NOVEL ITERATIVE MULTIPLE DESCRIPTION CODING FOR CORRELATED SOURCES

In this paper, we study a novel multiple description coding approach which uses a convolutional code to generate the individual descriptions. Contrary to most conventional multiple description schemes, which attempt to partially reconstruct the signal in the presence of a packet loss, we are interested in the question, whether the signal can be completely reconstructed at the receiver if one description is missing by exploiting the residual source correlation. Usually, audio-visual source encoders for digital mobile communications extract parameters that ­ due to delay and complexity constraints ­ exhibit some residual redundancy. This residual redundancy is exploited in a multiple description receiver by performing iterative source-channel decoding (ISCD). The source correlation required for near perfect reconstruction in case of a loss of one description is analyzed by means of EXIT charts and simulation results show the superior performance of the new approach. 1. INTRODUCTION Multiple Description Coding (MDC) [1, 2] is a tool to generate two (or more) descriptions of a signal which are then independently transmitted over a network with possible packet losses. If all descriptions are correctly received, the signal can be reconstructed with the best possible quality. If one or more descriptions of the signal are missing due to packet losses, the signal can still be reconstructed, however, with degraded overall quality. Multiple description coding can also be used for a more general kind of hierarchical coding: due to bottlenecks in the network, parts of the packets may be rejected, thus allowing a flexible rate adaptation.