A physical layer key generation method, system, electronic device and storage medium

Abstract

The present application provides a physical layer key generation method, system, electronic device and storage medium, which construct a training data set according to a first estimated value collected by a legitimate communication node A and a second estimated value collected by a legitimate communication node B; iteratively train the key generation network to be trained using a loss function constructed by the training data set; deploy a first target channel feature extraction network and a second target feature extraction network to the legitimate communication node A and the legitimate communication node B respectively; quantize multiple first target channel feature vectors obtained by processing each first estimated value using the first target channel feature extraction network through the legitimate communication node A to obtain a corresponding key sequence; quantize multiple second target channel feature vectors obtained by processing each second estimated value using the second target channel feature extraction network through the legitimate communication node B to obtain a corresponding key sequence.

Description

Technical Field

[0001] The present application relates to the field of physical layer security technology, and more specifically, to a physical layer key generation method, system, electronic device and storage medium. Background Art

[0002] With the rapid development and application of technologies such as 5G, the Internet of Things, and the Internet of Vehicles, the number of wireless connected devices has increased dramatically, which also means that more and more private data is transmitted through wireless channels. However, the inherent openness of wireless channels provides opportunities for malicious attacks, and the security risks of wireless transmission are becoming increasingly prominent. Secure and efficient key distribution is an important prerequisite for ensuring secure data transmission, but traditional key distribution schemes mostly rely on complex public key cryptographic algorithms, and the complexity increases geometrically with the increase in the number of wireless terminals. It is difficult to adapt to the security transmission requirements of typical wireless networks such as the Internet of Things and the Internet of Vehicles. Physical layer key generation technology can use the short-term reciprocity and spatiotemporal uniqueness of wireless channels to establish a shared key between the legitimate communicating parties without the assistance of a third party. It has the advantages of decentralization, lightweight, and information-theoretic security, and is an effective means to solve the key distribution problem in wireless communication networks.

[0003] The general process of physical layer key generation technology is as follows: First, the legitimate communicating parties alternately send channel detection signals containing common pilots in time division duplex mode, and use their respective received signals and known pilots to complete channel estimation. Then, by preprocessing and quantizing the channel estimate, it is converted into a binary sequence as the initial key. Ideally, as long as the time interval between the legitimate communicating parties sending channel detection signals does not exceed the channel coherence time, it can be guaranteed that both parties can obtain almost consistent channel estimates. However, in actual communication scenarios, due to the existence of non-reciprocal factors such as asynchronous channel detection and environmental noise, there will be differences in the channel estimates of both parties to varying degrees, resulting in more inconsistent bits in the initial keys of both parties, resulting in low key generation rate and reliability, and even key generation failure.

[0004] Therefore, in the process of physical layer key generation, how to overcome the non-reciprocity factor of the channel estimation between the two parties introduced by the actual communication environment and improve the key generation rate and reliability is a problem that urgently needs to be solved in this application. Summary of the invention

[0005] In view of this, the present application provides a physical layer key generation method, system, electronic device and storage medium, with the purpose of improving the physical layer key generation rate and reliability.

[0006] A first aspect of the present application provides a method for generating a physical layer key, the method comprising:

[0007] Constructing a corresponding training data set based on each first estimation value collected based on the legitimate communication node A and each second estimation value collected based on the legitimate communication node B; wherein the training data set includes a plurality of training sample pairs, and the training sample pairs include the first estimation value and the second estimation value;

[0008] Inputting the training data set into the key generation network to be trained, so that the key generation network to be trained constructs a corresponding loss function using the input training data set, and performs iterative training based on the loss function to obtain the key generation network;

[0009] Deploy a first target channel feature extraction network to a legitimate communication node A, and deploy a second target channel feature extraction network to a legitimate communication node B; wherein the first target channel feature extraction network is a first channel feature extraction network in a trained key generation network, and the second target channel feature extraction network is a second channel feature extraction network in a trained key generation network;

[0010] The legitimate communication node A uses the deployed first target channel feature extraction network to process the collected first estimation values ​​to obtain a plurality of first target channel feature vectors, and quantizes the plurality of first target channel feature vectors to obtain a key sequence of the legitimate communication node A;

[0011] The legal communication node B uses the deployed second target channel feature extraction network to process the collected second estimation values ​​to obtain multiple second target channel feature vectors, and quantizes the multiple second target channel feature vectors to obtain the key sequence of the legal communication node B.

[0012] Optionally, the key generation network to be trained is composed of two channel feature extraction networks, a shared decoder network and a mutual information evaluation network;

[0013] The step of inputting the training data set into the key generation network to be trained, allowing the key generation network to be trained to construct a corresponding loss function using the input training data set, and performing training based on the loss function to obtain the key generation network includes:

[0014] Input each training sample pair in the training data set into the key generation network to be trained;

[0015] For each of the training sample pairs, feature extraction is performed on a first estimated value in the input training sample pair through a first channel feature extraction network to obtain a first channel feature vector, and feature extraction is performed on a second estimated value in the input training sample pair through a second channel feature extraction network to obtain a second channel feature vector;

[0016] Reconstructing the first channel feature vector and the second channel feature vector through the shared decoder network to obtain a first channel reconstruction value and a second channel reconstruction value;

[0017] Performing evaluation using the first channel feature vector and the second channel feature vector through the mutual information evaluation network to obtain an evaluation value;

[0018] A loss function is constructed using each of the training sample pairs and their corresponding first channel feature vectors, second channel feature vectors, first channel reconstruction values, second channel reconstruction values ​​and evaluation values, and the loss function is used to adjust the parameters of the first channel feature extraction network, the second channel feature extraction network, the shared decoder network and the mutual information evaluation network to obtain a key generation network.

[0019] Optionally, performing channel reconstruction on the first channel feature vector and the second channel feature vector respectively through the shared decoder network to obtain a first channel reconstruction value and a second channel reconstruction value includes:

[0020] The first channel feature vector is processed by the shared decoder network to obtain a first initial vector, and the first initial vector is channel reshaped to obtain a first channel coarse reconstruction value; the first channel coarse reconstruction value is refined to obtain a first channel refined reconstruction value, and the first channel refined reconstruction value is convolved to obtain a first channel reconstruction value;

[0021] The second channel feature vector is processed by the shared decoder network to obtain a second initial vector, and the second initial vector is channel reshaped to obtain a second channel coarse reconstruction value; the second channel coarse reconstruction value is refined to obtain a second channel refined reconstruction value, and the second channel refined reconstruction value is convolved to obtain a second channel reconstruction value.

[0022] Optionally, the performing evaluation using the first channel feature vector and the second channel feature vector through the mutual information evaluation network to obtain an evaluation value includes:

[0023] The first channel eigenvector and the second channel eigenvector are concatenated to obtain a single channel matrix, and the single channel matrix is ​​input into a mutual information evaluation network, so that the mutual information evaluation network performs convolution processing on the single channel matrix, and reshapes and processes the data obtained after the convolution processing to obtain a corresponding evaluation value.

[0024] Optionally, constructing a loss function using each of the training sample pairs and their corresponding first channel feature vectors, second channel feature vectors, first channel reconstruction values, second channel reconstruction values, and evaluation values, and adjusting parameters of the first channel feature extraction network, the second channel feature extraction network, the shared decoder network, and the mutual information evaluation network using the loss function to obtain a key generation network includes:

[0025] Constructing a first loss function using each of the training sample pairs and their corresponding first channel feature vectors, second channel feature vectors, first channel reconstruction values, and second channel reconstruction values;

[0026] Constructing a second loss function using the evaluation value;

[0027] Construct a corresponding loss function according to the first loss function and the second loss function;

[0028] Initializing parameters of the key generation network to be trained;

[0029] Fixing the parameters of the first channel feature extraction network, the second channel feature extraction network, and the shared decoder network unchanged, and pre-training the mutual information evaluation network using the second loss function;

[0030] Determining whether the number of training iterations of the key generation network to be trained is greater than a target number of iterations;

[0031] If the number of training iterations of the key generation network to be trained is not greater than the target number of iterations, the parameters of the pre-trained mutual information evaluation network are fixed unchanged, and the parameters of the first channel feature extraction network, the second channel feature extraction network and the shared decoder network are adjusted using the loss function;

[0032] The parameters of the first channel feature extraction network, the second channel feature extraction network, and the shared decoder network after parameter adjustment are fixed unchanged, and the parameters of the pre-trained mutual information evaluation network are adjusted using the second loss function;

[0033] When the parameter adjustment of the pre-trained mutual information evaluation network is completed, the number of training iterations of the key generation network to be trained is controlled to increase by 1, and the execution is returned to determine whether the number of training iterations of the key generation network to be trained is greater than the target number of iterations, until the number of training iterations of the key generation network to be trained is greater than the target number of iterations, the key generation network is obtained.

