MPEG-4 VIDEO TRANSMISSION IN THE 5GHZ BAND THROUGH AN ADAPTIVE OFDM WIRELESS SCHEME

Transcription

1 MPEG-4 VIDEO TRANSMISSION IN THE 5GHZ BAND THROUGH AN ADAPTIVE OFDM WIRELESS SCHEME 1 D. Dardari, M.G. Martini, M. Milantoni, and M. Chiani D.E.I.S., University of Bologna Viale Risorgimento 2, Bologna, Italy {ddardari,mgmartini,mmilantoni,mchiani}@deis.unibo.it Abstract - Future wireless video transmission systems will consider OFDM (Orthogonal Frequency Division Multiplexing) as basic modulation technique due to its robustness and low complexity implementation in the presence of frequency selective channels. Recently, adaptive bit loading techniques have been applied to OFDM showing good performance gains in cable transmission systems. In this paper adaptive loading techniques are applied to a HIPERLAN2-like wireless system at 5GHz for efficient multimedia traffic transmission. Moreover, the classical water-filling algorithm is extended to the multi-layer case. The semi-analytical results obtained show a large improvement respect to the non-adaptive case, also taking different channel state information up-date rates into account. The impact of this technique in terms of video quality is also evaluated for MPEG-4 video transmission. Keywords - OFDM, adaptive modulation, bit loading, HIPERLAN2, MPEG-4. I. INTRODUCTION One of the main goal in the near future of communication systems is the development of multimedia efficient data coding, compression and transmission techniques that permit real-time mobile communications. The market requires a lot of services ( transmission, Internet browsing, E-Commerce, Multimedia Message Service, Location-Based Service, etc.) that combine a multitude of heterogenous networks. Market studies show an enhancement of the content when it is made available wherever the user wants to consume it. Mobility becomes an added value. Examples of this are the migration from POTS to mobile telephony and from the dial-up Internet to cellular data. The same is expected to happen with broadband multimedia data currently delivered over DSL or cable. Future networks (4G) will add mobility to this content as an enhancement. In this context, the major challenge is the integration of different categories of networks and wireless local area networks (WLAN). Systems have to be adaptive, i.e. they have to react to changing quality conditions, like varying channel capacity. In high speed wireless data applications the orthogonal frequency division multiplexing (OFDM) modulation scheme has been considered, because of its relatively simple receiver structure compared to single carrier transmission in frequency selective fading channels. OFDM modulation is adopted by IEEE for the extension of the wireless LAN standard to the 5GHz band (IEEE802.11a), providing data rates up to 54Mb/s [1]. ETSI adopted the OFDM scheme for the High Performance LAN physical layer standard (HIPERLAN2) too [2]. Recently, adaptive bit loading techniques have been applied to OFDM, showing good performance gains in cable [3], [4], [5] and wireless [6] transmission systems. The choice of a multicarrier modulation has the advantage of allowing the transmitting power and bit rate of each sub-channel to be changed dynamically according to channel selectivity variations. If the channel is slowly time varying, the receiver can provide a Channel State Information (CSI) to the transmitter using a robust feedback channel. Based on the CSI, an adaptive transmission technique has the possibility to modify dynamically the parameters of the modulator in order to improve the performance [7]. The theoretical channel capacity can be approached by distributing the total transmitted energy according to the water-filling principle. In this work, the Margin Adaptive (MA) bit loading algorithm [8] has been considered and applied to the HIPERLANlike type 2 physical layer working on the 5GHz band. An extension of the water-filling technique to the multi-layer case, that we will call in the following AULA (adaptive unequal loading algorithm), is presented to perform Unequal Error Protection (UEP) of layered video sources. The performance evaluation in terms of packet error probability (PEP) and P SNR (peak signal-to-noise ratio) for a MPEG-4 video transmission for outdoor wireless data service is addressed. Moreover, the effect on the performance of different CSI up-date rates has been also investigated in the case of a terminal speed of 3Km/s. Results show that good performance can be achieved if the CSI is up-dated at least every 1000 frames. II. REFERENCE TRANSMISSION SYSTEM In this paper we consider the HIPERLAN2 (or IEEE802.11a) physical layer standard as the reference transmission system [1], [2]. It employs the OFDM modulation technique as it is an efficient way to transmit in frequency selective channels. The OFDM scheme allows the transmission of N u parallel complex symbols A n (n = 1, 2,..N u ), that belong to a M n points constellation set, into N u parallel sub-channels (or sub-carriers). The symbol (or frame) duration is T s = 4µs [2]. The HIPERLAN2 standard sets M n = M for each subcarrier and M could be chosen between 2, 4, 16, 64. In order to grant the orthogonality between sub-carriers in ideal /02/$ IEEE PIMRC 2002

