Visual Media Coding and Transmission

Enhancement Schemes for Multimedia Transmission over Wireless Networks 381 Table 9.20 Modulation coding parameters for E GPRS [30] Scheme Header rate Code rate Data/Radio block MCS 1 0.53 0.53 176 MCS 2 0.53 0.66 224 MCS 3 0.53 0.80 296 MCS 4 0.53 1.0 352 MCS 5 0.33 0.37 448 MCS 6 0.33 0.49 592 MCS 7 0.36 0.76 2 Â 448 MCS 8 0.36 0.92 2 Â 544 MCS 9 0.36 1.0 2 Â 592 block. To ensure strong header protection, the header part of the radio block is independently convolutionaly coded from the data part of the radio block according to the code rate specified in Table 9.20. Finally, the header is interleaved over four bursts and transmitted. At the receiver, the header is de-interleaved and decoded first, followed by the rest of the radio block, which is decoded with the indicated transmission modes. The link-adaptation algorithms adapt each radio link to one of nine modulation coding schemes. Adaptation intervals and switching thresholds are determined by the particular algorithm used. Let CIRest(k) be the estimated channel condition at the kth radio block. Modulation coding scheme mode m is chosen if CIRest(k)2[jth(m), jth(m þ 1)], where jth(m) and jth(m þ 1) indicate channel CIR threshold values corresponding to mode m. Note that j0 ¼ 0 and j9 ¼ ¥. Two approaches to link adaptation, one based on source quality (referred to as the source (video) quality-based adaptation scheme, SQBAS) and the other based on system throughput (referred to as the throughput-based adaptation scheme, TBAS), are investigated. Table 9.21 depicts the differences between SQBAS and TBAS. 9.3.2.2 Link-adaptation Algorithms Source (Video) Quality-based Adaptation Scheme (SQBAS) SQBAS is designed to maximize the video quality by varying the source rate and channel- coding scheme accordingly, while maintaining fixed channel allocation throughout the transmission. Figure 9.35 shows the effects of channel errors upon the quality of MPEG-4-encoded video in a TU3 (typical urban multipath, mobile terminal velocity 3 kmph) propagation environment. Video sequences have been encoded for operation with three-timeslot usage. The video source Table 9.21 Characteristics of SQBAS and TBAS TBAS SQBAS Radio block level Operation at Video frame level Total system throughput Optimization based on Received bideo quality Fixed Video Source Rate Variable Variable Modulation Coding Scheme Variable Variable No. of Timeslots Allocation Fixed

382 Visual Media Coding and Transmissionaverage PSNR (dB) 32 30 28 26 24 22 20 18 MCS 1 MCS 5 16 MCS 2 MCS 3 MCS 6 14 5 10 15 20 25 30 35 40 CIR (dB) Figure 9.35 Video quality at TU3 900 MHz with ideal frequency hopping at three timeslot operation [31] rate for each channel-coding scheme has been set according to Table 9.22. Average perfor- mances over a number of different video sequences are shown. As can be seen in Figure 9.35, MCS-1 gives better performance than MCS-2 at all CIR values up to 22.5 dB. MCS-5 provides superior video quality to other schemes at CIR values better than 22.5 dB. Optimal video quality in these propagation conditions is therefore achieved by selecting MCS-1 when the channel CIR is lower than 22.5 dB, and MCS-5 otherwise. The proposed source quality-based link adaptation algorithm is summarized in Figure 9.36. Throughput-based Adaptation Scheme (TBAS) For TBAS, the switching threshold is selected so as to guarantee a target received video quality over a range of CIR. The scheme is designed to maximize the system throughput by selecting a channel-coding scheme with low protection and high throughput at better channel conditions for fixed-source rate operation. Table 9.22 EGPRS multislotting capacity for video (kbps) [31] Scheme 1 TS 2 TS 3 TS 4 TS 5 TS 6 TS 7 TS 8 TS MCS 1 7.5 15 22.5 30 37.5 45 52.5 60 MCS 2 9.6 19.2 28.8 38.4 48 57.6 67.2 76.8 MCS 3 12.6 25.2 37.8 50.4 63 75.6 88.2 100.8 MCS 4 15 30 45 60 75 90 105 120 MCS 5 19 38 57 76 95 114 133 152 MCS 6 25.2 50.4 75.6 100.8 126 151.2 176.4 201.6 MCS 7 38 76 114 152 190 228 266 304 MCS 8 446.2 92.4 138.6 184.8 231 277.2 323.4 369.6 MCS 9 50.3 100.6 150.9 201.2 251.5 301.8 352.1 402.4

Enhancement Schemes for Multimedia Transmission over Wireless Networks 383 Figure 9.36 Flow chart of the source (video) quality based link adaptation algorithm Figure 9.37 shows the results that are obtained for video transmission over the TU3 propagation environment with a fixed source rate of 38 kbps. The number of timeslots is allocated according to Table 9.22. For example, if target video quality in PSNR is 30 dB, an appropriate switching policy would be to select MCS-5 when the channel CIR is higher than 18 dB and MCS-1 otherwise, thereby achieving target video quality. This results in a saving of three timeslots when switching from MCS-1 to MCS-5. Even though four channel coding schemes (not MCS-3) are equally valid for use in the switching algorithm, for comparison purposes with SQBAS only MCS-1 and MCS-5 are used in the experiment.

384 Visual Media Coding and Transmission 34 100 32 30average PSNR (dB) MCS 1 28 BLER 26 MCS 5 24 MCS 2 22 MCS 3 20 MCS 6 18 10 1 5 10 2 10 15 20 25 30 MCS 1 10 3 5 10 15 20 25 30 35 MCS 2 0 CIR (dB) MCS 3 CIR (dB) MCS 5 (a) MCS 6 (b) 35 40 Figure 9.37 (a) Video quality at TU3 900 MHz with ideal frequency hopping at 38 kbps fixed source rate operation; (b) performance of EGPRS networks at TU3 900 MHz with ideal frequency hopping 9.3.2.3 Channel State Estimation The channel-estimation algorithm is constructed by partitioning the range of BLER measure- ments into a finite number of intervals, I, and mapping them on to the actual channel carrier-to- interference ratio. As Figure 9.37 illustrates, the measured BLER is a nonlinear function of channel CIR and also depends on the channel-coding scheme used. This is because the BLER flag is set if any bit of the radio block is in error after the channel decoding. Measured BLER is averaged over n radio blocks in order to reduce the effect of burst errors on the channel estimationX. Thus, the calculated BLER that is to be used in the channel-estimation algorithm is Bcal;j ¼ 1 n Bmeas; Bcal, Bmeas, n k¼1 j , where j and j are calculated and measured BLERs, respec- tively. Subscript j represents the channel-coding scheme used. In addition, the mean, Rmean, the variance, Rvar, and the gradient, Rgrad, of RSS are used in the channel prediction calculation. The measurement window size is set to be equal to the estimated processing delay, Dest, which is assumed to be constant for a given application. Using linear prediction, predicted RSS is Rpre ¼ RgradÁDest þ Rmean. Predicted RSS is also partitioned into I intervals. Let CIRest(k), Bcal, j(k), and Rpre(k) be the estimated channel condition, calculated BLER, and predicted RSS at kth radio block, and CIRi be the corre- sponding CIR value for ith (i2{1,2, . . ., I}) interval. 8 < CIRi; Bcal;jðkÞ 2 ½hthðiÞ; hthði þ 1ފ \\ RpreðkÞ 2 ½mthðiÞ; mthði þ 1ފ CIRestðkÞ ¼ : CIRestðk 1Þ; RvarðkÞ < g th CIRl Bcal;jðkÞ 2 ½hthðlÞ; hthðl þ 1ފ; RpreðkÞ 2 ½mthðmÞ; mthðm þ 1ފ where l < m ð9:29Þ where mth(i), hth(i), and gth indicate the corresponding RSS, BLER, and variance of RSS threshold values.

Enhancement Schemes for Multimedia Transmission over Wireless Networks 385 9.3.2.4 Feedback Techniques The feedback channel is used to carry measured channel information, RSS, and BLER. The measured RSS is quantized with eight-bit uniform quantization, and the measured BLER is represented by either zero or one. The nine bits of each quantized sample are further protected with a selected modulation coding scheme and are fed back to the transmitter via a noisy channel. The total information rate on the feedback channel is around 450 bps. This is very low compared with the forward data rate of 22 60 kbps. Therefore, in-band signaling is possible for two-way videotelephony. 9.3.2.5 Results and Discussion The QCIF test sequence Suzie was selected as the source signal. The video codec parameters used in the experiment are listed in Table 9.23. Each video frame is considered as one transport and network layer data payload unit in the EGPRS protocol implementation. This is because the size of each video frame is below the specified maximum LLC-PDU size, which is 1520 octets. Zero Feedback Delay The performance of SQBAS is compared to the performance of three other schemes. Two involve the use of fixed coding schemes; MCS-1 with source rate fixed at 22.5 kbps, and MCS-5 with source rate fixed at 57 kbps. The third (CBAS) uses both the MCS-1 and MCS-2 coding schemes, and switching between MCS-1 and MCS-5 is based on the actual CIR of the transmission channel. Figure 9.38(a) clearly illustrates the improvement in video quality under all channel conditions when the adaptive coding schemes (ACSs) are used. For example, the achieved quality improvements with SQBAS relative to MCS-1 and MCS-5 for an average channel CIR of 20.4 dB are 4.13 dB and 3.14 dB, respectively. The results also show that the proposed algorithm, SQBAS, slightly outperforms CBAS. For clarity, the video quality variation throughout a transmission is illustrated in Figure 9.38(b) as the cumulative distribu- tion functions (CDF) of averaged frame PSNR corresponding to the points a1, a2, a3, and a4 on Figure 9.38(a) (average CIR of 21.5 dB). The quality improvement is also visible when viewing the decoded video sequences (Figure 9.39). Table 9.23 Video codec parameters used in the experiment MPEG 4 10 fps Video Codec 600 bits Frame Rate Enabled Video Packet Size Enabled Reversible Variable Length Code (RVLC) 10 (fixed) Data Partitioning PSNR Number of intra MBs used in AIR MP4 Video Quality Measure Rate Adjustment

386 Visual Media Coding and Transmission 36 1 MCS−1 34 a1 0.9 MCS−5 32 a2 CBAS 30 a3 0.8 SQBAS 28 a4 26 MCS−1 24 MCS−5 CBAS SQBAS 22 12 14 16 18 20 22 24 26 28 30 average CIR (dB) (a) average PSNR (dB) 0.7 cdf 0.6 0.5 0.4 0.3 0.2 0.1 0 18 20 22 24 26 28 30 32 34 36 38 frame PSNR (dB) (b) Figure 9.38 Performance of the quality based adaptation scheme: (a) average frame PSNR vs. average channel CIR; (b) CDF of frame PSNR at CIR 21.5 dB Figure 9.39 Perceptual quality comparison

Enhancement Schemes for Multimedia Transmission over Wireless Networks 387 9.3.2.6 Performance of Link-adaptive Algorithms Performances of TBAS were measured in terms of system throughput savings, Thadapt,mcs1, and service quality improvements, Qadapt,mcs5, which are defined as: Qadapt;mcs5 ¼ Qadapt À Qmcs5 :100% ð9:30Þ Qmcs5 Thadapt;mcs1 ¼ Thmcs1 À Thadapt :100% ð9:31Þ Thmcs1 where Q and Th indicate the median frame PSNR of the decoded video sequence and the number of timeslots used in the transmission respectively. The resulting normalized quality improvement, Qadapt,mcs5, and normalized system throughput savings, Thadapt,mcs1, show a linear relationship for operation at a range of fixed source rates (Figure 9.40). The normalized quality improvement achieved tends to decrease with the increase in source rate. This can be explained by taking an example. Assume video sequences encoded at 38 and 60 kbps are transmitted over the same time-varying channel and that the same mode-switching threshold is used. The timeslot saving in switching from MCS-2 to MCS-5 for lower-rate transmission is 60%. The resulting timeslot saving is 50% for high-source rate transmission. In order to achieve the same timeslot saving for high-source rate transmission, the switching threshold should be lowered. This requires the transmission of more video data using low-quality MCS-5 mode; thus it reduces the normalized video quality seen in high-source rate transmission. Effect of Feedback Delay The difference in quality between zero-delay feedback (in terms of average frame PSNR) and finite-delay feedback is shown in Figure 9.40(b). The SQBAS is robust to feedback delay up to 180 ms. Above that, performance decreases with delay, resulting in 0.5 dB quality reduction at 240 ms. TBAS underperforms compared to SQBAS in terms of robustness to feedback delay, resulting in 0.5 dB quality reduction at 100 ms. However, as TBAS operates at radio block level, 1.4 CBAS 38kbps SQBAS normalized quality improvement 0.4 TBAS deviation reference to 0 delay (dB)TBAS 38kbps0.2 1.2 CBAS 60kbps 0 TBAS 60kbps −0.2 −0.4 1 −0.6 −0.8 0.8 −1 0.6 −1.2 0.4 0 0.2 0 50 100 150 200 250 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 feedback delay (ms) normalized throughput saving (a) (b) Figure 9.40 (a) Performance of throughput based adaptation; (b) effects of feedback delay on algorithm performance. Channel characteristics: ave. CIR, 14.4 dB; std. CIR, 4.2 dB. Channel allocation: three timeslots fixed for SQBAS; 3.8 timeslots on average for TBAS

