Adaptive Pre- and Post-Filtering for Block-Based Communication Systems

点击翻译按钮获取中文摘要

This research involves the theory, design, and efficient implementation of locally adaptive time-domain pre-/post-processing operators for efficient signal representation and robust communications. The time-domain interpretation leads to a powerful signal decomposition and signal reconstruction approach with an unprecedented level of adapting capability. This framework retains all flexible features of block-based approaches and adds on top a high level of adaptivity: data samples are processed with small pre-/post-processing block operators and each can be adapted on-the-fly if necessary. Furthermore, small local operators lend themselves nicely to parallel computing, and fast, VLSI-friendly, possibly multiplierless, implementations. Finally, the fact that time-domain pre-filtering is the closest link to the sensor and post-filtering to the display/renderer allows the highest level of integration flexibility as well as the most economical implementation with minimum software or hardware upgrade. This research is particularly geared toward low-complexity signal coding and communication algorithms or systems in resource-constrained, speed-critical, and real-time applications for wireless hand-held devices. In particular, the investigators study local adaptivity via adaptive Wiener filtering technique, providing the critical bridge between ad-hoc, but effective, pre-/post-filtering approaches and fundamental information-theoretic signal decomposition strategies. Target applications of such time-domain local operators under investigation include: (i) under-sampled pre-filtering and over-sampled post-filtering for low bit-rate coding and fast local compressed sensing;(ii) transform-based multiple description coding; (iii) error-resilient pre-/post-filters in error concealment for packet-switched erasure channels; (iv) optimal transform for distributed source coding in a dense sensor networks; and (v) design and application of adaptive two-dimensional (2-D) non-separable local decomposition for visual data representation and processing.