Tree-structured oversampled filterbanks as joint source-channel codes: application to image transmission over erasure channels

• Published in: IEEE transactions on signal processing Vol. 52; no. 9; pp. 2584 - 2599
• Main Authors: Motwani, R., Guillemot, C.
• Format: Journal Article
• Language: English
• Published: New York, NY IEEE 01.09.2004; Institute of Electrical and Electronics Engineers; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithm design and analysis; Algorithms; Applied sciences; Channels; Design engineering; Detection, estimation, filtering, equalization, prediction; Exact sciences and technology; Image coding; Image communication; Image processing; Image reconstruction; Information, signal and communications theory; Multimedia; Noise; Quantization; Reconstruction algorithms; Recovery; Robustness; Signal and communications theory; Signal processing; Signal, noise; Streaming media; Studies; Systems, networks and services of telecommunications; Telecommunications; Telecommunications and information theory; Time domain analysis; Transmission and modulation (techniques and equipments); Transport protocols; multiple description coding; Tree structured method; Joint source-channel coding; frames; Erasure resilient coding; oversampled filterbanks; Oversampling; Image reconstruction; Quantization noise; Coding; Filter bank; Performance analysis; Image transmission
• ISSN: 1053-587X; 1941-0476
• DOI: 10.1109/TSP.2004.832027

This paper studies oversampled filterbanks for robust transmission of multimedia signals over erasure channels. Oversampled filterbanks implement frame expansions of signals in l/sup 2/(Z). The dependencies between the expansion coefficients introduced by the oversampled filterbank are first characterized both in the z-domain and in the time-domain. Conditions for recovery of some typical erasure patterns like bursty erasure patterns are derived. The analysis leads to the design of two erasure recovery algorithms that are first studied without quantization noise. The reconstruction algorithm derived from the time-domain analysis exploits the fact that an oversampled filterbank represents signals with more than one set of basis functions. The erased samples are first reconstructed from the received ones, and then, signal space projection is applied. The effect of quantization noise on the reconstructed signal is studied for both algorithms. Using image signals, the theoretical results are validated for a number of erasure patterns, considering unequal error protection enabled tree-structured decompositions.