Unmanned Systems Technology Dec/Jan 2020 | Phoenix UAS | Sonar focus | Construction insight | InterGeo 2019 | Supacat ATMP | Adelan fuel cell | Oregon tour | DSEI 2019 | Copperstone Helix | Power management focus

7 Platform one Analog Devices is developing a 77 GHz silicon radar that will output a point cloud in the same way as a Lidar sensor (writes Nick Flaherty). That would allow a central processing module to use data from radar and Lidar in the same way and reduce the number of radar chips in a driverless vehicle. “A high-fidelity point cloud with a lot of data embedded in it provides these algorithms with the data they need,” said Chris Jacobs vice-president, autonomous transportation & automotive safety at Analog Devices. “For example, how do you make an AI system with functional safety? I think there is still a lot of work to do on designing an AI system that understands functional safety. We are still several years from seeing full AI systems underpinning our automotive platforms. “What is emerging now is a greater understanding of what the sensor fusion algorithms need in terms of data, and you need a sufficient amount of data to train the algorithms and give the correct response. This is leading to what he called ‘appropriate edge processing’, adding additional processing on the same chip as the radar built on 28 nm CMOS “You now have cooperative radar subsystems that synchronise and trade information without being physically connected,” he said. “If I can fuse these streams of data at the edge, I can add velocity data as metadata, and rather than needing 3 to 6Gsamples per second it can be a decimated version [one-tenth of that] that goes to the central controller.” That would still be too much data for the CAN bus but not for a Gigabit Ethernet connection. This approach avoids the need for an optical network link. “These days in radar sensors you have a millimetre-wave IC handling down- conversion and data conversion. Next to this chip is a radar processor that takes the MIPI or LVDS interface and does the object detection and tracking, outputting an object list to the CAN bus for the central engine control unit,” he said. “So what we are doing is using bulk CMOS 77 GHz technology, because we knew we would need a lot more processing at the edge with a lot more radars in the car – up to eight or 10. “This means the cost has to be reduced, and CMOS is an ideal platform for these types of products to add this edge processing, reducing the leakage current and power, and size. It then allows us to remove the standalone radar processor and output a lower bandwidth.” Jacobs said he sees cooperative radar modules that communicate as the future for these systems. “The challenge then is to produce the same quality of output using three radars rather than five. Say you have two modules with an overlapping field of view that speak to each other and share information – that gives a higher quality result without increasing the hardware cost. “For example, a front-facing radar needs a lot of computing power to get a 200 m range with cascaded modules all synchronised together with more power and computing requirements. A corner radar with a shorter range, with one radar and the processing block, is more manageable. Overlapping cooperative corner radars can implement a front-facing radar algorithmically, for example by reconfiguring the shape of the radar signal from short to mid- to long range on the fly, without having to allow for settling time.” “I think corner radar is where you will see a lot of optimisation over the next five years.” Jacobs added that ADI is defining the strategy for the development of the chips as a combination of fixed-function hard-coded digital algorithms and an embedded processor tailor-made to these applications, with functional safety built in but with some configurable software on an embedded processor. “We understand the mathematics for cooperative radar so we will have that in the mix,” he said. “There will be incrementally more edge processing, with the hooks to support both non- cooperative and cooperative processing. “We are targeting this for 2022 and beyond. There will initial deployments in 2021, but I think the main roll-out will be in 2024.” Silicon radar matches Lidar Sensors Unmanned Systems Technology | December/January 2020 x Adding computing algorithms to a radar processing chip would allow fewer such chips to be used in a driverless vehicle

RkJQdWJsaXNoZXIy MjI2Mzk4