Physical layer identity authentication system based on TimeGAN in wireless communication scene
By using the TimeGAN model to extract channel timing characteristics and mobile behavior characteristics in wireless communication scenarios, the problem of inaccurate identity authentication in the physical layer in wireless communication scenarios in the prior art is solved, and higher authentication accuracy and robustness are achieved. Suitable for high-speed mobile and dynamic environments.
Patent Information
- Application Number
- CN202411287422.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2024-12-24
AI Technical Summary
In wireless communication scenarios, the existing physical layer identity authentication method is inaccurate in the high-speed mobile state due to the dynamic changes in channel characteristics, and it is impossible to effectively identify the legitimate sender and malicious attacker.
Design a physical layer identity authentication system based on TimeGAN, including a channel fingerprint acquisition module, a timing feature learning module and an identity authentication module. The time series channel data is trained through the TimeGAN model, and the channel timing features and mobile behavior features are extracted, and these features are used. Conduct identity authentication to distinguish between legal sender and malicious attacker.
It improves the accuracy and robustness of physical layer identity authentication in wireless communication scenarios, can learn and adapt to changes in the communication environment in real time, and is suitable for high-speed mobile and dynamic communication environments.
Smart Images
Figure CN119183106A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of wireless communications, and in particular relates to a physical layer identity authentication system based on TimeGAN in a wireless communication scenario. Background Art
[0002] In wireless communication networks, channel characteristics can be used for physical layer authentication. Existing physical layer authentication methods are mainly based on static channel characteristics, such as received signal strength (RSS), channel state information (CSI), etc. However, in wireless communication scenarios, especially when the communication source is moving at high speed, static characteristics may cause authentication failure or misjudgment due to the change of channel characteristics over time and position. Therefore, an authentication method that can work effectively in a dynamic environment is needed to improve the security and robustness of the system. In recent years, generative adversarial networks (GANs) have been widely used in the modeling and generation of time series data, but their application in the field of wireless communications is not yet mature. Summary of the invention
[0003] 1. Technical issues to be resolved
[0004] The technical problem to be solved by the present invention is to design a physical layer identity authentication system to solve the problem of inaccurate physical layer identity authentication in wireless scenarios in the prior art.
[0005] (II) Technical solution
[0006] In order to solve the above technical problems, the present invention provides a physical layer identity authentication system based on TimeGAN in a wireless communication scenario, the system comprising: a channel fingerprint acquisition module, a temporal feature learning module and an identity authentication module;
[0007] The channel fingerprint collection module is responsible for collecting the channel fingerprint data of the legitimate sender Alice at different time points, including the received signal strength RSS and the channel state information CSI. The channel fingerprint data at multiple time points is called the time series channel data, and is also used to obtain the real-time channel fingerprint data of the unknown sender;
[0008] The time series feature learning module uses the TimeGAN model to train the time series channel data collected by the channel fingerprint acquisition module to extract Alice's channel time series features and mobile behavior features;
[0009] The identity authentication module is used to predict Alice's current channel fingerprint using the trained TimeGAN model during the authentication phase, and compare it with the channel fingerprint actually received from the unknown sender, and determine whether the unknown sender is the legitimate sender Alice or a malicious attacker based on the matching degree.
