Modelling Non-Stationary Signals by Time-Dependent AR Process with Time-Varying Gain

Abstract

In non-stationary signal modelling by time-dependent auto-regressive model, generally the gain, or the variance of the input signal is assumed to be constant. This paper considers the time-varying nature of gain and investigates its effect in estimating the evolutionary spectra in case of synthesized and natural signals. It reveals that if the change in gain is not considered it can distort the evolutionary spectrum to a large extent. The paper introduces a simple non-iterative method for estimating the time-varying gain, which performs satisfactorily for moderate variation of gain.

Indexing terms:

• Non-stationary signal
• Time-dependent model
• Auto-regressive model
• Time-varying gain
• Evolutionary spectrum