The stability of the LMF (Least Mean Fourth) adaptive algorithm is improved by using the median of the gradient instead of the raw gradient for updating the coefficients of the adaptive filter. This operation has two advantages: first it improves the performance of the LMF in impulse noise environments and secondly it enables the use of larger step-size of the adaptive algorithm especially when the signals are corrupted by noise. The disadvantages are a small raise in the computational complexity and slower convergence than the LMF. Two examples are given which illustrates the behavior of the Median LMF compared to the LMF algorithm.
|Title of host publication||NORSIG'98|
|Place of Publication||Vigsø|
|Publication status||Published - 1998|
|Event||3rd IEEE Nordic Signal Processing Symposium - Vigsø, Denmark|
Duration: 8 Jun 1998 → 11 Jun 1998
|Conference||3rd IEEE Nordic Signal Processing Symposium|
|Period||08/06/1998 → 11/06/1998|