A device identification method based on multidimensional statistical feature attributes

Abstract

The present invention discloses a device identification method based on multidimensional statistical feature attributes, which includes: the identification platform extracts multidimensional features from all radio frequency fingerprint original signal samples in radio frequency fingerprint original signal sample set A and radio frequency fingerprint original signal sample set B respectively and merges the extracted multidimensional features into radio frequency fingerprint feature vectors, and all obtained radio frequency fingerprint feature vectors constitute the corresponding radio frequency fingerprint feature sample set A′ and radio frequency fingerprint feature sample set B′ respectively; the training and testing of the classifier are completed according to the training set T1 and the test set T2 until the classifier meets the set recognition probability; the identification platform uses the classifier with the recognition probability reaching the standard to judge the identity type of the wireless transmitting device with unknown identity information. The present invention makes full use of the statistical characteristics of the complete signal in the time domain, uses the recognition algorithm to construct the classifier for training and recognition, thereby improving the accuracy of the identity recognition of the wireless transmitting device.

Description

Technical Field

[0001] The present invention relates to the technical field of identification of wireless communication equipment, and in particular to a method for identifying equipment based on multi-dimensional statistical feature attributes. Background Art

[0002] With the rapid development of 5G wireless networks, intelligent and ubiquitous connections are gradually being realized. However, wireless networks are broadcast and open, which makes the network vulnerable to eavesdropping and interference. The influx of a large number of micro terminals makes it difficult for traditional security solutions to meet the security requirements of low power consumption, low resources and high throughput. Radio frequency fingerprint recognition technology avoids the complex encryption calculations based on cryptography in the upper layer of the network, makes full use of the physical fingerprint characteristics of the device, and extracts the subtle radiation source characteristics caused by the hardware differences of wireless devices, thereby realizing low-complexity and lightweight device identification. Due to the difficulty of cloning radio frequency fingerprints, the device identity recognition mechanism established by it meets the lightweight access security identification of massive wireless terminals. However, how to construct an efficient and noise-resistant radio frequency fingerprint feature vector is an urgent problem to be solved. Summary of the invention

[0003] In view of this, the present invention provides a device identification method based on multidimensional statistical feature attributes, which makes full use of the statistical characteristics of the complete signal in the time domain, constructs anti-noise fingerprint features of the multidimensional statistical feature attributes, and then uses the recognition algorithm to build a classifier for training and recognition, thereby improving the accuracy of identity recognition of wireless transmitting devices.

[0004] The present invention discloses a device identification method based on multi-dimensional statistical feature attributes, which comprises:

[0005] Step 1: The identification platform collects radio frequency fingerprint signals from the signals transmitted by legal wireless devices and illegal wireless devices, and obtains radio frequency fingerprint original signal sample set A and radio frequency fingerprint original signal sample set B respectively;

[0006] Step 2: The recognition platform extracts multi-dimensional features from all RF fingerprint original signal samples in RF fingerprint original signal sample set A and RF fingerprint original signal sample set B respectively and merges the extracted multi-dimensional features into RF fingerprint feature vectors. All the obtained RF fingerprint feature vectors constitute the corresponding RF fingerprint feature sample set A′ and RF fingerprint feature sample set B′ respectively.

[0007] Step 3: The recognition platform divides the radio frequency fingerprint feature sample set A′ and the radio frequency fingerprint feature sample set B′ respectively to obtain a training set T1 and a test set T2;

[0008] Step 4: The recognition platform extracts a classifier using a recognition algorithm, and completes the training and testing of the classifier according to the training set T1 and the test set T2 until the classifier meets the set recognition probability; the recognition algorithm includes a fully connected neural network and a machine learning classification algorithm;

[0009] Step 5: The identification platform uses a classifier with a satisfactory identification probability to determine the identity type of the wireless transmitting device with unknown identity information.