2 channel conditions, the sub-channel subdivision is obtained by means of a N = 64 order inverse Fourier Transform (IFFT). Samples at the output of the IFFT block are converted from parallel to serial and transmitted every T c seconds (chip time). In practice, due to propagation effects, sub-channels do not still remain orthogonal so a cyclic prefix (guard interval T g = D T c ) is added to the OFDM frame (the IFFT output) in order to remove the inter-symbol interference (ISI) among sub-channels [9]. At the receiver side, the reverse process is performed. The cyclic prefix represents a redundancy, in fact only the time T u = N T c is dedicated to the transmission of useful symbols, whereas the total OFDM symbol (frame) time, T s, is T u + T g = T c (N + D). If the maximum multipath delay T d is greater than the guard interval T g, the received signal at the output of the FFT block can be written as [11] Video MPEG-4 CODER MPEG-4 Bit Stream Bit Loading Algorithm MOD COFDM 5 GHz Wireless Channel PSNR C.S.I. MPEG-4 DECODER Fig. 1. Transmission system considered DEMOD COFDM PER z n = H n w n A n + x n, (1) where H n is the channel transfer function gain related to the n-th sub-channel, and w n is a weight coefficient which allows non uniform power level allocation in the case where a bit loading algorithm is employed. The random variable x n represents the zero mean complex Gaussian thermal noise component with power σ 2 x = E[ x n 2 ] = 2N 0 /T u, (2) where N 0 is the single side power noise density. The average power, P n, dedicated to the n-th sub-channel is P n = E[ A n 2 ] wn 2 = 2(M n 1) wn 2, (3) 3 leading to a total average transmitted power P T N u P T = P n. (4) n=1 We have neglected the presence of pilot sub-carries allocated for channel estimation purposes. In the case where a M n -QAM signaling is adopted, assuming ideal phase offset compensation, perfect carrier recovery and synchronization, the bit error probability related to the n-th sub-channel can be approximated as follows P bn 2( M n 1) E = s 3 ε n H n erfc 2 η D, (5) Mn log 2 M n 2(M n 1) where Es denotes the average received symbol-energy-tonoise ratio, η D = N/(N + D) takes into account the loss due to the presence of the guard interval and ε n = P n /P T indicates the fraction of the power dedicated to the n-th subchannel. Obviously, it should be N u n=1 ε n = 1. Once the coding gain, R c, is fixed (if coding is present), it is possible to express Es as a function of the received average bitenergy-to-noise ratio E b. As can be noted, the performance at each sub-channel depends on H n, so severely attenuated sub-channels could compromise the performance. Generally, a suitable channel coding is necessary to improve the overall performance (Coded ODFM) [9]. In this paper a RS(96,54,8) block code is considered in order to make semi-analytical evaluations. As far as the channel model is concerned, in this work we refer to the 5Ghz E ETSI channel model [10] (outdoor in non line-of-sight condition) characterized by 18 Rayleigh fading paths. III. MULTI-LAYER ADAPTIVE BIT LOADING The HIPERLAN2 standard foresees a fixed bit loading scheme where, once taken the decision about the constellation size M, based on overall propagation conditions, each sub-channel utilizes the same size M n = M and the same weight w n = 1, independently by the single sub-channel condition. The basic principle of adaptive modulation techniques is the opportunity of modifying dynamically the modulation parameters according to the time-variant channel conditions [7]. This can be accomplished efficiently if the transmitter knows the channel state. A feedback channel should thus be available, as shown in fig.1, in order to pass the Channel State Information (CSI) to the transmitter. The rate of CSI depends on the channel variability, in particular on the channel coherence time, T ch. For example, if a maximum terminal speed of 3Km/h (14Hz Doppler effect ) is considered, the CSI has to be performed roughly every some milliseconds. In the numerical results, this aspect is taken into account. In the multi-carrier system considered, each sub-carrier bears 2 b n bits every T s seconds, where b n depends on the modulation scheme used and represents the amount of bits per dimension of the M n -QAM constellation (b n = 0.5 log 2 M n ). Adaptive solutions, realizable through software radio techniques, aim at dimensioning b n and the power profile of each sub-carrier (through the weights w n, or equivalently, ε n ) in order to get the best performance (fixed data rate) or the best data rate (fixed performance). The signal to noise ratio available at the n-th sub-carrier is

