Polarization fingerprint authentication model training method and system, and Internet of Things authentication method and system of wireless device

Abstract

The invention relates to a polarized fingerprint authentication model training method and system and an Internet of Things authentication method and system of wireless equipment. The first aspect provides a polarized fingerprint authentication model training method, which comprises the following steps: acquiring a polarized fingerprint, and constructing a polarized fingerprint frequency gradient matrix through the difference of the polarized fingerprint at a plurality of frequency points; decomposing the matrix into slices to obtain a training set; and inputting the training set into a neural network to obtain a trained polarized fingerprint authentication model. The second aspect provides an Internet of Things authentication method of the antenna device, comprising: acquiring a signal of the wireless device, the signal comprising a polarization fingerprint of the wireless device; authenticating the signal through an authentication protocol to obtain an application layer authentication result; authenticating the polarized fingerprint of the wireless equipment through the polarized fingerprint authentication model to obtain a physical layer authentication result; and comprehensively judging the wireless equipment according to the application layer authentication result and the physical layer authentication result.

Description

Technical Field

[0001] The present invention relates to a method for training a polarization fingerprint authentication model, a system, an Internet of Things authentication method and a system for wireless devices, and belongs to the field of computer technology. Background Art

[0002] Currently, there are mainly two measures for ensuring the identity security of wireless Internet of Things systems: one is various firewalls running on the network layer and the transport layer, which are used to filter devices that do not conform to the communication rules; the second is the identity authentication protocol running on the application layer, which verifies the identity of wireless Internet of Things devices by applying various types of encryption protocols. Currently, there is a problem that software protection used by Internet of Things devices at high levels such as the application layer is vulnerable to attacks. Due to low computing power and low-cost limitations, wireless Internet of Things devices usually can only use lightweight encryption algorithms, which are easily cracked by high-performance computing devices. In addition, many manufacturers do not build authentication technology into wireless Internet of Things devices to reduce costs, resulting in a lack of sufficient identity security guarantee for devices in the upper-layer communication protocol.

[0003] Therefore, implementing physical layer-based identity authentication in the underlying communication protocol has attracted wide attention in the academic and industrial fields. Since the defects generated in the manufacturing process of radio device hardware circuits are unique and stable, physical layer radio frequency fingerprints can uniquely characterize device identities, and have randomness and naturalness, making them suitable as authentication credentials. The identity authentication of wireless Internet of Things devices can be achieved through physical layer radio frequency fingerprints. Currently, radio frequency fingerprints are mainly constructed based on the influence of the non-uniformity of radio frequency devices on the time-domain and frequency-domain waveforms of wireless signals.

[0004] Common radio frequency fingerprint technologies for physical layer authentication include transient feature extraction and steady-state feature extraction. The transient part (such as the rising edge and the falling edge) does not carry communication information, and the waveform difference will not be covered by communication information. However, its signal-to-noise ratio is low and the duration is short, resulting in low fingerprint recognition accuracy and difficulty in realizing automation. For steady-state features, due to the small differences in carrier frequency offset and constellation vector offset, they are easily affected by noise, Doppler frequency shift and channel fading, making it difficult to effectively distinguish devices. In addition, with the improvement of radio frequency device manufacturing processes, the radio frequency fingerprint differences between devices of the same model are gradually decreasing, resulting in increased difficulty in discrimination, so that it is increasingly difficult for radio frequency fingerprints to distinguish devices of the same model. Summary of the Invention

[0005] The present invention adopts polarization fingerprint technology, which is easier to extract and has a higher signal-to-noise ratio. At the same time, since the wireless Internet of Things communication system does not modulate the polarization state, the polarization state is not affected by the carried information. Compared with radio frequency fingerprints affected by information, polarization fingerprints are more stable and have a longer duration. Polarization fingerprints can be extracted at all stages of communication, with a large sampling amount, facilitating the realization of automatic recognition, and improving the signal-to-noise ratio through cumulative noise reduction, effectively enhancing the feasibility of discrimination and reducing costs.

