Wireless network adjacent cell interference optimization method and system based on self-supervised learning

Abstract

The invention relates to the technical field of wireless communication network optimization, in particular to a wireless network neighbor cell interference optimization method and system based on self-supervised learning. Federal learning and global model updating; performing deep reinforcement learning spectrum allocation decision; advanced wireless communication technology application; the block chain guarantees execution of the smart contract; closed-loop feedback optimization driven by user experience; the method has the beneficial effects that the interference mode between the cells is automatically identified and predicted through the self-supervised learning model, and the spectrum allocation strategy is dynamically optimized in combination with deep reinforcement learning, so that the signal interference between the cells is effectively reduced, and the overall stability of the network is improved. Especially in a high-density user scene, the system can respond to interference changes in real time, and the problem of performance reduction caused by adjacent cell interference is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communication network optimization, and in particular to a method and system for optimizing wireless network neighboring area interference based on self-supervised learning. Background Art

[0002] With the rapid development of mobile communication technology and the widespread deployment of 5G networks, the demand for wireless network coverage in dense urban areas has increased significantly, and the dense deployment of base stations and the shortage of spectrum resources have become common phenomena. However, in densely deployed cell networks, the problem of interference between neighboring cells has become increasingly serious. Traditional static spectrum allocation and interference control methods lack flexibility and are difficult to adapt to dynamically changing network environments, resulting in reduced network performance, low spectrum utilization, and poor user experience.

[0003] Existing neighboring cell interference management technologies mainly rely on centralized interference control solutions and fixed spectrum allocation strategies. This approach is usually difficult to respond to interference changes between different cells in a timely manner when faced with complex and dynamic traffic demands in dense urban areas, resulting in a decrease in network performance. In addition, traditional optimization methods usually rely on manual configuration and centralized management, which cannot fully utilize the edge computing capabilities of the cell and the feedback from terminal devices, making it difficult to achieve adaptive resource optimization.

[0004] With the development of new technologies such as edge computing, machine learning, and massive MIMO, network optimization based on self-supervised learning and reinforcement learning has gradually become a research hotspot. Self-supervised learning models can automatically identify and learn interference patterns through big data analysis, while deep reinforcement learning can dynamically adjust spectrum allocation strategies to make the use of spectrum resources more flexible. At the same time, blockchain technology has advantages in data transparency and security, and can maintain the fairness and reliability of the sharing process in the scenario of spectrum sharing between cells.

[0005] Therefore, there is an urgent need for an intelligent neighboring cell interference optimization method based on self-supervised learning. Summary of the invention

[0006] The purpose of the present invention is to provide a wireless network neighboring cell interference optimization method and system based on self-supervised learning, which monitors and predicts interference in real time through edge computing and federated learning models, and combines deep reinforcement learning to realize dynamic sharing and intelligent allocation of spectrum resources, thereby improving the spectrum utilization of the network, reducing neighboring cell interference, and ensuring user experience, so as to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: a method for optimizing wireless network neighboring area interference based on self-supervised learning, the method comprising the following steps:

[0008] Data collection and local preprocessing;

[0009] Federated learning and global model updates;

[0010] Deep reinforcement learning for spectrum allocation decision-making;

[0011] Application of advanced wireless communication technology;

[0012] Blockchain security and smart contract execution;

[0013] User experience driven closed-loop feedback optimization.

[0014] Preferably, the specific operations of data collection and local preprocessing include:

[0015] Data collection: The edge node collects the network status data of the cell in real time, including user location information, traffic load, interference level, signal strength and user demand type. The collected data will be updated regularly to ensure the timeliness and accuracy of the input information;

[0016] Data preprocessing: Noise filtering, anomaly detection and standardization are performed on the raw data to format the data into a data format suitable for model training and inference. For missing data, the system interpolates or completes historical data to ensure data integrity.

[0017] Local reasoning and response: Edge nodes use deployed self-supervised learning models to make real-time predictions on interference patterns and traffic demands, identify high-interference periods and hotspots, and make resource allocation decisions immediately if sudden high interference or traffic surges are detected to avoid impacts on the global network.