[0034] Optionally, the step of processing the collected first estimation values ​​by the legitimate communication node A using the deployed first target channel feature extraction network to obtain a plurality of first target channel feature vectors, and quantizing the plurality of first target channel feature vectors to obtain a key sequence of the legitimate communication node A includes:

[0035] The legitimate communication node A uses the deployed first target channel feature extraction network to process the collected first estimation values ​​to obtain a plurality of first target channel feature vectors;

[0036] Counting the first data of each of the first target channel feature vectors in each dimension through the legal communication node A, and constructing a first empirical cumulative distribution function corresponding to each dimension according to each of the first data in each dimension;

[0037] Determine, by the legal communication node A, a mapping result corresponding to each first data on each dimension using the first empirical cumulative distribution function corresponding to each dimension and a preset data set, and convert the mapping result corresponding to each first data on each dimension into binary data;

[0038] The legitimate communication node A cascades the binary data corresponding to each first data in each dimension to obtain a key sequence of the legitimate communication node A.

[0039] Optionally, the step of processing the collected second estimation values ​​by the legitimate communication node B using a deployed second target channel feature extraction network to obtain a plurality of second target channel feature vectors, and quantizing the plurality of second target channel feature vectors to obtain a key sequence of the legitimate communication node B includes:

[0040] The legitimate communication node B processes each collected second estimation value by using the deployed second target channel feature extraction network to obtain a plurality of second target channel feature vectors;

[0041] Counting the second data of each of the second target channel feature vectors in each dimension through the legal communication node B, and constructing a second empirical cumulative distribution function corresponding to each dimension according to each of the second data in each dimension;

[0042] Determine, by the legal communication node B, a mapping result corresponding to each second data on each dimension using the second empirical cumulative distribution function corresponding to each dimension and a preset data set, and convert the mapping result corresponding to each second data on each dimension into binary data;

[0043] The legitimate communication node B cascades the binary data corresponding to each second data in each dimension to obtain a key sequence of the legitimate communication node B.

[0044] A second aspect of the present application provides a physical layer key generation system, the system comprising:

[0045] A training data set construction unit, configured to construct a corresponding training data set based on each first estimation value collected based on the legitimate communication node A and each second estimation value collected based on the legitimate communication node B; wherein the training data set includes a plurality of training sample pairs, and the training sample pairs include the first estimation value and the second estimation value;

[0046] A training unit, used for inputting the training data set into the key generation network to be trained, so that the key generation network to be trained constructs a corresponding loss function using the input training data set, and performs iterative training based on the loss function to obtain the key generation network;

[0047] A deployment unit, configured to deploy a first target channel feature extraction network to a legitimate communication node A, and to deploy a second target channel feature extraction network to a legitimate communication node B; wherein the first target channel feature extraction network is a first channel feature extraction network in a trained key generation network, and the second target channel feature extraction network is a second channel feature extraction network in a trained key generation network;

[0048] The legitimate communication node A is used to process the collected first estimation values ​​by using the deployed first target channel feature extraction network to obtain a plurality of first target channel feature vectors, and to quantize the plurality of first target channel feature vectors to obtain a key sequence of the legitimate communication node A;

[0049] The legitimate communication node B is used to use the deployed second target channel feature extraction network to process the collected second estimation values ​​to obtain multiple second target channel feature vectors, and quantize the multiple second target channel feature vectors to obtain the key sequence of the legitimate communication node B.

[0050] An electronic device provided in the third aspect of the present application includes: a processor and a memory, wherein the processor and the memory are connected via a communication bus; wherein the processor is used to call and execute a program stored in the memory; and the memory is used to store a program, wherein the program is used to implement the physical layer key generation method provided in the first aspect of the present application.

[0051] A fourth aspect of the present application provides a storage medium, in which computer executable instructions are stored, and the computer executable instructions are used to execute the physical layer key generation method provided by the first aspect of the present application.

[0052] The embodiment of the present application provides a physical layer key generation method, system, electronic device and storage medium, by constructing a corresponding training data set based on each first estimation value collected based on a legitimate communication node A and each second estimation value collected based on a legitimate communication node B; wherein the training data set includes a plurality of training sample pairs, and the training sample pairs include a first estimation value and a second estimation value; the training data set is input into a key generation network to be trained, so that the key generation network to be trained constructs a corresponding loss function using the input training data set, and iterative training is performed based on the loss function to obtain a key generation network; a first target channel feature extraction network is deployed to the legitimate communication node A, and a second target channel feature extraction network is deployed to the legitimate communication node B; wherein the first target The channel feature extraction network is the first channel feature extraction network in the trained key generation network, and the second target channel feature extraction network is the second channel feature extraction network in the trained key generation network; the legitimate communication node A uses the deployed first target channel feature extraction network to process the collected first estimation values ​​to obtain multiple first target channel feature vectors, and the multiple first target channel feature vectors are quantized to obtain the key sequence of the legitimate communication node A; the legitimate communication node B uses the deployed second target channel feature extraction network to process the collected second estimation values ​​to obtain multiple second target channel feature vectors, and the multiple second target channel feature vectors are quantized to obtain the key sequence of the legitimate communication node B. The technical means provided by the present application trains the key generation network to be trained by constructing a corresponding training data set in advance using each first estimation value collected based on the legitimate communication node A and each second estimation value collected based on the legitimate communication node B, and deploys the two channel feature extraction networks in the trained key generation network to the legitimate communication node A and the legitimate communication node B, respectively, so that the legitimate communication node A and the legitimate communication node B can respectively use the deployed channel feature extraction networks to effectively compensate for the non-reciprocity in the collected estimation values ​​due to time asynchrony, which can significantly enhance the reciprocity between the channel feature vectors extracted by the two legitimate communication nodes, thereby improving the reliability and efficiency of key generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0054] Figure 1 A flowchart of a method for training a key generation network provided in an embodiment of the present application;

[0055] Figure 2 A diagram showing an example structure of a key generation network to be trained provided in an embodiment of the present application;

[0056] Figure 3 A structural example diagram of a channel feature extraction network provided in an embodiment of the present application;

[0057] Figure 4 A structural example diagram of a residual network module ResNet in a channel feature extraction network provided in an embodiment of the present application;

[0058] Figure 5 A structural example diagram of a convolutional block attention module CBAM in a channel feature extraction network provided in an embodiment of the present application;

[0059] Figure 6 A structural example diagram of a shared decoder network provided in an embodiment of the present application;

[0060] Figure 7 A diagram showing an example structure of a mutual information evaluation network provided in an embodiment of the present application;

[0061] Figure 8 An example diagram of a curve of the change of NMSE (Normalized Mean Squared Error) versus SNR (Signal-to-Noise Ratio) between channel feature vectors of two legitimate communication nodes extracted by a physical layer key generation method and a comparison method provided in an embodiment of the present application;

[0062] Fig. 9 An example diagram of a curve of the correlation coefficient between the channel feature vectors of two legitimate communication nodes extracted by a physical layer key generation method and a comparison method provided in an embodiment of the present application versus SNR;

[0063] Fig.10 An example diagram of a curve of the inconsistency rate between key sequences of two legitimate communication nodes generated by a physical layer key generation method and a comparison method provided in an embodiment of the present application and a curve of the change of SNR;

[0064] Fig.11 An example diagram of a curve of average key length versus SNR generated by a physical layer key generation method and a comparison method provided in an embodiment of the present application;

[0065] Fig.12 A schematic diagram of the structure of a physical layer key generation system provided in an embodiment of the present application;

[0066] Fig.13 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0068] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0069] See also Figure 1 , shows a flow chart of a training method for a key generation network provided in an embodiment of the present application, the method specifically comprising the following steps:

[0070] S101: Construct a corresponding training data set based on each first estimation value collected based on the legitimate communication node A and each second estimation value collected based on the legitimate communication node B; wherein the training data set includes a plurality of training sample pairs, and the training sample pairs include a first estimation value and a second estimation value.

[0071] In the specific process of executing step S101, the legitimate communication node A and the legitimate communication node B communicate with each other at a preset time interval. Corresponding channel detection signals are sent alternately, so that the legitimate communication node A obtains a first estimated value of its own legitimate channel based on the received channel detection signal, and the legitimate communication node B obtains a second estimated value of its own legitimate channel based on the received channel detection signal; the first estimated value obtained by the legitimate communication node A and the second estimated value obtained by the legitimate communication node B form a channel estimation pair; when the collected When the channel estimation is correct, the collected Channel estimation pairs constitute a training sample pair, and finally use the collected training sample pairs to construct the corresponding training data set. is a positive integer greater than 1, is a positive integer greater than 1.

[0072] It should be noted that a corresponding preset time interval can be pre-set, wherein the preset time interval is less than the channel coherence time of the legitimate communication node A and the legitimate communication node B.

[0073] In actual application, it is assumed that the channel coherence time between the legitimate communication node A and the legitimate communication node B is , then you can set the preset time interval to ,in, Less than ; Control the legitimate communication node A and the legitimate communication node B to send channel detection signals successively at a preset time interval t, that is, control the legitimate communication node A to send a channel detection signal to the legitimate communication node B, and at a preset time interval When the legitimate communication node B is controlled to send a channel detection signal to the legitimate communication node A, the legitimate communication node A obtains a first estimated value of its own legitimate channel based on the received channel detection signal. , and the legitimate communication node B obtains a second estimated value of its own legitimate channel based on the received channel detection signal ,in, , , is the dimension of the estimated value (first estimate / second estimate); it will be displayed at a preset time interval First estimate received before and after and the second estimate Form a channel estimation pair When collected When the channel estimates are correct, A pair of channel estimates constitutes a training sample pair ,in, , ; Finally, use the collected training sample pairs to construct the corresponding training dataset .

