Joint and tandem source -channel coding with complexity and delay constraints.

Joint source-channel coding has been developed with the motivation that it can achieve better performance with less complexity and delay than tandem source-channel coding. However, little quantitative evidence for this claim has appeared in the literature. In this dissertation, we search for such evidence by quantitatively comparing representative systems of each type on the basis of distortion vs. complexity and distortion vs. delay when transmitting analog data samples across a binary symmetric channel. Channel-optimized transform coding is chosen as the system representative of joint source-channel coding, and transform coding with Reed-Solomon coding is chosen as the system representative of tandem source-channel coding. For each strategy, formulas for the mean-squared error, complexity and delay are found and used to minimize distortion subject to a complexity or a delay constraint, for data modeled as Gauss-Markov. The results of such optimizations suggest there are thresholds on complexity and delay such that when the available complexity or the permissible delay is above the threshold, tandem coding is better, and when below the threshold, channel-optimized transform coding is better. The investigation is expanded to unequal error protection in tandem source-channel coding. Providing different levels of error protection according to the importance of the transform coefficients, the performance of tandem source-channel coding is improved and the complexity threshold is reduced relative to that of an equal error protection system. Also, a fundamental problem in designing transform codes, rate allocation, is considered and an analytic expression for the optimal rate allocation is derived for noisy channels, based on high-resolution theory. The resulting asymptotically optimal rate allocation for noisy channels suggests that even more bits need to be allocated to the coefficients whose variances are greater than the geometric mean of such variances, compared to the optimal bit allocation for noiseless channels. On the other hand, even fewer bits are to be allocated to the coefficients whose variances are smaller than the geometric mean. The correspondence between experimental results and the asymptotic expression is observed to be good.PhDApplied SciencesElectrical engineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/125922/2/3016905.pd