1/4 [2]. RCPC is a family of convolutional codes derived from a mother code rate (1/4 in this case), a set of generating polynomials with memory, and a puncture table. In Fig. 5, bloc diagram of the RCPC encoder is depicted, where puncturing of the SRC encoder output allows different channel rates starting from 1/4 up to 1. Due to the flexibility in changing the code ratio, RCPC has been widely used in UEP FEC and hybrid ARQ/FEC protection. That is, by changing the puncture table of Table 1 to obtain different ratios, the level of protection can be altered. From Table 1, you can verify that, for all rates, the puncturing at a certain rate includes the puncturing of every bit of all lower rates plus additional bit(s), maing the code rate compatible. Another potential candidate for error correction, an iterative decoding based turbo code, has attracted lots of

interests due to its superior performance close to theoretical limits [4]. Fig. 6 shows the turbo encoder, composed of RSC (recursive systematic convolutional) encoder and interleaver. The optimum RSC encoder with 16 states, which minimizes the bit error probability by maximizing the effective free distance, is shown in Fig. 7 [9]. For a given RSC encoder, the error-correction performance of turbo code depends on the type and size of interleaver. Similar to the case of RCPC code, we can puncture the parity bits of turbo code to match the bandwidth constraint of underlying transmission channel. Table 2 shows the puncturing mechanism for RCPT code, where the channel code rate ranges from 8/24 to 8/8. V t (1) U t D D D D V t (2) V t (3) V t (4) Fig. 5: RCPC encoder. Data Upper RSC Encoder x p Parity 1 output p 2 Systematic output Puncturing Mechanism Interleaver Lower RSC Encoder Parity output A B Fig. 6: Turbo encoder. Systematic bit Input D D D D Parity bit Fig. 7: Optimum RSC encoder module for turbo encoder. Table 1: Puncturing tables for RCPC code (all values in hexadecimal representation). Rate γ 8/8 8/10 8/11 8/12 8/14 8/16 8/18 8/20 8/22 8/24 8/32 V t (1) FF FF FF FF FF FF FF FF FF FF FF V t (2) 00 88 A8 AA EE FF FF FF FF FF FF V t (3) 00 00 00 00 00 00 88 AA EE FF FF V t (4) 00 00 00 00 00 00 00 00 00 00 FF

Table 2: Puncturing tables for RCPT code (all values in hexadecimal representation). Rate γ 8/8 8/10 8/11 8/12 8/14 8/16 8/18 8/20 8/22 8/24 X FF FF FF FF FF FF FF FF FF FF P 1 00 80 88 88 AA AA EE FF FF FF P 2 00 40 40 44 44 55 55 55 77 FF As discussed in Section 2, if we want to apply the RCPT for each GOB of variable-length size, the RCPC should be byte-aligned with the synchronization codeword. It is thus necessary to modify the interleaver to support the bloc size variation per each bloc. However, the widely utilized random interleaver, nown to be the most efficient, is difficult to be converted since it does not exhibit any ind of regularity. On the other hand, we can easily mae the bytealigned RCPT code with the simple bloc interleaver. However, its performance is inferior due to the poor decorrelating property. Among the various types of interleavers, the JPL interleaver is nown to exhibit good interleaving performance with a special regularity suitable for variable size realization [7]. We have modified the JPL interleaver structure so as to match all sizes of byte-aligned GOB input. Fig. 8 illustrates the input-output relation for the modified JPL interleaver, where the repetitive tiling of regular patterns are shown. 4000 3500 3000 2500 2000 1500 1000 500 0 0 500 1000 1500 2000 2500 3000 3500 4000 Fig. 8: Input-output relation of the modified JPL interleaver (bloc size 4096). 3.2 Performance comparison of RCPC code and byte-aligned variable-length RCPT code The wireless channel model used to simulate the real-world situation includes the short-term fading, the long-term fading (i.e., shadowing), and the path loss. The multi-path phenomenon generates an amplitude variation of the transmitted channel signal that leads to the short-term fading. The path loss reflects the degradation of the signal strength over a distance. Fluctuation of the expected power level around the path-loss model is called the long-term fading effect and this long-term fading can be obtained by averaging the short-term fading SNR. In this paper, all simulations have been conducted assuming an urban micro-cell wireless environment [10]. The carrier frequency f c is set to 2 GHz and the mobile unit moves at a moderate velocity of 34 m/hr. The normalized Doppler frequency used to construct the fading is set to 10-3 while the short-term fading is modeled as the Rayleigh flat fading channel. For simplicity, it is assumed that power degradation due to the path loss is perfectly compensated by the power control method, and the SNR operational point is maintained throughout the wireless communication process. Thus, for a BPSK-modulated transmission, the fading model can be written as y a x n =, (1) where y is the received signal symbol at the decoder, a is the variation of Rayleigh flat fading amplitude, x is

