Wireless Physical Layer Group Key Generation Method and System Based on Group Consensus Mapping Network

Abstract

The present invention discloses a method and system for generating wireless physical layer group keys based on a group consensus mapping network, including a network training stage, an intra-group common key generation stage, and an inter-group common key generation stage. Through the group consensus mapping network implemented by using a CNN-LSTM model, multi-user groups can achieve the goal of intra-group consensus during the channel probing process, reducing the communication and time complexity of channel probing in multi-user scenarios, achieving a higher group key generation rate, and reducing the risk of information leakage.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence and information security, and mainly relates to a method and system for generating a wireless physical layer group key based on a group consensus mapping network. Background Art

[0002] The physical layer key generation technology based on wireless channel characteristics has received extensive attention as an alternative to symmetric cryptography. Most physical layer key generation technologies aim to generate pairwise keys between point-to-point devices. Compared with traditional encryption algorithms based on computational difficulty assumptions, this method uses the correlated channel variations between legitimate users as a common random factor to generate pairwise keys between two devices. As long as the distance between the eavesdropper and the legitimate device is greater than half a wavelength, the eavesdropper cannot infer information about the shared key.

[0003] The multi-user group key generation technology based on the physical layer has also been used as an alternative method for group key encryption algorithms. By utilizing the physical channel characteristics to generate and update group keys in a multi-user scenario, it has the potential to achieve information-theoretic security. However, there is still relatively little research on wireless physical layer group key generation technology at present, and most of the research focuses on generating physical layer keys between point-to-point users.

[0004] The existing wireless physical layer group key generation methods can be roughly divided into two types: the first is the group key generation based on pairwise keys, where shared keys are generated between pairwise users according to the physical layer key generation method, and then the pairwise keys between the pairwise users are used to generate the group key; the second is the group key generation based on group consensus, where channel probing is first performed between pairwise nodes, and then the nodes in the network exchange the observed channel measurement results, and all nodes reach a consensus on the random source used. However, there are certain defects in both methods at present. First, for the group key generation method based on pairwise keys, the group key needs to be updated frequently, and a large number of pairwise keys involved also need to be updated, resulting in very high computational and communication overheads, making this solution inefficient; the existing group consensus-based method uses the central node to broadcast the channel difference to achieve group consensus, but the central node broadcasting the channel difference still requires a large amount of communication and time overheads, resulting in a low group key generation rate and a risk of information leakage. Summary of the Invention

[0005] In view of the problems of low group key generation rate and relatively large security risks in the prior art, the present invention provides a wireless physical layer group key generation method and system based on a group consensus mapping network, including a network training stage, an intra-group common key generation stage, and an inter-group common key generation stage; in the network training stage, nodes perform channel estimation on the probe frames broadcast by the base station and preprocess the CSI data; the child nodes send the encrypted CSI data to the central node; the central node generates a data set based on the CSI data of the child nodes and itself; the central node uses the data set to train the group consensus mapping network for the child nodes and encrypts and sends the trained model to the child nodes; in the intra-group common key generation stage, all nodes perform channel estimation on the probe frames broadcast by the base station and preprocess the CSI data; the child nodes use the CSI data as the input of the trained group consensus mapping network model and obtain the predicted output of the model; the central node and the child nodes respectively use the vector quantization algorithm to quantize the collected CSI data and the predicted output of the model to obtain the original keys K c and K i ; the central node and the child nodes perform key negotiation on K c and K i to correct and obtain the group key K in ; in the inter-group common key generation stage, the central nodes of each group encrypt and send the intra-group common key K in to the base station, and the base station generates the inter-group common key K out by combining all the received intra-group common keys, and encrypts and broadcasts it to all nodes. By using the group consensus mapping network implemented by the CNN-LSTM model, the multi-user group can achieve the goal of intra-group consensus during the channel probing process, reducing the communication and time complexity of channel probing in the multi-user scenario and achieving a higher group key generation rate.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a wireless physical layer group key generation method based on a group consensus mapping network, characterized by including the following steps:

[0007] S1, network training stage: Nodes perform channel estimation on the probe frames broadcast by the base station and preprocess the CSI data; the child nodes send the encrypted CSI data to the central node, and the central node generates a data set based on the CSI data of the child nodes and itself; the central node uses the data set to train the group consensus mapping network for the child nodes and encrypts and sends the trained model to the child nodes;