388 Visual Media Coding and Transmission 31 2.5 0.9 0.8 2 0.7 1.5 0.6 mean PSNR loss (dB) cdf 1 0.5 0.5 0.4 0.3 0 TBAS 22.5kbps 0.2 TBAS 30kbps −0.5 TBAS 38kbps 0.1 TBAS 57kbps SQBAS −1 0 01234567 18 20 22 24 26 28 30 32 34 36 38 x 10 3 feedback error rate frame PSNR dB (a) (b) Figure 9.41 (a) Effect of noisy feedback: dotted line, ave. PSNR vs. feedback bit error rate; star marks, individual simulation points; (b) performance of SQBAS and TBAS for transmission of video over a time varying channel with an average CIR of 14.4 dB and a standard deviation of 4.2 dB the expected feedback delay is limited to the duration of a few radio blocks, which is less than 100 ms. Therefore, expected drop in quality is small for both schemes. Effect of Noisy Feedback The effect of noisy feedback on the performance of SQBAS is illustrated in Figure 9.41. The scheme is relatively robust to noisy feedback, with average PSNR degradation being around 0.5 dB. The effect of noisy feedback is similar to the effect of delayed feedback, in the sense that an inaccurate channel estimate for mode decision is made. In SQBAS, the mode decision interval, which equals to the video frame rate, is usually wide enough to accommodate reasonable prediction errors of channel states. Effect of Burst Errors The effect of burst error on the performance of the quality-based adaptation scheme is investigated in terms of standard deviation of frame PSNR over a number of runs for a set of simulated time-varying channels. The calculated standard deviation of frame PSNR is averaged over 1500 frames in each case, and results are depicted in Table 9.24. Fixed-rate MCS-1 is most robust to the burst errors, while MCS-5 shows least robustness. Adaptive algorithms show moderate robustness; however, due to the averaging property in the channel- estimation algorithm, SQBAS is more robust than CBAS. Table 9.24 Effect of burst channel errors Std (dB) Method 0.7516 1.5908 MCS 1 1.4999 MCS 5 1.0294 CBAS, variable rate SQBAS

Enhancement Schemes for Multimedia Transmission over Wireless Networks 389 Performance Comparison The video source rates and the channel-switching thresholds for TBAS were selected in such a way as to achieve on average three-timeslot allocation, thereby allowing easy comparison with SQBAS, which guarantees fixed three-timeslot operation. Figure 9.41(b) shows that the performance of SQBAS is much better than that of TBAS. The figure also shows the effect of source rates and channel coding rates on video performances over error-prone fixed-bandwidth channels. The video quality decreases with increase of source rate as provided channel protection decreases. 9.3.3 Link Adaptation for Streaming Video Communications Ideally, the design of link-adaptation algorithms for multimedia services should optimize the delivery of acceptable video quality, facilitating good-quality video services with the minimum possible demand on network resources. The previous section showed that appropriate design of link-adaptation algorithms results in improved perceptual quality for one-to-one conversa- tional services compared to non-adaptive schemes. In this section, a method of applying a link- adaptation technique to streaming (one-to-many) services is proposed. To achieve link adaptation for streaming applications, the algorithm should ideally be able to switch between a number of source coding rates without performing custom encoding for each user. One way to do this would be to use scalable coding. However, scalable coding tends to be much less compression-efficient than a single stream [32], and is therefore not very efficient in low-bit rate channels. Therefore, to improve the efficiency of the link adaptation, the scheme proposed in this paper switches between pre-encoded streams stored at the server. No extra user interaction is required and the link-adaptation algorithm is transparent to the end user. Figure 9.42 shows how the EGPRS packet-switched system architecture is modified for this proposal. The most important service entities for streaming are the content server and the streaming client, which together with the EGPRS core network entities ensure correct media delivery to the user. Link-quality and QoS-profile storage is the element that must be added to the system to provide link adaptation. It stores the most current measurements provided by the Figure 9.42 The system architecture: EGPRS packet switched streaming services

390 Visual Media Coding and Transmission base station controller (BSC) regarding the time-varying radio link. In addition, it stores hand- over and requested QoS parameters for each individual user. This recorded information can be used by a link-adaptation algorithm. 9.3.3.1 Stream-switched Link Adaptation The proposed link-adaptation algorithm is designed to improve video quality by varying the source rate and the channel-coding scheme according to measured channel condition, while using a fixed number of EGPRS timeslots. This means that although the source rate is allowed to vary, the allocated resources across the radio link do not change. This subsection describes how streams can be switched at the server, and how the link-adaptation algorithm operates. Stream Switching The most obvious way of providing different source rates for each user would be to use scalability. However, this is inefficient in terms of compression. An alternative is to encode ‘‘on-the-fly’’ for each user, which requires significant computation, and does not scale easily. The approach proposed here involves the use of pre-stored streams on the server (see Figure 9.43). The server buffers encoded video frames from each stream, and transmits the frame corresponding to the rate specified by the link-adaptation algorithm. Use of the AIR technique is critical to the success of this switching method. When switching is performed between streams, there is a mismatch between the encoder and decoder in terms of the reference frames used to predict future frames. This can potentially lead to drift [33]. However, the 1AIR technique limits the effects and prevents this from becoming a detectable problem. Video bit stream Content server buffer Video bit stream VFn−1 Video bit stream VFn+1 VFn VFn+2 VFn+1 VFn−2 VFn+3 VFn+2 VFn−1 VFn+4 VFn+3 VFn VFn+5 VFn+1 VFn+2 VFn+3 VFn+4 Bn+1.m+1 .VFn−1 Bn.m .VFn Bn+1.m .VFn+1 Bn.m VFn+2 Stream-switched video stream transmitted to streaming client. Figure 9.43 Example of stream switched link adaptation, where VFn is the nth video frame in the sequence, and B is the buffer

Enhancement Schemes for Multimedia Transmission over Wireless Networks 391 Link-adaptation Algorithm M  N buffers are allocated for each video sequence in the content server, where N is the number of timeslots in a radio frame (8 for EGPRS) and M is the number of modulation coding schemes (MCSs) supported (9 for EGPRS). Buffers are labeled: Bn;m ¼ where 1 n N and 1 m M ð9:32Þ In practice, it is not necessary to use all possible combinations of timeslots and MCSs. The minimum possible source rate at which the high-motion Foreman sequence can be encoded is 27 kbps. Therefore, MCS-1, which supports 22.5 kbps with three timeslots, cannot be used in the adaptive scheme for three-timeslot operation. Thus, MCS-2 and MCS-5 are used in the algorithm implementation. According to the video bit rate, Ri, and assuming MCS-5 is used, n number of timeslots, TSn,i, are selected for user i. As can be seen from Table 9.22, for selected TSn,i there are M different video source rates (corresponding to different MCS schemes) that can be applied. The link- adaptation algorithm is designed to adapt each radio link to one of M MCS schemes. Let CIRk,i be the measured channel condition at the kth radio block for user i. Modulation coding scheme mode m and video source rate Rn,m are chosen if: CIRk;i 2 ½jthðmÞ; jthðm þ 1ފ ð9:33Þ where jth(m) and jth(m þ 1) indicate corresponding channel threshold values. The main steps of the proposed link-adaptation algorithm are listed below: 1. Select n number of timeslots needed to satisfy the user’s source bit rate requirement. 2. Check for the start of the jth video frame. If start, go to step 3. 3. Estimate the channel condition for the next radio frame from the channel condition measurements. Check for channel-code switching conditions. Select the video source rate Rn,m and the MCS for the jth video frame according to the estimated channel condition. 4. Switch to buffer Bn,m and use the frame header information to find the start of the jth video frame. 5. Transmit data for the jth video frame from buffer Bn,m with modulation coding scheme m using n timeslots of the current radio frame. 6. If there are more video frames to be transmitted, return to step 2. The sequences are MPEG-4 coded using all of the error-resilience options and with AIR enabled. Sequences are compressed with different non-varying bit rate outputs using the TM5 rate-control algorithm. TM5 varies the quantization to ensure a particular target rate is achieved and does not drop frames. To achieve the lower bit rates, it is necessary to reduce the number of nonpredictive AIR macroblocks per frame. This means that while 10 AIR MBs are used for MCS-5, only 5 are used for MCS-2. 9.3.3.2 Results and Discussion CIR-based Link-adaptation Algorithm Tests were performed using the actual CIR value of the communications link. Accurate determination of the CIR is difficult in a practical system, so the performance of this link- adaptation algorithm scheme should be seen as ideal. No delay or feedback corruption was used.

392 Visual Media Coding and Transmission Figure 9.44 Simulation results for CIR based link adaptation algorithm: (a) Suzie; (b) Foreman The test results are shown in Figure 9.44. It should be noted that the CIR value on the x-axis of the graphs is the average CIR of a time-varying channel. The results clearly show that an ideal link-adaptation algorithm outperforms both of the fixed schemes over almost the entire CIR range. Figure 9.44 gives an indication of the difference in quality that can be expected from using the link-adaptation algorithm. The CIR link-adaptation algorithm improves PSNR by more than 2 dB for certain channel conditions. Another noticeable element of the results in Figure 9.44 is the similar performance of MCS-2 to MCS-5 at low average CIRs. This is due to the lower error-recovery properties of the MCS-2 bitstream: each frame contains fewer AIR blocks than in the MCS-5 sequence. BLER-based link-adaptation algorithm Further tests were carried out using BLER-based measurement of channel quality, which is a more practical measurement method for implementation. The results in Figure 9.45 show that the performance of the BLER link-adaptation algorithm is only slightly lower than that of the ideal CIR link-adaptation algorithm. For high average CIR, the BLER link-adaptation algorithm outperforms the CIR link-adaptation algorithm with Suzie. The gain in performance of the BLER link-adaptation algorithm over the fixed schemes is significant, and provides only slightly lower performance gain than the CIR link-adaptation algorithm. Sensitivity of Switching Threshold The switching threshold used in the experiments here was determined experimentally. However, the optimum threshold was slightly different for the Suzie and Foreman sequences, indicating that threshold switching can only be performed optimally by considering the content. In a practical implementation this is not feasible. Because precise optimization of the switching threshold according to the encoded content is not possible, it is important to examine the link-adaptation algorithm scheme’s sensitivity to different threshold values. Figure 9.46 shows the results of tests conducted using different switching thresholds for the CIR-based link-adaptation algorithm. For the Suzie sequence, the results show that 22.5 and 21.5 dB are both good choices for the switching threshold. For Foreman, the situation is more

Enhancement Schemes for Multimedia Transmission over Wireless Networks 393 Figure 9.45 Simulation results for BLER based link adaptation algorithm: (a) Suzie; (b) Foreman complex, as the best-performing threshold appears to change depending on the average CIR. However, the thresholds of 22.5 and 21.5 dB appear to be good choices. The results show that there is some tolerance for setting a suboptimal threshold value, and that it is preferable to set a threshold value slightly lower than optimal, rather than slightly higher. Sensitivity to Feedback Delay and Errors The BLER link-adaptation algorithm described here uses channel prediction estimates to minimize the effects of delay on the link-adaptation performance. Tests were conducted using a variety of delay values to evaluate the scheme’s sensitivity to delay. Additional tests were performed to evaluate the effects of feedback corruption on system performance by simulating Figure 9.46 Simulation results using different threshold values for CIR link adaptation algorithm switching: (a) Suzie; (b) Foreman

394 Visual Media Coding and Transmission Figure 9.47 Simulation results, at an average CIR of around 20 dB, showing the sensitivity of the BLER link adaptation algorithm to delay and error: (a) feedback delay (no feedback corruption); (b) feedback corruption (140 ms delay) transmission of the feedback bitstream using the same channel model described above. A delay of 140 ms was used in these additional tests. Figure 9.47(a) shows the results of feedback delay on the Suzie and Foreman sequences. The results differ slightly between the two sequences, but it is clear that the algorithm remains beneficial in the face of typical EGPRS network delays. However, even with this scheme it is preferable that delay is kept as low as possible. Figure 9.47(b) shows only minor variation with increasing BER. Comparisons with Power-control Techniques Power-control techniques for TDMA radio access systems typically use the same measurements as those used for the link-adaptation scheme presented here. This means that it is possible to use both schemes on the same transmission, where a decision would have to be made on whether to switch streams or change the power settings. In this section, the performance of the link-adaptation algorithm is compared to that of power-control techniques, and comments are made concerning the possible improvements that can be achieved by combining power control and link adaptation. The simple power-control algorithm implemented for this task closely follows the feedback- error-rate power-control algorithm proposed in [34]. However, instead of using the feedback error rate as an indication of channel quality, here the actual channel carrier-to-interference ratio is used in making the power-stepping decision. Therefore, the achieved performance should be considered as ideal. Another simplification of the algorithm compared to [34] is the use of a fixed power-stepping decision threshold and power-step size. Similar to [34], the dynamic range of the algorithm is limited by two settings: the maximum power (30 dBm) and the power transmission dynamic range (64 dB). An example of the performance of the power control algorithm is shown in Figure 9.48. The figure shows the CIR of the channel that would be experienced by the MS in the absence of power-control techniques. For the same channel, it also shows the channel conditions experienced by the MS and the transmission power variation in the power-control scenario.

