On the suitability of ML decoding in analog multiple description schemes

We consider analog multiple description joint source-channel schemes for the transmission of a Gaussian source over two independent AWGN parallel channels under a quadratic error distortion criteria. For such systems, Lines mapping was proved to be a simple and efficient mapping when MMSE decoding is used. However, MMSE decoding presents a high computational load. In order to reduce complexity of such schemes, this paper studies the suitability of replacing the MMSE decoders by the much simpler ML decoders. The scenario where the ML decoders have some additional side information (Wyner-Ziv scenario) is also analyzed. Their performances are compared with the corresponding theoretical bound and with the obtained simulation results of their MMSE counterparts.