[0018] Preferably, the specific operations of federated learning and global model updating include:

[0019] Local model training: Each edge node independently trains a self-supervised learning model, and learns the interference pattern and traffic distribution characteristics of the cell through a large amount of unlabeled data. The local model can identify traffic fluctuations and interference changes, providing important data support for subsequent spectrum allocation;

[0020] Model aggregation and global model update: Edge nodes regularly upload model parameters to the central server, which aggregates the parameters of all nodes to generate a global model. Through federated learning, the updated global model can more accurately reflect the interference pattern and traffic trend of the entire network.

[0021] Differential privacy protection and secure communication: Differential privacy technology is used when uploading parameters to protect user sensitive data, and encrypted transmission is carried out through a secure channel to ensure data security.

[0022] Preferably, the specific operations of deep reinforcement learning spectrum allocation decision-making include:

[0023] Definition of state space and action space: The system maps the network state to the state space of reinforcement learning, including channel utilization, neighboring interference level and user demand. The action space is defined as various decisions on spectrum resource allocation, such as adjusting frequency bands, borrowing spectrum, and sharing resource operations.

[0024] Deep reinforcement learning model training: Use the deep reinforcement learning DRL model to learn the optimal spectrum allocation strategy with network performance as the reward function, and optimize the decision-making process through the DQN, PPO or A3C algorithm to ensure that the spectrum sharing strategy selected by the model improves network performance while reducing interference to neighboring cells;

[0025] Real-time dynamic scheduling: The DRL model automatically generates spectrum allocation strategies under real-time network status changes. When high load or interference is detected, the system flexibly allocates resources between cells. In the case of insufficient spectrum resources, the model also supports a soft spectrum sharing mechanism, that is, under certain conditions, a cell temporarily borrows idle spectrum from neighboring cells.

[0026] Preferably, the specific operations of the advanced wireless communication technology application include:

[0027] Massive MIMO and beamforming: beamforming is performed through massive MIMO technology to transmit signal energy to user terminals in a directionally controlled manner, reduce interference to neighboring cells, dynamically adjust beam direction and strength, adapt to user location changes and traffic requirements in real time, and improve isolation between cells; cognitive radio spectrum sensing and allocation: the cognitive radio module automatically detects the occupancy of spectrum resources and identifies available idle frequency bands. When the system detects that the spectrum is idle, it automatically enables dynamic spectrum sharing and temporarily borrows idle frequency bands to make more efficient use of spectrum resources; millimeter wave and visible light communication support: millimeter wave or visible light communication modules are deployed in high-traffic areas to provide additional bandwidth to divert traditional frequency band traffic. Millimeter wave and visible light communications have strong anti-interference characteristics and are suitable for high-demand short-distance data transmission;

[0028] The specific operations of blockchain security and smart contract execution include: blockchain records the spectrum sharing process: the resource allocation, borrowing and release of the spectrum sharing process are recorded on the blockchain to ensure data transparency and non-tamperability. The time, frequency band and cell information of each resource allocation are all on the chain to facilitate subsequent review and traceability; smart contracts automatically execute sharing agreements: using blockchain smart contracts, the protocol parameters and conditions of spectrum sharing between cells are set in the contract. When the spectrum sharing conditions are met, the smart contract automatically executes resource allocation to ensure that all parties execute the spectrum sharing agreement fairly; breach detection and penalty mechanism: the smart contract sets a detection mechanism for breach of contract in the sharing process. Once a breach of contract occurs, the system will trigger automatic penalties and record the breach of contract;

[0029] The specific operations of closed-loop feedback optimization driven by user experience include: real-time collection and analysis of QoE data: real-time collection of QoE data from user terminal devices or applications, including download speed, latency, video fluency and call quality. The collected QoE data will be uploaded to edge nodes regularly as a basis for spectrum sharing and interference optimization; QoE-based optimization feedback mechanism: QoE data is fed back to edge computing nodes, and spectrum allocation strategies are dynamically adjusted through self-supervised learning and deep reinforcement learning models. When it is detected that the user experience in a certain area has declined, the system will prioritize allocating more resources to that area; closed-loop feedback iterative optimization: after each adjustment of the spectrum allocation strategy, the system monitors user QoE data in real time to verify the optimization effect. If the user experience improvement is not significant, the system will make a secondary adjustment to form a closed-loop feedback mechanism to ensure continuous optimization with user experience as the core.