the BPSK modulated signal, and n is the additive white Gaussian noise (AWGN), respectively. The resulting variation of the amplitude a under this urban micro-cell channel (normalized) is shown in Fig. 9. 2.5 2 1.5 1 0.5 0 1 1001 2001 3001 4001 5001 6001 7001 8001 9001 10001 11001 12001 13001 14001 15001 16001 17001 18001 19001 20001 21001 22001 23001 24001 25001 26001 27001 28001 29001 Fig. 9: Amplitude variation of time-varying correlated Rayleigh fading channel. For the RCPT code and the RCPC code, the BER performance has been computed using Monte Carlo method, as a function of variable pacet sizes. Fig. 10 and Fig. 11 illustrate BER and pacet-error rate performances of RCPC codes and proposed RCPT codes in the time-varying correlated Rayleigh-fading channel with a symbol to additive noise variance Es/No=15.0dB. As expected, RCPT code with the modified JPL interleaver shows superior performance to RCPC code. Moreover, the BER for the RCPC code decreases rapidly when we increase the pacet size. On the contrary, the pacet size has little effect on the BER performance for the RCPC code. In case of RCPT, this phenomenon continues even for pacet-error rates. That is, for RCPT code, pacet-error rates decreases as pacet-size increases. When we use the variable-size pacet to match the video segment (e.g., GOB), this means that GOB-pacet with intra-mb s can be better protected than GOB-pacet only with inter-mb s. Considering the error propagation, this can be another merit for using the proposed RCPT code. 1.00E-02 1.00E-03 BER 1.00E-04 1.00E-05 256 512 768 1024 1280 1536 1792 Pacet Size No Code RCPC (8/12) RCPC (8/14) RCPT (8/12) RCPT (8/14) Fig. 10: BER performance of RCPC code and RCPT code in the time-varying correlated Rayleigh-fading channel (Es/No: 15 db).

1.00E00 Pacet Error Rate 1.00E-01 1.00E-02 1.00E-03 256 512 768 1024 1280 1536 1792 Pacet Size No Code RCPC (8/12) RCPC (8/14) RCPT (8/12) RCPT (8/14) Fig. 11: Pacet-error rate performance comparison of RCPC code and RCPT code in the time-varying correlated Rayleigh-fading channel (Es/No: 15 db). 4. SIMULATION RESULTS AND CONCLUSION The performance comparison between RCPC code and byte-aligned variable-length RCPT code is continued here for the wireless video transmission. ITU-T H.263 video with simple error resilience and compression efficiency options is currently adopted. A variable-size pacet is generated for every GOB that contains a fixed number of MB s. Each pacet is then multiplexed based on the simplified H.223 Annex C. That is, each GOB-pacet of H.263 stream is used as a pacet of AL-SDU* illustrated in Fig. 2. This application-layer framed pacet, where compression-related segments are chosen as the delivery unit, has the potential of maing the error detection and concealment at the decoder more flexible and robust, hence improving the performance. Note that, if the pacet size is fixed, a channel pacet may contain irregular number of fragmented video segments and it complicates the decoder job. However, only simple error concealment based on the repetition is employed in our current evaluation, leaving the chance of future improvement. As mentioned in Section 2, the degree of error protection (i.e., the FEC code ratio) is controlled only by the pacet size and channel state. That is, UEP based on the layered priority and others has not been performed at current stage, since we want to focus on the performance comparison between RCPC and byte-aligned variable-length RCPT. Thus, the proactive protection is performed for each H.223 pacet based on the joint source-channel criteria, where the source and channel coding rates are jointly optimized as follows. To maintain a constant channel transmission rate R t, the available rate for the source should be reduced to Rs = γ R t, where γ [0,1] is a chosen channel code-ratio and the rate for the error correction becomes Rc ( = (1 γ ) Rt ). Assuming that the distortion D s and D c, which are induced by source compression and transmission error respectively, are independent of each other, the total distortion D can be expressed as D d ( c S s c c c γ, S ) = D ( R ( γ )) D ( R ( γ ), S ), (2) d where S c represents the channel status, which is determined by the signal-to-noise ratio in the AWGN or the time-

