Design of wavelet basis for ECG compression

The wavelet transform has proved to be a valuable tool for compression and analysis of ECG signals. While different techniques have been proposed for the compression itself if is not clear how to choose the particular basis to be used in the transform. In this study we look for wavelets that optimize the compression performance. In order to obtain certain independence from the entropy coding schemes and universal lossless compression strategies used, the performance index is the number of non-zero coefficients in the transform domain. This point is further assured by performing a technique similar to embedded zerotree coding. The basis to be compared are designed in the following ways: The first type of basis is obtained by optimizing a L₂ cost function, while the second type of basis is designed based on the regularity of the filters. We showed that the design of wavelets based on the local characteristics of the signal could increase the compression performance. Wavelets with maximum number of zero moments give extremely good results. While coding performance increases by filter length, based on general signal characteristics, sampling rate and filtering properties, we conclude that lengths 6 and 8 are enough for most applications