Enhancement Schemes for Multimedia Transmission over Wireless Networks 395 Figure 9.48 Example power control scenario: (a) comparison of CIR at MS with and without power control; (b) base station transmission power Power control simulations were carried out using fixed MCS-2 and MCS-5. Figure 9.49 shows the power-control results compared to the CIR link-adaptation algorithm. The results show that CIR link-adaptation algorithm performance is similar that to the best-case ideal power-control scenario for average channel CIR value up to 24 dB. However, at high CIR Figure 9.49 Comparison of the CIR link adaptation algorithm with power control

396 Visual Media Coding and Transmission values the power-control technique outperforms the link-adaptation technique. This is due to the fact that the power-control algorithm tends to prevent significant periods of low CIR, while the link-adaptation algorithm switches to MCS-2 at low CIR values, reducing the effective source bit rate. However, for channel conditions in the range of 12 24 dB, which is the average channel condition experienced by most mobile users, both techniques perform similarly. If only one scheme is implemented, the results appear to make the link-adaptation algorithm a more attractive proposition, given the problems with power-control mechanisms (e.g. reduced system capacity due to increased interference with other users). However, the results also suggest that combining both schemes would result in an improved link-adaptation scheme. 9.3.4 Link Adaptation for UMTS Adaptive modulation coding schemes provide a powerful means of exploiting the time-varying channel quality fluctuations of wireless channels. Spreading gain provides the key variable in determining user data rates and associated channel quality in CDMA-based communication systems. Therefore, in addition to channel coding schemes, adaptive spreading gain can also be used to exploit time-varying channels in CDMA systems. In this section, performance gain, which can be achieved by means of adaptive spreading gain control for real-time video communications over UMTS, will be discussed. Further perceptual quality enhancement is achieved via the joint application of unequal error-protection techniques and adaptive spreading gain control. A variety of adaptive rate schemes have been proposed in the literature for CDMA-based communications systems. Spread adaptive quadrature amplitude modulation was proposed in [36]. It exploited the time-variant channel quality of mobile channels by switching either the modulation mode or the spreading factor on a burst-by-burst basis. The multi-user joint detector and the successive interference cancellation receiver gain were analyzed and compared in the context of these adaptive schemes. Bit-rate adaptation with the aim of solving local coverage problems for uplink transmission was discussed in [37]. Evaluations were based on the estimated BLER experienced by the user, and the mobile output power. System-level simulation showed that an increment in cell coverage can be achieved with bit-rate adaptation while maintaining a requested BLER. Also, if the coverage is fixed then link quality improvement can be gained for uplink transmission. A dynamic spreading code assignment algorithm, which efficiently shares a WCDMA downlink between data traffic sources, and different QoS requirements have been presented in [38]. Both analytical and simulation results showed that the dynamic code allocation algorithm provides higher bandwidth utilization than a non-adaptive code allocation scheme. An adaptive rate and power allocation algorithm for uplink throughput maximization has been proposed in [39]. Results conclude that the optimum rate and power allocation performs significantly better than a scheme that uses power adaptation alone. The above-mentioned spreading gain control-related research was largely carried out at the system level. The main goal of the research was the improvement of system-level perfor- mances, which can be categorized into: . system capacity maximization . system throughput maximization . quality improvement in terms of average channel BER or BLER

Enhancement Schemes for Multimedia Transmission over Wireless Networks 397 . service flexibility and service multiplexing . higher system utilization . cell coverage. However, little attention was given to the application-level performances, such as received video quality in multimedia communications. As shown earlier in this chapter, perceptual video quality is a function of quantization distortion, concealment distortion, and distortion due to error propagation; thus received video quality greatly depends on the encoder format, error- resilience techniques, and error-concealment techniques applied. Therefore, it is necessary to investigate the effect of spreading gain control on multimedia performance at the application level, and to produce an optimum scheme for video application. 9.3.4.1 Adaptive Spreading Gain Control for Real-time Video Communications Adaptive spreading gain control techniques attempt to improve the received video quality by switching between different spreading codes levels depending on the state of the transmission channel. Source bit rate is varied according to the selected spreading factor within the TTI, while keeping the chip rate constant. In good channel conditions, quantization distortion becomes the dominating factor in received video quality. Therefore, in order to reduce the quantization distortion, a code with a low spreading factor, which supports a higher source rate, is selected in favorable channel conditions. Conversely, in hostile channel conditions a high spreading factor is used to minimize the channel distortion. The rate-switching threshold is selected according to the link-level simulation results. The transmission power is kept at a constant level for a certain channel SNR. This ensures that the interference power experienced by other users is not affected by the adaptive spreading gain control techniques. Eb/No and SNR are interrelated, and the derivation of SNR from Eb/No is as shown in Equation (9.34): SNR ¼ R Á Eb ð9:34Þ W Á No where W and R denote the channel bandwidth and source rate, respectively. Figure 9.50 shows video performance over UMTS channel in terms of average PSNR vs. SNR. As the figure illustrates, the condition used to switch between spreading factors (hence source rates) is set according to received video quality:  SF16 for SNR ! À 6 dB SF32 for SNR < À 6 dB Switching mode ¼ ð9:35Þ Due to the varying value of R, the spreading gain control technique results in a variation of transmit bit energy (Eb) over the duration of transmission. Selected transmission modes and corresponding source rates are listed in Table 9.25. Experiment and Results Similarly to the EGPRS time-varying channel model, a time-varying channel model is developed for the UMTS downlink. Table 9.26 lists the simulation parameters used. A detailed description of the time-varying channel model for UMTS considering multi-user scenarios is given Section 9.4.

398 Visual Media Coding and Transmission MPEG4 performance over Veh A cc 1/3 no pc, 10 fps 40 35 30 mean PSNR (dB) 25 20 15 10 sf32 sf16 5 sf8 −16 −14 −12 −10 −8 −6 −4 −2 02 S/N (dB) Figure 9.50 Video performance over vehicular A channel with 1/3 rate convolutional channel coding Perfect SNR estimation is assumed. Therefore, the obtained results show an upper bound of performance estimates. Full-error-resilience MPEG-4-coded video sequences were transmit- ted over simulated time-varying channels. The performance results in terms of average frame PSNR vs. average channel SNR (of time-varying channel) are presented in Figure 9.51. Each point is produced by averaging frame PSNR values for 50 different runs over at least 15 different channel profiles with the same mean SNR. For comparison, the performances of the non-adaptive schemes with spreading factors 32 and 16 are also shown in the figure. The adaptive spreading gain control scheme shows significant quality improvement for the transmission of both low- and high-motion video sequences. However, higher gain is achieved Table 9.25 Characteristics of transmission modes Mode 2 Mode 1 SF 16 Spreading Factor SF 32 137.4 kbps Source Rate 64.5 kbps Table 9.26 Parameter values used in UMTS channel simulation Veh A 50 km/h Propagation Environment 10 dB Mobile Speed 20 m Log Normal Variance 2 km De correlation Length Raleigh fading Hexagonal Cell Radius 0.6 Fading Characteristics CC1/3 Othorganality Factor Channel Coding

Enhancement Schemes for Multimedia Transmission over Wireless Networks 399 40average frame PSNR (dB) 34 sf=32 38 average frame PSNR (dB) 32 sf=16 36 sf=32 30 Adaptive scheme 34 sf=16 28 32 Adaptive scheme 26 30 24 28 −10 −5 0 5 10 15 22 −10 −5 0 5 10 15 26 20 average channel SNR (dB) 24 18 22 −15 (b) −15 average channel SNR (dB) (a) Figure 9.51 Performance of adaptive spreading gain control scheme: (a) Suzie; (b) Foreman for the low-motion sequence. As the theory suggests, the performance of the adaptive scheme gets closer to that of spreading factor 16 operation in good channel conditions. With poor channel conditions, the performance gets closer to that of spreading factor 32 operation. 9.3.4.2 Joint Adaptive Spreading Gain Control and Unequal Error-protection Scheme (JAS-UEP) Unequal error-protection schemes exploit the different levels of importance of transmitted data for perceptual quality, and apply different levels of channel protection accordingly in order to achieve maximum received quality for given channel conditions. The link-adaptation tech- niques, on the other hand, exploit time-varying channel quality by adaptive control of the transmission mode. Logically, a combination of these two techniques should result in improved performances. In this section, the combination of adaptive spreading gain control and unequal error-protection schemes is examined for real-time video communications over UMTS systems. An MPEG-4 data partition-based UEP scheme is combined with the developed adaptive spreading gain control scheme. UEP supports higher source bit rates than the 1/3 rate convolutional coding scheme. As given in Equation (9.34), channel SNR increases with increase of source rate. In order to maintain the same SNR, information bits should be transmitted with lower energy in a JAS-UEP scheme than in an adaptive spreading gain control scheme. This results in lower video quality for a given channel SNR. This problem is overcome with the use of unequal bit energy allocation for different bearers in the JAS-UEP scheme. The overall joint adaptive spreading gain control and unequal error-protection scheme is depicted in Figure 9.52. Each encoded video frame is separated into two streams based on MPEG-4 data partitioning. The high-priority data stream (the first partition) is channel-protected with 1/3 rate convolu- tional coding, while 1/2 rate convolutional code is used to protect the low-priority data stream (the second partition). Channel-coded streams are allocated with different transmit bit energy levels based on the JAS-UEP decision command. Finally, transport channels are multiplexed into the same physical channel for transmission over the wireless link. Feedback channel

400 Visual Media Coding and Transmission Figure 9.52 Realization of proposed JAS UEP over UMTS information is used to predict the channel conditions. According to the predicted channel conditions, the transmission mode selection decision is made. Two transmission modes are considered. The source rate supportable by the UEP scheme depends on the sequence characteristics, channel bit rate, and ratio between the size of the first partition and the size of the second partition: Rs ¼ ð1 þ wÞ Á Rch Á 1 ð9:36Þ Xch2 þ w Á Xch1 ð1 þ OHÞ where Rs, Rch, 1/Xch1 and 1/Xch2 denote source rate, channel bit rate, channel coding rate of the first stream, and channel coding rate of the second stream, respectively. w is the ratio between the amount of data in the first partition and the amount of data in the second partition.OH is the extra overhead incurred. Average bit energy for the transmission can be calculated from Equation (9.34) as: Eb ¼ SNR ÁW Á No ð9:37Þ Rs Let Eb1 and Eb2 denote the allocated bit energy for stream 1 and stream 2, respectively. In order to satisfy the average bit energy requirement: ð1 w wÞ Á Eb1 þ ð1 1 wÞ Á Eb2 ¼ Eb ð9:38Þ þ þ

Enhancement Schemes for Multimedia Transmission over Wireless Networks 401 Table 9.27 Operation modes of JAS UEP scheme Suzie Video sequence Mode Mode 1 Mode 2 Stream 1 Stream 2 Stream 1 Stream 2 Spreading Factor 32 CC 1/2 16 Source Rate (kbps) 88 Ebm1 1.25 190 Channel Code Bit Energy (dB) CC 1/3 CC 1/3 CC 1/2 Ebm1 þ 2.33 Ebm2 þ 3.15 Ebm2 1.25 Say Eb1 ¼ Eb þ x and Eb2 ¼ Eb À y, then: x ¼ 1 ð9:39Þ y w Based on the experimental results, y is selected to be 0.25Eb. The calculated source rate and transmit bit energy for the two operation modes for transmission of the Suzie sequence are listed in Table 9.27, where Ebm1 and Ebm2 indicate the average bit energy for mode 1 and mode 2, respectively. Experimental Results The experiments carried out in Section 9.3.4.2 are repeated for the transmission of the Suzie sequence over the vehicular A UMTS channel with the application of the JAS-UEP scheme. Results are plotted in Figure 9.53. The JAS-UEP scheme outperforms the adaptive spreading gain control scheme in good channel conditions. However, slightly lower performance is visible compared to the adaptive spreading gain control scheme with poor channel conditions. 40 38 average frame PSNR (dB) 36 34 32 30 28 26 sf 32 sf 16 24 AS JAS UEP 22 −10 −5 0 5 10 15 average channel SNR (dB) Figure 9.53 Performance of the JAS UEP scheme

