IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 3, JULY 2002 393 A New Resource Allocation Scheme Based on a PSNR Criterion for Wireless Video Transmission to Stationary Receivers Over Gaussian Channels Il-Min Kim and Hyung-Myung Kim, Senior Member, IEEE Abstract A new resource allocation scheme is proposed for video service over a Gaussian channel. For video quality rating, we introduce a suitable metric: the average peak signal-to-noise ratio (PSNR) or the average channel-induced PSNR degradation. Then, we model the end-to-end distortion considering the error propagation effect caused by motion compensation. On the basis of this model, we propose an efficient resource allocation scheme for real video service. Unlike conventional schemes, the new scheme satisfies the PSNR requirement of each video user by adjusting the bit-error rate level with the changes of image characteristics. Simulation results show that the proposed scheme utilizes the bandwidth more efficiently than conventional schemes. Index Terms Bit-error rate, distortion, H.263, peak signal-tonoise ratio, resource allocation, variable bit rate, wireless video. I. INTRODUCTION IN THE PAST decade a number of resource management schemes have been proposed in wireless communication systems [1] [7], and most of them have been designed to accommodate voice and/or data. In third-generation communication systems, however, a wide variety of multimedia services, such as video as well as voice and data, must be supported. For this, some advanced resource allocation schemes have been proposed to accommodate variable bit rate (VBR) image services [1] [3]. These schemes, however, have some limitations. First, they were designed and analyzed with simplified traffic models such as Poisson process or Markov chain [3] [6]. However, the real VBR traffic generated by H.261, H.263, and MPEG has more complex traffic characteristics [7]. Thus, the schemes may be inefficient in real applications. Next, although almost all the resource allocation schemes adopt bit-error rate (BER) as a quality-of-service (QoS) criterion, the BER does not reflect image quality with high fidelity. As video quality measuring metrics, there are more desirable criteria: subjective and objective. Subjective criterion is more important. However, it is difficult to do subject rating because it is not mathematically repeatable and it is always hard to collect a group of people to judge the quality of the decoded video. On the other hand, objective metric such as the peak signal to noise ratio (PSNR) has the advantage that it is mathematically repeatable Manuscript received May 1, 2000; revised January 1, 2001 and June 1, 2001; accepted August 1, 2001. The editor coordinating the review of this paper and approving it for publication is K. Chawla. The authors are with the Department of Electrical Engineering and Computer Science, Korea Advanced Institute of Science and Technology (KAIST), Taejon 305-701, Korea (e-mail: hmkim@csplab.kaist.ac.kr; min@csplab.kaist.ac.kr). Publisher Item Identifier 10.1109/TWC.2002.800538. and comparable, and of course, no human being is involved in the judgement. For this reason, for the video service, such an objective metric should be adopted as a QoS criterion instead of the BER. In this paper, on the basis of the PSNR criterion, we propose an efficient resource allocation scheme over a Gaussian channel. This paper is organized as follows. The system model and QoS criterions are described in Section II. In Section III, we present a problem of conventional resource allocation schemes for video application. In Section IV, a new capacity allocation scheme is proposed. The performances of the proposed and conventional schemes are compared in Section V. Finally, the summary and conclusions are given in Section VI. II. SYSTEM MODEL AND QOSCRITERIONS A. H.263 and Error Concealment Schemes H.263 is a new low-bit-rate video coding standard, which provides better picture quality at low-bit-rates with little additional complexity compared with H.261 [8]. H.263 includes four negotiable advanced coding modes: unrestricted motion vectors, advanced prediction, PB frames, and syntax-based arithmetic coding. However, we do not assume the PB frames in this paper. The bit stream syntax of H.263 is as follows. For quarter common intermediate format (QCIF) resolution (176 144 pixels), a frame is subdivided into nine group of blocks (GOBs), with 11 macro blocks each. H.263 provides a means to insert synchronization words at the GOB layer. When one GOB is corrupted, H.263 allows resynchronization at the beginning of the next GOB. An erroneous GOB may be detected by various techniques [9]. The decoder discards the corrupted GOB and replaces it with the error concealed image data [9]. Most current video coders compress the video data by