Skip to Main Content
This paper presents an alternate formulation of the least mean square (LMS) algorithm by using a set of angle variables monotonically related to the filter coefficients. The algorithm updates the angles directly instead of the filter coefficients and relies on quantities that can be realized by simple CORDIC rotations. Two architectures based on pipelined CORDIC unit are proposed which achieve efficiency either in time or in area. Further simplifications result from extending the approach to the sign-sign LMS algorithm. An approximate convergence analysis of the proposed algorithm, along with simulation results showing its convergence characteristics are presented.