Nyquist interpolation improves neuron yield in multiunit recordings

Multiunit electrodes, in particular tetrodes and polytrodes, are able to isolate action potentials from many neurons simultaneously. However, inaccuracies in the post-acquisition reconstruction of recorded spike waveforms can affect the reliability of spike detection and sorting. Here we show that bandlimited interpolation with sample-and-hold delay correction reduces waveform variability, leading to improved reliability of threshold-based event detection and improved spike sorting accuracy. Interpolation of continuously acquired data is, however, computationally expensive. A cost-benefit analysis was made of varying sampling rates from 12.5 kHz (no interpolation) to 100 kHz (eight times oversampling, with respect to the Nyquist frequency), taking into consideration the final application of the data. For most purposes, including spike sorting, sample rates below 25 kHz with bandlimited interpolation to 50 kHz were ideal, with negligible gains above this rate. A practical benefit, especially for large electrode arrays, is that the bandwidth and storage requirements can be greatly reduced by using data acquisition rates at or slightly above the Nyquist frequency.