Predictive Delay Metric for OLSR Using Neural Networks

In this paper, we propose an adaptability enhancement mechanism to be integrated with OLSR, and potentially any Mobile Ad Hoc Network (MANET) proactive routing protocol. The key of this mechanism is prediction and evaluation of the mean queuing delay as a routing metric. Neural network methods are used to predict delays. We investigated the pros and cons of using two types of neural networks, namely Multi-Layer Perceptron (MLP) and Radial Basis Function (RBF), in predicting nonstationary time series (e.g., mean queuing delay time series). We present TierUp -- our novel node-state routing table calculation algorithm, which is developed for the integration of delay prediction and OLSR. We name the extended version of OLSR as OLSRNN. We show through ns2 simulation that compared to OLSR, OLSRNN is able to increase data packet delivery ratio and reduce average end-to-end delay in scenarios with complex traffic patterns and various node mobility. Our simulation also shows the advantage of using neural network for delay prediction compared to moving average and exponential smoothing. The enhanced adaptability of OLSR_NN is further verified by the more balanced traffic observed in our simulation.