A Multi-Scenario UAV RF Dataset with Real-World Acquisition and Signal Processing Benchmarking

arXiv:2603.00106 · eess.SP, cs.CR · Submitted 2026-02-18 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Security Radio. Generated commentary on the latest security and cryptography papers.

Nadia: I'm Nadia, and with me are Elias and Priya, guest researcher.

Elias: Today's paper: "A Multi-Scenario UAV RF Dataset with Real-World Acquisition and Signal Processing Benchmarking".

Nadia: The gist The researchers present a real-world multi-scenario unmanned aerial vehicle (UAV) radio frequency (RF) dataset, namely DRFF-R2, which is collected using a dedicated acquisition platform under diverse operational conditions.

Elias: First, who's behind it and why it matters.

Paper summary: Elias: So to wrap up what we heard about "A Multi-Scenario UAV RF Dataset with Real-World Acquisition and Signal Processing Benchmarking," this paper is presenting DRFF-R2, a real world dataset collected under diverse operational conditions.

Priya: What's the main thing we should remember from all that talk about the different scenarios and configurations they set up?

Nadia: We should remember that the strength of this dataset is its systematic organization into seven subsets covering everything from ambient noise to indoor WiFi interference (<ref:2603.00106#pg8>). It’s a comprehensive resource for anyone looking into RF fingerprinting identification or model-level recognition (<ref:2603.00106#pg3>).

Elias: And the authors' focus on the hardware and configuration setup shows they are building something meant to be used by other researchers, ensuring consistency across different setups (<ref:2603.00106#pg1>).

Priya: For someone just listening to this show, what does this mean for the world in terms of UAV research?

Nadia: It means that future systems designed to detect or identify drones based on their radio signals will be much more robust if they are trained using data that mimics these varied real-world conditions (<ref:2603.00106#pg3>).

Elias: The paper itself is important because it addresses the critical gap in having large, diverse drone RF datasets out there (<ref:2603.00106#pg2>). They’re giving researchers the material they need to start building those recognition systems (<ref:2603.00106#pg8>).

Priya: So, the core implication is that having this structured data allows for studying how RF fingerprints change over time and in response to interference, which is what a lot of current work has been pointing toward (<ref:2603.00106#pg3>).

Nadia: That's right; it’s about providing the necessary foundation so we can actually test those advanced techniques on real data (<ref:2603.00106#pg8>).

Conclusion: Nadia: So we’ve looked at how they built this dataset called DRFF-R2, and now it’s time to talk about what that title actually means for us.

Elias: It sounds like they’re just collecting a bunch of drone radio signals, but the focus is on making sure those signals come from really messy real-world situations.

Priya: What I see is that the authors are trying to move away from clean lab tests and into something that actually mimics what happens when you fly a drone outside.

Nadia: Exactly. They’ve structured it into seven different scenarios, so you aren't just getting one type of signal, right?

Elias: And those scenarios cover everything from ambient noise to being inside an office with Wi-Fi traffic. That’s important for testing if a detection system is actually going to work in the wild.

Priya: From my side, it matters because it gives us a much better picture of how these signals behave when they are mixed up with other things that aren't just noise, but active interference.

Nadia: So, the main point is they’ve created a standardized way to test if we can actually use RF fingerprints to tell a drone apart reliably across different flight states.

Elias: And I’m wondering what assumptions they made about the signal acquisition hardware itself—like making sure all those different UAV models were tested against the same receiver setup.

Priya: That’s a key detail, Elias. If you don't control your testing environment, the results are just noise filtered through a weird lens. The authors spent time controlling that configuration before they started recording anything at all.

Nadia: It means this isn't just a random collection of data; it’s a curated resource designed for structured experimentation, which is what we need if we’re trying to build cheap detection tools.

Elias: Cheap tools require knowing exactly what the system is supposed to handle, so having that detailed metadata attached to every file really helps with the proof side of things.