[0006] The present invention aims to solve the application problem of polarization fingerprints in actual wireless networks, improve the accuracy of device identification, and reduce the overhead of device identification. The first key point of the present invention lies in the design of the polarization fingerprint frequency gradient (PFG) matrix. By describing the differences between polarization fingerprints at specific frequency points and their neighboring frequency points, the frequency characteristics of polarization fingerprints are effectively highlighted, the influence of polarization fingerprint changes caused by spatial distortion of legitimate devices is reduced, and the robustness of the scheme is improved. At the same time, the random fluctuation ranges of the physical layer characteristics described by polarization fingerprints are different at different frequencies, and the frequency components with a large random fluctuation range can provide higher discrimination. In order to achieve the efficient utilization of polarization fingerprints, it is necessary to highlight the influence of the part with a large random fluctuation range in the polarization fingerprints on the fingerprint recognition result, and weaken the part with a small random fluctuation range. Therefore, the second key point of the present invention lies in the design of the integrated CNN. The integrated CNN highlights the high-discrimination segments and weakens the influence of the low-discrimination segments on recognition by separating the polarization fingerprint PFG matrix respectively.

[0007] To achieve the above technical solution, the first aspect of the present invention provides a method for training a polarization fingerprint authentication model, and its steps include:

[0008] Obtain polarization fingerprints, and construct a polarization fingerprint frequency gradient matrix through the differences of the polarization fingerprints at multiple frequency points;

[0009] Decompose the matrix into slices to obtain a training set;

[0010] Input the training set into a neural network to obtain a trained polarization fingerprint authentication model.

[0011] Furthermore, the number of slices is equal to the number of frequency points of the corresponding polarization fingerprint.

[0012] Furthermore, the neural network is constructed by the following method:

[0013] Construct CNN sub-networks equal to the number of slices, and the sub-networks include a convolutional layer, a max pooling layer, an average pooling layer, a fully connected layer, and a data layer;

[0014] Perform soft voting on the results of the sub-networks, and use the results of the soft voting as the output results of the neural network model.

[0015] The second aspect of the present invention provides a system for training a polarization fingerprint authentication model, including:

[0016] A matrix construction module for obtaining polarization fingerprints and constructing a polarization fingerprint frequency gradient matrix through the differences of the polarization fingerprints at multiple frequency points;

[0017] A matrix slicing module for decomposing the matrix into slices to obtain a training set;

[0018] A data input module for inputting the training set into a neural network to obtain a trained polarization fingerprint authentication model.

[0019] Furthermore, the number of slices is equal to the number of frequency points of the corresponding polarization fingerprint.

[0020] Furthermore, the neural network is constructed by the following method:

[0021] Construct CNN sub-networks equal to the number of slices, and each sub-network includes a convolutional layer, a max pooling layer, an average pooling layer, a fully connected layer, and a data layer;

[0022] Perform soft voting on the results of the sub-networks, and use the result of the soft voting as the output result of the neural network model.

[0023] The third aspect of the present invention provides an Internet of Things authentication method for an antenna device, and its steps include:

[0024] Obtain the signal of the wireless device, and the signal includes the polarization fingerprint of the wireless device;

[0025] Authenticate the signal through an authentication protocol to obtain an application layer authentication result;

[0026] Authenticate the polarization fingerprint of the wireless device through the above polarization fingerprint authentication model to obtain a physical layer authentication result;

[0027] Perform a comprehensive determination on the wireless device through the application layer authentication result and the physical layer authentication result.

[0028] Furthermore, the comprehensive determination includes an application layer authentication result determination and a physical layer authentication result determination, and the physical layer authentication result determination includes:

[0029] Construct or obtain a device identity database;

[0030] Match the physical layer authentication result with the device identity database to complete the physical layer authentication result determination.

[0031] The fourth aspect of the present invention provides an Internet of Things authentication system for a wireless device, including:

[0032] A data acquisition module for acquiring the signal of the wireless device, and the signal includes the polarization fingerprint of the wireless device;

[0033] An application layer authentication module for authenticating the signal through an authentication protocol to obtain an application layer authentication result;

[0034] A physical layer authentication module is used to authenticate the polarization fingerprint of the wireless device through the polarization fingerprint authentication model described in claim 1, and obtain a physical layer authentication result;

[0035] A comprehensive determination module is used to comprehensively determine the wireless device based on the application layer authentication result and the physical layer authentication result.