[0010] Furthermore, the step 1 comprises:

[0011] Step 101: The legitimate wireless device continuously transmits radio frequency signals. The identification platform uses the receiving antenna to collect the original radio frequency fingerprint signal at the Nyquist sampling rate, and stores the collected signal into the register at time series intervals to obtain the radio frequency fingerprint original signal sample set A = [A1, A2, ..., A k ,…,A N ], where A k A represents the original signal sample of the RF fingerprint of the kth legal wireless device, where the value of k ranges from 1 to N; k =[x(1),x(2),…,x(i),…,x(n)], where x(i) represents the A corresponding to the i-th sampling point k The amplitude of i ranges from 1 to n;

[0012] Step 102: The legitimate wireless device stops transmitting RF signals to the identification platform, and the illegal wireless device starts to continuously transmit RF signals to the identification platform. The identification platform uses the receiving antenna to collect RF fingerprint signals at the Nyquist sampling rate, and stores the collected signals in the register according to the time series interval, and obtains the RF fingerprint original signal sample set B = [B1, B2, ..., B k ,…,B N ], where B k represents the original signal sample of the RF fingerprint of the kth illegal wireless device, where the value of k ranges from 1 to N; B k =[x'(1),x'(2),…,x'(i),…,x'(n)], where x'(i) represents the B corresponding to the i-th sampling point k The value of i ranges from 1 to n.

[0013] Furthermore, the step 2 comprises:

[0014] Take any RF fingerprint original signal sample in the RF fingerprint original signal sample set A and the RF fingerprint original signal sample set B, record it as X = [x[1], x[2], …x[i], …x[n]], extract the noise-resistant RF fingerprint features of X and combine them into a RF fingerprint feature vector of a single device and use it as the characteristic attribute of the RF fingerprint, complete the RF fingerprint feature vector extraction of all RF fingerprint original signal samples, and obtain the RF fingerprint feature sample set A′ and the RF fingerprint feature sample set B′ respectively.

[0015] Furthermore, the anti-noise radio frequency fingerprint features of X include:

[0016] The maximum value x in X max ;

[0017]

[0018] The minimum value x in X min ;

[0019]

[0020] The difference between the maximum and minimum values ​​in X p ;

[0021]

[0022] The average value of the absolute value of X arv ;

[0023]

[0024] The mean value μ of X;

[0025]

[0026] The first moment statistic σ of X 2 ;

[0027]

[0028] The square root of X, σ;

[0029]

[0030] The effective value of Xx RMS ;

[0031]

[0032] The ratio of the peak value of X to the RMS value k a ;

[0033]

[0034] The peak value of X p The average value of the absolute value x arv The average k p ;

[0035]

[0036] The peak value of X p and the square root amplitude x r The ratio k f ;

[0037]

[0038] The ratio of the fourth central moment of X to the fourth power of the standard deviation K;

[0039]

[0040] The effective value of Xx RMS With the rectified mean x arv The ratio k a ;

[0041]

[0042] The degree of asymmetry r of the probability distribution density curve of X relative to the mean;

[0043]

[0044] Furthermore, the radio frequency fingerprint feature vector of the single device is x=(x max ,x min ,...,r) T

[0046] Furthermore, x p As the peak value; arv As the rectified average, the signal is rectified, the negative values ​​are flipped, summed, and then averaged; μ is used as the first-order moment statistic of the signal; k a As the peak factor, it is used to characterize the extreme degree of the peak in the waveform; k p As the impulse factor, it is used to characterize whether there is an impact in the signal; k f As a margin factor; K is used as a kurtosis factor to statistically describe the flatness of the signal waveform and characterize the convexity of the peak of X; k a As the waveform factor; let r be the ratio of the third central moment of X to the cube of the standard deviation.

[0047] Furthermore, the step 3 comprises:

[0048] The recognition platform divides part of the samples in the radio frequency fingerprint feature sample set A′ into the training set T1, and divides the rest of the samples in the radio frequency fingerprint feature sample set A′ into the test set T2;

[0049] The recognition platform divides part of the samples in the radio frequency fingerprint feature sample set B′ into the training set T1, and divides the remaining samples in the radio frequency fingerprint feature sample set B′ into the test set T2, and finally obtains the training set T1 and the test set T2.

