CN114255487A - Internet of things equipment identity authentication method for semi-supervised radio frequency fingerprint extraction - Google Patents

Internet of things equipment identity authentication method for semi-supervised radio frequency fingerprint extraction Download PDF

Info

Publication number
CN114255487A
CN114255487A CN202111576682.7A CN202111576682A CN114255487A CN 114255487 A CN114255487 A CN 114255487A CN 202111576682 A CN202111576682 A CN 202111576682A CN 114255487 A CN114255487 A CN 114255487A
Authority
CN
China
Prior art keywords
signal
semi
radio frequency
internet
supervised
Prior art date
Legal status (The legal status 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 status listed.)
Pending
Application number
CN202111576682.7A
Other languages
Chinese (zh)
Inventor
童琳
李科
孙雨
汪佳
刘永一
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Anhui Agricultural University AHAU
Original Assignee
Anhui Agricultural University AHAU
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 Anhui Agricultural University AHAU filed Critical Anhui Agricultural University AHAU
Priority to CN202111576682.7A priority Critical patent/CN114255487A/en
Publication of CN114255487A publication Critical patent/CN114255487A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Abstract

The invention provides an Internet of things equipment identity authentication method for semi-supervised radio frequency fingerprint extraction, which comprises the following specific steps: receiving a signal from a communication device, and extracting features from the radio frequency signal; decomposing the received signal into a specified number of patterns using a VMD algorithm; then, reducing the dimension of the radio frequency fingerprint by using a semi-supervised dimension reduction method based on low-rank linear regression analysis; and finally, constructing a classification model by using a LightGBM algorithm. On the basis of the traditional radio frequency fingerprint extraction and identification method, the noise-containing signal is denoised by using a VMD algorithm, and semi-supervised learning and dimension reduction are realized on the radio frequency fingerprint by using a method based on low-rank linear regression analysis; and the LightGBM algorithm is used for training the acquired data, so that the training speed is high, the accuracy is high, and the identification precision of the equipment of the Internet of things is improved.

