Note on the estimation of the correlation function of neural spike trains

The use of time-bins in the estimation of the correlation function of neural spike trains has a filtering effect on the estimate and results in distortion and aliasing. Prior low-pass filtering of the spike trains, on the other hand, and computation of the correlation function of the emerging waveforms in the standard way result in an estimate that is also a filtered version of the original function but distortion- and alias -free. In addition, the correlation function so computed can be normalized. An analogous definition of the correlation coefficient for the first technique enables the comparison of these various correlation estimates and clarifies their properties.