A fast self-optimized LMS algorithm for non-stationary identification: application to underwater equalization

An adaptive algorithm called FOLMS is proposed. The algorithm has two novel characteristics: it is self-optimized, and it outperforms LMS (least-mean-square) and RLS (recursive-least-squares) algorithms in all the cases when the model to identify is alternatively stationary and nonstationary. Moreover, it requires a small computational cost (4N +3 add, 4N+5 mult). This algorithm is particularly interesting in nonstationary cases when the optimal step-size value has large variations, i.e. mainly when the minimum MSE is not only a function of the noise power but also of the model to identify impulse response, as in underwater equalization and all inverse identification problems