Timely system identification and change detection requires real-time signal processing techniques. Among the many tools that help to understand real-world signal characteristics, time-frequency decomposition analysis has received a lot of attention as it does not place restrictions on signal stationarity and periodicity. In this paper, a bio-inspired framework to conduct real-time time-frequency decomposition of arbitrary sensor signals is proposed. The procedure can be detailed into four steps: 1) passing the signal through a parallel filter bank with distinct characteristic frequencies; 2) applying the Hilbert transform; 3) inhibition to the filtered signals; 4) forming spectrogram representation based on the inhibited Hilbert transform signals. This proposed technique is then applied to decompose a benchmark signal with known theoretical time-frequency representation with the performance of the method compared to existing time-frequency methods including the Hilbert-Huang Transform, the short-time Fourier Transform and the Wavelet Transform. The results show that the proposed method produces spectrograms comparable to the existing solutions yet can do so in real-time.
doi: 10.12783/SHM2015/24