[0010] Preferably, the TimeGAN model consists of four modules:
[0011] 1). Encoder
[0012] Input: Original time series channel data X
[0013] Where X={x1,x2,...x T}, each x t represents the channel fingerprint data at time t, t=1,...,T;
[0014] Output: latent space vector H
[0015] The output of the encoder is a latent space vector H, which is used to capture the time-dependent characteristics of the input raw time series channel data, namely, channel time series characteristics and mobile behavior characteristics;
[0016] Working principle: The encoder uses a recurrent neural network to map the input raw time series channel data X to a latent space vector H;
[0017] 2).Decoder
[0018] Input: latent space vector H
[0019] The latent space vector H is the low-dimensional representation generated by the encoder;
[0020] Output: Reconstructed time series data X'
[0021] The decoder converts the latent space vector H back to time series channel data X' similar to the original time series channel data;
[0022] Working principle: The decoder reconstructs the latent space vector H into a time series channel data X' that is as similar as possible to the original input time series channel data X through another recurrent neural network structure;
[0023] 3) Generator
[0024] Input: Noise vector Z
[0025] The noise vector Z is a random vector sampled from a preset simple distribution and is used to generate latent space vectors;
[0026] Output: Generated latent space vector data H fake
[0027] The generator generates fake latent space vector data H based on the noise vector Z fake ;
[0028] Working principle: The generator uses the noise vector Z as input and generates latent space vector data H with similar distribution to the original time series channel data X through a deep neural network.fake ,The goal of the generator is to learn to generate fake data that can deceive the discriminator. It obtains the previous time series channel data and generates the channel fingerprint data at the current moment;
[0029] 4). Discriminator
[0030] Input: latent space vector H generated by the encoder and fake latent space vector data H generated by the generator fake
[0031] Output: True and false probabilities D(H) and D(H fake )
[0032] The output of the discriminator is a probability value indicating whether the input data is real or fake;
[0033] Working principle: The discriminator uses a deep neural network to classify the input latent space vector data. Its goal is to distinguish the real time series channel data H from the generated latent space vector data.
[0034] Preferably, the TimeGAN model has three loss functions. The temporal feature learning module uses the TimeGAN model to train the channel fingerprint data at multiple time points to minimize the sum of the three loss functions to meet the system requirements. The three loss functions are:
[0035] (1) Adversarial loss L GAN :
[0036] This loss function is optimized through adversarial learning between the generator and the discriminator; the generator tries to maximize the probability of the discriminator misclassifying, while the discriminator tries to minimize the probability of its misclassification;
[0037]
[0038] E in the loss function refers to expectation;
[0039] (2) Self-supervision loss L sup :
[0040] This loss function supervises the similarity of the time series data generated by the generator through the encoder's latent space vector representation. This loss ensures that the generated data retains the same time-dependent characteristics as the original data;
[0041]
[0042] (3) Reconstruction loss L rec :
[0043] The latent space vector representation is reconstructed back to the original input time series data through the decoder to ensure that the encoder and decoder can correctly capture and reconstruct the temporal dependency;
[0044]
[0045] Preferably, the TimeGAN model can not only learn the distribution characteristics of time series channel data, but also predict the channel fingerprint data at the current moment on the basis of maintaining time dependency and sequence characteristics.
[0046] Preferably, the encoder can extract features from the time series channel data and retain the temporal dependency as much as possible.
[0047] Preferably, the goal of the reconstruction process is to minimize the difference between the original time series channel data X and the reconstructed time series channel data X' to ensure that the latent space vector H maintains the time-dependent characteristics of the data.
[0048] Preferably, the discriminator is trained so that it can better distinguish between real and fake data, which in turn pushes the generator to produce more realistic data.
[0049] The present invention also provides a physical layer identity authentication method implemented by the system.
[0050] Preferably, the method comprises the following two stages:
[0051] Training phase:
[0052] Step 1: The channel fingerprint acquisition module collects the channel fingerprint data of the legitimate sender Alice at different time points and transmits it to the time series feature learning module;
[0053] Step 2: The time series feature learning module uses the TimeGAN model to train the channel fingerprint data at multiple time points to generate a generation model that can accurately predict the channel fingerprint at the current moment, that is, the trained TimeGAN model; the TimeGAN model consists of a generator, a discriminator, an encoder and a decoder, wherein the generator is used to generate a fitted channel fingerprint sequence, the discriminator is used to distinguish the channel fingerprint generated by the generator from the real channel fingerprint generated by the encoder, the encoder is used to extract the channel time series features and mobile behavior features of the input channel fingerprint, and the decoder is used to reconstruct the channel time series features and mobile behavior features into a channel fingerprint;
[0054] Certification stage:
[0055] Step 3: When the unknown sender attempts to communicate with the legitimate receiver Bob, Bob obtains the real-time channel fingerprint data of the unknown sender through the channel fingerprint collection module;
[0056] Step 4: The identity authentication module uses the trained TimeGAN model to predict the channel fingerprint of the legitimate sender Alice at the current moment;
[0057] Step 5: The identity authentication module compares the predicted channel fingerprint with the actually received channel fingerprint and calculates the matching degree; if the matching degree exceeds the preset threshold, the unknown sender is determined to be the legitimate Alice; otherwise, it is determined to be the malicious attacker Eve.
