Overcomplete Source Expansion Aided, Soft-bit Assisted Speech Transceiver

Abstract

This paper presents a scheme that invokes jointly optimized source and channel decoding, which exploits the redundancy imposed by the overcomplete expansion scheme. More explicitly, a structured redundancy based on discrete Fourier transform (DFT) is inserted to the speech signal prior to source encoding. At the receiver the soft-input soft-output (SISO) decoder exploits the source index correlation for error protection. Furthermore, the explicit redundancy introduced in the speech signal is exploited to reconstruct the received voice signal with the aid of pseudo-inverse of the overcomplete block transform. The performance of the proposed scheme is evaluated when communicating over Additive White Gaussian Noise (AWGN) channel. The specific scheme using an overall code rate of 1/2 benefitted from the exploitation of the residual redundancy with about 0.4 dB E-b/N-0 gain to its corresponding hard speech decoding based counterpart.