Algorithms and architectures for wireless image communication
Benjamin J. Belzer, John D. Villasenor · 1996
The goal of this research is to develop source and channel coding algorithms and architectures to enable two-way image communication over a network of portable, wireless multi-media workstations. Significant advances resulting from this research include: (1) A study of all biorthogonal wavelet filters with less than 36 combined analysis/synthesis taps to identify those filters with good deleted-band step and impulse responses, properties that predict filter performance for image compression. The search results include several shifted integer coefficient filters with good compression performance that are suitable for low power implementation, including the 2/6 tap filter bank used by RICOH Corporation in their CREW (compression with reversible embedded wavelets) combined lossless/lossy image codec. (2) Exploration of the space of orthogonal, complex-coefficient wavelet filters to find real, near perfect reconstruction, near orthogonal filters with linear phase and with good image compression performance. (3) A multiplierless, wavelet-based intra-frame video codec architecture. The codec provides video between 60 and 600 kilobits/sec with external video rate control and at channel bit error rates (BERs) of up to 0.02. This architecture has been successfully implemented in a single chip video codec by another UCLA researcher. (4) Development of new trellis-based joint source/channel coding (JSCC) algorithms that enable sharing of the trellis processor for source and channel coding. The thesis introduces symmetric trellis coded vector quantization (S-TCVQ) for low complexity source coding and shows that, for the important Laplacian source, S-TCVQ achieves the same quantization SNR performance at lower complexity when compared to trellis-based scalar vector quantization, the previously best known fixed wordlength quantizer. Also demonstrated is joint S-TCVQ/convolutional coding with optimal assignment of channel symbols so as to minimize overall distortion due to both quantization and channel errors; the resulting JSCC system can send good quality 7.5 frames/second 512 x 512 resolution intraframe video at channel BERs of 0.1 and at channel bandwidths of 600 kilobits/sec.