• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    The invention discloses a physical layer authentication method based on edge devices, which takes the advantages of the threshold and the machine learning. Firstly, the maximum repetitive authentication number and the authentication threshold of physical layer information packet are set. Then several terminals perform authentication shaking hands with the edge device. After the identity of terminals is confirmed, they access the edge computing system. Every terminal continuously sends packets to the edge device. When the edge device receives the packet sent by the jth terminal at the ith time slot, it extracts the current channel information matrix as the basis of packet authentication. The edge device adopts the threshold method to perform the physical layer information package authentication, while it carries out the machine learning of the physical layer information package authentication at the same time. After the machine learning authentication method is completed, this machine learning method is used for authentication by the edge device. The novel method improves the authentication rate of physical layer information package and avoids long-term training.
    公开号:NL2025648A
    申请号:NL2025648
    申请日:2020-05-22
    公开日:2021-02-01
    发明作者:Wen Hong；Chen Songlin；Song Huanhuan；Tang Jie
    申请人:Sichuan Shaoze Electronic Equipment Co Ltd；
    IPC主号:
    专利说明:

    -1- A physical layer authentication method for edge device combining threshold and machine learning Technical field The invention relates to the security protection of the Internet of Things system, in particular to an edge device physical layer authentication method combining threshold and machine learning. Background technique In order to meet various low-latency application scenarios such as industrial control, unmanned driving, and virtual reality, a new network architecture has emerged--a system architecture based on edge computing, which introduces edges through cloud computing servers and terminal device network layers. Computing equipment. Compared with cloud computing, edge computing brings the nearest data processing, reduces the network transmission volume, response delay, and also makes the security improved. It is called "the last mile of artificial intelligence." At the same time, the security protection of the edge computing system itself has become the key to its application. The secure access of the terminal and the verification of the packet sent by the terminal to the edge device are important links for the security of the edge computing system. Since the terminal is a device with limited energy and computing power, The use of high-strength security measures based on passwords is limited, and the security authentication technology of physical characteristics can make full use of the computing resources of edge devices to achieve lightweight security access and data packet authentication of terminals, which has become a technology that has received widespread attention.
    Summary of the invention The purpose of the present invention is to overcome the shortcomings of the prior art, and to provide a physical layer authentication method for edge devices that combines threshold and machine learning. Machine learning authentication. After completing machine learning authentication, machine learning methods are used for authentication, which improves the physical layer packet authentication rate and avoids long training.
    -2-
    The purpose of the present invention is achieved by the following technical solution: A method for authentication of the physical layer of an edge device combining threshold and machine learning, including the following steps:
    S1.Set CF as the maximum number of repetitions of physical layer packet authentication, 771s the authentication threshold of physical layer packet;
    S2. Several terminal L and the edge device R perform an authentication handshake, after the terminal identity is confirmed, access the edge computing system, and each terminal continuously sends information packets to the edge device;
    S3. The edge device extracts the current channel information matrix H : : when receiving the information packet sent by the {&% terminal at the {& moment;
    S4. Determine whether there is currently an authentication classifier trained using machine learning algorithms;
    If yes, go to step S8;
    If not, use the threshold authentication method to authenticate the extracted channel information matrix according to the set authentication threshold of the physical layer packet, and proceed to step S5;
    S5. When the threshold authentication method succeeds in authentication, the information packet is considered legitimate and the information packet is received;
    Threshold authentication method authentication is unsuccessful, the packet is considered illegal, and it is judged whether the maximum number of repetitions of authentication CF is reached:
    Reach the maximum number of repetitions of authentication CF, negate and discard the packet;
    If the maximum repetition number CF of authentication is not reached, return to step S3 for the next channel information extraction; S6. When the extracted channel information matrix H : : completes the packet authentication, the sample set is constructed as follows: TH = diff (RY, HE), i=2,0 Mj =1.2,-N, Among them, diff (Ht n SHE ) calculates the difference between H i and H i M is the maximum number of packets sent continuously, and N is the
    -3- number of terminals; H a ; represents the channel information matrix extracted by the edge device when the information packet sent by the §& terminal at time { — 2; H i represents the channel information matrix extracted by the edge device when the information packet sent by the ;#% terminal at time ; According to the authentication result of the threshold authentication method, Tf, i=2,---,M,j=12,---N is combined into a binary classification data set T={(x.y)(x.)....(x,.¥,).-}, in the training data set7 : x, = T : LL {+1, Successful threshold owthemtiontion 0 fm = 4, Thrashold omissie folies PELE That is, when y, = +1, it means that the corresponding data packet is legal; when y, = —1,it means that the corresponding data packet is illegal; K items in the data setT with y =+1 constitute the test set 7,4, , and the rest constitute the training set 7, ; S7. Use the classification algorithm in the machine learning algorithm to train according to the binary classification training data set 7}, to generate a classifier, and use the test set 7; to test the generated classifier. When the error rate E is less than a specific value £, the training is completed Certification classifier; Duetoy, =+1 ine, all the data in the set are successful in threshold authentication. If the classifier determines that the data packet corresponding to a certain data is an illegal data packet, it represents a judgment error, and divides the number of misjudged data by the total number of data in test set 7-4 we can get the error rate E.