[0050] Furthermore, the step 4 comprises:

[0051] The recognition platform marks the identity types of the samples from the radio frequency fingerprint feature sample set A′ and the radio frequency fingerprint feature sample set B′ in the training set T1, and extracts the classifier using the recognition algorithm. The classifier is trained and tested according to the training set T1 and the test set T2 until the classifier meets the set recognition probability. When the classifier is trained, the training set T1 is used as the input of the classifier, and the output is the real identity type label. When the classifier is tested, the test set T2 is used as the input of the classifier, and the expected output of the classifier is the identity ID number.

[0052] Furthermore, the step 5 comprises:

[0053] Collect the radio frequency fingerprint signal of the unknown wireless device, process it according to the method of steps 1 and 2, obtain the radio frequency fingerprint feature vector, and use the classifier with the recognition rate reaching the standard to judge the radio frequency fingerprint feature vector, so as to realize the legal identity judgment of the wireless transmitting device.

[0054] Due to the adoption of the above technical solution, the present invention has the following advantages:

[0055] (1) By analyzing the statistical characteristics of the complete signal in the time domain, dimensionless and dimensional statistical features are extracted respectively, thereby obtaining a lightweight multi-dimensional statistical feature attribute RF fingerprint feature vector, thereby improving the accuracy of wireless transmitter identification;

[0056] (2) The RF signal fingerprint recognition technology is used to reduce the complexity and computational overhead of the wireless device itself. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0058] Figure 1 A flow chart of a device identification method based on multi-dimensional statistical feature attributes;

[0059] Figure 2 The present invention is a flowchart of another device identification method based on multi-dimensional statistical feature attributes. DETAILED DESCRIPTION

[0060] The present invention is further described in conjunction with the accompanying drawings and embodiments, and the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field should fall within the scope of protection of the embodiments of the present invention.

[0061] like Figure 1 As shown, the present invention provides an embodiment of a device identification method based on multi-dimensional statistical feature attributes, comprising the following steps:

[0062] S1. The identification platform collects radio frequency fingerprint signals from legal and illegal wireless devices:

[0063] Identify legitimate wireless devices to collect RF fingerprint signals, and obtain RF fingerprint original signal sample set A at the same time interval; similarly, the identification platform collects RF fingerprint signals from simulated illegal wireless devices, and obtains RF fingerprint original signal sample set B at the same time interval;

[0064] S2. The recognition platform extracts radio frequency signal features from the radio frequency fingerprint original signal sample sets A and B:

[0065] Each sample in the original signal sample set A and B is transformed in time-frequency mode, and then sent to the convolutional neural network for processing to generate multi-dimensional statistical RF fingerprint feature sample sets A′ and B′;

[0066] S3. The recognition platform divides the RF fingerprint feature sample sets A′ and B′ into training set T1 and test set T2;

[0067] S4. The recognition platform extracts the classifier using the recognition algorithm, and completes the training and testing of the classifier according to the training set T1 and the test set T2 until the classifier meets the recognition probability;

[0068] S5. The identification platform uses a classifier with a satisfactory recognition rate to identify wireless transmitters of unknown identity information to achieve legal identity judgment.

[0069] In an embodiment of the present application, an edge computing device is used as an identification platform to perform radio frequency fingerprint feature identification of wireless devices.

