WLAN Channel Status Duration Prediction for Audio and Video Services Using Probabilistic Neural Networks

Due to massive increase in wireless access from smartphones, IoT devices, WLAN is aiming to improve its spectrum efficiency (SE) using many technologies. Some interesting techniques for WLAN systems are flexible allocation of frequency resource and cognitive radio (CR) techniques which expect to find more useful spectrum resource by modeling and then predicting of channel status using the captured statistics information of the used spectrum. This paper investigates the prediction accuracy of busy/idle duration of two major wireless services: audio service and video service using neural network based predictor. We first study the statistics distribution of their time-series busy/idle (B/I) duration, and then analyze the predictability of the busy/idle duration based on the predictability theory. Then, we propose a data categorization (DC) method which categorizes the duration of recent B/I duration according the their ranges to make the duration of next data be distributed into several streams. From the predictability analysis of each stream and the prediction performance using the probabilistic neural network (PNN), it can be confirmed that the proposed DC can improve the prediction accuracy of time-series data in partial streams.