Modified partial update EDS algorithms for adaptive filtering

Abstract

Partial update (PU) Euclidean direction search (EDS) algorithms have been developed to reduce the computational complexity of the full update EDS. In this paper, the PU EDS is modified to achieve better performance. The performance is analyzed for a time-invariant system and for a time-varying system. Theoretical steady-state mean and mean squared error results of the modified PU EDS are derived for both time-invariant system and time-varying system. Computer simulations are presented to support the theoretical analyses. The modified PU EDS can achieve similar performance to the full update EDS while reducing the computational complexity significantly. The performance of the modified PU EDS is also compared with the PU recursive least squares (RLS) algorithm and the PU conjugate gradient in computer simulations. The performance of modified PU EDS is comparable to PU RLS, and it needs less computational cost.