Traffic Prediction for Cognitive Networking in Multi-Channel Wireless Networks

One of the early applications of cognitive networking in wireless networks is the problem of dynamic channel selection in multi-channel wireless networks. Dynamic channel selection requires gathering past and current traffic across multiple channels and predict future traffic loads on each of the channels for deciding best channel for the Access Point (AP) to operate on for serving wireless clients. Traffic prediction can be performed by employing Multi-layer Feedforward Neural Network (MFNN) models for learning the effect of spatio-temporal-spectral parameters on traffic patterns and predicting future traffic loads on each of the channels. In this paper, we construct three kinds of traffic predictors that predict traffic at different time scales: MLP (Minute Level Prediction), MILP (Minute Interval Level Prediction), and HLP (Hourly Level Prediction) schemes and study their prediction accuracy in a campus wireless LAN environment. We found that one can simplify the cost of traffic prediction by carefully choosing input parameters of neural network model based on underlying traffic characteristics of environment under study. It is also observed that each location has its own unique traffic pattern, which makes it hard to reuse traffic predictor designed for an environment in a different environment.