Analysis of LMS-Newton adaptive filtering algorithms with variable convergence factor

Improved data-selective LMS-Newton adaptation algorithms

“…In this paper, two improved LMSN algorithms are described as alternatives for the LMSN algorithms proposed in [4]. The two algorithms yield a reduced steady-state misalignment relative to that in the known LMSN algorithms while requiring approximately the same number of iterations to converge.…”

“…These algorithms use an estimate of the inverse of the input-signal autocorrelation matrix to improve the accuracy of the updates of the weight vector. Unlike the LMS algorithm, the algorithms in [2][3][4] do not require a priori information about the input-signal statistics to control the learning rate. The recursive least-square (RLS) algorithm in [2] yields faster convergence than the LMS algorithm.…”

“…Unfortunately, it converges slowly especially when the input signal is colored. To improve the speed of convergence, several alternative algorithms have been proposed in [2][3][4]. These algorithms use an estimate of the inverse of the input-signal autocorrelation matrix to improve the accuracy of the updates of the weight vector.…”

“…The quasi-Newton algorithm reported in [3] is numerically robust but has poor tracking performance relative to the RLS algorithm [5]. The LMS Newton (LMSN) algorithms in [4], on the other hand, exhibit faster convergence and better numerical robustness relative to…”

“…The paper is organized as follows. In Section 2, the LMSN algorithms proposed in [4] are briefly reviewed. Then improved versions of these algorithms are described.…”