[0070] Wherein, the step S1 includes the following sub-steps:

[0071] S101. The legitimate wireless device continuously transmits radio frequency signals. The identification platform uses the receiving antenna to collect the original radio frequency fingerprint signal at the Nyquist sampling rate, and stores the signal into the register at time series intervals to obtain the radio frequency fingerprint original signal sample set A=[A1,A2,…,A k ,…,A N ], where A k A represents the original signal sample of the RF fingerprint of the kth legal wireless device, where the value of k ranges from 1 to N; k =[x(1),x(2),…,x(i),…,x(n)], where x(i) represents the A corresponding to the i-th sampling point k The amplitude of i ranges from 1 to n;

[0072] S102. The legitimate wireless device stops transmitting RF signals to the identification platform, and the illegal wireless device starts to continuously transmit RF signals to the identification platform. The identification platform uses the receiving antenna to collect RF fingerprint signals at the Nyquist sampling rate and stores the signals in the register according to the time series interval, and obtains the RF fingerprint original signal sample set B of the illegal wireless device = [B1, B2, ..., B k ,…,B N ], where B k represents the original signal sample of the RF fingerprint of the kth illegal wireless device, where the value of k ranges from 1 to N; B k =[x'(1),x'(2),…,x'(i),…,x'(n)], where x'(i) represents the B corresponding to the i-th sampling point k The value of i ranges from 1 to n.

[0073] The step S2 comprises the following sub-steps:

[0074] S201. For any RF fingerprint signal X in the RF fingerprint original signal sample set A and B, perform time domain analysis and form statistical features, extract the maximum signal x max ;

[0075]

[0076] S202. Extract the minimum value x in the complete sample signal min ;

[0077]

[0078] S203. Extract the difference x between the maximum and minimum values ​​in the complete sample signal p as the peak-to-peak value;

[0079]

[0080] S204. Extract the average value x of the absolute value of the complete sample signal arv As a rectified average, it can be understood as rectifying the signal, flipping the negative values, summing them, and then averaging them;

[0081]

[0082] S205. Extract the average value μ of the complete sample signal as the first-order moment statistic of the signal;

[0083]

[0084] S206. Extract the first-order moment statistic σ of the complete sample signal 2 ;

[0085]

[0086] S207. Extract the dynamic component of the complete sample signal energy and calculate the square root σ:

[0087]

[0088] S208. Extract the effective value x of the statistically complete sample signal RMS

[0089]

[0090] S209. Extract the ratio k of the peak value and the root mean square value of the complete sample signal a As a peak factor, it can characterize the extreme degree of the peak in the waveform;

[0091]

[0092] S210. Extract the peak value of the complete sample signal and the average value k of the absolute value of the signal p As an impulse factor, it can characterize whether there is an impact in the signal.

[0093]

[0094] S211. Extract the statistical complete signal peak value and square root amplitude value ratio k f as a margin factor.

[0095]

[0096] S212. Extract the ratio K of the fourth-order central moment of the signal and the fourth power of the standard deviation as the kurtosis factor, which can statistically describe the flatness of the signal waveform and characterize the convexity of the signal peak.

[0097]

[0098] S213. Extract the ratio k of the effective value and the rectified average value a as the shape factor.

[0099]

[0100] S214. Extract the asymmetry degree r of the probability distribution density curve of the statistical signal relative to the mean value as the ratio of the third-order central moment of the signal to the cube of the standard deviation.

[0101]

[0102] S215. Combining the 14 complete signal statistical features in the time domain into a single device RF fingerprint feature vector is x=(x max ,x min ,...,r) T As the characteristic attribute of radio frequency fingerprint.

[0103] Note: Since the statistical characteristics of the signal can reflect the properties of the signal itself and describe the overall trend and distribution, and noise usually does not represent the main trend of the data, and random noise is difficult to change the overall distribution characteristics of the original signal, the statistical characteristics can resist the interference of noise to a certain extent.

[0104] S216. Repeat steps S201 to S215 for each RF fingerprint signal in sample set A, complete the RF signal fingerprint vector extraction of all RF fingerprint signals, and obtain the RF fingerprint feature sample set A′;

[0105] Repeat steps S201 to S215 for each RF fingerprint signal in the sample set B to complete the extraction of RF signal fingerprint vectors of all RF fingerprint signals, and obtain the RF fingerprint feature sample set B′.

[0106] In the embodiment of the present application, in the process of extracting the features of the noise-resistant radio frequency fingerprint of each multi-dimensional statistical feature attribute in step S2, the combination mode can be determined according to actual needs, and can be selected as 2 to 14 dimensions.

