[go: up one dir, main page]

CN108566642B - Two-dimensional joint feature authentication method based on machine learning - Google Patents

Two-dimensional joint feature authentication method based on machine learning Download PDF

Info

Publication number
CN108566642B
CN108566642B CN201810239508.5A CN201810239508A CN108566642B CN 108566642 B CN108566642 B CN 108566642B CN 201810239508 A CN201810239508 A CN 201810239508A CN 108566642 B CN108566642 B CN 108566642B
Authority
CN
China
Prior art keywords
base station
information
channel information
sender
information sender
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.)
Active
Application number
CN201810239508.5A
Other languages
Chinese (zh)
Other versions
CN108566642A (en
Inventor
陈松林
文红
陈洁
郑烜
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.)
Chengdu Alaifu Information Technology Co ltd
Original Assignee
Chengdu Alaifu Information Technology Co ltd
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 Chengdu Alaifu Information Technology Co ltd filed Critical Chengdu Alaifu Information Technology Co ltd
Priority to CN201810239508.5A priority Critical patent/CN108566642B/en
Publication of CN108566642A publication Critical patent/CN108566642A/en
Application granted granted Critical
Publication of CN108566642B publication Critical patent/CN108566642B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W12/00—Security arrangements; Authentication; Protecting privacy or anonymity
    • H04W12/06—Authentication
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L27/00—Modulated-carrier systems
    • H04L27/26—Systems using multi-frequency codes
    • H04L27/2601—Multicarrier modulation systems
    • H04L27/2647—Arrangements specific to the receiver only
    • H04L27/2655—Synchronisation arrangements
    • H04L27/2689—Link with other circuits, i.e. special connections between synchronisation arrangements and other circuits for achieving synchronisation
    • H04L27/2695—Link with other circuits, i.e. special connections between synchronisation arrangements and other circuits for achieving synchronisation with channel estimation, e.g. determination of delay spread, derivative or peak tracking
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W12/00—Security arrangements; Authentication; Protecting privacy or anonymity
    • H04W12/08—Access security

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Computer Security & Cryptography (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

The invention discloses a two-dimensional combined feature authentication method based on machine learning, which comprises the steps that a base station B firstly carries out channel information acquisition on a legal information sender A and a simulated illegal information sender E, then, channel information difference values between continuous data frames are calculated according to channel information between the base station B and the information sender, and on the basis of the channel information difference values, amplitude-based test statistics are constructed and processed to obtain amplitude-based normalized LRT statistics; based on amplitude and phase combination, constructing test statistic, processing to obtain normalized LRT statistic based on amplitude and phase combination, then constructing a sample set of two-dimensional combination characteristics, establishing an authentication model, generating a classifier by using a machine learning method, training according to the sample set to obtain the classifier with the standard detection rate, and carrying out validity judgment on an information sender with unknown identity. Compared with a single characteristic dimension channel authentication method, the method has higher accuracy.

