| Wireless networks, which liberate people from conventional constraints of timeand space, provide ubiquitous services and applications, and play a fairly significantrole in national economy. Because of the time variation and unreliability of wirelesschannels, medium access control (MAC) in data link layer has a significant influenceupon the throughput, capacity and delay of wireless networks. This dissertation studiesthe traffic prediction and MAC algorithm in single-hop wireless networks, MACalgorithm in multi-hop wireless networks and MAC mechanism in cognitive radio adhoc networks.In the current study of MAC algorithms, the incorporation of the characteristics ofnetwork traffic for improving the performance of wireless networks is a fairly difficultissue because the volatility and self-similarity features of network traffic in wirelessnetworks pose great difficulty to network traffic prediction. Chapter2of thisdissertation hence is devoted to the study of the traffic prediction algorithm in wirelessnetworks. We propose a novel network traffic prediction scheme based on theFARIMA-GARCH model. A new method is presented to obtain a zero-mean trafficseries by a piecewise two-way CUSUM detection algorithm. Then the fractiondifference order is evaluated with good precision by the proposed bounded searchmethod. After obtaining the necessary model parameters, the innovation series aremodeled by GARCH to track the volatility of the network traffic. Finally, the meanprediction resulted from the model is compensated. The proposed prediction methodkeeps the same time complexity as the FARIMA model prediction method, and thesimulation results show that the prediction performance is better than the FARIMAprediction method.In chapter3, the prediction algorithm in chapter2is applied to MAC design insingle-hop wireless networks, which combines the collision resolution algorithm (CRA)with the prediction results to obtain optimal allocation of wireless channels, and wepropose two CRAs. First, by modeling the packet inter-arrival time, we propose aprediction-based hybrid CRA, of which the performance, as shown by simulation results is better than the FCFS algorithm. Then, because of the complexity in small scale traffic,we propose a prediction-based two-stage tree CRA by modeling the aggregate trafficflow. The simulation results show that the two-stage tree CRA performs better than thebinary-tree splitting algorithm in terms of network throughput, average delay andcollision resolution period.Because of the low allocation efficiency of FPRP for unicast traffic, Chapter4proposes an enhanced MAC based on FPRP for multi-hop wireless networks, whichimproves spatial reuse ratio by the introduction of one more reservation round anddifferentiated services. Simulations show that the enhanced MAC has betterperformance than FPRP. Then, with the ability of receiving multiple packets by thephysical layer, we propose a new reservation MAC for multi-hop wireless networks,which utilizes multi-way handshakes for exchanging control information and getting theinformation of node's ability on multi-packet reception, and derive the throughputestimation formula under ideal conditions. Numerical simulations show that ourapproach is effective.Chapter5studies the network capacity and optimization of network throughput incognitive radio ad hoc networks (CRAHN). First, we derive the closed-form expressionof the upper bound of network capacity for CRAHNs under underlay spectrum accessmodel, which shows that this upper bound is only determined by the space distributionof nodes. Then we present a novel cross-layer optimization algorithm for maximizingthe network throughput, which adopts genetic algorithm (GA) to achieve the optimalneighbor selection and power allocation. The simulation results show that the obtainednetwork throughput achieves a performance closely approximate to the upper bound ofnetwork capacity.Finally, chapter6summarizes this dissertation and presents the future researchdirections.
|
PDF Full Text Request