402 Visual Media Coding and Transmission This is due to the poor performance of MPEG-4 data partition-based UEP schemes in extreme channel conditions. Also, the transmit energy levels on two streams are kept constant for the selected transmission mode independent of the instantaneous channel quality. These energy levels can optimally be controlled according to the characteristics of the channel; thereby optimal performance gain can be obtained over a wide range of channel conditions. 9.3.5 Conclusion This section has explored application-level adaptive techniques, which can be applied to enhance video quality while media is being delivered. A novel link-adaptation scheme, based on measurable quantities in a practical cellular system, was introduced to exploit the time- varying nature of the mobile radio channel. Two modes of operation the source (video) quality-based adaptation scheme (SQBAS) and the throughput-based adaptation scheme (TBAS) were introduced for variable application source rate and fixed application source rate operation, respectively. The algorithms’ performances were evaluated for video communications over EGPRS systems. Our results reveal that when offered with similar traffic loads over similar channel environments, the quality-based scheme can provide noticeable improvements in video quality over the throughput-based scheme and any fixed coding scheme. The throughput-based adaptation scheme shows an approximately linear relationship between the system throughput savings and the video quality improvements. The proposed algorithms based on BLER, RSS, and first-order statistics of RSS perform as well as the CIR-based adaptive scheme. Further- more, the investigation shows that the proposed algorithms are robust against feedback delay, noisy feedback, and bursts of channel errors. These results reveal that proper control of link- adaptation mechanisms can be used to maximize the system throughput while maintaining adequate service quality for real-time video communications over wireless networks. A technique that provides delay-robust link adaptation for streaming video over EGPRS mobile networks was also presented. The proposed scheme uses the AIR technique in MPEG-4 to mitigate the drift effect resulting from stream switching. The switching is performed between two pre-encoded bitstreams, intended for use with two different modulation coding schemes. This removes the need for the encoder decoder interaction normally associated with link adaptation. By using a prediction method for future channel conditions, the technique is also robust to delay in the feedback channel. Tests were performed using EGPRS channel models, comparing the results of fixed modulation coding scheme scenarios to the link- adaptation method. Results show significant quality improvements with the adaptive scheme. Further tests reveal that the feedback data is acceptably robust to channel errors. The CIR-based link-adaptation algorithm is shown to perform similarly to the best-case ideal power control scenario, and the results indicate that combining power control with the proposed link- adaptation algorithm would result in a better scheme than ideal power control. Link-adaptation techniques in UMTS networks were also examined. An adaptive spreading gain control algorithm was proposed and analyzed for real-time video communications. Conducted experiments show significant improvements in received video quality. Further- more, an adaptive spreading gain control scheme was applied in conjunction with an unequal error-protection scheme. It was shown that the joint adaptive spreading gain control and unequal error-protection scheme can be used to achieve improved video performance in wireless networks.

Enhancement Schemes for Multimedia Transmission over Wireless Networks 403 9.4 User-centric Radio Resource Management in UTRAN (Portion reprinted, with permission, from O. Abdul Hameed, S. Nasir, H. Karim, T. Masterton, A.M. Kondoz, ‘‘Enhancing wireless video transmissions in virtual collaboration environ- ments’’, Mobile and Wireless Communications Summit, 2007. 16th IST, vol., no., pp.1 5, 1 5 July 2007. Ó2007 IEEE.) 9.4.1 Enhanced Call-admission Control Scheme In this section, an enhanced power-based downlink call-admission control scheme is presented. The aim of this scheme is to improve the blocking probability for high-priority new call requests such as emergency services call requests. The scheme was designed based on radio resource management works in [57,58]. It takes the user input into account at call-admission control time. The user input is obtained by classifying the new call requests, according to how much they are willing to pay for the services, into three service classes, being Enhanced, Moderate, and Normal. For each new call request, the new call’s service class is generated as 50 % for Normal, 30 % for Moderate, and 20 % for Enhanced. The scheme adopts quality optimization using QoS renegotiation or a pre-agreed service level agreement (SLA) such that lower-priority existing calls can give their radio network resources to higher-priority new call requests. Each user has a service quality profile for each traffic type, such as voice, video, or data, that defines maximum and minimum data rates. The scheme is illustrated in Figure 9.54. A new call request is accepted if there are enough radio network resources available. If not, the new call request’s service class is checked. If it is Enhanced, for example, the network tries first to step down the bit rate of Normal-service-class existing calls by the predefined SLA. If enough resources can be obtained after this step, the call is accepted. If not, the same process is applied on existing Moderate-service-class calls. If enough resources are still not available, the new call request is blocked. 9.4.2 Implementation of UTRAN System-level Simulator A system-level simulator for UMTS terrestrial radio access network (UTRAN) was imple- mented using the C þ þ programming language. The simulator’s features and functionalities are described below. 9.4.2.1 Cell Layout The simulated network area models a macro-cellular environment for the uplink and downlink directions [8]. The network topology consists of 49 hexagonal cells that are laid on a wraparound surface to avoid border effects, and hence each cell will have 6 neighbors, as shown in Figure 9.55. The cell radius is 300 m and an omnidirectional antenna is placed at the center of each cell, with one 5 MHz carrier (2100 MHz band) with reuse of one. 9.4.2.2 Traffic Generation The traffic models adopted in the simulator are for real-time services for voice and video. New call requests arrive at the 49 cell system according to a Poisson process of rate l call requests/ second.

404 Visual Media Coding and Transmission New call request Ptotal ≤ Yes Pmax CACthreshold? Accept No No N ormal ? Moderate ? Yes No Enhanced ? Yes Block Step down bit rate Yes of Normal users Step down bit rate of Normal users Ptotal ≤ Yes Ptotal ≤ CACthreshold ? Yes Pmax Pmax CACthreshold ? No No Step down bit rate Block of Moderate users Ptotal Yes Pmax ≤ CACthreshold? No Block Accept Figure 9.54 Flow chart illustrating the enhanced call admission control scheme For homogeneous load, the call requests (or user arrivals) are uniformly distributed over the simulated network area. For hot-spot simulations, about one quarter of the entire system’s new call requests will be initiated 10 m below the antenna in the hot-spot cell. No user mobility is considered. The inter arrival time between two consecutive call requests is exponentially distributed. The call duration is also exponentially distributed with mean call duration of 180 s. To generate the inter arrival times and the call durations the following approach is used:

Enhancement Schemes for Multimedia Transmission over Wireless Networks 405 42 43 44 45 46 47 48 35 36 37 38 39 40 41 28 29 30 31 32 33 34 21 22 23 24 25 26 27 14 15 16 17 18 19 20 7 8 9 10 11 12 13 0123456 Figure 9.55 Toroidal (wraparound) 49 cell system Let x be a random variable that follows an exponential distribution with rate a. The random variable x can be generated using the following equation: x ¼ À1 logeð1 À uÞ ð9:40Þ a where u is a uniformly-distributed random variable between 0 and 1. To generate the inter arrival times, l call requests/second is used for the rate parameter a. To generate the call duration times, m is used for the rate parameter a. 9.4.2.3 Propagation Model The propagation model adopted in the simulator includes large-scale fading, small-scale fading, and shadow fading. The attenuation between a mobile and the transmit antenna of a cell site, or between a mobile and each of the receiving antennas of a cell site, is modeled by: Attenuation ¼ g2 Á 101s0 Á 1 ð9:41Þ 101L0 where g2 is an exponentially-distributed random variable with unit mean, which accounts for Rayleigh fading; s is a Gaussian random variable with zero mean and a standard deviation of 5 dB, which introduces log normal shadowing due to the terrain irregularities; and L is the path loss. If the power transmitted from a NodeB towards a mobile is PTX, the power received at the mobile is PRX and is given by: PRX ¼ PTX Á Attenuation ð9:42Þ

406 Visual Media Coding and Transmission Fast-fading Model The fast fading is modeled using an exponentially-distributed random variable with unit mean, which accounts for Rayleigh fading [57]. Path-loss Model The path-loss component of the propagation model is calculated according to the path loss model for the vehicular test environment [8] as: L ¼ 128:1 þ 37:6 Á Log10ðRÞ ð9:43Þ where R is the NodeB mobile separation in kilometers. Shadow-fading Model The shadow-fading component of the propagation model is modeled as a Gaussian-distributed random variable with zero mean and standard deviation s. Xi may be expressed as the weighted sum of a component Z common to all cell sites and a component Zi that is independent from one cell site to the next. Both components are assumed to be Gaussian-distributed random variables with zero mean and standard deviation s independent from each other, so that: Xi ¼ aZ þ bZi ð9:44Þ such that a2 þ b2 ¼ 1. Typical parameters are s ¼ 8.9 and a2 ¼ b2 ¼ 1/2 for 50% correlation. The correlation is 0.5 between sectors from different cells, and 1.0 between sectors of the same cell [59]. 9.4.2.4 Interference The interference consists of three parts, being intra-cell interference, inter-cell interference, and thermal noise power. Intra-cell interference IOWN is the interference generated by users that are connected to the same base station. IOWN also includes interference caused by the perch channel and common channels, and their transmission powers are in total equal to 30 dBm for macro-cells [60]. Inter-cell interference IOTHER is the interference generated from the other cells. Thermal noise power N0 is equal to À99 dBm and is calculated for the 4.096 MHz band by assuming a 9 dB system noise figure. 9.4.2.5 Signal-to-interference Ratio (SIR) The signal-to-interference ratio (SIR) in the downlink direction at the user after dispreading is calculated every 10 ms for each active user in the simulator as: SIRDL ¼ a Á IOWN GP Á S þ N0 ð9:45Þ þ IOTHER where S is the received signal, GP is the processing gain (or SF), IOWN is the intra-cell interference, IOTHER is the inter-cell interference, and a is the orthogonality factor, which takes into account the fact that the downlink is not perfectly orthogonal due to multipath propagation. An orthogonality factor of 0 corresponds to perfectly orthogonal intra-cell users, while with a value of 1 the intra-cell interference has the same effect as inter-cell interference.

Enhancement Schemes for Multimedia Transmission over Wireless Networks 407 9.4.2.6 Power Control The downlink traffic channels are power-controlled using a simple SIR-based fast-inner-loop power control in order to maintain the SIR at a target value. Perfect power control is assumed, i. e. during the power-control loop each downlink traffic channel perfectly achieves the Eb/No target, assuming that the maximum NodeB transmission power is not exceeded. With the assumption of perfect power control, power-control error is assumed equal to 0% and power- control delay is assumed to be 0 seconds. UEs whose downlink traffic channel is not able to achieve the Eb/N0 target at the end of a power-control loop are considered in outage. Initial transmission power for the power-control loop of the downlink traffic channel is chosen randomly in the transmission power range. However, the initial transmission power should not affect the convergence process (power-control loop) to the target Eb/N0. Here it is assumed that the power control operates at a frequency of 100 Hz for faster simulations. The new power level is evaluated as: PTXðt ¼ PTXt Á SIRt arg et ð9:46Þ SIR 1Þ where PTX(t 1) is the transmitting power of the link in the (t À 1)th step of the simulation. The maximum power for each downlink traffic channel is 30 dBm in a macro-cellular environ- ment [60], and if the power control requires a power higher than the maximum value, the maximum value is adopted. In addition, if the sum of the powers required by the downlink channels exceeds the NodeB maximum transmission power, all the powers are proportionally reduced to limit the total power at the maximum value. 9.4.2.7 Call-admission Control (CAC) The call-admission control (CAC) scheme used in the simulator was a power-based CAC scheme that estimates the load of the radio network during the admission of a new call request generated by a user. The total power transmitted from the NodeB is a parameter that can be used to estimate the network load in downlink direction. A new call request is accepted as long as the output power level at the consulted NodeB stays below a certain predefined threshold. If the total power transmitted from a NodeB is Ptotal and the maximum downlink transmission power of the NodeB is Pmax, a new call request to this NodeB will be accepted if: Ptotal threshold ð9:47Þ Pmax and will be blocked if: Ptotal 1 threshold ð9:48Þ Pmax The idea of considering the ratio Ptotal/Pmax instead of the absolute threshold value Pthr is based on the assumption that it would be easier in a real system to have a simple tuning criterion based on a relative threshold [57]. The value of the threshold varies in the interval [0,1].

