Jump to navigation
Cross-layer schemes for enhancing H.264/AVC video quality over wireless channels
Rapid growth of video applications over wireless networks is overwhelming the wireless bandwidth. Since video applications demand large bandwidth and realtime transmission, supporting the rapidly increasing video traffic over the bandwidth-limited, error-prone, and time-varying wireless channels is very challenging. As a result, the video applications are likely to suffer packet losses over wireless networks which results in quality degradation. In this dissertation, we design a distortion prediction model for H.264/AVC compressed video streams, and use it for designing novel cross-layer protocols for enhancing the video quality by making more efficient use of the available wireless resources. The cumulative mean squared error (CMSE) is a widely used measure of video distortion. However, CMSE measurement is a time-consuming and computationally-intensive process which is not suitable for many video applications. A low-complexity and low-delay generalized linear model is proposed for predicting CMSE contributed by the loss of H.264 AVC encoded video slices. The model is trained over a video database by using a combination of video factors that are extracted during the encoding of the current frame, without using any data from future frames in the group of pictures (GOP). The slices are then prioritized within a GOP based on their predicted CMSE values. The accuracy of the CMSE prediction model is analyzed using cross-validation, analysis of variance, and correlation coefficients. The simulations are carried out to evaluate the performance of the CMSE prediction model for varying encoder configurations and bit rates of test videos. The CMSE slice prediction model is used to design an unequal error protection (UEP) scheme, using the rate-compatible punctured convolutional (RCPC) codes over wireless channels. This scheme provides protection to the video slices against the channel errors, based on their priority, in order to minimize the video distortion. An application of our slice prioritization is demonstrated by implementing a priority-aware slice discard scheme, where the low-priority slices are dropped from the router when the network experiences congestion. Additionally, the GOP-level slice prioritization is extended to the frame-level slice prioritization, and its performance is evaluated over the additive white Gaussian noise (AWGN) channels. The idea of using slice CMSE prediction is extended to adapt the video packet size to the wireless channel conditions, in order to minimize the video distortion. A real-time, priority-aware joint packet fragmentation and error protection scheme for real-time video transmission over Rayleigh fading channels is presented. The fragment error rates (FERs) are simulated for a combination of different fragment sizes and RCPC code rates. These FERs are then used to determine the optimal fragment sizes and code rates for packets of each priority class by minimizing the expected normalized predicted CMSE per GOP in H.264 video bit stream. An improvement in the received video quality over the conventional and priority-agnostic packet fragmentation schemes is observed. Next, a cross-layer, priority-aware scheduling scheme for real-time transmission of multiple video applications over a time-varying channel is developed. Each video application considered has different characteristics such as user priority, latency, distortion, size, and encoding bit rate. A cost function is optimized to determine the scheduling order for video frames. The performance of our scheme is compared with that of the CMSE based scheme, where the frames are rank-ordered for transmission using its CMSE per bit values, and with the earliest deadline first (EDF) scheme in which each user takes turns to transmit a frame. A collaborative effort with other researchers and developed two additional cross-layer error protection schemes. In the first scheme, a cross layer UEP scheme that jointly assigned FEC at both the Application layer (using Luby Transform) and the Physical layer (using RCPC codes) for prioritized video transmission is developed. The video distortion function is minimized by using the genetic algorithm (GA). The performance of our scheme is evaluated for different channel SNR values. In the second UEP scheme, a framework that combined the RCPC codes and concatenated it with hierarchical quadrature amplitude modulation (QAM) is investigated. Employing RCPC codes and hierarchical modulation jointly resulted in greater flexibility as some parts of the data can be protected only by the hierarchical modulation while others may be protected by a low FEC code rate. The performance of the proposed scheme is compared to the standard 8-QAM with symmetric constellation.
Computational Science Research Center
Doctor of Philosophy (Ph.D.) Claremont Graduate University and San Diego State University, 2016
© 2015 SDSU Library & Information Access. All Rights Reserved.