Spike communication of dynamic stimuli: rate decoding versus temporal decoding

Abstract

We compare two methods for extracting information about a dynamic analog stimulus from a spike train: a rate decoder, which is based on the number of spikes in a counting window, and a temporal decoder, which is based on interspike intervals. We derive analytical expressions for the distortion between a stimulus current that drives an integrate-and-fire spike generator and a reconstruction of that current based on the spike train. We consider intrinsic distortion, which arises due to the limitations of spike coding, and extrinsic distortion, which is due to explicit alterations of the spike train. We find that in the absence of extrinsic distortion, the temporal decoder is superior to the rate decoder. In the presence of extrinsic distortion, the temporal decoder is preferred when low latency is required, but the rate decoder is preferred when the latency requirement is relaxed.