Communication Frame Analysis to Differentiate Between Authorized and Unauthorized Drones of the Same Model

Highlights

What are the main fundings?

• RF analysis and characterization of drones: Analysis and characterization of drone communication signals based on physical layer data.
• Frame linking models: Development of an algorithm able to assign a frame to its drone based on data from the physical layer.

What are the implications of the main findings?

• Drone identification: Identify the communication frame of multiple drones in flight.
• Frame linking: Connect each drone communication physical layer signal to a registered drone.

Abstract

Unmanned aerial vehicle (UAV) applications are growing fast in different sectors, such as agricultural, commercial, academic, leisure, and health fields. However, drones pose a significant threat to public safety due to their ability to transmit information, particularly when used in an unauthorized or malicious manner. In fact, in order to protect citizens’ privacy and prevent accidents in high-traffic areas due to poorly controlled flights, no-fly zones for drones have been established in the legislation of a number of countries. Most common UAV detection techniques are based on radio frequencies, which identify drones and their models by monitoring radio frequency signals. However, differentiating between multiple UAVs of the same model is their main limitation. This article fills this gap by proposing a method for physically tracking the communication frames of a registered UAV in the presence of another UAV of the same model. A measurement campaign was conducted to collect real-world RF communication signals from two DJI MAVIC 2 Zoom, two DJI Air2S, and two DJI Phantom drones. This measurement was performed inside and outside an anechoic chamber in order to study the UAV’s communication without any interference and in the presence of other communications. Through detailed statistical analysis, we characterized features such as communication duration, time intervals between communications, signal strength, and patterns in communication timing sequences. Our analysis revealed unique, identifiable patterns for each UAV, even within identical models. Based on these results, we developed an automated system that links communication frames to the corresponding registered drones. The proposed method fills gaps in drone detection and surveillance models, providing valuable information for applications in the fields of security and airspace management. This research lays the foundation for drone identification solutions, thereby addressing a major limitation of current detection technologies.

1. Introduction

Unmanned aerial vehicles (UAVs), commonly known as drones, are rapidly becoming integral to modern technology ecosystems, finding applications in a diverse range of fields. These versatile devices leverage state-of-the-art technologies to perform tasks ranging from delivering medical supplies in remote areas to supporting disaster response efforts and enabling cutting-edge commercial applications such as aerial photography, surveying, and logistics [1,2,3]. The growing accessibility and decreasing costs of UAVs have democratized their usage, transforming them into indispensable tools for both professionals and hobbyists. However, this widespread adoption has also introduced significant challenges, particularly in the domain of public safety and security.

The ease of access to UAVs has inadvertently opened the door to their exploitation for malicious activities. Drones have been increasingly employed in drug trafficking, smuggling contraband across borders, illegal surveillance, and even in orchestrating bomb attacks [4,5,6]. These incidents underscore the pressing need for robust detection and mitigation strategies to address the security threats posed by unauthorized UAV operations. Developing effective detection systems is critical for safeguarding sensitive areas such as airports, government buildings, and public gatherings.

UAVs operate with varying levels of autonomy and can function in one of two primary modes: pre-programmed flights using Global Navigation Satellite System (GNSS) signals or manual control via remote operators [7]. These operational modes introduce a range of opportunities for detection and tracking, as UAVs often emit identifiable signatures. Various technologies, including RF signal analysis [8], imaging [9], radar sensors [10], and acoustic methods [11], have been explored to detect and track UAVs [12]. Among these, RF-based detection stands out for its potential to identify remotely controlled UAVs and those that transmit signals, such as live video feeds or positional data [13]. Furthermore, in contrast to vision- or acoustic-based approaches, RF-based analysis offers the advantage of continuous operation under varying environmental conditions, enabling reliable detection both during the day and at night.

Despite extensive research on UAV detection, most studies have focused on commercially available UAVs operating with proprietary protocols rather than standardized communication systems [14,15]. These efforts have successfully identified distinct communication protocols, including Standard Wi-Fi, Enhanced Wi-Fi, LightBridge, and Ocusync, allowing researchers to classify and pinpoint specific UAV models. However, a significant gap remains unaddressed: the differentiation of multiple UAVs of the same model. This capability is critical in scenarios where authorized drones operate alongside potentially threatening ones of identical make and model [16].

For instance, in securing sensitive environments such as public events, residential area, industrial sites, or highly frequented places, distinguishing between authorized drones and potential threats is essential. A failure to accurately identify and differentiate UAVs of the same model could result in delayed responses or inappropriate countermeasures, jeopardizing safety [17]. Therefore, tracking the communication frames of UAVs and accurately attributing frames to potentially unauthorized UAVs is a vital step in enhancing UAV detection systems.

Currently, to adress this challenge, machine learning approaches using RF data can detect drones and identify the brand and model of the drone and, therefore, its communication protocol. However, these approaches are limited when the measurement environment is polluted by other communications, as is the case in urban environments. To overcome this weakness in detection, algorithms such as You Only Look Once (YOLO) [18] make it possible to identify not only the communication frames belonging to a drone but also the drone model from an image representing radio signals in time-frequency domains. That being said, when several drones are communicating simultaneously, the latter are unable to differentiate whether a communication frame belongs to drone number 1 or drone number 2 if the drones are of the same models. This is important when, in an area where drone flying is prohibited, a drone has flight authorization but you want to be sure that no other drones are present. To meet these needs, some projects use RF fingerprinting [19] to identify drones. These approaches are interesting because they enable the detection and identification of drones based on their unique RF signature. However, it does have some limitations, such as classification accuracy, which can be mitigated by deep learning approaches, or overlap with other RF signals communicating in the same frequency band, which is often the case in cities.

