Interference signal generation method based on diffusion model
By generating interference signals using a diffusion model, the health risks and prior knowledge dependencies of high-power noise interference are addressed, enabling the generation of high-precision interference signals that effectively block unauthorized communications and improve the effectiveness and stability of the interference signals.
Patent Information
- Application Number
- CN202511481131.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-16
- Publication Date
- 2026-02-10
AI Technical Summary
Existing mobile communication signal jamming devices typically employ high-power white noise interference, which poses health risks. Furthermore, traditional and deep learning methods rely on prior knowledge, making it difficult to generate effective jamming signals in the absence of information, and conventional generation models are prone to failure.
An interference signal generation method based on a diffusion model is adopted. Noise perturbation signal is added by forward diffusion, and the original distribution is restored by conditional back diffusion to generate a high-precision interference signal. The conditional information is integrated by using Transformer to optimize KL divergence to generate a targeted interference signal.
Without prior knowledge, it generates high-precision, high-frequency-domain-fidelity interference signals to effectively block illegal communication links, avoid mode collapse, and improve interference effectiveness.
Smart Images
Figure CN121508731A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of wireless communication technology, specifically relating to a method for generating interference signals based on a diffusion model. Background Technology
[0002] For important national conferences and examinations, it is necessary to deploy mobile communication signal jamming equipment to shield 2G / 3G / 4G / 5G mobile communication signals in the conference areas and examination rooms. However, current mobile communication signal jamming equipment typically uses broadband high-power white noise interference, which has excessively high transmission power and can have harmful effects on the health of participants and examinees. By analyzing the waveform characteristics of mobile communication signals and designing jamming signals, the transmission power of the jamming signals can be significantly reduced, thereby improving the effectiveness of the jamming.
[0003] Currently, communication jamming methods can be mainly divided into traditional jamming methods, reinforcement learning methods, and deep learning methods. Traditional jamming methods utilize 5G's polar coding mechanism and physical layer Orthogonal Frequency Division Multiplexing (OFDM) structure to generate injected signals. These signals are 3.4 dB lower than legitimate signals and aim to mask the Synchronization Signal Block (SSB). This method first analyzes the SSB of the target signal, which contains important synchronization and timing information. By extracting this information, the method reconstructs prior knowledge of the target signal, such as its modulation scheme, frequency, and other parameters. It then uses this information to generate jamming signals similar to legitimate signals. These jamming signals are designed to disrupt communication while mimicking the characteristics of the target signal.
[0004] While reinforcement learning methods such as Q-learning, multi-armed bandit (MAB) algorithms, and game theory have been used to optimize traditional jamming strategies, these algorithms typically require extensive trial-and-error processes to achieve stable and effective jamming signal generation strategies. The aforementioned communication jamming methods often rely on the attacker's prior knowledge of the target communication signal, such as modulation scheme, frequency, and signal power. However, in real-world scenarios, communication jamming signals are usually used to interfere with non-cooperative systems where prior information is limited. Constructing such communication jamming signals is extremely complex, posing a significant challenge to the practical application of communication jamming.
[0005] With the continuous advancement of Artificial Intelligence Generated Content (AIGC) technology, the widespread application of various generative models has enabled the application of artificial intelligence in the field of communication, such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and Denoising Diffusion Probabilistic Models (DDPMs). However, GANs and VAEs primarily focus on expanding feature-level distributions. Since GANs generate data by learning the distribution of input data, they may struggle to accurately reconstruct the details of the target signal if the data distribution is too complex. On the other hand, VAEs compress input data into a latent space distribution through an encoder and then reconstruct the data from that distribution. This means that when generating new data, the model can only sample from this finite latent space, failing to fully capture the complex features of the target signal. Furthermore, GANs are trained through adversarial interactions between the generator and discriminator, which often leads to pattern collapse, i.e., the generated data lacks diversity and converges to a finite subset of patterns.
