Multiple description source coding: A geometric treatment and an application to correlated fading channels

A geometric treatment of the multiple description source encoding of a Gaussian random variable when two channels are available is presented. An achievable rate region is derived through a geometric approach which explicitly describes the steps followed by the encoding process. The trade-off between the performance when both links are working and that when a channel is down is given a geometric illustration.; Next, the geometric approach to quantization is extended to multiple description trellis coded quantization and to the geometry of the Voronoi regions of a trellis.; Finally, an application of multiple description source coding is presented. A multiple description transform speech coder is proposed and its performance is obtained for correlated Rayleigh fading channels. A coding algorithm utilizing an LPC analysis to achieve entropy-constrained quantization is described and it is shown that when applied to real speech data samples along with unequal error protection channel coding techniques, it offers significant improvements in performance over commonly proposed approaches for fading channels.