A variable step gain algorithm with fuzzy control

Abstract

AbstractNumerous methods for a parameter estimation in adaptive signal processing have been proposed. LMS, learning identification method, etc., for example, often are used due to easiness of their hardware construction, but their estimation speed and estimation accuracy vary by contrast depending on the magnitude of the coefficient called step gain. As a method for resolving it, a variable step (VS) algorithm in which the step gain is changed according to the parameter estimation condition has been proposed. However, since the variables in these algorithms that are intended to express the estimation conditions do not fully do so, it remains difficult to control the step gain appropriately.This paper defines anew variables to detect the change of parameters and disturbance and to comprehend the parameter estimating condition, and proposes a variable step algorithm for controlling the step gain by fuzzy controller with these variables as fuzzy input. Furthermore, computer simulation is implemented by applying this method to the learning identification method to show its effectiveness. In particular, with this algorithm we do not need to change the algorithm to deal with variations of the statistical properties of the input signal, and the computational complexity does not depend on the order of the filter; thus it is advantageous for higher‐order identification. Moreover, we have observed that it can cope with observation of noise, rapid change of unknown parameter, and it improves estimation speed by about 15 percent and estimation accuracy by about 20 percent compared with those of existing VS algorithms.