3 SNR n = P n H n 2 σx 2 = ε n g n, (6) having defined g n = E s N 0 H n 2. The SNR depends on the channel condition, and it may be opportunely shaped through different choices of ε n. The considered optimization algorithm is based on the water-filling technique, in its Margin Adaptive (MA) version assuming a constant bit rate [8]. The target of this technique is to minimize the total average transmitted power, P T, under the constraint of a fixed total bit rate, B r, or equivalently, maximize the performance if P T is kept constant. Since the capacity of a real system varies with the adopted modulationdemodulation encoding-decoding technique, and it is always below the Shannon theoretical limit, it is useful to introduce the gap concept, which is a figure for the analysis of real systems transmitting with bit rate below the channel capacity [8]. It is a measure of the distance, in terms of signal-to-noise ratio, between the system capacity and the theoretical channel capacity. The gap is dependent both on the modulation technique and on the channel coding scheme adopted. In a M-QAM modulation scheme the gap is practically constant with the variation of the SNR value when the performance, in terms of bit error probability, is kept constant [8]. In this work, we extend the classical water-filling algorithm to the multi-layer case, where several data streams have to be transmitted simultaneously with different performance requirements (unequal error protection) as typical in multimedia applications. In this case, the total number of sub-carriers is divided into L sets, each one, denoted with C(l), is associated to a different layer. Each layer requires a particular gap Γ l (related to the performance) and a specific bit rate B rl. We thus establish a constraint on the total bit-rate B r = L l=1 B r l = 2 b/t s, i.e. on the total number of bits per dimension, b, to be transmitted during a OFDM frame time T s b = 1 N u 2 log 2 (1 + ε n g n ), (7) Γ f(n) n=1 where the function f(n) returns the layer index number which the n-th sub-carrier belongs to. When the number of layers, L, is greater than one, the optimization of b n and ε n becomes a complex problem because the sub-carrier set, C(l), related to the l-th layer can be constructed only once the bit allocation b n (the output of the optimization problem) is known. We propose the following algorithm, that we call AULA, to solve this problem: 1) Set an initial guess bit allocation (e.g. uniform) 2) Order the sub-carriers for decreasing SNR n values 3) Construct the set C(l) for each layer starting from the most reliable sub-carriers (with higher SNR) 4) Perform the water-filling procedure with different gap values, Γ f(n) 5) Any changes in the resulting bit allocation b n? If yes, goes to 2. 6) Stop. At each algorithm step, the water-filling optimization gives the following solution for the power distribution ε n + Γ f(n) g n = constant (8) and constellation size for each sub-carrier M n = 2 2 bn = 1 + ε n g n Γ f(n). (9) It may happen that, after the optimization, a negative power value is assigned to some sub-carriers; this means that the optimal solution requires to turn off these sub-channels as affected by an excessive amount of noise. It is evident that this technique allocates the greatest amount of resources to the most reliable sub-carriers in terms of SNR. It should be noted that the solution could give fractional values for b n. In order to make the modulation technique more realistic, these values may be rounded to conventional values, considering that in this way the solution is only sub-optimal and the total bit rate is not exactly the one fixed. More dedicated algorithms could be exploited to deal with integer values for b n [8], even though we have verified that no large gains in terms of performance are obtained. IV. PHYSICAL LAYER PERFORMANCE EVALUATION In the numerical results a comparison has been performed between not adapted and adapted systems. A channel coding rate of R c = 9 16 has been considered in both cases. Packet oriented ATM-like transmission with 54 payload bytes per packet has been taken into account and the channel is assumed invariant during the transmission of each packet. The transmission of one packed requires 4 OFDM frames. The total capacity of the OFDM system is fixed to 48Mb/s (B r = 27 Mb/s useful bit rate). This choice corresponds, in the nonadapted system (like in the HIPERLAN2 standard), to a 16- QAM constellation scheme assigned to all the 48 data subcarriers. Once the channel transfer function is fixed, the packet error probability, P EP, is evaluated analytically by using the methodology presented in [11]. Then, for each parameters configuration, the average packet error probability is obtained by considering an observation time of 4 seconds. The channel transfer function samples have been normalized so that E[ H n 2 ] = 1. The optimization (bit loading) is performed, according to the temporal evolution of the channel, every N update OFDM frames, supposing that the channel state information is feedback with the same rate. It is advisable that N update T s < T ch,i.e. the channel correlation time. In Fig.2 the average packet error probability against E b /N 0 for single layer transmission in the case of good CSI rate (N update = 1000, as will be shown later) is reported. It may