[0058] The invention also provides a wireless communication method implemented based on the system.
[0059] (III) Beneficial effects
[0060] The present invention provides a physical layer identity authentication system based on a generative adversarial network, which is used to identify legitimate senders and malicious attackers in wireless communication scenarios. The authentication process implemented by the system includes a training phase and an authentication phase. The present invention utilizes the temporal feature learning capability of TimeGAN to effectively improve the accuracy and robustness of physical layer identity authentication in wireless communication scenarios. Moreover, the system can learn and adapt to changes in the communication environment in real time, and is suitable for high-speed mobile scenarios and dynamic communication environments. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Figure 1 A schematic diagram of the authentication process implemented by the system of the present invention;
[0062] Figure 2 This is a schematic diagram of the TimeGAN model design in this invention. DETAILED DESCRIPTION
[0063] In order to make the purpose, content and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below in conjunction with the drawings and examples.
[0064] The present invention provides a physical layer identity authentication system based on a generative adversarial network (GAN), in particular a physical layer identity authentication system based on a time generative adversarial network (TimeGAN), which is suitable for wireless communication scenarios. The authentication process implemented by the system includes a training phase and an authentication phase. Figure 1 In the training phase, the legitimate receiver (Bob) learns the temporal characteristics of the channel fingerprint of the legitimate sender (Alice) to extract and learn its mobility behavior characteristics; in the authentication phase, Bob predicts Alice's channel fingerprint at the future moment and compares it with the channel fingerprint of the unknown sender to determine whether the identity of the unknown sender is Alice.
[0065] System architecture:
[0066] The system consists of three main components: channel fingerprint acquisition module, temporal feature learning module (TimeGAN) and identity authentication module.
[0067] The channel fingerprint acquisition module is responsible for collecting the channel fingerprint data (i.e., channel characteristics) of the legitimate sender (Alice) at different time points, including received signal strength (RSS), channel state information (CSI), etc. It is also used to obtain real-time channel fingerprint data of unknown senders;
[0068] The time series feature learning module uses the TimeGAN model to train the time series channel data collected by the channel fingerprint acquisition module to extract Alice's channel time series features and mobile behavior features (the channel time series features and mobile behavior features are implicit in the collected time series channel data);
[0069] The identity authentication module is used to use the trained TimeGAN model to predict Alice's current channel fingerprint during the authentication phase, and compare it with the actually received channel fingerprint (the channel fingerprint from the unknown sender). Based on the matching degree, it determines whether the unknown sender is the legitimate sender Alice or a malicious attacker.
[0070] The certification process of this system includes the following two stages:
[0071] Training phase:
[0072] 1. The channel fingerprint acquisition module collects the channel fingerprint data of the legitimate sender Alice at different time points and transmits it to the time series feature learning module.
[0073] 2. The temporal feature learning module uses the TimeGAN model to train the channel fingerprint data at multiple time points to generate a generative model that can accurately predict the channel fingerprint at the current moment, that is, the trained TimeGAN model. The TimeGAN model consists of a generator, a discriminator, an encoder, and a decoder. The generator is used to generate a fitted channel fingerprint sequence, the discriminator is used to distinguish the channel fingerprint generated by the generator from the real channel fingerprint generated by the encoder, the encoder is used to extract the channel timing features and mobile behavior features of the input channel fingerprint, and the decoder is used to reconstruct the channel timing features and mobile behavior features into channel fingerprints.
