Single Chip Radar Processing for Object Detection

• Main Authors: Nagalikar, Shivasharan, Mody, Mihir, Baranwal, Ankur, Kumar, VC, Shankar, Prithvi, Farooqui, Mohammad Asif, Shah, Malav, Sangani, Nikhil, Rakesh, Yc, Karkisaval, Abhishek, Chellappan, Sriraj, Sm, Stalin, Pratti, Anita, Rosales, Randy, Dixit, Tanu, Murphy, Sean
• Format: Conference Proceeding
• Language: English
• Subjects: AM273x; Costs; Digital signal processors; Microcontrollers; mmwave Platform; Object detection; Radar detection; Radar sensing; Radio frequency; Software design
• ISSN: 2158-4001
• DOI: 10.1109/ICCE56470.2023.10043549

Radar sensing and object detection use cases becoming popular in real-world applications e.g. people counting, self-driving cars, water level detection, vital sign monitoring etc. Typical solution consists of multiple components i.e. Radio-RF Frontend, ADCs, Microcontrollers (MCU) and multiple Digital Signal Processors (DSP) cores to address higher compute performance needs for these applications including 2 RF chip cascade for better radar performance. The paper proposes a single chip solution (except RF Frontend sensing) to address performance needs (e.g. 2 RF chip cascade) at optimized cost & power. The Proposed solution consists of innovative techniques, namely, integration of multiple processing components in a single chip, an efficient system mapping & software design for radar processing and FFT HW Accelerator to reduce number of DSPs. The proposed solution is available as AM273x SoC from Texas Instruments and supports high compute radar processing scenarios e.g. object detect. The solution uses FFT HW Accelerator, which enable fast processing in less than 6.4 usec for 4\times 256 point complex FFT in this application, which is 4 times faster than DSP.