Description

Two-dimensional joint feature authentication method based on machine learning
Technical Field
The invention relates to an authentication technology of channel information, in particular to a two-dimensional joint feature authentication method based on machine learning.
Background
The 5G mobile communication system puts forward the requirements of high speed, high efficiency and high security, and when a plurality of mobile devices access to the wireless network simultaneously, the burden of identity authentication in the network is greatly increased. 5G involves the interconnection and communication between a large number of machines and devices, and therefore, in the context of dense application of the network, a lightweight authentication method is required, which is a prerequisite for the operation of the Internet of things. In the last decade, the development of physical layer security technology has brought new vitality to the wireless mobile communication field. Physical layer authentication is a non-password authentication because acquisition of channel information is easy and channel characteristics are difficult to forge, and thus physical layer authentication based on channel characteristics has been receiving wide attention from researchers.
Generally, channel information of a radio can be used to detect the validity of a sender identity in a wireless network. However, the authentication method using a single characteristic has a certain limitation, and only one channel characteristic is used as a division basis, so that the method has no high accuracy.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a two-dimensional joint feature authentication method based on machine learning, which is used for verifying a statistic T based on amplitudeABased on the normalized LRT statistic T based on the improved amplitudeaAnd in a test statistic T based on a combination of amplitude and phaseBBased on the improved normalized LRT statistic T based on amplitude and phase combinationbThen establishing a two-dimensional feature (T) baseda，Tb) Compared with the channel authentication method with single characteristic dimension, the authentication model has higher accuracy in channel authentication.
The purpose of the invention is realized by the following technical scheme: a two-dimensional joint feature authentication method based on machine learning comprises the following steps:
s1, the base station B carries out channel information acquisition on a legal information sender A and a simulated illegal information sender E to obtain a channel information data set of the legal information sender A
Figure GDA0003149919780000011
And a channel information data set simulating an illegal information sender E
Figure GDA0003149919780000012
S2, calculating a channel information difference value between continuous data frames for channel information between the base station B and an information sender;
s3, on the basis of the channel information difference value, constructing an amplitude-based test statistic T based on the amplitude difference of the subcarriersAAnd to TAProcessing to obtain a product based on TAImproved normalized LRT statistic TaTaking it as a first dimension feature:
wherein, the first dimension between the legal information sender A and the base station B is characterized in that
Figure GDA0003149919780000013
Simulating the first dimension characteristic between an illegal information sender E and a base station B as
Figure GDA0003149919780000021
S4, on the basis of the channel information difference value, constructing a test statistic T based on amplitude and phase combinationBTo TBProcessing to obtain a product based on TBImproved normalized LRT statistic TbAnd taking the two-dimensional feature as a second-dimensional feature:
wherein, the second dimension between the legal information sender A and the base station B is characterized in that
Figure GDA0003149919780000022
Simulating a second dimension between the illegal message sender E and the base station B as
Figure GDA0003149919780000023
S5, utilizing the channel information data set
Figure GDA0003149919780000024
And
Figure GDA0003149919780000025
constructing a two-dimensional union feature (T)a，Tb) As a sample set, two-dimensional joint features (T) are combined for data frames at the same timea，Tb) As a comprehensive basis for determinationConstructing an authentication model [ (T)a，Tb),y]Wherein:
the sample set of the two-dimensional joint characteristics between the legal information sender A and the base station B is as follows:
Figure GDA0003149919780000026
adding an identifier y to + 1;
the sample set for simulating the two-dimensional joint characteristics between the illegal information sender E and the base station B is as follows:
Figure GDA0003149919780000027
adding an identifier y to-1;
s6, adopting a machine learning method to construct a classifier, and according to the sample set TABAnd TEBTraining the classifier until the detection rate of the classifier reaches the standard;
and S7, the base station judges the legality of the information sender with unknown identity by using the classifier with the detection rate reaching the standard, so that the channel authentication of the two-dimensional combined characteristics based on machine learning is realized.
The step S1 includes the following sub-steps:
s101, a legal information sender A sends a signal to a base station B, and the base station B collects channel information of the legal information sender A
Figure GDA0003149919780000028
Figure GDA0003149919780000029
Wherein, N represents the number of frames,
Figure GDA00031499197800000210
channel information representing the estimation of the kth OFDM symbol between the base station B and the legitimate information sender a, k being 1,2, 3.
S102, simulating an illegal information sender E to send a signal to a base station B, and collecting signals simulating the illegal information sender E by the base station BChannel information
Figure GDA00031499197800000211
Figure GDA00031499197800000212
N represents the number of frames,
Figure GDA00031499197800000213
channel information indicating the estimation of the kth OFDM symbol between the base station B and the transmitter E of the analog illegal information, k being 1,2, 3.
Further, the channel information data frames of the legal information sender A and the simulated illegal information sender E are continuously sent, and the time interval between the two adjacent frames of data is collected within the relevant time, and the channel information has correlation.
In step S2, the channel information difference between consecutive data frames is calculated by the following formula:
Figure GDA0003149919780000031
in the formula,
Figure GDA0003149919780000032
representing the difference between the channel information data set to be calculated, the channel information of the (k + 1) th frame and the channel of the (k) th frame.
The step S3 includes the following sub-steps:
s301, on the basis of the channel information difference value, constructing an amplitude-based test statistic T based on the amplitude difference of subcarriersA：
Figure GDA0003149919780000033
Wherein,
Figure GDA0003149919780000034
m and n respectively represent the m-th row and the n-th column of the channel matrix; n represents total N frame data; sigma2Representing the noise power; n is a radical ofsRepresenting channel information of data frames, containing NsA frequency channel matrix, which is an OFDM symbol of an N-dimensional square matrix; x is NsA cumulative variable of the number of (1); k represents the kth frame data; the superscript XB represents an information sender X and a base station B, when the information sender X is a legal information sender A, XB (AB) in the formula is constructed to obtain amplitude-based test statistic of channel information between the legal information sender A and the base station B, and when the information sender X is a simulated illegal information sender E, XB (EB) in the formula is constructed to obtain amplitude-based test statistic of channel information between the simulated illegal information sender E and the base station B;
s302, test statistic TAProcessing to obtain a product based on TAImproved normalized LRT statistic TaTaking it as a first dimension feature:
Figure GDA0003149919780000035
wherein, the first dimension between the legal information sender A and the base station B is characterized in that
Figure GDA0003149919780000036
Simulating the first dimension characteristic between an illegal information sender E and a base station B as
Figure GDA0003149919780000037