[0030] A wireless network neighboring area interference optimization system based on self-supervised learning is applied to a wireless network neighboring area interference optimization method based on self-supervised learning, and the system includes:

[0031] Edge computing layer, used for data collection and local preprocessing;

[0032] Federated self-supervised learning module, used for federated learning and global model update;

[0033] Deep reinforcement learning spectrum sharing module, used for deep reinforcement learning spectrum allocation decision;

[0034] Advanced wireless communication modules, used for advanced wireless communication technology applications;

[0035] Security and fairness guarantee module, used for blockchain security and smart contract execution;

[0036] The user experience-driven closed-loop feedback module is used for user experience-driven closed-loop feedback optimization.

[0037] Preferably, the edge computing layer includes:

[0038] Local data collection and preprocessing: Edge computing nodes are deployed at each cell base station to collect local network data in real time and preprocess the data to reduce the transmission burden;

[0039] Edge reasoning and response: Perform reasoning directly on edge nodes based on pre-trained models, identify high-interference periods and traffic peak network status, make quick resource allocation decisions, and improve system response speed.

[0040] Preferably, the federated self-supervised learning module includes:

[0041] Local model training: Each edge node independently trains a self-supervised learning model, uses local data to learn the interference pattern and traffic characteristics of the cell, and automatically extracts key network features through the unlabeled data learning ability of the self-supervised learning model;

[0042] Model aggregation and global update: Each node regularly uploads model parameters to the central server, and aggregates the model through federated learning to form a global model, which improves prediction accuracy and ensures user data privacy.

[0043] Preferably, the deep reinforcement learning spectrum sharing module includes:

[0044] Spectrum allocation strategy optimization: Based on deep reinforcement learning (DRL) technology, the optimal spectrum allocation strategy is dynamically learned using network performance indicators as reward functions.

[0045] Dynamic resource scheduling: The DRL model adjusts spectrum sharing and dynamic allocation between cells according to the real-time network status, flexibly allocates resources, reduces interference between cells, and maximizes spectrum resource utilization.

[0046] Preferably, the advanced wireless communication module includes: large-scale MIMO and beamforming: using large-scale MIMO technology to isolate signals in the airspace, and transmitting signal energy in a directional manner through beamforming, thereby reducing interference to neighboring areas and improving signal coverage quality; cognitive radio technology: the equipment has spectrum sensing capabilities, can automatically detect the occupancy of spectrum resources, and dynamically select idle frequency bands to efficiently utilize spectrum resources; millimeter wave and visible light communication support: deploying millimeter wave or visible light communication in high-traffic areas, providing additional bandwidth support, and diverting the traffic pressure of traditional frequency bands;

[0047] The security and fairness guarantee module includes: Blockchain technology application: Through the decentralized characteristics of blockchain, the resource allocation and transaction information in the spectrum sharing process are recorded to ensure that the data is transparent and cannot be tampered with, and to provide reliable records of the sharing process; Smart contract execution: Smart contracts automatically execute spectrum sharing agreements to ensure that communities comply with spectrum sharing rules and maintain fairness and cooperation;

[0048] The user experience-driven closed-loop feedback module includes: QoE data collection and feedback: collecting user experience quality QoE data from user terminals, including download speed, signal strength, video fluency, and call quality, as feedback data for optimizing the system; feedback optimization strategy: dynamically adjusting spectrum allocation and interference control strategies based on QoE data to form a user experience-driven closed-loop feedback mechanism to ensure that the system performs adaptive optimization with user experience as the core.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The wireless network neighboring cell interference optimization method and system based on self-supervised learning proposed in the present invention automatically identifies and predicts the interference pattern between cells through the self-supervised learning model, and dynamically optimizes the spectrum allocation strategy by combining deep reinforcement learning, effectively reducing the signal interference between cells and improving the overall stability of the network. Especially in high-density user scenarios, the system can respond to interference changes in real time and reduce the performance degradation caused by neighboring cell interference.

