Research On Key Techniques For Multi-radio Multi-channel Wireless Mesh Network

Wireless Mesh Network(WMN) is s a new type of Multi-hop access network, which combines with self-organizing and self managing wireless architectures. The WMN is as the solution of "last kilometer" problem and solves the wireless network accessing ultimately. Much attention is focused on establishing the multi-radio multi-channel(MRMC) WMN which effectively decrease the interference and improve poor network performance that caused by the single channel WMN. The research on MRMC WMN at home and abroad is still in its infancy and no related literatures are reported about optimal channel assignment strategy, optimal routing and throughput prediction. In this thesis, in order to improve the throughput of MRMC WMN, 2 kinds of MRMC WMN are built with NS2 network simulation tool under Linux environment, 2 novel channel assignment methods named LB-TS(Load Balance Based on Tabu-search, LB-TS) and LB-GA(Load Balance Based on Genetic Algorithm, LB-GA) were proposed, one routing protocol which prefered to the 2 MRMC WMN established in this thesis was chosen, and based on Support Vector Regression(SVR), a MRMC WMN throughput prediction model was set up. In this theis, channel assignment algorithm, routing protocol performance analysis and network throughput prediction were analyzed to study key technologies in MRMC WMN. The main jobs and results are as follows:(1) The main restricting factors affecting WMN throughput were ascertained. Based on analysis of the structure and characteristics of WMN, hidden terminal problem, intra flow and inter flow interference are revealed as the key factor of restricting the throughput of WMN network.(2) The MRMC WMN simulation platform was set up. By comparison, Ramon’s scheme was refered as the effective sheme for extending single channel WMN to MRMC WMN. Using Ramon’s scheme, the network simulation tool NS2.33 code was modified to support MRMC WMN. Six-node and thirteen-node MRMC WMN model were built, with each mesh node equipped with two interface card and three available orthogonal channels to establish the simulation platform for channel assignment algorithms and routing protocol analysis.(3) Two novel channel assignment methods named LB-TS and LB-GA were proposed to decrease the interference between channels. Through applications of these two channel assignment algorithms, the traffic load on each link was estimated; distribution of channel flow was balanced; the bandwidth after channel assignment is close to the traffic demand of the link; the interference between channels was reduced. Comparing with the traditional TS algorithm, in the six-node MRMC WMN, MAC layer throughput were increased by 22.06% and 35.62% respectively by using the LB-TS algorithm or LB-GA algorithm; IP layer throughput were increased by 10.90% and 19.41% respectively by using the LB-TS algorithm and LB-GA algorithm. In the thirteen-node MRMC WMN, comparing with the traditional TS algorithm, MAC layer throughput were increased by 14.66% and 12.96% respectively by using the LB-TS algorithm or LB-GA algorithm; IP layer throughput were increased by 12.93% and 21.10% respectively by using the LB-TS algorithm and LB-GA algorithm.(4) A wireless routing protocol, which prefered to two MRMC WMN models was chosen as the default routing protocol in NS2 simulation.According to Ramon’s idea, the kernel codes of DSR, AODV and OLSR were modified to support MRMC WMN. Four routing protocol performance index in six-node and 13-node MRMC WMN were analyzed by the effect of LB-TS and LB-GA algorithm, with the change of hops between 1 and 4 and the change of traffic load among 0 Mbps and 77 Mbps. The results of experiments showed that the average network throughput of MRMC WMN was improved 1.78% and 3.25% by OLSR than AODV and DSR, and that LB-TS and LB-GA algorithms had no obvious difference with these three routing protocols. Therefore, OLSR routing protocol was chosen as the default routing protocol in NS2.(5) A new MRMC WMN throughput prediction model based on Support Vector Regression(SVR) was set up. According to the problem of low prediction accuracy of the present prediction models, Radial Basis Function(RBF) was chosen as the kernel function base on analysis of SVR kernel function. Six kinds of MRMC WMNs were established, LB-GA channel assignment algorithm was used in MAC layer, and OLSR routing protocol was loaded in network layer in NS2. Sample set and test set were established after data were collected from the simulation results of NS2 and key factors affecting MRMC WMN throughput were identified. The RBF kernel function parameters were optimized by using grid search-10 fold cross validation and the genetic algorithm. The kernel parameters where the optimization results have small Mean Square Error(MSE), were chosen to setup up the MRMC WMN prediction model based on SVR.(6) The MRMC WMN throughput prediction accuracy based on SVR was evaluated. BP neural network prediction model and EWMA prediction model were built with the data set adoped from NS2. MRMC WMN throughput was predicted and the prediction results ware compared by these three prediction models. The experimental results indicated that the prediction accuracy of the prediction model based on SVR is improved by 4.90% and 36.26% compared with BP neural network prediction model and EWMA prediction model. The MRMC WMN throughput prediction technology based on SVR provides a new idea for the network topology planning and network configuration optimization.The results above provide the technical support for decreasing the MRMC WMN interference, reducing the network congestion and improving the throughput prediction accuracy for MRMC WMN. It also breaks through the application bottlenecks of MRMC WMN and provids reference for applying the machine learning and optimization theory into the field of channel assignment and throughput prediction of MRMC WMN.