Autonomous Vehicles: MMW Radar Backscattering Modeling of Traffic Environment, Vehicular Communication Modeling, and Antenna Designs

Abstract

77 GHz Millimeter-wave (mmWave) radar serves as an essential component among many sensors required for autonomous navigation. High-fidelity simulation is indispensable for nowadays’ development of advanced automotive radar systems because radar simulation can accelerate the design and testing process and help people to better understand and process the radar data. The main challenge in automotive radar simulation is to simulate the complex scattering behavior of various targets in real time, which is required for sensor fusion with other sensory simulation, e.g. optical image simulation. In this thesis, an asymptotic method based on a fast-wideband physical optics (PO) calculation is developed and applied to get high fidelity radar response of traffic scenes and generate the corresponding radar images from traffic targets. The targets include pedestrians, vehicles, and other stationary targets. To further accelerate the simulation into real time, a physics-based statistical approach is developed. The RCS of targets are fit into statistical distributions, and then the statistical parameters are summarized as functions of range and aspect angles, and other attributes of the targets. For advanced radar with multiple transmitters and receivers, pixelated-scatterer statistical RCS models are developed to represent objects as extend targets and relax the requirement for far-field condition. A real-time radar scene simulation software, which will be referred to as Michigan Automotive Radar Scene Simulator (MARSS), based on the statistical models are developed and integrated with a physical 3D scene generation software (Unreal Engine 4). One of the major challenges in radar signal processing is to detect the angle of arrival (AOA) of multiple targets. A new analytic multiple-sources AOA estimation algorithm that outperforms many well-known AOA estimation algorithms is developed and verified by experiments. Moreover, the statistical parameters of RCS from targets and radar images are used in target classification approaches based on machine learning methods. In realistic road traffic environment, foliage is commonly encountered that can potentially block the line-of-sight link. In the second part of the thesis, a non-line-of-sight (NLoS) vehicular propagation channel model for tree trunks at two vehicular communication bands (5.9 GHz and 60 GHz) is proposed. Both near-field and far-field scattering models from tree trunk are developed based on modal expansion and surface current integral method. To make the results fast accessible and retractable, a macro model based on artificial neural network (ANN) is proposed to fit the path loss calculated from the complex electromagnetic (EM) based methods. In the third part of the thesis, two broadband (bandwidth > 50%) omnidirectional antenna designs are discussed to enable polarization diversity for next-generation communication systems. The first design is a compact horizontally polarized (HP) antenna, which contains four folded dipole radiators and utilizing their mutual coupling to enhance the bandwidth. The second one is a circularly polarized (CP) antenna. It is composed of one ultra-wide-band (UWB) monopole, the compact HP antenna, and a dedicatedly designed asymmetric power divider based feeding network. It has about 53% overlapping bandwidth for both impedance and axial ratio with peak RHCP gain of 0.9 dBi.