removing the implicit correlations using various techniques. However, this data compression aggravates the effects of bit errors. To overcome this problem, various error concealment schemes have been proposed [9] [12]. They can be categorized into two classes: temporal concealment [10], [11] and spatial concealment [12]. Spatial concealment has the advantage of providing smooth and consistent edges. However, it is computationally intensive and hard to implement in hardware. Temporal concealment is simple and performs best for most sequences except for those with too much active and/or irregular motion or with too many scene changes. The scheme in [10] uses another highly error-protected, but not error-free, channel over which motion 1536-1276/02$17.00 2002 IEEE
394 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 3, JULY 2002 vectors are delivered to the decoder. The decoder replaces the erroneous region by the motion compensated area of the reference frame. Since the performance of the temporal concealment schemes strongly depends on the availability of the motion vectors, the scheme in [10] can be considered simple and effective and, hence, is adopted in the system model considered in this paper. In the motion vectors channel, we assume that rate-compatible punctured convolutional (RCPC) coding [13], [14] is used although other error correction schemes can be used such as the Reed Solomon coding. Simulation details are explained in Section V. TABLE I SELECTED QP S AND THE CORRESPONDING AVERAGE PSNRS OF THE TEST IMAGES UNDER ERROR-FREE CONDITION B. VBR Video Transmission Each coded image sequence can be transmitted in a constant bit rate (CBR) or VBR mode. Although the communication system with the CBR mode can be managed more easily, VBR transmission has many other advantages [15]. Thus, we adopt the VBR mode and assume that each image sequence is coded with a fixed quantization step size. To accommodate the VBR data over the wireless code-division multiple-access (CDMA) channel, the variable processing gain-cdma (VPG-CDMA) [16], [17] technique can be used. C. QoS Criterions for Video Service As a QoS criterion for video service, we adopt the standard measure of objective image quality: the average PSNR (AP) which is defined by (1) where is the total number of coded frames and the distortion of the th frame. The distortion is measured as the mean squared error (MSE) between the original and the decoded frames. As another QoS criterion, we consider the average channelinduced PSNR degradation,, which is defined as follows: is the average PSNR under error-free channel condi- where tion. (2) III. THE CONVENTIONAL RESOURCE ALLOCATION SCHEME FOR VIDEO TRANSMISSION Almost all resource management schemes present a BER threshold as a QoS criterion, and the BER value is, in general, the same for users with the same service class such as voice, data, and video [1] [3]. In this paper, these schemes are referred to as the conventional scheme. In video service, however, the decoded image qualities may diverge even if the same BERs are given, because image characteristics such as the degree of motion and the complexity of the image can be different. To show this, we conduct a simulation with H.263 and real image sequences. In the simulation, we use six well-known QCIF images: three low motion images, Akiyo, Mother and daughter, Hall Monitor, and three high motion images, Silent, News, Foreman. These Fig. 1. Average PSNR versus bit-error probability P of the conventional scheme. images are normally used for the performance evaluation of very low-bit-rate codecs such as H. 263, H. 263+, and MPEG-4, which are suitable for wireless video service. Under error-free condition, the AP values of the image sequences can be adjusted as close as possible by varying the quantization parameters (QPs). The QP values and the corresponding AP values are listed in Table I. Each test sequence is composed of 300 image frames and grabbed at 30.0 Hz. However, the time duration of a sequence, 10 s, is much shorter than that of a call. Thus, we concatenate each image sequence ten times repeatedly. Each sequence is coded at ten frames per second. We assume that, in the conventional scheme, an optimum mode selection method [20] is adopted. All the simulation results are obtained by averaging ten runs. The curves of Fig. 1 represent AP versus BER for the six image sequences. It can be observed that the images with higher motion experience larger AP degradation than those with lower motion. This implies that the former is generally more susceptible to channel errors than the latter. Thus, if the various images are transmitted under the same BER condition as in the conventional scheme, there will be a large amount of deviation in the decoded image qualities. A similar phenomenon can be observed in [10].
