WO2021103206A1 - 基于机器学习算法的无线射频设备身份识别方法及系统 - Google Patents

基于机器学习算法的无线射频设备身份识别方法及系统 Download PDF

Info

Publication number
WO2021103206A1
WO2021103206A1 PCT/CN2019/126138 CN2019126138W WO2021103206A1 WO 2021103206 A1 WO2021103206 A1 WO 2021103206A1 CN 2019126138 W CN2019126138 W CN 2019126138W WO 2021103206 A1 WO2021103206 A1 WO 2021103206A1
Authority
WO
WIPO (PCT)
Prior art keywords
radio frequency
entropy
classification
machine learning
equipment
Prior art date
Application number
PCT/CN2019/126138
Other languages
English (en)
French (fr)
Inventor
邹玉龙
王后星
马云龙
朱佳
Original Assignee
南京邮电大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 南京邮电大学 filed Critical 南京邮电大学
Publication of WO2021103206A1 publication Critical patent/WO2021103206A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2431—Multiple classes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T1/00—General purpose image data processing
    • G06T1/0007—Image acquisition
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L63/00—Network architectures or network communication protocols for network security
    • H04L63/08—Network architectures or network communication protocols for network security for authentication of entities
    • H04L63/0876—Network architectures or network communication protocols for network security for authentication of entities based on the identity of the terminal or configuration, e.g. MAC address, hardware or software configuration or device fingerprint
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W12/00—Security arrangements; Authentication; Protecting privacy or anonymity
    • H04W12/06—Authentication

