Evaluation of Different Cutoff Frequencies of High-pass Filter for Online Spike Sorting

Abstract

Despite the ever-increasing demand for online spike sorting to be applied in various closed-loop neuromodulation experiments and treatments, the performance and bandwidth are still constrained by the strict requirement for time complexity. Initiatives for improving online spike sorting performance mostly started with the implementation and designing of sorting algorithms, assuming standardized data preprocessing operations are applicable to all cases and separable for evaluating sorting performance. However, we postulated that the cutoff frequency of the high-pass filter could affect the sorting performance, given that spike waveforms are informative in a broad band and would be distorted if the frequency characteristics of the filter and noise do not match. Based on this rationale, we have evaluated how cutoff frequency affects the spike sorting performance on both the synthetic and real datasets. It was demonstrated that, the cutoff frequency can have a huge impact on the sorting performance. Further, this impact was noise-dependent. For neural signals with homogeneous noise, the cutoff frequency would lead to greater disparity when the signal-noise ratio decreased. While for signals with different types of noise, when the noise was subject to a “1/f” power spectrum, higher cutoff frequencies would render better performance. However, lower cutoff frequencies were advantageous when the noise deviated from the “1/f” noise. Therefore, according to the evaluation, when the cutoff frequency of the high-pass filter was adaptively switchable, the spike sorting performance would be enhanced while sidestepping the challenges in designing sorting algorithms.