[0008] S2: Intra-group common key generation phase: All nodes perform channel estimation on the probe frames broadcast by the base station and preprocess the CSI data; the child nodes use the CSI data as the input of the trained group consensus mapping network model and obtain the predicted output of the model; the central node and the child nodes respectively use the vector quantization algorithm to quantize the collected CSI data and the predicted output of the model to obtain the original keys K c and K i ; the central node and the child nodes perform key negotiation on K c and K i to correct and obtain the group key K in ;

[0009] S3: Inter-group common key generation phase: Each group's central node encrypts and sends the intra-group common key K in to the base station, and the base station generates the inter-group common key K out by combining all the received intra-group common keys and broadcasts it to all nodes after encryption.

[0010] As an improvement of the present invention, the step S1 specifically includes:

[0011] S11: The base station broadcasts a probe frame containing a data packet index, and all nodes perform channel estimation to extract CSI data and preprocess it. The nodes need to meet the requirement that the time slot is greater than the coherence time when extracting CSI data from two adjacent data packets. All nodes synchronize the CSI data according to the data packet index;

[0012] S12: When all users have collected enough CSI data, all child nodes encrypt the CSI data with the intra-group common key generated in the previous time period and send it to the central node of their group. The central node decrypts it to obtain the CSI data of all child nodes in its group;

[0013] S13: The central node uses the CSI data it collected as the label and constructs a CSI data set with the child nodes' CSI data as the input. The input of the data set is the corresponding child nodes' CSI data, and the labels are all the central node's CSI data;

[0014] S14: The central node trains a group consensus mapping network for each child node according to the corresponding child nodes' CSI data sets. After the training is completed, the generated model is encrypted with the pairwise key between the central node and the corresponding child node and sent to the corresponding child node.

[0015] As an improvement of the present invention, the group consensus mapping network consists of a CNN module and an LSTM module. Among them, the CNN module consists of two convolutional layers and two pooling layers; the LSTM module consists of two long short-term memory layers and a fully connected layer. The training process of the network model is as follows: using the CSI data of the child nodes as the model input and the CSI data of the central node as the target output, calculating the error between the target output and the predicted output using the mean square loss function, and iteratively updating the network parameters using the backpropagation algorithm until the convergence requirement is met.

[0016] As another improvement of the present invention, in step S2, the child nodes use the CSI data as the input of the trained group consensus mapping network model. The trained group consensus mapping network is distributed to the child nodes by the central node in sequence. Each child node uses the deployed group consensus mapping network to map the CSI data of the central node to achieve the goal of group consensus and obtain the predicted output of the model.

[0017] As another improvement of the present invention, the specific process of the vector quantization algorithm in step S2 includes:

[0018] S21: The central node randomly selects k initial cluster centers;

[0019] S22: The central node calculates the Euclidean distance between each sample point of the CSI and each initial cluster center;

[0020] S23: The central node divides each sample point into the cluster center with the closest distance according to the Euclidean distance between the sample point and the cluster center;

[0021] S24: The central node recalculates the new k cluster centers according to the sample allocation result. If the cluster centers change, repeat step S22 to start a new round of iteration;

[0022] S25: The central node encrypts the finally obtained k cluster centers and broadcasts them to all child nodes within the group. The child nodes use the same method as in step S22 and step S23 to cluster the predicted output of the model;

[0023] S26: The central node and the child nodes quantify the samples according to the region index corresponding to the finally obtained sample points to obtain the original key sequences K c and K i .

[0024] As yet another improvement of the present invention, the least squares channel estimation method is used to implement channel estimation of the detection frame broadcast by the base station by the node, and the CSI data is calculated through the formula where Y is the long training symbol in the predefined preamble, X 1 and X 2respectively represent two long training symbols extracted from the received sounding frame.

[0025] As a further improvement of the present invention, the preprocessing of the CSI data at least includes stacking the real part and the imaginary part of the CSI matrix, stretching the stacked matrix into a one-dimensional vector of size 64, and normalizing the one-dimensional vector so that its data range is between -1 and 1.

[0026] As a further improvement of the present invention, the step S3 specifically includes:

[0027] S31: There are multiple groups in the current environment. The central nodes of each group use the pairwise key with the base station to encrypt the intra-group common key K in and sequentially send it to the base station;

[0028] S32: The base station decrypts using the corresponding pairwise key to obtain the intra-group common key K in , and the base station performs an exclusive OR operation on all received intra-group common keys K in to generate an inter-group common key K out ;

[0029] S33: The base station encrypts the new inter-group common key K out using all the inter-group common keys held by all users generated in the previous time period and broadcasts it to all user nodes. The user nodes decrypt using the inter-group common key of the previous time period to obtain the new inter-group common key K out , and thereafter, inter-group broadcast communication uses the new key K out .