Definitions

  • the invention relates to the technical field of equipment identification in wireless communication, in particular to a method and system for identification of wireless radio frequency equipment based on a machine learning algorithm.
  • Existing wireless device identification mainly by extracting the transient characteristic or steady-state characteristic parameter information of the radio frequency signal to establish the fingerprint characteristic information database of the device, when encountering an unknown device, the device is determined by comparing with the record in the fingerprint database identity of. Regardless of whether the transient characteristics or the steady-state characteristics are used, specific parameter information is extracted for device identification, and the image information of the radio frequency signal is not directly used.
  • some highly intelligent device identification methods add a radio frequency tag to the device, and then realize the identity of the device by identifying the radio frequency tag.
  • this method needs to transmit the radio frequency tag information in a wireless channel, and is vulnerable to eavesdropping attacks.
  • the purpose of the present invention is to overcome the deficiencies in the prior art, and propose a method and system for identification of radio frequency equipment based on machine learning algorithms, to solve the difficulty of parameter selection in the existing identification technology and the technology that tag information is easily tampered with. problem.
  • the present invention provides a wireless radio frequency device identification method based on a machine learning algorithm, which is characterized in that it includes the following processes:
  • the corresponding entropy value is calculated
  • the entropy value corresponding to the radio frequency device determine the final recognition result of the device.
  • the image size is 720*232 pixels.
  • the image is down-sampled.
  • the calculation of the corresponding entropy value according to the classification result of the radio frequency device includes:
  • the judging the final recognition result of the device according to the entropy value corresponding to the radio frequency device includes:
  • the mean value of entropy is compared with the preset threshold to determine the final recognition result of the device.
  • the present invention provides a wireless radio frequency equipment identification system based on a machine learning algorithm, which is characterized by including a radio frequency signal acquisition module, a signal conversion module, a classification identification module, an entropy calculation module, and a final result identification module; wherein :
  • the radio frequency signal acquisition module is used to collect the radio frequency signal sent by the radio frequency device to be identified;
  • the signal conversion module is used to convert the spectrogram of the radio frequency signal into an image
  • the classification recognition module is used to input the image into the preset convolutional neural network for classification, and obtain the classification result of the wireless radio frequency device;
  • the entropy value calculation module is used to calculate the corresponding entropy value according to the classification result of the wireless radio frequency device;
  • the final result recognition module is used to determine the final recognition result of the device according to the entropy value corresponding to the radio frequency device.
  • the image size is 720*232 pixels.
  • the classification and recognition module before the image is input into a preset convolutional neural network for classification, the image is down-sampled.
  • the calculation of the corresponding entropy value according to the classification result of the radio frequency device includes:
  • said judging the final identification result of this device according to the entropy value corresponding to the radio frequency device includes:
  • the mean value of entropy is compared with the preset threshold to determine the final recognition result of the device.
  • the beneficial effect achieved by the present invention is: the present invention uses the convolutional neural network (CNN) algorithm to train and test the image information of the radio frequency signal, without giving clear characteristic parameters, such as frequency offset, Information such as preamble and modulation characteristics, classifiers obtained through training and entropy calculation and comparison can effectively identify known devices and unknown devices, and can identify the specific identity of the device in the known device category to achieve wireless device security
  • the purpose of authentication is to improve the security of the wireless communication system.
  • Figure 1 is a flow chart of the method of the present invention
  • Fig. 2 is a schematic diagram of the probability of correct discrimination by applying the method of the present invention in an embodiment.
  • a wireless radio frequency equipment identification method based on a machine learning algorithm of the present invention is applied to the safety authentication of radio frequency equipment, as shown in Fig. 1, and includes the following steps:
  • Step 1 Collect radio frequency signals sent by multiple USRP devices, all of which have known device information
  • the transmitting end uses N USRP devices to transmit radio frequency signals.
  • the transmitting end device is called a USRP training device.
  • the N USRP devices know their device information (such as device type, device attributes, etc., as a classification label), and the receiving end uses A USRP device to receive radio frequency signals.
  • Step 2 Convert and store the radio frequency signal
  • the receiving end device Y sequentially collects the radio frequency signals of N training devices, and first normalizes the frequency spectrum of each radio frequency signal (the normalization method is the amplitude element of the radio frequency signal divided by the average value of the amplitude), and then uses Write the JPEG function in Labview to convert the normalized spectrogram into a visual normalized spectrogram.
  • the size of each image stored is 720*232 pixels. All radio frequency signals transmitted training apparatus X i unified reception processing method, in order to ensure uniformity of data collection.
  • Step 3 Input the converted image data into the convolutional neural network for training and learning, and obtain the classifier;
  • a convolutional neural network (CNN) algorithm (which is a machine learning algorithm) is used to construct a corresponding convolutional neural network, read image data transformed by N training devices, and input the convolutional neural network through two layers of convolution Pooling and training is performed to obtain a classifier. Since N training devices correspond to N types of labels, the output of the classifier is the label corresponding to a certain type of training equipment and the N type labels corresponding to each image data. Probability of decision.
  • the size of each image is 720*232 pixels.
  • the image data is down-sampled 4:1, and the image size after down-sampling is 180*58 pixels. In order to reduce the amount of training calculations.
  • Step 1 Entropy calculation
  • the transmitter uses M USRP equipment as the radio frequency equipment to be identified to transmit radio frequency signals.
