Physical layer security strategy generation method and system for uncertain air-ground environment

Abstract

This invention discloses a physical layer security strategy generation method and system for uncertain air-to-ground environments, belonging to the field of communication signal processing technology. The method includes: constructing a downlink multi-user MISO system model incorporating a UAV base station, legitimate users, and eavesdroppers, and using a bounded channel uncertainty model to describe channel errors; constructing an optimization model with the objective of maximizing security and confidentiality; constructing a generator network based on a conditional denoising diffusion probability model, including a channel encoder and a denoising U-Net network; performing two-stage training on the generator network: the first stage learns the conditional distribution of the strategy by minimizing noise prediction error; the second stage introduces a security and confidentiality loss function and a power constraint penalty term for targeted fine-tuning; in actual communication, the fine-tuned model is used to quickly generate beamforming and artificial noise strategies that satisfy power constraints from random noise. This invention significantly improves the physical layer security performance and response speed of air-to-ground communication systems.

Description

Technical Field

[0001] This invention belongs to the field of communication signal processing technology, specifically relating to a method and system for generating physical layer security policies for uncertain air-ground environments. Background Technology

[0002] With the rapid growth in the scale and data volume of sixth-generation (6G) wireless network services, the open wireless channels have significantly increased the risks of eavesdropping and illegal surveillance. Protecting confidential information in dense, dynamic, and heterogeneous network environments has become a critical issue. Physical layer security has attracted widespread attention due to its advantages such as low implementation complexity and secure communication. Among these methods, multi-antenna beamforming combined with artificial noise injection, utilizing spatial degrees of freedom to enhance legitimate links and weaken eavesdropping links, is an important means of achieving secure transmission.

[0003] However, when physical layer security technologies are extended to multi-user scenarios, the transmitter needs to simultaneously address factors such as inter-user interference, time-varying channel conditions, and complex scenario differences, significantly increasing the difficulty of transmission strategy design. Traditional solutions often achieve joint optimization of beamforming and artificial noise through approximation and iterative solutions, but such methods often have high computational costs, are difficult to deploy online inference, and may only obtain suboptimal solutions under complex non-convex coupling conditions, making it difficult to meet the requirements of low latency and high robustness for real-time secure communication.

[0004] To overcome the real-time bottleneck of iterative optimization, the industry has proposed data-driven learning methods for PLS policy design. Examples include using multilayer perceptrons for real-time interference management and security enhancement, or graph neural networks for learning multi-user precoding and secure beamforming. These methods typically belong to a discriminative modeling paradigm, tending to learn decision boundaries directly from observed channel characteristics. They are highly sensitive to noise, disturbances, or changes in channel statistics / network topology, and their generalization and robustness to unseen scenarios may decrease, thus affecting the stability and reliability of secure communication.

[0005] Meanwhile, generative artificial intelligence techniques such as diffusion models can provide new insights for wireless communication strategy design by learning complex high-dimensional data distributions. Compared to the training instability, pattern collapse, or reconstruction ambiguity that some generative adversarial networks or variational autoencoders may face, diffusion models achieve high-fidelity sampling generation through progressive noise addition and iterative denoising, and can learn the "distribution of solutions" rather than a single-point mapping under conditional constraints. Existing research shows that conditional diffusion models have strong adaptability in channel uncertainty and interference environments, and can be used for tasks such as channel denoising, device-awareness, and angle-of-arrival estimation, demonstrating robust modeling capabilities for complex scenarios.

[0006] For security-oriented transmission strategies, the diffusion model recovers structured policy vectors from random disturbances through iterative denoising, which meets the need to reconstruct near-optimal policies under uncertain and interference-constrained conditions. Moreover, compared with learning methods that only output a single solution, generative models can generate multiple candidate high-performance policies under the same channel conditions, providing space for subsequent selection, optimization and robust decision-making, thereby improving the system's ability to resist eavesdropping attacks and sudden changes in the scene.

[0007] Based on the above analysis, existing technologies for the "joint beamforming and artificial noise injection" problem in secure communication still generally suffer from the following shortcomings: First, traditional iterative optimization struggles to balance near-optimal performance with real-time inference; second, discriminative deep models have limited adaptability to unseen channel distributions, noise disturbances, and topology changes; third, security objectives (such as total privacy rate) are often tightly coupled with power / structure constraints, making it difficult for learned strategies to maintain high confidentiality gains while ensuring constraint feasibility. Therefore, a method is needed that can directly characterize the distribution of secure transmission strategies under channel conditions and rapidly generate feasible joint design schemes when new channels arrive, in order to achieve efficient, robust, and real-time deployable physical layer secure transmission in dynamic environments with multi-user interference and multiple eavesdroppers. Summary of the Invention

[0008] The technical problem to be solved by this invention is to address the shortcomings of the prior art by providing a physical layer security policy generation method and system for uncertain air-to-ground environments. By learning the conditional distribution of near-optimal policies through generative artificial intelligence, it achieves high confidentiality, strong generalization ability, and beamforming and artificial noise joint design that is adaptable to real-time communication. This addresses the technical problems of existing technologies in air-to-ground communication where the channel state information (CSI) is uncertain, the location of eavesdroppers is unknown, and the environment is dynamically changing. Traditional iterative optimization algorithms have extremely high computational complexity, which makes them unable to meet real-time requirements. Furthermore, policies designed based on perfect channel assumptions have poor robustness and severely reduced confidentiality under actual errors.

[0009] The present invention adopts the following technical solution: A method for generating physical layer security policies for non-deterministic air-to-ground environments includes the following steps: S1. Construct a downlink multi-user multiple-input single-output system model that includes a drone base station, multiple legitimate users, and multiple eavesdroppers; S2. Obtain the received signal expressions of legitimate users and eavesdroppers based on the system model, and calculate the privacy rate of the communication system based on the instantaneous signal-to-noise ratio of legitimate users and eavesdroppers. S3. Construct an optimization model with beamforming and artificial noise vector as optimization variables and maximizing the privacy rate as the objective function; S4. Construct a generator network based on a conditional denoising diffusion probability model, wherein the generator network includes a channel encoder for processing channel state information and a denoising U-Net network for performing the denoising process. S5. The conditional diffusion model is trained in the first stage based on the generator network. By minimizing the mean square error of noise prediction, the model learns to recover the conditional distribution of the joint beamforming and artificial noise strategy vectors from Gaussian noise. S6. Based on the pre-trained model trained in the first stage, a loss function with the goal of maximizing the confidentiality rate is introduced and a power constraint penalty term is added to perform targeted fine-tuning of the model in the second stage. S7. In actual communication, based on the channel state information acquired in real time, the model, which has been fine-tuned in the second stage, starts from random Gaussian noise and generates a beamforming and artificial noise transmission strategy that meets power constraints and has high security rate through a denoising process.

