Deep learning based intelligent interference management system for wireless communication networks

Abstract

The application discloses a wireless communication network intelligent interference management system based on deep learning and belongs to the technical field of communication network management; by processing and fusing multipath, mobility and environmental factors, the problem of inaccurate channel state information caused by the traditional model ignoring non-stationary interference can be solved; the constructed sequence simultaneously captures time trend and spatial cooperation, which can provide high-quality input for subsequent adaptive models; by obtaining cross-scene general parameters through meta-learning pre-training, realizing millisecond-level adaptive dynamic channels through online fine-tuning, and finally outputting accurate interference prediction, the adaptability of the model to complex channels and the real-time performance of interference estimation can be significantly improved, which provides key support for the anti-disturbance ability of trajectory control; through the cooperation of the above steps, the cross-scene generalization foundation is laid through the pre-training stage, the new scene is quickly adapted through the online fine-tuning stage, and the key compensation information is output in the interference prediction stage, forming a closed-loop logic of pre-training, fine-tuning and prediction.

Description

Technical Field

[0001] This invention relates to the field of communication network management technology, and more specifically to a deep learning-based intelligent interference management system for wireless communication networks. Background Technology

[0002] Wireless communication network interference management refers to the process of identifying, controlling, coordinating, or eliminating harmful interference between different signal sources in a wireless communication system through a series of technical means, in order to ensure communication quality, improve spectrum efficiency, enhance system capacity, and improve user experience.

[0003] Because wireless channels are affected by multipath effects, mobility, weather, obstacles, etc., channel state information fluctuates drastically within milliseconds. Most existing deep learning models assume that the data distribution is stable, making it difficult to adapt to the rapidly changing channel environment in real time, resulting in a sharp drop in the performance of interference prediction and suppression models. Summary of the Invention

[0004] The purpose of this invention is to provide a deep learning-based intelligent interference management system for wireless communication networks, which solves the technical problem that existing technical solutions are unable to adapt to rapidly changing channel environments in real time, leading to a sharp drop in the performance of interference prediction and suppression models.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A deep learning-based intelligent interference management system for wireless communication networks includes:

[0007] Real-time channel state information processing module: Based on the multipath effect, a tapped delay line model is constructed. The effective path characteristics are dynamically updated by fusing mobility parameters. The effective path gain is adjusted by fusing environmental interference factors. Based on the fused complex gain, a real-time channel state information sequence containing spatiotemporal correlation features is generated.

[0008] Meta-learning-driven learning processing module: Based on the generated real-time channel state information sequence, it obtains cross-scenario universal initialization parameters through meta-learning pre-training, and uses online incremental learning to quickly fine-tune new samples within milliseconds, outputting interference prediction results;

[0009] Interference prediction result analysis and dynamic resource adjustment module: performs data analysis on the output interference prediction results and dynamically adjusts the resource allocation scheme based on the analysis results.

[0010] Furthermore, when constructing a tapped delay line model based on multipath effects, multipath propagation is described using tapped delay lines, and the relevant expressions are:

[0011] ;in, is the multipath channel impulse response at time t; k is the effective path number. The number of valid paths; Let be the complex gain of the kth valid path; The propagation delay for the k-th valid path; For carrier frequency; is the Dirac delta function; j is the imaginary unit; t is a continuous time variable.

[0012] Furthermore, when dynamically updating effective path features by fusing mobility parameters, the relevant expression is:

[0013] ;in, Let $\frac{k}{k}$ be the rate of change of the delay of the k-th effective path over time. Let be the unit vector representing the arrival direction of the kth valid path; The instantaneous velocity vector of the user; It is the speed of light.

[0014] Furthermore, when adjusting the effective path gain by incorporating environmental interference factors, the relevant expression is:

[0015] ;in, The complex gain of the k-th effective path after the environment is merged; Let be the rain attenuation coefficient at time t; The obstacle occlusion coefficient at time t has a value range of (0,1], and is 1 when there is no occlusion; the environmental interference factor includes the rain attenuation coefficient and the obstacle occlusion coefficient.