To address this challenge, in addition to RF detection and classification approaches, we propose a study focusing on the RF signal characteristics of two UAVs of the same model. This research aims to explore their communication patterns, identify distinctive features, and develop a framework to trace communication frames effectively. The methodology involves building a comprehensive database through a controlled experimental setup, preprocessing the collected data, and conducting a detailed statistical analysis of the communication characteristics [20]. The insights gained from this analysis are used to design a robust system capable of identifying and tracking communication frames, providing a foundation for improved UAV detection and differentiation. The overall methodological process is illustrated in Figure 1.

This paper is structured as follows to elaborate on the proposed identification process. Section 2 details the experimental setup, including data collection and preprocessing methods. Section 3 presents a statistical analysis of the communication characteristics of the two UAVs. Section 4 describes the development of a RF tracking system based on the distinctive frame characteristics identified. Finally, Section 5 discusses the implications of our findings and outlines potential applications and future research directions.

2. Experimentation Configuration

This study established experimental setups comprising various equipment to capture, analyze, and characterize radio frequency (RF) signals emitted by UAVs. The components of this setup include a Software-Defined Radio (SDR) for signal acquisition designed by a French company named Inodesign group, a laptop to control the SDR and perform data processing using an Iron Python in a version 2.7, and an antenna for signal reception. Figure 2 shows the configuration in which measurements are acquired. The experiments were conducted under controlled and realistic conditions, providing a robust dataset for analysis.

The rest of this section describes the used equipment, two test benches (one inside an anechoic chamber and one outside), and the two operation modes (flight mode and stationary mode) employed during the study.

2.1. Equipment and Setup

2.1.1. Unmanned Aerial Vehicles (UAVs)

For this study, we selected commercially available UAVs to ensure relevance to real-world scenarios. Specifically, we used two DJI MAVIC 2 Zoom UAVs, two DJI AIR 2S UAVs, a DJI Phantom 4 UAV, and a DJI Phantom 3 UAV. This choice was guided by the need to explore the signal characteristics and communication protocols of both modern and earlier-generation UAV models that have been widely adopted in commercial and recreational applications.

The DJI MAVIC 2 Zoom features a communication system that employs proprietary protocols, notably OcuSync 2.0, which operates within the 2.4 GHz and 5.8 GHz frequency bands. Similarly, the DJI AIR 2S utilizes the more advanced OcuSync 3.0 communication protocol. This system operates within similar frequency bands. The DJI Phantom 3 Professional, on the other hand, employs the Lightbridge communication protocol, which operates primarily within the 2.4 GHz band; see Figure 3. These characteristics make these UAVs suitable for studying RF signal behaviors under controlled and variable conditions.

By focusing on models with proprietary communication protocols, the study directly addresses the challenges of detecting and identifying UAVs in real-world scenarios where such protocols dominate.

2.1.2. Software-Defined Radio (SDR)

The SDR utilized in this study is based on the ADRV9026 card, a versatile platform designed for advanced signal processing applications. The SDR was configured to monitor a 100 MHz frequency band centered at $2.45$ GHz, aligning with the UAV communication frequencies under study. The sampling rate was set to 122.88 MHz, and the signal acquisition was performed over a 40 ms time window. This configuration enabled the measurement of 4,915,198 frequency points within the specified band, providing high-resolution spectral data for analysis. During data acquisition, the system could acquire matrices of 4,915,198 × 2 per s, accounting for the time required for acquisition and storage. These matrices represent the in-phase (I) and quadrature-phase (Q) components of the captured signals, forming the basis for subsequent signal processing and analysis. This configuration ensured that the experimental setup could capture the high volume and quality of data necessary to characterize UAV communication signals and develop detection algorithms.

2.2. Measurement Benchs

2.2.1. Anechoic Chamber Bench

Conducting RF measurements of UAVs in an anechoic chamber is crucial for ensuring precise and reliable characterization of their communication systems under controlled and repeatable conditions. An anechoic chamber provides an environment that is free from external electromagnetic interference and unwanted reflections, simulating an idealized free-space scenario. This isolation is particularly important when evaluating the performance of radio frequency (RF) systems, as it allows researchers to measure signal strength, antenna patterns, and communication channel characteristics without the distortions that would typically occur in a real-world environment. By eliminating variables such as multipath propagation and environmental noise, the chamber enables accurate analysis of the UAV’s transmit and receive capabilities, modulation schemes, and power levels. Furthermore, a controlled environment is indispensable for calibrating communication systems, identifying potential hardware flaws, and verifying compliance with regulatory standards. It also facilitates detailed studies of how RF systems perform under various simulated scenarios, such as varying distances or angles between the UAV and ground stations, which can be systematically controlled within the chamber. These measurements provide valuable baseline data that can later be compared with tests conducted in outdoor or operational environments to account for real-world complexities. Ultimately, RF measurements in an anechoic chamber form the foundation for optimizing UAV communication systems, improving signal reliability, and ensuring robust performance in both controlled and unpredictable conditions. Figure 4 shows the measurement configuration in an anechoic chamber.

2.2.2. Outside Bench

After completing RF measurements in an anechoic chamber, conducting additional tests of UAVs in outdoor environments is a critical step to validate and refine their performance under real-world operating conditions. While anechoic chambers provide an idealized setting for isolating and evaluating RF parameters, the complexities of the real world, such as multipath propagation, terrain-induced reflections, atmospheric variations, and interference from other RF sources, cannot be fully replicated within such controlled spaces. Outdoor RF measurements enable researchers to assess the UAV’s ability to maintain reliable communication and navigation in these variable conditions, which are essential for its intended operational scenarios. These tests are especially important for UAVs operating in urban areas, mountainous terrains, or densely forested environments, where signal reflections and obstructions can cause significant challenges. By measuring RF performance outdoors, researchers can evaluate the effectiveness of system configurations established in the chamber, ensuring that they are robust enough to handle diverse environmental factors.

