Review of parameter estimation techniques for time-varying autoregressive models of biomedical signals

Biomedical signals are non-stationary and a research topic of practical interest as the signal has time varying statistics. The problem of time varying is usually circumvented by assuming local stationary over a short time interval, where stationary techniques are applied. However, features extracted from these methods are not always suitable and methods for non-stationary process are needed. Time varying signals are more accurately represented by time frequency methods and received most attention recently. Among the time frequency methods, parametric modeling such as TVAR has been promising over nonparametric methods with improved resolutions and able to trace strong non-stationary signal. Despite the success of TVAR in various applications it has drawbacks. This paper presents an extensive review on TVAR modelling techniques. Different approaches for TVAR modeling is presented and outlined. Principles, advantages, disadvantages of those techniques are presented concisely. And finally a new direction has been suggested briefly