[0016] Furthermore, a multi-antenna channel state information matrix is ​​constructed based on the fused complex gain. M represents the number of transmit antennas, N represents the number of receive antennas, and C indicates that the matrix elements belong to the complex field. The combined complex gain is given by m, where m represents the m-th transmitting antenna, m∈{1, 2, ..., M}; and n represents the n-th receiving antenna, n∈{1, 2, ..., N}.

[0017] The constructed multi-antenna channel state information matrix is ​​combined with the window length T to generate a channel state information sequence. .

[0018] Furthermore, a sequence sample set of channel state information under multiple scenarios was collected. ;in, For the scene The channel state information sequence; Total number of scenes; These are interference labels for the corresponding scenarios;

[0019] Employing a model-independent meta-learning framework, the model It consists of a CNN-LSTM feature extractor and a fully connected prediction head; among which, The parameters are general initialization parameters; the CNN-LSTM feature extractor is used to capture the spatiotemporal features of channel state information;

[0020] During meta-training, the general initialization parameters are updated by alternately optimizing the meta-loss. .

[0021] Furthermore, when using online incremental learning to rapidly fine-tune new samples within milliseconds, based on the generated real-time channel state information sequence... With general initialization parameters Using the initial model, a single-step gradient update is employed to adapt to new samples, resulting in online fine-tuned model parameters. .

[0022] Furthermore, the model was fine-tuned online. When reasoning about real-time channel state information sequences, the model is used. The CNN-LSTM feature extractor fuses spatial and temporal features on the preprocessed real-time channel state information sequence, with an output dimension of [missing information]. The global feature vector F; where, For feature dimensions;

[0023] The global feature vector F is input into two parallel output branches: the interference type output branch and the interference intensity output branch.

[0024] Furthermore, the interference type is predicted by using the interference type output branch. The interference intensity is predicted by using the output branch of the interference intensity. The predicted interference type and predicted interference intensity are integrated into a structured output. .

[0025] Furthermore, when dynamically adjusting the scheme based on the predicted interference type and predicted interference intensity in the output interference prediction results, if and If so, then the scheme of adjusting only the transmission power will be implemented;

[0026] like and Then, a joint adjustment of power and beamforming vector scheme will be implemented; Indicates Gaussian interference; Indicates pulse interference; The threshold for weak interference strength;

[0027] like If so, a subcarrier hopping power control scheme is implemented.

[0028] Compared to existing solutions, the beneficial effects achieved by this invention are:

[0029] This invention addresses the problem of inaccurate channel state information caused by neglecting non-stationary interference in traditional models by processing and fusing multipath, mobility, and environmental factors. The constructed sequence simultaneously captures temporal trends and spatial cooperation, providing high-quality input for subsequent adaptive models. Through the synergistic cooperation of the above steps, accurate dynamic channel input can be provided for subsequent meta-learning fine-tuning and interference suppression, which is the core foundation for improving system robustness.

[0030] This invention acquires cross-scenario general parameters through meta-learning pre-training, achieves millisecond-level adaptation to dynamic channels through online fine-tuning, and finally outputs accurate interference prediction. This can significantly improve the model's adaptability to complex channels and the real-time performance of interference estimation, providing key support for the anti-disturbance capability of trajectory control. Through the synergistic cooperation of the above steps, the pre-training stage lays the foundation for cross-scenario generalization, the online fine-tuning stage quickly adapts to new scenarios, and the interference prediction stage outputs key compensation information, forming a closed-loop logic of pre-training, fine-tuning, and prediction, supporting the anti-interference requirements of trajectory control.

[0031] This invention analyzes the predicted interference type and intensity in the interference prediction results and dynamically adjusts the optimal strategy based on the analysis results, which can effectively improve the suppression efficiency and significantly improve the communication quality of the system in complex interference scenarios. Attached Figure Description

[0032] The invention will now be further described with reference to the accompanying drawings.

[0033] Figure 1 This is a flowchart illustrating the operation of the intelligent interference management system for wireless communication networks based on deep learning, as described in this invention. Detailed Implementation

[0034] 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 embodiments of the present invention, and not all embodiments. 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.