KIM AND KIM: A NEW RESOURCE ALLOCATION SCHEME BASED ON A PSNR CRITERION 395 Because a communication system should specify the minimum guaranteed quality levels for all users as QoSs, the performance of the worst user is important. Thus, the conventional resource allocation scheme can be expressed as follows: The conventional resource allocation scheme: 1) When a minimum AP threshold,, is given, the maximum allowable BER,, for video users is determined by the following: where denotes the group of the video users, and the AP value of user when the BER is. 2) For each video user, the system resources are allocated so that is satisfied. For example, in the simulation results, is chosen as of Foreman.If is given as 25 db, is determined as 2 10. However, this BER value is much smaller than needed for the other image sequences. The BER is a function of the redundant bits used for forward error correction. Thus, in this case, too much of the system capacity is allocated to the other image sequences. Therefore, it is wasteful to assign a BER level for all video users equally, because the transmission channel must be able to satisfy the stringent BER requirement among all the users when a QoS criterion such as is given. When the average channel-induced PSNR degradation is given as a QoS criterion, can be determined by the following: where is an average channel-induced PSNR degradation threshold and the average channel-induced PSNR degradation of user with the BER. (3) (4) Let be the event that the ( )th GOB is corrupted. Then is given by where is the number of bits of the ( )th GOB and the number of GOBs in a frame. Assume denotes the distortion induced by the error concealment when the ( )th GOB is lost. is given by the sum of the squared differences between the ( )th GOBs image data encoded in the encoder and the corresponding error-concealed image data in the decoder. Then can be expressed as follows: For the inter-coded image sequence, the dependencies between the consecutive coded frames result in the propagation of the channel error-induced distortion to the subsequent frames. Therefore, we model the total distortion of each frame considering the expected error propagation effect on the future frames. First, we define the effective total distortion and the effective channel error-induced distortion of the th frame as follows: where represents the distortion component of the th frame propagated from for. We can rewrite as follows: (6) (7) (8) (9) (10) IV. THE PROPOSED RESOURCE ALLOCATION SCHEME FOR VIDEO TRANSMISSION To overcome the problem of the conventional scheme, we propose a new resource allocation scheme which attempts to make the decoded image qualities the same irrespective of the image characteristics. In this scheme, the BER level of each GOB is adjusted according to the characteristic of the image information contained in the GOB. Let be the BER level of the th GOB of the th frame, which will be denoted as the ( )th GOB. In the following, to derive, we model error propagation on frame level. A. Error Propagation Modeling on Frame Level The end-to-end distortion of an intra-coded image sequence is first considered. Under the assumption that the quantization distortion and the channel-induced distortion are statistically independent where is the quantization distortion, the channel errorinduced distortion of the th frame, and the number of image frames of a sequence. (5) where is defined as (11) Since represents the distortion propagation effect, caused by the inter-coding, from the ( )th frame to the th frame, it can be estimated as follows: (12) where is the total number of blocks (16 16 pixels) of a frame and the number of inter-coded blocks of the th frame. Recently, interframe error propagation has been widely studied. Steinbach et al. modeled the error propagation effect of each macro block and used it in the macro block INTRA refreshment [18]. Change et al. proposed a scheme that tracks the propagated errors by examining the backward motion dependency for each pixel [19]. For optimum mode selection, Côté et al. modeled error propagation on macro block level [20]. Considering error propagation along both temporal and
