Cross-frequency phase-amplitude coupling (CFC) refers to the co-modulation of the power of a fast oscillation and the phase of a slower one, and its use in human and animal electrophysiological studies has grown exponentially in the past decade. Here we show that conventional signal processing tools, and in particular the computation of instantaneous phase of filtered data, may introduce spurious CFC when repeated potentials exist in the data. To demonstrate this, we created two types of simulations. First, we created 1/f pink noise simulating electrophysiological data, and injected Gaussian potentials of various heights and widths with realistic jitters in their temporal separation. We show that upon filtering, the injected potentials cause a systematic shift in the phase of slower oscillations as compared to their original phase in the simulated data, such that the slow wave phase aligns with the timing of the potentials. The average time interval between the simulated potentials dictates the frequency-for-phase of the resulting CFC, whereas the width and height of the potentials mainly modulate the resulting range of the frequency-for-amplitude. Secondly, we used a real EEG dataset that does not feature CFC, and added to it similar potentials as described above. This resulted in strong spurious CFC, even when the added potentials were of low amplitude and were not readily detectable by visual inspection. Next, we show that similar patterns of activity can be found in electrocorticographic data, and suggest several tools that can help disclosing these confounds.
Meeting abstract presented at VSS 2014