[0035] like Figure 1 As shown, the present invention is a deep learning-based intelligent interference management system for wireless communication networks, including a real-time channel state information processing module, a meta-learning-driven learning processing module, and an interference prediction result analysis and resource dynamic adjustment module.

[0036] Real-time channel state information processing module: Based on multipath effects, a tapped delay line model is constructed. Effective path features are dynamically updated by fusing mobility parameters, and the effective path gain is adjusted by fusing environmental interference factors. Based on the fused complex gain, a real-time channel state information sequence containing spatiotemporal correlation features is generated. Specific steps include:

[0037] When constructing a tapped delay line model based on multipath effects, multipath propagation is described using tapped delay lines, and the relevant expressions are:

[0038] ;in, is the multipath channel impulse response at time t; k is the effective path number. The number of valid paths; Let the complex gain of the k-th valid path be... , Let be the initial complex gain of the k-th valid path; This refers to the channel coherence time; The dot product of user speed and path direction; The propagation delay for the k-th valid path; For carrier frequency; Let j be the Dirac delta function, and let t be the effective path delay time; j is the imaginary unit; t is the continuous time variable. For phase offset terms; For time-based positioning items;

[0039] It should be explained that the tapped delay line model is the core tool for wireless channel modeling. Its role is involved in the theoretical analysis, algorithm design and engineering implementation of communication systems. It can intuitively reflect the arrival time difference of multipath signals (through tapped delay) and energy distribution (through tapped gain), providing a basis for analyzing the impact of multipath fading.

[0040] Through changes over time and It simulates the Doppler frequency shift and multipath environment dynamic changes caused by the movement of mobile terminals or base stations, such as scene switching when a vehicle is moving; Fourier transform of h(t) can obtain the channel frequency response H(f), which can be directly used to analyze the channel frequency selectivity, such as inter-carrier interference and equalizer design requirements, and is the core analysis basis for broadband communication systems such as OFDM and 5G NR.

[0041] It should be noted that the criteria for determining an effective path are usually based on the ratio of the path signal strength to the direct path strength, or the energy percentage, as a threshold to filter out effective paths; for example, the ratio of the path signal strength to the direct path strength is greater than -20dB; the energy percentage is for example, contributing more than 90% of the total energy.

[0042] When dynamically updating effective path features by fusing mobility parameters, the relevant expression is:

[0043] ;in, Let $\frac{k}{k}$ be the rate of change of the delay of the k-th effective path over time. Let be the unit vector representing the arrival direction of the kth valid path; The instantaneous velocity vector of the user; The speed is the speed of light; the mobility parameter includes the user's instantaneous velocity vector;

[0044] It should be explained that this set of expressions is a core component of time-varying channel modeling. Its role spans from the underlying channel estimation to the upper-level algorithm optimization of mobile communication systems. It can accurately characterize the time-varying channel characteristics of high-speed mobile scenarios, solving the problems that static channel models cannot adapt to, such as high-speed rail and vehicle-to-everything (V2X) scenarios where user speeds can reach over 300 km / h, and the rate of change of multipath delay and gain is extremely high, with traditional static models having an error of over 50%. Combined with tapped delay line models, it can generate channel impulse responses that conform to real mobile scenarios, providing accurate input for communication algorithm simulation. Its implementation logic closely follows the physical propagation laws, and its core function covers the entire process from underlying channel estimation to upper-level algorithm optimization. Its beneficial effects are directly reflected in the improvement of communication reliability and the optimization of system resource utilization in high-speed mobile scenarios, and it is one of the key theoretical foundations supporting 4G LTE and 5G NR mobile communication technologies.