  • the transmitter equipment is called the USRP test equipment.
  • the information of the M USRP equipment is unknown, and the receiver uses a USRP equipment to receive the radio frequency. signal.
  • Step 2 Mean calculation
  • Step 3 Threshold comparison
  • the test device By comparing the mean value of entropy corresponding to the test device with a preset threshold (empirical value), when the mean value of entropy is greater than the set threshold, the test device is judged as an unknown device; when the mean value of entropy is less than the set threshold, then The test equipment is judged as a known equipment, and specific known equipment information (including equipment type) is obtained according to the label given by the classifier.
  • a preset threshold empirical value
  • the determination of the threshold is obtained by testing a large number of data sets.
  • the specific method is described as: collecting image data of N devices for training, testing these N known devices, and obtaining the entropy value of N devices through step 1.
  • the image data of other N devices are collected without participating in training, only testing is performed, and the entropy value of these N devices is obtained by the same method.
  • the entropy data of these 2N devices are drawn into a simulation diagram, and through observation, it can be found that the entropy distribution of the trained equipment (known equipment) and the non-training equipment (unknown equipment) has obvious layering phenomenon.
  • the mean entropy of the unknown device is significantly greater than the mean entropy of the known device.
  • An artificially set one between the mean entropy of the known device and the unknown device The middle value is used as the threshold.
  • USRP 1, 2, and 3 in the figure are known devices, that is, devices that have been trained.
  • the classifiers obtained correspond to the three types of tags, USRP 1, 2, and 3, and USRP 4, 5, and 6 are unknown devices, that is, unidentified devices.
  • the trained equipment is tested by collecting image data of the six USRP 1, 2, 3, 4, 5, and 6 devices. During the test, each device collects data K times, and the above-mentioned entropy calculation and threshold comparison are used to verify the test. Probability of correctness.
  • Figure 2(a) is the correct probability of the test when the test data is not grouped
  • Figure 2(b) is the test data collected K times into groups, every 10 data is divided into a group, and each is calculated in the entropy value.
  • the 10 instantaneous entropy values of the group are summed and averaged, and then compared with the set threshold.
  • the set of test data is judged as an unknown device; when the mean entropy is less than the set threshold, it is judged as a known device, and the specific known device is obtained according to the label given by the classifier.
  • the classifier gives the same label more than 6 times, then the determination result is the specific known device corresponding to the label.
  • a method for identifying wireless radio frequency equipment based on machine learning algorithms of the present invention uses software radio platforms USRP and LabVIEW to send and receive radio frequency signals from different equipment, and convert the radio frequency signals into image information, using Python And Matlab software is responsible for training and learning image information and testing, calculating the device entropy value through the classifier, and introducing a threshold comparison mechanism to realize the identification of wireless devices.
  • the recognition accuracy rate of the method for known devices reaches more than 98%, and the threshold comparison algorithm is further adopted, and the discrimination rate of unknown devices can reach more than 90%.
  • the present invention has the beneficial effects that the image information of the radio frequency signal can effectively identify the known device and the unknown device, and can identify the specific identity of the device in the known device category to achieve the wireless device
  • the purpose of security authentication is to improve the security of the wireless communication system.
  • the present invention uses the image information of the radio frequency signal to realize the identification of wireless equipment based on the electromagnetic environment.
  • the present invention is a method for identification of wireless equipment based on the machine learning algorithm. Improve the security of the communication system.
  • the present invention provides a wireless radio frequency equipment identification system based on a machine learning algorithm, which is characterized by including a radio frequency signal acquisition module, a signal conversion module, a classification identification module, an entropy calculation module, and a final result identification module; wherein :
  • the radio frequency signal acquisition module is used to collect the radio frequency signal sent by the radio frequency equipment to be identified;
  • the signal conversion module is used to convert the spectrogram of the radio frequency signal into an image
  • the classification recognition module is used to input the image into the preset convolutional neural network for classification, and obtain the classification result of the radio frequency device;
  • the entropy value calculation module is used to calculate the corresponding entropy value according to the classification result of the wireless radio frequency device;
  • the final result recognition module is used to determine the final recognition result of the device according to the entropy value corresponding to the radio frequency device.
  • the image size is 720*232 pixels.
  • the classification and recognition module before the image is input into a preset convolutional neural network for classification, the image is down-sampled.
  • the calculation of the corresponding entropy value according to the classification result of the radio frequency device includes:
  • said judging the final identification result of this device according to the entropy value corresponding to the radio frequency device includes:
  • the mean value of entropy is compared with the preset threshold to determine the final recognition result of the device.
  • this application can be provided as methods, systems, or computer program products. Therefore, this application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
  • computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
  • These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing equipment to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction device.
  • the device implements the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.
  • These computer program instructions can also be loaded on a computer or other programmable data processing equipment, so that a series of operation steps are executed on the computer or other programmable equipment to produce computer-implemented processing, so as to execute on the computer or other programmable equipment.
  • the instructions provide steps for implementing the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Signal Processing (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computer Security & Cryptography (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Power Engineering (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computing Systems (AREA)
  • Computer Hardware Design (AREA)
  • Image Analysis (AREA)

