Joint source and channel coding using a non-linear receiver

The joint optimization of the source and channel coders in a system consisting of a vector quantizer (VQ) whose output indices are mapped directly into points in the modulation signal space is considered. A decoder based on a nonlinear estimator is used to reconstruct the source signal. An iterative algorithm is introduced which jointly optimizes the VQ and the modulation signal set using as optimality criterion the minimum mean-square error (MSE) between the original and the reconstructed signals. It is shown that a jointly optimized system based on average channel characteristics significantly outperforms (by up to 5 dB) a reference system based on a VQ designed for the given source and a standard quadrature amplitude modulation (QAM) modulation signal set.<>