Abstract

This thesis examines the transmission of H.264 compressed video over an Additive White Gaussian Noise (AWGN) channel when Context-based Adaptive Binary Arithmetic Coding (CABAC) is used as the entropy code. First, several simulations were performed to examine the effect that bit errors have on H.264 decompression. The simulation results show that over 99% of the bit errors tested caused a semantic error to be detected. Further simulation results show that undetected bit errors (bit errors that don't cause semantic errors) were not observed to cause noticeable video quality degradation. Then, two transmission schemes are proposed. The first proposed scheme uses soft bit information (at the receiver) along with H.264 semantic verification to detect and conceal errors during decompression. Note that this scheme does not involve channel coding. The second proposed scheme is an iterative joint source-channel decoding scheme. It combines a rate-歔 convolutional code with the semantic verification strategy of the first scheme. Information is passed from the channel decoder to the source decoder by soft channel decoding. The source decoder then feeds information back to the channel decoder by modifying its original soft values. Simulation results show that both schemes offer significant improvements in terms of subjective quality and in terms of objective quality using PSNR and BER as measures. Keywords: Video transmission, H.264, CABAC, compression, slice candidates, joint source-channel decoding, semantic verification, soft values, semantic error detection