[0006] Against this backdrop, diffusion models play a crucial role in image and video generation, data augmentation, and other fields. Unlike GANs and VAEs, diffusion models use an iterative process of progressively adding and removing noise to capture complex data distributions. These models generate interfering signals in a data-driven manner, learning the latent distribution of the target signal through unsupervised learning methods without relying on specific prior knowledge. Traditional interference methods struggle to acquire prior knowledge such as the frequency and protocol characteristics of the target signal, while generative models like GANs and VAEs face pattern collapse issues, resulting in poor signal interference performance.
[0007] Considering the advantages of diffusion models, this invention proposes an interference signal generation method based on diffusion models. This method can generate interference signals with high time-domain accuracy and excellent frequency-domain fidelity. It can generate special interference signals for illegal receiving terminals and solve the problem of conventional generation models being prone to crashing. Summary of the Invention
[0008] To address the aforementioned problems, this invention proposes a method for generating interference signals based on a diffusion model. This method can generate specialized interference signals targeting unauthorized receiving terminals and also solves the problem of conventional generation models easily crashing. The details are as follows: Content 1: Forward Diffusion Process of Interference Signal Generation The forward diffusion process of a signal involves treating the signal as complex-valued time-series data. Noise is added to the signal in the time domain, and interference is introduced into the frequency domain to gradually disturb the original signal.
[0009] Content 2: Conditional Backdiffusion Process for Interference Signal Generation The received mobile communication base station signal from the environment is used as the input condition. The back-diffusion process mainly removes noise from the output of the forward diffusion process to restore the original distribution. Then, the interference signal is generated with the goal of minimizing the Kullback-Leibler (KL) divergence between the interference signal and the mobile communication base station signal.
[0010] The technical solution of this invention is as follows: An interference waveform generation method based on a diffusion model includes the following steps: when a signal shielding device... Mobile communication base stations were detected to illegal receiving terminal Send target signal At that time, the signal shielding device The target signal is generated using a diffusion model. interference waveform It was sent to the illegal receiving terminal. This makes the illegal receiving terminal Received interference signal This blocks the mobile communication base station. With the illegal receiving terminal The communication link between them; among which, It is a signal shielding device Transmission power, Indicates signal shielding equipment and illegal receiving terminals Channel gain between It is additive white Gaussian noise.
[0011] Preferably, the target signal is generated using a trained diffusion model. interference waveform The method for training the diffusion model is as follows: 21) Collect raw data transmitted by mobile communication base stations. ; 22) The diffusion model described above applies to the original data. Perform forward diffusion: convert the original data As complex-valued time series data, the original data is analyzed in the time domain. Noise is added and reparameterization techniques are used to represent the data distribution at each step of the forward process, introducing interference in the frequency domain to progressively perturb the original data. get Among them, from the perspective of time arrive The forward diffusion process is ;in, , , and It is the hyperparameter corresponding to time step t. It is Gaussian noise; 23) The diffusion model described above applies to Conditional backdiffusion is used to remove noise in order to restore the original distribution. ; 24) By minimizing and KL divergence between Optimize the diffusion model; wherein, Represents raw data The reverse process distribution, It is the predicted noise.
[0012] Preferred, from arrive The conditional probability distribution is ;in, It is the conditional probability distribution of the forward diffusion process, representing the signal obtained after the t-th diffusion step. The distribution yes scaling factor, It represents the variance of the noise components, and I represents the identity matrix.
[0013] Preferred, Indicates from noisy samples Predict the previous sample The process of distribution; among which, It is the mean of the previous state predicted based on the current state in step t of the back diffusion process.
[0014] Preferably, the diffusion model generates the target signal by minimizing the KL divergence between the generated interference signal s(t) and the target signal x(t). interference waveform .
[0015] The advantages of this invention are as follows: 1) Introducing a conditional mechanism into the conditional backpropagation process, and using Transformer to integrate conditional information, can generate interference signals for illegal receiving terminals, thereby effectively blocking the communication links of illegal receiving terminals. Its interference performance is better than traditional interference methods, improving the effectiveness of interference.
