Data Management in Vehicular Networks - Relevance-Aware Networking for Advanced Driver Assistance Systems

Future vehicles will exchange an increasing amount of data to increase their awareness beyond their local perception. This data is generated by the sensors of other vehicles, which share their local view of the environment. Compared to the data exchanged by today's vehicles, this data is much more fine-granular and, thus, changes more frequently, requiring much higher bandwidth to maintain an up-to-date view of the environment. The diverse level of accuracy or potential inaccuracy of vehicle-generated data, in conjunction with their increased bandwidth volume, poses considerable challenges for future vehicular networks. The potential inaccuracy of data provided by other vehicles necessitates a validation, which requires knowledge about the measuring sensors. Besides, the higher bandwidth consumption requires a more accurate consideration of each vehicle's interest in data, as not everything can be exchanged. The paradigm of Approximate Networks is particularly well suited for the provisioning of fine-granular data, as it allows to trade network and computation resources with the availability and quality of data. Our contributions in this thesis amount to developing mechanisms to apply the concept of approximate networks in the vehicular scenario. For this purpose, we first develop mechanisms for the assessment of data in these networks, which are the basis for our approach to approximate vehicular networks. As our first contribution, we propose an aggregation scheme to increase the data quality in the network. Our innovative aggregation scheme considers the heterogeneity of sensors and data-specific properties to adapt the influence of old measurements and increase the quality of the resulting aggregate. We then investigate the relevance of data to a specific vehicle as our second contribution, which relies on the prediction of the specific vehicle's future context. By combining the accuracy of the aggregate and its relevance, we determine the expected gain for a specific vehicle, the so-called impact. This impact is key for effective data prioritization and builds the foundation of our approximate vehicular network. As our third contribution, we design and implement an approximate vehicular network based on Diverse Prioritization and Treatment, aiming at improving network performance without increasing the resources consumed, as typically advocated under approximate networking. A probabilistic mechanism is proposed to properly modulate the redundancy of the messages in the network, leading to their increased overall availability to the interested vehicles without increasing the consumed resources. Finally, we design and develop our VEHICLE.KOM platform that is used to assess the effectiveness of the developed mechanisms under varying environmental conditions. We show that our aggregation scheme drastically reduces the false aggregates and adapts its behavior to lifetime and accuracy effectively. In addition, we demonstrate the effectiveness of our approach to approximate vehicular networking, by showing a drastic increase in the network performance under dynamic network conditions, especially when considering cooperation between vehicles