Rank reduction and James-Stein estimation
Jonathan H. Manton, Yingbo Hua · IEEE Transactions on Signal Processing · 1999
This correspondence addresses the problem of estimating the signal in a signal-plus-Gaussian-noise model when it is known that the signal lies in a given subspace. An alternative to rank reduction is presented. The new estimator has the remarkable property of having a smaller mean-square error than that of the maximum-likelihood (also least-squares) estimator for all parameter values.