Constrained Autoregressive Model for Burg Method

In this work, we propose a constrained autoregressive (CAR) model based Burg method for parametric spectral estimation. CAR Burg method is based on constraining the Mth reflection coefficient of an M-order AR model. The value of this constrained reflection coefficient can be used for controlling the frequency bias and improving the spectral resolution between two close peaks in the spectrum. The method guarantees stability of the model on certain condition. The benefits of the CAR model based Burg method are demonstrated by comparing its performance with other methods in applications such as pitch detection and separation of pitch from first formant in a voiced speech signal