408 Visual Media Coding and Transmission 9.4.2.8 Performance Metrics The system’s performance is evaluated using a number of performance metrics. The following parameters are obtained from the system-level simulator’s output: 1. Number of new call requests. 2. Number of accepted call requests. 3. Number of blocked call requests. 4. Number of quality dropped calls. 5. New-call-blocking probability (Pblocking), which is defined as the number of blocked call requests divided by the number of generated call requests. Blocking occurs if a new call request is denied access to the system. 6. Call-dropping probability (Pdropping), which is defined as the number of dropped calls divided by the number of accepted calls. After each power-control loop the actual SIR values experienced by each connection are evaluated, and dropping occurs when a connection’s SIR is more than 3 dB below the value of SIRtarget. 7. Accepted traffic (erlangs/cell), which is defined as the mean number of active calls per cell. 9.4.2.9 Simulation Techniques The steady-state simulation technique was used for the system-level simulations using the implemented system-level simulator. To gather steady-state simulation output requires statis- tical assurance that the simulation model has reached the steady-state. The main difficulty is to obtain independent simulation runs with the exclusion of the transient period. The two techniques commonly used for steady-state simulation are the method of batch means and the method of independent replication. Neither of these methods is superior in all cases. Their performance depends on the magnitude of the traffic intensity [61]. Method of Batch Means This method involves only one very long simulation run, which is subdivided into an initial transient period and n batches, as shown in Figure 9.56. Each batch is then treated as an independent run of the simulation experiment. No observations are made during the transient period, which is treated as a warm-up interval. Choosing a large batch interval size would effectively lead to independent batches and hence independent runs of the simulation; however, since the number of batches is small, one cannot invoke the central limit theorem to construct the needed confidence interval. On the other hand, choosing a small batch interval size would effectively lead to significant correlation between successive batches, therefore the results cannot be applied in constructing an accurate confidence interval. Suppose you have n equal batches of m observations each. The means of each batch is: P xij ; the sum is over j ¼ 1; 2; . . . ; m meani ¼ m ð9:49Þ The overall estimate is: ð9:50Þ P Estimate ¼ meani ; the sum is over i ¼ 1; 2; . . . ; n n

Enhancement Schemes for Multimedia Transmission over Wireless Networks 409 Figure 9.56 Method of batch means The 100(1 À a/2)% confidence interval using the Z-table (or T-table for n less than, say, 30) is: Estimate Æ Z Á S ð9:51Þ where the variance is: ð9:52Þ S2 ¼ P ðmeani À EstimateÞ2 ; the sum is over I ¼ 1; 2; . . . ; n ðn À 1Þ Method of Independent Replications This method is the most popularly used for systems with a short transient period. This method requires independent runs of the simulation experiment for different initial random seeds of the simulator’s random number generator. For each independent replication of the simulation run its transient period is removed, as shown in Figure 9.57. For the observed intervals after the transient period, data is collected and processed for the point estimates of the performance measure and for its subsequent confidence interval. Figure 9.57 Method of independent replications

410 Visual Media Coding and Transmission Suppose you have n replications with m observations each. The means of each replication is: P xij ; the sum is over j ¼ 1; 2; . . . ; m meani ¼ m ð9:53Þ The overall estimate is: ð9:54Þ P Estimate ¼ meani ; the sum is over i ¼ 1; 2; . . . ; n n The 100(1 À a/2)% confidence interval using the Z-table (or T-table for n less than, say, 30) is: Estimate Æ Z Á S ð9:55Þ where the variance is: P ðmeani À EstimateÞ2 ðn À 1Þ S2 ¼ ; the sum is over i ¼ 1; 2; ...; n ð9:56Þ In the simulations, we have adopted the method of independent replications. 9.4.3 Performance Evaluation of Enhanced CAC Scheme In order to evaluate the performance of the enhanced power-based CAC in the downlink direction, steady-state static system-level simulations were conducted using the implemented system-level simulator. Two kinds of stream services were considered in the simulations, being voice and video. The following traffic scenarios were considered: 1. Voice only: AMR voice service at 12.2 and 5.15 kbps using SFs of 128 and 256, respectively. 2. Video only: H.264/AVC video at 64 and 128 kbps using SFs of 32 and 16, respectively. The simulation time was set to 100 000 timeslots (each timeslot is of 10 ms duration) for each offered traffic value, and each simulation run was repeated 10 times and the average was taken. The results were collected after a warm-up period that was found to be 60 000 timeslots (or equivalently 600 s). The simulation parameters used in the system-level experiments are listed in Table 9.28. Table 9.28 System level simulation parameters WCDMA FDD downlink Vehicular A Radio Access 3.84 Mcps Environment 2.0 GHz Chip Rate 5.0 MHz Carrier Frequency 0.4 Bandwidth 43 dBm Orthogonality Factor 30 dBm Maximum NodeB Transmission Power Perch Channel and Common Channels Power 99 dBm Thermal Noise Power (Downlink) 300 m Cell Radius Mean ¼ 0 dB, standard deviation ¼ 5 dB Log Normal Shadowing

Enhancement Schemes for Multimedia Transmission over Wireless Networks 411 16accepted traffic (erl/cell) 14 12 original_0.5 10 voice, SF=128, CAC_thr=0.5 (validation) video, SF=32, CAC_thr=0.3 8 video, SF=16, CAC_thr=0.2 video, SF=16 to 32, CAC_thr=0.1 to 0.2 (scheme) 6 4 2 0 10 20 30 40 50 offered traffic (erl/cell) Figure 9.58 Accepted traffic versus offered traffic for different traffic types and data rates The first set of experiments was conducted to validate the system-level simulator. The traffic scenario considered was a voice service at 12.2 kbps using an SF of 128 for a CAC threshold value of 0.5. The simulation results are presented in Figure 9.58, which shows the accepted traffic versus the offered traffic curves in erlangs/cell for a 49 cell system and two different traffic types, being voice and video. The simulator’s performance was validated with [56]. The top two curves represent the accepted traffic in erlangs/cell versus the offered traffic in erlangs/cell for an all-voice traffic scenario at 12.2 kbps. The ‘‘original 0.5’’ in the figure’s legend represents the performance obtained in [56], whereas the ‘‘voice, SF ¼ 128, CAC thr ¼ 0.5 (validation)’’ represents the performance obtained using the implemented simulator. The bottom three curves are for video-only traffic at different data rates. The top one is for video traffic at 64 kbps, the bottom one for video at 128 kbps, whereas the middle one is obtained when using the scheme. The figure shows that, using the scheme, the accepted traffic is improved (or equivalently, lower blocking probability has been achieved) for video traffic users that have been classified into the three service classes. Therefore, the scheme can be used to provide lower blocking probability for the high-priority users, achieving the expected end-to- end QoS. 9.5 Conclusions The capabilities of 3G networks and the suitability of different radio bearer configurations for supporting real-time video and voice communications were given in Section 9.2. A bearer configuration with a spreading factor of 8 is clearly unsuitable for video applications due to high channel-induced errors and high transmission-power requirements. The lower bound of Eb/No that could support acceptable video quality is found to be 8.5 10 dB for a spreading factor of 8 with a 1/3 rate convolutional code channel in vehicular and pedestrian environments. This is not achievable in a multi-user system, unless very sophisticated diversity techniques and interference-suppression techniques are employed. The minimum Eb/No requirement was

412 Visual Media Coding and Transmission reduced down to 6.3 8 dB for spreading factor 32 and 16 channels. Fast power control can lower the minimum Eb/No requirements further by 0.8 3 dB in tested operating environments. Perceptual quality measurements show that spreading factor 32 outperforms others for video transmission over similar operating conditions. However, this limits the operating source rate to 64 and 97 kbps with 1/3 and 1/2 rate channel coding, respectively. System-level perfor- mances demonstrate the necessity of advanced antenna techniques, adaptive resource alloca- tion, and combined source- and channel-quality enhancement techniques in achieving 95% user satisfaction for real-time video services even for source rates as low as 64 kbps. In addition to the investigation of capabilities of 3G networks for the transmission of video and voice, perceptual video quality enhancement methods that exploit the adaptive source and network parameters for varying transmission conditions were explored in the work carried out in this chapter. The most efficient way to use limited radio resources is the deployment of class- or priority-based channel allocation. In this way, the most important information is transmitted over a high-priority channel, while a low-priority channel is used to transport low-priority data. The syntax of the video format and the video representation can be used in information prioritization. Data partition formatting and object-based video coding provide simple methods of data separation in multi-priority transmission. However, a perceptual quality estimation-based data separation algorithm provides an adaptive and flexible information prioritization mechanism which results in optimal video performance. The priority of the transmission channel can in fact be based on the channel protection, in terms of channel coding rate, modulation, spreading factor, and transmission-power allocation. A combination of these channel parameters provides optimal prioritized transmission. In addition, link adaptation, where source and network parameters are adjusted according to the time-varying channel condition at the receiver, is necessary in maintaining the perceptual video quality received by the end user. As discussed in Section 9.4, link adaptation can be designed either to enhance the received video quality for a given network configuration or to enhance the system performance in terms of system capacity or coverage, while guaranteeing end users’ QoS requirements. Either way provides improved performances compared to non- adaptive transmission schemes. Two link-adaptation algorithms are implemented. One is designed to maximize the received video quality, and is named ‘‘quality-based adaptation scheme’’. The other is designed to maximize the system throughput while guaranteeing a required video quality and is called ‘‘throughput-based adaptation scheme’’. Source- and network-parameter adaptation is conducted according to the estimated instantaneous channel quality, based on measured channel BLER, RSS, and the first-order statistics of RSS. Algorithm performances are demonstrated for video telephony over an EGPRS network. The adaptive schemes showed better performances than those of non-adaptive schemes. When offered similar traffic load and channel environments, the quality-based adaptation scheme outperforms the throughput-based scheme. Furthermore, the investigation shows that the proposed algorithms are robust against feedback delay, noisy feedback, and burst channel errors. The above-described link-adaptation algorithms are not suitable for video streaming applications as they require separate encoding of video sequences for each user. To overcome this problem, the video sequence is encoded at different output bit rates, which are stored in a buffer at the server. Link adaptation is performed by switching between pre-encoded streams according to the instantaneous channel condition. When switching is performed between streams, there is a mismatch between the encoder and decoder in terms of the reference frames used to predict future frames. This can potentially lead to drift. However, the proposed scheme

Enhancement Schemes for Multimedia Transmission over Wireless Networks 413 uses an appropriate selection of intra-coded blocks (AIR technique) in video frames to limit the drift effects, and prevents this becoming a detectable problem. Experiments performed using EGPRS channel models show significant quality improvements for the adaptive scheme compared with the results for fixed modulation coding schemes. Spreading gain provides the key variable in determining user data rates and associated channel quality in CDMA-based communication systems. Therefore, in addition to the channel coding schemes, adaptive spreading gain can be used to exploit time-varying channels in CDMA systems. Link-adaptation techniques, based on adaptive spreading gain control, are proposed and analyzed for real-time video communications in UMTS networks. Source rate and spreading code levels were varied depending on the state of the transmission channel. Adaptation is based on the actual channel signal-to-interference ratio, which is calculated at every TTI. The transmission power is kept at a constant level and the transmission bit energy is adjusted according to the selected spreading code level. This ensures that the interference power experienced by other users does not affect the adaptive spreading gain control techniques. The conducted experiments show 2 3 dB frame PSNR improvement compared to non-adaptive schemes. Further performance improvements are achieved by the combined application of adaptive spreading gain control and the unequal error-protection scheme. A joint system-link UMTS simulator, which combines a system-level simulator and the developed link-level simulator, is designed for the analysis of video performances in a multi- user downlink UMTS system. Video performances are investigated for the transmission of fully error resilience-enabled MPEG-4-coded video under constraints on the base station total transmit power (power budget) and the number of codes available per carrier (code budget). Video performance is measured in terms of the average frame PSNR of received video by individual users. System performances are shown in terms of the mean video quality, which is the average video quality received by users, and the number of satisfied users in the system. Three different resource-allocation schemes are implemented. Scheme 1 is a non-adaptive scheme, and it allocates equal transmit power for video users who require the same QoS. Scheme 2 adapts the allocated transmit power for each user according to their individual received channel quality. Scheme 3 combines an adaptive spreading gain control scheme with Scheme 2 to maximize the video quality received by individual users in the system. Experi- ments conducted over the simulated system show about 2 3 dB quality improvement in terms of average PSNR with adaptive power allocation (Scheme 2) compared to fixed resource allocation. Further performance enhancement, in terms of average quality and user satisfaction, is achieved with Scheme 3. Finally, an enhanced user-centric CAC scheme in the downlink direction of a UMTS system was presented and evaluated using an implemented and validated WCDMA UMTS UTRAN system-level simulator. The scheme can be used to achieve the expected end-to-end QoS for an Enhanced-service-class user, providing lower blocking probability. References [1] 3rd Generation Partnership Project, ‘‘Services and service capabilities: release 4,’’ TS 22.105, v. 4.3.0., Mar. 2002. [2] 3rd Generation Partnership Project, ‘‘Technical specification group services and system aspects: quality of service (QoS) concept and architecture: release 4,’’ TS 23.107, v. 4.6.0., Jan. 2003. [3] 3rd Generation Partnership Project, ‘‘Technical specification group terminals: common test environments for user equipment (UE) conformance testing: release 4,’’ TS 34.108, v. 4.7.0., Jun. 2003.

