A physical layer key generation method and device based on multi-task autoencoder

Through the multitasking autoencoder network, the channel estimation value is processed, the reciprocity characteristics are generated and quantified into physical layer keys, which solves the problems of high key inconsistency rate and leakage risk caused by weak channel correlation in the prior art, and improves the Communication security and key generation efficiency.

CN115134072BActive Publication Date: 2024-11-26TONGJI UNIV
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202210376521.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-11
Publication Date
2024-11-26
Estimated Expiration
2042-04-11

AI Technical Summary

Technical Problem

The existing physical layer security key generation mechanism is based on the assumption of strong channel correlation. In actual scenarios, weak channel correlation leads to a high key inconsistency rate and low generation rate. Traditional methods increase the risk of key leakage, especially when eavesdroppers approach legal communication nodes. Not applicable when it is.

Method used

Using a multi-task autoencoder network, through the gradient inversion layer between the encoder and the second decoder, combined with the Adam optimizer and the Leaky ReLU activation function, the channel estimates of the sending, receiving and eavesdropping ends are trained to generate reciprocal and quantified into physical layer keys.

Benefits of technology

It improves communication security, reduces the key inconsistency rate and risk of leakage, enhances the randomness and generation rate of keys, and is suitable for complex scenarios of wireless channels.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115134072B_ABST
    Figure CN115134072B_ABST
Patent Text Reader

Abstract

The present invention relates to a physical layer key generation method and device based on a multi-task autoencoder, the method comprising obtaining signal estimation values ​​of a transmitter, a receiver and an eavesdropper; inputting the channel estimation values ​​of the three into a multi-task autoencoder network for training, stopping the training when the loss function converges, and obtaining a trained multi-task autoencoder network; installing the trained multi-task autoencoder network into local processors of the transmitter and the receiver; obtaining the output of the hidden layer of the multi-task autoencoder network, obtaining the reciprocity characteristics of the transmitter and the receiver, and obtaining the physical layer key after quantization. Compared with the prior art, the present invention has the advantages of strong security and the like.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of communication security, and in particular to a physical layer key generation method and device based on a multi-task autoencoder. Background Art

[0002] Most of the existing physical layer security key generation mechanisms are based on the assumption of strong channel correlation, which is not appropriate for actual scenarios. Due to the influence of detection interval, channel estimation deviation and multipath effect, the channels observed by the two communicating parties are not highly correlated. The physical layer key generation under weakly correlated channels will have a great impact on the key inconsistency rate and generation rate. Based on the extraction of channel reciprocity features, the problem of low key inconsistency rate caused by weak channel correlation can be weakened to a certain extent, and the randomness of the key can be enhanced. In current research, there are few studies on efficient nonlinear transformation methods. Most schemes use linear transformation methods to transform domain mapping and dimension reduction of the obtained channel eigenvalues. However, in linear schemes, the selection of transformation matrices is not based on the characteristics of the data itself, resulting in the loss of some information in the original data, the inability to efficiently utilize the channel state information (CSI), and the low utilization rate of mutual information, which affects the security key capacity. At the same time, in the traditional physical layer key generation mechanism, in order to reduce the key inconsistency rate, most current studies use key negotiation mechanisms or the idea of ​​reciprocity information transmission, which will greatly increase the risk of key leakage.

[0003] In order to solve the problem of low security key capacity and key leakage under linear transformation, a nonlinear extraction method of channel reciprocity features using the multi-task autoencoder network architecture in deep learning was proposed. However, in practical applications, due to the uncertainty of the location, random fading and broadcast characteristics of the wireless channel, the channel conditions of the eavesdropper can be comparable to or even better than those of the legitimate communication parties, especially when the eavesdropper is very close to the legitimate communication node, even within half a wavelength, the eavesdropper's channel may have a high correlation with the legitimate communication channel. At this time, the traditional physical layer key extraction method is not applicable. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a physical layer key generation method and device based on a multi-task autoencoder.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A physical layer key generation method based on a multi-task autoencoder comprises the following steps:

[0007] S1, obtaining signal estimation values ​​of the transmitting end, the receiving end and the eavesdropping end;

[0008] S2, input the channel estimation values ​​of the three into the multi-task autoencoder network for training, and when the preset training requirements are met, stop the training to obtain a trained multi-task autoencoder network;

[0009] S3, installing the trained multi-task autoencoder network into the local processors of the transmitter and receiver respectively;

[0010] S4, obtaining the output of the hidden layer of the multi-task autoencoder network, obtaining the reciprocity characteristics of the transmitter and the receiver, and obtaining the physical layer key after quantization;

[0011] The multi-task autoencoder network includes an encoder, a first decoder and a second decoder, and a gradient reversal layer is arranged between the encoder and the second decoder.

