C-R bounds for the AR parameter estimation as a function of a data length

This correspondence presents the exact Cramer-Rao (C-R) bounds for unbiased parameter estimation of Gaussian autoregressive (AR) processes. In this particular instance, the computation of the exact bounds is actually simpler than that of some proposed asymptotic bounds. It is shown that in order to build the Fisher information matrix, the intractable analytical evaluation of the derivatives can be transformed into simple shift-matrix operations.