Real-Time detection, classification and DOA estimation of Unmanned Aerial Vehicle

The present work deals with a new passive system for real-time detection, classification and direction of arrival estimator of Unmanned Aerial Vehicles (UAVs). The proposed system composed of a very low cost hardware components, comprises two different arrays of three or six-microphones, non-linear amplification and filtering of the analog acoustic signal, avoiding also the saturation effect in case where the UAV is located nearby to the microphones. Advance array processing methods are used to detect and locate the wide-band sources in the near and far-field including array calibration and energy based beamforming techniques. Moreover, oversampling techniques are adopted to increase the acquired signals accuracy and to also decrease the quantization noise. The classifier is based on the nearest neighbor rule of a normalized Power Spectral Density, the acoustic signature of the UAV spectrum in short periods of time. The low-cost, low-power and high efficiency embedded processor STM32F405RG is used for system implementation. Preliminary experimental results have shown the effectiveness of the proposed approach.

💡 Research Summary

The paper presents a complete low‑cost, battery‑powered acoustic detection, classification, and direction‑of‑arrival (DOA) estimation system for unmanned aerial vehicles (UAVs). The hardware consists of either a three‑microphone or a six‑microphone omnidirectional array built from inexpensive commercial microphones (≈ 2.5 € each) and a non‑linear analog front‑end based on four cascaded MCP604 operational amplifiers. The analog chain provides a high‑pass filter at 80 Hz, logarithmic amplification to prevent saturation when a UAV flies close to the sensors, and a final low‑gain linear stage that scales the signal to a 0‑3 V range suitable for the on‑chip analog‑to‑digital converters (ADCs) of an STM32F405RG Cortex‑M4 microcontroller.

Each microphone is sampled by a 12‑bit SAR ADC running at 21 MHz. By applying software‑controlled oversampling (3×) and digital low‑pass filtering, the effective resolution is increased by 6 bits, yielding an 18‑bit equivalent dynamic range while the final sampling rate is 21.875 kHz. The microcontroller’s DMA engine streams the data to RAM, and an interrupt fires every 45 µs, allowing the CPU to process three (or six) streams with minimal load.

Signal processing is performed in real time. An energy‑based beamforming algorithm computes the signal energy over 200 ms windows; with three microphones this yields a DOA estimate at 5 Hz update rate, while with six microphones pairwise cross‑correlation provides finer angular resolution. To improve robustness against background noise, a weighted FFT is used: H(ω)=Pss(ω)/(Pss(ω)+Pnn(ω)), where Pss and Pnn are the power spectral densities of signal and noise, respectively.

For classification, the normalized power spectral density (PSD) of each UAV is stored as a 32‑dimensional feature vector in flash memory. A nearest‑neighbor (NN) classifier using a weighted Euclidean distance compares the current PSD to the stored templates. During training, mean and standard deviation for each frequency bin are computed to set decision thresholds. If the same UAV matches the template for two consecutive seconds with a distance below the threshold, the system declares a detection and reports the UAV type.

The system consumes about 130 mA at 3.6 V (≈0.47 W), making it suitable for portable battery operation. The total component cost is under 30 €, including the STM32 development board (≈ 15 €), microphones, and op‑amps. Experimental validation involved three different quadcopters (DJI P3, CX‑10, and a Sennheiser MKH‑8040 test rig). Power spectral density measurements up to 15 kHz showed distinct signatures for each platform, confirming that the PSD‑based NN classifier can differentiate UAVs even at low signal‑to‑noise ratios (≈ 0 dB). Real‑time tests in indoor and outdoor settings achieved detection ranges of 150–250 m, with 100 % detection rate in noise‑free conditions and reliable DOA estimation.

The authors conclude that the proposed architecture successfully integrates low‑cost hardware, efficient oversampling, energy‑based beamforming, and simple PSD‑NN classification to deliver a practical UAV acoustic monitoring solution. Future work will address more challenging acoustic environments (urban noise, multiple simultaneous UAVs) by incorporating multi‑dimensional scaling (MDS) for array calibration and exploring deep‑learning classifiers to improve robustness and scalability.