396 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 3, JULY 2002 spatial axes, Zhang et al. proposed an algorithm that recursively estimates the total distortion at pixel level precision [21]. Compared with these models, our proposed model requires less computational complexity because it is modeled on frame level. Stuhlmüller et al. analytically derived an model for the overall accumulated distortion under the assumption that, on average, the channel-induced error variance is the same for every frame [22], [23]. B. The Proposed Resource Allocation Scheme At the th frame, the encoder knows for but does not know for. However, these unknown values should be determined to estimate. We assume that image characteristics such as the number of inter-coded blocks in each frame are not changed abruptly. Under this assumption, we estimate for, as follows: (13) Thus, the estimate of, which will be denoted as,isgiven by (14) (15) The value of ( ) can be used as a measure of the decoded image quality. It is possible to make this measure the same for all video users at every frame as follows: the encoder of each video user adjusts so that becomes a prescribed value while is fixed as a constant (16) where is the threshold of the effective channel error-induced distortion of a frame. is used as a control parameter in the proposed scheme instead of the BER. In addition, the image qualities of all GOBs of a frame can be made identical by setting (17) Then from (7), (15) (17) (18) Thus, the required BER of the ( )th GOB is obtained as shown in (19) at the bottom of the page. Note that this derivation is neither codec specific nor error concealment scheme specific. All the parameters in the right-hand side of (19) except are known to the encoder. However, the encoder cannot calculate unless it knows the decoder s concealment scheme, which is generally unknown to the encoder because several concealment techniques can be used in real environments. Thus, at the call-setup stage, the encoder should request the information of the concealment scheme used in the decoder. However, the concealment schemes that are concurrently used in the wireless system may not be diverse, because various factors such as the complexity, the performance, and the suitability for the wireless service should be taken into account. In several studies [18], [20], it has been assumed that the encoder knows the concealment scheme of the decoder and calculates the distortion expected in the decoder. The determined BER level,, can be guaranteed by a few techniques as follows. First, we can use the RCPC codes, which was proposed by Hagenauer et al. for flexible error protection in time-varying channels [13], [14]. Next, BER-oriented power control schemes can be used [24], [25]. However, it may be difficult to apply the conventional joint inner-loop and outer-loop power control scheme, unless the outer-loop control [26], [27] works well enough to compensate for the variations of the channel conditions such as the power delay file, the number of resolvable paths, and the maximum Doppler frequency [28]. One advantage of choosing the RCPC codes is that the physical layer need not be involved at all. However, if we choose the power control schemes, the parameters such as,, and must be known to the physical layer. Thus, the encoder and the transmitter should be tightly coupled. We also note that a separate control channel is required to specify the coding scheme used for the RCPC codes or to deliver the information from each video user to the base station (BS) for the power control. In the performance comparison in Section V, therefore, we take account of the additional capacity consumed by the channel. Concerning the complexity, the proposed scheme increases the hardware complexity due to the RCPC coding and the required BER computation at every GOB by (19). Note that (19)
KIM AND KIM: A NEW RESOURCE ALLOCATION SCHEME BASED ON A PSNR CRITERION 397 Fig. 2. Average PSNR versus the effective channel error distortion threshold of the proposed scheme. D these are not needed in the conventional scheme. However, the computation required for the calculation of the BER is negligible compared with the other computations conducted in the encoder. On the other hand, in the conventional scheme with the optimal INTRA rate selection scheme, each block must be predicted, transformed, and quantized for the coding modes considered (skip, inter, and intra) [20] although these calculations are performed for one mode in the proposed scheme. Moreover, in the optimal INTRA scheme, the corresponding distortions and coding rates must be computed for the Lagrangian minimization. Generally, the additional computation for the optimal INTRA scheme is not