[0074] Certification stage:
[0075] 1. When an unknown sender attempts to communicate with the legitimate receiver Bob, Bob obtains the real-time channel fingerprint data of the unknown sender through the channel fingerprint collection module.
[0076] 2. The identity authentication module uses the trained TimeGAN model to predict the channel fingerprint of the legitimate sender Alice at the current moment.
[0077] The identity authentication module compares the predicted channel fingerprint with the actually received channel fingerprint and calculates the matching degree. If the matching degree exceeds the preset threshold, the unknown sender is determined to be the legitimate Alice; otherwise, it is determined to be the malicious attacker Eve.
[0078] TimeGAN model architecture design
[0079] refer to Figure 2 , the architecture of the TimeGAN model consists of four main modules:
[0080] 1. Encoder
[0081] 2. Decoder
[0082] 3. Generator
[0083] 4. Discriminator
[0084] The combination of these modules enables the TimeGAN model to not only learn the distribution characteristics of time series channel data, but also predict the channel fingerprint data at the current moment while maintaining time dependency and sequence characteristics.
[0085] 1. Encoder
[0086] Input: original time series channel data (ie time series data) X
[0087] Where X={x1,x2,...x T}, each x t Represents the channel fingerprint data at time t, t=1,...,T.
[0088] Output: latent space vector H
[0089] The output of the encoder is a low-dimensional latent space vector H, which is used to capture the time-dependent characteristics of the input original time series channel data, namely, channel time series characteristics and mobile behavior characteristics.
[0090] Working principle: The encoder uses a recurrent neural network (such as LSTM or GRU) to map the input raw time series channel data X to a latent space vector H. The role of the encoder is to extract features from complex time series channel data and retain the temporal dependencies as much as possible.
[0091] 2. Decoder
[0092] Input: latent space vector H
[0093] This is the low-dimensional representation generated by the encoder.
[0094] Output: Reconstructed time series data X'
[0095] The decoder converts the latent space vector H back into time series channel data X' which is similar to the original time series channel data.
[0096] Working principle: The decoder reconstructs the latent space vector H into time series channel data X' that is as similar as possible to the original input time series channel data X through another recurrent neural network structure (such as LSTM or GRU). The goal of the reconstruction process is to minimize the difference between the original time series channel data X and the reconstructed time series channel data X' to ensure that the latent space vector H maintains the time-dependent characteristics of the data.
[0097] 3. Generator
[0098] Input: Noise vector Z
[0099] The noise vector Z is a random vector sampled from a simple distribution (such as a normal distribution) and is used to generate the latent space vector.
[0100] Output: Generated latent space vector data H fake
[0101] The generator generates fake latent space vector data H based on the noise vector Z fake .
[0102] Working principle: The generator uses the noise vector Z as input and generates latent space vector data H with similar distribution to the original time series channel data X through a deep neural network. fake Its goal is to learn to generate fake data that can deceive the discriminator by obtaining the previous time series channel data to generate the channel fingerprint data at the current moment.
[0103] 4. Discriminator
[0104] Input: latent space vector H generated by the encoder and fake latent space vector data H generated by the generator fake
[0105] Output: True and false probabilities D(H) and D(H fake )
[0106] The output of the discriminator is a probability value indicating whether the input data is “real” (from the actual data distribution) or “fake” (from the generator).
[0107] Working principle: The discriminator uses a deep neural network to classify the input latent space vector data. Its goal is to distinguish between the real time series channel data H and the generated latent space vector data. The training of the discriminator enables it to better distinguish between real data and forged data, while in turn driving the generator to generate more realistic data.
[0108] Loss function of the TimeGAN model
[0109] The TimeGAN model has three loss functions. The TimeGAN model is used to train the channel fingerprint data at multiple time points through the time series feature learning module to minimize the sum of the three loss functions to meet the system requirements. The three loss functions are:
[0110] 1. Adversarial Loss L GAN :
[0111] This loss function is optimized through adversarial learning between the generator and the discriminator; the generator tries to maximize the probability that the discriminator misclassifies the object, while the discriminator tries to minimize the probability that it misclassifies the object.