    S8. Use the authentication classifier trained by the machine learning algorithm to authenticate the channel information matrix and determine whether the corresponding information packet is legal: If the packet is legal, receive the packet; If the information packet is illegal, it is determined whether the maximum number of repetitions of authentication CF is reached. When the maximum number of repetitions is reached, the information packet is rejected and discarded. If the maximum number of repetitions is not reached, the process returns to step S3 for the
    -4- next channel information extraction.
    The beneficial effects of the present invention are: (1) The present invention retains the results of the physical layer packet authentication by the edge device using the threshold method, and performs the machine learning authentication of the physical layer packet authentication, avoiding the long time offline that requires a large number of samples; (2) After completing machine learning authentication, the present invention adopts machine learning method for authentication, which improves the physical layer packet authentication rate; (3) The method proposed by the present invention has no restrictions on the terminal devices accessing the edge device and has good compatibility performance.
    Instruction with figures Figure 1 1s a flowchart of the method of the present invention; Figure 2 is a schematic diagram of the principle of the threshold authentication method.
    Specific implementation mode The technical solutions of the present invention are described in further detail below with reference to the drawings, but the protection scope of the present invention is not limited to the following.
    As shown in Figure 1, a physical layer authentication method for edge devices that combines threshold and machine learning includes the following steps: S1.Set CF as the maximum number of repetitions of physical layer packet authentication, 77 is the authentication threshold of physical layer packet; S2. Several terminal L and the edge device R perform an authentication handshake, after the terminal identity is confirmed, access the edge computing system, and each terminal continuously sends information packets to the edge device; S3. The edge device extracts the current channel information matrix H : : when receiving the information packet sent by the J£& terminal at the £&& moment; S4. Determine whether there is currently an authentication classifier trained using machine learning algorithms; If yes, go to step S8;
    -5-
    If not, use the threshold authentication method to authenticate the extracted channel information matrix according to the set authentication threshold of the physical layer packet, and proceed to step S5;
    S5. When the threshold authentication method succeeds in authentication, the information packet is considered legitimate and the information packet is received,
    Threshold authentication method authentication is unsuccessful, the packet 1s considered illegal, and it is judged whether the maximum number of repetitions of authentication CF is reached:
    Reach the maximum number of repetitions of authentication CF, negate and discard the packet;
    If the maximum repetition number CF of authentication is not reached, return to step S3 for the next channel information extraction,
    S6. When the extracted channel information matrix H i completes the packet authentication, the sample set is constructed as follows:
    LR or YY LR TLR} , T,; =diff Hg, H),i=2-M,j=12,---N,
    Among them, diff (Ri I.
    SH) calculates the difference between H a ;and H : , M is the maximum number of packets sent continuously, and N is the number of terminals; H oi ; represents the channel information matrix extracted by the edge device when the information packet sent by the j&& terminal at time i — 2;
    H i represents the channel information matrix extracted by the edge device when the information packet sent by the j&& terminal at time {;
    According to the authentication result of the threshold authentication method, Tf, i=2,---,M,j=12,---N is combined into a binary classification data set T={(x.y)(x.)....(x,.¥,).-}, in the training data set7 :
    x, =TH, {+1, Jucesssful threshold ontkennen mee xe ii, Thrashold omissie folies
    That 1s, when y, =+1, it means that the corresponding data packet is legal;
    when y, = —1,it means that the corresponding data packet is illegal; K items in the data
    -6- setT with y =+1 constitute the test set 7, , and the rest constitute the training set 7, ; S7. Use the classification algorithm in the machine learning algorithm to train according to the binary classification training data set 7,, to generate a classifier, and use the test set 7; to test the generated classifier. When the error rate E is less than a specific value g, the training is completed Certification classifier; Due toy, =+1 inlc, all the data in the set are successful in threshold authentication. If the classifier determines that the data packet corresponding to a certain data is an illegal data packet, it represents a judgment error, and divides the number of misjudged data by the total number of data in test set 7, we can get the error rate E.