Outdoor testing also provides valuable insights into range and coverage. For instance, UAVs often require long-range communication capabilities for remote operations, which can be affected by weather conditions, interference from other communication systems, or signal attenuation over distance. Testing in an open environment helps to determine whether the UAV’s antennas and RF systems can sustain connectivity over the required ranges and under varying orientations. Additionally, outdoor measurements help identify unforeseen issues, such as signal degradation caused by physical components of the UAV, like motors or payloads, interacting with external RF fields. They also allow for testing in compliance with regulatory frameworks that govern the use of specific frequency bands in actual environments, ensuring that the UAV does not inadvertently interfere with other devices or violate spectrum usage rules. By conducting outdoor RF tests, researchers bridge the gap between theoretical analysis and practical application, ensuring the UAV’s systems are not only optimized but also reliable and resilient for field deployment. This final stage of testing confirms the UAV’s readiness to perform its mission safely and effectively in real-world conditions. Figure 5 shows the measurement configuration outside an anechoic chamber.

2.3. Operation Modes

RF signal measurements for UAVs in flight mode and stationary mode are critical because the behavior of control and video signals can differ significantly between these two operational modes.

2.3.1. Stationary Mode

In stationary mode, the UAV remains physically static with its propellers turned off and all flight-related control systems inactive. This mode is typically used during initialization, diagnostics, or system checks, where the drone does not attempt to stabilize itself or respond to directional inputs. In this configuration, the communication between the UAV and the remote controller is limited to low-level status updates and basic telemetry signals, typically transmitted at regular intervals with predictable durations. Since the UAV’s flight control and stabilization systems are not engaged, the transmitted signals in this mode are less dynamic and generally less sensitive to real-time command variations. The transmission power requirements are minimal, as the UAV is not actively adjusting its control surfaces or motor speeds. Furthermore, video transmission, if active, tends to be stable with low latency, since the RF environment is stable and the UAV does not introduce mechanical vibrations or dynamic propagation effects.

2.3.2. Flight Mode

In flight mode, although the UAV remains physically stationary within the laboratory setup, since the UAV propellers are not attached to the drone, it is possible to send flight commands to the UAV while leaving it in the same position. The UAV directional, stabilization, and rotational functionalities are engaged, simulating a real-flight scenario where the UAV continuously responds to control commands as if it was navigating in the air. The activation of the engines introduces mechanical vibrations and electromagnetic noise, which can influence both the RF environment and the UAV internal systems. Control signals in this mode become more dynamic, as the UAV actively attempts to stabilize itself and interprets directional inputs from the remote controller in real time. These signals may vary in frequency and duration as the system compensates for simulated flight conditions. The communication link must therefore handle more frequent control updates and potentially higher command throughput. Additionally, the power consumption typically increases as the UAV motors and control systems are operational. UAV transmission in flight mode may face challenges such as increased mechanically induced RF noise and minor fluctuations in signal strength, necessitating more robust transmission strategies to maintain video quality and minimize latency.

By conducting RF measurements in both stationary and flight modes within the controlled laboratory environment, we can systematically characterize the impact of the UAV operational state on communication performance. This dual-mode analysis provides valuable insights into the differences in signal behavior, interference susceptibility, and power requirements, even when the UAV is not physically moving. Understanding these variations is critical for optimizing UAV communication systems, ensuring resilience and reliability across different operational scenarios, and developing more effective counter-UAV strategies that account for both passive and active system states.

3. Data Base

In order to build this study database, we obtained 40 ms recordings for the six drones studied. During these 40 ms of recording, seven downlink communication frames corresponding to signals from the drones to the radio control and eight uplink communication frames corresponding to frames from the radio control for the DJI MAVIC Zoom and DJI Air2S drones were measured and three downlink communication frames and eight uplink communication frames for the DJI Phantom. For each drone and for each bench, 50 recordings were made, to which recordings of two drones of the same model operating simultaneously in stationary mode and flight mode will be added. The number of recording and the number of frame studied are presented in

Table 1. Quantities of data in terms of number of records, number n of UAV frames and number of RC frames for the experiment in anechoic chamber.

ID	Mode	n of Records	n of UAV Frames	n of RC Frames
UAV 1	stationary mode	50	350	396
	flying mode	50	350	397
UAV 2	stationary mode	50	350	399
	flying mode	50	350	400
UAV 3	stationary mode	50	350	399
	flying mode	50	350	397
UAV 4	stationary mode	50	350	399
	flying mode	50	350	395
UAV 5	stationary mode	50	150	150
	flying mode	50	148	150
UAV 6	stationary mode	50	146	150
	flying mode	50	147	150
UAV 1 & 2	stationary mode	50	700	792
	flying mode	50	700	789
UAV 3 & 4	stationary mode	50	700	795
	flying mode	50	700	790
UAV 5 & 6	stationary mode	50	292	300
	flying mode	50	291	300

and

Table 2. Quantities of data in terms of number of records, number n of UAV frames and number of RC frames for the outside experiment.

