Distributed traffic estimation and throughput optimization in random access networks
Date of Issue2016-06-08
School of Electrical and Electronic Engineering
Satellite Engineering Centre
Due to the shared nature of the wireless channel in a random access network, the wireless nodes in a network naturally interact with each other and thus affect the throughput performance and energy consumption of each other. The aim of this thesis is to illustrate the use of local information for the traffic estimation and the throughput optimization in random access networks. Specifically, this thesis investigates various key issues on the design of random access networks, including transmission strategy in a slotted ALOHA network, distributed traffic characterization and throughput optimization in a CSMA network and optimum backoff protocol design in a non-persistent CSMA network. The traditional use of channel state information (CSI) in a slotted ALOHA network will lead to Braess Paradox, i.e., more information resulting in worse throughput performance. This thesis first proposes a new method to exploit the CSI named adaptive CSI method to eliminate the Braess Paradox and thus to improve the throughput performance. The use of acknowledgement is explored to estimate the transmission probabilities of other nodes in the network. It thus enables the distributed implementation of the proposed method. Moreover, this research generalizes two simple yet effective guidelines to determine the best transmission strategy under different system scenarios. Secondly, two novel distributed algorithms are proposed to estimate the on-going traffic in a heterogeneous CSMA network. In contrast to centralized Markov chain models, the proposed methods are able to characterize the traffic without prior information. They are also able to detect changes in the traffic. With the use of the traffic estimation method, a throughput adjustment algorithm is also proposed to tune the transmission rate of each node with respect to a desired throughput demand. Finally, the throughput performance of a non-persistent CSMA network under imperfect sensing is thoroughly examined. It is discovered that the traditional binary exponent backoff protocol accounts for the throughput inefficiency in CSMA networks. A novel backoff protocol named probabilistic backoff CSMA (PB-CSMA) is proposed, which is able to achieve optimum throughput performance regardless of the number of nodes as well as the level of sensing errors. Using the clear channel assessment results, the distributed implementation of PB-CSMA is also introduced. Simulation and experimental studies have been conducted to evaluate the proposed algorithms in terms of accuracy, convergence speed and computational complexity. The results show that with proper use of local information, the traffic intensity can be accurately estimated, the throughput can be significantly improved and the energy consumption can be further reduced for random access networks in a distributed manner.