Bekir Ahmet Dogrusoz

Soft Source Coding with Unequal Error Protecting Turbo Codes

Date: August 19, 2003

Abstract

It is possible to design source and channel coding systems separately without any loss in the overall system performance provided that there are no constraints on system complexity and delay. However, practical applications have finite complexity/delay constraints and likely to be operating over a time-varying channel, especially in wireless communication systems. Under these circumstances, the efficiency of the communication system relies on robustness and flexibility of both channel and source coding components. A robust and flexible system can be achieved with a combined source-channel coding method that is designed to allow for a dynamic trade-off between source encoding resolution and error protection capability of the overall system.

In this thesis, we have proposed and studied various systems based on the idea of combination of soft source decoding channel matched vector quantization (CMVQSD) and parallel concatenated recursive convolutional codes (turbo codes) with dynamic unequal error protection (UEP) capabilities. CMVQSD utilizes the reliability information (soft information) produced by the UEP turbo decoder. The proposed systems are studied over both additive white Gaussian noise (AWGN) and Rayleigh flat fading channels, for memoryless Gaussian and Gauss-Markov sources. We have shown that the proposed scheme (CMVQSD-UEP) is capable of adapting to changes in the channel parameters.

We have also extended the results to systems based on trellis coded quantization in addition to systems using plain vector quantization. This is achieved by applying soft decision decoding and channel matching to trellis coded quantization, and combining it with an UEP turbo code (CMTCQSD-UEP). This scheme is shown to have similar advantages over conventional TCQ as CMVQSD-UEP has over conventional VQ.

Adaptability, reliability, and robustness of the proposed systems are studied and it is concluded that employing the soft information provided by the decoder in an unequal error protecting turbo code in source coding can noticeably improve the performance of the overall system.

Committee:
Prof. David Brady
Prof. Dana Brooks
Prof. Masoud Salehi (Advisor)