Abstract
The unbiased minimum variance estimation of the correlation function, r xx (k), of a wide-sense stationary random signal x(k) is presented. However, the theoretical minimum variance estimator, r xx (k), for r xx (k) is a function of not only x(k) but also the unknown r xx (k) and thus is not computable. Additionally, r xx (k) is not computationally efficient. The authors propose a modified estimator, r xx (k), implemented by a three-step computationally efficient algorithm without need of r xx (k). Finally, they show some simulation examples using Capon's minimum variance spectral estimator in which r xx (k) is used for the correlation function estimate to indicate that r xx (k) leads to very good results on spectral estimation.