SBIR-STTR Award

Prototype, wavelet-based lossy compression algorithms will be developed, for compression of both transient and narrowband sonar signals. An appropriate wavelet-related signal energy into the transform will be designed, with the goal of optimally concentrating 1 fewest number of coefficients. Standard 2-band wavelets will be examined, as well as M-band wavelets (M>2), wavelet-packets, and general perfect reconstruction multirate filterbanks. The family of Aware-developed translation-invariant wavelet transforms will also be examined as a possibility. A bit allocation strategy will be developed, tuned to the particular characteristics of the sonar data to be compressed. Standard entropy coding will be implemented to reduce the remaining redundancy. Zero-run length, and Huffman coding will be implemented. A compression/decompression performance analysis will be performed, to determine the bandwidth savings provided by the compression algorithm. The effects of lossy compression on data quality will also be investigated. Distortions introduced by compression will be evaluated both quantitatively and subjectively. Computational algorithm complexity will be explored, and potential processing bottlenecks will be identified. A final report will be written, documenting the compression algorithm and detailing numerical results of the performance analysis.

Keywords:
Compression Wavelet Wavepacket Sonar

M-band wavelet-based transforms provide considerable flexibility for filter design, and frequency band decomposition. This flexibility is exploited for the design of a transform-based, one-dimensional data compression package. Signal-adaptive transform selection strategies will be developed. These strategies provide for tuning the transform to the input signal characteristics, for the purpose of concentrating the most amount of signal energy into the fewest number of transform coefficients. Time-adaptive quantization and adaptive Huffman modelling will be built in. This gives the capability to more accurately match the quantization and Huffman tables to the time-varying signal characteristics. Comutationally efficient transform structures will be implemented. Production quality C code will be developed in stand-alone form, and as a software toolkit. The software toolkit willenable seamless incorporation of compression into user-defined applications. A performance analysis will be undertaken to determine compression performance over a variety of signal classes. Compression effects on signal detection algorithms will be investigated. A simulation demonstration will be performed. Hardware requirements will be mapped out for transition to real-time implementation.

Keywords:
Compression Wavelets