Figure GDA0003149919780000041
Figure GDA0003149919780000042
The step S4 includes the following sub-steps:
s401, on the basis of the channel information difference value, based onAmplitude and phase combination to construct test statistic TB：
Figure GDA0003149919780000043
The upper standard XB represents an information sender X and a base station B, when the information sender X is a legal information sender A, XB (AB) in the formula is constructed to obtain the inspection statistic of channel information between the legal information sender A and the base station B based on the combination of amplitude and phase, and when the information sender X is a simulated illegal information sender E, XB (EB) in the formula is constructed to obtain the inspection statistic of channel information between the simulated illegal information sender E and the base station B based on the combination of amplitude and phase;
s402, testing statistic TBProcessing to obtain a product based on TBImproved normalized LRT statistic TbAnd taking the two-dimensional feature as a second-dimensional feature:
Figure GDA0003149919780000044
wherein, the second dimension between the legal information sender A and the base station B is characterized in that
Figure GDA0003149919780000045
Simulating a second dimension between the illegal message sender E and the base station B as
Figure GDA0003149919780000046
Figure GDA0003149919780000051
Figure GDA0003149919780000052
The step S5 includes the following sub-steps:
s501, sending the legal information from the sender A to the baseChannel information data set of station B
Figure GDA0003149919780000053
And (3) processing:
data set of channel information
Figure GDA0003149919780000054
Bringing in
Figure GDA0003149919780000055
Obtaining M frames of first dimension characteristics of channel information between a legal information sender A and a base station B in the calculation formula:
Figure GDA0003149919780000056
data set of channel information
Figure GDA0003149919780000057
Bringing in
Figure GDA0003149919780000058
Obtaining M frames of second dimension characteristics of channel information between a legal information sender A and a base station B in the calculation formula:
Figure GDA0003149919780000059
in the formula, M is N-2, that is, N frames of original data, and N-2 frames of feature data can be obtained;
s502, constructing a two-dimensional joint feature sample set T from a legal information sender A to a base station BAB：
Figure GDA00031499197800000510
Due to channel characteristics
Figure GDA00031499197800000511
And channel characteristics
Figure GDA00031499197800000512
Combined into a two-dimensional combined feature TABSo that the k frame data of the sample feature becomes
Figure GDA00031499197800000513
S503, collecting a sample set TABAs a legal data source, and for the sample set TABAdding an identifier y to +1 in each frame of data;
s504, simulating the channel information data set from the illegal information sender E to the base station B
Figure GDA00031499197800000514
And (3) processing:
data set of channel information
Figure GDA00031499197800000515
Bringing in
Figure GDA00031499197800000516
In the calculation formula, obtaining M frames of first dimension characteristics of channel information between the simulated illegal information sender E and the base station B:
Figure GDA00031499197800000517
data set of channel information
Figure GDA0003149919780000061
Bringing in
Figure GDA0003149919780000062
In the calculation formula, obtaining a second dimension characteristic of M frames simulating channel information between an illegal information sender E and a base station B:
Figure GDA0003149919780000063
in the formula, M is N-2, that is, N frames of original data, and N-2 frames of feature data can be obtained;
S505. constructing a two-dimensional joint feature sample set T simulating illegal information sender E to base station BEB:
Figure GDA0003149919780000064
Due to channel characteristics
Figure GDA0003149919780000065
And channel characteristics
Figure GDA0003149919780000066
Combined into a two-dimensional combined feature TEBSo that any k-th frame data of the sample feature becomes
Figure GDA0003149919780000067
S506, collecting the sample set TEBAs an illegal data source and for the sample set TEBAdds the identifier y-1 to each frame data in (1).
Further, the machine learning method in step S6 is any machine learning algorithm with a binary function, which is based on two-dimensional joint features (T)a,Tb) Carrying out classification training on data frames from a legal information sender A or a simulated illegal information sender E to obtain a corresponding classifier;
specifically, the step S6 includes the following sub-steps:
s601, establishing a training set and a testing set, and performing training and testing on a sample set TABAnd sample set TEBAdding each extracted partial data frame into training set, sample set TABAnd sample set TEBAdding the rest data frames into the test set;
s602, constructing a classifier by adopting a machine learning method, and training the classifier by utilizing data in a training set;
s603, testing the classifier obtained by training by using the data in the test set, judging whether the detection rate of the classifier reaches a set threshold value, and if so, outputting the classifier with the detection rate reaching the standard; if not, the process returns to step S1, and the sample set acquisition and classifier training of steps S1 to S6 are repeated.
The step S7 includes the following sub-steps:
s701, the base station B carries out channel information acquisition on the information sender C with unknown identity to obtain a channel information data set
Figure GDA0003149919780000068
Figure GDA0003149919780000069
Wherein, N represents the number of frames,
Figure GDA00031499197800000610
channel information representing the kth OFDM symbol estimate between the base station B and the information sender C of unknown identity, k being 1,2, 3.., N;
s702. for the channel information data set
Figure GDA00031499197800000611
On the basis of the channel information difference value, constructing an amplitude-based test statistic based on the amplitude difference of subcarriers, and processing the test statistic to obtain a normalized LRT statistic
Figure GDA00031499197800000612
As a first dimension characteristic of channel information between the base station B and the unknown identity information sender C:
Figure GDA0003149919780000071
data set of channel information
Figure GDA0003149919780000072
Bringing in
Figure GDA0003149919780000073
In the formula for calculating (a) of (b),obtaining M frames of first dimension characteristics of channel information between an unknown identity information sender C and a base station B:
Figure GDA0003149919780000074
M＝N-2；
s703. for the channel information data set
Figure GDA0003149919780000075
On the basis of the channel information difference value, constructing test statistic based on amplitude and phase combination, and processing the test statistic to obtain normalized LRT statistic
Figure GDA0003149919780000076
A second dimension characteristic as channel information between the base station B and the unknown identity information sender C:
Figure GDA0003149919780000077
data set of channel information
Figure GDA0003149919780000078
Bringing in
Figure GDA0003149919780000079
Obtaining M frames of second dimension characteristics of channel information between the unknown identity information sender C and the base station B in the calculation formula:
Figure GDA00031499197800000710
M＝N-2；
s704, constructing a two-dimensional joint feature sample set T of channel information between a base station B and an unknown identity information sender CCB：
Figure GDA00031499197800000711
S705, utilizing the classifier with the detection rate reaching the standard to perform T on the sample setCBIs judged to obtainWhen the y value is +1, the information sender C with unknown identity is legal; a value of-1 for y would represent that the sender C of the information of unknown identity is illegal.
The invention has the beneficial effects that: in an amplitude-based test statistic TABased on the normalized LRT statistic T based on the improved amplitudeaAnd in a test statistic T based on a combination of amplitude and phaseBBased on the improved normalized LRT statistic T based on amplitude and phase combinationbThen establishing a two-dimensional feature (T) baseda，Tb) The authentication model realizes channel authentication under two-dimensional combined characteristics by combining a machine learning algorithm, improves the defect of low accuracy of the existing authentication technology, and has higher accuracy compared with a channel authentication method with a single characteristic dimension.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