414 Visual Media Coding and Transmission [4] T. V. Lakshman, A. Ortega, and R. Reibman, ‘‘VBR video: tradeoffs and potentials,’’ Proc. IEEE, Vol. 86, No. 5, 952 973, May 1998. [5] 3rd Generation Partnership Project, ‘‘Codec for circuit switched multimedia telephony service: general descrip tion: release 4,’’ TS 26.110, v. 4.1.0., Apr. 2001. [6] 3rd Generation Partnership Project, ‘‘Packet switched conversational multimedia applications: default codecs: release 4,’’ TS 26.235, v. 4.0.0., Apr. 2001. [7] 3rd Generation Partnership Project, ‘‘Technical specification group terminals: medium access control (MAC) protocol specification: release 4,’’ TS 25.321, v. 4.7.0., Jan. 2003. [8] 3rd Generation Partnership Project, ‘‘Universal mobile telecommunications system (UMTS): selection proce dures for the choice of radio transmission technologies of the UMTS (UMTS 30.03 version 3.2.0),’’ TR 101.112, v. 3.2.0., Apr. 1998. [9] L. D. Soares, S. Adachi, and F. Pereira, ‘‘Influence of encoder parameters on the decoded video quality for MPEG 4 over W CDMA mobile networks,’’ International Conference on Image Processing, Vol. 2, pp. 148 151, Sep. 2000. [10] 3rd Generation Partnership Project, ‘‘Internet protocol (IP) multimedia call control protocol based on session initiation protocol (SIP) and session description protocol (SDP): stage 3: release 5,’’ TS 24.229, v. 5.0.0., Mar. 2002. [11] Y. Kikuchi, T. Normura, S. Fukunag, Y. Matsui, and H. Kimata,‘‘RTP payload format for MPEG 4 audio/visual streams’’ RFC 3016. [12] 3rd Generation Partnership Project, ‘‘AMR wideband speech codec: general description: release 6,’’ v. 6.0.0. [13] 3rd Generation Partnership Project, ‘‘Performance characterization of the AMR wideband speech codec: release 6,’’ v. 6.0.0. [14] 3rd Generation Partnership Project, ‘‘AMR wideband speech codec: test sequences, release 6,’’ v. 6.0.0. [15] http://www.pesq.org/. [16] 3rd Generation Partnership Project, ‘‘Logical link control (LLC) layer specification,’’ TS 04.64, v. 8.7.0., 2001. [17] Y. Eisenberg, C. E. Luna, T. N. Pappas, R. Berry, and A. K. Katsaggelos, ‘‘Joint source coding and transmission power management for energy efficient wireless video communications,’’ IEEE Trans. on Circuits and Systems for Video Technology, Vol. 12, No. 6, pp. 411 424, Jun. 2002. [18] I. M. Kim and H. M. Kim, ‘‘An optimum power management scheme for wireless video service in CDMA systems,’’ IEEE Trans. on Wireless Comm., Vol. 2, No. 1, Jan. 2003. [19] ISO/IEC JTC 1/SC 29/WG 11, ‘‘Information technology: generic coding of audio visual objects: part 2: visual,’’ ISO/IEC 14496 2, Jul. 2001. [20] 3rd Generation Partnership Project, ‘‘Physical layer procedures (FDD): release 4,’’ TS 25.214, v. 4.6.0., Apr. 2003. [21] R. Talluri, ‘‘Error resilient video coding in the ISO MPEG 4 standard,’’ IEEE Communications Magazine, pp. 112 119, Jun. 1998. [22] S. Worrall, A. Sadka, and A. Kondoz, ‘‘Optimal packetization of MPEG4 over mobile networks,’’ IEE Proc. Communications, Vol. 148, No. 4, 197 201, Aug. 2001. [23] 3rd Generation Partnership Project, ‘‘GSM/EDGE: radio transmission and reception (phase 2 þ ),’’ TS 05.05, v. 8.10.0., 1999. [24] S. Saunders, Antennas and Propagation for Wireless Communication Systems, John Wiley and Sons, Ltd., 1999. [25] C.C. Lee and R. Steele, ‘‘Signal to interference calculations for modern TDMA cellular communication systems,’’ IEEE Proc. Comms., Vol. 142, No. 1, Feb. 1995. [26] A. Furuskar, S. Mazur, F. Muller, and H. Olofsson, ‘‘EDGE, enhanced data rates for GSM and TDMA/136 evolution,’’ Ericsson Review, No. 1, 1999. [27] 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. Circuits, Sys. Video Tech., Vol. 10, pp. 1355 1363, Dec. 2000. [28] N.B. Mehta and A.J. Goldsmith, ‘‘Performance analysis of link adaptation in wireless data networks,’’ Proc. IEEE GLOBECOM ’00, pp. 1422 1426, 2000. [29] F. Tang, L. Deneire, M. Engels, and M. Moonen, ‘‘A general optimal switching scheme for link adaptation,’’ Proc. IEEE VTC ’01, pp. 1598 1602, 2001. [30] 3rd Generation Partnership Project, ‘‘GSM/EDGE: overall description of the GPRS radio interface: stage 2,’’ TS 03.64, v. 8.10.0., 2002. [31] S.N. Fabri,‘‘Multimedia communications over mobile packet networks,’’ PhD Thesis, University of Surrey, UK, 2001.

Enhancement Schemes for Multimedia Transmission over Wireless Networks 415 [32] X. Sun, F. Wu, S. Li, W. Gao, and Y. Zhang, ‘‘Macroblock based progressive fine granularity scalable (PFGS) video coding with flexible temporal SNR scalabilities,’’ Proc. International Conference on Image Processing, Vol. 2, pp. 1025 1028, Oct. 2001. [33] S. Dogan, A.H. Sadka, and A.M. Kondoz, ‘‘MPEG 4 video transcoder for mobile multimedia traffic planning,’’ IEE Conf. 3G Mobile Comm. Tech., London, UK, pp. 109 113, Mar. 2001. [34] P. Cherriman and L. Hanzo, ‘‘Power controlled H.263 based wireless videophone performance in interference limited scenarios,’’ IEEE PIMRC’96, Vol. 1, pp. 158 162, 1996. [35] S.W. Kim, ‘‘Adaptive rate and power DS/CDMA communications in fading channels,’’ Wireless Communica tions and Networking Conference, Vol. 3, pp. 1563 1566, Sep. 1999. [36] E.L. Kuan and L. Hanzo, ‘‘Comparative study of adaptive rate CDMA transmission employing joint detection and interference cancellation,’’ Proc. IEEE 51st Vehicular Technology Conference, Vol. 1, pp. 71 75, May 2000. [37] A. Fiorini ‘‘Uplink user bitrate adaptation for packet data transmissions in WCDMA,’’ IEEE 11th International Symposium on Personal, Indoor and Mobile Radio Communications, Vol. 2, pp. 1510 1514, Sep. 2000. [38] C.E. Fossa and N.J. Davis, ‘‘Dynamic code assignment improves channel utilization for bursty traffic in third generation wireless networks,’’ IEEE International Conference on Communications, Vol. 5, pp. 3061 3065, May 2002. [39] S.A. Jafar and A. Goldsmith, ‘‘Adaptive multirate CDMA for uplink throughput maximization,’’ IEEE Transactions on Wireless Communications, Vol. 2, No. 2, 218 228, Mar. 2003. [40] H. Holma and A. Toskala, WCDMA for UMTS: Radio Access for Third Generation Mobile Communications, John Wiley and Sons, Ltd., Revised Edition, 2001. [41] M. Amadei, U. Manzoli, and M.L. Merani, ‘‘On the assignment of Walsh and quasi orthogonal codes in a multi carrier DS CDMA system with multiple classes of users,’’ IEEE Global Telecommunications Conference, Vol. 1, pp. 841 845, Nov. 2002. [42] S. Oh and K.M. Wasserman, ‘‘Dynamic spreading gain control in multiservice CDMA networks,’’ IEEE Journal on Selected Areas in Communications, Vol. 17, No. 5, 918 927, May 1999. [43] Y. Lu, and R.W. Bordersen, ‘‘Integrating power control, error correction coding and scheduling for a CDMA downlink system,’’ IEEE Journal on Selected Areas in Communications, Vol. 17, No. 5, 978 989, Jun. 1999. [44] J.B. Kim, and M.L. Honig, ‘‘Resource allocation for multiple classes of DS CDMA traffic,’’ IEEE Transactions on Vehicular Technology, Vol. 49, No. 2, 506 519, Mar. 2000. [45] D.I. Kim, E. Hossain, and V.K. Bhargava, ‘‘Downlink joint rate and power allocation in cellular multirate WCDMA systems,’’ IEEE Transactions on Wireless Communications, Vol. 2 No. 1, pp. 69 80, Jan. 2003. [46] R. Fantacci and S. Nannicini, ‘‘Multiple access protocol for integration of variable bit rate multimedia traffic in UMTS/IMT 2000 based on wideband CDMA,’’ IEEE Journal on Selected Areas in Communications, Vol. 18, No. 8, pp. 1441 1454, Aug. 2000. [47] R. Hang, W.A. Krzymien, and C.L. Despins, ‘‘Improved MAC protocol for reverse link packet data transmission in wideband DS CDMA,’’ IEEE Transactions on Wireless Communications, Vol. 2, No. 1, pp. 162 174, Jan. 2003. [48] J. Perez Romero, O. Sallent, R. Agusti, and G. Pares, ‘‘A downlink admission control algorithm for UTRA FDD,’’ 4th International Workshop on Mobile and Wireless Communications Network, pp. 18 22, Sep. 2002. [49] W.S. Jeon and D.G. Jeong, ‘‘Call admission control for CDMA mobile communications systems supporting multimedia services,’’ IEEE Transactions on Wireless Communications, Vol. 1, No. 4, pp. 649 659, Oct. 2002. [50] K. Sipila, K. C. Honkasalo, J. Laiho Steffens, and A. Wacker, ‘‘Estimation of capacity and required transmission power of WCDMA downlink based on a downlink pole equation,’’ Proc. IEEE 51st Vehicular Technology Conference, Vol. 2, pp. 1002 1005, May 2000. [51] J. Cho and D. Hong, ‘‘Downlink throughput and fairness analysis in multi rate CDMA systems,’’ Proc. IEEE 53rd Vehicular Technology Conference, Vol. 2, pp. 1001 1006, May 2001. [52] K.K. Leung, P.F. Driessen, K. Chawla, and X. Qiu, ‘‘Link adaptation and power control for streaming services in EGPRS wireless networks,’’ IEEE J. Selected Areas Comms, Vol. 19, No. 10, pp. 2029 2039, Oct. 2001. [53] M.D. Anderson and S. Perreau, ‘‘Robust power control for cdma networks subject to modelization errors,’’ Proc. IEEE International Conference on Acoustics, Speech, and Signal Processing, Vol. 4, pp. IV 161 IV 164, Apr. 2003. [54] T. Minn and K.Y. Siu, ‘‘Dynamic assignment of orthogonal variable spreading factor codes in W CDMA,’’ IEEE Journal on Selected Areas in Communications, Vol. 18, No. 8, pp. 1429 1440, Aug. 2000.

416 Visual Media Coding and Transmission [55] M.D. Amico, M.L. Merani, and F. Maffioli, ‘‘Efficient algorithms for the assignment of OVSF codes in wideband CDMA,’’ IEEE International Conference on Communications, Vol. 5, pp. 3055 3060, Apr. May 2002. [56] R. Assarut, K. Kawanishi, U. Yamamoto, and Y. Onozato, ‘‘Performance evaluation of orthogonal variable spreading factor code assignment schemes based on UMTS/IMT 2000 in W CDMA,’’ IEEE 56th Vehicular Technology Conference, Vol. 3, pp. 1585 1589, Sep. 2002. [57] A. Capone and S. Redana, ‘‘Call admission control techniques for UMTS,’’ Proc. IEEE VTC 2001, Atlantic City, NJ, Oct. 2001. [58] S. E. Kim, H. Kim, and J.A. Copeland, ‘‘Dynamic radio resource allocation considering QoS in WCDMA network,’’ Mobile and Wireless Communications Network, 2002. [59] 3rd Generation Partnership Project, ‘‘CDMA 2000 evaluation methodology,’’ C.R1002 0, v. 1.0., Dec. 2004. [60] 3rd Generation Partnership Project, ‘‘Radio frequency (RF) system scenarios: release 7,’’ TR 25.942, v. 7.0.0., Mar. 2007. [61] http://home.ubalt.edu/ntsbarsh/simulation/sim.htm.