ID	Mode	n of Records	n of UAV Frames	n of RC Frames
UAV 1	stationary mode	50	350	392
	flying mode	50	350	399
UAV 2	stationary mode	50	350	398
	flying mode	50	350	397
UAV 3	stationary mode	50	350	400
	flying mode	50	350	399
UAV 4	stationary mode	50	350	399
	flying mode	50	350	397
UAV 5	stationary mode	50	147	150
	flying mode	50	149	150
UAV 6	stationary mode	50	148	150
	flying mode	50	149	150
UAV 1 & 2	stationary mode	50	700	795
	flying mode	50	700	792
UAV 3 & 4	stationary mode	50	700	794
	flying mode	50	700	796
UAV 5 & 6	stationary mode	50	294	300
	flying mode	50	295	300

.

4. RF Signal Preprocessing

4.1. Correlation

Correlation techniques are powerful tools that can be employed to effectively separate RC (radio control) signals and UAV signals embedded within a composite signal from IQ data normalized over a 40 ms time window. This separation is achieved by leveraging their inherent ability to identify and isolate distinct patterns present in different components of a signal. The composite signal in question typically contains both RC signals, used for controlling devices, and UAV signals, which are responsible for transmitting visual information. Each of these signal types exhibits unique characteristics in terms of their frequency spectra and modulation patterns, which serve as their distinct signatures. The process begins with the application of a correlation technique that involves comparing the composite signal against predefined templates or reference signals that represent the known characteristics of RC and UAV signals recorded in an anechoic chamber. This method, known as cross-correlation, measures the similarity between the composite signal and the reference signals at various points in time. The result of this process is a correlation curve, where peaks correspond to moments of high similarity that correspond to synchronization times. These peaks provide crucial information about the presence and timing of the individual signals within the composite. In essence, correlation techniques exploit the unique modulation and frequency characteristics of RC and video signals to distinguish them within a composite signal. The ability to cross-correlate with predefined reference signals ensures accurate identification and extraction, facilitating effective signal separation and processing. This approach is particularly advantageous in applications requiring precise signal isolation, such as in remote control systems or multimedia transmission. By applying this cross-correlation, we want to differentiate between signals corresponding to radio control and those corresponding to UAVs to extract their characteristics.

4.2. Reference Signal

We want to create a reference from the IQ files that will correspond to the RC signals and an IQ file that will correspond to the UAV signals. First, in order to identify the start and end of communication times, we determine a threshold value that will ensure that noise is not taken into account. Here, we have set the threshold value to 5% of the signal variability over a 40 ms window. This corresponds to a value of 0.05 for the IQ data normalized over this time window. We know that RC signal frames are short and transmitted between longer UAV signal frames. Once the times corresponding to the drone signals and RC signals have been identified in the file, we replace the values of the times corresponding to the UAV signals with 0 in order to create our reference file for the radio control, and we replace the times corresponding to the RC signals with 0 in order to create the reference file for the UAV signals.

4.3. RC Signal Extraction

The RC signal, for instance, often employs a specific modulation scheme designed for device control. By cross-correlating the composite signal with a reference template tailored to the RC signal’s characteristics, one can pinpoint its exact occurrence in the time domain. The distinctive peaks in the correlation output allow for the precise extraction of the RC signal. Once identified, this component can be isolated and separated from the composite signal for further processing or analysis.

4.4. UAV Signal Extraction

Similarly, the UAV signal, which typically occupies a different frequency range and employs a modulation pattern unique to its purpose, can also be identified through its correlation signature. The reference template for the video signal is designed to match its specific characteristics, ensuring that the cross-correlation process highlights the points in the composite signal where the UAV signal is present. By focusing on these correlation peaks, the video signal can be effectively separated from the composite.

4.5. Feature Extraction

To systematically characterize and compare the two UAVs, we construct a comprehensive database that incorporates several critical parameters. One of these parameters is the average power of control communications $\left(A_i\right)$, measured in decibels (dB), which reflects the average power level of signals sent to control the UAV. Similarly, the average power of communications from the UAV $\left(a_i\right)$, also measured in dB, indicates the average power of the signals transmitted by the UAV itself. In addition to power measurements, the database includes temporal metrics such as the time intervals between control communications $\left({\Delta }_i\right)$ and the time intervals between communications from the UAV $\left({\delta }_i\right)$, both represented in milliseconds (ms). These metrics capture the temporal gaps between consecutive transmissions in each communication type, providing insights into their regularity and frequency.

Furthermore, the database accounts for the duration of control communications $\left(T_i\right)$ and the duration of communications from the UAV $\left(t_i\right)$, both expressed in ms. These values indicate the length of individual transmissions, offering an additional layer of understanding about the communication dynamics.

The relationships between these parameters, specifically the duration and time interval indicators, are visually illustrated in Figure 6. This figure provides a clear representation of the timing and duration characteristics, serving as a foundation for the detailed analysis and comparison of the two UAVs. By incorporating these diverse metrics, the database enables a holistic understanding of communication behaviors and operational differences between UAVs.

5. Results and Discussion

5.1. Characterization of Drone RF Signal

For this study, our objective was to thoroughly characterize the communication behavior between the remote control and six UAVs, categorized into three sets: two MAVIC 2 Zoom (UAV 1 and UAV 2), two AIR 2S (UAV 3 and UAV 4), one Phantom 4, and one Phantom 3 Pro (UAV 5 and UAV 6). The study focused on analyzing the characteristics of video and control communications independently for each set of UAVs. A comparative analysis was also performed for these UAVs in two distinct operational modes: stationary mode, where the controller does not issue flight commands, and flying mode, where active motion commands are transmitted. These modes represent typical scenarios for UAV operation, allowing us to assess variations in communication behavior under static and dynamic conditions.