The technical solutions of the present invention are further described in detail below with reference to the accompanying drawings, but the scope of the present invention is not limited to the following.
As shown in fig. 1, a two-dimensional joint feature authentication method based on machine learning includes the following steps:
s1, the base station B carries out channel information acquisition on a legal information sender A and a simulated illegal information sender E to obtain a channel information data set of the legal information sender A
Figure GDA0003149919780000081
And a channel information data set simulating an illegal information sender E
Figure GDA0003149919780000082
S2, calculating a channel information difference value between continuous data frames for channel information between the base station B and an information sender;
s3, on the basis of the channel information difference value, constructing an amplitude-based test statistic T based on the amplitude difference of the subcarriersAAnd to TAProcessing to obtain a product based on TAImproved normalized LRT statistic TaTaking it as a first dimension feature:
wherein, the first dimension between the legal information sender A and the base station B is characterized in that
Figure GDA0003149919780000083
Simulating the first dimension characteristic between an illegal information sender E and a base station B as
Figure GDA0003149919780000084
S4, on the basis of the channel information difference value, constructing a test statistic T based on amplitude and phase combinationBTo TBProcessing to obtain a product based on TBImproved normalized LRT statistic TbAnd taking the two-dimensional feature as a second-dimensional feature:
wherein, the second dimension between the legal information sender A and the base station B is characterized in that
Figure GDA0003149919780000085
Simulating a second dimension between the illegal message sender E and the base station B as
Figure GDA0003149919780000086
S5, utilizing the channel information data set
Figure GDA0003149919780000087
And
Figure GDA0003149919780000088
constructing a two-dimensional union feature (T)a，Tb) As a sample set, two-dimensional joint features (T) are combined for data frames at the same timea，Tb) As a comprehensive judgment basis, an authentication model [ (T) is constructeda，Tb),y]Wherein:
the sample set of the two-dimensional joint characteristics between the legal information sender A and the base station B is as follows:
Figure GDA0003149919780000089
adding an identifier y to + 1;
the sample set for simulating the two-dimensional joint characteristics between the illegal information sender E and the base station B is as follows:
Figure GDA00031499197800000810
adding an identifier y to-1;
s6, adopting a machine learning method to construct a classifier, and according to the sample set TABAnd TEBTraining the classifier until the detection rate of the classifier reaches the standard;
and S7, the base station judges the legality of the information sender with unknown identity by using the classifier with the detection rate reaching the standard, so that the channel authentication of the two-dimensional combined characteristics based on machine learning is realized.
The step S1 includes the following sub-steps:
s101, a legal information sender A sends a signal to a base station B, and the base station B collects channel information of the legal information sender A
Figure GDA0003149919780000091
Figure GDA0003149919780000092
Wherein, N represents the number of frames,
Figure GDA0003149919780000093
channel information representing the estimation of the kth OFDM symbol between the base station B and the legitimate information sender a, k being 1,2, 3.
S102, the illegal information sender E is simulated to send signals to the base station B, and the base station B collects channel information of the illegal information sender E
Figure GDA0003149919780000094
Figure GDA0003149919780000095
N represents the number of frames,
Figure GDA0003149919780000096
channel information indicating the estimation of the kth OFDM symbol between the base station B and the transmitter E of the analog illegal information, k being 1,2, 3.
Further, the channel information data frames of the legal information sender A and the simulated illegal information sender E are continuously sent, and the time interval between the two adjacent frames of data is collected within the relevant time, and the channel information has correlation.
In step S2, the channel information difference between consecutive data frames is calculated by the following formula:
Figure GDA0003149919780000097
in the formula,
Figure GDA0003149919780000098
representing the difference between the channel information data set to be calculated, the channel information of the (k + 1) th frame and the channel of the (k) th frame.
The step S3 includes the following sub-steps:
s301, on the basis of the channel information difference value, constructing an amplitude-based test statistic T based on the amplitude difference of subcarriersA：
Figure GDA0003149919780000099
Wherein,
Figure GDA00031499197800000910
m and n respectively represent the m-th row and the n-th column of the channel matrix; n represents total N frame data; sigma2Representing the noise power; n is a radical ofsRepresenting channel information of data frames, containing NsA frequency channel matrix, which is an OFDM symbol of an N-dimensional square matrix; x is NsA cumulative variable of the number of (1); k represents the kth frame data; the superscript XB represents an information sender X and a base station B, when the information sender X is a legal information sender A, XB (AB) in the formula is constructed to obtain amplitude-based test statistic of channel information between the legal information sender A and the base station B, and when the information sender X is a simulated illegal information sender E, XB (EB) in the formula is constructed to obtain amplitude-based test statistic of channel information between the simulated illegal information sender E and the base station B;
s302, test statistic TAProcessing to obtain a product based on TAImproved normalized LRT statistic TaTaking it as a first dimension feature:
Figure GDA0003149919780000101
wherein, the first dimension between the legal information sender A and the base station B is characterized in that
Figure GDA0003149919780000102
Simulating the first dimension characteristic between an illegal information sender E and a base station B as
Figure GDA0003149919780000103
Figure GDA0003149919780000104
Figure GDA0003149919780000105
The step S4 includes the following sub-steps:
s401, on the basis of the channel information difference value, constructing a test statistic T based on amplitude and phase combinationB：
Figure GDA0003149919780000106
The upper standard XB represents an information sender X and a base station B, when the information sender X is a legal information sender A, XB (AB) in the formula is constructed to obtain the inspection statistic of channel information between the legal information sender A and the base station B based on the combination of amplitude and phase, and when the information sender X is a simulated illegal information sender E, XB (EB) in the formula is constructed to obtain the inspection statistic of channel information between the simulated illegal information sender E and the base station B based on the combination of amplitude and phase;
s402, testing statistic TBProcessing to obtain a product based on TBImproved normalized LRT statistic TbAnd taking the two-dimensional feature as a second-dimensional feature:
Figure GDA0003149919780000111
wherein, the second dimension between the legal information sender A and the base station B is characterized in that
Figure GDA0003149919780000112
Simulating a second dimension between the illegal message sender E and the base station B as
Figure GDA0003149919780000113
Figure GDA0003149919780000114
Figure GDA0003149919780000115
The step S5 includes the following sub-steps:
s501, channel information data set from legal information sender A to base station B
Figure GDA0003149919780000116
And (3) processing:
data set of channel information
Figure GDA0003149919780000117
Bringing in
Figure GDA0003149919780000118
Obtaining M frames of first dimension characteristics of channel information between a legal information sender A and a base station B in the calculation formula:
Figure GDA0003149919780000119
data set of channel information
Figure GDA00031499197800001110
Bringing in
Figure GDA00031499197800001111
Obtaining M frames of second dimension characteristics of channel information between a legal information sender A and a base station B in the calculation formula:
Figure GDA0003149919780000121
in the formula, M is N-2, that is, N frames of original data, and N-2 frames of feature data can be obtained;
s502, constructing a two-dimensional joint feature sample set T from a legal information sender A to a base station BAB：
Figure GDA0003149919780000122
Due to channel characteristics
Figure GDA0003149919780000123
And channel characteristics
Figure GDA0003149919780000124
Combined into a two-dimensional combined feature TABSo that the k frame data of the sample feature becomes
Figure GDA0003149919780000125
S503, collecting a sample set TABAs a legal data source, and for the sample set TABAdding an identifier y to +1 in each frame of data;
s504, simulating the channel information data set from the illegal information sender E to the base station B
Figure GDA0003149919780000126
And (3) processing:
data set of channel information
Figure GDA0003149919780000127
Bringing in
Figure GDA0003149919780000128
In the calculation formula, obtaining M frames of first dimension characteristics of channel information between the simulated illegal information sender E and the base station B:
Figure GDA0003149919780000129
data set of channel information
Figure GDA00031499197800001210
Bringing in
Figure GDA00031499197800001211
In the calculation formula, obtaining a second dimension characteristic of M frames simulating channel information between an illegal information sender E and a base station B:
Figure GDA00031499197800001212
in the formula, M is N-2, that is, N frames of original data, and N-2 frames of feature data can be obtained;
s505, constructing a two-dimensional joint feature sample set T simulating illegal information from a sender E to a base station BEB:
Figure GDA00031499197800001213
Due to channel characteristics
Figure GDA00031499197800001214
And channel characteristics
Figure GDA00031499197800001215
Combined into a two-dimensional combined feature TEBSo that any k-th frame data of the sample feature becomes
Figure GDA00031499197800001216
S506, collecting the sample set TEBAs an illegal data source and for the sample set TEBAdds the identifier y-1 to each frame data in (1).
Further, the machine learning method in step S6 is any machine learning algorithm with a binary function, which is based on two-dimensional joint features (T)a,Tb) Carrying out classification training on a data frame from a legal information sender A or a simulated illegal information sender E to obtain a corresponding classifier;
specifically, the step S6 includes the following sub-steps:
s601, establishing a training set and a testing set, and performing training and testing on a sample set TABAnd sample set TEBAdding each extracted partial data frame into training set, sample set TABAnd sample set TEBAdding the rest data frames into the test set;
s602, constructing a classifier by adopting a machine learning method, and training the classifier by utilizing data in a training set;
s603, testing the classifier obtained by training by using the data in the test set, judging whether the detection rate of the classifier reaches a set threshold value, and if so, outputting the classifier with the detection rate reaching the standard; if not, the process returns to step S1, and the sample set acquisition and classifier training of steps S1 to S6 are repeated.
In an embodiment of the present application, the step S603 specifically includes: sequentially inputting the data frames in the test set into a classifier obtained by training, judging each frame of data by the classifier to obtain a corresponding y value, counting and judging the correct number of data frames according to the marking information of the y value in the test set, dividing the correct number of data frames by the total number of the data frames in the test set to obtain the detection rate, if the detection rate is higher than a set threshold value, judging that the detection rate reaches the standard, outputting the classifier with the detection rate reaching the standard, if the detection rate is lower than the set threshold value, judging that the detection rate does not reach the standard, returning to the step S1, and repeatedly performing the sample set acquisition and the classifier training of the steps S1-S6.
The step S7 includes the following sub-steps:
s701, the base station B carries out channel information acquisition on the information sender C with unknown identity to obtain a channel information data set
Figure GDA0003149919780000131
Figure GDA0003149919780000132
Wherein, N represents the number of frames,
Figure GDA0003149919780000133
channel information representing the kth OFDM symbol estimate between the base station B and the information sender C of unknown identity, k being 1,2, 3.., N;
s702. for the channel information data set
Figure GDA0003149919780000134
On the basis of the channel information difference value, constructing an amplitude-based test statistic based on the amplitude difference of subcarriers, and processing the test statistic to obtain a normalized LRT statistic
Figure GDA0003149919780000135
As a first dimension characteristic of channel information between the base station B and the unknown identity information sender C:
Figure GDA0003149919780000136
data set of channel information
Figure GDA0003149919780000137
Bringing in
Figure GDA0003149919780000138
Obtaining M frames of first dimension characteristics of channel information between an unknown identity information sender C and a base station B in the calculation formula:
Figure GDA0003149919780000139
M＝N-2；
s703. for the channel information data set
Figure GDA00031499197800001310
On the basis of the channel information difference value, constructing test statistic based on amplitude and phase combination, and processing the test statistic to obtain normalized LRT statistic
Figure GDA0003149919780000141
A second dimension characteristic as channel information between the base station B and the unknown identity information sender C:
Figure GDA0003149919780000142
data set of channel information
Figure GDA0003149919780000143
Bringing in
Figure GDA0003149919780000144
Obtaining M frames of second dimension characteristics of channel information between the unknown identity information sender C and the base station B in the calculation formula:
Figure GDA0003149919780000145
M＝N-2；
s704, constructing a base station B and an unknown identity information senderTwo-dimensional joint characteristic sample set T of channel information between CCB：
Figure GDA0003149919780000146
S705, utilizing the classifier with the detection rate reaching the standard to perform T on the sample setCBJudging any frame data to obtain a corresponding y value, wherein the y value is +1 and represents that an information sender C with unknown identity is legal; a value of-1 for y would represent that the sender C of the information of unknown identity is illegal.
In conclusion, because the optimal judgment threshold value is difficult to find manually by the two-dimensional combined channel characteristics, when the dimension is increased and the legal or illegal identity is judged, a single limit is not defined according to a single standard any more, the invention comprehensively considers the two-dimensional combined characteristics to make the final judgment; in particular, in an amplitude-based test statistic TABased on the normalized LRT statistic T based on the improved amplitudeaAnd in a test statistic T based on a combination of amplitude and phaseBBased on the improved normalized LRT statistic T based on amplitude and phase combinationbThen establishing a two-dimensional feature (T) baseda，Tb) The authentication model of (1); the machine learning algorithm can be used for improving the authentication performance, input data of channel information is used as training data, each group of training data is added with a clear identifier, then a prediction model is determined, a learning process is established, the prediction result is compared with the actual result of the training data, the prediction model is continuously adjusted until the prediction result of the model reaches an expected accuracy, finally, tested channel information data is used as the input data, and an output label is given by a classifier generated by machine learning, so that the authentication is realized.

