Simulation has emerged as an alternative way of obtaining synthetic radar returns and antenna coupling for developing signal processing methods and detection algorithms used in Advanced Driver Assistance Systems (ADAS) [1]. One reason for this is that simulation is cheaper and less time consuming than measurement. Simulation can also be the only practical and safe approach when evaluating corner cases that would be dangerous or hard to realize in the field. However, obtaining synthetic radar returns through simulation presents unique challenges due to the massive electrical size of the computational domain at 77 GHz. A typical maximum range of 300m for commercially available radar sensors corresponds to approximately 75,000 wavelengths (λ) at 77 GHz. An average full-scale traffic scene with actors can easily be billions of cubic wavelengths (λ3) in size. Therefore, the use of full wave methods such as finite element method (FEM) becomes highly inefficient, if not impossible. Ray tracing asymptotic techniques employing different implementations of the shooting and bouncing ray (SBR) method have been used to address this challenge [1]. In one study, a facet-based, hybrid simulation was conducted to determine the impact of multipath propagation on the radar cross section of lowflying targets in a maritime scene [2], and in another, ray tracing techniques were used to obtain range-Doppler plots and the micro-Doppler response of a vehicle [3].

Although ray tracing techniques have made it possible to simulate electrically large electromagnetic problems, radar simulations present unique additional challenges. Specifically, the ray tracing, asymptotic techniques used are frequency domain solutions. Therefore, Doppler shifts cannot be directly computed from a single frequency sweep simulation across the bandwidth of an FMCW chirp signal. Instead, multiple successive static simulations need to be conducted to obtain enough data for a single coherent processing interval (CPI) [3]. Each of the static simulations are translated to reflect the distance moved within the pulse repetition interval (PRI). A single range-Doppler map is then obtained by conducting a 2D inverse fast Fourier transform (IFFT) [1]. Therefore, hundreds of simulations need to be conducted to obtain a single range-Doppler map. Another additional layer of complexity is introduced when simulating multiple-input multiple-output (MIMO) arrays. Assuming a MIMO array with NC virtual channels, the number of simulations is increased by a multiplicative factor of NC.