[0010] Preferably, in step S1, constructing the downlink multi-user multiple-input single-output system model specifically includes: Configure drone base station equipment Root antenna, service A single-antenna legitimate user, simultaneously existing A single-antenna eavesdropper; The base station sends the first The symbol for each user is The beamforming vector used is The injected artificial noise precoding matrix is The artificial noise vector is Construct the synthetic transmission signal vector of the base station. , For users, set the system to meet the total transmit power requirement. The constraints.

[0011] Preferably, in step S2, obtaining the received signal expressions of the legitimate user and the eavesdropper specifically includes: A bounded channel uncertainty model is adopted, in which the channel state information acquired by the base station is represented as the sum of the estimated channel and the channel error; Among them, drone base station to user Channel vector Represented as From drone base stations to eavesdroppers Channel vector Represented as ; in, and These represent the corresponding estimated channels. and These represent the corresponding channel estimation error vectors, and based on these, an expression for the received signal containing the channel error term is constructed.

[0012] Preferably, the step of calculating the privacy rate of the communication system based on the instantaneous signal-to-noise ratio between the legitimate user and the eavesdropper specifically includes: Calculate legitimate users achievable rate Its expression is determined based on the signal-to-noise ratio of the received signal; calculating the eavesdropper About users Information eavesdropping rate Its expression is determined based on the signal-to-noise ratio of the eavesdropping channel; the user's expression is calculated. privacy rate The privacy rate is determined by the worst-case eavesdropping channel, and the calculation formula is as follows:

[0013]

[0014] in, This represents the total number of eavesdroppers.

[0015] Preferably, in step S4, constructing the generator network based on the conditional denoising diffusion probability model specifically includes: The composite channel vector is defined as a concatenation of channels from the base station to all users and eavesdroppers, and the composite policy vector is defined as a concatenation of all beamforming vectors and artificial noise vectors. The composite policy vector in the complex domain is converted into a real-domain representation, and its real and imaginary parts are separated as data inputs to the diffusion model. The channel encoder uses a multilayer perceptron to map the complex channel state information into a fixed-dimensional real-valued embedding vector, which is used as conditional information input to the diffusion model. The denoising U-Net network adopts an encoder-decoder architecture, which includes multiple downsampling blocks and upsampling blocks, and connects the corresponding layers through skip connections.

[0016] Preferably, the denoising U-Net network further includes a cross-attention module, specifically configured as follows: Cross-attention modules are set at the bottleneck layer and key positions of the downsampling and upsampling blocks in the denoising U-Net network. In the cross-attention modules, the features of the current network layer are used as queries, and the channel embedding vector output by the channel encoder is provided as the key and value, so as to guide the denoising process by the channel conditions.

[0017] Preferably, in step S5, the first-stage training of the conditional diffusion model specifically includes: Perform a forward diffusion process, gradually adding Gaussian noise to the data via a Markov chain at time step [missing information]. Noisy samples Represented as:

[0018] in, Standard Gaussian noise, , These are the preset variance scheduling parameters; Perform the reverse denoising process by training the neural network. Predict the added noise; define the loss function in the pre-training phase as the mean square error between the predicted noise and the actual noise, and calculate the loss function at time step [number missing]. Optimization is achieved by uniform sampling within the specified interval.

[0019] Preferably, in step S6, the second-stage targeted fine-tuning of the model specifically includes: An unsupervised loss function that directly targets and maximizes the secrecy rate is adopted. The loss function includes a secrecy rate term and a power constraint penalty term. The power constraint penalty term is used to force the generated policy vector to meet the power limit. It is weighted by balancing the hyperparameters of the secrecy rate target and the power constraint. The model parameters are updated so that the generated distribution is concentrated in the solution space region with high secrecy rate, thereby improving the performance of the model under the specific optimization objective.

[0020] Preferably, in step S7, the generation of a beamforming and artificial noise transmission strategy that satisfies power constraints and has high security through the denoising process specifically includes: A noise reduction diffusion implicit model sampling acceleration method is adopted to sample the initial noise vector from a standard Gaussian distribution. Using a trained and finely tuned noise prediction network, iterative noise reduction is performed according to the reverse time step to estimate the sample of the previous time step, where the randomness of the sampling process is controlled. After the iteration is completed, the final denoised sample is obtained, and the real and imaginary parts are recombined by inverse transformation to obtain the beamforming and artificial noise strategy vector in the complex domain, which is directly applied to the signal transmission of UAV base stations.

[0021] Secondly, embodiments of the present invention provide a physical layer security policy generation system for uncertain air-to-ground environments, comprising: The model building module is used to build a downlink multi-user multiple-input single-output system model that includes a drone base station, multiple legitimate users, and multiple eavesdroppers; The rate calculation module is used to obtain the received signal expressions of legitimate users and eavesdroppers based on the system model, and to calculate the private rate of the communication system based on the instantaneous signal-to-noise ratio of legitimate users and eavesdroppers. An optimization construction module is used to construct an optimization model with beamforming and artificial noise vectors as optimization variables and maximizing the privacy rate as the objective function; A network construction module is used to construct a generative network based on a conditional denoising diffusion probability model. The generative network includes a channel encoder for processing channel state information and a denoising U-Net network for performing the denoising process. The first training module is used to perform a first-stage training of the conditional diffusion model based on the generator network. By minimizing the mean square error of noise prediction, the model learns to recover the conditional distribution of the joint beamforming and artificial noise policy vectors from Gaussian noise. The second fine-tuning module is used to perform targeted fine-tuning of the model in the second stage based on the pre-trained model trained in the first stage, by introducing a loss function with the goal of maximizing the confidentiality rate and adding a power constraint penalty term. The strategy generation module is used to generate beamforming and artificial noise transmission strategies that meet power constraints and have high security rates by using a model that has been fine-tuned in the second stage, starting from random Gaussian noise and generating them through a denoising process, based on the channel state information acquired in real time during actual communication.

[0022] Thirdly, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the above-described method for generating physical layer security policies for nondeterministic air-ground environments.

[0023] Fourthly, embodiments of the present invention provide a computer-readable storage medium including a computer program, which, when executed by a processor, implements the steps of the above-described method for generating physical layer security policies for nondeterministic air-ground environments.