Abstract

一种基于机器学习算法的无线射频设备身份识别方法及系统，属于无线通信技术领域。所述方法包括发射端发送射频信号，接收端接收处理信号，通过将射频信号转换为图像来训练学习卷积神经网络算法，进行身份识别。该方法借助机器学习算法，利用射频信号的图像信息可以有效地识别已知设备和未知设备，并能在已知设备类别中识别设备的具体信息，有效提高了无线通信系统的设备认证安全。

Description

基于机器学习算法的无线射频设备身份识别方法及系统 技术领域
本发明涉及无线通信中设备身份识别技术领域，具体涉及一种基于机器学习算法的无线射频设备身份识别方法及系统。
背景技术
现有的无线设备身份识别，主要通过提取射频信号的瞬态特征或稳态特征参数信息来建立设备的指纹特征信息库，当遇到未知设备时通过与指纹库中的记录进行对比来确定设备的身份。无论是利用瞬态特性还是稳态特征，都是通过提取特定的参数信息用于设备身份识别，没有直接利用射频信号的图像信息。
此外，一些智能化程度较高的设备识别方法通过给设备加入射频标签，进而通过识别射频标签来实现设备的身份识别，然而此方法需要在无线信道中传输射频标签信息，容易受到窃听攻击。
发明内容
本发明的目的在于克服现有技术中的不足，提出了一种基于机器学习算法的无线射频设备身份识别方法及系统，解决现有身份识别技术中参数选择的困难、标签信息易被篡改的技术问题。
为解决上述技术问题，本发明提供了一种基于机器学习算法的无线射频设备身份识别方法，其特征是，包括以下过程：
采集待识别无线射频设备发送的射频信号；
将射频信号的频谱图转换为图像；
将图像输入预设的卷积神经网络进行分类，获得此无线射频设备的分类结 果；
根据无线射频设备的分类结果，计算得到对应的熵值；
根据无线射频设备对应的熵值，判断此设备的最终识别结果。
进一步的，所述图像大小为720*232像素。
进一步的，所述图像输入预设的卷积神经网络进行分类之前，对图像进行下采样。
进一步的，所述根据无线射频设备的分类结果计算得到对应的熵值包括：
无线射频设备的分类结果为此无线射频设备属于第j(j＝1,2…N)类设备的概率，记为P j(j＝1,2…N)；
通过公式
Figure PCTCN2019126138-appb-000001
计算得到此无线射频设备对应的熵值。
进一步的，所述根据无线射频设备对应的熵值判断此设备的最终识别结果包括：
对无线射频设备进行多次分类，获得多个分类结果；
根据多个分类结果，计算得到对应的多个熵值；
根据多个熵值，计算得到对应的熵均值；
将熵均值与预设阈值进行比较判断，得到此设备的最终识别结果。
相应的，本发明提供了一种基于机器学习算法的无线射频设备身份识别系统，其特征是，包括射频信号采集模块、信号转换模块、分类识别模块、熵值计算模块和最终结果识别模块；其中：
射频信号采集模块，用于采集待识别无线射频设备发送的无线射频信号；
信号转换模块，用于将射频信号的频谱图转换为图像；
分类识别模块，用于将图像输入预设的卷积神经网络进行分类，获得此无 线射频设备的分类结果；
熵值计算模块，用于根据无线射频设备的分类结果，计算得到对应的熵值；
最终结果识别模块，用于根据无线射频设备对应的熵值，判断此设备的最终识别结果。
进一步的，信号转换模块中，所述图像大小为720*232像素。
进一步的，分类识别模块中，所述图像输入预设的卷积神经网络进行分类之前，对图像进行下采样。
进一步的，熵值计算模块中，所述根据无线射频设备的分类结果计算得到对应的熵值包括：
无线射频设备的分类结果为此无线射频设备属于第j(j＝1,2…N)类设备的概率，记为P j(j＝1,2…N)；
通过公式
Figure PCTCN2019126138-appb-000002
计算得到此无线射频设备对应的熵值。
进一步的，最终结果识别模块中，所述根据无线射频设备对应的熵值判断此设备的最终识别结果包括：
对无线射频设备进行多次分类，获得多个分类结果；
根据多个分类结果，计算得到对应的多个熵值；
根据多个熵值，计算得到对应的熵均值；
将熵均值与预设阈值进行比较判断，得到此设备的最终识别结果。
