Implementation and Validation of a Comprehensive, Noise-Aware FMCW Radar Simulation Framework

Abstract

Complete simulation environments for automotive scenarios require simulated sensor data. The widespread use of vision sensors led to the availability of comprehensive simulations of their measurements. The same level of detail is hardly attained for frequency-modulated continuous wave (FMCW) radar sensors, despite their numerous advantages in automotive applications. Standard radar simulations, in fact, solely focus on their ultimate output, i.e., point clouds with velocity information. Most state-of-the-art tools based on image-rendering and ray-tracing do not treat one or more important characteristics, such as reflection intensities, multiple antennas, and noise impairments. We present a comprehensive and high-fidelity simulation framework for FMCW radars, where images from an RGB-D camera model in the Unity game engine are manipulated to generate 3-D time-domain radar measurements. In addition, the framework provides thermal and phase noise (PN) modeling, radiation patterns, and corresponding LiDaR-like point clouds for ground truth. Signal processing techniques are performed on the generated data in the same way as on a real sensor, so that standard radar output is provided. The simulated data are compared and validated with real measurements collected in a parking garage, showing the accurate reproducibility of multiple scenarios. The overall characteristics of the proposed simulation are also compared with other simulator software in the literature.