10 Quality Optimization for Cross- network Media Communications 10.1 Introduction Future communication systems are intended to support a wide range of applications that require different levels of quality of service (QoS). Unlike the current single-service network, future multimedia applications will require multiple-service operation in integrated network environments. The beyond-3G communication network concept involves the coexistence of all the different network technologies. Thus, heterogeneity will play a major role in future networks and operating system environments. Highly-differentiated access technologies allow a user terminal to exploit a number of platforms to access heterogeneous services. Application data transmitted over heterogeneous networks will therefore experience systems with limited and varying capacity. Under this condition, the multimedia performance-enhancement algorithms will need to be optimized in order to achieve optimal QoS for multimedia over end-to-end systems. In particular, QoS parameter-mapping will become important as different networks provide different QoS. When multimedia is transmitted from a network with poor performance to a network with good performance, problems will arise in guaranteeing the required quality for the end user. In this chapter, the issue of supporting optimal quality for multimedia applications in heterogeneous network scenarios is addressed. First, a generic QoS infrastructure that guarantees multimedia quality in inter-networked operation is described. The QoS infrastruc- ture is designed in a generic manner to enable the accommodation of evolving wireless communication systems. The proposed infrastructure uses a layered architecture, which enables cross-protocol layer optimization as well as cross-network optimization for multi- media quality. Operation of the proposed infrastructure is demonstrated through the design of a QoS parameter-mapping emulator. Furthermore, MPEG-4-encoded video is transmitted over the simulated heterogeneous environment, and the results show the importance of such a QoS infrastructure in maintaining acceptable multimedia quality for the end user. Visual Media Coding and Transmission Ahmet Kondoz © 2009 John Wiley & Sons, Ltd. ISBN: 978-0-470-74057-6

418 Visual Media Coding and Transmission 10.2 Generic Inter-networked QoS-optimization Infrastructure Third-generation and beyond radio access technologies promise deployment in diverse environments, better quality and coverage, and more power and bandwidth than is possible in present systems, potentially enabling advanced multimedia communications in a cellular system. However, in order to realize widespread use of multimedia applications, several challenges remain to be addressed. A fundamental question is how to provide adequate QoS for multimedia applications, taking into consideration the effects of user mobility, cell handover, network handover, and so on. This section investigates the important issues of QoS provision- ing for multimedia traffic transmitted over heterogeneous wireless networks with adaptive QoS support and with the characteristics of the application taken into consideration. First, a brief overview of recent developments in QoS provisioning in mobile heterogeneous networks, enhancing the multimedia quality received by end users, is provided. Second, the design of a generic inter-networked QoS optimization infrastructure for multimedia delivery over hetero- geneous networks is described. 10.2.1 State of the Art In order to provide QoS in mobile heterogeneous systems, many issues must be addressed. Consider a user moving from one network to another. The user may interact with a variety of service providers with different service level agreements, network capacities, topologies, and policies. They may have a choice of wireless access technologies with different channel characteristics and QoS support capabilities. The user may adapt applications to meet new service requirements or network conditions. These factors can complicate end-to-end service provisioning and limit the ability of service adaptation. Current QoS-related research involves scalable inter-domain QoS architecture design for transmission of multimedia applications over heterogeneous systems [1 4]. Most of the work in this field focuses on developing QoS frameworks. A common approach is to combine advances in wireless, Internet, and reconfigurable technologies in order to achieve seamless service provisioning in heterogeneous, dynamically-varying computing and communication environ- ments. Integrated services (IntServs) and differentiated services (DiffServs) have shown QoS in best-effort Internet services. Some progress has recently been made in addressing wireless- related QoS issues, such as wireless access, mobility management, and portable devices. Even though some progress has been made in the context of individual architecture components, much less progress has been made in addressing the issue of an overall QoS architecture. The soft-QoS or adaptive-QoS control mechanisms have been addressed under a number of research proposals (with various approaches) as a way of maximizing the QoS for multimedia applications in wireless/wired communication systems. These QoS mechanisms can be identified under a number of different categories. The most popular approach is the service class-based adaptation [5,6], where the services are categorized into a number of different classes depending on their requirements. Each class is treated differently in call admission control, resource allocation, and QoS provisioning, in such a way as to achieve better overall service quality. Another approach that is used in adaptive-QoS mechanisms is the feedback-based adaptation [7]. The feedback information may take feedback from the client (user interactivity) regarding the quality of delivery, feedback from the system indicating time-to-time variation of system parameters, and feedback from the network nodes informing of changes in the physical network

Quality Optimization for Cross network Media Communications 419 status, such as channel variation, to assist the servers in making dynamic adjustments to the transmitted streams. Interaction between multimedia codecs and adaptive-QoS mechanisms has been considered in cross-protocol layer QoS optimization for an individual user [8]. However, most of these research efforts have been deficient in three key areas. First, they concentrate only on certain aspects of the overall QoS problem, certain entities of the transmission path, or certain layers of the communication model. Second, the characteristics of multimedia have not been considered for the improvement of the system performance (or QoS management at system level) in a multi-user multiservice environment. Also, media content adaptation and user feedback regarding the received quality have not been widely addressed in the literature for the purpose of QoS management. The content adaptation can be used to enable a user-centered customization of services while enhancing the service quality. Third, and more importantly, the research is focused on the provisioning and maintenance of QoS at network/transport layers and system level, and intends to achieve this by optimizing the objective parameters such as transmission rate, throughput, error rate, and transmission delay which are measured at the lower protocol layers. Little attention is given to the application-level parameters, such as received video quality. Particularly for emerging multimedia applications (e.g. networked virtual reality applications, mixed reality scenarios, and 2D/3D video communications), the received multimedia quality greatly depends not only on the channel characteristics but also on the coding/rendering techniques, encoding parameters, and error-resilience/concealment techniques applied. Therefore, it is necessary to consider media content adaptation, characteristics of the application, and user feedback in the design of QoS control algorithms to enhance the multimedia quality actually perceived by the end user. This means that the QoS decision at the system and lower protocol layers must be related to the degree of satisfaction of the end users, which is estimated at the application layer. One of the key issues to be addressed in the provision of multimedia services over future communication systems is the mapping of QoS from the user’s view down to the heterogeneous network infrastructure and terminal capabilities. The major goal of QoS mapping is the optimization of QoS provisioning and the support of efficient resource allocation, both in the end system and within the heterogeneous systems which include support of multiple services in multi-user multi-platform environments. The parameters to be considered in the QoS mapping should include user preferences, application parameters (e.g. service type, service class, source codec, required display mode, source parameters, etc.), network-related parameters (e.g. transmission delay, delay jitter requirement, bandwidth, etc.), and details of the revenue model (price that the user is willing to pay for the service). The issues regarding QoS parameter- mapping at various levels of communication systems for the provisioning of multimedia services have been addressed in a number of projects and research papers [8 10]. The design criteria of all these QoS-mapping algorithms are based on the direct computation of the correct solution given the constraints in place, using brute force. In order to reduce the complexity of the optimization problem, a linear approximation or simplification of the problem phase has been used to derive a solution in computationally-tractable time. The solutions found with these techniques often show a suboptimal resource allocation and experienced service quality. Due to the highly-fluctuating channel conditions, the varying system capacity, and the different transmission technologies encountered in a heterogeneous system, an adaptive-QoS or soft-QoS control mechanism is necessary to guarantee the heavy resource and quality

420 Visual Media Coding and Transmission requests deriving from multimedia users. Soft-QoS control parameters include a satisfaction index indicating users’ perceptual quality requirement, and an adaptive parameter profile function that captures the robustness of multimedia applications to network conditions. Thus, soft-QoS bridges the gap between the efficient provision of network-level QoS and users’ perceptual quality requirements of multimedia applications. 10.2.2 Optimisation of QoS for Heterogeneous Networks In heterogeneous network scenarios, the quality received by the user tends to be significantly impacted by the characteristics of the access networks encountered in the transmission path. The end-to-end path of a traffic flow typically traverses multiple dissimilar accesses networks offering a wide range of data rates, wireless segments with significantly variable channel conditions, and core networks operated by different providers. Different network connections provide different QoS support. The perceived quality is clearly a function of the QoS experienced in each segment of the path. Therefore, appropriate techniques should be followed in mapping QoS across different communication systems, in order to guarantee an acceptable multimedia quality to end users. Generic inter-networked QoS-optimized infrastructure is designed to enhance the multi- media quality received by end users in heterogeneous system. Therefore, the design is focused on the QoS guarantee across the user-plane protocol stacks of different access networks. This approach is significantly different from the conventional QoS infrastructure design in the literature, which is designed for the control-plane signaling. In order to have efficient QoS for different media, joint-layer architecture that combines the link-level, application-level, and system-level QoS provisioning is developed. Link-level adaptation is designed to optimize the allocation of link-level parameter values according to the media characteristics, and the ultimate goal is to provide each connection with the negotiated quality while using as few radio resources as possible. The application-level adaptation, which takes place while media is delivering, uses a feedback/estimated varying channel condition and dynamically adjusts the allocated link-level parameters. The system-level adaptation is used to optimize the allocation of the available system resources among different media users. The abovemen- tioned adaptive QoS provisioning results in cross-protocol layer optimization for delivered multimedia content. This layered QoS provisioning architecture is expanded across different networks. The QoS provisioning over heterogeneous network architecture has three main layers of optimization, namely “application QoS”, “intra-network (protocol-layer) QoS” and “inter- network QoS”, as shown in Figure 10.1. 10.2.2.1 Application QoS Application QoS addresses the process of matching the complexity of multimedia content to terminal resources. Advanced mapping techniques are needed to match the content bandwidth, media coding, source parameters (frame rate, resolution, quantization, etc.), and rendering complexity to the available terminal resources on the fly, while maximizing the overall perceived quality for a given network condition. Terminal reconfigurability will also be considered in the mapping process. Each application has a number of different objects

Quality Optimization for Cross network Media Communications 421 Application Layer Intra network QoS Network Layer Intra network QoSApplication Layer Transport Layer Data Link Layer Transport Layer Network Layer Physical Layer Network Layer Data Link Layer Data Link Layer Physical Layer Inter-network QoS Physical Layer Figure 10.1 End to end QoS performing a number of tasks while sharing the limited amount of terminal resources. The problem is to be formulated as a nonlinear optimization problem. 10.2.2.2 Intra-network QoS (Vertical QoS) In intra-network QoS, a user’s QoS specification is translated into values for QoS parameters supported by the different protocol layers of individual networks encountered in a heterogen- eous system. Each layer of the protocol stack is characterized by its own parameters and may offer different levels of QoS guarantee. QoS mapping will be designed to map between the parameters associated with given service classes at each of the layers. Between any two layers, it is necessary to find a mapping between the expected performance at the lower layer and its impact on QoS parameters that are meaningful to the higher layer. Thus, the design of the mapping algorithm will concern the parameter performances as seen from the perspective of the network (system), and also the degree of satisfaction of the end user. Moreover, the varying channel quality experienced in wireless links will also be considered to derive time-adaptive parameter mapping. 10.2.2.3 Inter-network QoS (Horizontal QoS) In intra-network QoS, QoS mapping is optimized separately for each intermediate network. Inter-network QoS performs the translation between point-to-point and end-to-end QoS guarantees. As different networks offer distinctly-defined QoS parameters and transmission parameters, reconciling these parameters into a common end-to-end characterization of service quality is complex, and it should be tackled with care to maintain the guaranteed service quality for multimedia transmission over a heterogeneous system. Several issues are to be considered in the inter-network QoS mapping process. Different network technologies may have different system capacities, and the available resources for a given service may vary from one network to another. Also, a subnetwork may offer various service differentiations. Another issue that will be considered is media scalability. In some communication platforms, streams with similar