varying correlated Rayleigh fading channels [11,12]. The normalized Doppler frequency for the time-varying correlated Rayleigh fading channel is assumed to be 10-3. For minimizing the total distortion D d, it is therefore important to understand how much we can improve the reliability at the cost of source rate. In other words, we need to find the optimum ratio γ that shows the best performance. For the evaluation of the video transmission system, it is necessary to average the distortion over the whole sequence in order to provide a single figure of merit. The video quality is measured by the mean-squared-error (MSE) distortion averaged over all encoded frames. The overall MSE D d for a whole sequence after decoding is also equivalent to the common PSNR measure, in fact, the average PSNR, 2 PSNR = 10log10 (255 / Dd ). For the joint optimization of the compression ratio and the error correction code ratio, we have chosen a total channel rate of R t = 96 bps. Finally, the end-to-end performance of wireless video transmission is evaluated for the AWGN channel with Es/No=1.5dB and the time-varying correlated Rayleigh-fading channel with Es/No=15.0dB. Fig. 12 and Fig. 13 illustrate the comparison performances for the RCPC code and the proposed RCPT code. We have used Foreman and Miss America sequences, which have relatively higher and lower frequency components, respectively. The proposed RCPT performs well even at channel code-ratio γ=8/14, while the performance for the RCPC degrades rapidly with the increase of channel code-ratio γ. The channel code-ratio γ for RCPC code should be lower than 0.44 for reliable wireless video, which means too much redundant bits under low bit rates. With proper joint source-channel rate control, we can verify from these results that the proposed byte-aligned variable-length turbo code can enhance the robustness of underlying wireless video transmission. We have proposed the use of byte-aligned variable-length turbo code, especially RCPT, for the robust wireless video transmission. Byte-aligned RCPT code with the modified JPL interleaver has been designed. Then, RCPT based on this interleaver is evaluated by its performance under the AWGN and time-varying correlated Rayleigh-fading channels. The simulation results demonstrate the superior performance as well as the potential of the proposed scheme. In our future wors, we will further investigate the performance of the proposed system under more elaborated wireless video transmission framewor, where the UEP based on the layered priority is performed with the intelligent receiver with better error detection and concealment. 35 30 25 PSNR(dB) 20 15 10 5 0 3 2.75 2.5 2.25 2 1.75 1.5 1/γ RCPT(Fading Channel, Es/No:15.0dB) RCPT(AWGN Channel, Es/No:1.5dB) RCPC(Fading Channel, Es/No:15.0dB) RCPC(AWGN Channel, Es/No:1.5dB) Fig. 12: PSNR performance comparison of RCPC and RCPT codes (as a function of 1/γ, Foreman).

45 40 35 30 PSNR(dB) 25 20 15 10 5 0 3 2.75 2.5 2.25 2 1.75 1.5 1/γ RCPT(Fading Channel, Es/No:15.0dB) RCPT(AWGN Channel, Es/No:1.5dB) RCPC(Fading Channel, Es/No:15.0dB) RCPC(AWGN Channel, Es/No:1.5dB) Fig. 13: PSNR performance comparison of RCPC and RCPT codes (as a function of 1/γ, Miss America). ACKNOWLEDGEMENTS This wor was supported by grant No. R02-2001-00908 from the Basic Research Program of the Korea Science & Engineering Foundation. REFERENCES 1. Y. Wang and Q. Zhu, Error control and concealment for video communication: A review, Proc. of the IEEE, vol. 86, pp. 974-997, May 1998. 2. ITU-T, Multiplexing protocol for low bitrate multimedia communication over highly error-prone channels, ITU-T recommendation H.223-Annex C, Dec. 1998. 3. ITU-T, Optional multiplexing protocol for low bitrate multimedia communication over highly error-prone channel, ITU-T recommendation H.223-Annex D, Dec. 1998. 4. C. Berrou, A. Glavieux and P. Thitimajshima, "Near Shannon limit error correcting coding and decoding: Turbo codes," in Proc. IEEE ICC '93, pp. 1064-1070, Geneva, Switzerland, May 1993. 5. S. A. Barbulescu, Iterative decoding of turbo codes and other concatenated codes, Ph.D. thesis, University of South Australia, Feb. 1996. 6. P. Cherriman, C. H. Wong, and L. Hanzo, Turbo- and BCH-coded wide-band burst-by-burst adaptive H.263- assisted wireless video telephony, IEEE Trans. on Circuits and Systems for Video Tech., vol. 10, pp. 1355-1363, Dec. 2000. 7. C. Heegard and S. B. Wicer, Turbo coding, Boston: Kluwer Academic Publisher, 1999. 8. N. Farber, B. Girod and J. Villasenor, Extension of ITU-T recommendation H.324 for error-resilient video transmission, IEEE Commun. Magazine, vol. 36, pp. 120-128, June 1998. 9. S. Benedetto and G. Montorsi, Design of parallel concatenated convolutional codes, IEEE Trans. on Commun., vol.

44, pp. 591-600, May 1996. 10. A. Anastasopoulos and K. M. Chugg, An efficient method for simulation of frequency selective isotropic Rayleigh fading, in Proc. Vehicular Technology Conference, pp. 2084-2088, Phoenix, AZ, May 1997. 11. K. Stuhlmuller, N. Farber, M Lin, and B. Girod, Analysis of video transmission over lossy channels, IEEE Journal on Selected Areas in Commun., vol. 18, pp. 1012-1032, June 2000. 12. M. Bystrom and J. W. Modestino, Combined source-channel coding schemes for video transmission over additive white Gaussian noise channel, IEEE Journal on Selected Areas in Commun., vol. 18, pp. 880-890, June 2000.