An efficient algorithm for two-dimensional autoregressive spectrum estimation

An algorithm similar to the Burg algorithm is given for autoregressive estimation in one dimension. Coefficients of the filter used to decorrelate the signal are found recursively from lower-order filters. The coefficients needed to get the larger-sized filters can be found at each stage of the recursion from the correlations of lower-order filter outputs. Thus, the resulting estimate does not assume knowledge of ensemble averages. It offers a computational saving of 25% over the commonly used method, in which an estimate of the correlation matrix is first computed and then inverted.<>