422 Visual Media Coding and Transmission performance requirements are aggregated into a single stream (transmit over shared channels), while others are treated separately (transmit over dedicated channels). A mapping algorithm will be designed to minimize the effect of network/system variations on the multimedia quality while maximizing the overall system utilization. Moreover, the influences of inter-network handover on the algorithm performance will be taken into account in QoS mapping optimization. The user-centricity plays a major part in the proposed QoS architecture as it is targeted to improve the multimedia quality received by the end user. Thus the user requirements and high QoE will be of primary concern for the development and assessment of the proposed technology. A block diagram of the proposed QoS architecture is depicted in Figure 10.2. The first step involves identification of QoE metrics for a given application. “Indicators of user satisfaction” (such as video/audio quality, picture size, playback speed, etc.) are identified for each application based on a number of user cases and application scenarios, in which knowledge-based information, such as personal information, social information, and user- defined preferences will be taken into consideration. These satisfaction indicators will be measured based on a graphical user interface that guides the user through a set of concept/ service-focused questions [11]. The answers are expressed in human terminology as “excellent”, “very good”, “good”, “fair”, or “poor”, or as scrolling bars, where the user can indicate the level of satisfaction for a set of attributes. QoE for the given service is calculated based on the gathered satisfaction indicator values following a statistical approach such as the service-level approach proposed in [12]. The next step is to map network QoS performance attributes on to the evaluated user-perceptible QoE performance targets. This will be done by deriving the relationship between network QoS and user QoEs based on statistical analysis. The derived QoS attributes will be fed to the heterogeneous communication systems. Application QoS, intra-networked QoS, and inter-networked QoS algorithms will be employed at the heterogeneous system to achieve the optimal QoS for the given application requirements. Moreover, a scheduling algorithm, which takes into account the characteristics of the media and user/channel profile information, will be developed to improve the performance of multimedia in heterogeneous systems. Finally, the proposed QoS enhancement approach is evaluated based on the end user’s assessment. The achieved QoE and targeted QoE will be compared, and derived QoE model parameters will be updated accordingly. Consequently, the proposed QoS architecture will help to optimize the end-to-end system performance, ensuring an excellent QoE for advanced multimedia users. 10.3 Implementation of a QoS-optimized Inter-networked Emulator This section describes the implementation of the proposed QoS architecture for EDGE-UMTS systems. The implementation follows a detailed description of each element of the QoS architecture, which is explained in Section 10.2. The generic QoS architecture described in Section 10.2 is applicable to any heterogeneous network system. However, the EDGE-UMTS system has been used as a case study due to its popularity in 3G wireless communication systems. Figure 10.3 shows the role of the QoS architecture in a practical operating environment, i.e. EDGE-UMTS. The system consists of five components: a transmitting mobile terminal, EDGE emulator, QoS mapping emulator, UMTS emulator, and receiving mobile terminal, as shown in Figure 10.4. The terminal emulator emulates the mobile terminal and is capable of transmitting and receiving multimedia data. An MPEG-4 data transmitter and decoder are integrated into

User group Identification of “Indicators of user User interactivity End user satisfaction” for a given application/service User profile/ behaviour/ application QoS enhancement QoS enhancement scenario User-centric soft-QoS control User-centric soft-QoS control QoS attributes Application QoS Application QoS Intra-network QoS Intra-network QoS Access Network A Access Network B QoS Parameter mapping Between Network A and B Inter-network QoS Middleware Multimedia transcoding Multimedia transmoding Heterogeneous communication system Figure 10.2 The proposed QoS architecture. Reproduced by Permission of Ó2005, 2007 IEEE

424 Visual Media Coding and Transmission Figure 10.3 EDGE UMTS parameter mapping emulator. Reproduced by Permission of Ó2005 IEEE the terminal emulator. The received data is decoded in real time and is displayed on the terminal emulator. The EDGE emulator consists of two parts: the GERAN data flow model and the EGPRS physical link layer model. The physical link layer model simulates the physical layer character- istics of the channel between the EDGE mobile terminal and a base station. The model includes forward error correction, modulation, and transmission over fading channels, equalization, and reception and detection of correctable and uncorrectable errors. The transmitted signal is subjected to a multipath fast-fading environment. The simulator model was built using the COSSAP stream simulation environment and closely follows the standard specification [13]. The EGPRS data flow simulator is implemented to carry out detailed examinations of the effects of channel errors upon applications. The model was implemented in Cþþ . The layers implemented include RTP/UDP/IP transport layers and GPRS SNDC, LLC, and RLC/MAC layer protocols. It must be emphasized that only the data flow properties of the protocols have been implemented in this model. This means that only the resulting effects on header sizes, packet- and stream-segmentation procedures, and flow-control have been implemented. Similar to the EDGE emulator, a UMTS emulator, which consists of the UTRAN data flow model and WCDMA physical layer, is implemented. The WCDMA physical layer model was developed in a generic manner that enables easy configuration of UTRAN link-level parameters such as channel structures, channel coding/decoding, spreading/de-spreading, modulation, transmission modeling, and propagation environments, and their corresponding data rates according to the 3GPP specifications [14]. The multipath-induced inter-symbol interference is implicit in the developed chip-level simulator. By adjusting the model parameters, the bit error and block error characteristics can be determined for a range of signal-to-noise ratios (or carrier-to-interference ratios) and for different physical layer configurations. A UMTS radio interface protocol model, which represents the data flow across the UTRAN protocol layers, is implemented in Visual Cþþ . It is integrated with the physical link layer model to emulate the actual radio interface experienced by users. This

Mob e C ent UMTS Emu ator QoS parameter-mapp ng EDGE Emu ator Mob e C ent (transm tter/rece ver) Conf gurat on Emu ator Conf gurat on Conf gurat on (transm tter/rece ver) Med a Conf gurat on & Conf gurat on & Conf gurat on & Med a Data Mon tor ng Too Mon tor ng Too Mon tor ng Too Data Contro & Contro & S gna ng EDGE Data S gna ng F ow Mode PS: QoS parameter- PS: IP/UDP/ mapp ng Emu ator EDGE Phys ca IP/UDP/ L nk Layer RTP UMTS Data Med a transcoder RTP F ow Mode UMTS EDGE UMTS Phys ca Data Data L nk Layer F ow F ow Mode Mode Figure 10.4 QoS mapping emulation system architecture

426 Visual Media Coding and Transmission allows for interactive testing of the effects of different parameter settings of the UTRAN upon the received multimedia quality. Also, it provides information on channel utilization and system complexity for the given connection. The QoS mapping emulator written in Visual Cþþ 6.0 is a relay program between the EDGE emulator and the UMTS emulator. To some extent, the QoS mapping emulator can be regarded as the combination of the EGPRS emulator and UMTS emulator. It receives the data sent by the EDGE emulator, changes it to UMTS format, and sends it the UMTS emulator, and vice versa. If necessary, a trans-coding function can be added to the parameter-mapping program. In the following, the structure of the QoS mapping emulation system and the function of each component of the emulation system are presented. 10.3.1 Emulation System Physical Link Layer Simulation Error pattern files produced by the EGPRS physical link layer simulator and UMTS physical link layer simulator are used to simulate the real mobile communication environment. The EGPRS and UMTS physical link layer simulators have been implemented using the COSSAD and Signal Processing WorkSystem (SPW) software simulation tools, respectively. The physical link layer parameters used in the simulators are listed in Tables 10.1 and 10.2. A detailed description of EGPRS and UMTS physical link layer simulators is given in Chapter 8. Table 10.1 EGPRS simulator parameters Parameter Settings Channel Coding Scheme GPRS CS 1 $ 4, HSCSD TCH/F9.6, TCH/F14.4, EGPRS PDTCH Supported MCS 1 $ 7 Interleaving Block rectangular over four frames for GPRS; as specified in GSM 05.01 for EGPRS Training Sequence Codes Eight codes available Modulation GMSK, 8 PSK Interference Characteristics User definable static C/I ratio for single co channel interferer. May also be disabled. No frequency offset Fading Characteristics Rayleigh fading for each path (Rice for one component of RA). Fading varies during one burst Multipath Characteristics TU, RA, HT propagation environments supported, as in GSM 05.05 Transmission Capabilities User definable. Can simulate no frequency hopping and ideal fre quency hopping (no correlation between successive bursts) Mobile Terminal Velocity User definable. Static > 250 kmph (for 900 MHz) Carrier Frequency User definable to 900 MHz or 1800 MHz Antenna Characteristics 0 dB gain for both transmitter and receiver. No antenna diversity Signal to Noise Characteristics AWGN source at receiver. User definable Eb/N0 ratio Burst Recovery Synchronization based on the cross correlation properties of the training sequence Equalizer 16 state soft output MLSE equalizer for GMSK; 16 state decision feedback MLSE equalizer for 8 PSK Channel Decoding Soft decision Viterbi convolutional decoder. Fire correction and de tection for CS 2 4 and MCS 1 9 Performance Measures Bit error patterns and block error patterns Simulation Length User definable. Most experiments run for 15 000 blocks/timeslot

Quality Optimization for Cross network Media Communications 427 Table 10.2 UMTS simulator parameters Parameters Settings CRC Attachment 24, 16, 12, 8, or 0 Channel Coding Scheme No coding, 1/2 rate convolutional coding, 1/3 rateconvolutional coding, Supported 1/3 rate turbo coding Interleaving 1st interleaving: block interleaver with inter frame permutation 2nd interleaving: block interleaver with inter columns permutation Rate Matching Permutation patterns are specified in [15] Simplified version of the algorithms (for convolutional rate matching TrCH Multiplexing and turbo rate matching) as specified in [15]. Rate matching ratio (repeat or puncturing ratio) is user definable Transport Format Experiments were conducted for two transport channels. Can be Detection extended to a higher number of transport channels Spreading Factor TFCI based detection Transmission Time Interval 512, 256, 128, 64, 32, 16, 8, 4 Pilot Bit Patterns 10 ms, 20 ms, 40 ms, 80 ms Interference/Noise Characteristics As specified in [14] Fading Characteristics User defined values are converted to the variance of AWGN source at receiver Multipath Characteristics Rayleigh fading mobile channel impulse response is updated 100 times Mobile Terminal Velocity for every coherence time interval Chip Rate Vehicular, pedestrian Carrier Frequency User definable. Constant for the simulation run Antenna Characteristics 3.84 Mcps Receiver Characteristics 2000 MHz 0 dB gain for both transmitter and receiver antenna Transmission Diversity Rake receiver with maximum ratio combining, equal gain combining or Characteristic selective combining. The number of rake fingers is user definable Space time block coding based transmit antenna Channel Decoding Diversity (STTD) as defined in [14] Closed loop fast power control [16] Performance Measures Soft decision Viterbi convolutional decoder Simulation Length Standard LogMap turbo decoder Number of turbo iterations is user definable. Bit error patterns and block error patterns 3000 6000 radio frames, equivalent to 30 60 s duration Using these physical link layer simulators, a set of error pattern files is produced for different radio bearer configurations. These error pattern files are used to emulate the effect of physical condition on transmitted data. The EGPRS emulator, which is described in Chapter 8, is used to simulate data transmission in real EGPRS networks. However, a few new features are added to the implementation presented in Chapter 8. These are: 1. New channel schemes (MCS-1 9) simulation in RLC layer. 2. Retransmission.

428 Visual Media Coding and Transmission 3. Connection with QoS mapping emulator or another EGPRS emulator. 4. Header corruption setting in each layer. 5. Timeslots setting. The UMTS emulator, which was designed to simulate data transmission in real UMTS networks, was also improved to support RLC AM (retransmission). 10.3.2 Emulation System Transmitter/Receiver Unit MPEG-4 File Transmitter (Figure 10.5) is used to transmit the encoded MPEG-4 video file (Ã.cmp) to other emulators or to MPEG-4 Decoder. MPEG-4 Decoder (Figure 10.6) is designed to decode the received MPEG-4 video packets and display them on the screen. Writing received packets to file (Ã.yuv) is supported. MPEG-4 Terminal combines the functions of MPEG-4 File Transmitter and MPEG-4 Decoder (Figure 10.7). This new program has a mobile handset shape. Users can press the keys to decode received MPEG-4 data or transmit MPEG-4 data to other emulators. Pressing the power key closes the program. Pressing the dial/answer key begins receipt of MPEG-4 data, while pressing the hang-up key stops receipt of MPEG-4 data. The message key, which is above the dial/answer key, is in control of sending MPEG-4 data to other emulators. Pressing the right mouse button will open a dialog to set the transmission and receiving parameters. The program can be run on Windows 2000 or XP. 10.3.3 QoS Mapping Architecture The QoS mapping emulator, written in Visual Cþþ 6.0, is a relay program between the EGRPS emulator and UMTS emulator. To some extent, the QoS mapping emulator can be regarded as the combination of the EGPRS emulator and UMTS emulator. It receives the encapsulated data sent by the EGPRS emulator, decapsulates it to MPEG-4 data, then encapsulates the MPEG-4 data in UMTS format and sends it to the UMTS emulator, and vice versa. Figure 10.5 MPEG 4 File Transmitter interface

Quality Optimization for Cross network Media Communications 429 Figure 10.6 MPEG 4 Decoder interface If necessary, transcoding function can be added to the program. To simulate the EGPRS and UMTS protocols precisely is impossible and unnecessary. Some simplifications and modifica- tions of the protocol are considered in the implementation. In Figure 10.8, the working process of the QoS mapping emulator is shown. Figure 10.7 MPEG 4 Terminal

430 Visual Media Coding and Transmission EGPRS QoS Mapping Emulator UMTSEmulator Application layer Application layer Application layer (MPEG-4 Codec) (MPEG-4 Codec) EGPRS Transport layer EGPRS Transport layer UMTS Transport layer UMTS Transport layer EGPRS SNDC layer EGPRS SNDC layer UMTS PDCP layer UMTS PDCP layer EGPRS LLC layer EGPRS LLC layer EGPRS RLC/MAC layer EGPRS RLC/MAC layer WCDMA RLC/MAC layer WCDMA RLC/MAC layer EGPRS Physical link layer simulator UMTS Physical link layer simulator (Using error pattern files to replace it (Using error pattern files to replace when emulator works) it when emulator works) Streams from EGPRS to Streams from UMTS to UMTS EGPRS Figure 10.8 Flow diagram of QoS mapping emulator