[0024] Fifthly, a chip includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the above-described method for generating physical layer security policies for nondeterministic air-ground environments.

[0025] In a sixth aspect, embodiments of the present invention provide an electronic device including a computer program, which, when executed by the electronic device, implements the steps of the above-described method for generating physical layer security policies for uncertain air-ground environments.

[0026] Compared with the prior art, the present invention has at least the following beneficial effects: A physical layer security policy generation method for nondeterministic air-to-ground environments is proposed. First, a conditional denoising diffusion probability model is used to learn the mapping distribution from noise to optimal beamforming and artificial noise policies in an unsupervised manner. Then, a physical layer security objective function and power constraints from the communications domain are introduced for supervised fine-tuning. This method combines the powerful distribution fitting capability of deep generative models with the specific objectives of communication optimization, avoiding the need for traditional iterative optimization algorithms to resolve complex non-convex problems with each channel change. By directly outputting the policy through the generative network, the inference speed is extremely fast, significantly reducing computational latency. Simultaneously, the two-stage training ensures that the model possesses both general channel feature extraction capabilities and specifically optimizes security performance, resolving the contradiction between real-time performance and optimality, making it particularly suitable for dynamically changing air-to-ground communication scenarios.

[0027] Furthermore, leveraging the multi-antenna advantage of drone base stations, beamforming focuses signal energy towards legitimate users while simultaneously using redundant degrees of freedom to generate artificial noise to interfere with eavesdroppers. Mathematically, a synthetic transmission signal vector incorporating useful signals, multi-user interference, artificial noise, and thermal noise is constructed, with a total power constraint set. This forms the foundational architecture for physical layer security, providing a clear variable space for subsequent optimization. By explicitly modeling the joint transmission signal of artificial noise and beamforming, this scheme maximizes resource utilization on power-constrained drone platforms. It not only protects the communication privacy of specific users but also actively suppresses the received signal-to-noise ratio of potential eavesdroppers through artificially created interference fields. This structured model definition ensures that the generated strategy meets actual hardware transmission requirements, avoiding the problem of theoretical algorithms being disconnected from actual equipment.

[0028] Furthermore, to address the Doppler shift, obstruction, and estimation errors caused by UAV movement in air-to-ground environments, an error model is incorporated to force the generated beamforming and artificial noise strategies to have a certain degree of redundancy or safety margin. Even in the worst-case scenario where channel estimation is flawed, a certain level of secure communication capability can still be maintained. This is crucial for UAV air-to-ground communication under rapidly changing channel conditions, effectively solving the problem of a large gap between theoretical performance and actual implementation results.

[0029] Furthermore, by embodying the Max-Min principle, the optimization target is locked onto the worst-case eavesdropping scenarios, ensuring that legitimate users' communications are secure regardless of which eavesdropper is in the optimal receiving position. In drone scenarios, the eavesdropper's location is unknown and may move. This design ensures that the system provides a deterministic security lower bound guarantee at any time and against any distributed group of eavesdroppers. This results in a generated strategy with an extremely high level of defense, meeting the stringent security requirements of military or highly classified communications.

[0030] Furthermore, by leveraging key features extracted by the channel encoder, the generative network can adaptively adjust its output strategy for different channel environments. The separation of real and imaginary parts avoids the complexity of complex number operations in neural networks, improving training stability and convergence speed. This overall architecture enables the model to efficiently process large-scale channel data generated by high-dimensional antenna arrays, providing a feasible technical path for applying this technology in large-scale MIMO systems.

[0031] Furthermore, the introduction of a cross-attention mechanism significantly improves the accuracy and adaptability of the generated strategy. The model no longer mechanically applies a template but can dynamically adjust the beam direction and the distribution of artificial noise based on subtle changes in the real-time channel state. This is particularly crucial in uncertain spatial environments, where channel errors can cause drastic shifts in the optimal solution. The attention mechanism helps the model quickly locate the correct solution space region, avoiding the generation of invalid or inefficient strategies, thereby further improving the system's security and confidentiality performance.

[0032] Furthermore, the first-stage pre-training provides excellent initialization parameters for subsequent fine-tuning, preventing the model from getting trapped in local optima. It significantly reduces the reliance on labeled data, as obtaining large amounts of perfect labeled data is extremely costly in communication optimization. Through unsupervised learning, the model has acquired the ability to generate reasonable policies, which not only accelerates the convergence speed of the second stage but also enhances the model's generalization ability, enabling it to generate feasible solutions that conform to basic physical laws even when facing unseen channel scenarios.

[0033] Furthermore, the second-stage fine-tuning transforms the general generative model into a dedicated security policy generator. The advantage is that the generated policy truly approximates the optimal solution mathematically, directly maximizing the system's security and confidentiality rates. Simultaneously, explicit power constraint penalties eliminate the need for post-processing, ensuring that the output policy can be directly used in the transmitter. This targeted fine-tuning enables the invention to approach or even surpass the performance of traditional iterative optimization algorithms while maintaining extremely low inference latency.

[0034] Furthermore, sampling acceleration technology addresses the latency issues in real-time communication applications of generative models, compressing the generation process, which might otherwise take seconds, to milliseconds, fully meeting the real-time requirements of high-speed drone movement. Combined with policy reconstruction, executable launch parameters can be directly output, achieving end-to-end automation from channel awareness to policy generation. This makes AI-based physical layer security strategies truly feasible for engineering implementation, capable of responding to sudden security threats.

[0035] It is understood that the beneficial effects of the second to sixth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0036] In summary, the method of this invention applies the channel-aware conditional diffusion model to secure communication scenarios where eavesdropping is possible. By learning the conditional distribution of near-optimal strategies through generative artificial intelligence, it achieves a beamforming and artificial noise co-design with high confidentiality, strong generalization ability, and adaptability to real-time communication.

[0037] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0038] Figure 1 A model diagram of the wireless transmission system constructed for this invention; Figure 2 This is a flowchart of the method of the present invention; Figure 3 This is a schematic diagram illustrating the relationship between the privacy and speed of the system of this invention and the number of legitimate users; Figure 4 This is a schematic diagram illustrating the relationship between the privacy and transmission rate of the system of the present invention as a function of transmission power; Figure 5 A schematic diagram of a computer device provided in an embodiment of the present invention; Figure 6 This is a block diagram of a chip provided according to an embodiment of the present invention.

