Investigation of Influence of APCMA-based Wireless Communication on Neural Computation in Wireless Spiking Neural Networks

In this paper, we investigate the influence of pulse density and misdetection of the asynchronous pulse code multiple access (APCMA) to reservoir computing in a wireless spiking neural network (WSNN). In a WSNN, sets of neurons are connected with each other by wireless communication. Therefore, the computational capability of a WSNN depends on the performance of wireless communication. APMCA accomplishes immediate and reliable wireless communication by encoding information as intervals of successive pulses, but it suffers from misdetection of messages especially when the pulse density is high. By using NARMA2 as a benchmark task, we showed that the synchronicity between modules was more important than the decrease in the pulse density to achieve the higher performance.