[0012] Furthermore, the training steps of the multi-task autoencoder network are as follows:

[0013] The channel estimation value of the transmitting end is used as the input of the encoder, the channel estimation value of the receiving end is used as the output of the first decoder, and the channel estimation value of the eavesdropping end is used as the output of the second decoder to train the multi-task autoencoder network.

[0014] Furthermore, the activation function of the multi-task autoencoder network is Leaky ReLU.

[0015] Furthermore, the multi-task autoencoder network is optimized using the Adam optimizer.

[0016] Furthermore, the method for obtaining the signal estimation value of the eavesdropping end is as follows:

[0017] The correlation matrix between the channel characteristics of the transmitting end and the receiving end is calculated, and the correlation matrix between the transmitting end and the eavesdropping end is obtained by multiplying the correlation matrix by the correlation coefficient. The signal estimation value of the eavesdropping end is generated according to the correlation matrix between the transmitting end and the eavesdropping end.

[0018] A physical layer key generation device based on a multi-task autoencoder comprises a memory and a processor; the memory is used to store a computer program; the processor is used to implement the following method when executing the computer program:

[0019] S1, obtaining signal estimation values ​​of the transmitting end, the receiving end and the eavesdropping end;

[0020] S2, input the channel estimation values ​​of the three into the multi-task autoencoder network for training, and when the preset training requirements are met, stop the training to obtain a trained multi-task autoencoder network;

[0021] S3, installing the trained multi-task autoencoder network into the local processors of the transmitter and receiver respectively;

[0022] S4. Obtain the output of the hidden layer of the multi-task autoencoder network, obtain the reciprocity characteristics of the transmitter and the receiver, and obtain the physical layer key after quantization.

[0023] Compared with the prior art, the present invention has the following advantages:

[0024] 1. The present invention adopts a multi-task autoencoder network to perform training according to the channels of the transmitting end, the receiving end and the eavesdropping end. By setting a gradient reversal layer between the encoder and the second decoder, due to the existence of the gradient reversal layer between them, the parameters of the encoder will be optimized in the direction of increasing the difference between the input and the output, so that the hidden layer features generated by the encoder contain fewer components related to the eavesdropping end, generate a physical layer key, and thus improve the security of the overall communication.

[0025] 2. The present invention uses the Adam optimizer to optimize the multi-task autoencoder network, which is computationally efficient and simple to implement. Leaky ReLU is used as the activation function of the multi-task autoencoder network. Compared with the traditional ReLU activation function, it solves the problem of neuron necrosis during runtime and further improves the stability of the multi-task autoencoder network operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic diagram of the process of the present invention.

[0027] Figure 2 This is the multi-task autoencoder network architecture training diagram of the present invention.

[0028] Figure 3 This is a network parameter configuration diagram of the multi-task autoencoder of the present invention.

[0029] Figure 4 This is a configuration diagram of the simulation experiment parameters of the present invention.

[0030] Figure 5 It is a comparison diagram of the mutual information between the present invention and the principal component analysis method.

[0031] Figure 6 This is a comparison diagram of mean square error between the present invention and the principal component analysis method.

[0032] Figure 7 A comparison diagram of the communication node correlation between the present invention and the conventional multi-task autoencoder network.

[0033] Figure 8 The figure is a flow chart of the present invention applied to base station user communication. DETAILED DESCRIPTION

[0034] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0035] This embodiment provides a physical layer key generation method based on a multi-task autoencoder, such as Figure 1 As shown, the specific steps include:

[0036] Step S1: Obtain signal estimation values ​​of the transmitting end, the receiving end and the eavesdropping end, and extract amplitude and phase thereof.

[0037] Step S2: The estimated signal value of the transmitting end As input, the signal estimate at the receiving end and the signal estimate of the eavesdropping end As the target value, the multi-task autoencoder network training is input. When the loss function, network weight and activation function all reach the ideal value, that is, when the preset training requirements are met, the training is stopped to obtain the trained multi-task autoencoder network.

[0038] Step S3: Install the trained multi-task autoencoder network into the local processors of the transmitting end and the receiving end.

[0039] Step S4, obtain the output z of the hidden layer of the multi-task autoencoder network, obtain the reciprocity characteristics of the transmitter and the receiver, and obtain the physical layer key after quantization.