[0074] S102: Inputting the training data set into the key generation network to be trained, so that the key generation network to be trained constructs a corresponding loss function using the input training data set, and performs iterative training based on the loss function to obtain the key generation network.

[0075] In the specific process of executing step S102, after constructing the corresponding training data set, the key generation network to be trained can be trained using each training sample pair in the training data set input to obtain the corresponding key generation network; for each training sample pair, the key generation network to be trained respectively extracts features from the first estimated value and the second estimated value in the training sample pair to obtain a first channel feature vector and a second channel feature vector; performs channel reconstruction on the first channel feature vector and the second channel feature vector to obtain a first channel reconstruction value and a second channel reconstruction value, and performs evaluation using the first channel feature vector and the second channel feature vector to obtain an evaluation value; finally, a loss function is constructed using the training sample pair, the first channel feature vector, the second channel feature vector, the first channel reconstruction value, the second channel reconstruction value and the evaluation value, so as to use the constructed loss function to perform corresponding parameter adjustment to obtain the key generation network.

[0076] In the embodiment of the present application, two channel feature extraction networks, a shared decoder network and a mutual information evaluation network can be used in advance to form a key generation network to be trained, such as Figure 2 shown.

[0077] Optionally, the training data set is input into the key generation network to be trained, so that the key generation network to be trained uses the input training data set to construct a corresponding loss function, and is trained based on the loss function. The process of obtaining the key generation network can be specifically as follows: input each training sample pair in the training data set into the key generation network to be trained; for each training sample pair, feature extraction is performed on the first estimated value in the input training sample pair through the first channel feature extraction network to obtain a first channel feature vector, and feature extraction is performed on the second estimated value in the input training sample pair through the second channel feature extraction network to obtain a second channel feature vector; channel reconstruction is performed on the first channel feature vector and the second channel feature vector respectively through the shared decoder network to obtain a first channel reconstruction value and a second channel reconstruction value; the first channel feature vector and the second channel feature vector are evaluated through the mutual information evaluation network to obtain an evaluation value; a loss function is constructed using each training sample pair and its corresponding first channel feature vector, second channel feature vector, first channel reconstruction value, second channel reconstruction value and evaluation value, and the loss function is used to adjust the parameters of the first channel feature extraction network, the second channel feature extraction network, the shared decoder network and the mutual information evaluation network to obtain a key generation network.

[0078] In some embodiments, after each training sample pair in the training data set is input into the key generation network to be trained, for each training sample pair, a first channel feature extraction network is used to extract the first estimated value to obtain a first channel feature vector ; Extract the characteristics of the second estimated value through the second channel feature extraction network to obtain the second channel feature vector .

[0079] in, is the function between the input and output of the channel feature extraction network, is the first estimate of the training sample pair, is the parameter of the first channel feature extraction network, is the second estimated value in the training sample pair, Parameters of the feature extraction network for the second channel.

[0080] It is worth noting that the structures of the two channel feature extraction networks (the first channel feature extraction network and the second channel feature extraction network) are the same, both of which are feedforward neural networks. The channel feature extraction network is composed of the residual network module ResNet, the convolution layer Conv1, the convolution block attention module CBAM, the convolution layer Conv2, the convolution layer Conv3 and the fully connected layer FC1 from the input layer to the output layer. Figure 3As shown in the figure, the residual network module ResNet contains 3 convolutional layers and 1 shortcut connection, and adopts Leaky ReLU activation function; the convolutional block attention module CBAM is composed of channel attention units and spatial attention units connected in sequence, which are used to learn the weight coefficients in the channel dimension and spatial dimension respectively, and then multiply the weight coefficients with the input data to obtain the corresponding output data.

[0081] Specifically, after the training sample pair is input into the key generation network to be trained, the first estimated value in the training sample pair can be processed by the residual network module ResNet in the first channel feature extraction network to obtain output data with the same dimension as the first estimated value, wherein the dimension of the first estimated value can be ; The output data of ResNet passes through a convolution kernel of size , the step length is And the convolution layer Conv1 with batch normalization has 16 channels, height and width The output data of Conv1 is input into the convolution block attention module CBAM to obtain the output data with the same dimension as the input data, where the dimension of the input data at this time is ; After the output data of CBAM is processed by the Leaky ReLU activation function, it passes through two convolutional layers Conv2 and Conv3 with batch normalization and Leaky ReLU activation function. The convolution kernel sizes of the convolutional layers Conv2 and Conv3 are both , the former step length is , the latter step length is , the dimensions of the output data of convolutional layers Conv2 and Conv3 are compressed to ; The output data of Conv3 is reshaped into a dimension of Finally, the output data of Conv3 passes through a fully connected layer FC1 to obtain an output dimension of The first channel eigenvector of ,in, It can be 64, which is not limited in this embodiment of the application.

[0082] It should be noted that the process of extracting features from the second estimated value in the training sample pair through the second channel feature extraction network to obtain the second channel feature vector is the same as the process of extracting features from the first estimated value in the training sample pair through the first channel feature extraction network to obtain the first channel feature vector. Please refer to the process of extracting features from the first estimated value in the training sample pair through the first channel feature extraction network to obtain the first channel feature vector, and the embodiments of the present application will not be repeated here.

[0083] In the embodiment of the present application, the residual network module ResNet in the channel feature extraction network (the first channel feature extraction network or the second channel feature extraction network) includes three convolutional layers, such as Figure 4 As shown, the convolution kernel size of each layer is , the step length is , the number of channels of the output data are 8, 16 and 2 respectively. By filling the input data of each layer with an appropriate number of 0 elements, the height and width of the output data of each layer remain unchanged. The output data of the first two convolutional layers are processed by batch normalization and Leaky ReLU activation function in turn. The output data of the last convolutional layer is added to the input data of ResNet after batch normalization, and then processed by Leaky ReLU activation function to obtain the output data of RestNet. In short, the relationship between the input data and output data of ResNet is shown in formula (1):

[0084] (1)

[0085] in, is the output data of ResNet, is the input data of ResNet, represents the Leaky ReLU activation function, Represents the processing of 3 convolutional layers in RestNet.

[0086] In the embodiment of the present application, the structure of the convolutional block attention module CBAM in the channel feature extraction network is as follows Figure 5 As shown in the figure, after the estimated value (first estimated value or second estimated value) in the training sample pair of the key generation network to be trained is processed by the residual network module ResNet to obtain the corresponding output data, the output data of ResNet is passed through the convolution layer Conv1, and the obtained output data is used as the input data of the convolution attention module , so that the convolutional attention module CBAM is concerned with the input data The specific processing process is as follows:

[0087] A1: Convolutional attention module on input data Performing maximum pooling and average pooling on both the height and width dimensions, we get two dimensions: Vector and ,in, represents the maximum pooling process in the height and width dimensions, Represents average pooling in height and width dimensions;

[0088] A2: Convert vector and Input them into the same multi-layer perceptron MLP respectively, so that MLP outputs 2 weight vectors according to the input vector and ,in, Indicates the processing of a multilayer perceptron MLP. The multilayer perceptron MLP is a fully connected neural network with a single hidden layer. The number of neurons in the hidden layer can be 8. The ReLU activation function is used. The number of neurons in the input and output layers can be 16.

[0089] A3: weight vector and After addition, the sigmoid activation function is used to process the resulting dimension. The channel attention weight ,in, Represents the sigmoid activation function.

[0090] A4: Channel attention weight Each element in is multiplied by all elements on the corresponding channel of the input data F to obtain the channel attention output data ,in, Indicates that the elements of x are copied in the height and width dimensions. Represents element-wise multiplication.

[0091] A5: Output data for channel attention Perform maximum pooling and average pooling on the channel dimension, and get two dimensions respectively: The matrix and ,in Represents the maximum pooling processing in the channel dimension, Represents average pooling processing in the channel dimension.

[0092] A6: Convert the two matrices and Each as a channel, cascaded into a dimension of Dual-channel feature map data ,Then After a convolutional layer, the spatial attention weight is output , where the number of output channels of this convolutional layer is 1, with batch normalization processing, and the convolution kernel size is , the step length is , and fill the input data of the convolutional layer with an appropriate number of 0 elements so that the height and width of the output data are consistent with the input data, using the sigmoid function;

[0093] A7: Spatial attention weights Each element in the channel attention output data The output data of the convolutional block attention module CBAM is obtained by multiplying the elements of all channels at the corresponding height and width positions. ,in, Indicates that The elements of are replicated across the channel dimension.

[0094] Finally, the output data of the convolutional block attention module CBAM After being processed by the Leaky ReLU activation function, it passes through two convolutional layers Conv2 and Conv3 with batch normalization and Leaky ReLU activation function to obtain vector, and The vector passes through a fully connected layer FC1 and the output dimension is The channel feature vector / .

