An Improved Natural Gradient Algorithm for Blind Source Separation

This paper proposes an improved natural gradient algorithm for blind source separation (BSS) based on the constrained optimization method. The improved algorithm introduces a scaling factor that restricts the training process by the balance spot, which adds little computational complexity and overcomes the conflict between the convergence rate and the steady- state accuracy. Therefore, the new algorithm exhibits fast convergence and excellent performance. Computer simulation results show that the new algorithm is effective. And compared with the conventional natural gradient algorithm and the adaptive step- size algorithm, the performance of the improved algorithm is obviously better.