Practical comparison of adaptive filter algorithms

Linear filters can be designed to pass signals of interest and suppress undesired signal components. Wiener [1942] originated work to determine conditions for an optimal linear filter. The present paper reviews the problem of optimally reconstructing one signal as a filtered version of a second signal. This problem has applications in noise reduction, echo or interference cancellation from a signal of interest, system modeling, and equalization. Also included is an overview of methods to implement optimal, adaptive linear filters. The focus of the paper is on contrasting two methods for the particular problem of restoring speech signals in loud audio backgrounds. The discussion includes consideration of the many problems encountered in applications of adaptive processing. Trade-offs include sacrificing steady-state MSE for adaptation rate, controlling errors introduced by round-off selecting filter order, and operation count (complexity)