| Hi-index | 0.00 |
In this dissertation, two analytical models are presented in order to investigate issues regarding data throughput performance in wireless environments and multi-hop wireless sensor networks (WSNs). The wireless channel is highly unreliable, time varying, and has memory due to multipath fading. Losses are highly correlated and the channel characteristics change randomly in time on both slow and fast time scales. The wireless channel can be considered to be a scarce, shared resource. In a multi-hop wireless network, spatial channel reuse is desirable to improve channel utilization, and energy efficiency needs to be achieved to prolong the lifetime of the network. The first model presented is an analytic model for Transmission Control Protocol (TCP) performance over wireless channel with highly correlated fading characteristics. The wireless channel packet loss process is modeled using a Linear Algebraic Queueing Theory representation of a hidden Markov chain that can incorporate autocorrelations in successive packet losses. The packet loss model is then used in the development of a discrete time Markov chain representation of the evolution of the TCP congestion window. Transient and steady-state performance measures such as the mean and variance of the congestion window size and throughput for varying channel autocorrelation are presented. The second model is an approximation model for the throughput performance of multi-hop WSNs using queueing network theory. WSNs are modeled as multipoint-to-point queueing networks where blocking arises before transmission due to dependencies in trans missions among the neighboring nodes. Queueing networks with blocking are generally difficult to treat, and, except for special cases, an exact analysis is not usually attainable. Thus, an analytical approximation technique is used such that the network is partitioned into two or three nodes and product form solution techniques are applied. The dependencies in transmissions are captured based on the transmission periods of neighboring nodes. Fixed-point approximations are used to obtain the unknown probabilities that the neighboring nodes are idle. Queue length distributions are iteratively solved using matrix geometric solutions. Based on the queue length probability distribution, other performance measures, such as mean queue length, throughput, and delay, can be readily obtained.