Design and Analysis of Supervised and Decision-Directed Estimators of the MMSE/LCMV Filter in Data Limited Environments

摘要

We consider sample-matrix-inversion (SMI)-type estimates of the minimum-mean-square-error (MMSE) and the linearly constrained-minimum-variance (LCMV) linear filters obtained from data records of limited size. We quantify theoretically the (detrimental) effect of the desired-signal energy level on the mean square (MS) filter estimation error and the normalized output signal-to-interference-plus-noise ratio (SINR) by deriving a new exact analytical expression and a lower bound, respectively. For cases where accumulation of pure disturbance observations is not possible, we show theoretically how certain intuitive, pilot-assisted, and decision-directed adaptive filter implementations that utilize desired-signal-present data/observations perform close to their desired-signal-absent counterparts. Simulation studies illustrate our theoretical developments in the context of spread-spectrum communications over multipath fading channels under perfect and nonperfect synchronization.