    S8. Use the authentication classifier trained by the machine learning algorithm to authenticate the channel information matrix and determine whether the corresponding information packet is legal: If the packet is legal, receive the packet; If the information packet is illegal, it is determined whether the maximum number of repetitions of authentication CF is reached. When the maximum number of repetitions is reached, the information packet is rejected and discarded. If the maximum number of repetitions is not reached, the process returns to step S3 for the next channel information extraction.
    In the embodiments of the present application, the extraction method of the channel information matrix may be an LS (Least Square) channel information estimation method, or a MMSE (Minimum Mean Square Error) channel information estimation method, or may be other channel information estimation methods; Classifiers trained through machine learning algorithms include but are not limited to AdaBoost classifier, bagging classifier and boosting classifier; The specific value ¢ is preset according to the working conditions of the system, and the value of £ satisfies: € <1%,; The value of Kin the step S6 satisfies the following condition: M-N/10<k<M-N/2.
    As shown in FIG. 2, in the embodiment of the present application, the specific principle of the threshold authentication method is as follows: First, suppose that the edge device R and the legal access node L have "handshake", the edge device R has completed the storage of the first frame data packet, extract the channel estimation
    -7- matrix H - on the legal path and line it in the time domain Averaging and extracting channel information (CSI, channel state information) H oe Hf (LL) L H(LN,) HI" (1) A HF (20) L Hf (2.N) | gee _| H5(2) (1) M M M =! M Hf (NI) L Hf(N,N,) HN) Then, after receiving the second data frame, R also obtains the second frame H oF from the unknown node U to be accessed according to equation (1), and compares H 3 and H - . Assuming that there is a strong correlation between consecutive data frames, the time interval between two frames is within the coherent time, if the two are close, it means that the sender is a legitimate access node L; If the two are not close to each other, it indicates that the node to be accessed is an illegal access node E, and an authentication is completed.
    Suppose R has successfully authenticated that the ¥&& data frame is sent by L, and the error-free channel information is H,"; the identity of the sender of the & 3 ita data frame is unclear as U, and the error-free channel information is H on The following hypothesis can be made through the binary hypothesis test [55]: Lr UR _ LR H, : Ha - H, 2) + . IJ UR LR = H, : Hi, + H, The null hypothesis $£g: the error-free channel information in frame & + ft% and the error-free channel information in frame #&% are equal, and the node to be accessed is the legal access node L; Alternative hypothesis +; : The error-free channel information in frame & 2 1£& and the error-free channel information in frame &&% are not equal, and the node to be accessed is an illegal access node E. In formula (2), H A and H EE are error-free channel information, and the actual obtained channel information estimates Hand ZEE : B; = H* +W, (3)
    UR UR Hia = Hea + Wa
    -8- Among them, W, and¥. 4; are independent and identically distributed complex Gaussian noises, subject to &¥ (S, 2) distribution.
    In order to reduce the complexity of authentication, the two equations in equation (3) are subtracted, the physical layer authentication based on channel characteristics is changed to the comparison of the "difference" of the channel information estimation value between consecutive frames, and the equation (3) becomes: H,: diff (Zi , ar) <7 © H,: diff (B.
    A) > In the above formula, diff (4 UR IR ) represents the method used to calculate the "difference" between H," and ZJ, and 77 is the threshold.
    Zero hypothesis H , : the difference between the channel information in frame & + 1¢&% and the channel information in frame &&% is less thanthreshold 77 , which is the legal access node L; Alternative hypothesis#H, : The difference between the channel information in frame & + 3&& and the channel information in frame &&% is greater than threshold 77 , which is an illegal access node E; If the channel information "difference" between consecutive frames is abbreviated as T , then equation (4) becomes: # 0 T=dijf (Bf HA) (5) H, In the general scheme of physical layer packet authentication, 77 is the threshold, which is generally determined by experiments or empirical methods.
    The general scheme of physical layer packet authentication 1s also called the threshold method of physical layer packet authentication, but the accuracy rate of this authentication method is currently needs improvement.