[0016] 2) Even in the absence of prior information, this method can model complex time-frequency relationships in signals, thus generating interference signals against unauthorized receiving terminals without relying on prior knowledge.
[0017] 3) It effectively avoids the pattern collapse caused by the game between the generator and discriminator in GANs, and the model training is more stable than that of GANs. Attached Figure Description
[0018] Figure 1 This is a diagram of the network topology model architecture.
[0019] Figure 2 This is a flowchart of forward and reverse diffusion. Detailed Implementation
[0020] The present invention will now be described in further detail with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.
[0021] This invention is the first to introduce a diffusion model into the field of wireless communication interference signal generation, and designs a novel interference signal generation method based on the diffusion model in conjunction with the Transformer architecture. The network topology model architecture diagram of the method described in this invention is shown below. Figure 1 As shown, three key entities are involved: mobile communication base stations. Illegal receiving terminal and signal shielding equipment The signal jamming device uses the method described in this invention to generate interference signals, thereby blocking the communication link between mobile communication base stations and illegal receiving terminals, and preventing information leakage in important meeting venues or illegal activities such as cheating on exams using mobile communication networks.
[0022] The specific implementation process is as follows: 1. Training Phase The training phase involves pre-collecting mobile communication base station signals in the environment to construct raw data, and then using a forward diffusion process to process the raw signals. The process consists of two steps: raw data acquisition and forward diffusion.
[0023] (1) Collect raw data .
[0024] First, mobile communication base station signals in the environment are collected, covering all 2G / 3G / 4G / 5G mobile communication frequency bands. The collected data is represented as raw data. .
[0025] (2) The forward diffusion process yields During the forward propagation of a signal, the signal can be considered as complex-valued time-series data. Let the received raw data... Follow a specific distribution During forward diffusion, the original signal can be gradually perturbed by adding noise in the time domain and introducing interference in the frequency domain. This iterative diffusion modifies the original data. Distribution in the time and frequency domains, from time arrive The forward diffusion process can be represented as: in and These represent the Fourier transform and its inverse transform, respectively. This represents Gaussian convolution used for spectral blurring, and * represents the circular convolution operator. These are predefined parameters. It is Gaussian noise. According to the convolution theorem, it can be simplified to: in .
[0026] During the forward diffusion process, Gaussian noise is emitted from each step. Gradients are added gradually. However, the process becomes increasingly complex as gradients propagate through the diffusion steps. The gradients at each layer are affected by accumulated noise from previous steps, which slows down convergence during training. To ensure the feasibility of the forward diffusion process, it is necessary to be able to directly add gradients from previous steps. Add noise at any time step t, where To simplify this process, a reparameterization technique is used to represent the data distribution at each step of the forward process. This is achieved through: The simplified equation is: in: .
[0027] because and It is the hyperparameter corresponding to time step t. and It is the constant corresponding to time step t. Therefore, from arrive The conditional probability distribution can be expressed as: in It is the conditional probability distribution of the forward diffusion process, representing the signal obtained after the t-th diffusion step. The distribution of the initial signal. . yes The scaling factor, and It represents the variance of the noise components, and I represents the identity matrix.
[0028] 2. Reasoning Stage During the inference phase, the signal jamming device receives and analyzes signals from the mobile communication base station. Based on the time-frequency characteristics, a diffusion model is used to generate a high-precision interference signal in the time domain. At the same time, it ensures excellent fidelity in the frequency domain, and is divided into two steps: conditional back diffusion process and interference signal generation.