[0036] Further, the comprehensive determination includes an application layer authentication result determination and a physical layer authentication result determination. The physical layer authentication result determination includes:

[0037] Construct or obtain a device identity database;

[0038] Match the physical layer authentication result with the device identity database to complete the physical layer authentication result determination.

[0039] The beneficial effects of the present invention are as follows:

[0040] (1) In the existing wireless Internet of Things communication system, the modification required for applying polarization fingerprints is small. Since the transmitted polarization state is completely determined by the antenna and does not involve devices before the RF front-end, applying polarization fingerprints in the existing wireless Internet of Things system only requires replacing the antenna, while RF fingerprints are usually affected by devices before the RF front-end.

[0041] (2) The extraction of polarization fingerprints is more difficult and the signal-to-noise ratio is higher. Since the current wireless Internet of Things communication system does not modulate the polarization state, the polarization state will not be affected by the carried information. Compared with RF fingerprints that are usually affected by the carried information, polarization fingerprints are more stable and have a longer duration in wireless signals. Therefore, the extraction of polarization fingerprints is not only less difficult but can also be carried out at all stages of communication. Because the polarization fingerprint has a long duration, a large sampling amount can be achieved for the polarization fingerprint, and noise reduction can be achieved through accumulation to improve the signal-to-noise ratio.

[0042] (3) The universality of polarization fingerprints is better. Polarization fingerprints are not affected by the modulation method. For wireless Internet of Things communication systems that may adopt different modulation methods, the extraction method of polarization fingerprints is the same, while the RF fingerprints of different wireless communication systems often require designing specific extraction algorithms. Therefore, polarization fingerprints have better universality.

[0043] (4) Polarization fingerprints have unique spatial domain characteristics. Polarization fingerprints describe the physical layer characteristics in the electric field intensity vector of wireless signals. Therefore, compared with RF fingerprints, polarization fingerprints have unique spatial domain characteristics, which makes the polarization fingerprints of the same model of devices in different spatial domains show obvious differences, effectively solving the problem of too small differences between the RF fingerprints of the same model of devices, resulting in a high recognition error rate. Description of the Drawings

[0044] Figure 1 Schematic diagram for constructing the polarization fingerprint frequency gradient matrix of the present invention.

[0045] Figure 2 Schematic diagram of the structure of the basic CNN of the present invention.

[0046] Figure 3 Schematic diagram of the structure of the integrated CNN of the present invention.

[0047] Figure 4 Schematic diagram of the identity authentication scheme of the present invention. Detailed implementation manners

[0048] The present invention will be further described in detail below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not intended to limit the scope of the present invention.

[0049] In this embodiment, the polarization fingerprint refers to the element extracted from the signal of the antenna after surface grooving. The surface of the antenna is grooved, so that the polarization state of the antenna is changed. The frequency range of the signal is f1, f2, …, f n , then at the frequency point f k , the complex polarization state can be expressed as

[0050]

[0051] According to the values of the complex polarization state of the signal at different frequency points, the polarization fingerprint of the antenna is expressed as:

[0052]

[0053] In this embodiment, the wireless device refers to various devices that realize data transmission and connection through wireless communication technology.

[0054] This embodiment proposes a polarization fingerprint frequency gradient (PFG) matrix, which effectively highlights the frequency characteristics of the polarization fingerprint by describing the difference between the polarization fingerprint at a specific frequency point and its neighboring frequency points, reduces the influence of the polarization fingerprint change caused by the spatial distortion of the legitimate device, and improves the robustness. As Figure 1 shown, the PFG matrix describes the difference between the polarization fingerprint at a specific frequency point and its neighboring frequency points. Specifically, for device q, the PFG matrix can be defined as:

[0055] PFG q =(g q1 ,g q2 ,…g qm ) T

[0056] where m is the number of all frequency points of the PF of device q, g qkThe PF difference of frequency segment k between frequency i and its 2r+1 adjacent frequency points is defined as:

