UCLA

Large-scale wireless network design and analysis generally falls between two extreme approaches. The information theoretic approach trades scenario fidelity for analytical tractability. Its results often reveal fundamental design tradeoffs and performance limits under very generic assumptions. Numerical simulation campaigns capture the details of a particular network but at a high computational cost, precluding large-scale scenarios with many nodes. Because personal safety is at stake, the design of vehicular wireless networks poses a particularly daunting challenge, requiring both large-scale and highly detailed models. This dissertation aims to push information theoretic analysis towards reality and pull out detailed simulation in lieu of abstractions that improve its efficiency and, therefore, scalability.

The first analysis, however, facilitates the detailed simulation of a specific scenario on the highway. I present novel expressions for the time-domain correlations due to Doppler spectra proposed in the literature for modeling the highway propagation environment. The spectra are also re-interpreted to yield novel and intuitive non-isotropic angle-of-arrival scatterer profiles. Finally, I propose a sum-of-sinusoids software channel model that accurately synthesizes the fading paths' random processes.

In the second analysis, I take a semi-analytic approach to abstract the impact of the time-varying channel on effective throughput, or goodput, which accounts for physical layer over-head and re-transmission of lost packets. Based on an extensive simulation of a standard wire-less vehicular network physical layer, I derive a set of expressions that can predict the goodput given a number of signal, packet, and channel parameters. Further, I derive a metric that delimits the (nearly) optimal packet size (in the sense that it maximizes goodput) given the channel and signal parameters.

The third analysis extends the information theoretic statistics of an idealized orthogonal frequency division multiplexing (OFDM) system to one with practical constraints on channel estimation overhead. Incorporating heretofore separate results on channel estimation error, outage probability approximation, and goodput analysis, we identify a new tradeoff between the channel frequency-diversity and the training required to exploit that diversity. From this perspective, I find a novel optimization of the signal bandwidth of time-scalable OFDM transmissions to maximize goodput.

The final analysis derives new, closed form bounds and an approximation for the variance of narrowband vector transmissions in time-selective channels with training overhead. The bounds and approximation are shown to be accurate over a large range of channel and signal parameters. The expressions reveal the fundamental dependency of the variance on the product of the maximum Doppler shift and the vector duration.