negligible. Regarding the required memory, the additional memory of the proposed scheme may be smaller than that of the conventional scheme because, in the proposed scheme, only the number of inter-coded blocks of each frame must be stored while, in the conventional scheme, the block error probability and the error concealment distortion of each block must be stored. For both the conventional and the proposed schemes, the decoder error concealment must be performed at the encoder for each block and the incurred concealment distortion must be computed. Of course, in the conventional scheme, the computational complexity and the required memory can be considerably reduced if the optimal INTRA rate selection scheme is not employed. In this case, however, the PSNR variation becomes larger at smaller BER values because the video codec is not optimized with respect to each image sequence. Regarding the architectural complexity of the system, the proposed scheme may increase the complexity because the information on must be conveyed from the mobile terminal to the BS. Fig. 2 shows the AP versus when the BER level of each GOB is adjusted by (19). By varying, the AP degradation of the worst user can be adjusted. From the results, we check whether the adopted error propagation model is accurate. If the adopted model is accurate, the real AP should be given (20) where is the quantization distortion. In Fig. 2, is calculated as about 25 because the error-free PSNR is about 34 db. After putting the value into (20), we observe that the AP calculated by (20) matches well the real AP in Fig. 2 for every value. Note that the PSNR ranges of the worst users in Figs. 1 and 2 are the same. In addition, the PSNR degradations are also the same. Thus, these two figures can be compared and we can observe that the AP difference of the images in Fig. 2 is quite smaller than that of Fig. 1. For example, in Fig. 2, the difference becomes about 4 db when the AP of the worst user drops by 14 db while, in Fig. 1, the difference is about 13 db. Using (19), we propose a new resource allocation scheme as follows: The proposed resource allocation scheme: 1) When is specified, we find the maximum value of the channel error-induced distortion threshold,, from the following: (21) where is the AP value of user with. 2) At every GOB, of (19) is calculated using. 3) For every video user, the system resource is allocated so that the calculated value is satisfied at every GOB. By this scheme, the image qualities of the video users can be maintained very close. When the average channel-induced is deter- PSNR degradation is given as a QoS criterion, mined by the following: (22) where is the average PSNR degradation of user with. For example, let 20 db. Then is given by 0.5 10 for the conventional scheme and by 500 for the proposed scheme. Fig. 3 shows the PSNR versus the frame number of 400 to 450 with the and values. From Fig. 3, we can observe the following: the conventional scheme wastes the system resource in order to provide unnecessarily high PSNRs for some users; however, the proposed scheme efficiently utilizes the system resource while the required is satisfied for every user. Moreover, the proposed scheme considerably reduces the PSNR fluctuation of each sequence with time. Fig. 4 shows the subjective results for frame 400 of Mother and Daughter and Foreman when 20 db. Thus, the selected and values are the same as in Fig. 3. We can see that, in the proposed scheme, the difference of the subjective qualities of the images is smaller. Fig. 4 can be interpreted as follows: in a resource utilization point of view, two acceptable sequences which satisfy a required QoS but consume the system resource minimally are better than one acceptable sequence and one good sequence which consumes the system resource unnecessarily much. V. PERFORMANCE COMPARISON In this section, we show that the wireless bandwidth can be utilized more efficiently by the proposed scheme in CDMA systems. We assume a perfectly power controlled CDMA