[0112]
[0113] The E in the loss function refers to the expected (average) value.
[0114] 2. Self-supervised loss L sup :
[0115] This loss function supervises the similarity of the time series data generated by the generator through the latent representation of the encoder (latent space vector representation). This loss ensures that the generated data retains the same time-dependent characteristics as the original data.
[0116]
[0117] 3. Reconstruction loss L rec :
[0118] The latent representation is reconstructed back to the original input time series data through the decoder to ensure that the encoder and decoder can correctly capture and reconstruct the temporal dependencies.
[0119]
[0120] It can be seen that the present invention provides a physical layer identity authentication system based on a generative adversarial network (GAN), and the authentication process implemented by the system includes a training phase and an authentication phase. In the training phase, the legitimate receiving end (Bob) collects the channel fingerprint data of the legitimate sending end (Alice) at different time points, and uses the TimeGAN model to train these data to learn the timing characteristics and mobile behavior characteristics of Alice's channel fingerprint; in the authentication phase, when the unknown sending end attempts to communicate with Bob, Bob uses the trained TimeGAN model to predict Alice's channel fingerprint at a future time point, and compares it with the channel fingerprint actually received, so as to determine whether the identity of the unknown sending end is Alice; if the channel fingerprint match is high, it is determined to be Alice; otherwise, it is determined to be a malicious attacker Eve. The system can effectively improve the security of the physical layer, prevent the malicious attacker (Eve) from impersonating the legitimate sending end, and has high accuracy and reliability. Further, the present invention uses the time series feature learning ability of TimeGAN to effectively improve the accuracy and robustness of physical layer identity authentication in wireless communication scenarios. And the system can learn and adapt to changes in the communication environment in real time, and is suitable for high-speed mobile scenarios and dynamic communication environments.
[0121] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. A physical layer identity authentication system based on TimeGAN in a wireless communication scenario, characterized in that: The system includes: a channel fingerprint acquisition module, a time series feature learning module and an identity authentication module; The channel fingerprint collection module is responsible for collecting the channel fingerprint data of the legitimate sender Alice at different time points, including the received signal strength RSS and the channel state information CSI. The channel fingerprint data at multiple time points is called the time series channel data, and is also used to obtain the real-time channel fingerprint data of the unknown sender; The time series feature learning module uses the TimeGAN model to train the time series channel data collected by the channel fingerprint acquisition module to extract Alice's channel time series features and mobile behavior features; The identity authentication module is used to predict Alice's current channel fingerprint using the trained TimeGAN model during the authentication phase, and compare it with the channel fingerprint actually received from the unknown sender, and determine whether the unknown sender is the legitimate sender Alice or a malicious attacker based on the matching degree.
2. The system according to claim 1, characterized in that: The TimeGAN model consists of four modules: 1). Encoder Input: Original time series channel data X Where X = {x1, x2, ... x T }, each x t represents the channel fingerprint data at time t, t=1,...,T; Output: latent space vector H The output of the encoder is a latent space vector H, which is used to capture the time-dependent characteristics of the input raw time series channel data, namely, channel time series characteristics and mobile behavior characteristics; Working principle: The encoder uses a recurrent neural network to map the input raw time series channel data X to a latent space vector H; 2).Decoder Input: latent space vector H The latent space vector H is the low-dimensional representation generated by the encoder; Output: Reconstructed time series data X' The decoder converts the latent space vector H back to time series channel data X' similar to the original time series channel data; Working principle: The decoder reconstructs the latent space vector H into a time series channel data X' that is as similar as possible to the original input time series channel data X through another recurrent neural network structure; 3) Generator Input: Noise vector Z The noise vector Z is a random vector sampled from a preset simple distribution and is used to generate latent space vectors; Output: Generated latent space vector data H fake The generator generates fake latent space vector data H based on the noise vector Z fake ; Working principle: The generator uses the noise vector Z as input and generates latent space vector data H with similar distribution to the original time series channel data X through a deep neural network. fake ,The goal of the generator is to learn to generate fake data that can deceive the discriminator. It obtains the previous time series channel data and generates the channel fingerprint data at the current moment; 4). Discriminator Input: latent space vector H generated by the encoder and fake latent space vector data H generated by the generator fake Output: True and false probabilities D(H) and D(H fake ) The output of the discriminator is a probability value indicating whether the input data is real or fake; Working principle: The discriminator uses a deep neural network to classify the input latent space vector data. Its goal is to distinguish the real time series channel data H from the generated latent space vector data.