Description

Internet of things equipment identity authentication method for semi-supervised radio frequency fingerprint extraction
Technical Field
The invention relates to the technical field of communication, in particular to an identity authentication method for equipment of the Internet of things by semi-supervised radio frequency fingerprint extraction.
Background
Because the signal extraction of the equipment of the Internet of things is more complex, the equipment is very challenging; the traditional algorithm EMD is lack of a complete theoretical basis; in practice, a large number of unmarked samples are easy to obtain, and the number of marked samples is limited, so that the traditional supervised and unsupervised feature dimension reduction method cannot achieve good effect.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides an identity authentication method for equipment of the Internet of things by semi-supervised radio frequency fingerprint extraction.
The invention is realized by the following technical scheme:
an identity authentication method for Internet of things equipment for semi-supervised radio frequency fingerprint extraction comprises the following steps:
A. receiving a signal from the Internet of things equipment by using a communication signal receiver, carrying out frequency conversion on a signal spectrum to a proper frequency by using the receiver, and obtaining a digital signal by using analog-to-digital conversion and filtering operation;
B. decomposing a received signal into a specified number of modes by using a VMD-based method, determining a high-noise component by using an array entropy, and reconstructing a low-noise component to realize a noise reduction function, thereby realizing the extraction of nonlinear dynamic characteristics of the signal;
C. reducing the dimension of the radio frequency fingerprint by using a semi-supervised dimension reduction method based on low-rank linear regression analysis;
D. constructing a classification model by using a LightGBM algorithm, and training the acquired data;
E. the simplified features are clustered to identify the Internet of things equipment.
Preferably, after the communication signal in the step a receives a signal from the internet of things device, the frequency spectrum of the signal is converted to an appropriate frequency through a radio frequency filter device, so that the interference quotation mark is attenuated, and then the digital signal is obtained by analog-to-digital conversion and filtering operation.
Preferably, the step B uses a VMD and low rank regression analysis based method, the VMD decomposition adaptively decomposes the original signal into a plurality of Intrinsic Mode Functions (IMFs), selects a noise component according to an autocorrelation criterion, and then implements an optimal estimation of the transient electromagnetic signal using an independent component analysis algorithm based on kurtosis maximization.
(1) The multi-modal decomposition process of the VMD algorithm on the signal is as follows:
the VMD transfers the acquisition process of the signal components into a variation frame, adopts a non-recursive processing strategy, and realizes the decomposition of the original signal into IMF components with the designated number by constructing and solving a constraint variation problem; in the VMD algorithm, the intrinsic mode function IMF is redefined as an amplitude-frequency modulated (AM-FM) signal based on the modulation criteria, i.e.
uk(t)＝AK(t)cos(φk(t))(1) (1)
φk(t) is ukPhase of (t) & phi'k(t)≥0，AK(t) is uk(t) instantaneous amplitude at time t and AK(t) is not less than 0; and the instantaneous amplitude uk(t) the amplitude envelope A and instantaneous angular frequency B vary slowly compared to each other, i.e. over a sufficiently long time range t-delta, t + delta]Within, in the middle of them, (delta ≈ 2 π/φ'k(t))，uk(t) is approximately one having an amplitude of AK(t) and a frequency of wk(t) harmonic signals.
Continuously updating the frequency center and the bandwidth of each component in the iterative solution process of the variational model; finally, carrying out self-adaptive segmentation on the signal frequency band according to the frequency characteristics of the signal to obtain a plurality of narrow-band IMF components; assuming that the VMD decomposes the original signal into K IMF components, the corresponding constrained variational model expression is as follows:
Figure BDA0003425410870000021
{uk}＝{u1,u2,...,ukk IMF components for VMD method decomposition, { wk}＝{w1,w2,...,wkIs ukThe center of frequency of each IMF assembly,
Figure BDA0003425410870000022
is the partial derivative of function time, δ (t) is the unit pulse function, j is the imaginary unit, and is the convolution method.
Adaptively decomposing the mixed signal into relatively pure IMF components through the decomposition process; after VMD decomposition is carried out on the signals, IMF components with a limited number of different time scales are obtained, and the frequencies of the IMF components are arranged from low to high; the sharp part or noise of the signal corresponds to the large order IMF and the low frequency part of the signal corresponds to the small order IMF.