398 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 3, JULY 2002 activity is neglected. For each video user, however, multiple subchannels should be used for additional traffics such as voice and signaling. Thus, we assume each video user occupies subchannels each of which is separated from the others by the multicode CDMA technique [2], [3], and each of the subchannels accommodates VBR traffic using VPG-CDMA. After allocating capacities to video and voice users, residual capacity is consumed by data users. For each data user, one CBR channel is assumed. Thus, each data user can transmit just one data packet at a radio frame. Our aim is to assign transmit power to all users such that their sum is as small as possible, and such that the QoS requirements of all the users are met. It can be shown that at the th radio frame, a unique solution to this problem exists if and only if Fig. 3. PSNR versus frame number of 400 to 450 with AP = 20 db. P = 0.5 2 10 for the conventional scheme and D = 500 for the proposed scheme. where, (24) Fig. 4. Subject results for frame 400 with AP = 20 db. (a) Akiyo with the conventional scheme, P = 0.5 2 10. (b) Akiyo with the proposed scheme, D = 500. (c) Foreman with the conventional scheme, P = 0.5 2 10. (d) Foreman with the proposed scheme, D = 500. system with a large number of intra- and inter-cell interferers in flat fading environments. Then the proposed resource allocation scheme, which is based on Gaussian channel assumption, can be applied. As a measure of the performance comparison, we define the average data throughput as follows: (23) where is the number of the total running radio frames and the data throughput at the th radio frame. The procedure of the calculation of is described in the following. Let and denote video and voice users in the system, respectively. Each voice user has single CBR channel, and voice : the number of total running radio frames. : the bit rate and the required energy per bit to interference plus noise density ratio (EINR) of the th subchannel of the th video user at the th radio frame, respectively. : the bit rate and the required EINR of voice and data users at each radio frame, respectively., : peak transmit power permitted to the th subchannel of the th user and to the th user s channel, respectively. : pathloss from the th user to the BS at the th radio frame. : intercell interference plus background noise power at the th radio frame. This constraint is similar to the results in [30] and [31], but, multiple subchannels of a user are not taken into consideration in the papers. We derived (24) using the procedure in [31] and considering the fact that no interference exists between the subchannels of a user since the subchannels are separated by orthogonal codes. We also note that the capacity constraint of the multicode CDMA system in [32] is similar to (24). However, unlike the results in [32], multirate and multi QoS subchannels are adopted in (24), and (24) reduces to the results in [32] without
KIM AND KIM: A NEW RESOURCE ALLOCATION SCHEME BASED ON A PSNR CRITERION 399 TABLE II SYSTEM PARAMETERS fading if we make the bit rates and EINRs of all the subchannels in (24) be identical. After allocating the capacities for video users, we assume call admission control (CAC) is performed such that voice users are accepted by the BS as many as possible without violating the constraint of (24). As a CAC scheme, various techniques can be used such as [29]. Then the maximum value of is calculated as (25) where denotes the largest integer less than or equal to. Although the system does not know the values of,, and in advance, the peak values are assumed to be known to the system. In the simulation, for the performance comparisons, we fix the number of voice users as given by conventional proposed (26) Note that of the conventional and the proposed schemes may be different, even if or is the same. Next, the data throughput,, at the th radio frame is obtained as shown in (27) at the bottom of the page. In the simulation, we assume each video user has four subchannels ( 4). The first subchannel transmits image packets and the second one motion vector packets. A motion vector packet contains the motion vector ingredients of a GOB, and corresponding image packet does the remained image data of the GOB. To deliver motion vectors safely for the error concealment at the decoder, the EINR of the second subchannel is sufficiently high: 12 db in the simulation. When motion vectors are corrupted, they are estimated using the (27)
