Deep Learning-Based Spectrum Prediction Collision Avoidance for Hybrid Wireless Environments

With a growing number of connected devices relying on the Industrial, Scientific, and Medical radio bands for communication, spectrum scarcity is one of the most important challenges currently and in the future. The existing collision avoidance techniques either apply a random back-off when spectrum collision is detected or assume that the knowledge about other nodes' spectrum occupation is known. While these solutions have shown to perform reasonably well in intra-Radio Access Technology environments, they can fail if they are deployed in dense multi-technology environments as they are unable to address the inter-Radio Access Technology interference. In this paper, we present Spectrum Prediction Collision Avoidance (SPCA): an algorithm that can predict the behavior of other surrounding networks, by using supervised deep learning; and adapt its behavior to increase the overall throughput of both its own Multiple Frequencies Time Division Multiple Access network as well as that of the other surrounding networks. We use Convolutional Neural Network (CNN) that predicts the spectrum usage of the other neighboring networks. Through extensive simulations, we show that the SPCA is able to reduce the number of collisions from 50% to 11%, which is 4.5 times lower than the regular Multiple Frequencies Time Division Multiple Access (MF-TDMA) approach. In comparison with an Exponentially Weighted Moving Average (EWMA) scheduler, SPCA reduces the number of collisions from 29% to 11%, which is a factor 2.5 lower.