[0039] Among them, 60. Computer equipment; 61. Processor; 62. Memory; 63. Computer program; 600. Electronic device; 610. Processing unit; 620. Storage unit; 6201. Random access memory unit; 6202. Cache memory unit; 6203. Read-only memory unit; 6204. Program / utility; 6205. Program module; 630. Bus; 640. Display unit; 650. Input / output interface; 660. Network adapter; 700. External device. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0042] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0043] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this invention generally indicates that the preceding and following objects have an "or" relationship.

[0044] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0045] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0046] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0047] This invention provides a physical layer security policy generation method for nondeterministic air-to-ground environments. First, a generator network based on a conditional denoising diffusion probability model is constructed. A denoising U-Net network and a channel encoder are used to solve complex non-convex optimization problems from a data-driven perspective. Through a two-stage training strategy of pre-training and fine-tuning, the model first learns the conditional distribution of the near-optimal policy, and then optimizes it for the security target, thereby generating a high-quality beamforming and artificial noise joint policy. This avoids the high computational complexity and difficulty in real-time processing of traditional iterative optimization algorithms, and also overcomes the poor generalization ability and insufficient robustness of traditional deep learning discriminative models. This invention can quickly generate transmission policies that meet power constraints and have high security, significantly improving the physical layer security performance of multi-user multiple-input single-output systems in dynamic environments.

[0048] Please see Figure 2 This invention provides a method for generating physical layer security policies for uncertain air-to-ground environments, comprising the following steps: S1. Construct a downlink multi-user multiple-input single-output system model that includes a drone base station, multiple legitimate users, and multiple eavesdroppers; Please see Figure 1 Considering a multi-user, multi-input, single-output communication system with multiple eavesdroppers, its signal transmission process is defined as follows: The UAV base station is equipped with... Root antenna, service A single-antenna legitimate user, simultaneously existing A single-antenna eavesdropper, a drone base station, sends signals to users.

[0049] S2. Based on the wireless transmission system model constructed in step S1, obtain the expressions for the signals received by legitimate users and eavesdroppers, and calculate the privacy rate of the communication system based on the instantaneous signal-to-noise ratio of legitimate users and eavesdroppers. The base station sends the first The symbol for each user is The beamforming vector used is To confuse eavesdroppers, the base station also injects artificial noise, and its precoding matrix is ​​as follows: The artificial noise vector is The composite transmission signal vector of the base station is represented as follows: ,in, For a user aggregation, the system must meet the total transmit power requirement. The constraints are expressed as:

[0050] in, It is a matrix The Column vector.

[0051] Considering the channel estimation error in real-world communication environments, the system adopts a bounded channel uncertainty model. The channel state information acquired by the base station consists of the estimated channel and the channel error, specifically expressed as follows:

[0052]

[0053] in, It is a drone base station to user The channel vector, From drone base stations to eavesdroppers The channel vector, and These represent the corresponding channel estimation error vectors.

[0054] The reachable rate of a legitimate user is represented as

[0055] in, Thermal noise power, Indicates conjugate transpose; eavesdropper About users The rate at which information is eavesdropped is expressed as:

[0056] By clearly defining the signal model, beamforming vector, and artificial noise precoding matrix construction methods of a multi-user, multi-input, single-output system, and explicitly providing total transmit power constraints, subsequent policy generation and optimization can be carried out within a physically realizable set of constraints, avoiding problems such as power overruns or inconsistent physical meanings in the generated policies. At the same time, the above modeling describes the mechanism of improving the achievable rate of legitimate links and suppressing the rate of eavesdropping links with a unified vector signal expression, providing clear and computable targets and constraint boundaries for the diffusion model to learn the joint policy distribution, thereby improving the feasibility and deployability of policy generation.

[0057] S3. Construct an optimization model with beamforming and artificial noise vector as optimization variables and maximizing the privacy rate obtained in step S2 as the objective function. user privacy rate The calculation formula is determined by the worst-case eavesdropping channel:

[0058] The optimization model with maximizing privacy rate as the objective function is as follows:

[0059]

[0060] in, It is the beamforming matrix at the base station, and the constraint ensures that the total power of the normalized beamforming and artificial noise vector remains within the unit budget.

[0061] S4. Construct a generative network based on a conditional denoising diffusion probability model, which includes a channel encoder for processing channel state information and a denoising U-Net network for performing the denoising process. The channel encoder and data preprocessing include the following: defining the composite channel vector as a concatenation of channels from the base station to all users and eavesdroppers:

[0062] Define the composite strategy vector as a concatenation of all beamforming vectors and artificial noise vectors:

[0063] To adapt to neural network inputs, the policy vector in the complex field is converted into a representation in the real field:

[0064] in, and Representing the real and imaginary parts respectively; the channel encoder uses a multilayer perceptron to map complex channel state information into a fixed dimension. The real-valued embedding vector is processed as follows: The embedding vector The denoising U-Net network is input as conditional information into the diffusion model. The network incorporates a cross-attention mechanism to fuse channel state information. Specific structural features include: an encoder-decoder architecture containing multiple downsampling and upsampling blocks connected to corresponding layers via skip connections; cross-attention modules are introduced at the bottleneck layer of the U-Net and at key locations in the downsampling and upsampling blocks; in the cross-attention module, the features of the current network layer serve as the query, and the channel embedding vector provides the key and value.

[0065] By constructing a concatenated representation of composite channel vectors and composite policy vectors, and separating the real and imaginary parts of the complex-domain policy vectors to transform them into real-domain inputs, while employing a multilayer perceptron to map complex channel information into fixed-dimensional real-valued embedding vectors as conditional information, the model's interface consistency and scalability across different network sizes and antenna / user configurations can be significantly improved. This processing presents complex phase and amplitude information to the generator network with a unified numerical scale, reducing the training difficulty caused by numerical instability and dimensional coupling, and enabling the diffusion model to learn policy distributions conditionally using channel embeddings, thereby enhancing its generalization ability to unseen channel statistics or noisy environments.

[0066] The denoising backbone is set as a U-Net structure with a cross-attention mechanism, and cross-attention modules are introduced at the bottleneck layer and key upsampling and downsampling positions. This allows the network to dynamically adjust the energy distribution of beamforming and artificial noise according to the channel embedding at different scales of denoising iteration, so as to achieve "directional obfuscation" of eavesdropping links and "directional enhancement" of legitimate links. This multi-scale convolutional feature extraction and attention conditional fusion can simultaneously capture the spatial correlation in the antenna dimension and the local high-frequency details and global low-frequency structure in the policy vector. In principle, it makes up for the problem that simple MLP cannot take into account both local and global features. Therefore, it forms a higher quality joint policy generation capability and brings better security gain.