Claims (9)

1. A two-dimensional joint feature authentication method based on machine learning is characterized in that: the method comprises the following steps:
s1, the base station B carries out channel information acquisition on a legal information sender A and a simulated illegal information sender E to obtain a channel information data set of the legal information sender A
Figure FDA0003149919770000011
And a channel information data set simulating an illegal information sender E
Figure FDA0003149919770000012
S2, calculating a channel information difference value between continuous data frames for channel information between the base station B and an information sender;
s3, on the basis of the channel information difference value, constructing an amplitude-based test statistic T based on the amplitude difference of the subcarriersAAnd to TAProcessing to obtain a product based on TAImproved normalized LRT statistic TaTaking it as a first dimension feature:
wherein, the first dimension between the legal information sender A and the base station B is characterized in that
Figure FDA0003149919770000013
Simulating the first dimension characteristic between an illegal information sender E and a base station B as
Figure FDA0003149919770000014
S4, on the basis of the channel information difference value, constructing a test statistic T based on amplitude and phase combinationBTo TBProcessing to obtain a product based on TBImproved normalized LRT statistic TbAnd taking the two-dimensional feature as a second-dimensional feature:
wherein, the second dimension between the legal information sender A and the base station B is characterized in that
Figure FDA0003149919770000015
Imitating illegal information sender EThe second dimension between the base station B and the base station B is characterized in that
Figure FDA0003149919770000016
S5, utilizing the channel information data set
Figure FDA0003149919770000017
And
Figure FDA0003149919770000018
constructing a two-dimensional union feature (T)a，Tb) As a sample set, two-dimensional joint features (T) are combined for data frames at the same timea，Tb) As a comprehensive judgment basis, an authentication model [ (T) is constructeda，Tb),y]Wherein:
the sample set of the two-dimensional joint characteristics between the legal information sender A and the base station B is as follows:
Figure FDA0003149919770000019
adding an identifier y to + 1;
the sample set for simulating the two-dimensional joint characteristics between the illegal information sender E and the base station B is as follows:
Figure FDA00031499197700000110
adding an identifier y to-1;
where M-N-2, N denotes a channel information data set
Figure FDA00031499197700000111
And
Figure FDA00031499197700000112
the number of frames;
s6, adopting a machine learning method to construct a classifier, and according to the sample set TABAnd TEBTraining the classifier until the detection rate of the classifier reaches the standard;
and S7, the base station judges the legality of the information sender with unknown identity by using the classifier with the detection rate reaching the standard, so that the channel authentication of the two-dimensional combined characteristics based on machine learning is realized.
2. The machine learning-based two-dimensional joint feature authentication method according to claim 1, wherein: the step S1 includes the following sub-steps:
s101, a legal information sender A sends a signal to a base station B, and the base station B collects channel information of the legal information sender A
Figure FDA0003149919770000021
Figure FDA0003149919770000022
Wherein, N represents the number of frames,
Figure FDA0003149919770000023
channel information representing the estimation of the kth OFDM symbol between the base station B and the legitimate information sender a, k being 1,2, 3.
S102, the illegal information sender E is simulated to send signals to the base station B, and the base station B collects channel information of the illegal information sender E
Figure FDA0003149919770000024
Figure FDA0003149919770000025
N represents the number of frames,
Figure FDA0003149919770000026
channel information indicating the estimation of the kth OFDM symbol between the base station B and the transmitter E of the analog illegal information, k being 1,2, 3.
3. The machine learning-based two-dimensional joint feature authentication method according to claim 1, wherein: the channel information data frames of the legal information sender A and the simulated illegal information sender E are continuously sent, and the time interval between the two adjacent frames of data is collected within the relevant time, and the channel information has correlation.
4. The two-dimensional joint feature authentication method based on machine learning according to claim 2, wherein: in step S2, the channel information difference between consecutive data frames is calculated by the following formula:
Figure FDA0003149919770000027
in the formula,
Figure FDA0003149919770000028
representing the difference between the channel information data set to be calculated, the channel information of the (k + 1) th frame and the channel of the (k) th frame.
5. The machine learning-based two-dimensional joint feature authentication method according to claim 4, wherein: the step S3 includes the following sub-steps:
s301, on the basis of the channel information difference value, constructing an amplitude-based test statistic T based on the amplitude difference of subcarriersA：
Figure FDA0003149919770000029
Wherein,
Figure FDA00031499197700000210
m and n respectively represent the m-th row and the n-th column of the channel matrix; n represents total N frame data; sigma2Representing the noise power; n is a radical ofsRepresenting data frame channel information, packetsContaining NsEach frequency channel matrix is an OFDM symbol of an N-dimensional square matrix; x is NsA cumulative variable of the number of (1); k represents the kth frame data; the superscript XB represents an information sender X and a base station B, when the information sender X is a legal information sender A, XB (AB) in the formula is constructed to obtain amplitude-based test statistic of channel information between the legal information sender A and the base station B, and when the information sender X is a simulated illegal information sender E, XB (EB) in the formula is constructed to obtain amplitude-based test statistic of channel information between the simulated illegal information sender E and the base station B;
s302, test statistic TAProcessing to obtain a product based on TAImproved normalized LRT statistic TaTaking it as a first dimension feature:
Figure FDA0003149919770000031
wherein, the first dimension between the legal information sender A and the base station B is characterized in that
Figure FDA0003149919770000032
Simulating the first dimension characteristic between an illegal information sender E and a base station B as
Figure FDA0003149919770000033
Figure FDA0003149919770000034
Figure FDA0003149919770000035
6. The machine learning-based two-dimensional joint feature authentication method according to claim 5, wherein: the step S4 includes the following sub-steps:
s401, on the basis of the channel information difference value, constructing a test statistic T based on amplitude and phase combinationB：
Figure FDA0003149919770000036
The upper standard XB represents an information sender X and a base station B, when the information sender X is a legal information sender A, XB (AB) in the formula is constructed to obtain the inspection statistic of channel information between the legal information sender A and the base station B based on the combination of amplitude and phase, and when the information sender X is a simulated illegal information sender E, XB (EB) in the formula is constructed to obtain the inspection statistic of channel information between the simulated illegal information sender E and the base station B based on the combination of amplitude and phase;
