Adaptive signal processing using a modified gradient estimation technique

An algorithm based on the gradient descent approach is proposed for adaptive signal processing. The algorithm outperforms the least mean square algorithm (LMS) in terms of convergence speed and misadjustment noise, and meets the performance characteristics of the conjugate gradient (OG) and recursive least squares (RLS) methods without the attendant computational complexity. It is simple to implement and lends itself to real time processing. The algorithm can be implemented using lattice or tap-delay line structures. In this paper extensive computational results are presented to illustrate the algorithm performance in terms of convergence properties, and its applications in adaptive noise cancelling and adaptive spectral analysis.