WLAN Throughput Prediction Using Deep Learning with Throughput, RSS, and COR

Abstract

Throughput prediction of wireless LAN (WLAN) is important technology for effective use of frequency spectrum. In conventional throughput prediction methods, the future throughput is predicted by learning variations of throughput and some related information such as Received Signal Strength (RSS). On the other hand, the WLAN throughput is highly affected by Channel Occupancy Ratio (COR) due to carrier sense multiple access with collision avoidance. Therefore, this paper proposes simultaneous learning of throughput, RSS, and COR to learn the latent cause of the throughput variation. We compare the prediction accuracy of several prediction models, and it is confirmed that the accuracy is improved by the proposed simultaneous learning regardless of the network structure.