A realizable mean square error estimator applied to rank selection

This paper describes a practical method for estimating the mean square error (MSE) given a finite amount of sample data. The new estimator is applied to the critical problem of rank selection for reduced rank adaptive filters by selecting the rank that minimized the MSE estimate. Since this approach estimates the optimum filter rank, independent of the signal rank, it will work for any reduced rank algorithm, and in particular offers a working solution for non-eigen-based reduced rank techniques. The estimator's performance is simulated for the problem of detecting the optimum rank for identifying plane waves impinging a line array.