400 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 3, JULY 2002 Fig. 5. Average data throughput T versus average channel-induced PSNR degradation threshold 1AP. surrounding available motion vectors. If no surrounding motion vectors are available, the corrupted motion vectors are set to zero. The third subchannel delivers voice packets, and the fourth one controls packets which have the information about instantaneous bit rates of the first and the second subchannels at each radio frame. The first and the second subchannels are VBR, and the third and the fourth CBR. In the proposed scheme, the rate information of the RCPC code is also delivered via the fourth subchannel. Thus, the bit rate of the fourth subchannel,, is 1.0 kbits/s in the conventional scheme and 1.5 kbits/s in the proposed one. Peak transmit power, pathloss, and intercell interference plus background noise are assumed to be the same for all subchannels, channels, and time, i.e.,,,, and is chosen to be 0.1 as in [30]. The system parameters used are summaried in Table II. Fig. 5 shows the versus the average channel-induced PSNR degradation threshold. From the results it can be observed that the proposed scheme increases the average data throughput. VI. CONCLUSION For voice, data, and image services over wireless channels, many researchers have proposed various resource management schemes which adopt BER as a QoS criterion. However, for the image quality rating, the BER is not suitable and there are more desirable QoS criterions such as the average PSNR or the average channel-induced PSNR degradation. In this paper, we have proposed a new resource allocation scheme which guarantees the required PSNR or the allowable PSNR degradation for each video user. This scheme adjusts the BER level of each GOB according to the image characteristic of the GOB. Numerical results show that the proposed scheme achieves higher bandwidth utilization than the conventional scheme. REFERENCES [1] I. F. Akyildiz, D. A. Levine, and I. Joe, A slotted CDMA protocol with BER scheduling for wireless multimedia networks, IEEE/ACM Trans. Networking, vol. 7, pp. 146 158, Apr. 1999. [2] Z. Liu, M. J. Karol, M. E. Zarki, and K. Y. Eng, A demand assignment access control for multicode DS-CDMA wireless packet (ATM) networks, in Proc. INFOCOM 96, 1996, pp. 713 721. [3] I.-M. Kim, B.-C. Shin, Y. J. Kim, J. K. Kim, and I. Han, Throughput improvement scheme in multicode CDMA, Electron. Lett., vol. 34, pp. 963 964, May 1998. [4] I.-M. Kim and H.-M. Kim, Efficient power management schemes for video service in CDMA systems, Electron. Lett., vol. 36, pp. 1149 1150, June 2000. [5] J. Wu and R. Kohno, A wireless multimedia CDMA system based on transmission power control, IEEE J. Select. Areas Commun., vol. 14, pp. 683 691, May 1996. [6], Performance evaluation of wireless multimedia CDMA networks using adaptive transmission control, IEEE J. Select. Areas Commun., vol. 14, pp. 1688 1697, Dec. 1996. [7] J. C. Chen, K. M. Sivalingam, P. Agrawal, and R. Acharya, Scheduling multimedia services in a low-power MAC for wireless and mobile ATM networks, IEEE Trans. Multimedia, vol. 1, pp. 187 201, June 1999. [8] G. Côté, B. Erol, M. Gallant, and F. Kossentini, H.263+: video coding at low bit rates, IEEE Trans. Circuits Syst. Video Technol., vol. 8, pp. 849 866, Nov. 1998. [9] Y. Wang and Q.-F. Zhu, Error control and concealment for video communication: a review, Proc. IEEE, vol. 86, pp. 974 997, May 1998. [10] M. Ghanbari and V. Seferidis, Cell loss concealment in ATM video codecs, IEEE Trans. Circuits Syst. Video Technol., vol. 3, pp. 238 247, June 1993. [11] W. Luo and M. E. Zarki, Analysis of error concealment schemes for MPEG-2 video transmission over ATM based networks, in SPIE, vol. 2501, 1995, pp. 1358 1368. [12] J. W. Park, J. W. Kim, and S. U. Lee, DCT coefficients recovery-based error concealment technique and its application to the MPEG-2 bit stream error, IEEE Trans. Circuits Syst. Video Technol., vol. 7, pp. 845 854, Dec. 1997. [13] J. Hagenauer, Rate-compatible punctured convolutional codes (RCPC codes) and their applications, IEEE Trans. Commun., vol. 36, pp. 389 400, Apr. 1988. [14] J. Hagenauer, N. Seshadri, and C.-E. W. Sundberg, The performance of rate-compatible punctured convolutional codes for digital mobile radio, IEEE Trans. Commun., vol. 38, pp. 966 980, July 1990. [15] T. V. Lakshman, A. Ortega, and A. R. Reibman, VBR video: tradeoffs and potentials, Proc. IEEE, vol. 86, pp. 952 973, May 1998. [16] C. L. I and R. D. Gitlin, Variable spreading gain CDMA with adaptive control for true packet switching wireless networks, in Proc. ICC 95, 1995, pp. 725 730. [17] I.-M. Kim and H.