s402, testing statistic TBProcessing to obtain a product based on TBImproved normalized LRT statistic TbAnd taking the two-dimensional feature as a second-dimensional feature:
Figure FDA0003149919770000041
wherein, the second dimension between the legal information sender A and the base station B is characterized in that
Figure FDA0003149919770000042
Simulating a second dimension between the illegal message sender E and the base station B as
Figure FDA0003149919770000043
Figure FDA0003149919770000044
Figure FDA0003149919770000045
7. The machine learning-based two-dimensional joint feature authentication method according to claim 1, wherein: the step S5 includes the following sub-steps:
s501, channel information data set from legal information sender A to base station B
Figure FDA0003149919770000046
And (3) processing:
data set of channel information
Figure FDA0003149919770000047
Bringing in
Figure FDA0003149919770000048
Obtaining M frames of first dimension characteristics of channel information between a legal information sender A and a base station B in the calculation formula:
Figure FDA0003149919770000049
data set of channel information
Figure FDA0003149919770000051
Bringing in
Figure FDA0003149919770000052
Obtaining M frames of second dimension characteristics of channel information between a legal information sender A and a base station B in the calculation formula:
Figure FDA0003149919770000053
in the formula, M is N-2, namely N frames of original data, and N-2 frames of feature data are obtained;
s502, constructing a two-dimensional joint feature sample set T from a legal information sender A to a base station BAB：
Figure FDA0003149919770000054
Due to channel characteristics
Figure FDA0003149919770000055
And channel characteristics
Figure FDA0003149919770000056
Combined into a two-dimensional combined feature TABSo that the k frame data of the sample feature becomes
Figure FDA0003149919770000057
S503, collecting a sample set TABAs a legal data source, and for the sample set TABAdding an identifier y to +1 in each frame of data;
s504, simulating the channel information data set from the illegal information sender E to the base station B
Figure FDA0003149919770000058
And (3) processing:
data set of channel information
Figure FDA0003149919770000059
Bringing in
Figure FDA00031499197700000510
In the calculation formula, obtaining M frames of first dimension characteristics of channel information between the simulated illegal information sender E and the base station B:
Figure FDA00031499197700000511
data set of channel information
Figure FDA00031499197700000512
Bringing in
Figure FDA00031499197700000513
In the calculation formula, obtaining a second dimension characteristic of M frames simulating channel information between an illegal information sender E and a base station B:
Figure FDA00031499197700000514
in the formula, M is N-2, namely N frames of original data, and N-2 frames of feature data are obtained;
s505, constructing a two-dimensional joint feature sample set T simulating illegal information from a sender E to a base station BEB:
Figure FDA00031499197700000515
Due to channel characteristics
Figure FDA00031499197700000516
And channel characteristics
Figure FDA00031499197700000517
Combined into a two-dimensional combined feature TEBSo that any k-th frame data of the sample feature becomes
Figure FDA00031499197700000518
S506, collecting the sample set TEBAs an illegal data source and for the sample set TEBAdds the identifier y-1 to each frame data in (1).
8. The machine learning-based two-dimensional joint feature authentication method according to claim 1, wherein: the step S6 includes the following sub-steps:
s601, establishing a training set and a testing set, and performing training and testing on a sample set TABAnd sample set TEBAdding each extracted partial data frame into training set, sample set TABAnd sample set TEBThe rest of the data frames are added into the test set；
S602, constructing a classifier by adopting a machine learning method, and training the classifier by utilizing data in a training set;
s603, testing the classifier obtained by training by using the data in the test set, judging whether the detection rate of the classifier reaches a set threshold value, and if so, outputting the classifier with the detection rate reaching the standard; if not, the process returns to step S1, and the sample set acquisition and classifier training of steps S1 to S6 are repeated.
9. The machine learning-based two-dimensional joint feature authentication method according to claim 6, wherein: the step S7 includes the following sub-steps:
s701, the base station B carries out channel information acquisition on the information sender C with unknown identity to obtain a channel information data set
Figure FDA0003149919770000061
Figure FDA0003149919770000062
Wherein, N represents the number of frames,
Figure FDA0003149919770000063
channel information representing the kth OFDM symbol estimate between the base station B and the information sender C of unknown identity, k being 1,2, 3.., N;
s702. for the channel information data set
Figure FDA0003149919770000064
On the basis of the channel information difference value, constructing an amplitude-based test statistic based on the amplitude difference of subcarriers, and processing the test statistic to obtain a normalized LRT statistic
Figure FDA0003149919770000065
As between the base station B and the unknown identity information sender CFirst dimension characteristic of channel information:
Figure FDA0003149919770000066
data set of channel information
Figure FDA0003149919770000067
Bringing in
Figure FDA0003149919770000068
Obtaining M frames of first dimension characteristics of channel information between an unknown identity information sender C and a base station B in the calculation formula:
Figure FDA0003149919770000069
s703. for the channel information data set
Figure FDA00031499197700000610
On the basis of the channel information difference value, constructing test statistic based on amplitude and phase combination, and processing the test statistic to obtain normalized LRT statistic
Figure FDA00031499197700000611
A second dimension characteristic as channel information between the base station B and the unknown identity information sender C:
Figure FDA0003149919770000071
data set of channel information
Figure FDA0003149919770000072
Bringing in
Figure FDA0003149919770000073
In the calculation formula (2), obtaining the unknown identity informationM frames of second dimension characteristics of channel information between sender C and base station B:
Figure FDA0003149919770000074
s704, constructing a two-dimensional joint feature sample set T of channel information between a base station B and an unknown identity information sender CCB：
Figure FDA0003149919770000075
S705, utilizing the classifier with the detection rate reaching the standard to perform T on the sample setCBJudging any frame data to obtain a corresponding y value, wherein the y value is +1 and represents that an information sender C with unknown identity is legal; a value of-1 for y would represent that the sender C of the information of unknown identity is illegal.
CN201810239508.5A 2018-03-22 2018-03-22 Two-dimensional joint feature authentication method based on machine learning Active CN108566642B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810239508.5A CN108566642B (en) 2018-03-22 2018-03-22 Two-dimensional joint feature authentication method based on machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810239508.5A CN108566642B (en) 2018-03-22 2018-03-22 Two-dimensional joint feature authentication method based on machine learning