[0030] To achieve the above object, the technical solution adopted by the present invention is also: a wireless physical layer group key generation system based on a group consensus mapping network, including a computer program, and when the computer program is executed by a processor, it implements the steps of any one of the above methods.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention provides a wireless physical layer group key generation method based on a group consensus mapping network. Through the group consensus mapping network implemented by using a CNN-LSTM model, the sub-nodes within the group map the reference channel through the group consensus mapping network, and all nodes within the group reach an agreement on the channel characteristics for quantization. The multi-user group can achieve the goal of in-group consensus during the channel probing process. During the channel probing process, only the base station needs to broadcast the probing frame, and there is no need to perform channel probing between paired users, reducing the communication complexity of channel probing to a constant level, which is independent of the in-group user scale, greatly reducing the communication complexity, and effectively improving the group key generation rate. In addition, by using the vector quantization algorithm, the utilization rate of channel characteristics is improved, effectively improving the wireless physical layer group key generation rate, saving channel probing time, and reducing the risk of information leakage. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is the flowchart of the steps of the wireless physical layer group key generation method based on the group consensus mapping network in the present invention;

[0033] Figure 2 is the working framework diagram of the wireless physical layer group key generation method based on the group consensus mapping network in the present invention;

[0034] Figure 3 is the schematic diagram of the topological structure of the group consensus mapping network model in the present invention;

[0035] Figure 4 is the training flowchart of the group consensus mapping network in Example 1 of the present invention;

[0036] Figure 5 is the flowchart of the vector quantization algorithm in Example 1 of the present invention;

[0037] Figure 6 is the schematic diagram of the two-dimensional vector quantization algorithm in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] The following further clarifies the present invention in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and not to limit the scope of the present invention.

[0039] Example 1

[0040] A wireless physical layer group key generation method based on a group consensus mapping network, as Figure 1 shown, includes the following steps:

[0041] Step S1: Network training phase: Nodes perform channel estimation on the probe frames broadcast by the base station and preprocess the CSI data; the child nodes send the encrypted CSI data to the central node; the central node generates a dataset based on the CSI data of the child nodes and itself; the central node uses the dataset to train a group consensus mapping network for the child nodes and sends the trained model to the child nodes after encryption, as Figure 2 shown.

[0042] S11: The base station broadcasts IEEE 802.11n probe frames containing packet indices, and all nodes use the least squares channel estimation method to extract CSI data, which is calculated by the formula where Y is the long training symbol in the predefined preamble, and X 1 and X 2 represent two long training symbols extracted from the received probe frames respectively.

[0043] The requirement for nodes to extract CSI data from two adjacent packets is that the time slot is greater than the coherence time, and the coherence time is calculated by the formula where f d is the Doppler shift, f c is the carrier frequency of the wireless signal, v is the object moving speed, and c is the speed of light propagation.

[0044] In this example, the used wireless carrier frequency is 3.7 Ghz, the object moving speed in the environment is 5 km / h, and the calculated channel coherence time is 25 ms. By setting the CSI sampling period to 30 ms, the CSI sampling rate can be obtained as approximately 33 sample / sec. CSI data is synchronized between different nodes according to the packet indices.

[0045] S12: Each node collects 3000 CSI samples for training the model. Each CSI sample consists of 64 subcarriers, and we only use the first 32 subcarriers. Since the neural network cannot process complex numbers, it is necessary to preprocess the CSI data. Stack the real and imaginary parts of the CSI matrix and stretch the stacked matrix into a one-dimensional vector of size 64. The j-th CSI sample of the central node and child node i can be expressed as:

[0046] C G (j) = [real(a(j,1)), …, real(a(j,32)), imag(a(j,1)), …, imag(a(j,32))]

[0047] C i (j) = [real(b(j,1)), …, real(b(j,32)), imag(b(j,2)), …, imag(b(j,32))]

[0048] After the central node and child node i have collected 3000 CSI sample data, child node i encrypts and sends the CSI data stream C i ={C i (1), C i (2), …, C i (3000)} to the central node using a paired key, and the central node obtains the CSI data of all child nodes after decryption. All child nodes perform the same operations as child node i.

