Prediction-compensated multiple description coding with optimal weighted reconstruction

In this paper, a weighted reconstruction method for the prediction-compensated multiple description coding (PC-MDC) with two descriptions is proposed. Different from the original PC-MDC, where the redundant data are simply discarded when both descriptions are received, we use a weighted average of all available data to recover the signal. The closed form solution of the optimal weighting factors and the corresponding bit allocations are derived for M-band perfect reconstruction filter bank based PC-MDC. The asymptotic performance of the method is also studied. Theoretical analysis shows that the proposed method further reduces the expected distortion and the distortion product of the non-weighted PCMDC method.