与现有技术相比，本发明所达到的有益效果是：本发明利用卷积神经网络(CNN)算法来训练和测试射频信号的图像信息，不需要给出明确的特征参数，如频偏、前导和调制特征等信息，通过训练学习得出的分类器以及熵值计算和比较可以有效地识别已知设备和未知设备，并能在已知设备类别中识别设备的 具体身份，达到无线设备安全认证的目的，提高了无线通信系统的安全性。
附图说明
图1为本发明方法的流程框图；
图2为实施例中应用本发明方法的判别正确概率示意图。
具体实施方式
下面结合附图对本发明作进一步描述。以下实施例仅用于更加清楚地说明本发明的技术方案，而不能以此来限制本发明的保护范围。
本发明的一种基于机器学习算法的无线射频设备身份识别方法，应用于无线射频设备的安全认证，参见图1所示，包括以下步骤：
1)训练过程：
步骤1：采集多个USRP设备发送的射频信号，此多个USRP设备均已知其设备信息；
利用现有的USRP设备(Universal Software Radio Peripheral，通用软件无线电外设)作为无线射频设备进行无线射频信号的收发。发射端采用N台USRP设备来发射无线射频信号，发射端设备称为USRP训练设备，此N台USRP设备已知其设备信息(如包括设备类型、设备属性等，作为分类标签)，接收端采用一台USRP设备来接受射频信号。
将N台USRP训练设备X i(i＝1,2,...N)(例如N取值3)置于某一位置，N台训练设备在相同频点处分别发送相同的射频信号，接收端USRP设备Y依次采集N个训练设备的射频信号，其中设备X i和Y为同一厂家生产的同一型号设备。
步骤2：对射频信号进行转换存储处理；
接收端设备Y依次采集到N台训练设备的射频信号，将各射频信号的频谱 图先经过归一化处理(归一化方法为射频信号的幅值元素除以幅值的均值)，再利用Labview中写入JPEG函数将归一化后的频谱图转换为可视的归一化频谱图，存储的每张图像大小为720*232像素。对所有训练设备X i发送的射频信号采用统一的接收处理方法，以保证数据采集的统一性。
步骤3：将转换后的图像数据输入卷积神经网络进行训练学习，获取分类器；
本发明中运用卷积神经网络(CNN)算法(是一种机器学习算法)构建相应的卷积神经网络，读取N台训练设备转化得到的图像数据，输入卷积神经网络，经过两层卷积池化并进行训练，得到分类器，由于N台训练设备对应N类标签，该分类器的输出为每张图像数据被判别为某一类训练设备所对应的标签以及N类标签分别对应的判决概率。
进一步，由于数据采集量较大，每张图像大小为720*232像素，在读取图像数据训练之前，将图像数据按照4:1进行下采样，下采样后的图像大小为180*58像素，以减小训练的计算量。
进一步，由于USRP设备中的振荡器频偏、相位噪声、调制器的调制误差、功放的非线性失真等因素，导致了各射频信号的频谱存在细微差异，为了充分学习不同训练设备发送的射频信号差异，需要对不同时刻、不同频段的训练设备信号进行采集、处理和训练，以得到更通用的分类器。
2)测试过程(即待识别无线射频设备的识别过程)：
步骤1：熵值计算；
测试时，发射端采用M台USRP设备作为待识别无线射频设备来发射无线射频信号，发射端设备称为USRP测试设备，此M台USRP设备信息未知，接收端采用一台USRP设备来接受无线射频信号。
每台测试设备的图像数据采集K次，利用上述训练得到的分类器对每台测试设备采集到的图像数据进行分类，进而可以得到此测试设备的数据判为第j(j＝1,2…N)类训练设备的概率，记为P j(j＝1,2…N)。通过公式
Figure PCTCN2019126138-appb-000003
计算得到此测试设备对应的熵值。
步骤2：均值计算；
因为每台测试设备的图像数据采集K次，相应的可得到K次测试结果，进而计算得到每次测试的熵值，通过公式
Figure PCTCN2019126138-appb-000004
计算得到该测试设备对应的熵均值，其中H i(i＝1,2…K)表示第i次测试时该测试设备对应的熵值。
步骤3：阈值比较；
通过将测试设备对应的熵均值与预设的阈值(经验值)进行比较，当熵均值大于设定的阈值时，则此测试设备判为未知设备；当熵均值小于设定的阈值时，则此测试设备判为已知设备，并根据分类器给出的标签获取具体的已知设备信息(包括设备类型)。