[0095] In the embodiment of the present application, after obtaining the first channel feature vector and the second channel eigenvector After that, the first channel feature vector and the second channel feature vector can be respectively input into the shared decoder network, so that the shared decoder network performs channel reconstruction on the first channel feature vector and the second channel feature vector respectively to obtain the first channel reconstruction value and the second channel reconstruction value .in, A function that represents the relationship between the input and output of the shared decoder network, are the parameters of the shared decoder network.

[0096] Optionally, the process of performing channel reconstruction on the first channel eigenvector and the second channel eigenvector respectively through a shared decoder network to obtain a first channel reconstruction value and a second channel reconstruction value can be specifically as follows: processing the first channel eigenvector through a shared decoder network to obtain a first initial vector, and performing channel reshaping on the first initial vector to obtain a first channel coarse reconstruction value; refining the first channel coarse reconstruction value to obtain a first channel refined reconstruction value, and convolving the first channel refined reconstruction value to obtain the first channel reconstruction value; processing the second channel eigenvector through a shared decoder network to obtain a second initial vector, and performing channel reshaping on the second initial vector to obtain a second channel coarse reconstruction value; refining the second channel coarse reconstruction value to obtain a second channel refined reconstruction value, and convolving the second channel refined reconstruction value to obtain a second channel reconstruction value.

[0097] It is worth noting that the shared decoder network is a feedforward neural network with the following structure: Figure 6As shown in the figure, it specifically includes a fully connected layer FC2, a Reshape, 2 sequentially connected ResNets and a convolutional layer Conv4.

[0098] Specifically, the first channel feature vector and the second channel feature vector are respectively input into the fully connected layer FC2 of the shared decoder network, so that the fully connected layer FC2 can decode the first channel feature vector. Process and output a dimension of The vector (the first initial vector) is then reshaped into a vector with a dimension of The first channel rough reconstruction value is input into two sequentially connected ResNets, and the first channel refined reconstruction value is output, wherein the dimension of the first channel refined reconstruction value is the same as the dimension of the input first channel rough reconstruction value, and the structure of ResNet is the same as the ResNet structure in the channel feature extraction network; the first channel refined reconstruction value passes through a convolution layer Conv4, and the output dimension is The first channel reconstruction value , where the convolution kernel size of the convolution layer Conv4 is , the step length is , with batch normalization processing, using sigmoid activation function. Among them, the process of processing the second channel feature vector through the shared analytical code network is the same as the process of processing the first channel feature vector through the shared analytical code, which can be seen from the process of processing the first channel feature vector through the shared analytical code, and will not be repeated in this embodiment of the present application.

[0099] In the embodiment of the present application, the first channel feature vector and the second channel eigenvector The first channel feature vector can be input into the shared decoder network separately. and the second channel eigenvector Input the mutual information evaluation network together, so that the mutual information evaluation network uses the first channel feature vector and the second channel eigenvector Evaluate and get the corresponding evaluation value ,in, A function that represents the relationship between the input and output of the mutual information evaluation network, Parameters of the network evaluated for mutual information.

[0100] Optionally, the process of evaluating the first channel eigenvector and the second channel eigenvector through the mutual information evaluation network to obtain the evaluation value can be specifically as follows: concatenating the first channel eigenvector and the second channel eigenvector to obtain a single-channel matrix, and inputting the single-channel matrix into the mutual information evaluation network, so that the mutual information evaluation network performs convolution processing on the single-channel matrix, and reshapes and processes the data obtained after the convolution processing to obtain the corresponding evaluation value.

[0101] It is worth noting that the mutual information evaluation network is a feedforward neural network with the following structure: Figure 7 As shown, it specifically includes a 4-layer convolutional neural network ConvNet, a Reshape and a fully connected layer.

[0102] Specifically, the mutual information evaluation network transforms the first channel feature vector and the second channel eigenvector Spliced ​​into a dimension Single channel matrix , and convert the single channel matrix Input to a 4-layer convolutional neural network ConvNet, so that the convolutional neural network ConvNet is a single channel matrix Perform convolution processing and get the dimension The output data of the convolutional neural network ConvNet is shown in Table 1 in order from front to back. The output data of the convolutional neural network ConvNet is reshaped into a dimension of The vector is processed by a fully connected layer to obtain the corresponding evaluation value.

[0103] Table 1:

[0104]

[0105] In an embodiment of the present application, after obtaining the first channel eigenvector, the second channel eigenvector, the first channel reconstruction value, the second channel reconstruction value and the evaluation value of each training sample pair, the loss function of the key generation network to be trained can be constructed using each training sample pair and its corresponding first channel eigenvector, second channel eigenvector, first channel reconstruction value, second channel reconstruction value and evaluation value, and the key generation network to be trained is trained by sampling a stochastic gradient descent algorithm to minimize the loss function of the key generation network to be trained, thereby obtaining the key generation network.

[0106] Optionally, a first loss function is constructed using each pair of training samples and their corresponding first channel feature vectors, second channel feature vectors, first channel reconstruction values, and second channel reconstruction values; a second loss function is constructed using the evaluation value; a corresponding loss function is constructed based on the first loss function and the second loss function; and the loss function is used to adjust the parameters of the first channel feature extraction network, the second channel feature extraction network, the shared decoder network, and the mutual information evaluation network to obtain a key generation network.

[0107] As an implementation method of an embodiment of the present application, the parameters of the key generation network to be trained are initialized; the parameters of the first channel feature extraction network, the second channel feature extraction network and the shared decoder network are fixed unchanged, and the mutual information evaluation network is pre-trained using the second loss function; it is determined whether the number of training iterations of the key generation network to be trained is greater than the target number of iterations; if the number of training iterations of the key generation network to be trained is not greater than the target number of iterations, the parameters of the pre-trained mutual information evaluation network are fixed unchanged, and the parameters of the first channel feature extraction network, the second channel feature extraction network and the shared decoder network are adjusted using the loss function; the parameters of the first channel feature extraction network, the second channel feature extraction network and the shared decoder network after the parameter adjustment are fixed unchanged, and the parameters of the pre-trained mutual information evaluation network are adjusted using the second loss function; when the parameter adjustment of the pre-trained mutual information evaluation network is completed, the number of training iterations of the key generation network to be trained is controlled to increase by 1, and the execution is returned to determine whether the number of training iterations of the key generation network to be trained is greater than the target number of iterations, until the number of training iterations of the key generation network to be trained is greater than the target number of iterations, the key generation network is obtained.

[0108] In some embodiments, the number of training rounds for adjusting the parameters of the first channel feature extraction network, the second information feature extraction network, and the shared decoder network using the loss function while the parameters of the pre-trained mutual information evaluation network are fixed unchanged can be pre-set, as well as the number of training rounds for adjusting the parameters of the pre-trained mutual information evaluation network using the second loss function while the parameters of the first channel feature extraction network, the second channel feature extraction network, and the shared decoder network after the parameter adjustment are fixed unchanged; that is, by fixing the parameters of the pre-trained mutual information evaluation network unchanged, when the number of times the parameters of the first channel feature extraction network, the second information feature extraction network, and the shared decoder network are adjusted using the loss function reaches the corresponding training rounds, the parameters of the first channel feature extraction network, the second channel feature extraction network, and the shared decoder network after the parameter adjustment are fixed unchanged, and the parameters of the pre-trained mutual information evaluation network are adjusted using the second loss function until the number of adjustments reaches the corresponding training round, and the number of training iterations of the key generation network to be trained is controlled to increase by 1.

[0109] In the actual application process, the loss function is constructed using each training sample pair and its corresponding first channel feature vector, second channel feature vector, first channel reconstruction value, second channel reconstruction value and evaluation value, and the loss function is used to adjust the parameters of the first channel feature extraction network, the second channel feature extraction network, the shared decoder network and the mutual information evaluation network. The detailed process of obtaining the key generation network is as follows:

[0110] B1: Construct the first loss function using the first channel feature vector, the second channel feature vector, the first channel reconstruction value and the second channel reconstruction value of each training sample pair, as shown in formula (2):

[0111] (2)

[0112] in, is the first loss function, is the number of training sample pairs in a training batch, represents the Euclidean norm.

[0113] B2: Use the evaluation value to construct the second loss function, as shown in formula (3):

[0114] (3)

[0115] in, is the number of training sample pairs in a batch of training, is the first training sample pairs The corresponding evaluation value is is the second loss function, >0, is the mutual information regularization term weight, which can be set .

[0116] B3: Construct the corresponding loss function according to the first loss function and the second loss function, as shown in formula (4):

[0117] (4)

[0118] in, is the loss function, , is the regularization term weight.

