Pulse Amplitude Modulation for Electro-optical Spiking Neural Networks

Abstract

Spiking neurons represent the most accurate model of the neural cells by using pulses and timing for information processing and adaptation. Visible light communication can be leveraged to establish a wireless link between neurons in spiking networks even when neural areas are in relative motions. Typically, parallel transmission in electro-optical spiking neural networks is performed using wavelength division multiplexing, which is limited by the number of wavelengths used and multiple bandpass optical filters. This paper explores the possibility of using multi-level pulse amplitude modulation (PAM) in multi-input-optical-axons (MIOA) integrated by the parallel neural paths in a spiking neuron network (SNN). To evaluate PAM-MIOA, we implement an electro-optical SNN that controls the force of two anthropomorphic fingers actuated by the shape memory alloy (SMA)-based actuators. The voltage threshold level in PAM is automatically adjusted based on the reference optical power. Results show that the electro-optical SNN is able to hold an object when using PAM-MIOA even with the link misalignment.