Safety has been an increasingly important issue for the development of autonomous cars. Within the 3SA project, the team is working on a simulation platform with the vocation to solve the problems related to the safety of the autonomous car. Regarding the perception, we propose methods for modeling ADAS sensors in constrained environments. Radars are part of these sensors. By their ability to detect targets with different parameters as range, speed or even angular position over a range up to 200 meters but also by their low sensitivity to meteorological conditions, radars provide several advantages for perception. However, some environmental parameters disturb the radar and affect its accuracy and reliability. Our team, a joint collaboration between academic (IMT Atlantique-IRT SYSTEMX) and industrial (Renault, AVSimulation, OKTAL-SE, STELLANTIS) partners, is working on the development of a physic based virtual radar model and on stress patterns in synthetic scenes. This paper addresses a particular validation step: the construction of a virtual radar model based on real characterization. To obtain an accurate radar model, our method combined theorical radar concepts together with the use of a RF reference bench. We will see how real data of characterization of canonical forms obtained by the bench and the knowledge of the analytic RCS results of these forms allow to justify the results obtained with the real radar and to build the virtual model with a better level of accuracy.