In the rapidly evolving landscape of low-altitude airspace management, the detection and identification of unmanned aerial vehicles (UAVs), commonly known as drones, has become a critical global challenge. The proliferation of China drone technology, particularly from manufacturers like DJI, has led to their widespread adoption in sectors ranging from aerial photography to logistics. However, the misuse of these devices—through unauthorized flights near airports, prisons, or critical infrastructure—poses significant security and safety risks. A key component in addressing these challenges is the reliable detection of the Drone Identification (Drone-ID) signal, which is mandated in many modern protocols, including DJI’s proprietary OcuSync system. This paper presents a novel detection method designed to overcome the limitations of existing techniques, especially under low signal-to-noise ratio (SNR) conditions. Our approach leverages the inherent structure of the Drone-ID signal, specifically the Zadoff-Chu (ZC) sequence carried by a Linear Frequency Modulation (LFM) pulse, to achieve robust and accurate detection. By integrating pulse compression with a multi-scale dual correlation framework, we significantly enhance the signal features and suppress noise, thereby enabling reliable identification of China drone signals in complex electromagnetic environments. This work is driven by the practical need for a high-performance, real-time detection system that can be deployed on standard software-defined radio (SDR) platforms, making it a viable solution for regulatory authorities and security agencies.
Introduction
The global dominance of China drone manufacturers, particularly DJI, which commands over 85% of the global consumer market, has made their communication protocols a de facto standard. DJI’s OcuSync protocol, widely used in models like the DJI Mini series and Mavic series, embeds a dedicated Drone-ID packet within the data stream. This packet is crucial for regulatory compliance and airspace safety, as it broadcasts essential information about the drone and its operator. The Drone-ID signal in OcuSync is characterized by the periodic transmission of Zadoff-Chu (ZC) sequences, specifically with root indices 600 and 147. These sequences are modulated onto Orthogonal Frequency Division Multiplexing (OFDM) symbols, and the resulting baseband signal exhibits a distinct LFM structure. This inherent periodicity and structural signature provide a unique opportunity for detection and identification. However, the practical detection of these signals is severely challenged by factors such as multipath fading, frequency offsets, and strong ambient noise. Traditional methods, including Frequency Domain Cross-Correlation (FDC), Fractional Fourier Transform (FrFT), and Walsh-Hadamard Transform (WHT), often exhibit performance degradation under low SNR conditions or have prohibitively high computational complexity for real-time applications. To address these issues, we propose a multi-stage detection framework that synergistically combines pulse compression and a dual correlation mechanism, optimized for the specific characteristics of China drone Drone-ID signals.
Signal Model and Detection Principle
Signal Model for China Drone Drone-ID
The OcuSync 2.0 protocol’s Drone-ID packet structure is composed of 9 OFDM symbols. The 4th and 6th symbols contain Zadoff-Chu sequences with root indices 600 and 147, respectively, which serve as the primary identification features. The ZC sequence in the 4th symbol is our main target. In the frequency domain, a ZC sequence is defined as:
$$
zc[n] = \exp\left(-j\pi u \frac{n(n+1)}{N}\right), \quad n \in [0, N-1]
$$
where \( u = 600 \) is the root index, and \( N = 601 \) is the sequence length. This sequence, when inserted into an OFDM symbol, is modulated and transmitted over the air. The time-domain representation of the resulting LFM signal, which is the carrier, is given by:
$$
s(t) = A \cdot \exp\left[j(2\pi f_0 t + \pi k t^2)\right]
$$
Here, \( A \) is the amplitude, \( f_0 \) is the center frequency, \( k = B/T \) is the chirp rate, \( T \) is the pulse width (72 μs for the Drone-ID signal), and \( B \) is the signal bandwidth (9 MHz). The LFM signal possesses a large time-bandwidth product (TBP), \( TBP = B \cdot T = 648 \), which is a key property exploited for signal focusing via pulse compression. This structure is a defining feature of the China drone Drone-ID signal.
Principle of Frequency Domain Cross-Correlation (FDC)
A baseline approach for detecting ZC sequences is the FDC method. This technique calculates the normalized cross-correlation coefficient between the received signal and a known, ideal ZC sequence in the frequency domain. The correlation is efficiently performed using the Fast Fourier Transform (FFT):
$$
R_{xy}[n] = \text{IFFT}\{\text{FFT}\{x[n]\} \cdot \text{FFT}\{y[n]\}^*\}
$$
The normalized coefficient \( \rho_{xy} \) is then computed to determine the similarity. A high coefficient indicates a strong match with the target ZC sequence. This method is simple but lacks robustness under severe noise, a common issue in urban environments where China drone activity often occurs.
