Robust Multitaper Reassignment Vectors for Enhanced Time-Frequency Representation

Time-frequency reassignment, commonly used to analyze time-varying signals, is a post-processing method that sharpens time-frequency representations. However, the reassign-ment method is sensitive to noise, making it unsuitable for time-frequency analysis when the signal-to-noise ratio is low. To increase the robustness of the reassignment technique, we develop a novel reassigned multitaper spectrogram (ReMuS). Two options for multitaper weights are explored: optimal weights and equal weights, where the optimal weights are designed to minimize the variance of the reassignment vectors in the presence of noise. Simulations suggest that the novel multitaper reassignment vectors have reduced variance in white Gaussian noise in comparison to the spectrogram reassignment vectors. Furthermore, it is demonstrated that the multitapers mitigate the influence of spectrogram zeros on the reassignment. Finally, the advantages of the ReMuS compared to the reassigned spectrogram are illustrated on the benchmark “bat signal”.