[0040] Unlike ordinary autoencoders, the autoencoder network used in this embodiment retains the original encoder (feature extractor or encoder) structure while increasing the number of decoders to two, namely, the first decoder decoder b With the second decoder e , and in the encoder and decoder e Add a gradient reversal layer. The gradient reversal layer will cause the gradient to be multiplied by a negative constant after passing through it during back propagation training. The training process is as follows Figure 2 As shown, the final multi-task autoencoder network parameter configuration is as follows Figure 3 As shown in the figure, the Adam optimizer is used to optimize the network, and Leaky ReLU is used as the activation function of the network.

[0041] In this embodiment, the method for obtaining the signal estimation value of the eavesdropping end is as follows:

[0042] This embodiment assumes that the eavesdropping end is close to the transmitting end of the legitimate communicating parties. Therefore, the correlation matrix between the channel characteristics of the transmitting end and the receiving end is calculated first, and the correlation matrix between the transmitting end and the eavesdropping end is obtained by multiplying the correlation matrix by the correlation coefficient (selected as 0.7 in this embodiment). The signal estimation value of the eavesdropping end is generated according to the correlation matrix between the transmitting end and the eavesdropping end.

[0043] In order to verify the performance of this embodiment, the LTE-A Uplink Link Level Simulator in the Vienna LTE-A Simulators is used to simulate the channel environment when the vehicle is driving. The specific parameters are as follows: Figure 4 As shown in the figure, mean squared error (MSE), mutual information (MI) and Pearson correlation coefficient (Pearson correlation coefficient) are selected as evaluation indicators of the multi-task autoencoder network.

[0044] As a more intuitive method to measure the "average error", mean square error can evaluate the degree of deviation between the data of both parties. This paper calculates MSE to reflect the deviation between the reciprocity eigenvalues ​​obtained by the sender and the receiver.

[0045] The number of bits of the key obtained after the reciprocal feature quantization can be reflected by calculating the mutual information of the features extracted by the sender and the receiver. A higher mutual information combined with an appropriate quantization method can generate a pair of keys with a lower inconsistency rate, a higher generation rate and better randomness.

[0046] The Pearson correlation coefficient is mainly used to measure the correlation between two variables. It can indicate the correlation between the features extracted between two communication nodes.

[0047] The comparison of mean square error and mutual information between the method (AE) for generating a key in this embodiment and the method (PCA) for generating a key using principal component analysis is shown in the figure below: Figure 5 and Figure 6 As shown, it can be clearly seen from the curve in the figure that the reciprocity feature extraction scheme proposed in this embodiment is better than the principal component analysis scheme.

[0048] like Figure 7As shown, it is a scatter plot of the correlation of each communication node using the multi-task autoencoder network (Critical state) of this embodiment and the conventional autoencoder (Baseline). It can be clearly seen from the figure that the correlation between the sending end and the receiving end obtained by the method of this embodiment is better than that of the conventional method, and the correlation between the sending end and the eavesdropping end is also reduced.

[0049] This embodiment can be applied to communications in wireless networks. The communicating parties can be a base station and a user. When the base station communicates with the user, the base station sends a pilot signal to the user, and the user obtains channel state information (CSI) by detecting the pilot signal. Similarly, the user also sends a pilot signal to the base station, and the base station obtains the channel state information by receiving the pilot signal. There is a third-party channel encryption platform that collects the channel state information detected by the base station and the user within a certain period of time, and inputs it into the self-encoding network for training. After the training is completed, the trained encoder end is installed on the base station end and the user end respectively. At this time, the base station and the user can obtain the reciprocity feature z of the channel state information, and then quantize it to obtain the preliminary security key Q. After the security key has been checked for consistency by the communicating parties, it can be directly used for encryption. The process is as follows Figure 8 shown.

[0050] This embodiment also proposes a physical layer key generation device based on a multi-task autoencoder, including a memory and a processor; the memory is used to store a computer program; the processor is used to implement the following method when executing the computer program:

[0051] Step S1: Obtain signal estimation values ​​of the transmitting end, the receiving end and the eavesdropping end, and extract amplitude and phase thereof.

[0052] Step S2: The estimated signal value of the transmitting end As input, the signal estimate at the receiving end and the signal estimate of the eavesdropping end As the target value, the multi-task autoencoder network training is input. When the loss function, network weight and activation function all reach the ideal value, that is, when the preset training requirements are met, the training is stopped to obtain the trained multi-task autoencoder network.

[0053] Step S3: Install the trained multi-task autoencoder network into the local processors of the transmitting end and the receiving end.

