Nonnegative least-mean-fourth algorithm

Abstract

Adaptive filtering techniques have been widely utilized in system identification. In some applications, nonnegativity is a desired constraint on the parameters to estimate because of the inherent physical characteristics of unknown systems. To address this problem, a useful adaptive algorithm, called the nonnegative least-mean-square (NNLMS), has been proposed for nonnegative system identification. In this paper, we propose a nonnegative adaptive algorithm, which is called the nonnegative least-mean-fourth (NNLMF). Compared to the NNLMS, the NNLMF shows good performance in the cases where the measurement noise is sine wave, uniformly distributed sequence, or square wave.