[0107] In step S4, a part of the samples in the radio frequency fingerprint feature sample set A′ is divided into the training set T1, and another part of the samples is divided into the test set T2. Similarly, a part of the samples in the radio frequency fingerprint feature sample set B′ is divided into the training set T1, and another part of the samples is divided into the test set T2; the method of dividing the training set and the test set includes but is not limited to the cross-validation method, the hold-out method and the bootstrap method.

[0108] In an embodiment of the present application, step S4 includes:

[0109] The recognition platform performs identity type labeling (identity ID) on the samples from sample set A′ and sample set B′ in the training set, and extracts the classifier using the recognition algorithm, completes the training and testing of the classifier according to the training set T1 and the test set T2, until the classifier meets the recognition probability; the recognition algorithm includes but is not limited to fully connected neural networks and machine learning classification algorithms. Note: During training, T1 is used as the model input, and the output is the real identity type label (identity ID). During testing, T2 is used as the model input, and the model expects to output the judged identity ID number. Machine learning classification algorithms include SVM, clustering, decision tree, naive Bayes, and Softmax regression algorithms.

[0110] In the examples of this application, see Figure 2 The step S5 includes: collecting the radio frequency fingerprint signal of the unknown wireless device, performing statistical characteristic analysis on it according to the processing method of the radio frequency fingerprint signal X in steps S201 to S215, and extracting the anti-noise radio frequency signal fingerprint vector of the multi-dimensional statistical characteristic attribute, and using the classifier with the recognition rate reaching the standard to judge the extracted radio frequency signal fingerprint vector, so as to realize the legitimacy identity judgment of the wireless transmitting device.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A device identification method based on multi-dimensional statistical feature attributes, characterized in that: include: Step 1: The identification platform collects radio frequency fingerprint signals from the signals transmitted by legal wireless devices and illegal wireless devices, and obtains radio frequency fingerprint original signal sample set A and radio frequency fingerprint original signal sample set B respectively; Step 2: The recognition platform extracts multi-dimensional features from all RF fingerprint original signal samples in RF fingerprint original signal sample set A and RF fingerprint original signal sample set B respectively and merges the extracted multi-dimensional features into RF fingerprint feature vectors. All the obtained RF fingerprint feature vectors constitute the corresponding RF fingerprint feature sample set A′ and RF fingerprint feature sample set B′ respectively. Step 3: The recognition platform divides the radio frequency fingerprint feature sample set A′ and the radio frequency fingerprint feature sample set B′ respectively to obtain a training set T1 and a test set T2; Step 4: The recognition platform extracts a classifier using a recognition algorithm, and completes the training and testing of the classifier according to the training set T1 and the test set T2 until the classifier meets the set recognition probability; the recognition algorithm includes a fully connected neural network and a machine learning classification algorithm; Step 5: The identification platform uses a classifier with a satisfactory identification probability to determine the identity type of the wireless transmitting device with unknown identity information.

2. The method according to claim 1, characterized in that The step 1 comprises: Step 101: The legitimate wireless device continuously transmits radio frequency signals. The identification platform uses the receiving antenna to collect the original radio frequency fingerprint signal at the Nyquist sampling rate, and stores the collected signal into the register at time series intervals to obtain the radio frequency fingerprint original signal sample set A = [A1, A2, ..., A k ,…,A N ], where A k A represents the original signal sample of the RF fingerprint of the kth legal wireless device, where the value of k ranges from 1 to N; k =[x(1),x(2),…,x(i),…,x(n)], where x(i) represents the A corresponding to the i-th sampling point k The amplitude of i ranges from 1 to n; Step 102: The legitimate wireless device stops transmitting RF signals to the identification platform, and the illegal wireless device starts to continuously transmit RF signals to the identification platform. The identification platform uses the receiving antenna to collect RF fingerprint signals at the Nyquist sampling rate, and stores the collected signals in the register according to the time series interval, and obtains the RF fingerprint original signal sample set B = [B1, B2, ..., B k ,…,B N ], where B k represents the original signal sample of the RF fingerprint of the kth illegal wireless device, where the value of k ranges from 1 to N; B k =[x'(1),x'(2),…,x'(i),…,x'(n)], where x'(i) represents the B corresponding to the i-th sampling point k The value of i ranges from 1 to n.