Priya: It moves us closer to understanding how these signals change over time and in different physical spaces, which is essential for any privacy or measurement work we do on drones.

Nadia: So, essentially, this paper gives researchers a robust playground where they can finally test if their detection algorithms can actually distinguish between different types of drone activity under realistic interference.

School of Cyber Science and Engineering, Southeast University · Chien-Shiung Wu College, Southeast University · National Mobile Communications Research Laboratory, Southeast University · Centre for Wireless Innovation (CWI), Queen’s University Belfast

eess.SP, cs.CR

Submitted: 2026-02-18

Updated: 2026-10-09

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 78/100

The gist: The gist The researchers present a real-world multi-scenario unmanned aerial vehicle (UAV) radio frequency (RF) dataset, namely DRFF-R2, which is collected using a dedicated acquisition platform

Key concepts

DRFF-R2 Dataset: A comprehensive collection of radio frequency (RF) recordings from 26 different UAV units spanning 8 models. It was captured using a unified framework to ensure consistency in hardware and environment, providing a rich resource for research.
Multi-Scenario Organization: The dataset is systematically divided into seven distinct subsets representing different operational conditions. These scenarios include ambient noise, controlled propagation indoors, outdoor flight at various altitudes/distances, indoor interference with WiFi, and multi-drone mixed operations.
RF Fingerprinting (RFF): A technique used to identify specific UAV models or states by analyzing their unique radio frequency signatures. The dataset is designed to support RFF identification and model-level recognition by comparing recorded signals against known patterns.
Time-Varying RFF Study: The ability to track how the RF fingerprint of a UAV changes over time. This involves analyzing the signal characteristics across different flight states, altitudes, and acquisition days to understand temporal variations in the radio signal.

Terminology

Consistency — 4× in this paper · explained in 3 episodes
Fingerprinting — 4× in this paper · explained in 2 episodes
Benchmark — 2× in this paper · explained in 2 episodes
Benchmarking — 2× in this paper · explained in 4 episodes
Unified Framework — 2× in this paper · explained in 5 episodes
Feature Engineering — 1× in this paper · explained in 3 episodes
Wavelet Transform — 1× in this paper · explained in 2 episodes

Summary

The gist The researchers present a real-world multi-scenario unmanned aerial vehicle (UAV) radio frequency (RF) dataset, namely DRFF-R2, which is collected using a dedicated acquisition platform under diverse operational conditions.

Dataset Overview and Scope

The dataset is systematically organized into seven well-defined subsets corresponding to different operational and signal composition scenarios to facilitate structured experimentation<ref:2603.00106#pg8>. The dataset contains RF recordings from 26 UAV units spanning 8 distinct models, captured across varying flight states, altitudes, speeds, acquisition days, and receiver configurations<ref:2603.00106#pg2>. By covering diverse acquisition settings and signal compositions, the dataset provides a comprehensive resource for future UAV RF signal research<ref:2603.00106#pg3>. The dataset comprises RF signals from 26 UAV units spanning 8 distinct models and is systematically organized into seven well-defined subsets, each tailored to facilitate different research objectives and application scenarios<ref:2603.00106#pg8>.

Data Acquisition and Hardware Configuration

The dataset was constructed through a controlled RF reception procedure comprising environmental baseline measurement, scenario-specific data collection and structured data storage<ref:2603.00106#pg4>. The acquisition system consisted of the UAVs under test, RF receivers, a data storage subsystem, and auxiliary field equipment<ref:2603.00106#pg5>. Specifically, the hardware components were configured and tested prior to data collection<ref:2603.00106#pg5>. The acquisition system included UAVs, an SDRbased receiver consisting of two Ettus Research universal software radio peripheral (USRP) 2943 devices, a Windows 11 laptop with 16GB RAM and a 2TB external storage device, and a dual-band omnidirectional antenna with a gain of 5 dBi<ref:2603.00106#pg5>. The sampling rate was fixed at Fs = 100 MS/s, resulting in each acquisition segment containing approximately 1.4 × 108 complex I/Q samples<ref:2603.00106#pg5>.