(2) The method based on independent component analysis is as follows:
the normalized autocorrelation function of a group of IMF components obtained after VMD decomposition of an original signal is a group of random sequences, the energy concentration degree of a discrete signal sequence in the area near the zero point is determined by calculating the energy concentration ratio of the area near the zero point, and the noise content degree of the IMF components is further determined; and reconstructing the noise domain into an imaginary noise component, and taking the imaginary noise component and the source signal as an input matrix of an independent component analysis algorithm, thereby obtaining the optimal estimation of the source signal.
Preferably, the semi-supervised dimension reduction method of introducing low rank regression analysis in the step C can more effectively mine the unmarked samples.
Will be respectively taken as Xl＝[xl,...,xl]TAs a sample set of labels, Xu＝[xl+1,...,xN]Is an unlabeled sample set; in addition to this, the present invention is,
Figure BDA0003425410870000031
wherein A isKIs a sample set of class k; from the above symbols, the matrix XlAnd XuIs a sample point and each column of the matrix is a sample point.
We can use independent component analysis to explore the relationship between these sample sets:
Figure BDA0003425410870000032
Figure BDA0003425410870000033
the matrix T represents the relationship between a set of labeled samples and a set of unlabeled samples, and S represents the relationship between unlabeled samples; the h and j columns of the matrix represent the ith unlabeled sample Xl+iJth unlabeled sample Xl+jThe relationship between; firstly, preprocessing a matrix S, normalizing each line of the S, calculating the absolute value of each element, and then recording a result matrix as R;
second, we define the following tag propagation:
Lk+1＝βRLk+(1-β)L0,k＝1,L (5)
wherein β is a parameter that balances these two terms; l is0Abs (t) L, L is a matrix containing label information of the marked sample;
finally, the labels of the unlabeled samples are predicted in such a way that if the k-th component in line L X i is the largest in its i-th line, we assume that the i-th unlabeled sample Xl+iBelonging to class k, high dimensional data is projected to the desired dimension.
The semi-supervised method based on low-rank linear regression can effectively extract linear structures in a data set, and contributes to dimension reduction of extracted signal features by using unmarked data.
Preferably, in the step D, the features after dimension reduction are divided into a training set and a test set, and a supervised training set is given
Figure BDA0003425410870000034
Finding a certain function f using LightGBM*(x) Is approximated by
Figure BDA0003425410870000035
Multiple training runs minimize the expected value of a particular loss function L (y, f (x)), as follows:
Figure BDA0003425410870000036
LightGBM (LightGBM) decision tree algorithm based on Histogram, and integrates a plurality of T regression trees
Figure BDA0003425410870000037
To approximate the final model, i.e.
Figure BDA0003425410870000041
The regression tree may be represented as wq(x)Q ∈ {1, 2., J }, where J denotes the number of leaves, q denotes the decision rule of the tree, and w denotes the sample weight of the leaf node.
Therefore, the LightGBM will be trained in additive form, as follows:
Figure BDA0003425410870000042
in LightGBM, the objective function is quickly approximated by newton's method, and then for simplicity, using the Leaf growth strategy of Leaf-wise with depth constraint, after removing the constant terms in (7), the formula can be transformed into:
Figure BDA0003425410870000043
wherein g isiAnd hiFirst and second order gradient statistics representing a loss function; let Ij denote the sample set j of the leaf and can be converted into:
Figure BDA0003425410870000044
for a particular tree structure q (x), the optimal leaf weight score for each leaf node
Figure BDA0003425410870000045
And extreme value ΓKThe following can be solved:
Figure BDA0003425410870000046
inputting the data of the test set into a LightGBM model to obtain the accuracy of the test set; the LightGBM model has good effect when using the training set; the advantages become more and more apparent as the sample size increases.
Preferably, in the step E, the features of the obtained reduced-dimension signal are input into the trained LightGBM model to realize cluster recognition, and the results of model recognition and classification are obtained, which have high conformity with the actual situation.
Compared with the prior art, the invention has the following advantages:
a radio frequency fingerprint (RF-fingerprint) authentication model of the communication equipment accurately and effectively identifies a signal radiation source of the communication equipment by extracting a radio frequency fingerprint of a communication signal, analyzes the signal and classifies the characteristics of the signal through a machine learning method.