[0119] B4: Parameters in the key generation network to be trained Initialize;

[0120] B5: Determine whether the number of pre-training times of the mutual information evaluation network is greater than the target number of pre-training times; wherein the target number of pre-training times may be 100, which is not limited in this embodiment of the present application;

[0121] B6: If the number of pre-training times of the mutual information evaluation network is not greater than the target number of pre-training times, the parameters of the mutual information evaluation network are adjusted using the second loss function sampling stochastic gradient descent algorithm, and the adjustment method is shown in formula (5);

[0122] (5)

[0123] in, The learning rate of the mutual information evaluation network can be set ;

[0124] B7: When the mutual information evaluation network parameter adjustment is completed, the pre-training times of the mutual information evaluation network are controlled to increase by 1, and the process returns to execute B5 until the pre-training times of the mutual information evaluation network are greater than the target pre-training times, and the pre-training of the mutual information evaluation network is determined to be completed, and then B8 can be executed;

[0125] B8: Determine whether the number of training iterations of the key generation network to be trained is greater than the target number of iterations, where the target number of iterations can be pre-set to ;

[0126] B9: If the number of training iterations of the key generation network to be trained is not greater than the target number of iterations, the parameters of the pre-trained mutual information evaluation network are fixed unchanged, and the parameters of the first channel feature extraction network, the second channel feature extraction network, and the shared decoder network are adjusted using the stochastic gradient descent algorithm with adaptive clipping using the loss function. The specific adjustment method is shown in formula (6), where the number of training rounds for adjusting the parameters of the first channel feature extraction network, the second channel feature extraction network, and the shared decoder network using the stochastic gradient descent algorithm with adaptive clipping using the loss function can be: , , Indicates the floor symbol;

[0127] (6)

[0128] in, is the learning rate of the first channel feature extraction network, the second channel feature extraction network and the shared decoder network, and They are loss functions About parameters and The adaptive clipping gradient of is shown in formula (7):

[0129] (7)

[0130] Among them, the regularization term weight can be set as , the initial learning rates of the first channel feature extraction network, the second channel feature extraction network and the shared decoder network can be set to , Represents the loss function The first loss function contained in and the second loss function Regarding network parameters The ratio of the gradient norm of ; This means that the above two parts of loss and Regarding network parameters The ratio of the gradient norm of .

[0131] B10: The parameters of the first channel feature extraction network, the second channel feature extraction network, and the shared decoder network after parameter adjustment are fixed, and the learning rate of the pre-trained mutual information evaluation network is set to , the parameters of the pre-trained mutual information evaluation network are adjusted by using the second loss function and the stochastic gradient descent algorithm. The specific adjustment method is shown in formula (8); wherein, the number of training rounds for adjusting the parameters of the pre-trained mutual information evaluation network by using the second loss function and the stochastic gradient descent algorithm can be , , Indicates the floor symbol;

[0132] (8)

[0133] B10: After the parameter adjustment of the pre-trained mutual information evaluation network is completed, the number of training iterations of the key generation network to be trained is controlled to increase by 1, and the process returns to execute B8 until the number of training iterations of the key generation network to be trained is greater than the target number of iterations, thereby obtaining the key generation network.

[0134] S103: deploying a first target channel feature extraction network to a legitimate communication node A, and deploying a second target channel feature extraction network to a legitimate communication node B; wherein the first target channel feature extraction network is a first channel feature extraction network in a trained key generation network, and the second target channel feature extraction network is a second channel feature extraction network in a trained key generation network.

[0135] In the specific process of executing step S103, after obtaining the key generation network, the first channel feature extraction network in the trained key generation network (for the sake of distinction, the first channel feature extraction network in the trained key generation network is referred to as the first target channel feature extraction network) can be deployed to the legitimate communication node A, and the second channel feature extraction network in the trained key generation network (for the sake of distinction, the second channel feature extraction network in the trained key generation network is referred to as the second target channel feature extraction network) can be deployed to the legitimate communication node B.

[0136] S104: The legitimate communication node A uses the deployed first target channel feature extraction network to process the collected first estimation values ​​to obtain a first target channel feature vector, and quantizes multiple first target channel feature vectors to obtain a key sequence of the legitimate communication node A.

[0137] In the specific execution of step S104, after the first target channel feature extraction network of the trained key generation network is deployed to the legitimate communication node A, the legitimate communication node A and the legitimate communication node B are connected at a preset time interval. Corresponding channel detection signals are sent alternately, so that the legitimate communication node A obtains a first estimation value of its own legitimate channel based on the received channel detection signal, and the legitimate communication node B obtains a second estimation value of its own legitimate channel based on the received channel detection signal; the legitimate communication node A is controlled to extract each collected first estimation value by using the first target feature extraction network deployed by itself to obtain multiple first target channel feature vectors, and the multiple first target channel feature vectors are quantized to obtain the key sequence of the legitimate communication node A.

[0138] It should be noted that the process of extracting features from each first estimated value through the first target feature extraction network is the same as the process of extracting features from the first estimated value in the training sample pair through the first channel feature extraction network disclosed in the above step S102. Please refer to the corresponding content in the above step S102, and the embodiments of the present application will not be repeated here.

[0139] Optionally, the process of quantizing multiple first target channel feature vectors through the legitimate communication node A to obtain the key sequence of the legitimate communication node A can be specifically as follows: the legitimate communication node A uses the deployed first target channel feature extraction network to process the collected first estimation values ​​to obtain multiple first target channel feature vectors; the legitimate communication node A counts the first data of each first target channel feature vector in each dimension, and constructs the first empirical cumulative distribution function corresponding to each dimension based on the first data in each dimension; the legitimate communication node A uses the first empirical cumulative distribution function corresponding to each dimension and a preset data set to determine the mapping result corresponding to each first data in each dimension, and converts the mapping result corresponding to each first data in each dimension into binary data; the binary data corresponding to each first data in each dimension are cascaded to obtain the key sequence of the legitimate communication node A.

[0140] Specifically, the key sequence of the legitimate communication node A is explained below by way of example:

[0141] C1: collected through legitimate communication node A first estimated values; for each collected first estimated value, control the legitimate communication node A to extract the first estimated value using the deployed first target channel feature extraction network, and obtain a dimension of of The first target channel feature vector, where The first target channel feature vector The first data in the dimension is ;

[0142] C2: Statistics The first target channel feature vector The first data in each dimension and constructed from the first data in each dimension First Empirical Cumulative Distribution Function ;

[0143] C3: Use the first empirical cumulative distribution function corresponding to each dimension and preset data sets , determine the mapping result corresponding to each first data in each dimension, wherein the specific method is as shown in formula (9);

[0144] ,like (9)

[0145] in, For the The dimension The mapping result corresponding to the first data, , , The number of binary bits after each data is quantized.

[0146] C4: Convert the mapping result corresponding to each first data in each dimension into binary data , that is, get The quantization result of each first data in the dimension; and The quantization results of each first data in the dimension are cascaded to obtain the first The first quantitative result in dimension ;in, Function that converts a decimal number to a binary number.

[0147] C5: The first quantization results on the dimensions are cascaded to obtain the key sequence of the legitimate communication node A. .

[0148] S105: The legitimate communication node B uses the deployed second target channel feature extraction network to process the collected second estimation values ​​to obtain multiple second target channel feature vectors, and quantizes the multiple second target channel feature vectors to obtain the key sequence of the legitimate communication node B.

[0149] In the specific process of executing step S105, after the second target channel feature extraction network of the trained key generation network is deployed to the legitimate communication node B, the legitimate communication node B can be controlled to use the first target feature extraction network deployed by itself to extract the collected first estimation values ​​to obtain multiple second target channel feature vectors, and the multiple second target channel feature vectors are quantized to obtain the key sequence of the legitimate communication node B.

[0150] It should be noted that the process of performing feature extraction on each second estimation value through the Bth target feature extraction network is the same as the process of performing feature extraction on the second estimation value in the training sample pair through the second channel feature extraction network disclosed in the above step S102. Please refer to the corresponding content in the above step S102, and the embodiments of the present application will not be repeated here.

[0151] Optionally, the process of quantizing multiple second target channel feature vectors through the legitimate communication node B to obtain the key sequence of the legitimate communication node B can be specifically as follows: the legitimate communication node B uses the deployed second target channel feature extraction network to process the collected second estimation values ​​to obtain multiple second target channel feature vectors; the legitimate communication node B counts the second data of each second target channel feature vector in each dimension, and constructs the second empirical cumulative distribution function corresponding to each dimension according to the second data in each dimension; the legitimate communication node B uses the second empirical cumulative distribution function corresponding to each dimension and a preset data set to determine the mapping result corresponding to each second data in each dimension, and converts the mapping result corresponding to each second data in each dimension into binary data; the binary data corresponding to each second data in each dimension are cascaded to obtain the key sequence of the legitimate communication node B.

[0152] Specifically, the key sequence of the legitimate communication node B is explained in an example manner as follows:

[0153] D1: collected by legitimate communication node B second estimated values; for each collected second estimated value, control the legitimate communication node B to extract the second estimated value using the deployed second target channel feature extraction network, and obtain a dimension of of The second target channel feature vector, where The second target channel feature vector The first data in the dimension is ;

[0154] D2: Count J second target channel feature vectors The first data in each dimension and constructed from the second data in each dimension Second Empirical Cumulative Distribution Function ;

[0155] D3: Use the first empirical cumulative distribution function corresponding to each dimension and preset data sets , determine the mapping result corresponding to each second data in each dimension, wherein the specific method is as shown in formula (10);

[0156] ,like (10)

[0157] in, For the The dimension The mapping result corresponding to the second data, , , The number of binary bits after each data is quantized.