[0051] A dynamic spectrum sharing mechanism based on deep reinforcement learning is adopted to support flexible sharing and adjustment of spectrum resources between cells, significantly improving spectrum utilization. In addition, with the help of cognitive radio technology, the equipment can sense the idleness of the spectrum in real time and use the available spectrum during peak hours, alleviating the pressure of tight spectrum resources.

[0052] By combining edge computing with federated learning, the system has the ability to self-learn and adapt, and can intelligently adjust resource allocation according to the dynamic changes in network load and interference. This adaptive optimization capability not only reduces the need for manual intervention, but also effectively responds to complex changes in dense urban network environments, realizing intelligent optimization management of wireless networks.

[0053] The spectrum sharing process is recorded through blockchain technology, and the sharing agreement is automatically executed using smart contracts to ensure the transparency and immutability of the spectrum allocation process, thereby maintaining the fairness of resource allocation. This mechanism effectively prevents problems such as spectrum abuse, ensures the trust relationship between each cell in the spectrum sharing process, and promotes multi-party cooperation.

[0054] Through a closed-loop feedback system driven by user experience, the quality of user experience (QoE) is taken as the core optimization goal. The system collects QoE data in real time and dynamically adjusts spectrum allocation and interference management strategies to ensure continuous improvement of user experience. This optimization method centered on user experience has significantly improved user satisfaction with network use and enhanced the network's service quality.

[0055] Self-supervised learning and federated learning are used to achieve distributed data training, and model training and reasoning are performed directly on edge nodes, which protects user data privacy while reducing data transmission costs and delays. The model aggregation method of federated learning improves the accuracy of the global model, enables the system to better identify and respond to network interference, and ensures the efficiency and security of data utilization.

[0056] The integration of massive MIMO, beamforming, millimeter wave and visible light communication technologies has effectively improved network coverage and anti-interference performance. Directed signal transmission through massive MIMO and beamforming technologies reduces the incidence of interference in neighboring areas, while millimeter wave and visible light communications provide additional bandwidth for high-traffic areas, further improving the overall capacity and coverage quality of the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose and technical solution of the present invention clearly and completely described, and the advantages more clearly understood, the embodiments of the present invention are further described in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the present invention, rather than all of the embodiments, and are only used to explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] For example, see Figure 1 The present invention provides a technical solution: a method for optimizing wireless network neighboring area interference based on self-supervised learning, the method comprising the following steps:

[0060] 1. Data collection and local preprocessing

[0061] 1.1 Data Collection

[0062] Edge nodes (such as base stations) collect network status data for their cells in real time, including user location information, traffic load, interference level, signal strength, and user demand type (such as high-speed data or low-latency requirements). The collected data is updated regularly to ensure the timeliness and accuracy of the input information.

[0063] 1.2 Data Preprocessing

[0064] The raw data is filtered for noise, detected for anomalies, and standardized to format the data into a data format suitable for model training and inference. For missing data, the system ensures data integrity by interpolation or historical data completion.

[0065] 1.3 Local Reasoning and Response

[0066] The edge nodes use the deployed self-supervised learning model to make real-time predictions on interference patterns and traffic demand, and identify high-interference periods and hotspots. If sudden high interference or traffic surges are detected, the edge nodes make resource allocation decisions immediately to avoid impacting the global network.

[0067] 2. Federated Learning and Global Model Update

[0068] 2.1 Local Model Training

[0069] Each edge node independently trains a self-supervised learning model, and learns the interference pattern and traffic distribution characteristics of the cell through a large amount of unlabeled data. The local model can identify traffic fluctuations and interference changes, providing important data support for subsequent spectrum allocation.

[0070] 2.2 Model Aggregation and Global Model Update

[0071] Edge nodes regularly upload model parameters (such as gradients or weights) to the central server, which aggregates the parameters of all nodes to generate a global model. Through federated learning, the updated global model can more accurately reflect the interference patterns and traffic trends of the entire network.