[0049] S13: The central node uses C i and C G as the training data and labels of the dataset respectively, and obtains the dataset S i corresponding to child node i containing 3000 training samples and labels. Both the training samples and labels are one-dimensional vectors of 1×64.

[0050] S14: The dataset S i is divided into a training set, a validation set, and a test set in a ratio of 7:2:1 using the non-repetitive random sampling technique. Among them, the non-repetitive random sampling technique is implemented through the train_test_split auxiliary function on scikit-learn.

[0051] The central node trains a group consensus mapping network model for each child node based on the corresponding child node training set. The group consensus mapping network is a neural network used to achieve channel mapping and reach group consensus. The meaning of group consensus is that in the wireless physical layer key generation technology, users within a group need to hold the same or similar channel feature sequences in order to obtain similar bit sequences in the subsequent quantization step. However, in a multi-user scenario, it is impossible to achieve that all users hold similar channel features only by channel probing, and additional steps are needed to achieve the goal of reaching group consensus. This patent realizes this goal based on deep learning. The group consensus mapping network consists of a CNN module and an LSTM module. Among them, the CNN module consists of two convolutional layers and two pooling layers, which are used to extract the spatial features of the input sequence and reduce the data dimension; the LSTM module consists of two long short-term memory layers and a fully connected layer, which are used to extract the temporal features. The training process of the model is as follows: The CSI data of the child node is used as the model input, and the CSI data of the central node is used as the target output. The mean square loss function is used to calculate the error between the target output and the predicted output, and the backpropagation algorithm is used to iteratively update the network parameters until the convergence requirement is met.

[0052] The topological structure of the group consensus mapping network model is as Figure 3As shown in the figure, it consists of 2 convolutional layers, 2 pooling layers, 2 long short-term memory layers and 1 fully connected layer. Among them, the number of filters in the convolutional layers are 64 and 128 respectively, the pooling windows of the two pooling layers are both 2, the number of neurons in the long short-term memory layers are 128 and 64 respectively. The number of neurons in the fully connected layer is 64, using the linear activation function, and the output sequence is a one-dimensional vector of 1×64.

[0053] Referring to Figure 4 , the training process of the group consensus mapping network is as follows:

[0054] S141: Input the training set of the CSI data of the child nodes;

[0055] S142: Network initialization: Determine the parameters and structure of the network before network training, determine the topological structure of the group consensus mapping network, and initialize the weights between layers and the thresholds of each neuron;

[0056] S143: Calculate the output of the CNN module based on the connection weights and the input vector;

[0057] S144: Calculate the output of the LSTM module based on the connection weights and the output of the CNN module;

[0058] S145: Calculate the error between the target output and the predicted output according to the mean square loss function;

[0059] S146: Use the backpropagation algorithm to calculate the output errors of each layer, and iteratively update the connection weights between layers until the convergence requirement is met, so that the network can realize channel mapping.

[0060] After the training is completed, the generated model is encrypted using the paired key and sent to the corresponding child nodes.

[0061] Step S2: Intra-group common key generation phase: All nodes perform channel estimation on the probe frames broadcast by the base station and preprocess the CSI data; the child nodes use the CSI data as the input of the trained group consensus mapping network model and obtain the predicted output of the model; the central node and the child nodes respectively use the vector quantization algorithm to quantize the collected CSI data and the predicted output of the model to obtain the original keys K c and K i ; the central node and the child nodes perform key negotiation to correct K c and K i to obtain the group key K in .

[0062] In the process of preprocessing CSI data, the following steps are included: Stack the real part and the imaginary part of the CSI matrix, and stretch the stacked matrix into a one-dimensional vector of size 64. Then, normalize the one-dimensional vector so that its data range is between -1 and 1.

[0063] S21: The base station broadcasts an IEEE 802.11n probe frame containing a packet index, and all nodes use the least squares channel estimation method to extract CSI data;

[0064] S22: All nodes preprocess the collected CSI data. Assume that the CSI data obtained by the central node and the child nodes are respectively and The child node takes as the input of the group consensus mapping network trained by the central node in the network training stage, and obtains the model prediction output

[0065] S23: The central node uses the preprocessed CSI data as the channel feature, and the child node i uses the model prediction output as the channel feature. The central node and the child node i respectively perform vector quantization on and to obtain the original keys K c and K i , K c and K i There are some inconsistent bit positions.

