A fast technique for finding the stationary autocorrelation matrix from a perturbational measurement of the gradient

摘要

It is shown that, when using a synchronously spaced FIR (finite impulse response) data equalizer with complex weights, one can estimate the autocorrelation matrix by empirically measuring the gradient of the MSE (mean square error) performance surface using only the real portion of the maximal time-delay weight located at the far end of the FIR filter. Since one only needs to perturb one weight of the FIR filter to find the autocorrelation matrix, this represents a minimal computation technique. This technique is of particular usefulness when one does not have access to the tap weight data vector.