Asymptotic behavior of order statistic least mean square (OSLMS) algorithms in nonGaussian environments

These algorithms modify the ordinary LMS algorithm by applying an OS filtering operation to the instantaneous gradient estimate. The OS operation in OSLMS can reduce the bias on filter coefficient estimates (relative to LMS) when operating in non-Gaussian environments and can also reduce the average squared parameter error when in steady state operation. Some supporting analysis is presented for these effects, and simulation studies are provided. Guidelines are suggested for the selection of the OSLMS algorithms based on the expected noise environment.<>