Digital filters for firing rate estimation

Abstract

When a rate histogram is used to represent the firing pattern of a neuron there is the potential for serious error due to aliasing, and because of this the rate histogram is a very poor way to represent neural activity. It is theoretically possible to encode a signal in a spike train and decode it without error by filtering and sampling. There is no natural optimal filter design for this problem, but it is possible to specify the characteristics of a good rate estimating filter heuristically and design a filter with these characteristics. Two rate estimating filters are described here. Their performance has been tested, and compared to the rate histogram and the French-Holden rate estimating algorithm, by measuring their ability to recover signals encoded as impulse sequences by Integral Pulse Frequency Modulation (IPFM). These filters are simple to implement and perform well. They should be used in preference to the rate histogram.