[0029] (1) Conditional backdiffusion process The reverse process removes noise to restore the original distribution. It can be represented as the following Markov chain: in, Indicates that it has learnable parameters The distribution of the reverse diffusion process. Unlike the forward diffusion process, the reverse diffusion process cannot be directly calculated. Therefore, model training is used to minimize... and KL divergence between:
[0030] in, Represents raw data The reverse process distribution. The above two equations can be reformulated as follows to improve the quality of the generated signal: in It is the predicted noise. Therefore, this invention minimizes... and The mean squared error (MSE) between them is used to reduce the KL divergence. Meanwhile, The distribution of ) can be represented as: in It is the mean of the previous state predicted based on the current state at time step t in the back diffusion process. In prediction It plays a key role in the mean of the distribution. ) indicates from noisy samples Predict the previous sample The process of distribution, which is achieved through the distribution of... Remove estimated Gaussian noise To achieve this.
[0031] (2) Interference signal generation Signal shielding devices receive mobile communication base station signals in the environment. , target signal As a conditional input, to maximize the interference effect, it is necessary to generate a signal that is similar to the target signal. Interference signals with highly similar statistical characteristics This means maximizing the probability of generating the optimal interference signal given the target signal. Specifically, similarity is measured by minimizing the distribution differences between signals: The goal of generating the interference signal is to minimize the KL divergence between the generated interference signal s(t) and the target signal x(t): in, Indicates the optimization parameters. and These are the probability distributions of the target signal and the generated interference signal, respectively.
[0032] In summary, this invention proposes a novel interference signal generation method based on a diffusion model. This method uses the target signal as a conditional input to generate a highly targeted interference signal. Experimental results show that the proposed algorithm can generate interference signals without prior knowledge and outperforms traditional interference methods and VAE-based interference signal generation methods.
[0033] Although specific embodiments of the invention have been disclosed for illustrative purposes to aid in understanding and implementing the invention, those skilled in the art will understand that various substitutions, variations, and modifications are possible without departing from the spirit and scope of the invention and the appended claims. Therefore, the invention should not be limited to the content disclosed in the preferred embodiments, and the scope of protection claimed by the invention is defined by the claims.
Claims
1. A method for generating interference waveforms based on a diffusion model, comprising the following steps: When signal shielding device Mobile communication base stations were detected to illegal receiving terminal Send target signal At that time, the signal shielding device The target signal is generated using a diffusion model. interference waveform It was sent to the illegal receiving terminal. This makes the illegal receiving terminal Received interference signal This blocks the mobile communication base station. With the illegal receiving terminal The communication link between them; among which, It is a signal shielding device Transmission power, Indicates signal shielding equipment and illegal receiving terminals Channel gain between It is additive white Gaussian noise.
2. The method according to claim 1, characterized in that, The target signal is generated using the trained diffusion model. interference waveform The method for training the diffusion model is as follows: 21) Collect raw data transmitted by mobile communication base stations. ; 22) The diffusion model described above applies to the original data. Perform forward diffusion: convert the original data As complex-valued time series data, the original data is analyzed in the time domain. Noise is added and reparameterization techniques are used to represent the data distribution at each step of the forward process, introducing interference in the frequency domain to progressively perturb the original data. get ; among which, from time arrive The forward diffusion process is ;in, , , and It is the hyperparameter corresponding to time step t. It is Gaussian noise; 23) The diffusion model described above applies to Conditional backdiffusion is used to remove noise in order to restore the original distribution. ; 24) By minimizing and KL divergence between Optimize the diffusion model; wherein, Represents raw data The reverse process distribution, It is the predicted noise.
3. The method according to claim 2, characterized in that, from arrive The conditional probability distribution is ;in, It is the conditional probability distribution of the forward diffusion process, representing the signal obtained after the t-th diffusion step. The distribution, yes scaling factor, It represents the variance of the noise components, and I represents the identity matrix.
4. The method according to claim 3, characterized in that, Indicates from noisy samples Predict the previous sample The process of distribution; among which, It is the mean of the previous state predicted based on the current state in step t of the back diffusion process.
5. The method according to claim 1, characterized in that, The diffusion model generates the target signal by minimizing the KL divergence between the generated interference signal s(t) and the target signal x(t). interference waveform .