[0066] Vector quantization means that on the basis of the joint multi-dimensional channel features, the quantization region is extended to the multi-dimensional level, and the multi-dimensional channel features are transformed into a binary bit sequence. Referring to Figure 5 , the process of vector quantization is as follows:

[0067] S231: The central node randomly selects k initial cluster centers;

[0068] S232: The central node calculates the Euclidean distance between each sample point of the CSI and each initial cluster center;

[0069] S233: The central node divides each sample point into the cluster center with the closest distance according to the Euclidean distance between the sample point and the cluster center;

[0070] S234: The central node recalculates the new k cluster centers according to the sample allocation result. If the cluster centers change, repeat step S232 to start a new round of iteration;

[0071] S235: The central node encrypts the finally obtained k clustering centers and broadcasts them to all the sub-nodes within the group. The sub-node i clusters the model prediction outputs using the same method as in steps S232 and S233;

[0072] S236: The central node and the sub-node i quantify the samples according to the region indexes corresponding to the finally obtained sample points to obtain the original key sequences K c and K i ;

[0073] As Figure 6 shown, it is a schematic diagram of the two-dimensional vector quantization algorithm in this example when k = 4. S24: The central node and the sub-node i use the BCH error correction code to perform key negotiation to correct K c and K i to obtain the group key K in . In step S2, each sub-node takes the same operations as the sub-node i.

[0074] Step S3: Group-external public key generation phase: Each group's central node encrypts and sends the group-internal public key K in to the base station. The base station generates the group-external public key K out by combining all the received group-internal public keys, and broadcasts it to all the nodes after encryption.

[0075] S31: There are 3 groups in the current environment, and the number of users in each group is 4. The central nodes of each group use the pairwise keys to encrypt the group-internal public key and send it to the base station in sequence, where j represents the group number;

[0076] S32: The base station decrypts using the pairwise key to obtain the group-internal public keys of each group The group-external public key K out is calculated by the formula ;

[0077] S33: The base station encrypts the key K out using all the group-external public keys held by the users generated in the previous time period, and broadcasts it to 12 user nodes. The user nodes decrypt using the group-external public key in the previous time period to obtain the new group-external public key K out . After that, the group-external broadcast communication uses the new key K out .

[0078] The network training phase in step S1 of this method belongs to a one-time process. After the network training is completed, it is no longer necessary to train the network for generating group keys in the subsequent long period. Each time the group key is generated subsequently, only step S2 for generating the group-internal public key and step S3 for generating the group-external public key need to be performed, which effectively improves the wireless physical layer group key generation rate and saves the channel detection time.

[0079] Therefore, this method uses deep learning to assist in the generation of wireless physical layer group keys, reducing the communication and time complexity of channel detection in multi-user scenarios and effectively improving the generation rate of physical layer group keys. In addition, the present invention introduces a vector quantization algorithm as the quantization algorithm in the generation of physical layer group keys. By jointly quantifying multi-dimensional channel features, the channel utilization rate is improved, and to a certain extent, the generation rate of group keys can also be improved.

[0080] It should be noted that the above content only illustrates the technical idea of the present invention and cannot be used to limit the protection scope of the present invention. For those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements all fall within the protection scope of the claims of the present invention.

Claims

1. Wireless Physical Layer Group Key Generation Method Based on Group Consensus Mapping Network, characterized in that it includes the following steps: S1. Network training phase: Nodes perform channel estimation on the probe frames broadcast by the base station and preprocess the CSI data; The child nodes send the encrypted CSI data to the central node, and the central node generates a data set based on the CSI data of the child nodes and itself; The central node uses the data set to train a group consensus mapping network model for the child nodes and encrypts and sends the trained model to the child nodes; S2, Intra-group common key generation phase: All nodes perform channel estimation on the detection frames broadcast by the base station and preprocess the CSI data; the child nodes use the CSI data as the input of the trained group consensus mapping network model and obtain the predicted output of the model; the central node and the child nodes respectively use the vector quantization algorithm to quantize the collected CSI data and the predicted output of the model to obtain the original keys K c and K i ; the central node and the child nodes conduct key negotiation on K c and K i to correct and obtain the intra-group common key K in ; the child nodes use the CSI data as the input of the trained group consensus mapping network model. The trained group consensus mapping network model is sequentially distributed by the central node to the child nodes. Each child node uses the deployed group consensus mapping network model to implement the mapping of the central node's CSI data to achieve the goal of intra-group consensus and obtain the predicted output of the model; S3, Generation stage of the inter-group public key: Each group's central node encrypts and sends the intra-group public key K in to the base station. The base station generates the inter-group public key K out by combining all the received intra-group public keys, and broadcasts it to all nodes after encryption.

