On frequency dependencies of sliding window correlation

Sliding window correlation (SWC) is one of the most popular methods to study the dynamics of functional connectivity from resting-state functional MRI scans. These scanned signals are often non-stationary and normally bandpass filtered before the SWC analysis, so there are more than one frequency component present in the correlating signals. In this paper we study the effects of two frequencies in such signals by extending the findings of an earlier study that considered only a single frequency. Although the fluctuations in the SWC are reduced when the length of the window approaches the minimum window length, as reported in the earlier study, but the ability of the SWC to give the stationary correlation value dependents on the ratio of the two frequencies rather than the minimum frequency only. Furthermore, some undesirable components other than the modulating frequency would be extracted from non-stationary analysis under certain conditions.