3. The method according to claim 1, characterized in that The step 2 comprises: Take any RF fingerprint original signal sample in the RF fingerprint original signal sample set A and the RF fingerprint original signal sample set B, record it as X = [x[1], x[2], …x[i], …x[n]], extract the noise-resistant RF fingerprint features of X and combine them into a RF fingerprint feature vector of a single device and use it as the characteristic attribute of the RF fingerprint, complete the RF fingerprint feature vector extraction of all RF fingerprint original signal samples, and obtain the RF fingerprint feature sample set A′ and the RF fingerprint feature sample set B′ respectively.

4. The method according to claim 3, characterized in that The noise-resistant RF fingerprint features of X include: The maximum value x in X max ; The minimum value x in X min ; The difference between the maximum and minimum values ​​in X p ; The average value of the absolute values ​​of X arv ; The mean value μ of X; The first moment statistic σ of X 2 ; The square root of X, σ; The effective value of Xx RMS ; The ratio of the peak value of X to the RMS value k a ; The peak value of X p The average value of the absolute value x arv The average k p ; The peak value of X p and the square root amplitude x r The ratio k f ; The ratio of the fourth central moment of X to the fourth power of the standard deviation K; The effective value of Xx RMS With the rectified mean x arv The ratio k a ; The degree of asymmetry r of the probability distribution density curve of X relative to the mean; 5. The method according to claim 4, characterized in that The radio frequency fingerprint feature vector of a single device is x=(x max ,x min ,...,r) T .

6. The method according to claim 4, characterized in that x p As the peak value; arv As the rectified average, the signal is rectified, the negative values ​​are flipped, summed, and then averaged; μ is used as the first-order moment statistic of the signal; k a As the peak factor, it is used to characterize the extreme degree of the peak in the waveform; k p As the impulse factor, it is used to characterize whether there is an impact in the signal; k f As a margin factor; K is used as a kurtosis factor to statistically describe the flatness of the signal waveform and characterize the convexity of the peak of X; k a As the waveform factor; let r be the ratio of the third central moment of X to the cube of the standard deviation.

7. The method according to claim 1, characterized in that The step 3 comprises: The recognition platform divides part of the samples in the radio frequency fingerprint feature sample set A′ into the training set T1, and divides the rest of the samples in the radio frequency fingerprint feature sample set A′ into the test set T2; The recognition platform divides part of the samples in the radio frequency fingerprint feature sample set B′ into the training set T1, and divides the remaining samples in the radio frequency fingerprint feature sample set B′ into the test set T2, and finally obtains the training set T1 and the test set T2.

8. The method according to claim 1, characterized in that The step 4 comprises: The recognition platform marks the identity types of the samples from the radio frequency fingerprint feature sample set A′ and the radio frequency fingerprint feature sample set B′ in the training set T1, and extracts the classifier using the recognition algorithm. The classifier is trained and tested according to the training set T1 and the test set T2 until the classifier meets the set recognition probability. When the classifier is trained, the training set T1 is used as the input of the classifier, and the output is the real identity type label. When the classifier is tested, the test set T2 is used as the input of the classifier, and the expected output of the classifier is the identity ID number.

9. The method according to claim 1, characterized in that: The step 5 comprises: Collect the radio frequency fingerprint signal of the unknown wireless device, process it according to the method of steps 1 and 2, obtain the radio frequency fingerprint feature vector, and use the classifier with the recognition rate reaching the standard to judge the radio frequency fingerprint feature vector, so as to realize the legal identity judgment of the wireless transmitting device.

Images