[0045] When adjusting the effective path gain by incorporating environmental interference factors, the relevant expression is:

[0046] ;in, The complex gain of the k-th effective path after the environment is merged; Let be the rain attenuation coefficient at time t. ;in, This is a frequency-dependent attenuation coefficient, which is positively correlated with the carrier frequency, for example, at 5GHz. ≈0.012, at 28GHz ≈0.15; Let be the instantaneous rainfall rate at time t; The rainfall rate index is set to 0.75 by default for convective rain and 0.65 by default for stratiform rain. The length of the signal propagation path; It is a natural exponential function; Let be the obstacle occlusion coefficient at time t, with a value ranging from (0,1], and 1 when there is no occlusion. ;in, Free space path loss; The additional loss due to obstacles at time t is 10 for brick wall obstacles, 5 for tree obstacles, and 3 for glass obstacles. When there are multiple obstacles, it is the sum of the losses of each obstacle. The environmental interference factor includes the rain attenuation coefficient and the obstacle shading coefficient.

[0047] It should be explained that this expression is a quantitative model of the attenuation effect of the wireless channel environment. By superimposing environmental factors such as rainfall and obstacles, it performs a secondary correction on the channel complex gain after the influence of mobility, thereby achieving a full-dimensional characterization of channel characteristics. Its implementation logic closely matches the physical essence of environmental attenuation. Its core function is to construct a full-dimensional channel model for real-world scenarios. The beneficial effects are directly reflected in the improvement of communication system robustness, the reduction of engineering costs, and the support for the implementation of intelligent communication. It is one of the core tools supporting the design and optimization of mobile communication systems in complex scenarios.

[0048] Constructing a multi-antenna channel state information matrix based on the fused complex gain M represents the number of transmit antennas, N represents the number of receive antennas, and C indicates that the matrix elements belong to the complex field. The combined complex gain is given by m, where m represents the m-th transmitting antenna, m∈{1, 2, ..., M}; and n represents the n-th receiving antenna, n∈{1, 2, ..., N}.

[0049] The constructed multi-antenna channel state information matrix is ​​combined with the window length T to generate a channel state information sequence. ;

[0050] In this embodiment of the invention, by processing and fusing multipath, mobility, and environmental factors, the problem of inaccurate channel state information caused by neglecting non-stationary interference in traditional models can be solved; the constructed sequence simultaneously captures temporal trends and spatial cooperation, which can provide high-quality input for subsequent adaptive models; through the synergistic cooperation of the above steps, accurate dynamic channel input can be provided for subsequent meta-learning fine-tuning and interference suppression, which is the core foundation for improving system robustness.

[0051] Meta-learning-driven learning processing module: Based on the generated real-time channel state information sequence, it obtains cross-scenario universal initialization parameters through meta-learning pre-training, and uses online incremental learning to quickly fine-tune new samples within milliseconds, outputting interference prediction results; the specific steps include:

[0052] When obtaining cross-scenario universal initialization parameters through meta-learning pre-training, it is necessary to collect channel state information sequence sample sets under multiple scenarios. ;in, For the scene The channel state information sequence; Total number of scenes; Interference labels for corresponding scenarios, such as noise power labels and interference type labels; multiple scenarios include but are not limited to indoor, outdoor, dynamic occlusion, etc.

[0053] Employing a model-independent meta-learning framework, the model It consists of a CNN-LSTM feature extractor and a fully connected prediction head; among which, The parameters are general initialization parameters; the CNN-LSTM feature extractor is used to capture the spatiotemporal features of channel state information;

[0054] During meta-training, the general initialization parameters are updated by alternately optimizing the meta-loss. The expression involved is ;in, The meta-loss function; For expectation operators; For task sample pairs; L is the task loss function; f is the deep learning model to be trained, such as CNN-LSTM; For internal fine-tuning step size; Gradient operator for parameters; Used as training samples for the task; Label the task training; This serves as a test sample for the task. Tag for task testing;

[0055] It should be explained that this expression is the core mathematical expression of model-independent meta-learning, which aims to train a general model initialization parameter through the paradigm of "learning how to learn," so that it can quickly adapt to new tasks with only a small amount of data and fine-tuning. Through the logic of "alternating optimization of the inner and outer loops," it realizes the paradigm upgrade from single-task learning to learning how to learn. Its core function is to generate general initialization parameters, allowing the communication model to adapt quickly in small sample scenarios. The beneficial effects are directly reflected in the performance improvement when data is scarce, the enhanced cross-scenario generalization ability, and the reduction of engineering costs.