Publications (2)

Publication Number Publication Date
CN108566642A CN108566642A (en) 2018-09-21
CN108566642B true CN108566642B (en) 2021-08-13

Family

ID=63532052

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810239508.5A Active CN108566642B (en) 2018-03-22 2018-03-22 Two-dimensional joint feature authentication method based on machine learning

Country Status (1)

Country Link
CN (1) CN108566642B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110944002B (en) * 2019-12-06 2020-08-21 深圳供电局有限公司 Physical layer authentication method based on exponential average data enhancement
CN111083156B (en) * 2019-12-25 2021-11-02 中国联合网络通信集团有限公司 Authentication method, apparatus, electronic device and storage medium
CN113076941A (en) * 2021-04-20 2021-07-06 上海阿莱夫信息技术有限公司 Single pointer dial reading identification method based on video frame fusion

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102223637A (en) * 2011-07-20 2011-10-19 北京邮电大学 Identity authentication method and system based on wireless channel characteristic
CN102256249A (en) * 2011-04-02 2011-11-23 电子科技大学 Identity authentication method and equipment applied to wireless network
CN102739402A (en) * 2012-06-06 2012-10-17 天津大学 Strong safety certification method based on HB+ in RFID (Radio Frequency Identification Devices) system

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8726023B2 (en) * 2005-02-03 2014-05-13 Nokia Corporation Authentication using GAA functionality for unidirectional network connections

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102256249A (en) * 2011-04-02 2011-11-23 电子科技大学 Identity authentication method and equipment applied to wireless network
CN102223637A (en) * 2011-07-20 2011-10-19 北京邮电大学 Identity authentication method and system based on wireless channel characteristic
CN102739402A (en) * 2012-06-06 2012-10-17 天津大学 Strong safety certification method based on HB+ in RFID (Radio Frequency Identification Devices) system

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
基于信道信息的无线通信接入认证技术研究;张金玲;《中国优秀硕士学位论文全文数据库》;20180215;第四章 *
对TS 36.300中的NB-IoT描述的更正;英特尔公司，华为，CATT;《3GPP TSG-RAN WG2 Meeting #95 R2-165808》;20160823;全文 *

Also Published As

Publication number Publication date
CN108566642A (en) 2018-09-21

Similar Documents

Publication Publication Date Title
CN108566642B (en) Two-dimensional joint feature authentication method based on machine learning
Pan et al. Threshold-free physical layer authentication based on machine learning for industrial wireless CPS
CN113014524B (en) Digital signal modulation identification method based on deep learning
CN108304877B (en) A method for physical layer channel authentication based on machine learning
Chen et al. Physical‐Layer Channel Authentication for 5G via Machine Learning Algorithm
CN105160678A (en) Convolutional-neural-network-based reference-free three-dimensional image quality evaluation method
CN102148987B (en) Compressed Sensing Image Reconstruction Method Based on Prior Model and l0 Norm
CN109117747A (en) Radar signal classification method based on cyclo-stationary Yu depth convolutional neural networks
CN113726711B (en) OFDM receiving method and device, channel estimation model training method and device
CN112784690B (en) Broadband signal parameter estimation method based on deep learning
CN101848482A (en) Method and device for acquiring interference matrix
CN113284046A (en) Remote sensing image enhancement and restoration method and network based on no high-resolution reference image
CN104392086B (en) A kind of signal deteching circuit and method based on Pearson came rand variate coefficient correlation
CN113704737A (en) Small sample physical layer equipment authentication method, system, terminal and storage medium
Ouyang et al. Channel estimation for underwater acoustic OFDM communications: An image super-resolution approach
Kosuru et al. Digital image steganography with error correction on extracted data
CN111192206A (en) A method to improve image clarity
CN105744548A (en) PCI optimization method and apparatus
Bhuiyan et al. An improved image steganography algorithm based on PVD
CN117081806B (en) A channel authentication method based on feature extraction
CN112240964B (en) Method for identifying fault type of power distribution network
CN114401400B (en) Video quality evaluation method and system based on visual saliency coding effect perception
CN115310476A (en) Multi-sensor combined signal detection and time-frequency positioning method and device
CN114584441A (en) Digital signal modulation identification method based on deep learning
CN114143147A (en) Unauthorized large-scale Internet of things equipment detection method based on deep learning

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
GR01 Patent grant
GR01 Patent grant