Capturing the Essential Aspects of Reliable Vehicular Communications

Vehicular communications promise to be an essential driver for roadside safety in the future, as well as an important building block of autonomous driving. Communication standards that consider this communication mode specifically have been published as early as 2007 and continue to be refined. However, the nature of this application makes it especially challenging to achieve a required reliability, with channels having large delay spreads and large Doppler spectra. Low-latency requirements are in place to ensure that emergency information is disseminated sufficiently quickly, favoring ad-hoc network setups. High ehicular densities can however cause problems through hidden node communications in such an environment. Testing all these aspects for a given communication scheme at once is a daunting task. It is infeasible to deploy enough vehicles to replicate all aspects in dense traffic scenarios. Hence, many people resort to simulations. These simulations have to introduce simplifications, and not all of these simplifications are valid. This thesis aims at identifying the aspects of vehicular communications that have to be modeled, from the physical channel over channel estimation to network performance. I do this by integrating mathematical models with software defined radios that act as transmitter and receiver for IEEE 802.11p, as well as vehicular channel emulators. The first step is to analyze the vehicular channel and identify the required emulator complexity. I work with vehicular channel measurements, and apply sparse techniques and Akaike’s information criterion to identify the number of multipath clusters observed in a vehicular channel. The results show that modeling four to six clusters already provides a representative channel compared to the measurements. Then, I use sparsely fitted measurement data, as well as stationary channel models defined by ETSI, to measure the performance of IEEE 802.11p in a single link communication setup. This evaluation proves that considering the underlying channel is essential, as the achievable packet performance depends strongly on the delay-Doppler configuration of the small-scale fading channel. I further include mobility simulations
of cars to simulate dense communication networks. Based on this network data, I introduce an
algorithm to reduce the network complexity, which allows me to measure the network effects
at the same time as the channel effects with a small number of communication nodes. Finally,
I introduce a stochastic approach to modeling burst-behavior based on the Gilbert-Elliott
model. I use maximum-likelihood estimates for the Gilbert-Elliott model, which is modified
to become nonstationary, and combine them with modeling of the interference. The results
show that the channel characteristics and the interference have to be modeled in tandem, as
neglecting one means the simulation is overestimating the achievable performance