Comparative analysis of discrete wavelet transform and singular spectrum analysis in signal trend identification

Abstract

Trend identification gives a general tendency of a signal and as such can represent the first step in signal change detection. This is often a method for extracting useful features from the measured signal that can be transferred to meaningful conclusions, which can serve as a basis for process control or predicting the future behavior of the process. Fourier analysis, owing to its simplicity, is traditionally applied in time series processing; however, it has shortcomings in the analysis of signals that are not periodic. As a result, a number of techniques with better performance in non-stationary signal analysis have been proposed. In this paper, we carry out comparative analysis of two techniques for trend identification of non-stationary signals: Discrete Wavelet Transform (DWT) and Singular Spectrum Analysis (SSA). The analysis involves examining the applicability of these techniques both in offline and in real-time cases. Comparative analysis, as a result, gives support for the decision in selecting the technique of choice depending on the application.