Algoritmo adaptativo tipo-LMS com soma do erro

Abstract

In this paper, implemented a new lter similar to the LMS, but, with a coast function based in the sum of the error. As a result, we obtain a very simple function, producing a rapid convergence and a small mismatch when compared with the LMS algorithm and other algorithms. The adaptive lter is based on non-linear functions such as estimation of the gradient of a surface performance. We use the gradient algorithm to update the weights. this update is based on high-order statistics to obtain information about the signs involved in the process, in order to improve the performace of the adaptive lter. Derive the equations based on Taylor series of non-linear functions, to achieve the criteria that ensures their convergence. We also do a weight vector covariance study in steady state and determine the equations that calculate the time constants in an adaptive process. Here the algorithm proposed, which uses a cost function and were made simulacoes Monte Carlo with real signals to validate the theory presented. In this role the α coefficients have been optimized to provide increased stability and better performance in its convergence speed.