Video Transmission for Third Generation Wireless Communication Systems

Journal of Research of the National Institute of Standards and Technology, March-April, 2001 by H. Gharavi, S. M. Alamouti

We should point out, however, that partitioning could result in a significant increase in the overall bitrate. This is caused by not only the additional partitioning overhead required for synchronization, but also by the coding inefficiencies that can arise from separating the DCT coefficients prior to variable length (VL) coding. For this reason, we argue that it is advantageous if partitioning can be accomplished in terms of the VL-coded DCT coefficients. In addition, we are concerned with developing a video stream partitioning scheme based on constant bitrate (CBR) transmission. We note, further- more, that there are no fundamental requirements for the partitions to be of equal size. In fact, the partitions are often of differing sizes depending on the error protection strategy with respect to the channel conditions.

3.1 VLC-based Partitioning

As mentioned earlier, the separation of the quantised DCT coefficients into different protection classes can be arranged either before or after VL-coding. In the former case, i.e., when partitioning before VL-coding, which we refer to as fixed-zone partitioning, a given number of lower frequency DCT coefficients generated by the standard zigzag scanning process [16] is earmarked for transmission over a higher-integrity subchannel, which can be constituted, for example, by the higher-integrity transmission channel.

In our experiments we observed, however, that this arrangement could significantly increase bits per block, mainly due to the breakup of the run-level symbol near the DCT cutoff region [14]. Alternatively, if the block is partitioned after VLC coding, by selecting a fixed number of VLCs (i.e., VLC-based partitioning) no additional bits would be needed. The main concern with this approach, however, is that the number of DCT coefficients within the upper-zone may change from one block to the next. This is due the nature of run-level coding, where each VLC can represent a different number of DCT coefficients. Consequently, this affects the progression of noise at the receiver as the lower priority partition (second partition) is often expected to be lost. For a better clarification of the above argument, let us assume that the upper-zone, as shown in Fig. 7a, corresponds to the number of DCT coefficients m that is selected for the current coding block. In addition, suppose that, when the motion compensated block (reference block) was partitioned during the transmission of the previous frame, the number of its DCT coefficients for the first partition resulted in an "n" number of coefficients where n [less than]m. The upper-zone difference between the two blocks is shown in Fig. 7b, where the darker shaded area represents the selected upper-zone region of the reference block. Recognizing that the coefficients within the zone difference belong to the second partition of the reference block, now consider the situation when the second partition had been corrupted by errors during the transmission of the previous frame. Under such a condition, the upper zone DCT representing the first partition of the current interframe block cannot be properly reconstructed due to the loss of DCT information within the zone difference. This would, consequently, cause a drift between the local and remote decoder, regardless of how well the first partition is protected. The visual impact of such distortion depends on the number as well as the magnitudes of the non-zero coefficients that fall within the zone difference, m -- n [14]. However, the progression of such distortion and its visual effects does not appear to be of grave concern-as far as the intraframe reset can be accomplished in a reasonable period (see the results section).