Normalized LMS Algorithm Degradation Due to Estimation Noise

The steady-state weight vector derived by either the least mean square (LMS) or normalized least mean square (NLMS) algorithms has random deviations from the optimum values. These deviations increase the steady-state residue power. A previous paper derived the LMS weight noise effects for a multiple sidelobe canceller (MSLC) application. This paper describes the NLMS weight noise effects. It is shown that for a thermal noise environment, the weight noise effect for the LMS algorithm is insignificant but is quite significant for the NLMS algorithm. Calculations for example noise plus interference environments imply that the NLMS weight noise effects are always larger than that for LMS.