3. The system according to claim 2, characterized in that: The TimeGAN model has three loss functions. The temporal feature learning module uses the TimeGAN model to train the channel fingerprint data at multiple time points to minimize the sum of the three loss functions to meet the system requirements. The three loss functions are: (1) Adversarial loss L GAN : This loss function is optimized through adversarial learning between the generator and the discriminator; The generator tries to maximize the probability that the discriminator misclassifies the class, while the discriminator tries to minimize the probability that it misclassifies the class; E in the loss function refers to expectation; (2) Self-supervision loss L sup : This loss function supervises the similarity of the time series data generated by the generator through the encoder's latent space vector representation. This loss ensures that the generated data retains the same time-dependent characteristics as the original data; (3) Reconstruction loss L rec : The latent space vector representation is reconstructed back to the original input time series data through the decoder to ensure that the encoder and decoder can correctly capture and reconstruct the temporal dependency; 4. The system according to claim 2, characterized in that The TimeGAN model can not only learn the distribution characteristics of time series channel data, but also predict the channel fingerprint data at the current moment while maintaining time dependency and sequence characteristics.
5. The system according to claim 2, characterized in that The encoder can extract features from time-series channel data and preserve the temporal dependencies as much as possible.
6. The system according to claim 2, characterized in that The goal of the reconstruction process is to minimize the difference between the original time series channel data X and the reconstructed time series channel data X' to ensure that the latent space vector H maintains the time-dependent characteristics of the data.
7. The system according to claim 2, characterized in that The training of the discriminator enables it to better distinguish between real and fake data, which in turn pushes the generator to produce more realistic data.
8. A physical layer identity authentication method implemented by the system as claimed in any one of claims 1 to 7.
9. The method according to claim 8, characterized in that The method consists of the following two stages: Training phase: Step 1: The channel fingerprint acquisition module collects the channel fingerprint data of the legitimate sender Alice at different time points and transmits it to the time series feature learning module; Step 2: The time series feature learning module uses the TimeGAN model to train the channel fingerprint data at multiple time points to generate a generation model that can accurately predict the channel fingerprint at the current moment, that is, the trained TimeGAN model; the TimeGAN model consists of a generator, a discriminator, an encoder and a decoder, wherein the generator is used to generate a fitted channel fingerprint sequence, the discriminator is used to distinguish the channel fingerprint generated by the generator from the real channel fingerprint generated by the encoder, the encoder is used to extract the channel time series features and mobile behavior features of the input channel fingerprint, and the decoder is used to reconstruct the channel time series features and mobile behavior features into a channel fingerprint; Certification stage: Step 3: When the unknown sender attempts to communicate with the legitimate receiver Bob, Bob obtains the real-time channel fingerprint data of the unknown sender through the channel fingerprint collection module; Step 4: The identity authentication module uses the trained TimeGAN model to predict the channel fingerprint of the legitimate sender Alice at the current moment; Step 5: The identity authentication module compares the predicted channel fingerprint with the actually received channel fingerprint and calculates the matching degree; if the matching degree exceeds the preset threshold, the unknown sender is determined to be the legitimate Alice; otherwise, it is determined to be the malicious attacker Eve.
10. A wireless communication method implemented based on the system according to any one of claims 1 to 7.