Wavelet signal processing is broadly used for analysis of non-stationary data particularly, real-time seismic signals. In the geophysical analysis, numerous wavelet filters are developed to realize the signal characteristics by multi-level spectral synthesis. However, the selection of optimal wavelet family and wavelet filter for seismic wave analysis is a major issue and no rationale exists for choosing the appropriate wavelet filter. Our paper aims to solve this problem through evaluating various wavelet filters by two computational analyses, the first one is a descriptive statistical measure of spectral synthesis of seismic signals, and the second one is perfect reconstruction error of different wavelet filters. The test dataset contains 520 samples, which include normal tremors (i.e., ground motion signal with ambient vibrations), local-mining blasts and earthquake signals. These signals were subjected to single level decomposition by adopting Haar, Daubechies, Symlet, Coiflet and Biorthogonal wavelets families. Descriptive statistical measures (mean, standard deviation, skewness and kurtosis) are used to evaluate approximation (signal passed through low pass filter) and detail (signal passed through high pass filter) coefficients. Statistical results reveal applications of the descriptive statistics for characterising the seismic signals and understanding the ground response. The analysis of perfect reconstruction error of wavelets suggesting the strength of the wavelet filter is related to reducing the data redundancy. Based on these analysis, we found that Daubechies (db3 and db4), Symlet (sym3), Coiflet (Coif1), and Biorthogonal (BIOR3.5 and BIOR5.5) are the best wavelet filters to perform the seismic signal analysis.