[0158] D4: Convert the mapping result corresponding to each second data in each dimension into binary data , that is, get The quantization result of each second data in the dimension; and The quantization results of each second data in the dimension are cascaded to obtain the Second quantification result in dimension ;in, Function that converts a decimal number to a binary number.

[0159] D5: The first quantization results on the dimensions are cascaded to obtain the key sequence of the legitimate communication node B .

[0160] The embodiment of the present application provides a physical layer key generation method, which can effectively remove the non-reciprocal noise in the estimated values ​​collected by both legitimate communication nodes by designing two channel feature extraction networks based on the residual network module and the convolutional block attention module in the key generation network; further, combined with the design of the autoencoder structure of the shared decoder network, and the loss function design of the mean square error and mutual information regularization for the mean of the channel estimation samples of both parties, the channel feature extraction network in the present method can effectively compensate for the non-reciprocity caused by the asynchronous channel detection of the legitimate communication nodes in the channel estimation of both parties. Thereby solving the problem in the prior art that due to the non-reciprocal factors in the estimated values ​​of both legitimate communication nodes, the initial keys of both legitimate communication nodes have more inconsistent bits, resulting in low key generation rate and reliability, and even key generation failure, and significantly enhancing the reciprocity between the channel feature vectors extracted by both legitimate communication nodes, thereby improving the reliability and efficiency of key generation.

[0161] Furthermore, in the embodiment of the present application, in order to verify the performance of a physical layer key generation method provided by the embodiment of the present application, the present application selects two similar physical layer key generation methods as comparison methods. Comparison method 1 and comparison method 2 both use an autoencoder structure of "encoder network + decoder network", the encoder network consists of 6 convolutional layers and 1 fully connected layer, and the decoder network structure is the same as the shared decoder network structure in the method of the present application. The difference between the two comparison methods is that: Comparison method 1 adopts a single autoencoder structure, and the channel estimation samples of one of the two legal communicating parties are used as the input data of the autoencoder during training. The loss function is the mean square error between the output data of the autoencoder and the channel estimation samples of the other party. The same network is deployed on both legal communicating parties to extract the channel feature vector, and the deployed network is the encoder network in the trained single autoencoder; Comparison method 2 adopts a dual autoencoder structure. During training, the channel estimation samples of the two legal parties are respectively input into two corresponding autoencoders. The loss function is composed of the mean square error between the input and output of each of the two autoencoders and the mean square error between the channel feature vectors output by the encoder networks in the two autoencoders. The encoder networks in the two trained autoencoders are respectively deployed to the corresponding legal communicating parties to extract the channel feature vector.

[0162] Figure 8The curves of the variation of NMSE (Normalized Mean Squared Error) between the channel feature vectors of the two legitimate communication nodes extracted according to a physical layer key generation method provided by the present application, comparison method 1 and comparison method 2 as a inverse of SNR (Signal-to-Noise Ratio) are shown. Except for comparison method 1, the NMSE of the channel feature vectors extracted by the physical layer key generation method and comparison method 2 provided by the present application are lower than the NMSE of the original channel estimation sample, which indicates that the physical layer key generation method provided by the present application can improve the similarity between the channel observations of the two legitimate communication nodes by extracting the channel feature vectors; moreover, from Figure 8 It can also be seen that the NMSE performance of the channel feature vector extracted by the physical layer key generation method provided by the present application is significantly lower than that of the two comparison methods, especially in the low signal-to-noise ratio area, the performance advantage of the physical layer key generation method provided by the present application is more obvious.

[0163] Fig. 9 The correlation coefficient between the channel feature vectors of the two legitimate communication nodes extracted by the physical layer key generation method provided by the present application and the comparison method 1 and the comparison method 2 is shown as a function of SNR. The correlation coefficient between the channel feature vectors of the two legitimate communication nodes extracted by the physical layer key generation method provided by the present application and the two comparison methods is higher than the correlation coefficient between the legitimate double-transmitted original channel estimation samples, and Fig. 9 It can also be seen that the correlation coefficient of the channel feature vector extracted by the physical layer key generation method provided by the present application is significantly higher than that of the two comparison methods, which means that the physical layer key generation method provided by the present application can significantly enhance the correlation between the channel observations of both legitimate communication nodes by extracting the channel feature vector; in addition, even in the low signal-to-noise ratio area, the physical layer key generation method provided by the present application can still obtain a correlation coefficient close to 1.

[0164] comprehensive Figure 8 and Fig. 9 It can be seen from the results that, whether from the perspective of error or correlation, the physical layer key generation method provided by the present application can significantly improve the reciprocity between the channel feature vectors of the legitimate parties, which is conducive to the generation of a more consistent key sequence by both parties.

[0165] Fig.10The curves showing the inconsistency rate between the key sequences of the two legal communication nodes generated by the physical layer key generation method provided by the present application and the comparison method 1 and the comparison method 2 as a function of SNR are shown. The inconsistency rate of the key sequence generated by the physical layer key generation method provided by the present application is significantly lower than that of the two comparison methods, especially in the low signal-to-noise ratio area, and the physical layer key generation method provided by the present application has a more significant performance advantage.

[0166] Fig.11 The curves showing the variation of the average key length with SNR according to the physical layer key generation method provided by the present application and the comparison method 1 and the comparison method 2 are shown. The length of the key sequence that can be generated by the physical layer key generation method provided by the present application by averaging the estimated value (first estimated value and second estimated value) of each training sample pair is significantly greater than that of the two comparison algorithms, especially in the low signal-to-noise ratio area, the average key length obtained by the physical layer key generation method provided by the present application is significantly greater than that of the two comparison algorithms.

[0167] comprehensive Fig.10 and Fig.11 It can be seen from the results that the physical layer key generation method provided by the present application can extract channel feature vectors with better reciprocity, so that the physical layer key generation method provided by the present application can achieve more reliable and efficient key generation.

[0168] Based on the physical layer key generation method provided in the above embodiment of the present application, correspondingly, the embodiment of the present application also provides a physical layer key generation system, such as Fig.12 As shown, the physical layer key generation system includes:

[0169] A training data set construction unit, configured to construct a corresponding training data set based on each first estimated value collected based on the legitimate communication node A and each second estimated value collected based on the legitimate communication node B; wherein the training data set includes a plurality of training sample pairs, and the training sample pairs include the first estimated value and the second estimated value;

[0170] A training unit, used for inputting a training data set into a key generation network to be trained, so that the key generation network to be trained constructs a corresponding loss function using the input training data set, and performs iterative training based on the loss function to obtain a key generation network;

[0171] A deployment unit, used to deploy a first target channel feature extraction network to a legitimate communication node A, and to deploy a second target channel feature extraction network to a legitimate communication node B; wherein the first target channel feature extraction network is a first channel feature extraction network in a trained key generation network, and the second target channel feature extraction network is a second channel feature extraction network in a trained key generation network;

[0172] The legitimate communication node A is used to process the collected first estimation values ​​by using the deployed first target channel feature extraction network to obtain a plurality of first target channel feature vectors, and to quantize the plurality of first target channel feature vectors to obtain a key sequence of the legitimate communication node A;

[0173] The legitimate communication node B is used to process the collected second estimation values ​​using the deployed second target channel feature extraction network to obtain multiple second target channel feature vectors, and quantize the multiple second target channel feature vectors to obtain the key sequence of the legitimate communication node B.

[0174] The specific principles and execution processes of each unit in the physical layer key generation system disclosed in the above-mentioned embodiment of the present application are the same as the physical layer key generation method disclosed in the above-mentioned embodiment of the present application. Please refer to the corresponding parts of the physical layer key generation method disclosed in the above-mentioned embodiment of the present application, and will not be repeated here.

[0175] The embodiment of the present application provides a physical layer key generation system, which constructs a corresponding training data set based on each first estimation value collected based on a legitimate communication node A and each second estimation value collected based on a legitimate communication node B; wherein the training data set includes a plurality of training sample pairs, and the training sample pairs include a first estimation value and a second estimation value; the training data set is input into a key generation network to be trained, so that the key generation network to be trained constructs a corresponding loss function using the input training data set, and iterative training is performed based on the loss function to obtain a key generation network; a first target channel feature extraction network is deployed to the legitimate communication node A, and a second target channel feature extraction network is deployed to the legitimate communication node B; wherein the first target channel feature extraction network The network is a first channel feature extraction network in a trained key generation network, and the second target channel feature extraction network is a second channel feature extraction network in a trained key generation network; the legitimate communication node A uses the deployed first target channel feature extraction network to process each collected first estimation value to obtain multiple first target channel feature vectors, and quantizes the multiple first target channel feature vectors to obtain a key sequence of the legitimate communication node A; the legitimate communication node B uses the deployed second target channel feature extraction network to process each collected second estimation value to obtain multiple second target channel feature vectors, and quantizes the multiple second target channel feature vectors to obtain a key sequence of the legitimate communication node B. The technical means provided in the present application constructs a corresponding training data set in advance using each first estimation value collected based on the legitimate communication node A and each second estimation value collected based on the legitimate communication node B to train the key generation network to be trained, and deploys the two channel feature extraction networks in the trained key generation network to the legitimate communication node A and the legitimate communication node B, respectively, so that the legitimate communication node A and the legitimate communication node B can respectively use the deployed channel feature extraction networks to effectively remove the non-reciprocity caused by noise in the collected estimation values, and compensate for the non-reciprocity caused by time asynchrony in the collected estimation values, which can significantly enhance the reciprocity between the channel feature vectors extracted by the two legitimate communication nodes, thereby improving the reliability and efficiency of key generation.

