Encoding and Decoding Mixed Bandlimited Signals using Spiking Integrate-and-Fire Neurons

Conventional sampling focuses on encoding and decoding bandlimited signals by recording signal amplitudes at known time points. Alternately, sampling can be approached using biologically-inspired schemes. Among these are integrate-and-fire time encoding machines (IF-TEMs). They behave like simplified versions of spiking neurons and encode their input using spike times rather than amplitudes. Moreover, when multiple of these neurons jointly process a set of mixed signals, they form one layer in a feedforward spiking neural network. In this paper, we investigate the encoding and decoding potential of such a layer. We propose a setup to sample a set of bandlimited signals, by mixing them and sampling the result using different IF-TEMs. We provide conditions for perfect recovery of the set of signals from the samples in the noiseless case, and suggest an algorithm to perform the reconstruction.
Comment: To appear in ICASSP 2020. Code is available at https://github.com/karenadam/Multi-Channel-Time-Encoding