阈值的确定通过大量数据集的测试得到，具体方法描述为：采集N台设备的图像数据进行训练，并对这N台已知设备进行测试，通过步骤1得到N台设备的熵值。同时采集另外N台设备图像数据不参与训练，只进行测试，用同样方法得到这N台设备的熵值。将这2N台设备的熵值数据绘制仿真图，通过观察可以发现已训练设备(已知设备)和未参与训练设备(未知设备)的熵值分布有明显的分层现象。通过计算N台已知设备和N台未知设备的熵均值，可以发现未知设备的熵均值要明显大于已知设备的熵均值，在已知设备和未知设备的熵均值之间人为地设定一个中间值作为阈值。
结合图2，图中USRP1、2、3为已知设备，即已被训练过的设备，获得的分类器对应USRP1、2、3这三类标签，USRP4、5、6为未知设备，即未被训练过的设备，通过采集USRP1、2、3、4、5、6这六台设备的图像数据进行测试，测试时每台设备采集数据K次，利用上述熵值计算和阈值比较来验证测试正确概率。图2(a)是测试数据未分组情况下的测试正确概率，图2(b)是将K次采集的测试数据进行分组，每10个数据分为一组，在熵值计算中将每一组的10个瞬时熵值求和并取平均，再与设定的阈值进行比较。当熵均值大于设定的阈值，将该组测试数据判为未知设备；当熵均值小于设定的阈值，即判为已知设备，并根据分类器给出的标签获得具体已知设备，若一组数据中分类器给出相同标签6次以上，则判定结果为该标签对应的具体已知设备。
综上所述，本发明的一种基于机器学习算法的无线射频设备身份识别方法，通过软件无线电平台USRP和LabVIEW来发送和接收不同设备的射频信号，并将射频信号转换为图像信息，利用Python和Matlab软件负责训练学习图像信息并进行测试，通过分类器计算设备熵值大小，并引入门限比较机制来实现无线设备身份识别。通过大量的数据采集及训练，所述方法对已知设备的识别准确率达到98％以上，进一步采用阈值比较算法，可得未知设备的判别率达到90％以上。
与传统的设备识别方法相比，本发明的有益效果为：利用射频信号的图像信息可以有效地识别已知设备和未知设备，并能在已知设备类别中识别设备的具体身份，达到无线设备安全认证的目的，提高了无线通信系统的安全性。
综上所述，本发明利用射频信号的图像信息实现了基于电磁环境中的无线设备身份识别，相较于传统的设备识别方法，本发明的一种基于机器学习算法的 无线射频设备身份识别方法提高了通信系统的安全性。
相应的，本发明提供了一种基于机器学习算法的无线射频设备身份识别系统，其特征是，包括射频信号采集模块、信号转换模块、分类识别模块、熵值计算模块和最终结果识别模块；其中：
射频信号采集模块，用于采集待识别无线射频设备发送的射频信号；
信号转换模块，用于将射频信号的频谱图转换为图像；
分类识别模块，用于将图像输入预设的卷积神经网络进行分类，获得此无线射频设备的分类结果；
熵值计算模块，用于根据无线射频设备的分类结果，计算得到对应的熵值；
最终结果识别模块，用于根据无线射频设备对应的熵值，判断此设备的最终识别结果。
进一步的，信号转换模块中，所述图像大小为720*232像素。
进一步的，分类识别模块中，所述图像输入预设的卷积神经网络进行分类之前，对图像进行下采样。
进一步的，熵值计算模块中，所述根据无线射频设备的分类结果计算得到对应的熵值包括：
无线射频设备的分类结果为此无线射频设备属于第j(j＝1,2…N)类设备的概率，记为P j(j＝1,2…N)；
通过公式
Figure PCTCN2019126138-appb-000005
计算得到此无线射频设备对应的熵值。
进一步的，最终结果识别模块中，所述根据无线射频设备对应的熵值判断此设备的最终识别结果包括：
对无线射频设备进行多次分类，获得多个分类结果；
根据多个分类结果，计算得到对应的多个熵值；
根据多个熵值，计算得到对应的熵均值；
将熵均值与预设阈值进行比较判断，得到此设备的最终识别结果。
本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程 或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
以上所述仅是本发明的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本发明技术原理的前提下，还可以做出若干改进和变型，这些改进和变型也应视为本发明的保护范围。