[0067] S5. Based on the network obtained in step S4, the conditional diffusion model is trained in the first stage. By minimizing the mean square error of noise prediction, the model learns to recover the conditional distribution of the joint beamforming and artificial noise strategy vectors from Gaussian noise. The model pre-training is based on a forward diffusion process and a reverse denoising process. The forward diffusion process gradually adds Gaussian noise to the data through a Markov chain at time steps. Noisy samples This is represented as:

[0068] in, Standard Gaussian noise, , The pre-defined variance scheduling parameters are used; the inverse denoising process is achieved by training a neural network. To predict the added noise, the loss function during the pre-training phase is defined as the mean square error between the predicted noise and the actual noise:

[0069] in, exist Uniform sampling within the interval.

[0070] A pre-training mechanism employing forward diffusion to gradually add noise and reverse denoising to predict noise, with the mean square error between predicted and actual noise as the training objective, enables the model to approach the near-optimal policy distribution through a stable learning process, avoiding problems such as instability and pattern collapse in traditional generative adversarial training. This noise prediction training also encourages the model to continuously learn high-frequency details and residual errors in the data distribution in subsequent stages, thereby improving the fine-grained reconstruction capability and overall generation quality of policy samples, providing a good initialization for subsequent fine-tuning aimed at maintaining confidentiality.

[0071] S6. Based on the pre-trained model obtained in step S5, introduce a loss function with the goal of maximizing the confidentiality rate, perform targeted fine-tuning of the model in the second stage, and add a power constraint penalty term. The fine-tuning step employs an unsupervised loss function that directly targets and maximizes confidentiality, specifically in the form of:

[0072] in, It is a prediction policy vector generated by the diffusion model. The first term is the calculated security rate; the second term is the power constraint penalty term, used to force the generated policy vector to meet the power limit. The normalized target norm corresponds to unit power. To balance the security target with the power constraint, the hyperparameters are fine-tuned to update the model parameters so that the generated distribution is concentrated in the solution space region with high security.

[0073] By employing a direct-oriented and security-maximizing unsupervised fine-tuning loss and incorporating a power constraint penalty term, the security objective can be explicitly injected into the generation process without relying on additional labels, aligning the model parameter update direction with the improvement of system security performance. The power penalty term is used to concentrate the generation strategy within the feasible power set, avoiding unfeasible power out-of-bounds solutions in the pursuit of high security. Furthermore, since the neighborhood close to the optimal solution has been explored during the pre-training phase, the fine-tuning process can further shrink the generation distribution and concentrate it in the solution space region with higher security, thereby obtaining continuous and interpretable security gains under complex conditions such as multiple eavesdroppers and multiple users.

[0074] S7. Based on the model obtained in step S6, in the actual communication process, according to the channel state information obtained in real time, the trained model is used to generate a beamforming and artificial noise transmission strategy that meets power constraints and has high security rate by starting from random Gaussian noise and through the denoising process.

[0075] The online inference and policy generation steps employ a denoising diffusion implicit model sampling acceleration method. Specifically, the process involves sampling an initial noise vector from a standard Gaussian distribution. Using a trained and finely tuned noise prediction network, denoising is iterated in reverse time steps, and the samples from the previous time step are estimated using the following formula:

[0076] in, Control the randomness of the sampling process; after completing the iteration, obtain the final denoised sample. Furthermore, by recombining the real and imaginary parts through inverse transformation, the beamforming and artificial noise strategy vectors in the complex domain are obtained:

[0077] The generated policy vector is the optimal security policy for actual transmission.

[0078] A denoising diffusion implicit model is employed for sampling acceleration. By reducing the number of reverse denoising time steps, high-quality policy generation can be achieved, significantly reducing online inference latency and meeting the timeliness requirements of real-time physical layer processing links. Simultaneously, the inference complexity of sampling increases linearly with the number of sampling steps, facilitating deployment on different computing platforms and allowing for latency / performance trade-offs. Combined with the noise-to-policy generation characteristics of the diffusion model, a joint beamforming and artificial noise policy that meets power constraints can be quickly output during the online phase, enabling low-latency secure transmission in dynamic wireless environments.

[0079] For scenarios with multiple eavesdroppers, the user confidentiality rate is defined using the worst-case eavesdropper criterion, and the generation of the dataset is based on maximizing the confidentiality rate. This makes the generation strategy protective against the most unfavorable eavesdropping conditions, maintaining robust security gains even when the number of eavesdroppers increases or the eavesdropping channel becomes stronger. At the same time, LogSumExp is introduced to smooth the non-differentiable maximum operator, which can transform the originally non-smooth confidentiality rate objective into a differentiable form, facilitating gradient optimization and neural network training, reducing training oscillations and improving convergence stability. This makes it easier for the model to learn effective interference injection and beam suppression rules in the high-dimensional joint policy space.

[0080] In another embodiment of the present invention, a physical layer security policy generation system for nondeterministic air-ground environments is provided. This system can be used to implement the above-mentioned physical layer security policy generation method for nondeterministic air-ground environments. Specifically, the physical layer security policy generation system for nondeterministic air-ground environments includes a model building module, a rate calculation module, an optimization construction module, a network building module, a first training module, a second fine-tuning module, and a policy generation module.

[0081] Among them, the model building module is used to build a downlink multi-user multiple-input single-output system model that includes a drone base station, multiple legitimate users, and multiple eavesdroppers; The rate calculation module is used to obtain the received signal expressions of legitimate users and eavesdroppers based on the system model, and to calculate the private rate of the communication system based on the instantaneous signal-to-noise ratio of legitimate users and eavesdroppers. An optimization construction module is used to construct an optimization model with beamforming and artificial noise vectors as optimization variables and maximizing the privacy rate as the objective function; A network construction module is used to construct a generative network based on a conditional denoising diffusion probability model. The generative network includes a channel encoder for processing channel state information and a denoising U-Net network for performing the denoising process. The first training module is used to perform a first-stage training of the conditional diffusion model based on the generator network. By minimizing the mean square error of noise prediction, the model learns to recover the conditional distribution of the joint beamforming and artificial noise policy vectors from Gaussian noise. The second fine-tuning module is used to perform targeted fine-tuning of the model in the second stage based on the pre-trained model trained in the first stage, by introducing a loss function with the goal of maximizing the confidentiality rate and adding a power constraint penalty term. The strategy generation module is used to generate beamforming and artificial noise transmission strategies that meet power constraints and have high security rates by using a model that has been fine-tuned in the second stage, starting from random Gaussian noise and generating them through a denoising process, based on the channel state information acquired in real time during actual communication.