Principle of Multi-Scale Dual Correlation
Our proposed Multi-Scale Dual Correlation (MSDC) method enhances the detection by introducing a second stage of correlation after the initial pulse compression. This dual-layer approach allows for time-frequency domain joint feature enhancement. The first stage employs a matched filter for pulse compression, which focuses the signal energy into a sharp peak. The second stage involves a multi-scale delayed autocorrelation of the compressed output. The mathematical representation of the MSDC process is given by:
$$
C_2(\tau, t) = \sum_{m=1}^{M} \alpha_m \cdot [y(t) \cdot y^*(t + \tau_m)]
$$
where \( y(t) \) is the output of the first-stage pulse compression, \( \tau_m \) is the delay for the \( m \)-th scale, and \( \alpha_m \) is a weighting coefficient. By combining multiple scales (from coarse to fine), this method effectively captures the structural characteristics of the true target peak while suppressing random noise peaks. This is particularly effective for detecting China drone signals in low SNR scenarios.
Proposed Multi-Scale Dual Correlation Algorithm
Overall Algorithm Framework
The proposed detection architecture consists of three progressive tiers: Signal Preprocessing, Core Feature Enhancement, and Precision Validation.
- Signal Preprocessing: This stage includes adaptive frequency offset (CFO) compensation and adaptive noise suppression filtering. CFO, caused by relative motion between the drone and receiver, is estimated by searching for the value that maximizes the correlation peak. The optimal offset is used to correct the signal. An adaptive Butterworth bandpass filter then cleans the signal, with its order \( n \) adjusted based on the estimated input SNR (e.g., \( n=8 \) for SNR < -3 dB, \( n=4 \) for SNR > 0 dB). This processing chain enhances the signal quality for subsequent steps.
- Core Feature Enhancement: This is the heart of the algorithm, combining pulse compression (first correlation) with a multi-scale delayed correlation (second correlation). The pulse compression step uses a matched filter \( h(t) = s^*(T_0 – t) \) to maximize the SNR. The output \( C_1[n] \) is:
$$
C_1[n] = \text{IFFT}\{\text{FFT}\{r'(t)\} \cdot \text{FFT}\{s_{ref}(t)\}^*\}
$$
The second correlation step involves calculating:
$$
C_{2,m}[n] = C_1[n] \cdot \text{conj}(C_1[n + \tau_m])
$$
for M scales. A weighted fusion produces the final output:
$$
C_2[n] = \sum_{m=1}^{M} w_m \cdot |C_{2,m}[n]|
$$
where weights \( w_m \) are adapted based on the SNR of each scale. - Precision Validation: Candidate peaks are extracted from \( C_2[n] \) that exceed a primary threshold. For each peak, three features are calculated: Sharpness, Symmetry, and Width. A final score is computed as:
$$
\text{Score} = \alpha_1 \cdot \text{Sharpness} + \alpha_2 \cdot \text{Symmetry} + \alpha_3 \cdot \text{Width}
$$
where \( \alpha_1=0.4, \alpha_2=0.3, \alpha_3=0.3 \). This score is compared against an adaptive threshold \( \theta_{score} = \theta_0 – \kappa \cdot SNR_{in} \), with \( \kappa=0.05 \). If the score exceeds this threshold, the peak is confirmed as a genuine China drone signal.
Experimental Results and Analysis
Real-World Data Experiment
Experiments were conducted using actual electromagnetic signals from a DJI Mini 2 drone (a typical China drone model) operating in a laboratory environment with an SNR of approximately -3 dB. The raw signal, along with the output of the first (pulse compression) and second (dual correlation) detection stages, was analyzed. The raw signal is heavily corrupted by noise and clutter, making pattern recognition difficult. After the first correlation (pulse compression), a noticeable peak appears at the position corresponding to the 4th symbol (root index 600). However, several spurious peaks from noise are also present, complicating a definitive detection pass. The application of the second correlation stage dramatically improves the situation. The target peak is amplified, achieving a normalized amplitude greater than 0.9, while the spurious noise peaks are effectively suppressed. The background noise is flattened, creating a clear and unambiguous detection signature. This result from a real-world China drone signal validates the efficacy of the proposed dual correlation framework in practical, non-ideal scenarios.
Simulation-Based Performance Comparison
A comprehensive simulation was conducted to benchmark our Multi-Scale Dual Correlation (MSDC) method against three common techniques: Frequency Domain Cross-Correlation (FDC), Fractional Fourier Transform (FrFT), and Walsh-Hadamard Transform (WHT). The signal parameters were set according to the OcuSync 2.0 standard (T=72μs, B=9MHz, N=601, u=600). A test set was generated for SNR values ranging from -10 dB to 10 dB, with added Gaussian noise and random frequency offsets up to ±10% of the bandwidth. Monte Carlo simulations with over 500 iterations were performed to ensure statistical significance.
