A thesis submitted for the degree of Doctor of Philosophy at the Department of Electronic Systems Engineering, University of Essex
Mohammad Mahdi Ghandi, January 2006
Supervisors: Professor Mohammad Ghanbari and Dr Ed Jones
Examiners: Dr Abdul-Hamid Sadka and Dr Martin Fleury

Abstract

In this thesis prioritisation and layering techniques for reliable video transmission are investigated. Two layering methods are considered: 1) the data partitioning technique of the H.264 standard and 2) a new proposed H.264 SNR scalability method. For the appropriate protection of the layered coded video, methods such as prioritised forward error correction coding (FEC with turbo coding in this thesis) or hierarchical quadrature amplitude modulation (HQAM) are employed. With data-partitioning, the problem is that the ratio between the rates of the high priority (HP) and low priority (LP) layers is highly variable with the contents of the pictures. However, with a conventional HQAM configuration, the HP and LP capacities require a constant ratio. Also, with fmmixed HP and LP coding ratios in FEC, a highly variable channel rate will be required. This thesis presents switching methods to provide flexible FEC and HQAM to tackle this problem. Furthermore, it is shown that a combination of HQAM and FEC will yield an even better performance. The variable HP/LP ratio problem does not exist in the proposed, drift-free H.264 SNR scalable codec, because the rates of the individual priority layers can be effectively controlled. This is done in a proposed Lagrangian rate controller where quantisation parameters of the individual macroblocks are adjusted and the frames' Lagrangian multiplier is adaptively tuned. To provide an efficient scalability scheme, the enhancement layer encoder has the flexibility to use three different prediction modes. In addition, the coding methods and the bitstream structure have new modifications. We also propose an efficient error-concealment method for the scalable codec to efficiently restore the lost information at the decoder. With the prioritised transmission, we show that the proposed scalable codec can perform better than data partitioning. Finally, in applications where there is knowledge of the transmission channel, a superior solution is presented which adapts the modulation mode, channel coding ratio and the source rate according to the channel conditions. We show that even in feedback-based adaptive transmission systems, the source encoder must be resilient to probable transmission errors, e.g. by means of layering.