[0082] This invention provides a terminal device comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve corresponding method flows or corresponding functions. The processor described in this embodiment can be used for the operation of a physical layer security policy generation method for uncertain air-ground environments, including: A downlink multi-user multiple-input single-output (MIMO) system model is constructed, comprising a drone base station, multiple legitimate users, and multiple eavesdroppers. Based on the system model, the received signal expressions for legitimate users and eavesdroppers are obtained, and the privacy rate of the communication system is calculated based on the instantaneous signal-to-noise ratio (SNR) of legitimate users and eavesdroppers. An optimization model is constructed with beamforming and artificial noise vectors as optimization variables and maximizing the privacy rate as the objective function. A generator network based on a conditional denoising diffusion probability model is constructed, comprising a channel encoder for processing channel state information and a denoising U-Net network for performing the denoising process. Based on the generator network... The network performs a first-stage training on the conditional diffusion model, minimizing the mean square error of noise prediction to enable the model to learn the conditional distribution of the joint beamforming and artificial noise strategy vectors from Gaussian noise. Based on the pre-trained model trained in the first stage, a loss function with the objective of maximizing security and confidentiality is introduced, and a power constraint penalty term is added to perform a second-stage targeted fine-tuning of the model. In actual communication, based on the real-time channel state information, the model fine-tuned in the second stage starts from random Gaussian noise and generates a beamforming and artificial noise transmission strategy that satisfies the power constraint and has high security and confidentiality through a denoising process.

[0083] Please see Figure 5 The terminal device is a computer device. In this embodiment, the computer device 60 includes a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When executed by the processor 61, the computer program 63 implements the physical layer security policy generation method for uncertain air-ground environments described in this embodiment. To avoid repetition, these details are not elaborated here. Alternatively, when executed by the processor 61, the computer program 63 implements the functions of each model / unit in the physical layer security policy generation system for uncertain air-ground environments described in this embodiment. To avoid repetition, these details are not elaborated here.

[0084] Computer device 60 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. Computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will understand that... Figure 5 This is merely an example of computer device 60 and does not constitute a limitation on computer device 60. It may include more or fewer components than shown, or combine certain components, or different components. For example, computer device may also include input / output devices, network access devices, buses, etc.

[0085] The processor 61 may be a Central Processing Unit (CPU), or other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0086] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or memory of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the computer device 60.

[0087] Furthermore, the memory 62 may include both internal storage units of the computer device 60 and external storage devices. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 can also be used to temporarily store data that has been output or will be output.

[0088] Please see Figure 6 The terminal device is an electronic device 600, which is manifested in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including storage unit 620 and processing unit 610), a display unit 640, etc.

[0089] The storage unit stores program code, which can be executed by the processing unit 610 to perform the steps described in the method section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 610 can perform actions such as... Figure 2 The steps are shown in the figure.

[0090] Storage unit 620 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 6201 and / or cache memory 6202, and may further include a read-only memory (ROM) 6203.

[0091] Storage unit 620 may also include a program / utility 6204 having a set (at least one) program module 6205, such program module 6205 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0092] Bus 630 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the multiple bus structures.

[0093] Electronic device 600 can also communicate with one or more external devices 700 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 600, and / or with any device that enables electronic device 600 to communicate with one or more other computing devices (e.g., router, modem). This communication can be performed via input / output interface 650. Furthermore, electronic device 600 can also communicate with one or more networks (e.g., local area network, wide area network, and / or public network, such as the Internet) via network adapter 660. Network adapter 660 can communicate with other modules of electronic device 600 via bus 630. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms.

[0094] Example 4 This invention also provides a storage medium, specifically a computer-readable storage medium, which is a memory device in a terminal device for storing programs and data. It is understood that the computer-readable storage medium here can include both built-in storage media in the terminal device and extended storage media supported by the terminal device; it can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). More specific examples of the computer-readable storage medium include: an electrical connection with one or more wires, a portable disk, a hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical fiber, portable compact disk read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0095] Computer-readable storage media also include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium can also be any readable medium other than a readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, radio frequency, etc., or any suitable combination thereof.

[0096] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0097] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the physical layer security policy generation method for non-deterministic air-ground environments in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor to perform the following steps: A downlink multi-user multiple-input single-output (MIMO) system model is constructed, comprising a drone base station, multiple legitimate users, and multiple eavesdroppers. Based on the system model, the received signal expressions for legitimate users and eavesdroppers are obtained, and the privacy rate of the communication system is calculated based on the instantaneous signal-to-noise ratio (SNR) of legitimate users and eavesdroppers. An optimization model is constructed with beamforming and artificial noise vectors as optimization variables and maximizing the privacy rate as the objective function. A generator network based on a conditional denoising diffusion probability model is constructed, comprising a channel encoder for processing channel state information and a denoising U-Net network for performing the denoising process. Based on the generator network... The network performs a first-stage training on the conditional diffusion model, minimizing the mean square error of noise prediction to enable the model to learn the conditional distribution of the joint beamforming and artificial noise strategy vectors from Gaussian noise. Based on the pre-trained model trained in the first stage, a loss function with the objective of maximizing security and confidentiality is introduced, and a power constraint penalty term is added to perform a second-stage targeted fine-tuning of the model. In actual communication, based on the real-time channel state information, the model fine-tuned in the second stage starts from random Gaussian noise and generates a beamforming and artificial noise transmission strategy that satisfies the power constraint and has high security and confidentiality through a denoising process.

[0098] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0099] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0100] To verify the effectiveness of the method of this invention, an air-to-ground communication simulation platform centered on a UAV base station was built. The UAV base station was equipped with 8 transmitting antennas, the number of legitimate users varied in gradients from 2 to 8, the number of eavesdroppers was fixed at 4, and the transmit power was adjusted from 0 to 30 dBm. A 5% channel estimation error was introduced to simulate an uncertain air-to-ground environment. The method of this invention was compared with traditional iterative optimization algorithms and zero-forcing beamforming algorithms. The privacy and data rate of the system were tested as a function of the number of legitimate users and transmit power, respectively. The robustness under channel uncertainty was also verified. Please refer to [link to relevant documentation]. Figure 3 The experimental results are as follows: With a fixed transmit power of 20dBm and 5% channel uncertainty, the system privacy and data rate were tested as the number of legitimate users increased from 2 to 8: When the number of legitimate users is 2, the privacy and bandwidth of the method of this invention is 13.1 bps / Hz, the traditional iterative algorithm is 13.3 bps / Hz, and the zero-forcing algorithm is 9.1 bps / Hz. The present invention improves upon the zero-forcing algorithm by 43.9%. When the number of legitimate users is 5, the privacy and bandwidth of the method of this invention is 27.5 bps / Hz, the traditional iterative algorithm is 27.7 bps / Hz, and the zero-forcing algorithm is 21.0 bps / Hz. The present invention improves upon the zero-forcing algorithm by 30.9%. When the number of legitimate users is 8, the privacy and speed of the method of this invention is 33.2bps / Hz, the traditional iterative algorithm is 33.4bps / Hz, and the zero-forcing algorithm is 26.9bps / Hz. The present invention improves the speed by 23.4% compared with the zero-forcing algorithm.

