On feed-through terms in the lms algorithm

Abstract

The well known least mean squares (LMS) algorithm is studied as a control system. When applied in a noise canceller a block diagram approach is used to show that the step size has two upper limits. One is the conventional limit beyond which instability results. The second limit shows that if the step size is chosen to be too large then feed-through terms consisting of signal times noise will result in an additive term at the noise canceller output. This second limit is smaller than the first and will cause distortion at the noise canceller output.