Scenario Design and Signal Collection Procedures

Data was collected across multiple operational scenarios to capture realistic propagation dynamics and interference conditions<ref:2603.00106#pg6>. The seven subsets are designed to facilitate different research objectives and application scenarios<ref:2603.00106#pg8>. These scenarios include:

Ambient Noise Scenario, which records background RF signals to characterize the baseline electromagnetic environment<ref:2603.00106#pg6>.
Controlled Propagation Scenario, conducted inside an enclosure lined with RF absorbing material to constrain multipath propagation and external interference<ref:2603.00106#pg6>.
Outdoor Flight Scenarios, involving controlled flight maneuvers at predefined distances (e.g., 50 m horizontal separation from the receiver) and altitudes (10 m, 30 m, and 50 m)<ref:2603.00106#pg5>.
Indoor Interference Scenarios, where the UAV operated within enclosed indoor environments with active WiFi traffic in the 2.4 GHz band<ref:2603.00106#pg6>.
Multi-Drone Mixed Scenarios, where multiple UAVs were operated simultaneously within the same frequency band during data collection<ref:2603.00106#pg6>.

The signal acquisition procedure followed a predefined workflow covering UAV operation, SDR configuration, and data recording<ref:2603.00106#pg5>. The UAV operators were responsible for drone operation and flight-state control<ref:2603.00106#pg5>. The acquisition program operator was responsible for SDR configuration and data saving, ensuring key parameters like sampling rate, center frequency, and receiver gain were set according to the designated scenario prior to recording<ref:2603.00106#pg5>.

Data Organization and Metadata

The drone RF signal data are stored in MAT format<ref:2603.00106#pg8>. A structured naming convention is used for each file, such as mavic3 1 Ascend c17 u1 d2.mat, which encodes key metadata including UAV model, operational state, acquisition scenario, and timestamp<ref:2603.00106#pg9>. This naming convention enables efficient indexing and retrieval of the data<ref:2603.00106#pg9>. Each data file contains a fixed internal structure with predefined variables such as RF0 I In-phase (I) component of the received signal, Fs Sampling rate (Hz), CenterFrequence Center frequency (Hz), Gain Receiver gain (dB), State UAV operational state label, Distance Horizontal distance between the UAV and the receiver, Height Flight altitude at the time of signal recording, and FlightMode Description of the UAV flight mode<ref:2603.00106#pg9>.

Technical Validation

To demonstrate data quality, a time-frequency representation was computed using short-time Fourier transform (STFT) representations for representative RF signal segments from dataset-2, dataset-3, and dataset-6<ref:2603.00106#pg10>. The magnitude spectrogram is computed as S(m, k) = X(m, k)<ref:2603.00106#pg10>. Representative STFT spectrograms exhibit structured spectral patterns across scenarios, with observable differences in spectral occupancy, temporal continuity, and interference structure among distinct signal compositions<ref:2603.00106#pg10>. Furthermore, a baseline supervised classification experiment was conducted using dataset-1 to assess label consistency and class separability<ref:2603.00106#pg11>. The confusion matrix in Figure 4 shows the distribution of predicted labels across the 26 operational states, indicating observable correspondence between the recorded RF signals and the annotated flight state labels<ref:2603.00106#pg12>.

