How do I design a compact radar sensor for a robotic collision avoidance system?
Compact Radar Design for Robotic Collision Avoidance
Radar-based collision avoidance is becoming standard in warehouse robots (AGVs), delivery robots, cleaning robots, and industrial cobots. Unlike ultrasonic sensors (limited range, narrow field of view) or lidar (expensive, limited in some environmental conditions), radar provides robust all-condition detection at low cost.
Radar SoC Selection Guide
- TI IWR6843 (60 GHz): 3TX/4RX, integrated DSP and MCU, 4 GHz bandwidth, range resolution 3.75 cm. Designed for industrial sensing. $15-25 in volume
- TI AWR1843 (77 GHz): 3TX/4RX, integrated DSP, 4 GHz bandwidth. Originally for automotive but widely used in robotics. $10-20 in volume
- Infineon BGT60TR13C (60 GHz): 1TX/3RX, lower cost, simpler configuration. Suitable for short-range applications (<10 m). $5-10 in volume
- NXP TEF82xx (77 GHz): 3TX/4RX, high-performance MIMO. Designed for automotive ADAS. $15-30 in volume
PCB Antenna Design
At 77 GHz (lambda = 3.9 mm), a half-wave patch element is approximately 1.3 x 1.3 mm (on Rogers RO3003, Er = 3.0). A 2x4 array of 8 elements fits in approximately 6 x 12 mm and provides approximately 12 dBi gain. The antenna feed network uses microstrip corporate feed with Wilkinson dividers. PCB substrate requirements: Er tolerance < 2%, loss tangent < 0.003 at 77 GHz, copper roughness < 0.5 um RMS. Standard FR4 is not suitable above approximately 10 GHz; Rogers or Isola high-frequency materials are required.
Array gain: G = 10 log(N_elements) + G_element ~ 12 dBi for 2x4 array
Detection range: R = (P_tx G^2 lambda^2 sigma / ((4pi)^3 SNR_min))^(1/4)
For 10 dBm TX, 12 dBi, 1 m^2 RCS, 10 dB SNR: R ~ 50 m
Frequently Asked Questions
Can this radar distinguish between a person and a wall?
Yes, with appropriate signal processing. A moving person has a non-zero Doppler signature (walking speed 1-2 m/s creates a 350-700 Hz Doppler shift at 77 GHz) while a wall has zero Doppler. Micro-Doppler analysis can further classify targets: a walking person has characteristic leg swing patterns, a rolling cart has a constant velocity signature. Machine learning classifiers trained on radar signatures can distinguish people, vehicles, shopping carts, and static structures with 85-95% accuracy.
How much does a complete radar module cost?
In volume (10K+ units): radar SoC $10-25, PCB with antenna $3-8, passives and power management $2-5, assembly $2-5. Total BOM: $17-43. At production volumes of 100K+, costs drop to $12-30 per module. This is dramatically lower than lidar ($100-500) and competitive with multi-zone ultrasonic arrays ($15-40 for the equivalent detection coverage).
What is the minimum detectable target size?
A 77 GHz radar with 4 GHz bandwidth and 12 dBi antenna can detect a human hand (RCS approximately 0.001-0.01 m^2) at 3-10 meters, a person (RCS approximately 0.5-2 m^2) at 15-50 meters, and a vehicle (RCS approximately 10-100 m^2) at 50-200 meters. The minimum detectable RCS depends on the radar's SNR at the given range, which is set by the transmit power, antenna gain, receiver noise figure, and processing gain.