Database of RF fingerprinting on use case IoT devices

This document is a dataset of radiofrequency signals. It is composed of 1000 signals coming emitted by 10 different devices. This dataset was developped for benchmarking machine learning methods on an Internet of Things classification task: recognizing which device emitted a signal. This dataset is in an adaptation of the dataset collected by Basak et al. in “Drone classification from RF fingerprints using deep residual nets” (IEEE COMSNETS conference, 2021). Basak et al. collected signals from six commercial drones, three drone radio-controllers and one WiFi router. The conducted the measurements in an anechoic chamber, using a universal software radio peripheral (USRP X310) placed seven meters apart from the devices . The signals were all in the 2.4 GHz ISM band and the whole 100 MHz band was received instantaneously using a receiving sampling rate of 100 MSps (i.e. the system down-converted the signal frequencies to the 0-100 MHz band to sample them correctly). While the original dataset by Basak et al. consisted in spectrograms of 256 frequency bins by 256 time frames, we have converted in into averaged spectra of 256 frequency bins. Furthermore, while Basak et al. have considered several noise levels, here we only consider the lowest noise level available (-60 dBm). The database is stored in an h5 file, a format adapted to databases. Inside the file there are two datasets: the signals (‘Signals’) and the targets (‘Targets’). The targets correspond to the ten different classes of signals: Parrot Disco (0), Q205 (1), Tello (2), MultiTx (3), Nine Eagles (4), Spektrum DX4e (5), Spectrum DX6i (6), Wltoys (7), S500 (8) and Linkys router (9). This dataset corresponds to the Deliverable D6.2 of the RadioSpin EU funded project.

本数据集为射频信号数据集，共包含10种不同设备发射的1000路信号。本数据集专为物联网（Internet of Things, IoT）分类任务下的机器学习方法基准测试而构建，目标为识别信号所属的发射设备。本数据集改编自Basak等人于2021年IEEE COMSNETS会议发表的论文《基于深度残差网络的无人机射频指纹分类》中的采集数据集。
Basak等人的原始数据集采集了6款商用无人机、3款无人机遥控器与1台WiFi路由器的信号，测试环境为微波暗室（anechoic chamber），使用通用软件无线电外设（universal software radio peripheral, USRP X310）在距被测设备7米处完成信号采集。所有信号均处于2.4 GHz ISM频段，系统以100 MSps的采样率瞬时捕获整段100 MHz带宽的信号，即系统将信号频率下变频至0-100 MHz频段以实现正确采样。
原始数据集的格式为256个频点×256个时间帧的语谱图（spectrograms），本数据集将其转换为256个频点的平均频谱。此外，Basak等人的研究考虑了多种噪声水平，而本数据集仅保留了最低噪声水平（-60 dBm）。
本数据集以h5格式存储，该格式适配大规模数据库的存储需求。文件内包含两个数据集：分别为"Signals"（信号数据）与"Targets"（标签数据）。标签对应10个信号类别：Parrot Disco（0）、Q205（1）、Tello（2）、MultiTx（3）、Nine Eagles（4）、Spektrum DX4e（5）、Spectrum DX6i（6）、Wltoys（7）、S500（8）与Linkys路由器（9）。
本数据集对应欧盟RadioSpin资助项目的可交付成果D6.2。

搜集汇总

背景与挑战

背景概述

该数据集包含10种物联网设备的1000个射频信号，用于机器学习分类任务基准测试。数据集改编自Basak等人的研究，转换为256频率箱的平均频谱，并仅包含最低噪声水平（-60 dBm）的信号。