The bicepstral distance between random signals: A new tool for comparison of ARMA models identification methods based on higher-order statistics

This paper deals with distance measures for signal processing and pattern recognition. It proposes a new distance between stationary random signals, called the bicepstral one, which can be easily converted in a distance between ARMA models. This distance is based on higher order statistics, and therefore is not phase blind. Thus, it provides a good tool for comparison of ARMA model identification methods based on higher order statistics.