[0176] Optionally, the key generation network to be trained is composed of two channel feature extraction networks, a shared decoder network and a mutual information evaluation network, and the training unit includes:

[0177] An input unit, used for inputting each training sample pair in the training data set into the key generation network to be trained;

[0178] A channel feature vector extraction unit is used to extract features from a first estimated value in an input training sample pair through a first channel feature extraction network for each training sample pair to obtain a first channel feature vector, and to extract features from a second estimated value in an input training sample pair through a second channel feature extraction network to obtain a second channel feature vector;

[0179] A channel reconstruction unit, used to perform channel reconstruction on the first channel feature vector and the second channel feature vector respectively through a shared decoder network to obtain a first channel reconstruction value and a second channel reconstruction value;

[0180] An evaluation unit, configured to perform evaluation using the first channel feature vector and the second channel feature vector through a mutual information evaluation network to obtain an evaluation value;

[0181] The first training subunit is used to construct a loss function using each training sample pair and its corresponding first channel feature vector, second channel feature vector, first channel reconstruction value, second channel reconstruction value and evaluation value, and use the loss function to adjust the parameters of the first channel feature extraction network, the second channel feature extraction network, the shared decoder network and the mutual information evaluation network to obtain a key generation network.

[0182] Optionally, a channel reconstruction unit comprises:

[0183] A first channel reconstruction subunit is used to process the first channel feature vector through a shared decoder network to obtain a first initial vector, and perform channel reshaping on the first initial vector to obtain a first channel coarse reconstruction value; perform refinement processing on the first channel coarse reconstruction value to obtain a first channel refined reconstruction value, and perform convolution processing on the first channel refined reconstruction value to obtain a first channel reconstruction value;

[0184] The second channel reconstruction subunit is used to process the second channel feature vector through the shared decoder network to obtain a second initial vector, and perform channel reshaping on the second initial vector to obtain a coarse reconstruction value of the second channel; refine the coarse reconstruction value of the second channel to obtain a refined reconstruction value of the second channel, and perform convolution processing on the refined reconstruction value of the second channel to obtain a second channel reconstruction value.

[0185] Optional, evaluation unit, specifically for:

[0186] The mutual information evaluation network concatenates the first channel feature vector and the second channel feature vector to obtain a single channel matrix, and inputs the single channel matrix into the mutual information evaluation network, so that the mutual information evaluation network performs convolution processing on the single channel matrix, and reshapes and processes the data obtained after the convolution processing to obtain a corresponding evaluation value.

[0187] Optionally, the first training subunit includes:

[0188] A first loss function construction unit, used to construct a first loss function using each training sample pair and its corresponding first channel feature vector, second channel feature vector, first channel reconstruction value and second channel reconstruction value;

[0189] A second loss function construction unit, used to construct a second loss function using the evaluation value;

[0190] A loss function construction unit, used to construct a corresponding loss function according to the first loss function and the second loss function;

[0191] An initialization unit, used for initializing the parameters of the key generation network to be trained;

[0192] A pre-training unit, used to fix the parameters of the first channel feature extraction network, the second channel feature extraction network and the shared decoder network, and pre-train the mutual information evaluation network using a second loss function;

[0193] A judging unit, used to judge whether the number of training iterations of the key generation network to be trained is greater than the target number of iterations;

[0194] A first parameter adjustment unit is used to fix the parameters of the pre-trained mutual information evaluation network unchanged if the number of training iterations of the key generation network to be trained is not greater than the target number of iterations, and adjust the parameters of the first channel feature extraction network, the second channel feature extraction network and the shared decoder network by using a loss function;

[0195] A second parameter adjustment unit is used to fix the parameters of the first channel feature extraction network, the second channel feature extraction network and the shared decoder network after the parameter adjustment, and adjust the parameters of the pre-trained mutual information evaluation network by using a second loss function;

[0196] The self-increment unit is used to control the number of training iterations of the key generation network to be trained to be incremented by 1 after the parameter adjustment of the mutual information evaluation network is completed, and return to the execution judgment unit until the number of iterations of the key generation network to be trained is greater than the target number of iterations, so as to obtain the key generation network.

[0197] Optionally, the legal communication node A is specifically used for:

[0198] The collected first estimation values ​​are processed by using the deployed first target channel feature extraction network to obtain multiple first target channel feature vectors; the first data of each first target channel feature vector in each dimension are counted, and the first empirical cumulative distribution function corresponding to each dimension is constructed according to the first data in each dimension; the mapping result corresponding to each first data in each dimension is determined by using the first empirical cumulative distribution function corresponding to each dimension and a preset data set, and the mapping result corresponding to each first data in each dimension is converted into binary data; the binary data corresponding to each first data in each dimension are cascaded to obtain the key sequence of the legitimate communication node A.

[0199] Optionally, the legal communication node B is specifically used for:

[0200] The collected second estimation values ​​are processed by using the deployed second target channel feature extraction network to obtain multiple second target channel feature vectors; the second data of each second target channel feature vector in each dimension are counted, and the second empirical cumulative distribution function corresponding to each dimension is constructed according to the second data in each dimension; the mapping result corresponding to each second data in each dimension is determined by using the second empirical cumulative distribution function corresponding to each dimension and a preset data set, and the mapping result corresponding to each second data in each dimension is converted into binary data; the binary data corresponding to each second data in each dimension are cascaded to obtain the key sequence of the legitimate communication node B.

[0201] The present application also provides a storage medium, in which program instructions are stored. When the program instructions are loaded and executed by a processor, any one of the above-mentioned physical layer key generation method embodiments is implemented.

[0202] The present application also provides an electronic device, such as Fig.13 As shown, the device includes a processor 1301 and a memory 1302; the memory stores program instructions; the processor calls the program instructions in the memory to execute any one of the above-mentioned physical layer key generation method embodiments.

[0203] The processor in this article can be a CPU of a terminal, or an MCU integrated in the terminal, or a combination of a CPU and an MCU. Moreover, the processor includes a kernel, which calls a corresponding program from a memory, and one or more kernels can be set.

[0204] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0205] For the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0206] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The modules and sub-modules in the methods and terminals of each embodiment of the present application can be merged, divided and deleted according to actual needs.

[0207] In the several embodiments provided in the present application, it should be understood that the disclosed terminals, devices and methods can be implemented in other ways. For example, the terminal embodiments described above are only schematic, for example, the division of modules or submodules is only a logical function division, and there may be other division methods in actual implementation, for example, multiple submodules or modules can be combined or integrated into another module, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0208] The modules or submodules described as separate components may or may not be physically separated, and the components of the modules or submodules may or may not be physical modules or submodules, that is, they may be located in one place, or they may be distributed on multiple network modules or submodules. Some or all of the modules or submodules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0209] In addition, each functional module or submodule in each embodiment of the present application may be integrated into one processing module, or each module or submodule may exist physically separately, or two or more modules or submodules may be integrated into one module. The above-mentioned integrated modules or submodules may be implemented in the form of hardware or in the form of software functional modules or submodules.

[0210] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0211] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0212] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A physical layer key generation method, characterized in that: The method comprises: Constructing a corresponding training data set based on each first estimation value collected based on the legitimate communication node A and each second estimation value collected based on the legitimate communication node B; wherein the training data set includes a plurality of training sample pairs, and the training sample pairs include the first estimation value and the second estimation value; Inputting the training data set into the key generation network to be trained, so that the key generation network to be trained constructs a corresponding loss function using the input training data set, and performs iterative training based on the loss function to obtain the key generation network; Deploy a first target channel feature extraction network to a legitimate communication node A, and deploy a second target channel feature extraction network to a legitimate communication node B; wherein the first target channel feature extraction network is a first channel feature extraction network in a trained key generation network, and the second target channel feature extraction network is a second channel feature extraction network in a trained key generation network; The legitimate communication node A uses the deployed first target channel feature extraction network to process the collected first estimation values ​​to obtain a plurality of first target channel feature vectors, and quantizes the plurality of first target channel feature vectors to obtain a key sequence of the legitimate communication node A; The legitimate communication node B uses the deployed second target channel feature extraction network to process the collected second estimation values ​​to obtain a plurality of second target channel feature vectors, and quantizes the plurality of second target channel feature vectors to obtain a key sequence of the legitimate communication node B; The key generation network to be trained is composed of two channel feature extraction networks, a shared decoder network and a mutual information evaluation network; The step of inputting the training data set into the key generation network to be trained, allowing the key generation network to be trained to construct a corresponding loss function using the input training data set, and performing training based on the loss function to obtain the key generation network includes: Input each training sample pair in the training data set into the key generation network to be trained; For each of the training sample pairs, feature extraction is performed on a first estimated value in the input training sample pair through a first channel feature extraction network to obtain a first channel feature vector, and feature extraction is performed on a second estimated value in the input training sample pair through a second channel feature extraction network to obtain a second channel feature vector; Reconstructing the first channel feature vector and the second channel feature vector through the shared decoder network to obtain a first channel reconstruction value and a second channel reconstruction value; Performing evaluation using the first channel feature vector and the second channel feature vector through the mutual information evaluation network to obtain an evaluation value; A loss function is constructed using each of the training sample pairs and their corresponding first channel feature vectors, second channel feature vectors, first channel reconstruction values, second channel reconstruction values ​​and evaluation values, and the loss function is used to adjust the parameters of the first channel feature extraction network, the second channel feature extraction network, the shared decoder network and the mutual information evaluation network to obtain a key generation network.

