Algorithm for Estimating AR(P) Model Parameters Using Burg's Method

This article discusses the algorithm for estimating AR(P) model parameters using Burg's method. The implementation utilizes forward prediction error (ef) and backward prediction error (bf) calculations, with mse representing the mean squared error of predictions. The algorithm efficiently computes reflection coefficients through minimization of forward and backward prediction errors, employing recursive updates similar to the Levinson-Durbin algorithm. In the final program output, prediction errors for all model orders are systematically arranged in a lower triangular matrix format, allowing for easy comparison of different model complexities. While the parameters estimated by this method may contain some inherent approximation errors, practical applications can improve model accuracy and predictive capability through algorithm optimization techniques such as proper order selection and regularization. During experimental phases, practitioners can compare results using different parameter values and model orders to determine the optimal configuration for specific datasets, potentially implementing cross-validation routines to assess performance.