4 PEP adaptive, ideal CSI non adaptive CSI update=1000 CSI update =2000 CSI update =5000 CSI update = Eb/No (db) Fig. 2. Performance evaluation of the adaptive scheme with different CSI update rates. Also the non-adapted case is reported as reference. PEP single layer case beta=1, layer 1 beta=1, layer 2 beta=10/7, layer 1 beta=10/7, layer 2 beta=2, layer 1 beta=2, layer Eb/No (db) Fig. 3. Performance of the two layers adaptive loading algorithm with different values of β be noted that the adaptive technique produces a gain, in terms of the required SNR, of about 4 db with respect to the not adaptive case, having fixed the average P EP to The effect on the performance of different CSI up-date rates has been investigated and the results are reported in fig. 2, where the average packet error probability against E b /N 0 is shown for different values of N update. The results show that a CSI up-date at least every 1000 frames is sufficient not to have a significant degradation of the performance. This means that the overhead due to CSI can be small. The multi-layer optimization is investigated in fig. 3 in the case where two streams are considered, fixing B r1 = B r2 = 12 Mbit/s. The ratio β = Γ 1 /Γ 2 takes into account different gap requirements about the two layers. The single layer performance is reported for comparison. As can be noted, also in the case of β = 1 there is a performance un-balance of about 1.5 db in terms of SNR between layer 1 and layer 2. This is due to the sub-carriers set construction strategy adopted (point 3 of the algorithm) which assigns the most reliable sub-carriers to the first layer. The performance un-balance can be further enhanced by considering different values for β. The curves reported in fig. 3 have been obtained for β = 1, β = 10/7 and β = 2. As β increases, the performance gap between layer 1 and layer 2 becomes more relevant. Compared to the balanced case (one layer), it is interesting to observe that the gain obtained, in terms of SNR, for the layer 1 is greater than the loss paid for the layer 2. This behavior permits to realize a unequal error protection scheme through the bit loading algorithm. V. MPEG-4 VIDEO TRANSMISSION In order to evaluate the performance of video transmission with the proposed technique, we focused on MPEG-4 [12], the latest ISO/IEC standard for video compression. The MPEG-4 standard utilizes the concept of object-based coding, allowing interactivity, and layered coding. The MPEG-4 bitstream is basically structured in Video Objects (VO s), Video Object Layers (VOL s), i.e. the information related to an object in a scalability layer, Video Object Planes (VOP s), i.e. the instance of an object in a frame and, optionally, Groups Of Video object planes (GOV s) and packets. As most video compression standards, it extensively relies on prediction and entropy coding and it is consequently very sensitive to channel errors. With the goal of transmission over error prone channels, some error resilience tools have been added to the MPEG-4 standard: Reversible Variable Length Codes (RVLC), Header Extension Codes (HEC), resync markers and data partitioning help adding robustness to the MPEG-4 bitstream. With the use of Resync markers, the MPEG-4 bitstream results composed of packets which are of almost the same length, separated by start codes, unique words recognizable from any sequence of variable length codewords, but not robust to channel errors. Regardless of these tools, MPEG-4 video transmission over wireless channels is thus still critical: for this reason, studies aimed at efficiently transmit MPEG-4 video over wireless channels are currently being performed. A performance evaluation of MPEG-4 video transmission over an HIPERLAN2-like system with the adaptive technique described above has been here performed and results have been compared with the non adaptive case. As in [13] we coded, according to the MPEG-4 standard, the first 10 frames of a video sequence (the foreman test sequence in CIF format) at a bit rate of 644 kbit/sec. As the fixed total bit-rate is 27 Mbit/s, we supposed to send a packet each 42, considering others multimedia streams to be transmitted in the remaining time. The error probability is analytically evaluated for different channel conditions in terms of E b /N 0, both for the adaptive and for the non-adaptive case. According to these results, a noise realization has been generated for each channel condition. The channel realization ob-

