Tel Aviv-based Arbe has developed an automotive-grade (AEC-Q100) high-resolution radar chipset designed to help autonomous vehicles, as well as passenger vehicles equipped with advanced driver-assistance systems (ADAS), detect and identify objects. The technology can separate, identify, and track hundreds of objects in high horizontal and vertical resolution at a long range in a wide field of view.
Arbe claims its radar chipset generates an image 100 times more detailed than any other solution on the market today. The system is then able to take those images and simultaneously localize and map the environment.
The company’s proprietary millimeter-wave automotive-grade radar RFIC chipset includes a transmitter chip with 24 output channels and a receiver chip with 12 input channels. This processor is capable of processing 30 Gb/s of data, representing a virtual array of over 2,300 virtual channels. It can support over 100,000 detections per frame and is designed in accordance with ISO 26262 functional safety, enabling ASIL B (Automotive Safety Integrity Level) qualification.
The radar processing chip makes it possible to integrate smart detection algorithms, post-processing, and SLAM (simultaneous localization and mapping) into the chip. SLAM is the computational problem of constructing or updating a map of an unknown environment while simultaneously keeping track of an object’s location within it. AI-based post-processing and SLAM can be leveraged to identify and track objects, differentiate dynamic objects from their surroundings, predict multiple object trajectory, and conduct sensor fusion with camera and other, parallel, automotive sensor systems.
Achieving High-Res Object Separation
Arbe’s chipset transmits and receives signals from multiple antennas. By converting information from time to frequency domains, the system provides a 4D image with high azimuth and elevation resolution while simultaneously sensing the environment in long range with a wide field of view in real time. 4D imaging radar uses echolocation and the principle of time-of-flight measurement to capture a space in 3D. In addition to azimuth, elevation angle, and slant range, it takes into account Doppler frequency as a fourth measure. Arbe’s processing chip generates 30 frames/s of 4D image, with equivalent processing throughput of 3 Tb/s.
A radar system that has high-resolution object separation in azimuth and elevation will in theory lead to more accurate decision-making. When distance, height, depth, and speed are simultaneously assessed, radar can be repositioned from a supportive role to the backbone of the sensing suite.
Indeed, Arbe is so confident in its radar chipset that CEO Kobi Marenko says it will enable Level 3 automation in passenger vehicles without requiring LiDAR (light detection and ranging radar). Level 3 is a designation by SAE that means conditional automation in which a driver must still be prepared to intervene.
Arbe’s baseband processing chip integrates a radar processing unit (RPU) architecture with embedded radar signal-processing algorithms to convert massive amounts of raw data in real time while maintaining low silicon power consumption. The chip is built to provide typical power consumption of less than 4 W. Usually, today’s radar power consumption is around 10 to 20 W.
Dealing with Interference
As the number of deployed radars increase in the automotive sector, the likelihood that one radar’s transmission is received by another radar has also increased. Interference results in a host of issues, such as a degradation in the noise floor, leading to missed detections, or blind spots, at certain ranges or directions. It can also create ghost objects in certain cases—ghost objects are targets seen by the radar that don’t exist.
Reliable radar performance requires methods to identify and mitigate interference, or avoid it altogether. Many of these radars transmit chirps (a signal in which the frequency increases [up-chirp] or decreases [down-chirp] with time) on the same frequency bands, a reality that leads to signal mixing and increased collision rates.
The processor chip empowers an imaging radar system that produces detailed images, and separates, identifies, and tracks hundreds of objects in high horizontal and vertical resolution, on top of range and Doppler resolution. This radar aims to achieve 1- and 2-deg. azimuth and elevation. The horizontal field of view (HFoV) and vertical field of view (VFoV) are 30 and 100 deg., respectively. Range is about 300 m. With regard to latency, Arbe said 30 frames/s provides real-time frames every 33 ms, which enables a maximum latency (end of a frame until it’s received at the main ECU) of 34 ms.
Through enhanced frequency-modulated continuous-wave (FMCW) technology, Arbe’s radar actively avoids and mitigates chirp transmitter interference. This keeps channels clear to correctly track objects to ensure the road ahead is safe for drivers, passengers, and pedestrians.
The inability to distinguish threats from false alarms is a leading cause of autonomous-vehicle accidents. False alarms trigger radar to report phantom objects, which in turn perpetuate false positives and false negatives. Arbe claims its FMCW enhancement, channel separation, and advanced post-processing reduce false alarms with close to zero instances of phantom objects, eliminating both false-positive and false-negative scenarios.
Founded in 2015 by a team of semiconductor engineers, radar specialists, and data scientists, Arbe has secured $55 million from leading investors. Its chips will be made by Global Foundry using 22-nm depleted silicon-on-insulator (SOI) technology.