To quantify communication performance, key indicators were identified and denoted as $\overline{A}$, $\overline{a}$, $\overline{\Delta }$, $\overline{\delta }$, $\overline{T}$, and $\overline{t}$. These indicators capture average values of the power of controller communications, the power of communications from the UAV, the time intervals between control communications, the time intervals between communications from the UAV, the duration of control communications, and the duration of communications from the UAV, respectively. These metrics are summarized in

Table 3. Average values of $\overline{A}$, $\overline{a}$ in dB and $\overline{\Delta }$, $\overline{\delta }$, $\overline{T}$, $\overline{t}$ in ms, where the UAVs are operating in an anechoic chamber.

		dB		ms
ID	Mode	$\overline{\mathit{A}}$	$\overline{\mathit{a}}$	$\overline{\Delta }$	$\overline{\mathit{\delta }}$	$\overline{\mathit{T}}$	$\overline{\mathit{t}}$
UAV 1	stationary mode	55.3822	71.4989	4.4625	1.1491	0.5119	3.6182
	flying mode	55.1249	71.3536	4.3294	1.2414	0.5210	3.5506
UAV 2	stationary mode	57.8625	72.1214	4.5874	1.1025	0.4944	3.6711
	flying mode	58.7050	72.2870	4.4130	1.3551	0.5111	3.5367
UAV 3	stationary mode	50.2711	66.7506	4.9143	1.3164	0.5010	1.9120
	flying mode	52.5168	69.1394	4.5118	1.2253	0.5050	3.6552
UAV 4	stationary mode	51.3170	67.0678	4.7460	1.2841	0.5023	2.1140
	flying mode	54.2701	69.4841	4.6014	1.1921	0.5091	3.7247
UAV 5	stationary mode	58.2450	77.01372	11.8292	4.2116	2.1770	9.8867
	flying mode	59.7410	78.2241	11.0148	4.0166	2.3416	9.9708
UAV 6	stationary mode	58.9130	77.8752	11.7701	4.2330	2.2167	9.7462
	flying mode	60.3760	79.5371	10.9104	4.1520	2.4783	9.8133

and

Table 4. Average values of $\overline{A}$ and $\overline{a}$ in dB and $\overline{\Delta }$, $\overline{\delta }$, and $\overline{T}$, $\overline{t}$ in ms, where the UAVs are operating outside of the anechoic chamber.

		dB		ms
ID	Mode	$\overline{\mathit{A}}$	$\overline{\mathit{a}}$	$\overline{\Delta }$	$\overline{\mathit{\delta }}$	$\overline{\mathit{T}}$	$\overline{\mathit{t}}$
UAV 1	stationary mode	55.5115	71.5062	4.5528	1.3659	0.5824	3.7261
	flying mode	55.7586	71.7041	4.6684	1.4433	0.5849	3.7487
UAV 2	stationary mode	58.4667	72.4944	4.5758	1.5496	0.5674	3.6522
	flying mode	59.3555	72.6024	4.7786	1.6381	0.5716	3.6503
UAV 3	stationary mode	50.7428	67.1945	5.6016	1.3653	0.5579	2.2809
	flying mode	52.8791	69.5513	4.9578	1.4684	0.5743	3.8416
UAV 4	stationary mode	51.6803	67.7132	5.3215	1.9826	0.5235	2.3201
	flying mode	54.9368	69.7634	5.0276	1.7456	0.5635	3.9162
UAV 5	stationary mode	58.5955	77.6857	11.8932	4.3783	2.3241	10.0141
	flying mode	60.2437	78.4709	11.5718	4.1476	2.5611	10.2809
UAV 6	stationary mode	59.4635	78.3755	11.8218	4.3462	2.8363	10.1029
	flying mode	60.7714	79.8402	11.2722	4.2119	2.8781	10.3342

, providing a clear overview of the communication characteristics for each set of UAVs and their respective modes.

To statistically evaluate differences in these indicators, the study employed the Wilcoxon–Mann–Whitney test (MW test) [21,22]. The Wilcoxon–Mann–Whitney test works by comparing ranks of the sampling. The purpose of this test is to show the difference in the average values of two samples and to highlight a difference in distribution. As the value is based on ranks, it allows differences between two distributions to be highlighted without requiring the samples to be normally distributed. This non-parametric test designed to compare the distributions of two independent samples is suitable for analyzing UAV data that may not conform to normality. For each comparison, the null hypothesis ($H_0$) posited that the medians of the distributions were equal (${\eta }_1-{\eta }_2=0$), while the alternative hypothesis ($H_1$) suggested a difference (${\eta }_1-{\eta }_2\ne 0$). A significance threshold, $\alpha =0.05$ was used to determine whether the observed differences were statistically significant. A p-value less than or equal to $\alpha$ led to the rejection of $H_0$, indicating a meaningful difference in the distributions.

Table 5. Wilcoxon–Mann–Whitney test, where the UAVs are operating in an anechoic chamber; the values are rounded to ${10}^{-4}$ decimal points.