The radio frequency fingerprint identification technology can be applied to fault detection of equipment and management of wireless spectrum, and if abnormal frequency enters, illegal invasion is detected by the technology, so that the safety of the Internet of things is improved.
Drawings
FIG. 1 is a flowchart of the steps of a method for identity authentication of Internet of things equipment by semi-supervised radio frequency fingerprint extraction;
fig. 2 is a decision tree algorithm block diagram based on Histogram in LightGBM of an internet of things device identity authentication method for semi-supervised radio frequency fingerprint extraction;
fig. 3 is a block diagram of a Leaf growth strategy of Leaf-wise with depth limitation in LightGBM of an internet of things device identity authentication method for semi-supervised radio frequency fingerprint extraction.
Detailed Description
The following examples are given for the detailed implementation and specific operation of the present invention, but the scope of the present invention is not limited to the following examples.
As shown in fig. 1 to 3, the method for authenticating the identity of the internet of things device by extracting the semi-supervised radio frequency fingerprint provided by the embodiment includes the following steps:
A. receiving a signal from the Internet of things equipment by using a communication signal receiver, carrying out frequency conversion on a signal spectrum to a proper frequency by using the receiver, and obtaining a digital signal by using analog-to-digital conversion and filtering operation;
B. decomposing a received signal into a specified number of modes by using a VMD-based method, determining a high-noise component by using an array entropy, and reconstructing a low-noise component to realize a noise reduction function, thereby realizing the extraction of nonlinear dynamic characteristics of the signal;
C. reducing the dimension of the radio frequency fingerprint by using a semi-supervised dimension reduction method based on low-rank linear regression analysis;
D. constructing a classification model by using a LightGBM algorithm, and training the acquired data;
E. the simplified features are clustered to identify the Internet of things equipment.
Preferably, after the communication signal in the step a receives a signal from the internet of things device, the frequency spectrum of the signal is converted to an appropriate frequency through a radio frequency filter device, so that the interference quotation mark is attenuated, and then the digital signal is obtained by analog-to-digital conversion and filtering operation.
Preferably, the step B uses a VMD and low rank regression analysis based method, the VMD decomposition adaptively decomposes the original signal into a plurality of Intrinsic Mode Functions (IMFs), selects a noise component according to an autocorrelation criterion, and then implements an optimal estimation of the transient electromagnetic signal using an independent component analysis algorithm based on kurtosis maximization.
(1) The multi-modal decomposition process of the VMD algorithm on the signal is as follows:
the VMD transfers the acquisition process of the signal components into a variation frame, adopts a non-recursive processing strategy, and realizes the decomposition of the original signal into IMF components with the designated number by constructing and solving a constraint variation problem; in the VMD algorithm, the intrinsic mode function IMF is redefined as an amplitude-frequency modulated (AM-FM) signal based on the modulation criteria, i.e.
uk(t)＝AK(t)cos(φk(t))(1) (1)
φk(t) is ukPhase of (t) & phi'k(t)≥0，AK(t) is uk(t) instantaneous amplitude at time t and AK(t) is not less than 0; and the instantaneous amplitude uk(t) the amplitude envelope A and instantaneous angular frequency B vary slowly compared to each other, i.e. over a sufficiently long time range t-delta, t + delta]Within, in the middle of them, (delta ≈ 2 π/φ'k(t))，uk(t) is approximately one having an amplitude of AK(t) and a frequency of wk(t) harmonic signals.
Continuously updating the frequency center and the bandwidth of each component in the iterative solution process of the variational model; finally, carrying out self-adaptive segmentation on the signal frequency band according to the frequency characteristics of the signal to obtain a plurality of narrow-band IMF components; assuming that the VMD decomposes the original signal into K IMF components, the corresponding constrained variational model expression is as follows:
Figure BDA0003425410870000061
{uk}＝{u1,u2,...,ukk IMF components for VMD method decomposition, { wk}＝{w1,w2,...,wkIs ukThe center of frequency of each IMF assembly,
Figure BDA0003425410870000062
is the partial derivative of function time, δ (t) is the unit pulse function, j is the imaginary unit, and is the convolution method.
