Joint Source-Channel Coding by Means of an Oversampled Filter Bank Code
- PDF / 753,193 Bytes
- 12 Pages / 600.03 x 792 pts Page_size
- 109 Downloads / 199 Views
Joint Source-Channel Coding by Means of an Oversampled Filter Bank Code Slavica Marinkovic and Christine Guillemot IRISA-INRIA, Campus de Beaulieu, 35042 Rennes Cedex, France Received 1 September 2004; Revised 5 April 2005; Accepted 7 April 2005 Quantized frame expansions based on block transforms and oversampled filter banks (OFBs) have been considered recently as joint source-channel codes (JSCCs) for erasure and error-resilient signal transmission over noisy channels. In this paper, we consider a coding chain involving an OFB-based signal decomposition followed by scalar quantization and a variable-length code (VLC) or a fixed-length code (FLC). This paper first examines the problem of channel error localization and correction in quantized OFB signal expansions. The error localization problem is treated as an M-ary hypothesis testing problem. The likelihood values are derived from the joint pdf of the syndrome vectors under various hypotheses of impulse noise positions, and in a number of consecutive windows of the received samples. The error amplitudes are then estimated by solving the syndrome equations in the least-square sense. The message signal is reconstructed from the corrected received signal by a pseudoinverse receiver. We then improve the error localization procedure by introducing a per-symbol reliability information in the hypothesis testing procedure of the OFB syndrome decoder. The per-symbol reliability information is produced by the soft-input soft-output (SISO) VLC/FLC decoders. This leads to the design of an iterative algorithm for joint decoding of an FLC and an OFB code. The performance of the algorithms developed is evaluated in a wavelet-based image coding system. Copyright © 2006 Hindawi Publishing Corporation. All rights reserved.
1.
INTRODUCTION
Various joint source-channel coding approaches, guided by an optimum tradeoff between compression efficiency and error and/or erasure resilience depending on the link characteristics, have been considered in order to improve multimedia signal transmission over noisy channels. Here, we focus on JSCC techniques based on quantized redundant signal expansions by means of OFB. As the signal representation in this approach is redundant, an OFB-encoded stream has error-resilient features. Error-control coding and signal decomposition are thus integrated in a single block. The errorcorrecting code thus allows also to suppress some quantization noise effects. So far, the research in this area has been concentrated mainly on the study of oversampled transform codes (OTC) which are OFB codes with polyphase filter orders equal to zero. The OTC can be viewed as real-number block codes, while the OFB codes can be associated to real-number convolutional codes. Decoding of real-number block codes has been considered by many authors [1–7]. Oversampled block transforms like DFT codes have been shown to be BCH codes over the real field [8, 9]. DFT or DCT codes have also been considered as joint source-channel block codes to obtain
robustness to erasures [4–6, 10] an
Data Loading...
