Unsupervised and Supervised Machine Learning Model for Cross-Terms Suppression in Wigner-Ville Distribution of Non-Stationary Signals

In this paper, we present a novel method based on unsupervised and supervised machine learning model for cross- term suppression in the Wigner-Ville distribution (WVD) of non-stationary multi-component signals. In this study, we considered the linear frequency modulated (LFM) mono-components that are either overlapped or non-overlapped in the time-frequency (T-F) plane. In this methodology, the WVD of the multi-component signal is computed. Then the k-means clustering-based unsupervised learning algorithm is applied to the computed WVD to form the clusters in the T-F plane. The clusters are grouped into mono-components and cross-terms. Further, the linear regression model is used to estimate the missing samples of mono-components in WVD. The efficacy of the proposed method is evaluated by comparing the normalized Renyi entropy under different signal-to-noise (SNR) conditions with another existing method. The simulation results show that the proposed method is superior to the compared method when mono-components are overlapped and comparable when mono-components are non-overlapped in the T-F plane in terms of normalized Renyi entropy.