A new autoregressive method for high-performance spectrum analysis

In this paper, a new method of autoregressive (AR) spectrum estimation is presented. It shall be called two-sided autoregressive spectrum estimation, because an interpolative or smoothing model is postulated, as opposed to the predictive (one-sided) model used in AR modeling. The matrix equations arising in the estimation procedures proposed in this paper exhibit a special structure. The exploitation of these structures leads to fast solutions that reduce the total number of computations by an order of magnitude compared with straightforward approaches. Also, special attention is directed to the constrained two-sided AR model. Simulation examples show higher resolution capability of the proposed method relative to the least-squares AR method.