Development of Low-Cost Single-Chip Automotive 4D Millimeter-Wave Radar

Yongjun Cai, Jie Bai, Hui‐Liang Shen, Libo Huang, B. Dharma Rao, Haiyang Wang

TL;DR

This paper presents a novel, low-cost, single-chip 4D automotive millimeter-wave radar system that overcomes the limitations of traditional 3D radars by incorporating target height information. The developed system achieves high accuracy in detecting both dynamic and static objects, paving the way for enhanced safety and reliability in autonomous driving.

What Problem Does This Paper Address?

Current automotive radar systems, often limited to 3D (range, azimuth, velocity), struggle to differentiate objects with similar horizontal positions but different heights. This ambiguity can lead to critical misinterpretations, such as mistaking a manhole cover for a road sign or vice-versa, posing a significant safety risk. While 4D radars (adding elevation) offer a solution by providing height information, existing high-performance 4D radar systems are often complex and expensive, hindering widespread adoption. This paper addresses the need for a cost-effective, single-chip 4D radar solution that can accurately perceive the environment in complex scenarios, improving autonomous driving safety.

Methods

The researchers developed a single-chip 4D millimeter-wave radar system with a focus on system architecture, antenna optimization, and signal processing. Key methodological contributions include:

  • Single-Chip Design: A highly integrated hardware platform utilizing a sparse MIMO antenna array was designed to reduce cost and form factor. The system employs the Texas Instruments AWR2944 chip, integrating essential RF components and enabling advanced radar operations.
  • Antenna Optimization: A comb-shaped radiator antenna was optimized for gain and radiation efficiency. Current distribution control was implemented to suppress sidelobe levels, achieving a 60° horizontal beamwidth and 16° elevation beamwidth.
  • Signal Processing:
    • FMCW Radar Signal Processing: Standard Fast Fourier Transform (FFT) techniques are used for range and velocity estimation. Angle estimation is performed using an FFT on the synthetic virtual array formed by the MIMO configuration.
  • Rain Clutter Identification: A novel method is proposed that statistically analyzes the velocity and distance distribution of raindrops to distinguish them from actual targets.
  • Noise Point Suppression: An algorithm based on the variance of angular FFT peak amplitudes is introduced to filter out noise points (e.g., ghost reflections, ground clutter) while preserving genuine targets. This method leverages the distinct amplitude differences between peaks in the FFT profiles of legitimate targets versus noise.

Datasets

The paper describes extensive field testing and data collection. Specific datasets mentioned include:

  • Rainfall Data: Approximately 40 minutes of data were collected under normal rainy conditions using a compact vehicle equipped with the radar. This data was used to analyze raindrop velocity and distance distributions.
  • Road Testing Data: Data was collected from various road surfaces (pavement, gravel, dirt) and in different scenarios, including:
  • Testing with two vehicles in adjacent lanes at different longitudinal distances.
  • Testing with a variety of targets including trucks, cars, electric bicycles, pedestrians, overpasses, and metal signage.
  • Specific scenarios for evaluating elevation performance with pedestrians and ground-mounted corner reflectors.

The paper does not explicitly mention training datasets for machine learning models, as the focus is on signal processing algorithms and hardware design.

Key Insights and Ideas

  • Height is Crucial for Ambiguity Resolution: The paper strongly emphasizes that adding elevation data (4D radar) is critical for resolving ambiguities that 3D radars cannot, leading to more robust object detection and safer autonomous driving.
  • Single-Chip Integration is Key to Cost-Effectiveness: By moving to a single-chip design with a sparse MIMO array, the researchers demonstrate a path towards significantly reducing the cost of 4D radar systems, making them more accessible for mass deployment.
  • Targeted Noise Suppression is Essential: The proposed methods for rain clutter and general noise point suppression are vital for improving the reliability of automotive radar in real-world, often challenging, conditions. The amplitude-based noise filtering is particularly interesting for its simplicity and effectiveness.
  • Elevation Angle Provides Powerful Discrimination: The distinct elevation angle characteristics of ground-level objects (corner reflectors) versus above-ground objects (pedestrians) offer a robust way to differentiate them, even at close ranges.

Key Results

  • High Angular Accuracy: The developed radar achieves a maximum measurement error of approximately ±0.2° for azimuth angles within ±30° and ±0.4° for elevation angles within ±13°.
  • Effective Target Detection: The radar reliably detected dynamic targets (vehicles, pedestrians, bicycles) and static infrastructure (overpasses, traffic signs) in extensive road tests.
  • Rain Clutter Rejection: The proposed raindrop identification method showed promising results in distinguishing raindrops from actual targets, aligning with precipitation intensity.
  • Noise Point Elimination: The noise suppression method successfully identified over 90% of unwanted points in test scenarios, with a false identification rate of less than 5% for legitimate targets.
  • Ground Clutter Rejection: The algorithm achieved a >70% ground clutter rejection rate.
  • Vehicle Multipath Detection: The system demonstrated ~90% accuracy in detecting vehicle multipath.
  • Detection Range: The radar achieved a maximum detection range of 210 m.

Limitations and Future Work

  • Elevation Angle Resolution: The paper notes that beyond 80 m, the limited elevation angle resolution (6.1°) and multipath reflections can make it difficult to fully resolve elevated targets from subsurface ghosts, leading to indistinguishable targets at certain ranges.
  • Cost vs. Performance Trade-offs: While the single-chip design offers a cost advantage, the paper acknowledges that achieving ideal imaging effects with this approach is challenging compared to more complex, multi-chip solutions.
  • Future Work: The authors suggest that the developed single-chip 4D radar represents a promising direction for future forward-looking radar systems, anticipating its widespread adoption in vehicles due to its cost, performance, and commercialization advantages. Further research could focus on improving elevation angle resolution at longer ranges and exploring more advanced signal processing techniques for even greater accuracy and robustness.

Summary updated

Abstract

Traditional 3D millimeter-wave radars lack target height information, leading to identification failures in complex scenarios. Upgrading to 4D millimeter-wave radars enables four-dimensional information perception, enhancing obstacle detection and improving the safety of autonomous driving. Given the high cost-sensitivity of in-vehicle radar systems, single-chip 4D millimeter-wave radar solutions, despite technical challenges in imaging, are of great research value. This study focuses on developing single-chip 4D automotive millimeter-wave radar, covering system architecture design, antenna optimization, signal processing algorithm creation, and performance validation. The maximum measurement error is approximately ±0.2° for azimuth angles within the range of ±30° and around ±0.4° for elevation angles within the range of ±13°. Extensive road testing has demonstrated that the designed radar is capable of reliably measuring dynamic targets such as vehicles, pedestrians, and bicycles, while also accurately detecting static infrastructure like overpasses and traffic signs.
BibTeX
@misc{cai2025development,
  author = {Yongjun Cai and Jie Bai and Hui‐Liang Shen and Libo Huang and B. Dharma Rao and Haiyang Wang},
  title = {{Development of Low-Cost Single-Chip Automotive 4D Millimeter-Wave Radar}},
  year = {2025},
  doi = {10.3390/s25154640},
}

Loading related content...

Figure