[0056] When using online incremental learning to rapidly fine-tune new samples within milliseconds, based on the generated real-time channel state information sequence... With general initialization parameters For the initial model, a single-step gradient update is used to adapt to the new samples. The relevant expressions are:

[0057] ;in, These are the model parameters after online fine-tuning; Temporary interference labels for real-time channel state information sequences are obtained through short-term statistics or prior knowledge; For online fine-tuning of step size;

[0058] It should be explained that this expression is a key step in applying meta-training to real-time communication scenarios. Through lightweight single-step / few-step gradient updates, it updates the pre-trained general initialization parameters. It can quickly adapt to the current real-time channel environment and solve the pain points of dynamic changes in interference and lagging data labeling in communication scenarios. Its implementation logic is achieved through lightweight single-step updates and automatic temporary tag generation. Its core function is to balance real-time performance and support edge deployment. Its beneficial effects are directly reflected in the real-time improvement of interference prediction accuracy, the order-of-magnitude reduction of response time, and the significant reduction of resource consumption.

[0059] Using the online fine-tuned model When reasoning about real-time channel state information sequences, the model is used. The CNN-LSTM feature extractor fuses spatial and temporal features on the preprocessed real-time channel state information sequence, with an output dimension of [missing information]. The global feature vector F; where, For feature dimensions;

[0060] The global feature vector F is input into two parallel output branches: the interference type output branch and the interference intensity output branch.

[0061] Among them, the interference type output branch outputs 2D logits through a fully connected layer, corresponding to Gaussian or impulse interference, with a shape of [1,2].

[0062] The interference intensity output branch outputs a 1D regression value through a fully connected layer, which corresponds to the noise power spectral density and has a shape of [1,1].

[0063] When converting the model output into interpretable disturbance types and intensities, a Softmax operation is performed on the logits of the disturbance type output branch, and the category with the highest probability is taken as the final disturbance type. The relevant expression is:

[0064] ;in, To predict the type of interference, the output is 0 or 1, where 0 corresponds to Gaussian interference and 1 corresponds to impulse interference, indicating the main type of interference in the current channel as determined by the model. The maximum index function selects the index corresponding to the category with the highest probability from the input probability distribution as the final prediction type; The Softmax activation function converts the unnormalized score output by the model into a probability distribution between 0 and 1, satisfying that the sum of the probabilities of all classes is 1. The unnormalized score for the type branch is the original value output by the interference type prediction branch in the model, reflecting the model's confidence in different interference types. The larger the value, the higher the confidence in the corresponding type.

[0065] Furthermore, the regression value of the interference intensity output branch is inversely normalized to restore the actual noise power spectral density, and the relevant expression is as follows:

[0066] ;in, To predict the interference intensity, the output is the actual noise power spectral density, which represents the magnitude of the current channel interference predicted by the model. The normalized output of the intensity branch is the normalized value of the interference intensity prediction branch output in the model. It usually takes a value range of 0~1 or -1~1 and is the model's original prediction result of the interference intensity. The standard deviation of the intensity training data is specifically the standard deviation obtained from historical interference intensity data during the pre-training phase, which is used to restore the numerical range during inverse normalization. The average value of the intensity training data is specifically the average value obtained from historical interference intensity data during the pre-training phase, compared with... This will help achieve inverse normalization;

[0067] It needs to be explained that the predicted interference type With predicted interference intensity The processing of interference prediction models is a key link from feature extraction to practical application. Its synergistic effect is reflected in type determination, selection of suppression methods, intensity calculation, and adjustment of resource allocation, forming a complete interference processing flow. It can also be quickly extended to more interference types, such as multipath interference and neighboring cell interference, simply by adding the logits dimension and category mapping rules. The output results fully comply with the interference index definition of the 3GPP standard and can be directly connected to the MAC layer and PHY layer algorithms of the base station without large-scale modification of the existing system.

[0068] These two operations transform the abstract output of the AI ​​model into executable decisions for the communication system, realizing full-process intelligentization of interference perception, intelligent decision-making, and resource optimization. They are the core links in improving the robustness and resource utilization of the communication system.