Claims (10)

  1. 一种基于机器学习算法的无线射频设备身份识别方法，其特征是，包括以下过程：
    采集待识别无线射频设备发送的射频信号；
    将射频信号的频谱图转换为图像；
    将图像输入预设的卷积神经网络进行分类，获得此无线射频设备的分类结果；
    根据无线射频设备的分类结果，计算得到对应的熵值；
    根据无线射频设备对应的熵值，判断此设备的最终识别结果。
  2. 根据权利要求1所述的一种基于机器学习算法的无线射频设备身份识别方法，其特征是，所述图像大小为720*232像素。
  3. 根据权利要求1所述的一种基于机器学习算法的无线射频设备身份识别方法，其特征是，所述图像输入预设的卷积神经网络进行分类之前，对图像进行下采样。
  4. 根据权利要求1所述的一种基于机器学习算法的无线射频设备身份识别方法，其特征是，所述根据无线射频设备的分类结果计算得到对应的熵值包括：
    无线射频设备的分类结果为此无线射频设备属于第j(j＝1,2…N)类设备的概率，记为P j(j＝1,2…N)；
    通过公式
    Figure PCTCN2019126138-appb-100001
    计算得到此无线射频设备对应的熵值。
  5. 根据权利要求1所述的一种基于机器学习算法的无线射频设备身份识别方法，其特征是，所述根据无线射频设备对应的熵值判断此设备的最终识别结果包括：
    对无线射频设备进行多次分类，获得多个分类结果；
    根据多个分类结果，计算得到对应的多个熵值；
    根据多个熵值，计算得到对应的熵均值；
    将熵均值与预设阈值进行比较判断，得到此设备的最终识别结果。
  6. 一种基于机器学习算法的无线射频设备身份识别系统，其特征是，包括射频信号采集模块、信号转换模块、分类识别模块、熵值计算模块和最终结果识别模块；其中：
    射频信号采集模块，用于采集待识别无线射频设备发送的射频信号；
    信号转换模块，用于将射频信号的频谱图转换为图像；
    分类识别模块，用于将图像输入预设的卷积神经网络进行分类，获得此无线射频设备的分类结果；
    熵值计算模块，用于根据无线射频设备的分类结果，计算得到对应的熵值；
    最终结果识别模块，用于根据无线射频设备对应的熵值，判断此设备的最终识别结果。
  7. 根据权利要求6所述的一种基于机器学习算法的无线射频设备身份识别系统，其特征是，信号转换模块中，所述图像大小为720*232像素。
  8. 根据权利要求6所述的一种基于机器学习算法的无线射频设备身份识别系统，其特征是，分类识别模块中，所述图像输入预设的卷积神经网络进行之前，对图像进行下采样。
  9. 根据权利要求6所述的一种基于机器学习算法的无线射频设备身份识别系统，其特征是，熵值计算模块中，所述根据无线射频设备的分类结果计算得到对应的熵值包括：
    无线射频设备的分类结果为此无线射频设备属于第j(j＝1,2…N)类设备的概率，记为P j(j＝1,2…N)；
    通过公式
    Figure PCTCN2019126138-appb-100002
    计算得到此无线射频设备对应的熵值。
  10. 根据权利要求6所述的一种基于机器学习算法的无线射频设备身份识别系统，其特征是，最终结果识别模块中，所述根据无线射频设备对应的熵值判断此设备的最终识别结果包括：
    对无线射频设备进行多次分类，获得多个分类结果；
    根据多个分类结果，计算得到对应的多个熵值；
    根据多个熵值，计算得到对应的熵均值；
    将熵均值与预设阈值进行比较判断，得到此设备的最终识别结果。
PCT/CN2019/126138 2019-11-27 2019-12-18 基于机器学习算法的无线射频设备身份识别方法及系统 WO2021103206A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911177591.9A CN110647951A (zh) 2019-11-27 2019-11-27 基于机器学习算法的无线射频设备身份识别方法及系统
CN201911177591.9 2019-11-27

Publications (1)

Publication Number Publication Date
WO2021103206A1 true WO2021103206A1 (zh) 2021-06-03

Family

ID=69014694

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/126138 WO2021103206A1 (zh) 2019-11-27 2019-12-18 基于机器学习算法的无线射频设备身份识别方法及系统

Country Status (2)

Country Link
CN (1) CN110647951A (zh)
WO (1) WO2021103206A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116541693A (zh) * 2023-07-07 2023-08-04 河北省水利工程局集团有限公司 开放式施工环境人员检测方法、劳务人员在岗管理系统
CN117200932A (zh) * 2023-09-28 2023-12-08 中国南方电网有限责任公司超高压输电公司电力科研院 基于神经网络的无人机频射信号的识别方法和系统

