A Simple Scheme for Adapting Time-Frequency Representations

Abstract

Signal-dependent time-frequency representations, by adapting their functional form to fit the signal being analyzed, offer many performance advantages over conventional representations. In this paper, we propose a simple, efficient technique for continuously adapting time-frequency representations over time. The procedure computes a short-time quality measure of the representation for a range of values of a free parameter and estimates the optimal parameter value maximizing the quality measure via interpolation. Many representations, including the short-time Fourier transform, the cone-kernel distribution, and the continuous wavelet transform, support adaptation, at a computational cost of only a few times that of the corresponding static representations.