Blind source separation algorithm based on modified bacterial colony chemotaxis

Abstract

Most blind source separation (BSS) algorithm use single-point optimization method which always have the disadvantage of slow convergence speed, bad separate precision and easily getting into the local optimization. In view of these disadvantages, recently, Chen proposed a multiple-point optimization algorithm for BSS named DPBCC, which overcome these disadvantages at a certain extent. But DPBCC uses the superior bacterial random perturbation strategy to solve the problem of local convergence, which cannot ensure that after random perturbation it will be an ergodic search of the domain. So the ability of global convergence still has to improve. This paper proposes a modified bacterial colony chemotaxis algorithm (CBCC) for BSS, combined the strategy of chaos search with the strategy of neighborhood random search, reaching to an ergodic search of the entire domain, solving the local convergence better, improving the convergence speed and separate precision further. Take the BSS algorithm based on kurtosis under the instantaneous linear model for example to do computer simulation. The results validate the superiority of CBCC by comparing with the existing ones.

Blind Source Separation Method Based on Neural Network with Bias Term and Maximum Likelihood Estimation Criterion

“…Many studies have proposed performance metrics, such as PI [ 29 , 30 ], [ 31 ], and signal-to-noise ratio (SNR) [ 32 ], for simulation experiments. These experiments and their evaluation metrics are useful for the study of algorithms because the simulated (or even experimental) signal carries information that accurately characterizes the system (i.e., the source signal and its statistical characteristics, and the system mixing matrix).…”

“…Second, the optimization algorithm of NN can overcome the problems of conventional BSS with a fixed step size, as well as batch processing, using a small batch variable step size algorithm. Based on the above-mentioned advantages of NNs and considering the slow convergence and low separation performance of conventional BSS methods in engineering applications [ 31 ], a BSS method that combines the maximum likelihood estimation (MLE) criterion and an NN with a bias term is proposed in this study. The method addresses problems in engineering applications in the following three ways: Improving the cost function for BSS: According to the MLE cost function, the training parameters of the feedforward NN ( W and b ) are added to the cost function as penalty terms through regularization to improve the learning performance.…”