An Energy-Based Method for Signal Compression and Reconstruction with Wavelets.
NAVAL UNDERSEA WARFARE CENTER NEWPORT DIV RI
Pagination or Media Count:
Wavelets and wavelet transforms have recently emerged as a promising alternative to traditional Fourier based spectral decompositions in a variety of signal processing applications. With the expected exponential increase in data traffic volume and the consequent overloading of storage capacity and transmission channels, the need for improved signal compression is essential. A new energy based method for selection of wavelet coefficients for signal compression is proposed. The number of coefficients selected from a particular level of the wavelet decomposition tree is proportional to the mean energy contained in the coefficients at that level. In experimental tests, this method provided significantly improved performance over conventional global thresholding based wavelet selection techniques. The performance index used is the signal to error ratio, which is a measure of the quality of the reconstructed signal from its compressed representation compared to the original. For highly nonstationary signals, a crossover effect is observed that is, the energy based selection is out performed by the conventional method at higher percent retention levels. In such a situation, a segmentation strategy is proposed wherein the signal is decomposed into segments of similar characteristics prior to compression.