Advancing Connected Vehicle Technologies by Improving Vehicular Channel Model Accuracy and Safety Performance Measures

摘要

Wireless communications technologies allow vehicles to exchange information and thus create connected vehicle networks that enable safety applications, such as accident avoidance, thereby reducing damage and injuries caused by moving vehicle collisions. The most promising technologies in the U.S. that will enable such a vehicular ad hoc network (VANET) are collectively referred to as Dedicated Short-Range Communications (DSRC). While standards evaluation units exist, deployment has been limited to prototype testing, forcing VANET researchers to rely on simulation tools and supporting models, with mixed results. Results from inaccurate models can threaten the evaluation of safety applications, with existing performance metrics often only evaluating communications Quality of Service (QoS) measures while ignoring vehicular mobility.;In this dissertation, we explore common deterministic and stochastic vehicular channel models (VCMs), comparing their performance to measurement data from a real-world testbed, and evaluating their impact to safety performance metrics. First, we contribute to the ns-3 simulator an implementation of a VCM that supports obstacle shadowing using geodata and provide simulation results that compare the performance of the deterministic obstacle shadowing model to other common stochastic fading and shadowing models. Second, we study the packet-level performance of DSRC safety message receptions among vehicular encounters as derived from a large deployment of nearly 3000 DSRC-equipped vehicles operating near Ann Arbor, MI. We find that packet losses for many vehicle-to-vehicle (V2V) encounters differ significantly from traditional, static-node networks. Around Ann Arbor, packet losses exhibit temporal correlations when inter-packet gaps are 400ms or longer, but are mostly uncorrelated for shorter gaps. Third, we evaluate the performance of several existing VCMs and show that the UMTRI large-scale Ann Arbor testbed exhibits significant shadowing and fading effects that traditional VCMs often fail to capture. Fourth, we introduce BUR-GEN, a packet generation algorithm that improves upon other, common models in terms of packet burst pattern generations. Fifth, we propose SafeRelay, a safety message dissemination technique that floods geo-addressed safety messages within a nearby flooding zone, and evaluate packet delivery effectiveness using a new metric, probability of safety awareness, that combines packet delivery effectiveness with vehicular mobility. Sixth, we conduct a safety assessment that compares BUR-GEN and i.i.d.-based packet loss models to the observations found within the Ann Arbor testbed. We find that our burst-aware packet generation model improves awareness probability for maximum safety tolerance by a factor of 31. Finally, we motivate additional studies of VCMs that avoid the pitfalls we observe within models that are based on i.i.d. assumptions and instead employ bursty packet generator functions. The lessons learned from our studies motivate advances in connected vehicle technologies by improving vehicular channel model accuracy and safety performance measures.