[0057] g qk ={p(f ik )-p(f jk ),i-r≤j≤i+r}

[0058] In summary, the PFG matrix PFG corresponding to the polarization fingerprint of device n q is an m×(2r+1) matrix. Among them, r is preset by the user for the recognition module according to the composition of the polarization fingerprint frequency points (usually r = 2). Although the polarization fingerprint will change due to different spatial positions, its frequency characteristics are relatively stable. Therefore, the PFG matrix can effectively highlight the frequency characteristics of the polarization fingerprint. Since the fingerprint database usually stores the template fingerprints under a specific spatial distribution, and the spatial position of the legitimate device may change in actual applications, the polarization fingerprint processed by the PFG matrix enhances the tolerance to the changes in the polarization fingerprint caused by spatial distortion.

[0059] In this embodiment, according to the above PFG matrix, a polarization fingerprint authentication model and its training method are proposed.

[0060] In this embodiment, it is implemented using a CNN network. First, a basic CNN is designed, and its structure is as Figure 2 shown and the sizes of each layer in the basic CNN are determined according to the PFG structure, including three convolutional layers, two max pooling layers, one average pooling layer, a fully connected layer, and an output layer.

[0061] In this embodiment, according to the uneven feature distribution characteristics of the polarization fingerprint in the frequency domain, the PFG matrix is decomposed into n slices, and an ensemble CNN is designed. n is consistent with the number of frequency points of the polarization fingerprint.

[0062] In this embodiment, the slices are used to form a slice training sample set for training the ensemble CNN.

[0063] In this embodiment, the ensemble CNN is composed of n sub-networks fused by soft voting, and its structure is as Figure 3 shown. Each sub-network is trained from the sample set of different slices. The structure of the sub-network is the same as that of the basic CNN, and the input is a one-dimensional array. In order to balance the effectiveness of the results of different sub-networks, the results of all sub-networks are combined through soft voting to obtain the final recognition result. Each element in the weight vector of soft voting is determined proportionally by the variance of the corresponding slice training sample set. The larger the variance of a slice, the larger the range of feature fluctuations and the higher the resolution ability, so the result of the corresponding sub-network is considered more effective.

[0064] In this embodiment, a polarization fingerprint database is constructed according to the received polarization fingerprint to be trained.

[0065] This embodiment provides a comprehensive, secure and flexible authentication solution for Internet of Things (IoT) wireless devices, as Figure 4 shown. The signal of the wireless device is obtained. Based on the existing wireless IoT authentication protocol, a parallel physical layer authentication mechanism based on polarization fingerprints is added. The physical layer authentication result is generated by obtaining and analyzing the polarization characteristic information of the wireless device. At the same time, the system retains and runs the existing application layer authentication protocol to generate the application layer authentication result. The dual authentication results from the physical layer and the application layer are received, and a comprehensive determination is made according to the preset verification algorithm, and finally the identity authentication result is output. This solution effectively improves the security and reliability of the identity authentication of wireless IoT devices through a multi-level authentication mechanism.

[0066] In this embodiment, for the existing application layer authentication protocol, the technical methods include password-based authentication, certificate-based authentication, and digital signature-based authentication.

[0067] In this embodiment, the input polarization fingerprints are classified according to the trained polarization fingerprint authentication model, and the classification result and the corresponding prediction probability are output. The classification result is determined whether to be accepted by comparing the probability value of the prediction result with the threshold. If the classification result is accepted, the classification result is matched with the device identity ID in the polarization fingerprint database. If the match is successful, "registered device" is output, and the physical layer authentication is completed.

[0068] In this embodiment, if the classification result is not accepted by the recognition module or does not match the device identity ID, "unregistered device" is output, and the administrator decides whether to register this device in the polarization fingerprint database and start a new round of training of the polarization fingerprint authentication model.