2. The wireless physical layer group key generation method based on group consensus mapping network according to claim 1, characterized in that: The specific steps of step S1 include: S11: The base station broadcasts a probe frame containing a packet index, and all nodes perform channel estimation to extract CSI data and preprocess it. When nodes extract CSI data from two adjacent packets, it is required that the time slot is greater than the coherence time. All nodes synchronize the CSI data according to the packet index; S12: When all users have collected enough CSI data, all child nodes encrypt the CSI data with the intra-group common key generated in the previous time period and send it to the central node of this group. The central node decrypts it to obtain the CSI data of all child nodes in this group; S13: The central node uses the CSI data it collected as labels and constructs a CSI data set with the child node CSI data as inputs. The input of the data set is the corresponding child node CSI data, and the labels are all the central node CSI data; S14: The central node trains a group consensus mapping network model for each child node according to the corresponding child node CSI data set. After training, the generated model is encrypted and sent to the corresponding child node using the pairwise key between the central node and the corresponding child node.

3. The wireless physical layer group key generation method based on group consensus mapping network according to claim 2, characterized in that: The group consensus mapping network consists of a CNN module and an LSTM module. Among them, the CNN module consists of two convolutional layers and two pooling layers; The LSTM module consists of two long short-term memory layers and a fully connected layer; The training process of the network model is as follows: Use the child node CSI data as the model input, use the central node CSI data as the target output, calculate the error between the target output and the predicted output using the mean square loss function, and use the backpropagation algorithm to iteratively update the network parameters until the convergence requirement is met.

4. The wireless physical layer group key generation method based on group consensus mapping network according to claim 3, characterized in that: The specific process of the vector quantization algorithm in step S2 includes: S21: The central node randomly selects k initial cluster centers; S22: The central node calculates the Euclidean distance between each sample point of the CSI and each initial cluster center; S23: The central node divides each sample point into the cluster center with the closest distance according to the Euclidean distance between the sample point and the cluster center; S24: The central node recalculates k new cluster centers according to the sample assignment result. If the cluster centers change, repeat step S22 to start a new round of iteration; S25: The central node encrypts the finally obtained k clustering centers and broadcasts them to all the sub-nodes within the group. The sub-nodes cluster the model prediction outputs using the same method as in steps S22 and S23. S26: The central node and the child nodes perform quantization on the samples according to the region indexes corresponding to the finally obtained sample points to obtain the original key sequence K c and K i .

5. The wireless physical layer group key generation method based on group consensus mapping network according to claim 4, characterized in that: The least - squares channel estimation method is used to implement channel estimation of the probe frame broadcast by the node to the base station, and the CSI data is calculated through the formula ; where Y is the long training symbol in the predefined preamble, and X 1 and X 2 respectively represent two long training symbols extracted from the received probe frame.

6. The wireless physical layer group key generation method based on group consensus mapping network according to claim 5, characterized in that: The preprocessing of the CSI data at least includes stacking the real part and the imaginary part of the CSI matrix, stretching the stacked matrix into a one-dimensional vector of size 64, and performing normalization processing on the one-dimensional vector so that its data range is between -1 and 1.

7. The wireless physical layer group key generation method based on group consensus mapping network according to claim 1, characterized in that: The specific steps of step S3 include: S31: There are multiple groups in the current environment. The central nodes of each group use the pairwise key with the base station to encrypt the group public key K in and send it to the base station in sequence; S32: The base station decrypts using the corresponding paired key to obtain the intra-group common key K of each group in , and the base station performs an exclusive OR operation on all the received intra-group common keys K in to generate an inter-group common key K out ; S33: The base station encrypts the new out-of-group public key K with all the out-of-group public keys held by users generated in the previous time period, out and broadcasts it to all user nodes. The user nodes decrypt it with the out-of-group public key of the previous time period to obtain the new out-of-group public key K, out and thereafter, the out-of-group broadcast communication uses the new key K. out .

8. A wireless physical layer group key generation system based on group consensus mapping network, including a computer program, characterized in that: When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1-7 above.

Images