2. The method according to claim 1, characterized in that The performing channel reconstruction on the first channel feature vector and the second channel feature vector respectively through the shared decoder network to obtain a first channel reconstruction value and a second channel reconstruction value includes: The first channel feature vector is processed by the shared decoder network to obtain a first initial vector, and the first initial vector is channel reshaped to obtain a first channel coarse reconstruction value; the first channel coarse reconstruction value is refined to obtain a first channel refined reconstruction value, and the first channel refined reconstruction value is convolved to obtain a first channel reconstruction value; The second channel feature vector is processed by the shared decoder network to obtain a second initial vector, and the second initial vector is channel reshaped to obtain a second channel coarse reconstruction value; the second channel coarse reconstruction value is refined to obtain a second channel refined reconstruction value, and the second channel refined reconstruction value is convolved to obtain a second channel reconstruction value.

3. The method according to claim 1, characterized in that The step of performing evaluation using the first channel feature vector and the second channel feature vector through the mutual information evaluation network to obtain an evaluation value includes: The first channel eigenvector and the second channel eigenvector are concatenated to obtain a single channel matrix, and the single channel matrix is ​​input into a mutual information evaluation network, so that the mutual information evaluation network performs convolution processing on the single channel matrix, and reshapes and processes the data obtained after the convolution processing to obtain a corresponding evaluation value.

4. The method according to claim 1, characterized in that: The method comprises: constructing a loss function by using each pair of the training samples and the first channel feature vector, the second channel feature vector, the first channel reconstruction value, the second channel reconstruction value and the evaluation value corresponding thereto, and adjusting the parameters of the first channel feature extraction network, the second channel feature extraction network, the shared decoder network and the mutual information evaluation network by using the loss function to obtain a key generation network, including: Constructing a first loss function using each of the training sample pairs and their corresponding first channel feature vectors, second channel feature vectors, first channel reconstruction values, and second channel reconstruction values; Constructing a second loss function using the evaluation value; Construct a corresponding loss function according to the first loss function and the second loss function; Initializing parameters of the key generation network to be trained; Fixing the parameters of the first channel feature extraction network, the second channel feature extraction network, and the shared decoder network unchanged, and pre-training the mutual information evaluation network using the second loss function; Determining whether the number of training iterations of the key generation network to be trained is greater than a target number of iterations; If the number of training iterations of the key generation network to be trained is not greater than the target number of iterations, the parameters of the pre-trained mutual information evaluation network are fixed unchanged, and the parameters of the first channel feature extraction network, the second channel feature extraction network and the shared decoder network are adjusted using the loss function; The parameters of the first channel feature extraction network, the second channel feature extraction network, and the shared decoder network after parameter adjustment are fixed unchanged, and the parameters of the pre-trained mutual information evaluation network are adjusted using the second loss function; When the parameter adjustment of the pre-trained mutual information evaluation network is completed, the number of training iterations of the key generation network to be trained is controlled to increase by 1, and the execution is returned to determine whether the number of training iterations of the key generation network to be trained is greater than the target number of iterations, until the number of training iterations of the key generation network to be trained is greater than the target number of iterations, the key generation network is obtained.

5. The method according to claim 1, characterized in that The method of processing the collected first estimation values ​​by the legitimate communication node A using the deployed first target channel feature extraction network to obtain a plurality of first target channel feature vectors, and quantizing the plurality of first target channel feature vectors to obtain a key sequence of the legitimate communication node A includes: The legitimate communication node A processes the collected first estimation values ​​by using the deployed first target channel feature extraction network to obtain a plurality of first target channel feature vectors; Counting the first data of each of the first target channel feature vectors in each dimension through the legal communication node A, and constructing a first empirical cumulative distribution function corresponding to each dimension according to each of the first data in each dimension; Determine, by the legal communication node A, a mapping result corresponding to each first data on each dimension using the first empirical cumulative distribution function corresponding to each dimension and a preset data set, and convert the mapping result corresponding to each first data on each dimension into binary data; The legal communication node A cascades the binary data corresponding to each first data in each dimension to obtain a key sequence of the legal communication node A.

6. The method according to claim 1, characterized in that The method of processing the collected second estimation values ​​by the legitimate communication node B using the deployed second target channel feature extraction network to obtain a plurality of second target channel feature vectors, and quantizing the plurality of second target channel feature vectors to obtain a key sequence of the legitimate communication node B includes: The legitimate communication node B uses the deployed second target channel feature extraction network to process the collected second estimation values ​​to obtain multiple second target channel feature vectors; Counting the second data of each of the second target channel feature vectors in each dimension through the legal communication node B, and constructing a second empirical cumulative distribution function corresponding to each dimension according to each of the second data in each dimension; Determine, by the legal communication node B, a mapping result corresponding to each second data on each dimension using the second empirical cumulative distribution function corresponding to each dimension and a preset data set, and convert the mapping result corresponding to each second data on each dimension into binary data; The legitimate communication node B cascades the binary data corresponding to each second data in each dimension to obtain a key sequence of the legitimate communication node B.

7. A physical layer key generation system, characterized in that: The system comprises: A training data set construction unit, configured to construct a corresponding training data set based on each first estimation value collected based on the legitimate communication node A and each second estimation value collected based on the legitimate communication node B; wherein the training data set includes a plurality of training sample pairs, and the training sample pairs include the first estimation value and the second estimation value; A training unit, used for inputting the training data set into the key generation network to be trained, so that the key generation network to be trained constructs a corresponding loss function using the input training data set, and performs iterative training based on the loss function to obtain the key generation network; A deployment unit, configured to deploy a first target channel feature extraction network to a legitimate communication node A, and to deploy a second target channel feature extraction network to a legitimate communication node B; wherein the first target channel feature extraction network is a first channel feature extraction network in a trained key generation network, and the second target channel feature extraction network is a second channel feature extraction network in a trained key generation network; The legitimate communication node A is used to process the collected first estimation values ​​by using the deployed first target channel feature extraction network to obtain a plurality of first target channel feature vectors, and to quantize the plurality of first target channel feature vectors to obtain a key sequence of the legitimate communication node A; The legitimate communication node B is used to process the collected second estimation values ​​by using the deployed second target channel feature extraction network to obtain a plurality of second target channel feature vectors, and to quantize the plurality of second target channel feature vectors to obtain a key sequence of the legitimate communication node B; The key generation network to be trained is composed of two channel feature extraction networks, a shared decoder network and a mutual information evaluation network, and the training unit includes: An input unit, used for inputting each training sample pair in the training data set into the key generation network to be trained; A channel feature vector extraction unit is used to perform feature extraction on a first estimated value in the input training sample pair for each of the training sample pairs through a first channel feature extraction network to obtain a first channel feature vector, and perform feature extraction on a second estimated value in the input training sample pair through a second channel feature extraction network to obtain a second channel feature vector; a channel reconstruction unit, configured to perform channel reconstruction on the first channel feature vector and the second channel feature vector respectively through the shared decoder network to obtain a first channel reconstruction value and a second channel reconstruction value; an evaluation unit, configured to perform evaluation using the first channel feature vector and the second channel feature vector through the mutual information evaluation network to obtain an evaluation value; The first training subunit is used to construct a loss function using each of the training sample pairs and their corresponding first channel feature vectors, second channel feature vectors, first channel reconstruction values, second channel reconstruction values ​​and evaluation values, and use the loss function to adjust the parameters of the first channel feature extraction network, the second channel feature extraction network, the shared decoder network and the mutual information evaluation network to obtain a key generation network.

8. An electronic device, characterized in that: include: A processor and a memory, wherein the processor and the memory are connected via a communication bus; wherein the processor is used to call and execute a program stored in the memory; The memory is used to store a program, and the program is used to implement the physical layer key generation method as described in any one of claims 1-6.

9. A storage medium, characterized in that: The storage medium stores computer executable instructions, and the computer executable instructions are used to execute the physical layer key generation method as described in any one of claims 1-6.

Images