Adaptively decomposing the mixed signal into relatively pure IMF components through the decomposition process; after VMD decomposition is carried out on the signals, IMF components with a limited number of different time scales are obtained, and the frequencies of the IMF components are arranged from low to high; the sharp part or noise of the signal corresponds to the large order IMF and the low frequency part of the signal corresponds to the small order IMF.
(2) The method based on independent component analysis is as follows:
the normalized autocorrelation function of a group of IMF components obtained after VMD decomposition of an original signal is a group of random sequences, the energy concentration degree of a discrete signal sequence in the area near the zero point is determined by calculating the energy concentration ratio of the area near the zero point, and the noise content degree of the IMF components is further determined; and reconstructing the noise domain into an imaginary noise component, and taking the imaginary noise component and the source signal as an input matrix of an independent component analysis algorithm, thereby obtaining the optimal estimation of the source signal.
Preferably, the semi-supervised dimension reduction method of introducing low rank regression analysis in the step C can more effectively mine the unmarked samples.
Will be respectively taken as Xl＝[xl,...,xl]TAs a sample set of labels, Xu＝[xl+1,...,xN]Is an unlabeled sample set; in addition to this, the present invention is,
Figure BDA0003425410870000063
wherein A isKIs a sample set of class k; from the above symbols, the matrix XlAnd XuIs a sample point and each column of the matrix is a sample point.
We can use independent component analysis to explore the relationship between these sample sets:
Figure BDA0003425410870000064
Figure BDA0003425410870000071
the matrix T represents the set of labeled samples and unlabeled samplesRelationships between this set, and S represents relationships between unlabeled exemplars; the h and j columns of the matrix represent the ith unlabeled sample Xl+iJth unlabeled sample Xl+jThe relationship between; firstly, preprocessing a matrix S, normalizing each line of the S, calculating the absolute value of each element, and then recording a result matrix as R;
second, we define the following tag propagation:
Lk+1＝βRLk+(1-β)L0,k＝1,L (5)
wherein β is a parameter that balances these two terms; l is0Abs (t) L, L is a matrix containing label information of the marked sample;
finally, the labels of the unlabeled samples are predicted in such a way that if the k-th component in line L X i is the largest in its i-th line, we assume that the i-th unlabeled sample Xl+iBelonging to class k, high dimensional data is projected to the desired dimension.
The semi-supervised method based on low-rank linear regression can effectively extract linear structures in a data set, and contributes to dimension reduction of extracted signal features by using unmarked data.
Preferably, in the step D, the features after dimension reduction are divided into a training set and a test set, and a supervised training set is given
Figure BDA0003425410870000072
Finding a certain function f using LightGBM*(x) Is approximated by
Figure BDA0003425410870000073
Multiple training runs minimize the expected value of a particular loss function L (y, f (x)), as follows:
Figure BDA0003425410870000074
LightGBM (LightGBM) decision tree algorithm based on Histogram, and integrates a plurality of T regression trees
Figure BDA0003425410870000075
To approximate the final model, i.e.
Figure BDA0003425410870000076
The regression tree may be represented as wq(x)Q ∈ {1, 2., J }, where J denotes the number of leaves, q denotes the decision rule of the tree, and w denotes the sample weight of the leaf node.
Therefore, the LightGBM will be trained in additive form, as follows:
Figure BDA0003425410870000077
in LightGBM, the objective function is quickly approximated by newton's method, and then for simplicity, using the Leaf growth strategy of Leaf-wise with depth constraint, after removing the constant terms in (7), the formula can be transformed into:
Figure BDA0003425410870000081
wherein g isiAnd hiFirst and second order gradient statistics representing a loss function; let Ij denote the sample set j of the leaf and can be converted into:
Figure BDA0003425410870000082
for a particular tree structure q (x), the optimal leaf weight score for each leaf node
Figure BDA0003425410870000083
And a value of fKThe following can be solved:
Figure BDA0003425410870000084
inputting the data of the test set into a LightGBM model to obtain the accuracy of the test set; the LightGBM model has good effect when using the training set; the advantages become more and more apparent as the sample size increases.
Preferably, in the step E, the features of the obtained reduced-dimension signal are input into the trained LightGBM model to realize cluster recognition, and the results of model recognition and classification are obtained, which have high conformity with the actual situation.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims (7)

