A Blind Source Separation Algorithm of Non-stationary Signals Based on Local Polynomial Fourier Transform

摘要

The separation of non-stationary signals is a hot topic. The blind source separation algorithms based on time-frequency analysis can not only describe the time-domain feature and the frequency-domain feature of signals but also extract the time-varying feature of non-stationary signals. Blind source separation owns better performance in solving the problem of non-stationary signals separation. In traditional methods based on linear time-frequency analysis, the time-frequency resolution is not high. Meanwhile, in traditional methods based on bilinear time-frequency, there is the problem of the cross-term interference. According to these problems, this paper proposes a blind source separation algorithm to separate the source signals based on local polynomial Fourier transform that adjusts the time-frequency resolution and the cross-term interference. First, the observed signals that are received by the senors are processed through whitening. Then, the whitened signals are transformed to get time-frequency distribution matrices through local polynomial Fourier transform. What's more, a strategy of choosing time-frequency distribution matrices is proposed to detect the matrices that own the underlying diagonal or off-diagonal structure and remove other matrices. The thresholds of traditional strategies in other papers are not needed in this paper. Therefore, the strategy of this paper can avoid the blindness of choosing the threshold. Finally, a joint diagonalization of a combined set of time-frequency distribution matrices is done to estimate the mixing matrix and the source signals. The simulation results indicate that the proposed algorithm of this paper can extract the non-stationary source signals effectively and own better performance than other comparison algorithms.