Skip to Main Content
Auto-regressive modeling of measured data is commonly used in numerous signal processing applications. When aiming for high accuracy, Burg's method has been found to give a suitable model. It has been shown that when the signal energy is non-uniformly distributed in a frequency range, the use of a modified frequency scale is advantageous. This is often the case with audio signals. We introduce a frequency warped version of Burg's method for calculating the auto-regressive filter parameters. A bilinear frequency mapping can be embedded in Burg's method by replacing the unit-delays of the lattice structure used in Burg's method with first-order allpass filters. The benefits of the frequency-warped Burg's method are demonstrated by comparing its signal modeling performance against those of the conventional Burg's method and the warped Yule-Walker method.