Implementation of Wireless Spiking Neural Network for Classifications on MNIST Dataset

This paper presents the implementation of a handwritten digit recognition system using the Wireless Brain-Inspired Computing (WiBIC). Utilizing Asynchronous Pulse Code Multiple Access for wireless communication enables the transmission of encoded pulse trains while maintaining the firing time information of spiking neural networks. The relationship between the misdetection probability of APCMA and the system performance is evaluated by adjusting the input time intervals of image signals into the network and varying the APCMA pulse density. Experimental results demonstrate that the WiBIC system can realize the inherent learning capability of SNN in the wireless domain without impairment if the misdetection probability is in the acceptable range.