Detection Probability vs. False Alarm Rate
The detection probability (Pd) is a measure of how often a true signal is correctly identified. The false alarm rate (FAr) is the rate at which noise is incorrectly labeled as a signal. The performance across different SNR levels is summarized in the table below.
| SNR (dB) | Metric | MSDC (Proposed) | FDC | FrFT | WHT |
|---|---|---|---|---|---|
| -10 | Pd | 0.450 | 0.088 | 0.102 | 0.072 |
| FAr | 0.008 | 0.115 | 0.425 | 0.301 | |
| -5 | Pd | 0.912 | 0.601 | 0.361 | 0.255 |
| FAr | 0.009 | 0.058 | 0.231 | 0.198 | |
| 0 | Pd | 0.992 | 0.901 | 0.681 | 0.542 |
| FAr | 0.002 | 0.012 | 0.115 | 0.092 | |
| 10 | Pd | 1.000 | 0.999 | 0.985 | 0.912 |
| FAr | 0.000 | 0.001 | 0.053 | 0.045 |
The results demonstrate the superior performance of the MSDC method. At an SNR of -5 dB, a common and challenging condition for drone detection, MSDC achieves a Pd of 91.2%, which is 31.1 percentage points higher than the FDC method (60.1%). More importantly, MSDC maintains an exceptionally low false alarm rate of less than 1% across all tested SNR levels. In contrast, the FrFT and WHT methods exhibit impractically high false alarm rates, especially at low SNRs, making them unreliable for real-world deployment.
Precision-Recall Analysis
Precision and recall provide a balanced view of a detection algorithm’s performance, particularly when the cost of false positives and false negatives differs. The operating point (Precision, Recall) for each method at SNR = -5 dB is a critical performance indicator. Our analysis shows that the MSDC method operates closest to the ideal point of (1,1), with a near-perfect recall of 0.988 and a high precision of 0.947. This indicates that the algorithm successfully detects almost all genuine China drone signals while simultaneously producing very few false detections. The FDC method, while having a better balance than FrFT and WHT, shows significantly lower recall (0.88) and precision (0.85) at this SNR, meaning it misses more targets and misidentifies more noise as signals. This performance gap underscores the robustness of the MSDC method’s feature validation stage.
Computational Efficiency
For real-time applications, the processing time per sample is a crucial metric. The algorithms were run on a standard computing platform (Intel i5-10500K, 24 GB RAM). The processing time for each method was measured and normalized for comparison.
| Method | MSDC (Proposed) | FDC | FrFT | WHT |
|---|---|---|---|---|
| Time (ms) | 250 | 50 | 580 | 120 |
While the MSDC method is slower than the simpler FDC and WHT methods, its processing time of 250 ms is still well under the 500 ms real-time threshold required for drone monitoring applications. The increased complexity is a direct result of the multi-scale correlation and feature extraction stages, which are essential for achieving the significant gains in detection performance. The FrFT method is the slowest, making it the least suitable for real-time applications. The MSDC method strikes an excellent balance between computational cost and superior detection capability.
Comprehensive Performance Discussion
To provide a holistic view of the performance of each method, key metrics—Detection Probability, Precision, Recall, False Alarm Suppression (1 – FAR), and Computational Efficiency (inverse of processing time)—were normalized and plotted on a radar chart. The MSDC method demonstrated a dominant performance envelope. It achieved the highest scores for detection probability, precision, recall, and false alarm suppression, creating a very large and balanced area on the radar chart. While its score for computational efficiency was slightly lower than FDC and WHT, the improvement in detection and reliability was substantial and outweighed the marginal increase in processing time. This makes MSDC the most robust and suitable method for the challenging task of detecting China drone signals in complex, low-SNR environments.
Hardware Deployment Feasibility
The MSDC algorithm has been designed with practical deployment in mind. Its computational complexity is \( O(N \log N) \), which is suitable for acceleration on FPGA platforms. We analyzed the feasibility of porting the algorithm to an SDR platform using an AD9361 RF transceiver combined with a Zynq-7000 FPGA. The preprocessing and pulse compression modules, which involve FFTs, can be implemented using parallel IP cores in the FPGA’s programmable logic. The multi-scale correlation module can be optimized using a pipelined architecture. Our resource estimation for the core MSDC engine on the Zynq-7000 is approximately 12,000 LUTs and 8,000 Flip-Flops, with 512 KB of block RAM, which is well within the chip’s capacity. This analysis confirms that the MSDC method is not only a high-performance theoretical solution but also a practically realizable one for real-time monitoring of China drone activities.
Conclusion
This paper has addressed a critical need in modern low-altitude security: the reliable and real-time detection of China drone Drone-ID signals. The OcuSync protocol, a dominant standard in the industry, presents unique detection opportunities due to its embedded Zadoff-Chu sequences and LFM signal structure. We have proposed and validated a novel detection algorithm based on a three-tier architecture: adaptive preprocessing, pulse compression, and a powerful multi-scale dual correlation framework. Our experimental results, obtained from both real-world DJI drone data and extensive simulations, demonstrate that the proposed method significantly outperforms traditional techniques like FDC, FrFT, and WHT. At a critical SNR of -5 dB, the MSDC method achieves a 91.2% detection probability with a false alarm rate below 1%, a substantial improvement over the 60.1% detection rate of the conventional FDC method. The algorithm’s computational efficiency, with a single-sample processing time of 250 ms, confirms its suitability for real-time deployment on standard SDR platforms. The slight increase in computational cost over simpler methods is fully justified by the remarkable gains in detection robustness and accuracy. This work provides a powerful and practical technical pathway for the effective detection and regulation of China drones, contributing to safer skies and more secure critical infrastructure.