(a) Two DJI MAVIC 2 Zoom UAVs
	A		a		$\Delta$		$\delta$		T		t
	p-value	B	p-value	B	p-value	B	p-value	B	p-value	B	p-value	B
Stationary mode
UAV 1 vs. 2	0	1	0	1	0.0003	1	0.0002	1	0	1	0.0815	0
Flying mode
UAV 1 vs. 2	0	1	0	1	0.3616	0	0.5059	0	0.0576	0	0.9891	0
Flying mode vs. Stationary mode
UAV 1	0.0955	0	0	1	0.1791	0	0.6301	0	0.0010	1	0.3694	0
UAV 2	0.0014	1	0	1	0.0511	0	0.2664	0	0	1	0.8739	0
(b) Two AIR 2S UAVs
	A		a		$\Delta$		$\delta$		T		t
	p-value	B	p-value	B	p-value	B	p-value	B	p-value	B	p-value	B
Stationary mode
UAV 3 vs. 4	0.0004	1	0.0001	1	0.0002	1	0	1	0	1	0	1
Flying mode
UAV 3 vs. 4	0.1363	0	0.3115	0	0.0027	1	0.4834	0	0	1	0.4396	0
Flying mode vs. Stationary mode
UAV 3	0.0701	0	0.1363	0	0.0277	1	0.1642	0	0.4115	0	0.1944	0
UAV 4	0.0007	1	0.0001	1	0.2322	0	0	1	0.4518	0	0	1
(c) Two DJI Phantom (2 and 3)
	A		a		$\Delta$		$\delta$		T		t
	p-value	B	p-value	B	p-value	B	p-value	B	p-value	B	p-value	B
Stationary mode
UAV 5 vs. 6	0.0002	1	0.0001	1	0.1902	0	0.2106	0	0.3668	0	0.1940	0
Flying mode
UAV 5 vs. 6	0.0810	0	0.0106	1	0.2801	0	0.2157	0	0.1516	0	0.1368	0
Flying mode vs. Stationary mode
UAV 5	0.0086	1	0.0173	1	0.4602	0	0.4210	0	0.3345	0	0.3093	0
UAV 6	0.0029	1	0.0001	1	0.4458	0	0.4225	0	0.3612	0	0.1294	0

and

Table 6. Wilcoxon–Mann–Whitney test, where the UAVs are operating outside of the anechoic chamber; the values are rounded to ${10}^{-4}$ decimal points.

(a) Two DJI MAVIC 2 Zoom UAVs
	A		a		$\Delta$		$\delta$		T		t
	p-value	B	p-value	B	p-value	B	p-value	B	p-value	B	p-value	B
Stationary mode
UAV 1 vs. 2	0	1	0	1	0.0005	1	0.0003	1	0	1	0.07421	0
Flying mode
UAV 1 vs. 2	0	1	0	1	0.2561	0	0.4142	0	0.0384	0	0.6427	0
Flying mode vs. Stationary mode
UAV 1	0.06921	0	0	1	0.2716	0	0.8219	0	0.0021	1	0.3147	0
UAV 2	0.0024	1	0	1	0.0630	0	0.4263	0	0	1	0.6591	0
(b) Two AIR 2S UAVs
	A		a		$\Delta$		$\delta$		T		t
	p-value	B	p-value	B	p-value	B	p-value	B	p-value	B	p-value	B
Stationary mode
UAV 3 vs. 4	0.0001	1	0.4251	0	0.0571	0	0.0004	1	0	1	0	1
Flying mode
UAV 3 vs. 4	0.0009	1	0.1207	0	0.2172	0	0.3077	0	0	1	0.3404	0
Flying mode vs. Stationary mode
UAV 3	0.3388	0	0.4849	0	0.2419	0	0.0130	1	0.1250	0	0.0003	1
UAV 4	0.0320	1	0.0810	0	0.3496	0	0.2783	0	0.0388	1	0.1352	0
(c) Two DJI Phantom (2 and 3)
	A		a		$\Delta$		$\delta$		T		t
	p-value	B	p-value	B	p-value	B	p-value	B	p-value	B	p-value	B
Stationary mode
UAV 5 vs. 6	0.0002	1	0.0001	1	0.3291	0	0.6480	0	0.3392	0	0.2076	0
Flying mode
UAV 5 vs. 6	0.0046	1	0.0018	1	0.0944	0	0.2681	0	0.0576	0	0.2488	0
Flying mode vs. Stationary mode
UAV 5	0.3957	0	0.0070	1	0.4910	0	0.2704	0	0.2216	0	0.2092	0
UAV 6	0.3525	0	0.0029	1	0.3164	0	0.3036	0	0.4633	0	0.3629	0

summarize the results, presenting both the p-values and a Boolean indicator (B) that takes a value of 1 if $H_0$ is rejected and 0 otherwise.

The results reveal notable differences in the communication characteristics between the sets of UAVs. To highlight the ability of our study to accurately identify UAV communication frames, we describe the result for the 2 MAVIC 2 Zoom. For instance, in stationary mode, significant differences were observed between the two MAVIC 2 Zoom UAVs for most criteria, except in video communication times, which showed no statistically significant variation. This consistency in video communication under stationary conditions suggests stable handling of video signals irrespective of minor hardware or software variations within the same UAV model.

In flying mode, where the UAVs received active motion commands, the communication intervals and times were slightly different between the two MAVIC 2 Zoom UAVs. This finding implies that the method and frequency of transmitting motion commands are fully consistent within this set of UAVs, likely reflecting the design and operational protocols.

An intriguing pattern emerged when analyzing the behavior of individual UAVs in the two modes. For the same UAV, the transmission times of the control signals, $\overline{T}$, exhibited notable changes when transitioning from stationary to flying mode. Furthermore, the intervals between these control signal transmissions, $\overline{\Delta }$, also vary between modes, indicating that the timing rhythm of the command transmission is affected by the type of operational mode.

The average power of controller communications $\overline{A}$ is used as an additional metric to identify communication frames associated with the same UAV. This parameter demonstrated strong potential as an additional indicator for distinguishing UAVs based on their unique communication characteristics. However, the study excluded signal strength from the analysis despite observing significant differences in this parameter between MAVIC 2 Zoom UAVs. This exclusion was attributed to the stationary nature of the experimental setup, which did not involve UAV mobility. In future studies, incorporating dynamic scenarios with varying spatial separations and environmental factors could provide a more comprehensive understanding of the role of signal strength in UAV communication.