[0072] 2.3 Differential Privacy Protection and Secure Communication

[0073] Differential privacy technology is used when uploading parameters to protect user sensitive data, and encrypted transmission is carried out through a secure channel to ensure data security.

[0074] 3. Deep reinforcement learning spectrum allocation decision

[0075] 3.1 Definition of State Space and Action Space

[0076] The system maps the network status to the state space of reinforcement learning, including channel utilization, neighboring interference level, user demand, etc. The action space is defined as various decisions on spectrum resource allocation, such as adjusting frequency bands, borrowing spectrum, sharing resources and other specific operations.

[0077] 3.2 Deep reinforcement learning model training

[0078] The deep reinforcement learning (DRL) model is used to learn the optimal spectrum allocation strategy with network performance (throughput, latency, and interference minimization) as the reward function. The decision-making process is optimized through algorithms such as DQN, PPO, or A3C to ensure that the spectrum sharing strategy selected by the model improves network performance while reducing interference in neighboring areas.

[0079] 3.3 Real-time dynamic scheduling

[0080] The DRL model automatically generates spectrum allocation strategies under real-time network status changes. When high load or interference is detected, the system flexibly allocates resources between cells. In the case of insufficient spectrum resources, the model also supports a soft spectrum sharing mechanism, that is, under certain conditions, a cell temporarily borrows idle spectrum from neighboring cells.

[0081] 4. Application of advanced wireless communication technology

[0082] 4.1 Massive MIMO and Beamforming

[0083] Through large-scale MIMO technology, beamforming is performed to transmit signal energy to user terminals in a directional manner, reducing interference to neighboring cells. The beam direction and strength are dynamically adjusted to adapt to user location changes and traffic requirements in real time, improving isolation between cells.

[0084] 4.2 Cognitive Radio Spectrum Sensing and Allocation

[0085] The cognitive radio module automatically detects the occupancy of spectrum resources and identifies available idle frequency bands. When the system detects that the spectrum is idle, it automatically enables dynamic spectrum sharing and temporarily borrows idle frequency bands to make more efficient use of spectrum resources.

[0086] 4.3 mmWave and visible light communication support

[0087] Deploy millimeter wave or visible light communication modules in high-traffic areas (such as shopping malls and transportation hubs) to provide additional bandwidth to divert traditional frequency band traffic. Millimeter wave and visible light communications have strong anti-interference characteristics and are suitable for high-demand short-distance data transmission.

[0088] 5. Blockchain security and smart contract execution

[0089] 5.1 Blockchain records spectrum sharing process

[0090] The resource allocation, borrowing and release of the spectrum sharing process are recorded on the blockchain to ensure data transparency and immutability. The time, frequency band, and cell information of both parties of each resource allocation are all recorded on the blockchain to facilitate subsequent review and traceability.

[0091] 5.2 Smart Contract Automatically Executes Sharing Agreement

[0092] Using blockchain smart contracts, the protocol parameters and conditions for spectrum sharing between cells are set in the contract. When the spectrum sharing conditions are met, the smart contract automatically executes resource allocation to ensure that all parties implement the spectrum sharing agreement fairly.

[0093] 5.3 Breach Detection and Penalty Mechanism

[0094] The smart contract sets up a detection mechanism for breach of contract (such as spectrum abuse) during the sharing process. Once a breach occurs, the system will trigger automatic penalties and record the breach.

[0095] 6. User experience driven closed-loop feedback optimization

[0096] 6.1 Real-time Collection and Analysis of QoE Data

[0097] Collect QoE data in real time from user terminal devices or applications, including download speed, latency, video fluency, call quality, etc. The collected QoE data will be uploaded to edge nodes regularly as a basis for spectrum sharing and interference optimization.

[0098] 6.2 QoE-based Optimization Feedback Mechanism

[0099] The QoE data is fed back to the edge computing nodes, and the spectrum allocation strategy is dynamically adjusted through self-supervised learning and deep reinforcement learning models. For example, when it is detected that the user experience in a certain area is degraded (such as signal weakening or increased lag), the system will prioritize allocating more resources to that area.

