Radars are the crucial perception sensors for active automotive safety, UAVs, smart city, and many other applications. Radars are capable of running in the night and harsh weather conditions. The information provided is distance, azimuth, elevation, and the relative speed of the obstacles. In most cases, radars data perfectly complements cameras.

Conventional radar implements either RF part, low- and high-level processing. And those are performed with limited computing power as the sensors have strict cost constraints. Radar data is also processed separately if a radar works along with cameras and lidars in sensor fusion systems. It happens due to legacy constraints: radar should be capable of operating as a single sensor.

This is how a conventional data acquisition pipeline works:

Now and on, the radars are used with external processing units that merge the data from several sources (cameras, radars, lidars, positioning sensors, acceleration sensors). They are called sensor fusion systems. To fit sensor fusion systems, we suggest radar has no high-level processing. It becomes a source of raw data, like a camera or a lidar. Low-level processing becomes complex to fit sensor fusion system demand. This is how we changed the radar architecture to fit this approach

Below is a proposed radar data acquisition pipeline:

Such an approach allows solving some of the critical challenges that limited the radar penetration ratio:

• Radar is not a black box anymore. It provides a sensor fusion system with consistent Ethernet output for feature- or early (data) sensor fusion.
• The novel low-level engine allows extracting more detailed information on the sensor fusion system demand.
• The radar is simplified, and its cost is reduced compared to conventional radars.

Thus radar becomes compliant with 4th and 5th generations automotive E/E architecture:

• The entire high-level processing (clustering and further) is done by the CPU.
• Data acquisition and specific low-level processing are left to the sensors.

The typical configuration looks like this.

We believe it makes radar a key low-cost sensor for the ADAS, 4th and 5th generation E/E architecture.