This paper presents a six-element stereo array system for detecting and positioning low-altitude drones using acoustic-seismic signals. The system combines Mel-frequency cepstral coefficients (MFCC) with Support Vector Machines (SVM) for target identification and employs Time Difference of Arrival (TDOA) for 3D spatial localization. Field tests demonstrate 88.89% detection accuracy, validating its effectiveness in battlefield situational awareness applications for low altitude drone monitoring.
Introduction
Battlefield situational awareness critically depends on detecting mobile targets, especially in dynamic environments. Current detection methods include optical, infrared, magnetic, radar, acoustic, and seismic technologies. While radar offers long-range capabilities, its high power consumption, cost, and inability to penetrate obstacles limit effectiveness. Acoustic-seismic detection provides significant advantages: low cost, minimal power requirements, and strong obstacle penetration. Although its range is shorter than radar, it exceeds infrared and magnetic methods, making it ideal for low altitude UAV detection.
Principle of Target Detection
Acoustic-Seismic Signal Characteristics
Acoustic and seismic signals from low altitude UAVs exhibit similar time-frequency characteristics. Field measurements of DJI M300 drones show dominant spectral energy at 1kHz. Environmental noise sources include:
| Noise Source | Frequency Range |
|---|---|
| Wind (≤6m/s) | ≤200Hz |
| Animal sounds | 100-1000Hz |
| Rainfall | ≤20Hz (LF), ≥1kHz (HF) |
The distinct spectral separation between low altitude UAV signatures and environmental noise enables reliable feature extraction.
Detection Methodology
The MFCC-SVM processing pipeline:
- Pre-emphasis: Compensates high-frequency attenuation
$$H(z) = 1 – az^{-1}, \quad a \in (0.9,1.0)$$ - Framing: Segments signals into 4s windows (10kHz sampling)
- Mel transformation: Converts to perceptually relevant scale
$$\text{mel}(f) = 2595 \times \log_{10}\left(1 + \frac{f}{500}\right)$$ - SVM classification: Uses radial basis function kernel for nonlinear separation
Localization Principle
The six-element acoustic array enables 3D localization of low altitude UAVs under far-field conditions (\( r \geq 2d^2/\lambda_{\text{min}} \)):
| Parameter | Calculation |
|---|---|
| Azimuth (θ) | $$\theta = \arccos\left(\frac{S_v \times \tau_{ij}}{d_{ij}}\right) – \beta_{ij}$$ |
| Elevation (φ) | $$\phi = \arccos\left(\frac{S_v \times \tau_{16}}{d_{16}}\right)$$ |
Sensor spacing optimization balances localization accuracy and signal quality. For typical low altitude UAV frequencies (500Hz-10kHz), optimal spacing ranges 1.7-34.6cm.
System Design and Simulation
The hexagonal array configuration with coordinates:
| Sensor | Position |
|---|---|
| Mic1 | (0, d, 0) |
| Mic2 | (d, 0, 0) |
| Mic3 | (0, -d, 0) |
| Mic4 | (-d, 0, 0) |
| Mic5 | (0, 0, 0) |
| Mic6 | (0, 0, d) |
Simulation results for different spacings (360 sources at 100m radius):
| Spacing | Azimuth Error | Elevation Error |
|---|---|---|
| 0.05m | 4.9% | 4.8% |
| 0.1m | 1.5% | 2.4% |
| 0.2m | 0.4% | 0.08% |
Optimal spacing of 0.1m balances accuracy and signal-to-noise ratio for practical low altitude UAV detection.
Experimental Validation
Field tests with DJI M300 drones approaching from multiple axes:
| Approach Direction | Azimuth Error | Elevation Error |
|---|---|---|
| Y-axis positive | ≤5° | Within theoretical range |
| X-axis positive | ≤2° | Within theoretical range |
| Y-axis negative | ≤8° | Within theoretical range |
Confusion matrix for low altitude UAV detection:
| Actual/Predicted | UAV | Noise |
|---|---|---|
| UAV | 409s | 63s |
| Noise | 45s | 455s |
Accuracy calculation:
$$\text{Accuracy} = \frac{409 + 455}{409 + 63 + 45 + 455} \times 100\% = 88.89\%$$
Conclusion
This system effectively detects and localizes low altitude UAVs using acoustic-seismic signatures. Key achievements:
- 88.89% detection accuracy in field tests
- Azimuth errors ≤8° across approach vectors
- Real-time processing capability
The integration of MFCC feature extraction with SVM classification and TDOA localization provides a robust solution for monitoring low altitude drones in battlefield environments. Future work will optimize elevation accuracy through multi-plane sensor configurations.