Abstract:

The problem of simultaneous blind separation of suband superGaussian signals is addressed. Firstly, a mixture density model suitable for the case where these two types signals exist simultaneously is proposed. Then the nonlinear function is computed according to this density model. A unifying algorithm for separating these two type signals is obtained in the end. On the other hand, the fuzzy inference system to determine the step length of the proposed adaptive algorithm on line is applied. Therefore, the learning algorithm has faster convergence speed and smaller steady state error. Simulation experiments of separating phonetic signals and artificial synthetical signals demonstrate the validity of the proposed algorithm, and the interferencetosignal ratio (ISR) is calculated to describe the extension of the improvement of the proposed algorithm.