A new quasi-Newton adaptive filtering algorithm

摘要

A new algorithm for FIR adaptive filters based on the quasi-Newtonnclass of optimization algorithms is described. A series of theoremsndemonstrating the stability of the algorithm, boundedness and positivendefiniteness of the estimated autocorrelation matrix of the input signalnare provided. The internal variables of the algorithm and their effectnare also investigated in order to provide a better insight of thenalgorithm's behavior. Extensive simulation results are presented fornfixed- and floating point implementation which show that the proposednalgorithm has comparable convergence speed and superior robustnessnrelative to other known Newton-type algorithms. Furthermore, robustnessnis guaranteed for highly correlated or even nonpersistently excitingninput signals, which makes the proposed algorithm a powerful alternativento the LMS and the RLS algorithms