Joint Optimization of Iterative Source and Channel Decoding Using Over-Complete Source-Mapping

Abstract

The intentionally imposed or inherent unintentional residual redundancy found in source encoded bitstreams can be exploited for supporting joint SoftBit-Source Decoding and Channel Decoding, which has the potential of improving both the error correcting capability as well as the subjective audio or video quality of communication systems. As a potent error concealment technique, the softbit-based source decoding, proposed by Adrat, Vary and Spittka exploits the residual redundancy or correlation inherent in the source codec parameters for mitigating the effects of transmission errors. However, when using efficient source encoders, limited source redundancy is left in the source-encoded bitstream. In this scenario the SoftBit-Source Decoder (SBSD) may have a limited extrinsic information contribution, which results in negligible system performance improvements beyond two decoding iterations. In our novel approach, we partition the total available bit-rate budget between the source and channel codecs in order to improve the attainable error correcting capability and hence to maximize the overall system's performance. More explicitly, the inherent redundancy in the encoded bitstream is intentionally increased with the aid of over-complete mapping. By analysing the extrinsic information transfer function characteristics of the softbit-aided source decoder, we present a set of overcomplete mappings, which maximize the iterative source-channel decoding performance. As a main contribution of this paper, we propose an algorithm for designing a suitable overcomplete mapping for iterative source-channel decoding. Our experimental results show that at the same overall code-rate of the system, the iterative source-channel coding scheme using a suitable overcomplete mapping may lead to an approximately $2dB$ signal-to-noise ratio gain in comparison to the conventional iterative source-channel coding benchmarker.

More information

Identifiers

ORCID for L-L Yang: orcid.org/0000-0002-2032-9327

ORCID for L Hanzo: orcid.org/0000-0002-2636-5214

Catalogue record

Export record