[0100] 6.3 Closed-loop feedback iterative optimization

[0101] After each adjustment of the spectrum allocation strategy, the system monitors user QoE data in real time to verify the optimization effect. If the user experience does not improve significantly, the system will make a second adjustment to form a closed-loop feedback mechanism to ensure continuous optimization with user experience as the core.

[0102] Embodiment 2, based on embodiment 1, proposes a wireless network neighboring cell interference optimization method system based on self-supervised learning, which includes the following main modules: edge computing layer, federated self-supervised learning module, deep reinforcement learning spectrum sharing module, advanced wireless communication module, security and fairness guarantee module and user experience driven closed-loop feedback module. Each module works together to achieve intelligent management of neighboring cell interference optimization and spectrum sharing.

[0103] Edge computing layer

[0104] Local data collection and preprocessing: Edge computing nodes are deployed at each cell base station to collect local network data (such as signal strength, interference level, user location, traffic demand, etc.) in real time, and preprocess the data to reduce the transmission burden.

[0105] Edge reasoning and response: Based on the pre-trained model, reasoning is performed directly at the edge node to identify network conditions such as high interference periods and traffic peaks, and resource allocation decisions can be made quickly to improve the system's response speed.

[0106] Federated Self-Supervised Learning Module

[0107] Local model training: Each edge node independently trains a self-supervised learning model, using local data to learn the interference pattern and traffic characteristics of the cell. Through the unlabeled data learning ability of the self-supervised learning model, key network features are automatically extracted.

[0108] Model aggregation and global update: Each node regularly uploads model parameters to the central server, and aggregates the model through federated learning to form a global model, which improves prediction accuracy and ensures user data privacy.

[0109] Deep reinforcement learning spectrum sharing module

[0110] Spectrum allocation strategy optimization: Based on deep reinforcement learning (DRL) technology, network performance indicators (such as throughput, latency, and interference level) are used as reward functions to dynamically learn the optimal spectrum allocation strategy.

[0111] Dynamic resource scheduling: The DRL model adjusts spectrum sharing and dynamic allocation between cells according to the real-time network status, flexibly allocates resources, reduces interference between cells, and maximizes spectrum resource utilization.

[0112] Advanced wireless communication module

[0113] Massive MIMO and beamforming: Use massive MIMO technology to isolate signals in the spatial domain and transmit signal energy in a direction through beamforming, thereby reducing interference to neighboring cells and improving signal coverage quality.

[0114] Cognitive radio technology: The device has spectrum perception capabilities, can automatically detect the occupancy of spectrum resources, and dynamically select idle frequency bands to efficiently utilize spectrum resources.

[0115] Millimeter wave and visible light communication support: Deploy millimeter wave or visible light communication in high-traffic areas to provide additional bandwidth support and divert traffic pressure from traditional frequency bands.

[0116] Safety and fairness guarantee module

[0117] Application of blockchain technology: Through the decentralized characteristics of blockchain, resource allocation and transaction information in the spectrum sharing process are recorded to ensure that the data is transparent and cannot be tampered with, and provide reliable records of the sharing process.

[0118] Smart contract execution: Smart contracts automatically execute spectrum sharing agreements, ensuring that communities comply with spectrum sharing rules and maintain fairness and cooperation.

[0119] User experience driven closed-loop feedback module

[0120] QoE data collection and feedback: Collect user experience quality (QoE) data from user terminals, including download speed, signal strength, video fluency, call quality, etc., as feedback data for optimizing the system.

[0121] Feedback optimization strategy: Dynamically adjust spectrum allocation and interference control strategies based on QoE data to form a closed-loop feedback mechanism driven by user experience, ensuring that the system performs adaptive optimization with user experience as the core.

[0122] Embodiment 3, based on embodiment 2, proposes the following contents:

[0123] 1. Edge computing layer design

[0124] 1.1 Local data collection and preprocessing

[0125] Data type: including user location information, traffic load, interference level, signal strength, user demand type (such as high-speed data, low-latency connection), etc.

[0126] Real-time collection: Edge computing nodes are deployed at each cell base station to collect data related to network status in real time to ensure the timeliness of the data.

