c****m
发帖数: 91 | 1 Well, the theoretical analysis of LMS is quite involving. As far as I know,
normalized LMS still has to resort to independent analysis (e.g., Simon
Haykin's Adaptive Filter Theory) or stochastic differential equations
(e.g., Solo&Kong Adaptive Signal Processing Algorithms). From your
observations, the reason for possible divergence of some parameters could
be 1. fast time-varying 2. bad autocorrelation structure 3. model mismatch
4. unmodeled errors ...
You might want to try: 1. change your step |
|