[0069] The predicted interference type and predicted interference intensity are integrated into a structured output. ;

[0070] When performing smoothing and validity checks on the prediction results of the structured output, the prediction interference intensity over five consecutive windows is considered. Taking the moving average involves the following expression:

[0071] ;in, This is the smoothed predicted value of the interference intensity; For the first The original predicted value of the interference intensity of the window; These are the current time window index and the summation variable, respectively.

[0072] It should be explained that the moving average eliminates short-term noise interference by averaging the original predicted values ​​of five consecutive windows, which can make the interference intensity prediction results smoother and avoid misjudgment of subsequent control decisions due to instantaneous fluctuations. For example, if the predicted value of a certain window is abnormally high due to sudden pulse interference, the moving average can dilute the abnormality by using the normal values ​​of adjacent windows, ensuring that the output result is closer to the long-term trend of the real interference.

[0073] like If , then it will be truncated to 0;

[0074] If the type prediction changes continuously, such as Gaussian → Impulse → Gaussian, then the majority voting method is used to correct the result;

[0075] Final result Write to shared memory or a message queue.

[0076] In this embodiment of the invention, cross-scenario general parameters are obtained through meta-learning pre-training, and online fine-tuning is used to achieve millisecond-level adaptation to dynamic channels, ultimately outputting accurate interference prediction. This can significantly improve the model's adaptability to complex channels and the real-time performance of interference estimation, providing key support for the anti-disturbance capability of trajectory control. Through the synergistic cooperation of the above steps, the pre-training stage lays the foundation for cross-scenario generalization, the online fine-tuning stage quickly adapts to new scenarios, and the interference prediction stage outputs key compensation information, forming a closed-loop logic of pre-training, fine-tuning, and prediction, supporting the anti-interference requirements of trajectory control.

[0077] Interference prediction result analysis and dynamic resource adjustment module: This module analyzes the output interference prediction results and dynamically adjusts the resource allocation scheme based on the analysis results. Specific steps include:

[0078] When dynamically adjusting the scheme based on the predicted interference type and predicted interference intensity in the output interference prediction results, if and If so, then the scheme of adjusting only the transmission power will be implemented;

[0079] like and Then, a joint adjustment of power and beamforming vector scheme will be implemented; Indicates Gaussian interference; Indicates pulse interference; The threshold for weak interference intensity is preset based on system characteristics;

[0080] like If so, then the subcarrier hopping power control scheme is implemented;

[0081] It should be noted that the implementation of the scheme of adjusting only the transmit power, the scheme of jointly adjusting the power and beamforming vector, and the scheme of subcarrier hopping plus power control can all be implemented based on the existing adjustment schemes, and the specific adjustment content is not limited.

[0082] In this embodiment of the invention, by performing data analysis on the predicted interference type and predicted interference intensity in the interference prediction results, and dynamically adjusting and selecting the optimal strategy based on the analysis results, the suppression efficiency can be effectively improved and the communication quality of the system under complex interference scenarios can be significantly improved.