[0127] Data preprocessing: Perform noise filtering, anomaly detection, and standardization on the collected data, and format the data for training and prediction to improve the efficiency and accuracy of model training.

[0128] 1.2 Edge Reasoning and Real-time Response

[0129] Model deployment: Deploy self-supervised learning models and reinforcement learning models on edge nodes for local reasoning and real-time response.

[0130] Real-time response: The model predicts interference patterns and traffic peaks in real time. Once high interference or traffic surge is detected in the cell, the edge node immediately makes resource allocation optimization decisions to improve response speed.

[0131] Local caching: Frequently used model parameters and important data are cached at the edge to reduce the frequency of central transmission and improve resource utilization efficiency.

[0132] 2. Design of Federated Self-Supervised Learning Module

[0133] 2.1 Local Model Training

[0134] Self-supervised learning: Based on a large amount of unlabeled data, the self-supervised learning model automatically learns the interference patterns and traffic demand characteristics between cells. By predicting changes in network status (such as load peaks and interference areas), the system can perceive potential problems in advance.

[0135] Unlabeled data learning: Use unlabeled data for multi-task self-supervised learning to obtain features such as user behavior and signal change trends, which are used as basic data support for spectrum sharing and interference optimization.

[0136] 2.2 Model Aggregation for Federated Learning

[0137] Parameter aggregation: The model parameters (such as gradients or weights) trained by each edge node are regularly uploaded to the central server. The server aggregates the parameters of each node to generate a global model to avoid direct uploading of private data.

[0138] Model update frequency: Set different update frequencies according to network status and data distribution to ensure that the global model can adapt to the dynamic changes of the network in the region and ensure the accuracy of learning.

[0139] 2.3 Model Security and Privacy Protection

[0140] Differential privacy protection: Differential privacy processing is used on uploaded model parameters to prevent sensitive data leakage and ensure user privacy.

[0141] Model security isolation: Encrypt communication between edge nodes and central servers, transmit model parameters through secure channels, and prevent potential network attacks.

[0142] 3. Design of spectrum sharing module based on deep reinforcement learning

[0143] 3.1 Deep reinforcement learning spectrum allocation strategy

[0144] State definition: Map the real-time state of the system (such as channel utilization, neighboring interference level, user demand) into the state space of reinforcement learning.

[0145] Action space: includes specific operations such as spectrum resource allocation, frequency band adjustment, and spectrum sharing between cells. The DRL model selects the optimal action based on different states.

[0146] Reward function: The reward function is designed with the goals of maximizing network throughput, minimizing neighboring cell interference, and maximizing spectrum utilization, and the effectiveness of the decision is evaluated in real time.

[0147] 3.2 Dynamic Resource Scheduling

[0148] Policy training: Spectrum allocation strategy training is performed through deep Q learning (DQN) or policy gradient-based algorithms (such as PPO and A3C) to learn the optimal dynamic resource allocation method.

[0149] Real-time spectrum allocation: When the DRL model detects changes in network status, it dynamically adjusts the allocation of spectrum resources to ensure maximum spectrum utilization and minimize interference.

[0150] Soft spectrum sharing mechanism: allows cells to temporarily borrow idle spectrum from neighboring cells when resources are insufficient to meet peak demand and ensure flexible use of spectrum resources.

[0151] 4. Advanced wireless communication module design

[0152] 4.1 Massive MIMO and Beamforming

[0153] Beamforming strategy: Use massive MIMO to perform beamforming so that signal energy is transmitted in a directionally controlled manner, reducing interference to neighboring cells.

[0154] Beam adaptive adjustment: Based on real-time traffic and user location distribution, the beam direction and strength are dynamically adjusted to minimize interference and improve signal quality and coverage.

[0155] 4.2 Cognitive Radio Technology

[0156] Spectrum sensing and detection: The device has spectrum sensing capabilities, automatically detects the usage status of surrounding frequency bands, and identifies idle spectrum resources.

[0157] Dynamic frequency band selection: When it is detected that the channel load of the neighboring area is lower than a certain threshold, the system can temporarily borrow idle frequency bands to improve spectrum utilization.