Data Availability

The dataset is publicly hosted on the Science Data Bank platform at https://doi.org/10.57760/sciencedb.36815 [17], ensuring long-term preservation, accessibility, and reproducibility. The dataset is distributed under the CC BY 4.0 license. The signal acquisition program and associated experimental code are publicly available at https://doi.org/10.57760/sciencedb.36815<ref:2603.00106#pg10>. The work was supported in part by the National Key Research and Development Program of China under Grant 2024YFE0200700, in part by National Science Foundation of China (NSFC) under Grants 62371131, and in part by Undergraduate Training Programs for Innovation of Jiangsu Province<ref:2603.00106#pg10>. The references include works such as DroneRFa [3] and DroneRFb-DIR [4]<ref:2603.00106#pg12>. The references also cover studies on time-varying RFF characteristics [15][16]<ref:2603.00106#pg3>. The references include studies on RF fingerprinting over time [15] and hybrid time-varying distortions [16]<ref:2603.00106#pg10>. The references also cover drone detection and identification using deep learning approaches [8]. The references also cover studies on recognition of UAV video signal using RF fingerprints in the presence of WiFi interference [11] and detection and classification of UAVs using RF fingerprints in the presence of WiFi and Bluetooth interference [12]<ref:2603.00106#pg10>. The references also cover studies on waveform analytics for RF-based UAV detection and identification system using machine learning [14]<ref:2603.00106#pg10>. The references also cover studies on drone RF signal based UAV detection and mode classification [10]<ref:2603.00106#pg10>. The references also cover studies on radio frequency control and video signal recordings of drones [7]<ref:2603.00106#pg10>. The references also cover studies on drone identification via distilled RF fingerprints-based LLM in ISAC networks [5]<ref:2603.00106#pg10>. The references also cover studies on radio frequency fingerprinting over time [15] and hybrid time-varying distortions [16]<ref:2603.00106#pg10>.

Improvements for AI systems

textbfSample-based Model Recognition Under Interference: Introduce a Unified Multi-Scenario Classifier for Heterogeneous Drones with Wi-Fi Noise Effect Analysis. This system can perform drone recognition under interference conditions by leveraging the dataset6-wifi mixed data to train a classifier that specifically identifies drone models while quantifying the degradation caused by co-channel and adjacent-channel interference, directly addressing existing limitations in recognizing drones when operating near Wi-Fi traffic [11][12].
textbfReal-time Flight State Awareness with Enhanced Temporal Resolution: Implement a deep learning model optimized for the dataset3-single drone hover data to perform UAV flight state awareness by analyzing the time-varying RFF characteristics. This system will utilize the STFT analysis described in Section 4 to extract features from segments like mavic3 1 hover c1 u2 d1, allowing it to accurately distinguish between different states (e.g., Hovering (Video) vs. Hovering (Occluded)) based on subtle spectral patterns, thereby improving the robustness of state recognition beyond the baseline EfficientNetB0 model [8][9][10].
textbfAdaptive Signal Processing for Time-Varying RFF: Develop a time-varying RFF analysis module capable of tracking signal changes over time. This module will be trained on data exhibiting time-varying RFF characteristics [15][16] to identify and model the evolution of RF fingerprints as a UAV changes flight conditions (speed, altitude, trajectory), enabling predictive maintenance or dynamic threat assessment capabilities.

Abstract

We present a real-world multi-scenario unmanned aerial vehicle (UAV) radio frequency (RF) dataset, namely DRFF-R2, which is collected using a dedicated acquisition platform under diverse operational conditions. All signals are acquired within a unified framework to ensure consistency in hardware configuration and environmental settings. The dataset is systematically organized into seven well-defined subsets corresponding to different operational and signal composition scenarios to facilitate structured experimentation. Each file follows a clearly annotated naming convention to enable convenient data indexing and reproducible analysis. The dataset contains RF recordings from 26 UAV units spanning 8 distinct models, captured across varying flight states, altitudes, speeds, acquisition days, and receiver configurations. By covering diverse acquisition settings and signal compositions, the dataset provides a comprehensive resource for future UAV RF signal research, including RF fingerprinting (RFF) identification, model-level recognition, flight state analysis, time-varying RFF study, and interference-aware signal processing.

Sources

RFUAV: A Benchmark Dataset for Unmanned Aerial Vehicle Detection and Identification