Sample Partial Autocorrelation Function
S. Degerine · IEEE Transactions on Signal Processing · 1993
A new estimation procedure for a partial autocorrelation coefficient of a stationary time series based on a natural geometrical analysis of the sample data is proposed. A method for autoregressive parameter estimation, which is midway between the constrained least squares procedure of Burg and the unconstrained one given by the for- ward-backward least squares method, is then obtained. The method provides a stable filter and operates in a recursive model-order fash- ion. Simulation results indicate that the method eliminates some short- comings of classical least squares procedures and stays close to the ex- act maximum likelihood estimation method.