1. An identity authentication method for Internet of things equipment for semi-supervised radio frequency fingerprint extraction comprises the following steps:
A. receiving a signal from the Internet of things equipment by using a communication signal receiver, carrying out frequency conversion on a signal spectrum to a proper frequency by using the receiver, and obtaining a digital signal by using analog-to-digital conversion and filtering operation;
B. decomposing a received signal into a specified number of modes by using a VMD-based method, determining a high-noise component by using an array entropy, and reconstructing a low-noise component to realize a noise reduction function, thereby realizing the extraction of nonlinear dynamic characteristics of the signal;
C. reducing the dimension of the radio frequency fingerprint by using a semi-supervised dimension reduction method based on low-rank linear regression analysis;
D. constructing a classification model by using a LightGBM algorithm, and training the acquired data;
E. the simplified features are clustered to identify the Internet of things equipment.
2. The identity authentication method for the equipment of the internet of things for semi-supervised radio frequency fingerprint extraction according to claim 1, wherein the identity authentication method comprises the following steps: after the communication signal in the step A receives the signal from the Internet of things equipment, the frequency spectrum of the signal is converted to a proper frequency through a radio frequency filter device, so that the interference quotation mark is attenuated, and then the digital signal is obtained by utilizing analog-to-digital conversion and filtering operation.
3. The identity authentication method for the equipment of the internet of things for semi-supervised radio frequency fingerprint extraction according to claim 1, wherein the identity authentication method comprises the following steps: and step B, using a method based on VMD and low-rank regression analysis, performing self-adaptive decomposition on the original signal into a plurality of Intrinsic Mode Functions (IMFs) by VMD decomposition, selecting a noise component according to an autocorrelation criterion, and then realizing optimal estimation on the transient electromagnetic signal by using an independent component analysis algorithm based on kurtosis maximization.
4. The identity authentication method for the equipment of the internet of things for semi-supervised radio frequency fingerprint extraction according to claim 1, wherein the identity authentication method comprises the following steps: and C, a semi-supervised dimension reduction method of low-rank regression analysis is introduced to effectively extract a linear structure in the data set, so that unmarked samples are effectively mined, and dimension reduction of the extracted signal features by using the unmarked data is facilitated.
5. The identity authentication method for the equipment of the internet of things for semi-supervised radio frequency fingerprint extraction according to claim 1, wherein the identity authentication method comprises the following steps: and D, dividing the features subjected to dimension reduction into a training set and a testing set, giving a supervision training set, training for multiple times, and inputting the data of the testing set into the LightGBM model to obtain the accuracy of the testing set.
6. The identity authentication method for the equipment of the internet of things for semi-supervised radio frequency fingerprint extraction according to claim 1, wherein the identity authentication method comprises the following steps: and E, inputting the characteristics of the obtained reduced-dimension signals into the trained LightGBM model to realize cluster recognition and obtain the results of model recognition and classification.
7. The identity authentication method for the equipment of the internet of things for semi-supervised radio frequency fingerprint extraction according to the claims 1-6, characterized in that:
the multi-modal decomposition process of the signal by the VMD algorithm is as follows:
the VMD transfers the acquisition process of the signal components into a variation frame, adopts a non-recursive processing strategy, and realizes the decomposition of the original signal into IMF components with the designated number by constructing and solving a constraint variation problem; in the VMD algorithm, the intrinsic mode function IMF is redefined as an amplitude-frequency modulated (AM-FM) signal based on the modulation criteria, i.e.
uk(t)＝AK(t)cos(φk(t))(1) (1)
φk(t) is ukPhase of (t) & phi'k(t)≥0，AK(t) is uk(t) instantaneous amplitude at time t and AK(t) is not less than 0; and the instantaneous amplitude uk(t) the amplitude envelope A and instantaneous angular frequency B vary slowly, i.e. over a sufficiently long time range t-delta, t + delta]Within, in the middle of them, (delta ≈ 2 π/φ'k(t))，uk(t) is approximately one having an amplitude of AK(t) and a frequency of wk(t) a harmonic signal;
continuously updating the frequency center and the bandwidth of each component in the iterative solution process of the variational model; carrying out self-adaptive segmentation on a signal frequency band according to the frequency characteristic of the signal to obtain a plurality of narrow-band IMF components;
assuming that the VMD decomposes the original signal into K IMF components, the corresponding constrained variational model expression is as follows:
Figure FDA0003425410860000021
