Delay Prediction for Intelligent Routing in Wireless Networks Using Neural Networks

Research on wireless routing protocols have been striving to achieve two important but contradicting goals: adaptability to the dynamic network conditions and efficient routing information diffusion. A possible approach to solve this issue is through intelligent use of the nodes´ past experiences of the network traffic conditions, and making predictions on the future network traffic conditions based on the experiences. Delay is a typical routing metric. In this paper, we present a scheme for predicting mean per-packet one-hop delays using neural network approaches. The predicted one-hop delays are then used by the nodes to participate in routing information diffusion. By experiments, we prove the feasibility of predicting mean delays as a time series using either tapped-delay-line multi-layer perceptron (MLP) network or tapped-delay-line radial basis function (RBF) network. Two types of inputs for prediction are used: a) the mean delay time series itself only, b) the mean delay time series together with the corresponding traffic loads. The advantages and limitations of these neural network approaches are discussed