Fig. 4: Time-frequency analysis of EEG and acute electrophysiology signals. | Nature Communications

Fig. 4: Time-frequency analysis of EEG and acute electrophysiology signals.

From: Time-frequency super-resolution with superlets

Fig. 4

Data was recorded over occipital electrode Oz for EEG (a, b) and in mouse visual cortex for acute electrophysiology (ce). a Global time-frequency EEG power spectrum around stimulus onset computed using Fourier analysis (STFT; top), wavelets (CWT; middle), and adaptive additive superlets (ASLT; bottom). b Zoom-in analysis over the γ-frequency band (30–150 Hz) of data from a using STFT with various windows (top), CWT with different number of Morlet cycles (middle), and adaptive multiplicative superlets (bottom). Representations in a were first logarithmized (base 10) and both those in a and b were baselined (z-score) to 500 ms pre-stimulus period. Representations are averages across 61 trials. c Fourier (STFT; top), adaptive multiplicative superlets (ASLT; middle), and wavelet power spectra (CWT; bottom) around stimulus onset on mouse electrophysiology data. Representations were first logarithmized (base 10) and then baselined (z-score) to pre-stimulus period. d Zoom-in on a γ-burst from data in c, induced by the passage of the grating through the receptive field of cortical neurons. The SLT used multiplicative superlets of order 7 and c1 = 2, optimized to provide high temporal and frequency resolution (bottom left). By comparison, individual wavelets optimized for time (top left), frequency (bottom right) or a compromise between the two (top right) cannot reveal all the details evidenced by the superlet. e Further zoom-in on a detail from data in d provided by the superlet (top left) reveals two time neighboring packets (NP1 and NP2), a higher frequency packet (HP), and a lower ongoing rhythm (LOR). Tuned wavelets on 10–40 Hz band-passed data indicate roughly the presence of the temporal (left bottom) and frequency (top right) components. The location of HP cannot be determined by wavelet analysis in the average time-frequency spectrum, but is recovered by single-trial analysis, indicating that superlets can correctly reveal very fine time-frequency details, which are smeared out in the average spectra by the other methods. Absolute power shown (linear scale, no baselining) in d and e. Representations in c and d are averages across ten trials.

Back to article page