{uk}＝{u1,u2,...,ukk IMF components for VMD method decomposition, { wk}＝{w1,w2,...,wkIs ukThe center of frequency of each IMF assembly,
Figure FDA0003425410860000023
is the partial derivative of the function time, δ (t) is the unit pulse function, j is the imaginary unit, is the convolution method;
adaptively decomposing the mixed signal into relatively pure IMF components through the decomposition process; after VMD decomposition is carried out on the signals, IMF components with a limited number of different time scales are obtained, and the frequencies of the IMF components are arranged from low to high; the sharp part or noise of the signal corresponds to the IMF of large order, and the low-frequency part of the signal corresponds to the IMF of small order;
the method based on independent component analysis is as follows:
the normalized autocorrelation function of a group of IMF components obtained after VMD decomposition of an original signal is a group of random sequences, the energy concentration degree of a discrete signal sequence in the area near the zero point is determined by calculating the energy concentration ratio of the area near the zero point, and the noise content degree of the IMF components is further determined; reconstructing the noise domain into an imaginary noise component, and using the imaginary noise component and the source signal as an input matrix of an independent component analysis algorithm so as to obtain the optimal estimation of the source signal;
and (3) mining unmarked samples by introducing a semi-supervised dimension reduction method of low-rank regression analysis, namely: will be respectively taken as Xl＝[xl,...,xl]TAs a sample set of labels, Xu＝[xl+1,...,xN]Is an unlabeled sample set; in addition to this, the present invention is,
Figure FDA0003425410860000022
wherein A isKIs a sample set of class k; from the above symbols, the matrix XlAnd XuEach row of (a) is a sample point, and each column of the matrix is a sample point;
independent component analysis was used to explore the relationship between these sample sets:
Figure FDA0003425410860000031
Figure FDA0003425410860000032
the matrix T represents the relationship between a set of labeled samples and a set of unlabeled samples, and S represents the relationship between unlabeled samples; the h and j columns of the matrix represent the ith unlabeled sample Xl+iJth unlabeled sample Xl+jThe relationship between; first, we preprocess the matrix S, and attribute to each row of SNormalizing, calculating the absolute value of each element, and recording the result matrix as R;
second, we define the following tag propagation:
Lk+1＝βRLk+(1-β)L0,k＝1,L (5)
wherein β is a parameter that balances these two terms; l is0Abs (t) L, L is a matrix containing label information of the marked sample;
finally, the labels of the unlabeled samples are predicted in such a way that if the k-th component in line L X i is the largest in its i-th line, we assume that the i-th unlabeled sample Xl+iBelonging to the kth class, and projecting high-dimensional data to a required dimension;
the low-rank linear regression-based semi-supervised method is used for extracting a linear structure in a data set, and is beneficial to using unmarked data to reduce the dimension of the extracted signal characteristics;
dividing the features after dimension reduction into a training set and a test set, and giving a supervision training set
Figure FDA0003425410860000033
Finding a certain function f using LightGBM*(x) Is approximated by
Figure FDA0003425410860000034
Multiple training runs minimize the expected value of a particular loss function L (y, f (x)), as follows:
Figure FDA0003425410860000035
LightGBM (LightGBM) decision tree algorithm based on Histogram, and integrates a plurality of T regression trees
Figure FDA0003425410860000036
To approximate the final model, i.e.
Figure FDA0003425410860000037
Regression tree canTo be represented as wq(x)Q ∈ {1, 2.,. J }, where J denotes the number of leaves, q denotes a decision rule of the tree, and w denotes a sample weight of a leaf node;
therefore, the LightGBM will be trained in additive form, as follows:
Figure FDA0003425410860000038
in LightGBM, the objective function is quickly approximated by newton's method, and then for simplicity, using the Leaf growth strategy of Leaf-wise with depth constraint, after removing the constant terms in (7), the formula can be transformed into:
Figure FDA0003425410860000041
wherein g isiAnd hiFirst and second order gradient statistics representing a loss function; let Ij denote the sample set j of the leaf and can be converted into:
Figure FDA0003425410860000042
for a particular tree structure q (x), the optimal leaf weight score for each leaf node
Figure FDA0003425410860000043
And extreme value ΓKThe following can be solved:
Figure FDA0003425410860000044
and inputting the data of the test set into a LightGBM model to obtain the accuracy of the test set.
CN202111576682.7A 2021-12-22 2021-12-22 Internet of things equipment identity authentication method for semi-supervised radio frequency fingerprint extraction Pending CN114255487A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111576682.7A CN114255487A (en) 2021-12-22 2021-12-22 Internet of things equipment identity authentication method for semi-supervised radio frequency fingerprint extraction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111576682.7A CN114255487A (en) 2021-12-22 2021-12-22 Internet of things equipment identity authentication method for semi-supervised radio frequency fingerprint extraction