In conclusion, this study emphasizes the significance of characterizing the behavior of UAV communication across various operating modes and sets of UAVs. The findings offer valuable insight into the nuances of video and control signal transmission, paving the way for optimizing UAV communication systems for diverse applications. The use of robust statistical methods and well-defined metrics ensures the reliability of the results, making this research a solid foundation for further exploration in UAV communication analysis.

5.2. Automatic Linking

Our comprehensive study of two MAVIC 2 Zoom UAVs, two AIR 2S UAVs, a Phantom 4 UAV, and a Phantom 3 UAV has provided critical insights into the inherent differences and similarities among these UAVs within the same model family. This investigation is particularly significant for understanding how variations in design, technology, and operational parameters influence their characteristics of communication signals. The ability to differentiate and analyze these signals is crucial for various applications, including UAV identification, tracking, and ensuring secure communication protocols.

In associating communication signals with their respective UAVs, a fundamental challenge arises from the distinct differences between signals emitted by the UAVs themselves and those emitted by their controllers. These differences are often influenced by hardware specifications, signal modulation techniques, and frequency allocations, which vary between UAV models and their controllers. Addressing this challenge is vital for ensuring accurate identification and reducing potential errors in signal attribution.

Our approach involves a two-step systematic process for tracking and associating signals. The first step focuses on differentiating between signals emitted by the controllers and those from the UAVs. This distinction is achieved through a combination of signal analysis techniques, such as spectral analysis, modulation scheme identification, and machine learning classifiers trained to recognize unique patterns associated with controller and UAV transmissions. By isolating the specific signals for the UAVs, we create a foundation for a more precise analysis.

In the second step, the differentiated signals are subjected to a detailed examination to establish a strong link between the communication signals and the corresponding UAVs. This involves analyzing the temporal, spectral, and spatial characteristics of the signals. Furthermore, by leveraging advanced signal correlation techniques and real-time processing algorithms, we can accurately associate each communication signal with its source UAV. This step is crucial for applications such as swarm management, intrusion detection, and forensic investigations involving UAV operations.

The Drone Frame Identification and Association Algorithm is presented in Algorithm 1, with a detailed explanation provided in the subsequent subsections. For this algorithm, the complexity is given by

$$Complexity:T\left(n\right)=n\left[O\right(L)+O(1)+O(LlogL+K\left)\right]$$

Algorithm 1 Drone Frame Identification and Association Algorithm

1: $F_{\mathrm{ref}}\leftarrow \mathrm{Reference}\mathrm{drone}\mathrm{frame}$   2: ${\mathcal{I}}_{ref}\leftarrow \mathrm{Reference}\mathrm{emmission}\mathrm{Interval}$   3: $\left\{F_i\right\}\leftarrow \mathrm{Incoming}\mathrm{signal}\mathrm{frames}$   4: ${\mathcal{I}}_i\leftarrow \mathrm{Incoming}\mathrm{emmission}\mathrm{Interval}$   5: ${\mathcal{C}}_{ref}\leftarrow \mathrm{Unique}\mathrm{signal}\mathrm{characteristics}\mathrm{Reference}$   6: ${\mathcal{C}}_i\leftarrow \mathrm{Incoming}\mathrm{unique}\mathrm{signal}\mathrm{characteristics}$   7: $\alpha \leftarrow \mathrm{Threshold}\mathrm{value}$   8: for new $F_i$ do   9:      Step 1: Identify Drone Frame Appearance 10:      $R_{F_i{F}_{\mathrm{ref}}}\stackrel{\Delta }{=}E\left[F_i{F}_{ref}^H\right]\mathrm{where}H\mathrm{denotes}\mathrm{an}\mathrm{Hermitian}\mathrm{transposition}$ 11:      if $R_{F_i{F}_{\mathrm{ref}}}\ge \alpha$ then 12:            Mark $F_i$ as a potential drone frame 13:      else 14:            Continue to the next frame 15:      end if 16:      Step 2: Verify Emission Timing 17:      if ${\mathcal{I}}_i\vDash {\mathcal{I}}_{ref}$ then 18:           Confirm $F_i$ as a valid drone frame 19:      else 20:            Discard $F_i$ as invalid 21:      end if 22:      Step 3: Associate Frame to UAV 23:      Extract unique signal 24:      ${\mathcal{C}}_i\sim {\mathcal{C}}_{ref}$ 25:      Assign $F_i$ to the UAV with the closest match 26: end for 27: Output: List of associated UAVs and their detected frames

5.2.1. Detecting Drone Frames via Cross-Correlation

In this initial step, the system aims to detect the presence of drone signals by comparing incoming frames with a pre-stored reference frame. This comparison is performed using cross-correlation, a mathematical technique that measures the similarity between two signals as a function of the time lag applied to one of them. Cross-correlation is particularly effective in scenarios with noise or interference, as it enhances the matching process by focusing on signal patterns rather than exact values.

The reference frame, which represents the characteristic signal structure of a known UAV communication pattern, serves as a template for identification. By computing the cross-correlation between each incoming frame and the reference frame, the system identifies peaks in the correlation output. These peaks indicate the presence of a signal resembling the reference frame, thus confirming the detection of a potential drone communication frame. This method ensures robustness against variations in signal strength or environmental noise, making it suitable for real-world UAV signal detection tasks.