    Some scholars have proposed to use the machine learning method instead of the threshold decision method of formula (5), but it requires large label samples and long training time.
    Therefore, this patent proposes a combination of a threshold and a machine
    -9- learning edge device physical layer authentication method, which retains the results of the edge device using the threshold method for physical layer packet authentication, and performs machine learning authentication for physical layer packet authentication, which avoids It takes a long time for a large number of samples to be offline; after completing machine learning authentication, the machine learning method is used for authentication, which improves the physical layer packet authentication rate; at the same time, the method proposed by the present invention has no restrictions on terminal devices that access edge devices, and Compatibility.
    The above is the preferred embodiment of the present invention, it should be understood that the present invention is not limited to the form disclosed herein, and should not be seen as an exclusion from other embodiments, but can be used in other combinations, modifications and environments, and can be Within the scope of the concept, the above teaching or related field of technology or knowledge can be modified.
    Changes and changes made by those skilled in the art without departing from the spirit and scope of the present invention should fall within the scope of protection of the appended claims of the present invention.
    权利要求:
    Claims (5)
    [1]
    1. An authentication method at the physical layer of an edge device that combines threshold and machine learning, characterized by the following steps: SI. CF is set as the maximum number of iterations of the packet authentication on the physical layer, 77 is the authentication threshold of the packet on the physical layer; S2. Several terminals L and the edge device R perform an authentication handshake after the identity of the terminal is confirmed, connect to the edge computer system, and each terminal continuously sends information packets to the edge device; S3. The edge device obtains the current channel information matrix H i when it receives the information packet sent by jth number ith. S4. It is determined whether there is currently an authentication classifier suitable for the learning algorithms; If so, continue with step S8; If not, the threshold authentication method is used to authenticate the obtained channel information matrix, according to the set authentication threshold of the packet on the physical layer. Proceed to step SS; S5. If the authentication of the threshold authentication method succeeds, the information package is considered legitimate and the information package is received; If the authentication of the threshold authentication method fails, the packet is illegal and it is assessed whether the maximum number of authentication repetitions, CF, has been reached: When the maximum number of authentication repetitions, CE, is reached, the packet is ignored and deleted, If the maximum number of iterations of authentication, C '# has not yet been reached, go back to step S3 to obtain the channel information again; S6. If the obtained channel information matrix H i completes the packet authentication, then
    11 - a sample set is composed as follows: TF = diff (HX, JH) i = 2,, M, j = 12, N, Here diff (B 0 JH) calculates the difference between HE, and H; ‚M is the maximum number of packets continuously transmitted and N is the number of terminals, HX ui represents the channel information matrix obtained by the edge device when the information packet, which was sent by jth number i - 1. H iN represents the channel information matrix that was obtained by the edge device when the information packet, which was sent by jth number i.
    According to the authentication result of the threshold authentication method, Ti = 2, M, j = 12,: N is combined into a binary classification data set T = {(x. 3) 0 Jo) Va) "+}, In the training data set 7: xX, = Te; Jor = i Threshold Authentication Success CS m = 12..M- Ns —1, Threshold Authentication Failed That is, if y, = + 1, then the corresponding data packet is legal and if y, = -1, then it is corresponding data packet illegal, K items in the data set T with y = + 1 make up the test set 7p, and the rest make up the training set 7, S7. The classification algorithm in the machine learning algorithm is used to learn according to the binary classification training data set 7 ,, to generate a classification and use the test set 7.4 to test the generated classification If the error rate E is less than a specific value g, the training is completed.
    Certification Classification; Because y ,, = + 1 in 7, all data in the set has passed the threshold authentication. The classification decides that the data package belongs to that
    - 12 - if certain data is an illegal data package, then there is a judgment error and the number of incorrectly evaluated data is divided by the total number of data in the 7-4, The quotient is the error rate E.
    S8. The authentication classification, learned from the machine learning algorithm, is used to authenticate the channel information matrix and determine whether the corresponding information packet is legal: If the packet is legal, the packet is received; If the information package is illegal, it is determined whether the maximum number of iterations of the authentication, CF, has been reached. If so, the information package will be rejected and removed. If the maximum number of repetitions has not been reached, the process returns to step S3 to obtain the next channel information.
