Joint source-channel decoding of predictively and nonpredictively encoded sources: A two-stage estimation approach

David Jonathan Miller, Elias S G Carotti, Yu Wei Wang, Juan Carlos De Martin

    Research output: Contribution to journalArticle

    1 Scopus citations

    Abstract

    A common joint source-channel (JSC) decoder structure for predictively encoded sources involves first forming a JSC decoding estimate of the prediction residual and then feeding this estimate to a standard predictive decoding (synthesis) filter. In this paper, we demonstrate that in a JSC decoding context, use of this standard filter is suboptimal. In place of the standard filter, we choose the synthesis filter coefficients to give a least-squares (LS) estimate of the original source, based on given training data. For first-order differential pulse-code modulation, this yields as much as 0.65-dB gain in reconstructing first-order Gauss-Markov sources. More gains are achieved with modest additional complexity by increasing the filter order. While performance can also be enhanced by increasing the source's Markov model order and/or the decoder's lookup table memory, complexity grows exponentially in these parameters. For both predictive and non-predictive coding, our LS approach offers a strategy for increasing the estimation accuracy of JSC decoders while retaining manageable complexity.

    Original languageEnglish (US)
    Pages (from-to)1575-1584
    Number of pages10
    JournalIEEE Transactions on Communications
    Volume52
    Issue number9
    DOIs
    StatePublished - Sep 1 2004

    All Science Journal Classification (ASJC) codes

    • Electrical and Electronic Engineering
    • Computer Networks and Communications

    Fingerprint Dive into the research topics of 'Joint source-channel decoding of predictively and nonpredictively encoded sources: A two-stage estimation approach'. Together they form a unique fingerprint.

  • Cite this