Impact of Street Geometry on Target Tracking in Automotive Radar Networks

We propose a metric calledtarget trackingprobability for the analysis of automotive radar networks using stochastic geometry. Contrary to the classical one-shot detection probability, the tracking probability characterizes the run-length of successful detections. Technically, due to the correlation of run-lengths, this characterization is a challenging task, which we solve using de Moivre's generating functions. In particular, we investigate the impact of non-homogeneous street structures on the tracking probability by leveraging the recently introduced binomial line process and the corresponding Cox process driven by it. Based on the derived framework, we study a radar MAC (Medium Access Control) protocol in which automotive radars share the same channel. Finally, we suggest the optimal MAC parameter that optimizes the target tracking probability for different locations of the ego radar and different interfering vehicle densities. Our research reveals several system design insights that lead to optimal tracking of the target by adaptive selection of the radar parameters for various use-cases.