Imaging radar is a key perception technology for automotive and industrial applications. A lot of progress has been made with high channel count systems, deploying, for example, 12 transmit and 16 receive channels with cascaded monolithic microwave integrated circuit solutions. Nevertheless, fully automated driving requires even higher angular resolution for drive-under/drive-over decisions and exact predictions of object trajectories in dense urban driving scenarios. Both problems can be solved by increasing the antenna size and building larger radars. However, there is a physical limit to what can be placed on the front of a car, and manufacturing very large arrays is quite difficult. Thus, coherent automotive radar networks are a way to achieve high spatial resolution and obtain the complete velocity vector of an object from a single measurement.
This solution is commercially attractive, as the sensor can remain relatively small, and complexity can be moved from the physical hardware to algorithms and processing. Two different test setups, each comprised of two multiple-input multiple-output radar units in the 76-77 GHz band, are presented in this article. To obtain azimuth and additional elevation information, the setups use a 1D and a 2D antenna array, respectively. Processing-based coherent evaluation is employed to create an additional radar image with doubled azimuth resolution and improved signal-to-noise ratio and to enable the estimation of vectorial target velocities. These benefits are presented and compared with optical reference images in traffic scenarios.