[0101] Depend on Figure 3 As can be seen, the sum and privacy rates of the three schemes generally show an upward trend with the increase of the number of legitimate users. Among them, the performance of the scheme proposed in this invention is consistently better than the zero-forcing algorithm and almost the same as the traditional iterative algorithm. This confirms that fine-tuning enhances the implicit interference cancellation strategy of the model, indicating that the learned artificial noise injection is more effective than the fixed null space design in adapting to multiple eavesdropping channels. The scheme proposed in this invention, by jointly generating beamforming and artificial noise strategies under channel condition constraints, enables the transmitted energy to be more effectively concentrated in the legitimate link direction while satisfying power constraints, and preferentially projects artificial noise into the subspace that is more sensitive to the eavesdropping link. Thus, it can still achieve the dual goals of weakening the eavesdropping reception quality when multi-user interference is enhanced.

[0102] Please see Figure 4 No channel error (solid line) When the transmit power is 5dBm, the method of this invention achieves 7.7bps / Hz, the traditional iterative algorithm achieves 7.8bps / Hz, and the zero-forcing algorithm achieves 4.9bps / Hz. When the transmit power is 15dBm, the method of this invention achieves 19.2bps / Hz, the traditional iterative algorithm achieves 19.4bps / Hz, and the zero-forcing algorithm achieves 13.1bps / Hz. When the transmit power is 20dBm, the method of this invention achieves 24.2bps / Hz, the traditional iterative algorithm achieves 24.4bps / Hz, and the zero-forcing algorithm achieves 18.2bps / Hz.

[0103] 5% channel uncertainty (dashed line) When the transmit power is 5dBm, the method of this invention achieves 6.8bps / Hz, the traditional iterative algorithm achieves 6.7bps / Hz, and the zero-forcing algorithm achieves 4.1bps / Hz. When the transmit power is 15dBm, the method of this invention achieves 14.0bps / Hz, the traditional iterative algorithm achieves 13.7bps / Hz, and the zero-forcing algorithm achieves 9.1bps / Hz. When the transmit power is 20dBm, the method of this invention achieves 16.0bps / Hz, the traditional iterative algorithm achieves 15.5bps / Hz, and the zero-forcing algorithm achieves 11.6bps / Hz.

[0104] Depend on Figure 4 As can be seen, with increasing transmit power, the sum and security rates of all three schemes show an upward trend, and the performance of the proposed scheme consistently outperforms the zero-forcing algorithm and is nearly identical to that of the traditional iterative algorithm. More importantly, the results highlight the robustness of the proposed scheme under conditions of incomplete channel information. When a 5% uncertainty (dashed line) is introduced, the proposed scheme consistently outperforms both the traditional iterative algorithm and the zero-forcing algorithm. This indicates that through training on different data, the proposed scheme has learned a more robust strategy that is less susceptible to channel estimation errors compared to the traditional iterative algorithm. This makes the proposed scheme particularly suitable for practical wireless systems, as channel information is inevitably associated with errors.

[0105] In summary, this invention provides a physical layer security policy generation method and system for uncertain air-to-ground environments. It deeply integrates generative artificial intelligence with physical layer secure communication. By constructing a system model including UAV base stations, legitimate users, and eavesdroppers, and combining a bounded channel uncertainty model to accurately calculate the privacy rate, it builds a denoising U-Net containing a channel encoder and a cross-attention module. The conditional denoising diffusion probabilistic model generator network, after two-stage training including pre-training and targeted fine-tuning of the confidentiality rate with power constraint penalty terms, generates transmission strategies through a denoising diffusion implicit model sampling acceleration method. This effectively solves the technical problems of poor real-time performance, insufficient robustness of discriminative models, and difficulty in balancing security objectives and power constraints in traditional iterative optimization. The generated strategies can accurately concentrate transmission energy on legitimate links and weaken eavesdropping links. It exhibits excellent privacy and rate performance in multi-user and different transmission power scenarios, and still maintains strong robustness under 5% channel estimation error. The strategy generation time is significantly reduced to meet real-time communication requirements, and it can be directly adapted to UAV base station engineering applications. It can also generate multiple candidate strategies to improve the system's anti-attack and anti-mutation capabilities, comprehensively improving the physical layer security transmission capability of multi-user multi-input single-output systems in non-deterministic air-to-ground environments, and providing a practical and advanced technical solution for 6G wireless network air-to-ground communication security.

[0106] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0107] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0108] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0109] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0110] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0111] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0112] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random-access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0113] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0114] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0115] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0116] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for generating physical layer security policies for uncertain air-to-ground environments, characterized in that, Includes the following steps: S1. Construct a downlink multi-user multiple-input single-output system model that includes a drone base station, multiple legitimate users, and multiple eavesdroppers; S2. Obtain the received signal expressions of legitimate users and eavesdroppers based on the system model, and calculate the privacy rate of the communication system based on the instantaneous signal-to-noise ratio of legitimate users and eavesdroppers. S3. Construct an optimization model with beamforming and artificial noise vector as optimization variables and maximizing the privacy rate as the objective function; S4. Construct a generator network based on a conditional denoising diffusion probability model, wherein the generator network includes a channel encoder for processing channel state information and a denoising U-Net network for performing the denoising process. S5. The conditional diffusion model is trained in the first stage based on the generator network. By minimizing the mean square error of noise prediction, the model learns to recover the conditional distribution of the joint beamforming and artificial noise strategy vectors from Gaussian noise. S6. Based on the pre-trained model trained in the first stage, a loss function with the goal of maximizing the confidentiality rate is introduced and a power constraint penalty term is added to perform targeted fine-tuning of the model in the second stage. S7. In actual communication, based on the channel state information acquired in real time, the model, which has been fine-tuned in the second stage, starts from random Gaussian noise and generates a beamforming and artificial noise transmission strategy that meets power constraints and has high security rate through a denoising process.