    [2]
    2. The physical layer authentication method for an edge device according to claim 1 combines threshold and machine learning, while in step S3, the method for obtaining the channel information matrix includes the LS channel information estimation method and the MMSE channel information estimation method, without up to those estimation methods have to be limited.
    [3]
    3. The method for authenticating the physical layer of an edge device according to claim 1 combines threshold and machine learning, while in step S4 the classification trained by the machine learning algorithms contains the AdaBoost classification, the Bagging classification and the boosting classification. without being limited to those classifications.
    [4]
    4. The authentication method on the physical layer of an edge device combines according to claim 1 threshold and machine learning, while in step S7 the specific value ¢ is set according to the operating conditions of the system and the value of e-Compliance: e <1 %.
    - 13 -
    [5]
    5. The authentication method on the physical layer of an edge device according to claim 1 combines threshold and machine learning, while the value of K in step S6 satisfies the condition: M -N / 10 <K <M -N / 2.
    类似技术:
    公开号 | 公开日 | 专利标题
    Rigaki et al.2018|Bringing a gan to a knife-fight: Adapting malware communication to avoid detection
    Morimae2014|Verification for measurement-only blind quantum computing
    CN108932535B|2019-07-02|A kind of edge calculations clone&#39;s node recognition methods based on machine learning
    CN108304877B|2021-10-08|Physical layer channel authentication method based on machine learning
    CN108924836B|2019-07-02|A kind of edge side physical layer channel authentication method based on deep neural network
    Jankowski et al.2016|On efficiency of selected machine learning algorithms for intrusion detection in software defined networks
    Tang et al.2020|MF-Adaboost: LDoS attack detection based on multi-features and improved Adaboost
    CN107124327B|2019-04-02|The method that JT808 car-mounted terminal simulator reverse-examination is surveyed
    Palau et al.2020|DNS tunneling: A deep learning based lexicographical detection approach
    CN104935548B|2019-05-24|Auth method, apparatus and system based on intelligent equipment of tatooing
    NL2025648A|2021-02-01|A physical layer authentication method for edge device Combining threshold and machine learning
    WO2018057691A1|2018-03-29|Unsupervised classification of web traffic users
    Chen et al.2011|Detecting bots via incremental LS-SVM learning with dynamic feature adaptation
    Acar et al.2019|A privacy‐preserving multifactor authentication system
    CN109614780A|2019-04-12|Biological information certifying method and device, storage medium, electronic equipment
    Wu et al.2010|Detecting DDoS attack towards DNS server using a neural network classifier
    Abdalla et al.2021|Algebraic adversaries in the universal composability framework
    CN108600145A|2018-09-28|A kind of method and device of determining ddos attack equipment
    Muntean et al.2010|A novel intrusion detection method based on support vector machines
    CN110381509B|2022-01-11|Combined authentication method and server suitable for dynamic connection scene
    Man et al.2021|Intelligent intrusion detection based on federated learning for edge-assisted internet of things
    CN110944002A|2020-03-31|Physical layer authentication method based on exponential average data enhancement
    KR20190028880A|2019-03-20|Method and appratus for generating machine learning data for botnet detection system
    US10805318B2|2020-10-13|Identification of a DNS packet as malicious based on a value
    Sun et al.2017|A distinction method of flooding DDoS and flash crowds based on user traffic behavior
    同族专利:
    公开号 | 公开日
    CN110149342B|2020-01-10|
    AU2020100826A4|2020-07-16|
    NL2025648B1|2021-10-05|
    CN110149342A|2019-08-20|
    LU101814B1|2020-09-23|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    CN101848479B|2010-04-09|2012-08-08|西安电子科技大学|Simulation system and simulation method of wireless network security protocol|
    US9848442B2|2014-11-10|2017-12-19|Newracom Inc.|Method for transmitting and receiving frame in wireless local area network|
    EP3226641A4|2014-11-19|2018-09-05|Newracom, Inc.|Method and apparatus for processing ppdu based on bbs identification information in high efficiency wireless lan|
    CN107046468B|2017-06-14|2020-10-02|电子科技大学|Physical layer authentication threshold determination method and system|
    CN108304877B|2018-02-02|2021-10-08|电子科技大学|Physical layer channel authentication method based on machine learning|
    法律状态:
    优先权:
    申请号 | 申请日 | 专利标题
    CN201910463221.5A|CN110149342B|2019-05-30|2019-05-30|Edge device physical layer authentication method combining threshold and machine learning|
    [返回顶部]