Blind Wiener filtering: estimation of a random signal in noise using little prior knowledge

The authors present a method for estimating a random signal component from a data vector consisting of a piece of a narrowband random sequence corrupted with additive noise. The correlation structure of the sequence is unknown. The method is based on rank reduction principles presented by Scharf and Tufts (1987). It achieves a lower mean squared estimation error than an unbiased minimum variance estimator at the expense of introducing bias into the estimate. Its superior performance over short data records makes it useful in rapidly changing signal environments. The performance of the method is analyzed and simulations to demonstrate its effectiveness are presented.<>