The key operation of this step is the correlation. The complexity $O\left(L\right)$ per frame is multiplied by the number of frames n:

$$Complexity:O(n·L).$$

5.2.2. Verifying Emission Timing with Database

Once a potential drone frame has been detected, the next step is to verify its temporal validity. UAVs typically emit signals at regular intervals or within characteristic time windows determined by their communication protocols. The system leverages a pre-compiled database containing the expected emission intervals for various UAV models.

By cross-referencing the timestamp of the detected frame against the database, the system ensures that the signal aligns with the expected timing characteristics. If the frame’s emission time falls within the predefined interval, it is validated as a genuine UAV signal. Conversely, frames falling outside the expected intervals are discarded as potential false positives, such as noise or unrelated signals. This time-based validation step is critical for maintaining the accuracy and reliability of the detection system, especially in environments with high signal activity.

This is a simple comparison of values. The complexity is a constant per recording:

$$Complexity:O\left(1\right)per frame.$$

5.2.3. Decision: Linking Frame to UAV

After validating a frame as a UAV signal, the final step is to associate it with the specific UAV that generated it. Each UAV model, and in some cases an individual UAV, has unique communication signal characteristics. These characteristics include parameters such as frequency bands, modulation schemes, emission power levels, and specific patterns in signal structure.

By analyzing the similarities and differences between the detected signal’s characteristics and the database entries using a cross-correlation technique, the frame is assigned to the UAV with the closest match. This association process not only identifies the source UAV but also helps track its activity over time.

The frame association follows a process similar to UAV communication signals originating from the remote controller, with the primary distinction lying in the reference frame used and the communication interval. This method ensures that signals are matched accurately to their respective UAVs by analyzing signal patterns and communication properties. Feature extraction ($C_i$) depends on the method. Here, it is $O\left(LlogL\right)$. The comparison of a signal and a reference signal depends on the number of reference UAVs (K). If the signal is compared to K drones, the complexity is $O\left(K\right)$. For this step, the total complexity is

$$Complexity:O(K+LlogL).$$

5.2.4. Evaluation

In order to evaluate the performance of our communication frame allocation, we will calculate the error ratio obtained from the recordings in which two UAVs communicate with their RC. The error ratio is defined by Equation (1), where $F_N$ represents the frame of the reference UAV allocated to the second UAV, $F_P$ represents the frame of the second UAV allocated to the reference UAV, $T_P$ represents the frame of the reference UAV, correctly allocated, and $T_N$ represents the frame of the second UAV, correctly allocated.

$$ErrorRatio={\displaystyle \frac{F_N+F_P}{F_N+F_P+T_N+T_P}}=1-Accuracy$$

(1)

When two UAVs operate simultaneously, this approach successfully allocates the detected signal frames to the corresponding drone without misassociation errors. This high level of accuracy is facilitated by the stability of the UAV’s unique communication characteristics when tested in a controlled, anechoic chamber environment. The controlled testing environment minimizes external noise and interference, allowing the identification process to rely solely on each UAV signal’s stable and predictable characteristics. However, in the context of the RC’s communication signals during outside experimentation, there is a minor observed error rate of $6.9\%$, which is distributed between the DJI MAVIC 2 Zoom and the DJI Air 2S. The MAVIC 2 Zoom error rate is 7.8%, and the DJI Air 2S error rate is 8.7%. The confusion matrices are shown in Figure 7.

This suggests that while the UAV’s signal properties are stable and consistent, the remote controller’s signals are more prone to variability or overlap, resulting in occasional misassociation. This percentage quantifies the probability of errors occurring during the association of remote controller communication signals.

6. Conclusions

The purpose of this paper is to link communication frames belonging to authorized drones based solely on communication time characteristics. To do this, two test benches were used to measure the short-range (5 m) communication signals between the UAV and its RC in an anechoic chamber and in a room of our laboratory.

Initially, an analysis of communication frame characteristics revealed differences in patterns between drones of the same model. The studied parameters are the average power of control communications $\left(A\right)$, the average power of communications from the UAV $\left(a\right)$, the time intervals between control communications $(\Delta )$, the time intervals between communications from the UAV $\left(\delta \right)$, the duration of control communications $\left(T\right)$ and the duration of communications from the UAV $\left(t\right)$.

The decision to use time parameters was made because it offers a certain advantage. Once a pattern has been identified in the communication times of the UAVs and RC, all that is needed is to wait for the start time of the next frame to link it to the corresponding UAV. The other advantage is that when a dropout occurs in the frame allocation due to an error, all that is needed to reallocate the frames to the UAVS is to re-identify the pattern.

The algorithm we have developed makes it possible to take advantage of these characteristics in order to assign communication frames to drones whose characteristics are present in a reference database. Despite good performance in assigning communication frames ($\mathrm{error}\mathrm{rate}=6.9\%$), the procedure has only been tested for close-range communication (5 m). Due to flight authorization restrictions, we have not studied the algorithm ability to assign communication frames for low power levels measured at the monitoring antenna. In this type of configuration, one of the difficulties would be identifying frames by correlation. Variations that are too small compared to the interference caused by other communications could prevent communication frames from being tracked.

To strengthen the conclusions and improve the robustness of the analysis, future research should incorporate a larger dataset with more measurements across a wider range of scenarios. This would involve testing additional UAV models and UAVs employing different communication protocols. The inclusion of a variety of UAVs and communication standards would allow for a more comprehensive evaluation of the method’s performance and adaptability across diverse operational conditions. It would also be possible to use deep learning approaches (which are more robust for identifying low-power communications in noisier environments) to identify transmission times instead of autocorrelation.

By addressing these limitations, future work can provide a more generalized and reliable framework for UAV signal identification and frame association, ensuring applicability in practical, real-world deployments.