Channel Model for Spiking Neural Networks Inspired by Impulse Radio MIMO Transmission

Abstract

In analogy with impulse radio counterpart in wireless communication, a channel model for spiking neural networks (SNNs) is proposed. The neuron channel model includes fanout and fanin with weighted sum in axons, synapses, and dendrites. It is based on a wireless channel model with multiple-input-multiple-output (MIMO) air and digital interfaces, but is orders of magnitude scaled down in size and the group velocity of signaling waves for broadcasting and superposition. The neuron channel matrix can flexibly encode weights and delays per synaptic connection to facilitate native processing of temporal SNNs by wave passive propagation dynamics over the physical media.