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111310700B (zh) * 2020-02-27 2024-02-13 电子科技大学 一种用于辐射源指纹特征识别的中频采样序列处理方法
CN112528695B (zh) * 2020-12-18 2024-03-15 北京元心科技有限公司 识别方法、装置、设备及计算机可读存储介质
CN112911597B (zh) * 2021-03-24 2022-11-01 上海电机学院 基于射频信号精细画像的物联网物理层多级特征提取方法
CN114374975A (zh) * 2022-03-22 2022-04-19 南京邮电大学 一种可抗多径衰落的射频指纹识别方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101893704A (zh) * 2010-07-20 2010-11-24 哈尔滨工业大学 一种基于粗糙集的雷达辐射源信号识别方法
US20110248827A1 (en) * 2010-04-08 2011-10-13 The United States Government in the name of the Secretary of the Navy System and Method for Radio-Frequency Fingerprinting as a Security Layer in RFID Devices
CN102540164A (zh) * 2012-01-04 2012-07-04 中国电子科技集团公司第二十八研究所 一种基于增量式学习的雷达辐射源分类建模方法
CN110011932A (zh) * 2019-04-18 2019-07-12 清华大学深圳研究生院 一种可识别未知流量的网络流量分类方法和终端设备
CN110018447A (zh) * 2019-04-09 2019-07-16 电子科技大学 基于双谱分析与卷积神经网络的信号发射机个体识别方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110248827A1 (en) * 2010-04-08 2011-10-13 The United States Government in the name of the Secretary of the Navy System and Method for Radio-Frequency Fingerprinting as a Security Layer in RFID Devices
CN101893704A (zh) * 2010-07-20 2010-11-24 哈尔滨工业大学 一种基于粗糙集的雷达辐射源信号识别方法
CN102540164A (zh) * 2012-01-04 2012-07-04 中国电子科技集团公司第二十八研究所 一种基于增量式学习的雷达辐射源分类建模方法
CN110018447A (zh) * 2019-04-09 2019-07-16 电子科技大学 基于双谱分析与卷积神经网络的信号发射机个体识别方法
CN110011932A (zh) * 2019-04-18 2019-07-12 清华大学深圳研究生院 一种可识别未知流量的网络流量分类方法和终端设备

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116541693A (zh) * 2023-07-07 2023-08-04 河北省水利工程局集团有限公司 开放式施工环境人员检测方法、劳务人员在岗管理系统
CN116541693B (zh) * 2023-07-07 2023-09-12 河北省水利工程局集团有限公司 开放式施工环境人员检测方法、劳务人员在岗管理系统
CN117200932A (zh) * 2023-09-28 2023-12-08 中国南方电网有限责任公司超高压输电公司电力科研院 基于神经网络的无人机频射信号的识别方法和系统

Also Published As

Publication number Publication date
CN110647951A (zh) 2020-01-03

Similar Documents

Publication Publication Date Title
WO2021103206A1 (zh) 基于机器学习算法的无线射频设备身份识别方法及系统
CN110086737B (zh) 一种基于图神经网络的通信信号调制方式的识别方法
WO2018176889A1 (zh) 一种数字通信信号调制方式自动识别方法
CN111428817B (zh) 一种面向无线电信号识别对抗攻击的防御方法
CN111832417B (zh) 基于cnn-lstm模型和迁移学习的信号调制样式识别方法
CN112702294B (zh) 一种基于深度学习的多层次特征提取的调制识别方法
Oyedare et al. Estimating the required training dataset size for transmitter classification using deep learning
CN111144522B (zh) 一种基于硬件本征差异的电网nfc设备指纹认证方法
CN111160424B (zh) 一种基于cnn图像识别的nfc设备指纹认证方法及系统
CN112367273B (zh) 基于知识蒸馏的深度神经网络模型的流量分类方法及装置
CN111050315B (zh) 一种基于多核双路网络的无线发射机识别方法
Gong et al. Multi-task based deep learning approach for open-set wireless signal identification in ISM band
CN115294615A (zh) 一种基于机器学习的射频指纹识别方法
CN114157539A (zh) 一种数据知识双驱动的调制智能识别方法
CN114980122A (zh) 一种小样本射频指纹智能识别系统与方法
CN114867018A (zh) 利用载波频率偏移的设备识别认证方法、系统及存储介质
Wang et al. Specific emitter identification based on deep adversarial domain adaptation
CN113869156A (zh) 一种射频指纹识别方法
CN117131436A (zh) 面向开放环境的辐射源个体识别方法
CN116776227A (zh) 基于特征融合的卫星身份识别方法和装置
CN116738251A (zh) 一种基于生成对抗网络的射频指纹识别训练数据生成方法
CN110826425A (zh) 一种基于深度神经网络的vhf/uhf频段无线电信号调制方式识别方法
Xie et al. A novel method for few-shot specific emitter identification in non-cooperative scenarios
CN115809426A (zh) 一种辐射源个体识别方法及系统
CN109587136A (zh) 一种基于双极大值的射频指纹特征提取和识别方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19953739

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19953739

Country of ref document: EP

Kind code of ref document: A1