Long-Range Drone Detection of 24 G FMCW Radar with E-plane Sectoral Horn Array

Byung-Gil Choi, Daegun Oh, Sunwoo Kim, Jong-Wha Chong, Ying-chun Li2018 · DOI

TL;DR

This paper presents a novel 24-GHz FMCW radar system designed for long-range drone detection (up to 1 km). It introduces a 3D subspace-based algorithm for joint range, azimuth, and Doppler estimation, demonstrating its effectiveness in real-world outdoor experiments with quadcopter drones.

What Problem Does This Paper Address?

The increasing prevalence of civilian drones poses significant threats to public safety and national security, necessitating effective surveillance and detection systems. While various drone detection methods exist (audio, visual), radar offers advantages for long-range and all-weather capabilities. Existing radar solutions for drone detection often have limited effective ranges (tens to a few hundred meters) or only provide 1D or 2D parameter estimation (range, Doppler, or range-azimuth). There's a clear need for radar systems capable of detecting drones at longer distances (up to 1 km) and simultaneously estimating their range, azimuth, and velocity in a 3D space.

Methods

The core of this work lies in two main components: a custom-built 24-GHz FMCW radar system and a novel 3D subspace-based algorithm.

Radar System: * Transmitter: A 24-GHz FMCW radar system with a high-gain (25-dBi) lens horn antenna is used for transmission. A power amplifier (PA) circuit is employed to achieve a high output power (38 dBm). * Receiver: A two-element E-plane sectoral horn antenna array is used for reception. These antennas are spaced by one wavelength to achieve a specific field of view and provide high gain (14-dBi). * Transceiver: A 24-GHz transceiver module (Analog Devices EV-RADAR-MMIC2) is utilized, along with baseband adapter boards and a data-logging platform.

3D Subspace-Based Algorithm: * Signal Model: The received FMCW signals are modeled considering range, azimuth, and Doppler phase shifts. * 3D Auto-Correlation Matrix: A novel 3D auto-correlation matrix is constructed by leveraging the phase shifts between adjacent elements in the received signals across range, azimuth, and Doppler dimensions. This approach aims to reduce computational complexity compared to traditional methods. * Eigenvalue Decomposition (EVD): The 3D auto-correlation matrix is decomposed using EVD to separate signal and noise subspaces. * MDL Criterion: The Minimum Description Length (MDL) criterion is used to estimate the number of targets. * 3D Pseudo-Spectrum: A 3D pseudo-spectrum is calculated using the signal subspace. * Peak Detection: Peak detection on the pseudo-spectrum identifies the K targets, yielding their estimated range, azimuth, and Doppler (velocity). * Stationary Clutter Mitigation: A subspace projection approach is employed to mitigate undesired stationary clutter (e.g., reflections from walls) that can mask drone signals.

Datasets

The paper does not explicitly mention pre-existing datasets for training. Instead, it describes experiments conducted using two quadcopter drones (DJI manufactured) in outdoor environments. The raw radar data collected during these experiments serves as the dataset for evaluating the proposed algorithm.

Key Insights and Ideas

  • Long-Range Drone Detection is Feasible with 24-GHz FMCW: The paper demonstrates that a 24-GHz FMCW radar system, when properly designed with high-gain antennas and powerful transmitters, can achieve effective drone detection at ranges up to 1 km, significantly extending the capabilities of previous systems.
  • 3D Parameter Estimation is Crucial for Drone Surveillance: Moving beyond 1D or 2D estimations, the proposed 3D subspace-based algorithm's ability to jointly estimate range, azimuth, and Doppler provides a more comprehensive understanding of the drone's state, which is vital for effective tracking and identification.
  • Novel 3D Auto-Correlation Matrix Reduces Complexity: The proposed method's construction of a 3D auto-correlation matrix, which implicitly incorporates range, azimuth, and Doppler information, offers a more computationally efficient approach to joint parameter estimation compared to methods that might require separate processing steps.
  • Stationary Clutter Mitigation is Essential for Realistic Environments: The successful application of a subspace projection technique to mitigate stationary clutter highlights the practical challenges of real-world radar deployment and the importance of robust signal processing for reliable drone detection.

Key Results

  • Detection Range: The implemented radar system successfully detected quadcopter drones at ranges up to 1 km.
  • Parameter Estimation Accuracy: The algorithm provided estimations for range, azimuth angle, and radial speed. For example, in Experiment 1, a drone was detected at approximately 966 m with an azimuth angle of 13.5° and a radial speed of -0.7 m/s. In Experiment 2, a drone was detected at approximately 1005 m with an azimuth angle of 4.5° and a radial speed of 3.4 m/s.
  • Clutter Mitigation Effectiveness: Figure 16 visually demonstrates the significant improvement in detection results after applying the stationary clutter mitigation approach, where the drone signal becomes clearly visible.
  • Algorithm Performance: The proposed 3D subspace-based algorithm was verified experimentally and showed good performance in complex outdoor environments.

Limitations and Future Work

  • Hardware Implementation: The authors acknowledge that implementing the full algorithm on hardware (FPGA/DSP) for real-time processing remains a challenge due to the matrix operations and 3D spectrum searching involved. Currently, the processing is done offline using MATLAB on a high-performance PC.
  • Moving Platform: The current radar system assumes a stationary platform. Future work should focus on developing new signal models and algorithms for scenarios where the radar system is placed on a moving platform.
  • Wind Influence: The experiments showed that wind can affect drone flight stability, particularly at higher altitudes, which can influence the accuracy of the estimated flight paths.
  • Narrow Beam Width: At large azimuth angles, the detection results were influenced by the narrow beam width of the Tx antenna.

Summary updated

BibTeX
@misc{choi2018long,
  author = {Byung-Gil Choi and Daegun Oh and Sunwoo Kim and Jong-Wha Chong and Ying-chun Li},
  title = {{Long-Range Drone Detection of 24 G FMCW Radar with E-plane Sectoral Horn Array}},
  year = {2018},
  doi = {10.3390/s18124171},
}

Loading related content...

Figure