5 FOREMAN SEQUENCE, average Eb/No=15 db, IPPPPPPPPP adaptive loading non adaptive loading modulation, similarly as done through convolutional codes in [16], [13]. A further remarkable quality improvement is thus expected. PSNR frame number Fig. 4. PSNR versus frame number. E b /N 0 = 15dB. The upper curve represents the adaptive loading case. VI. CONCLUSIONS In this paper an adaptive loading technique for multi-carrier modulation is analyzed and applied to the HIPERLAN2 physical layer system. An extension to the multi-layer case of the classical MA bit loading algorithm is proposed in order to perform UEP at modulation level. Remarkable improvements have been achieved with the adaptive technique over non-adaptive HIPERLAN2-like system. A gain up to 4 db in terms of average E b /N 0 for P EP = 10 6 has been obtained with respect to the non-adaptive case considering a CSI update cycle every 1000 frame. The technique has also been applied to MPEG-4 video transmission with good performance gain results. VII. ACKNOWLEDGEMENTS The financial support of the European Community (JOCO project) and of CNR-MIUR (project multimedialita 5%) are gratefully acknowledged. Fig. 5. Foreman frame no.6, received with E b /N 0 = 15dB. On the left, the non-adaptive case: PSNR=20.25 db. On the right, the adaptive loading case: PSNR=21.58 db. tained has been considered for the transmission of the MPEG- 4 sequence. The MoMusys MPEG-4 decoder [14] has been used, with some modifications in order to improve the robustness to errors. Regardless of the improved robustness, we had to avoid noise addition in correspondence of Start Codes in order to perform the decoding: the decoder crashes if errors are present in Start codes, due to the extremely high error sensitivity of this bitstream portion, as described above. Noise has thus been added selectively to the bitstream, assuming start codes and GOV/VOP headers reception error free. Alternatively, the technique described in [15] could be considered. Also the first frame is supposed to be received error free, in order to allow concealment of the subsequent frames; we may in fact retransmit the frame in the case errors occur, as a small delay may be tolerated at the beginning of the bitstream. Results have been obtained in terms of P SNR vs. frame number for several channel conditions in terms of E b /N 0 and show that an evident improvement may be obtained with the adaptive technique respect to the non-adaptive case. In fig.4 the results obtained with adaptive loading for E b /N 0 = 15dB are compared with the non-adaptive case; an improvement of up to 2dB in terms of PSNR is achieved. Also the visual quality improves with the adaptive loading technique considered, as shown in fig.5. The two-layer extension of the loading algorithm proposed in this paper may be advantageously applied for providing unequal error protection of the MPEG-4 bitstream through REFERENCES [1] IEEE802.11, part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications: High Speed Physical Layer in the 5 GHz Band, P802.11a/D7.0, July [2] ETSI TS , Broadband Radio Access Networks (BRAN) Hiperlan Type2 - Physical layer, V April [3] J.A.C.Bingham, ADSL, VDSL, and Multicarrier Modulation, John Wiley and Sons, [4] I. Kalet, The Multitone Channel, IEEE Transactions on Communications, v. 37, n. 2, February [5] H. Zheng and K.J.R Liu, Robust Image and Video Transmission over Spectrally Shaped Channels Using Multicarrier Modulation, IEEE Trans. on Multimedia, Vol. 1, No.1, pp , March [6] L. Van der Perre, S. Thoen, P. Vandenameele, Adaptive loading strategy for a high speed OFDM-based WLAN, IEEE Globecom 1998, vol.1, pp [7] L.Hanzo, P.J.Cherriman, J. Streit, Wireless Wideo Communications: Second to Third Generation Systems and Beyond, IEEE press [8] John M. Cioffi, EE379C: Advanced Digital Communication, Lecture notes, Stanford University. [9] S.B.Weinstein, P.M. Ebert, Data Transmission by Frequency Division Multiplexing using the Discrete Fourier Transform, IEEE Trans. Comm., Vol.19, Oct [10] J.Medbo, H.Andersson, P.Schramm, Channel Models for Hiperlan/2 in different indoor scenarios, COST 259, Apr [11] D. Dardari, V. Tralli, High-Speed Indoor Wireless Communications at 60 GHz with Coded OFDM, IEEE Transactions on Communications, Vol.47, No.11, Nov. 99. [12] MPEG-4 standard, MPEG-4 Video Group, Final Draft of International Standard, ISO-IEC/JTC1/SC29/WG11 N2502a, Atlantic City, October [13] M.G.Martini, M.Chiani, Wireless transmission of MPEG-4 video: performance evaluation of Unequal Error Protection over a block fading channel, IEEE Veh. Technol. Conf. (VTC 2001), Rhodes, May [14] MoMuSys project website: PROJECTS/ac098.htm [15] M. G. Martini, M. Chiani, Robust Transmission of MPEG 4 Video: Start Codes Substitution and Length Field Insertion Assisted Unequal Error Protection, Picture Coding Symposium - PCS 2001, Seoul, April [16] M.Budagavi, W. Rabiner Heinzelman, J. Webb, R. Talluri, Wireless MPEG-4 Video Communication on DSP Chips, IEEE Signal Processing Magazine, January 2000.