[0054] Step S4, obtain the output z of the hidden layer of the multi-task autoencoder network, obtain the reciprocity characteristics of the transmitter and the receiver, and obtain the physical layer key after quantization.

[0055] This embodiment further provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by the processor, the physical layer key generation method based on the multi-task self-encoder mentioned in the above embodiment is implemented, and any combination of one or more computer-readable media can be used. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0056] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A physical layer key generation method based on a multi-task autoencoder, characterized in that: The following steps are involved: S1, obtaining signal estimation values ​​of the transmitting end, the receiving end and the eavesdropping end; S2, input the channel estimation values ​​of the three into the multi-task autoencoder network training. When the loss function, network weight and activation function all reach the ideal value, stop the training and obtain the trained multi-task autoencoder network; The training steps of the multi-task autoencoder network are as follows: The channel estimation value of the transmitting end is used as the input of the encoder, the channel estimation value of the receiving end is used as the output of the first decoder, and the channel estimation value of the eavesdropping end is used as the output of the second decoder to train the multi-task autoencoder network; S3, installing the trained multi-task autoencoder network into the local processors of the transmitter and receiver respectively; S4, obtaining the output of the hidden layer of the multi-task autoencoder network, obtaining the reciprocity characteristics of the transmitter and the receiver, and obtaining the physical layer key after quantization; The multi-task autoencoder network includes an encoder, a first decoder and a second decoder, and a gradient reversal layer is arranged between the encoder and the second decoder.

2. A physical layer key generation method based on a multi-task autoencoder according to claim 1, characterized in that: The activation function of the multi-task autoencoder network is Leaky ReLU.

3. According to claim 1, a physical layer key generation method based on a multi-task autoencoder is characterized in that: The multi-task autoencoder network is optimized using the Adam optimizer.

4. A method for generating a physical layer key based on a multi-task autoencoder according to claim 1, characterized in that: The method for obtaining the signal estimation value of the eavesdropping end is as follows: The correlation matrix between the channel characteristics of the transmitting end and the receiving end is calculated, and the correlation matrix between the transmitting end and the eavesdropping end is obtained by multiplying the correlation matrix by the correlation coefficient. The signal estimation value of the eavesdropping end is generated according to the correlation matrix between the transmitting end and the eavesdropping end.

5. A physical layer key generation device based on a multi-task autoencoder, characterized in that: The invention comprises a memory and a processor; the memory is used to store a computer program; the processor is used to implement the following method when executing the computer program: S1, obtaining signal estimation values ​​of the transmitting end, the receiving end and the eavesdropping end; S2, input the channel estimation values ​​of the three into the multi-task autoencoder network training. When the loss function, network weight and activation function all reach the ideal value, stop the training and obtain the trained multi-task autoencoder network; The training steps of the multi-task autoencoder network are as follows: The channel estimation value of the transmitting end is used as the input of the encoder, the channel estimation value of the receiving end is used as the output of the first decoder, and the channel estimation value of the eavesdropping end is used as the output of the second decoder to train the multi-task autoencoder network; S3, installing the trained multi-task autoencoder network into the local processors of the transmitter and receiver respectively; S4, obtaining the output of the hidden layer of the multi-task autoencoder network, obtaining the reciprocity characteristics of the transmitter and the receiver, and obtaining the physical layer key after quantization; The multi-task autoencoder network includes an encoder, a first decoder and a second decoder, and a gradient reversal layer is arranged between the encoder and the second decoder.

6. A physical layer key generation device based on a multi-task autoencoder according to claim 5, characterized in that: The activation function of the multi-task autoencoder network is Leaky ReLU.

7. A physical layer key generation device based on a multi-task autoencoder according to claim 5, characterized in that: The multi-task autoencoder network is optimized using the Adam optimizer.

8. The physical layer key generation device based on a multi-task autoencoder according to claim 5, characterized in that: The method for obtaining the signal estimation value of the eavesdropping end is as follows: The correlation matrix between the channel characteristics of the transmitting end and the receiving end is calculated, and the correlation matrix between the transmitting end and the eavesdropping end is obtained by multiplying the correlation matrix by the correlation coefficient. The signal estimation value of the eavesdropping end is generated according to the correlation matrix between the transmitting end and the eavesdropping end.

Citation Information

Patent Citations

  • High-consistency physical key generation method based on neural network

    CN108365951A

  • Physical layer key generation method and system based on deep learning

    CN113746628A

  • Encoding and reconstructing inputs using neural networks

    US20190171929A1