Application of the 2D Local Entropy Information in Sparse TFD Reconstruction

Abstract

This paper investigates a method for information extraction from the time-frequency distributions (TFDs) based on the local Rényi entropy (LRE) calculated in 2-dimensional (2D) time-frequency (TF) regions. The obtained entropy map information has been projected on the time and frequency axes, estimating the local number of components. The local number of components obtained in this way has been compared to the existing 1D estimation method and applied in the shrinkage operator of a sparse TFD reconstruction algorithm. The obtained results confirm that the estimation based on the 2D entropy map achieves higher accuracy on the considered synthetic and real-life signals corrupted by noise when compared to the accuracy of the 1D method, improving both the shrinking operator classification and the TFD reconstruction performance.