Trigonometric decomposition of time-frequency distribution kernels

Trigonometric decomposition of the time-frequency distribution (TFD) kernels refers to representing each row of the kernel in the time time-lag domain by a finite number of cosinusoidal terms. This allows the local autocorrelation function to be recursively updated, yielding a efficient TFD. Unlike the spectrogram where the t-f desirable properties are compromised, the proposed trigonometric decomposition preserves both the support and marginal properties and permits data-dependent kernel design