Abstract:In modern spectral estimation, because in some cases the steady random sequence which in Yule-Walker spectrum estimated is short, and the power spectral density of the ARMA random process is hard to approaches the real density, so kalman filtering was introduced to deal with this problem when we estimates the AR model parameter with the autocorrelation. To estimates the finally power spectrum, we take the AR model parameter and the white gaussian noise as filter’s input and partial parameters. The experimental result indicated that in the Yule-Walker spectrum estimated that joins the kalman filter, its computational accuracy and stability had certain enhancement, it can be one of the methods to resolve relevant issues.