Study on idle slot availability prediction for WLAN using a probabilistic neural network

Abstract

We have recently proposed a multi-band wireless local area network (WLAN) system as a solution to the increasingly crowded frequency space. Efficiency can be improved by an agile transceiver that transmits on an idle channel on either or both bands concurrently, and a busy/idle (B/I) predictor will form part of the sensing unit for such a system. A probabilistic neural network (PNN) is studied here for predicting upcoming WLAN B/I status based on pattern matching and classification of previous state patterns. IEEE 802.11 wireless data frames were captured at two hot-spots on multiple channels and the B/I status estimated. The prediction performance is compared for two different locations, channels, prediction matrix dimensions, B/I vs channel occupancy ratio (COR) input types, and frequency of retraining. Results show that the PNN has good potential to estimate the number of idle slots in the upcoming 20 slots and the performance improves with regular retraining.