dc.description.abstract
Advanced driver assistance systems (ADAS) and autonomous vehicles (AV) aim to reduce accidents and congestion by improving their perception and decision-making algorithms. Equipped with various sensors such as LIDAR, RADAR, and cameras,ADAS and AV collect information about their surroundings. In addition, by sharing this sensor information, as well as vehicle kinematic data, between vehicles and other road infrastructure the sensor horizon of a single vehicle can be extended, overcoming the limits of line-of-sight sensor systems. This data sharing is enabled by vehicle-to-everything (V2X) communication. The vehicular communication channel is affected by several factors, including multipath propagation, which leads to Doppler and delay dispersion, shadowing from other moving objects, and signal obstruction by large buildings. Due to the movement of road users, the radio wave propagation effects can change rapidly. Therefore, extensive testing and verification is required to ensure that a given V2X communication link can reliably exchange information under vehicular radio channel propagation conditions.In this thesis we investigate methods for testing the reliability of vehicular com-munication channels and for verifying V2X communication hardware as well as the used transmit and receive signal processing algorithms. As a basis for testing, we present a hardware-in-the-loop (HiL) framework. This framework includes a channel emulator and a given transmitter-receiver modem pair. The channel emulator is controlled by a time-varying, non-stationary impulse response, which can be determined either by a numerical geometric radio wave propagation simulation or by empirical radio channel measurements. For the purpose of this thesis, IEEE 802.11pmodems are used in the HiL framework, though it is compatible with any other standards as well (LTE-V2X, 5G C-V2X, IEEE 802.11bd, etc.). The HiL framework abstracts the physical and data-link layer and provides the frame error rate (FER)as a communication reliability measure.Considering that real-world V2X communication scenarios involve more than onecommunication link, it is crucial to verify its reliability at the system-level. Therefore, in this thesis, we present a low-complexity yet accurate method for estimating the FER based on the HiL framework. For this purpose, we use a geometry-based stochastic channel model (GSCM) to enable a representation of the non-stationary vehicular fading process. Furthermore, we show an efficient way of computing the time-variant condensed radio channel parameters, which mainly determine the FERtime- and frequency limited region, i.e., the stationarity region, namely path loss,root mean square delay spread, Doppler bandwidth, K-factor, and line-of-sight Doppler shift. Using the HiL framework, the FER is measured for different values of the condensed radio channel parameters, stored in a FER table and looked up during the run-time of a system-level simulation.The FER is an important parameter for ADAS and AV algorithms. However, other vehicles transmit only a few frames per second, so predicting the FER based on the impulse response sequence of a frame is of great interest for direct implementation in a V2X modem. For this purpose, we investigate two different machine learning (ML) algorithms, namely deep neural network (DNN) and convolutional neural network(CNN). Both models are trained in a supervised manner, and the ground truth of the used dataset is obtained by the HiL setup and the GSCM model. Finally, taking into account the time and computational cost of training and retraining the entire models from scratch, we further investigate methods for adapting the ML models to new datasets of limited size while maintaining their original accuracy.
en