A stochastic V2V LOS/NLOS model using neural networks for hardware-in-the-loop testing

Many of the envisioned applications based on Vehicle-to-Vehicle (V2V) communication require a certain amount of information received from other road users. Urban scenarios pose a particular challenge to the communication quality for Vehicular Ad-Hoc Networks (VANETs) as obstacles such as buildings, foliage, and infrastructure attenuate the signal. These challenges have to be taken into account already at the development stage of applications. In this paper we introduce a wall-clock time test approach which is capable of emulating the availability of information depending on the topology of an urban scenario. To this end, we make use of a neural network to predict LOS/NLOS probabilities which can then in turn be used to predict packet success rates. Our method achieves a high prediction accuracy that enables the realistic testing of a device-under-test in terms of communication and computational load.