[0083] In the several embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for example, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0084] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0085] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0086] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the essential characteristics of the present invention.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A deep learning-based intelligent interference management system for wireless communication networks, characterized in that, include: Real-time channel state information processing module: Based on the multipath effect, a tapped delay line model is constructed. The effective path characteristics are dynamically updated by fusing mobility parameters. The effective path gain is adjusted by fusing environmental interference factors. Based on the fused complex gain, a real-time channel state information sequence containing spatiotemporal correlation features is generated. In the tapped delay line model based on multipath effects, multipath propagation is described by the tapped delay line, and the relevant expressions are as follows: ;in, is the multipath channel impulse response at time t; k is the effective path number. The number of valid paths; Let the complex gain of the k-th valid path be... , Let be the initial complex gain of the k-th valid path; This refers to the channel coherence time; The dot product of user speed and path direction; The propagation delay for the k-th valid path; For carrier frequency; Let j be the Dirac delta function, and let t be the effective path delay time; j is the imaginary unit; t is the continuous time variable. For phase offset terms; For time-based positioning items; When dynamically updating effective path features by fusing mobility parameters, the relevant expression is: ;in, Let $\frac{k}{k}$ be the rate of change of the delay of the k-th effective path over time. Let be the unit vector representing the arrival direction of the kth valid path; The instantaneous velocity vector of the user; The speed is the speed of light; the mobility parameter includes the user's instantaneous velocity vector; When adjusting the effective path gain by incorporating environmental interference factors, the relevant expression is: ;in, The complex gain of the k-th effective path after the environment is merged; Let be the rain attenuation coefficient at time t. ;in, This is the frequency-dependent attenuation coefficient, which is positively correlated with the carrier frequency; Let be the instantaneous rainfall rate at time t; Rainfall rate index; The length of the signal propagation path; It is a natural exponential function; Let be the obstacle occlusion coefficient at time t, with a value ranging from (0,1], and 1 when there is no occlusion. ;in, Free space path loss; The additional loss due to obstacles at time t; environmental interference factors include rain attenuation coefficient and obstacle occlusion coefficient; Meta-learning-driven learning processing module: Based on the generated real-time channel state information sequence, it obtains cross-scenario universal initialization parameters through meta-learning pre-training, and uses online incremental learning to quickly fine-tune new samples within milliseconds, outputting interference prediction results; Interference prediction result analysis and dynamic resource adjustment module: performs data analysis on the output interference prediction results and dynamically adjusts the resource allocation scheme based on the analysis results.

2. The intelligent interference management system for wireless communication networks based on deep learning according to claim 1, characterized in that, Constructing a multi-antenna channel state information matrix based on the fused complex gain M represents the number of transmit antennas, N represents the number of receive antennas, and C indicates that the matrix elements belong to the complex field. The combined complex gain is given by m, where m represents the m-th transmitting antenna, m∈{1, 2, ..., M}; and n represents the n-th receiving antenna, n∈{1, 2, ..., N}. The constructed multi-antenna channel state information matrix is ​​combined with the window length T to generate a channel state information sequence. .

3. The intelligent interference management system for wireless communication networks based on deep learning according to claim 2, characterized in that, Collect channel state information sequence sample sets under multiple scenarios ;in, For the scene The channel state information sequence; Total number of scenes; These are interference labels for the corresponding scenarios; Employing a model-independent meta-learning framework, the model It consists of a CNN-LSTM feature extractor and a fully connected prediction head; among which, The parameters are general initialization parameters; the CNN-LSTM feature extractor is used to capture the spatiotemporal features of channel state information; During meta-training, the general initialization parameters are updated by alternately optimizing the meta-loss. .

4. The intelligent interference management system for wireless communication networks based on deep learning according to claim 3, characterized in that, When using online incremental learning to rapidly fine-tune new samples within milliseconds, based on the generated real-time channel state information sequence... With general initialization parameters Using the initial model, a single-step gradient update is employed to adapt to new samples, resulting in online fine-tuned model parameters. .

5. The intelligent interference management system for wireless communication networks based on deep learning according to claim 4, characterized in that, Using the online fine-tuned model When reasoning about real-time channel state information sequences, the model is used. The CNN-LSTM feature extractor fuses spatial and temporal features on the preprocessed real-time channel state information sequence, with an output dimension of [missing information]. The global feature vector F; where, For feature dimensions; The global feature vector F is input into two parallel output branches: the interference type output branch and the interference intensity output branch.

6. The intelligent interference management system for wireless communication networks based on deep learning according to claim 5, characterized in that, Predict the type of interference by using the output branch based on the type of interference. The interference intensity is predicted by using the output branch of the interference intensity. The predicted interference type and predicted interference intensity are integrated into a structured output. .

7. The intelligent interference management system for wireless communication networks based on deep learning according to claim 6, characterized in that, When dynamically adjusting the scheme based on the predicted interference type and predicted interference intensity in the output interference prediction results, if and If so, then the scheme of adjusting only the transmission power will be implemented; like and Then, a joint adjustment of power and beamforming vector scheme will be implemented; Indicates Gaussian interference; Indicates pulse interference; The threshold for weak interference strength; like If so, a subcarrier hopping power control scheme is implemented.

Images