Publications (1)

Publication Number Publication Date
CN114255487A true CN114255487A (en) 2022-03-29

Family

ID=80794029

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111576682.7A Pending CN114255487A (en) 2021-12-22 2021-12-22 Internet of things equipment identity authentication method for semi-supervised radio frequency fingerprint extraction

Country Status (1)

Country Link
CN (1) CN114255487A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115186714A (en) * 2022-07-15 2022-10-14 中国人民解放军国防科技大学 Network card frequency spectrum fingerprint feature amplification method based on feature correlation and self-adaptive decomposition
CN115510924A (en) * 2022-11-17 2022-12-23 中铁第一勘察设计院集团有限公司 Radio frequency fingerprint identification method based on improved variational modal decomposition

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115186714A (en) * 2022-07-15 2022-10-14 中国人民解放军国防科技大学 Network card frequency spectrum fingerprint feature amplification method based on feature correlation and self-adaptive decomposition
CN115186714B (en) * 2022-07-15 2023-07-21 中国人民解放军国防科技大学 Network card frequency spectrum fingerprint feature amplification method based on feature correlation and self-adaptive decomposition
CN115510924A (en) * 2022-11-17 2022-12-23 中铁第一勘察设计院集团有限公司 Radio frequency fingerprint identification method based on improved variational modal decomposition

Similar Documents

Publication Publication Date Title
CN114255487A (en) Internet of things equipment identity authentication method for semi-supervised radio frequency fingerprint extraction
CN100507971C (en) Independent component analysis based automobile sound identification method
CN109890043B (en) Wireless signal noise reduction method based on generative countermeasure network
CN107392123B (en) Radio frequency fingerprint feature extraction and identification method based on coherent accumulation noise elimination
CN109307862A (en) A kind of target radiation source individual discrimination method
CN108680796B (en) Electromagnetic information leakage detection system and method for computer display
CN114564982B (en) Automatic identification method for radar signal modulation type
Jeong et al. Spectrogram-based automatic modulation recognition using convolutional neural network
CN108567418A (en) A kind of pulse signal inferior health detection method and detecting system based on PCANet
CN111222442A (en) Electromagnetic signal classification method and device
CN111310656A (en) Single motor imagery electroencephalogram signal identification method based on multi-linear principal component analysis
CN111680737A (en) Radar radiation source individual identification method under differential signal-to-noise ratio condition
CN107616780A (en) A kind of brain electro-detection method and device using wavelet neural network
Huang et al. Radar waveform recognition based on multiple autocorrelation images
CN111652132B (en) Non-line-of-sight identity recognition method and device based on deep learning and storage medium
CN111983569B (en) Radar interference suppression method based on neural network
CN115770044B (en) Emotion recognition method and device based on electroencephalogram phase amplitude coupling network
CN111814703B (en) HB-based signal joint feature extraction method under non-reconstruction condition
CN114626412A (en) Multi-class target identification method and system for unattended sensor system
Shi et al. A radio frequency fingerprint identification method for wireless devices based on ShuffleNet-SVM
CN107341519A (en) A kind of SVMs identification optimization method based on multiresolution analysis
CN111404852A (en) Modulation mode identification method based on amplitude and spectral amplitude characteristics
Fan et al. A Hybrid Approach for Signal Modulation Recognition Using Deep Learning Methods
CN112132007B (en) Modulation mode identification method and system based on ResNet-SVM
US11849347B2 (en) Time axis correlation of pulsed electromagnetic transmissions

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination