Grasshopper Optimization Algorithm for Blind Source Separation Based on Independent Component Analysis

Abstract

The Blind Source Separation (BSS) refers to the task of recovering the source signal from a known mixed signal (also called the observation signal). The core of BSS is to find a separation matrix W and Independent Component Analysis (ICA) has been intensively studied for BSS. However, when using traditional ICA, it is easy to fall into the local optimum and the convergence speed is slow. Moreover, the accuracy of speech separation remains inadequate. For this reason, we propose that Grasshopper Optimization Algorithm (GOA) is employed to search for the separation matrix W for the BSS in conjunction with the Negative Entropy maximization function. The results show that effective separation can be achieved by our method (GOA_BSS) for different types of data including the human speech and bird sounds in various scenarios considered. Specifically, Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) are used to compare with GOA. GOA is superior to the two algorithms in separation efficiency, separation stability, and convergence speed. In summary, GOA_BSS has achieved an efficient separation success rate (S-Rate) in the problem of BSS, and GOA_BSS has good generalization capability.