2. The method for generating physical layer security policies for uncertain air-to-ground environments according to claim 1, characterized in that, In step S1, constructing the downlink multi-user multiple-input single-output system model specifically includes: Configure drone base station equipment Root antenna, service A single-antenna legitimate user, simultaneously existing A single-antenna eavesdropper; The base station sends the first The symbol for each user is The beamforming vector used is The injected artificial noise precoding matrix is The artificial noise vector is Construct the synthetic transmission signal vector of the base station. , For users, set the system to meet the total transmit power requirement. The constraints.

3. The physical layer security policy generation method for non-deterministic air-to-ground environments according to claim 1, characterized in that, In step S2, obtaining the received signal expressions of the legitimate user and the eavesdropper specifically includes: A bounded channel uncertainty model is adopted, which represents the channel state information obtained by the base station as the sum of the estimated channel and the channel error; Among them, drone base station to user Channel vector Represented as From drone base stations to eavesdroppers Channel vector Represented as ; in, and These represent the corresponding estimated channels. and These represent the corresponding channel estimation error vectors, and based on these, an expression for the received signal containing the channel error term is constructed.

4. The method for generating physical layer security policies for uncertain air-to-ground environments according to claim 3, characterized in that, The calculation of the communication system's privacy rate based on the instantaneous signal-to-noise ratio between legitimate users and eavesdroppers specifically includes: Calculate legitimate users achievable rate Its expression is determined based on the signal-to-noise ratio of the received signal; calculating the eavesdropper About users Information eavesdropping rate Its expression is determined based on the signal-to-noise ratio of the eavesdropping channel; the user's expression is calculated. privacy rate The privacy rate is determined by the worst-case eavesdropping channel, and the calculation formula is as follows: in, This represents the total number of eavesdroppers.

5. The method for generating physical layer security policies for uncertain air-to-ground environments according to claim 1, characterized in that, In step S4, constructing the generator network based on the conditional denoising diffusion probability model specifically includes: The composite channel vector is defined as a concatenation of channels from the base station to all users and eavesdroppers, and the composite policy vector is defined as a concatenation of all beamforming vectors and artificial noise vectors. The composite policy vector in the complex domain is converted into a real-domain representation, and its real and imaginary parts are separated as data inputs to the diffusion model. The channel encoder uses a multilayer perceptron to map the complex channel state information into a fixed-dimensional real-valued embedding vector, which is used as conditional information input to the diffusion model. The denoising U-Net network adopts an encoder-decoder architecture, which includes multiple downsampling blocks and upsampling blocks, and connects the corresponding layers through skip connections.

6. The method for generating physical layer security policies for uncertain air-to-ground environments according to claim 5, characterized in that, The denoising U-Net network also includes a cross-attention module, specifically configured as follows: Cross-attention modules are set at the bottleneck layer and key positions of the downsampling and upsampling blocks in the denoising U-Net network. In the cross-attention modules, the features of the current network layer are used as queries, and the channel embedding vector output by the channel encoder is provided as the key and value, so as to guide the denoising process by the channel conditions.

7. The method for generating physical layer security policies for uncertain air-to-ground environments according to claim 1, characterized in that, Step S5, specifically the first-stage training of the conditional diffusion model, includes: Perform a forward diffusion process, gradually adding Gaussian noise to the data via a Markov chain at time step [missing information]. Noisy samples Represented as: in, Standard Gaussian noise, , These are preset variance scheduling parameters; Perform the reverse denoising process by training the neural network. Predict the added noise; define the loss function in the pre-training phase as the mean square error between the predicted noise and the actual noise, and at time step... Optimization is achieved by uniform sampling within the specified interval.

8. The method for generating physical layer security policies for uncertain air-to-ground environments according to claim 1, characterized in that, In step S6, the second stage of targeted fine-tuning of the model specifically includes: An unsupervised loss function that directly targets and maximizes the secrecy rate is adopted. The loss function includes a secrecy rate term and a power constraint penalty term. The power constraint penalty term is used to force the generated policy vector to meet the power limit. It is weighted by balancing the hyperparameters of the secrecy rate target and the power constraint. The model parameters are updated so that the generated distribution is concentrated in the solution space region with high secrecy rate, thereby improving the performance of the model under the specific optimization objective.

9. The method for generating physical layer security policies for uncertain air-to-ground environments according to claim 1, characterized in that, In step S7, the generation of a beamforming and artificial noise transmission strategy that satisfies power constraints and has high security through the denoising process specifically includes: A noise reduction diffusion implicit model sampling acceleration method is adopted to sample the initial noise vector from a standard Gaussian distribution. Using a trained and finely tuned noise prediction network, iterative noise reduction is performed according to the reverse time step to estimate the sample of the previous time step, where the randomness of the sampling process is controlled. After the iteration is completed, the final denoised sample is obtained, and the real and imaginary parts are recombined by inverse transformation to obtain the beamforming and artificial noise strategy vector in the complex domain, which is directly applied to the signal transmission of UAV base stations.

10. A physical layer security policy generation system for uncertain air-to-ground environments, characterized in that, include: The model building module is used to build a downlink multi-user multiple-input single-output system model that includes a drone base station, multiple legitimate users, and multiple eavesdroppers; The rate calculation module is used to obtain the received signal expressions of legitimate users and eavesdroppers based on the system model, and to calculate the private rate of the communication system based on the instantaneous signal-to-noise ratio of legitimate users and eavesdroppers. An optimization construction module is used to construct an optimization model with beamforming and artificial noise vectors as optimization variables and maximizing the privacy rate as the objective function; A network construction module is used to construct a generative network based on a conditional denoising diffusion probability model. The generative network includes a channel encoder for processing channel state information and a denoising U-Net network for performing the denoising process. The first training module is used to perform a first-stage training of the conditional diffusion model based on the generator network. By minimizing the mean square error of noise prediction, the model learns to recover the conditional distribution of the joint beamforming and artificial noise policy vectors from Gaussian noise. The second fine-tuning module is used to perform targeted fine-tuning of the model in the second stage based on the pre-trained model trained in the first stage, by introducing a loss function with the goal of maximizing the confidentiality rate and adding a power constraint penalty term. The strategy generation module is used to generate beamforming and artificial noise transmission strategies that meet power constraints and have high security rates by using a model that has been fine-tuned in the second stage, starting from random Gaussian noise and generating them through a denoising process, based on the channel state information acquired in real time during actual communication.

Images