Automotive Radar Simulator Based on Ray-Tracing and Machine Learning

In this paper, we have designed software that can simulate a driving scenario with several vehicles or pedestrians. Each target has 4 parameters: its type, speed, distance, and RCS. The software generates simulated signals using Ray-tracing and Neural Networks, which are trained with a real radar dataset. The Neural Network part uses a Long Short-Term Memory (LSTM) Network for Scenarios with more than two targets and a simple multilayer perceptron (MLP) network for simulating scenarios with just one target. After starting the simulation, it shows an animation of the movement of different targets on a road. In addition, it can save the radar signals received by the radar receiver and draw four plots from signals as the real location of targets in the Range-Doppler map (RD map), 1D FFT of the signal, 2D FFT image, and estimated location of targets. It can produce a defined number of random scenarios for dataset creation in machine learning research, while also considering the impact of antenna patterns on the received radar signals.