A flexible and approximate computing approach for time-frequency distributions

Eigen-decomposition of the kernel allows fast computation of time-frequency distributions. This paper proposes a scheme which exploits further approximation by adapting dynamically the accuracy of spectrograms to the eigenvalue magnitudes. The approach employs the wavelet transform followed by block-recursive Fourier transform and a compensation network. It provides a set of parameters for the users to trade accuracy against computational complexity so that the implementation can be tailored to a specific application.