[0069] In this embodiment, the comprehensive determination refers to the dual authentication judgment of the signal of the wireless device through the application layer authentication and the physical layer authentication. When the device passes the dual authentication, the device passes the identity authentication of the system. If any authentication fails, the device fails the identity authentication of the system. The upper layer protocol of the device is combined with the polarization fingerprint recognition without changing the existing wireless IoT authentication protocol, and a parallel physical layer authentication mechanism based on polarization fingerprints is introduced. Based on the dual identity authentication of the application layer and the physical layer, according to factors such as the system security level, authentication priority, channel state, and delay requirement, it is flexibly selected whether to merge and use or dynamically adjust the weights of the application layer and physical layer recognition in the authentication to complete the final comprehensive identity verification.

[0070] Although specific embodiments of the present invention are disclosed for illustrative purposes, which are intended to help understand the content of the present invention and implement it accordingly, those skilled in the art can understand that various substitutions, changes, and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the present invention should not be limited to the content disclosed in the best embodiments, and the scope of protection claimed by the present invention shall be subject to the scope defined by the claims.

Claims

1. A method for training a polarization fingerprint authentication model, the steps of which include: Obtain polarization fingerprints, and construct a polarization fingerprint frequency gradient matrix through the differences of the polarization fingerprints at multiple frequency points; Decompose the matrix into slices to obtain a training set; Input the training set into a neural network to obtain a trained polarization fingerprint authentication model.

2. The method according to claim 1, wherein The number of the slices is equal to the number of frequency points of the corresponding polarization fingerprint.

3. The method according to claim 1, wherein The neural network is constructed by the following method: Construct CNN sub-networks equal to the number of slices, and the sub-networks include a convolutional layer, a max pooling layer, an average pooling layer, a fully connected layer, and a data layer; Perform soft voting on the results of the sub-networks, and use the results of the soft voting as the output results of the neural network model.

4. A system for training a polarization fingerprint authentication model, including: A matrix construction module for obtaining polarization fingerprints and constructing a polarization fingerprint frequency gradient matrix through the differences of the polarization fingerprints at multiple frequency points; A matrix slicing module for decomposing the matrix into slices to obtain a training set; A data input module for inputting the training set into a neural network to obtain a trained polarization fingerprint authentication model.

5. The system according to claim 4, wherein The number of the slices is equal to the number of frequency points of the corresponding polarization fingerprint.

6. The system according to claim 4, characterized in that, The neural network is constructed by the following method: Construct CNN sub-networks equal to the number of slices, and the sub-networks include a convolutional layer, a max pooling layer, an average pooling layer, a fully connected layer, and a data layer; Perform soft voting on the results of the sub-networks, and use the results of the soft voting as the output results of the neural network model.

7. An Internet of Things authentication method for an antenna device, the steps of which include: Obtain the signal of the wireless device, and the signal includes the polarization fingerprint of the wireless device; Authenticate the signal through an authentication protocol to obtain an application layer authentication result; Authenticate the polarization fingerprint of the wireless device through the polarization fingerprint authentication model described in claim 1 to obtain a physical layer authentication result; Perform a comprehensive determination on the wireless device through the application layer authentication result and the physical layer authentication result.

8. The method according to claim 7, wherein The comprehensive determination includes an application layer authentication result determination and a physical layer authentication result determination, and the physical layer authentication result determination includes: Construct or obtain a device identity database; Match the physical layer authentication result with the device identity database to complete the physical layer authentication result determination.

9. An Internet of Things authentication system for a wireless device, including: A data acquisition module for obtaining the signal of the wireless device, and the signal includes the polarization fingerprint of the wireless device; An application layer authentication module for authenticating the signal through an authentication protocol to obtain an application layer authentication result; A physical layer authentication module for authenticating the polarization fingerprint of the wireless device through the polarization fingerprint authentication model described in claim 1 to obtain a physical layer authentication result; A comprehensive determination module for performing a comprehensive determination on the wireless device through the application layer authentication result and the physical layer authentication result.

10. The system according to claim 9, wherein The comprehensive determination includes an application layer authentication result determination and a physical layer authentication result determination, and the physical layer authentication result determination includes: Construct or obtain a device identity database; Match the physical layer authentication result with the device identity database to complete the determination of the physical layer authentication result.

Images