Research Of Routing Protocols Based On Traffic Prediction In Ad Hoc Networks

An ad hoc network is a temporary multi-hop system composed of a group of mobile terminals with wireless transceivers. Because the mobile terminals may move in the network with arbitrary speed and mobility pattern, the routing protocols of ad hoc networks have to adapt to the real-time network topology change due to nodal mobility. However, the traditional on-demand routing protocol AODV selects the route with the shortest path criterion. In networks with high traffic, this criterion may cause packet loss, because the high traffic is transmitted through a small number of nodes. Therefore, the load balancing problem is a significant topic in ad hoc networks.The main work and results in this thesis are as follows:(1) It proposes a routing protocol based on load balancing LBR-AODV (Load Balancing Reversed AODV). The proposed protocol extends the R-AODV protocol and introduces a load balancing scheme for single routes and a cross-layer nodal congestion control algorithm. Hence, in the proposed protocol, the number of hops is not the only criterion for selecting the routes. The proposed protocol also avoids selecting routes containing congesting nodes, which subsequently balances the load of the whole network and improves the network performance.(2) It proposes a routing protocol based on nodal traffic prediction with wavelet neural networks WNNP-LBRP (Load Balancing Routing Protocol Based on Wavelet Neural Network Prediction). It first analyses the characteristics of the network traffic, and finds the prediction model which best matches the problem and has the highest prediction precision. After analysis, a BP neural network with a wavelet function as the transfer function is selected, a wavelet neural network prediction model is built, and relevant algorithms are developed and implemented. Thereafter, the prediction algorithm is embedded into the AODV protocol. The traffic is measured with the queue length of the MAC interface. The prediction model can precisely predict the traffic load in real-time. The proposed routing protocol compares the load with a dynamic threshold, and avoids heavily loaded nodes acting as relays. Before the network arrives at the congested state, the routing table is updated, which improves the performance of the network.(3) The performance of the routing protocols WNNP-LBRP, LBR-AODV and AODV are analyzed with NS-2simulations. The simulation results show that WNNP-LBRP and LBR-AODV outperforms AODV in terms of packet delivery rate, end-to-end delay and the routing overhead. The WNNP-LBRP protocol with prediction performs the best.