-M. Kim, Scheme for enhancing performance of smart antenna in variable processing gain packet CDMA systems, Electron. Lett., vol. 35, pp. 518 519, Apr. 1999. [18] E. Steinbach, N. Färber, and B. Girod, Standard compatible extension of H.263 for robust video transmission in mobile environments, IEEE Trans. Circuits Syst. Video Technol., vol. 7, pp. 872 881, Dec. 1997. [19] P.-C. Chang and T.-H. Lee, Precise and fast error tracking for errorresilient transmission of H.263 video, IEEE J. Select. Areas Commun., vol. 18, pp. 600 607, June 2000. [20] G. Côté, S. Shirani, and F. Kossentini, Optimal mode selection and synchronization for robust communications over error-prone networks, IEEE J. Select. Areas Commun., vol. 18, pp. 952 965, June 2000. [21] R. Zhang, S. L. Regunathan, and K. Rose, Video coding with optimal inter/intra-mode switching for packet loss resilience, IEEE J. Select. Areas Commun., vol. 18, pp. 966 976, June 2000. [22] N. Färber, K. Stuhlmüller, and B. Girod, Analysis of error propagation in hybrid video coding with application to error resilience, in Proc. ICIP 99, 1999, pp. 550 554. [23] K. Stuhlmüller, N. Färber, M. Link, and B. Girod, Analysis of video transmission over lossy channels, IEEE J. Select. Areas Commun., vol. 18, pp. 1012 1032, June 2000. [24] P. S. Kumar, R. D. Yates, and J. Holtzman, Power control based on bit-error rate (BER) measurements, in Proc. MILCOM 95, 1995, pp. 617 620. [25] P. Cherriman and L. Hanzo, Error-rate-based power-controlled multimode H.263-assisted video telephony, IEEE Trans. Veh. Technol., vol. 48, pp. 1726 1738, Sept. 1999. [26] A. Sampath, P. S. Kumar, and J. M. Holtzman, On setting reverse link target SIR in a CDMA system, in Proc. VTC 97, 1997, pp. 929 933. [27] H. Kawai, H. Suda, and F. Adachi, Outer-loop control of target SIR for fast transmit power control in turbo-coded W-CDMA mobile radio, Electron. Lett., vol. 35, pp. 699 701, Apr. 1999. [28] J. V. Andrew, CDMA, Principles of Spread Spectrum Communication. Reading, MA: Addison-Wesley, 1995.
KIM AND KIM: A NEW RESOURCE ALLOCATION SCHEME BASED ON A PSNR CRITERION 401 [29] I.-M. Kim, B.-C. Shin, and D.-J. Lee, SIR-based call admission control by intercell interference prediction for DS-CDMA systems, IEEE Commun. Lett., vol. 4, pp. 29 31, Jan. 2000. [30] S. Ramakrishna and J. M. Holtzman, A scheme for throughput maximization in a dual-class CDMA system, IEEE J. Select. Areas Commun., vol. 16, pp. 830 844, Aug. 1998. [31] A. Sampath, P. S. Kumar, and J. M. Holtzman, Power control and resource management for a multimedia wireless CDMA system, in Proc. PIMRC 95, 1995, pp. 21 25. [32] S. J. Lee, H. W. Lee, and D. K. Sung, Capacities of single-code and multicode DS-CDMA systems accommodating multiclass services, IEEE Trans. Veh. Technol., vol. 48, pp. 376 384, May 1999. Il-Min Kim received the B.S. degree in electronics engineering from Yonsei University, Seoul, Korea, in 1996, and the M.S. and Ph.D. degrees in electrical engineering from the Korea Advanced Institute of Science and Technology (KAIST), Taejon, Korea, in 1998 and 2001, respectively. From July 1997 to August 2001, he was a Member of the Technical Staff at the Electronics and Telecommunications Research Institute (ETRI), Taejon, Korea. Since October 2001, he has been employed as a Postdoctoral Associate at the Department of Electrical Engineering and Computer Sciences, Massachusetts Institute of Technology, Cambridge, MA. His research interest include space-time codes, resource management, wireless video, power control, and smart antennas. Dr. Kim was awarded the 2001 Samsung Humantech Gold Prize. Hyung-Myung Kim (S 86 M 86 SM 99) received the B.S. degree in electronics engineering from Seoul National University, Seoul, Korea, in 1974, and the M.S. and Ph.D. degrees in electrical engineering from the University of Pittsburgh, Pittsburgh, PA, in 1982 and 1985, respectively. Since 1986, he has been with the Department of Electrical Engineering and Computer Science, Korea Advanced Institute of Science and Technology (KAIST), Taejon, Korea, where he is currently a Professor. During the summer of 1997, he was on a sabbatical leave as a Visiting Researcher at the Department of Electrical Engineering, Pennsylvania State University, University Park, Pa. His research interests include digital signal/image processing, digital transmission of voice, communications data and image, and multidimensional system theory. Dr. Kim was the Treasurer of the IEEE Taejon Section, in 1992. He has been an Editorial Board Member of Multidimensional Systems and Signal Processing since 1990.