[0158] 4.3 mmWave and visible light communication support

[0159] High-density data traffic offload: In hot spots with high traffic demand, millimeter wave and visible light communications are supported to provide additional bandwidth to alleviate the pressure on traditional frequency bands.

[0160] Interference-free high-frequency communication: Millimeter wave and visible light technologies have strong anti-interference characteristics, suitable for short-distance communications in high-demand areas, and improve the network load carrying capacity.

[0161] 5. Security and fairness guarantee module design

[0162] 5.1 Blockchain technology ensures transparency of spectrum sharing

[0163] Spectrum sharing records: The spectrum sharing process between cells is recorded through blockchain, including resource borrowing and release, usage time, etc., to ensure the transparency of the sharing process.

[0164] Unalterable transaction records: Once shared records are recorded on the blockchain, they cannot be tampered with, ensuring the reliability and security of spectrum allocation.

[0165] 5.2 Automatic execution of smart contracts

[0166] Smart contract deployment: Based on blockchain smart contracts, define the spectrum sharing agreement between cells, and set frequency band borrowing conditions and default handling mechanisms.

[0167] Automated execution and supervision: When the spectrum demand between cells reaches the set conditions, the smart contract automatically executes the spectrum borrowing agreement to ensure transparent and fair execution of the rules.

[0168] 6. User experience driven closed-loop feedback module design

[0169] 6.1 Real-time Collection of User Experience Data

[0170] QoE data type: collects user-side experience quality data such as download speed, latency, signal strength, video fluency, and call quality as optimization feedback data.

[0171] Terminal feedback mechanism: Regularly feed back QoE data to edge computing nodes through applications or terminal devices to provide a basis for optimization decisions.

[0172] 6.2 QoE-based Feedback Optimization

[0173] User experience driven optimization: Based on QoE data, the system dynamically adjusts spectrum allocation and interference control strategies to ensure optimization centered on user experience.

[0174] Closed-loop feedback mechanism: After each optimization strategy, the optimization effect is fed back through QoE data to continuously improve the effectiveness of the model and realize a closed-loop feedback system driven by user experience.

[0175] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A wireless network neighboring cell interference optimization method based on self-supervised learning, characterized in that: The method comprises the following steps: Data collection and local preprocessing; Federated learning and global model updates; Deep reinforcement learning for spectrum allocation decision-making; Application of advanced wireless communication technology; Blockchain security and smart contract execution; User experience driven closed-loop feedback optimization.

2. The method for optimizing wireless network neighboring area interference based on self-supervised learning according to claim 1, characterized in that: The specific operations of data collection and local preprocessing include: Data collection: The edge node collects the network status data of the cell in real time, including user location information, traffic load, interference level, signal strength and user demand type. The collected data will be updated regularly to ensure the timeliness and accuracy of the input information; Data preprocessing: Noise filtering, anomaly detection and standardization are performed on the raw data to format the data into a data format suitable for model training and inference. For missing data, the system interpolates or completes historical data to ensure data integrity. Local reasoning and response: Edge nodes use deployed self-supervised learning models to make real-time predictions on interference patterns and traffic demands, identify high-interference periods and hotspots, and make resource allocation decisions immediately if sudden high interference or traffic surges are detected to avoid impacts on the global network.

3. The method for optimizing wireless network neighboring area interference based on self-supervised learning according to claim 1, characterized in that: The specific operations of federated learning and global model update include: Local model training: Each edge node independently trains a self-supervised learning model, and learns the interference pattern and traffic distribution characteristics of the cell through a large amount of unlabeled data. The local model can identify traffic fluctuations and interference changes, providing important data support for subsequent spectrum allocation; Model aggregation and global model update: Edge nodes regularly upload model parameters to the central server, which aggregates the parameters of all nodes to generate a global model. Through federated learning, the updated global model can more accurately reflect the interference pattern and traffic trend of the entire network. Differential privacy protection and secure communication: Differential privacy technology is used when uploading parameters to protect user sensitive data, and encrypted transmission is carried out through a secure channel to ensure data security.

4. The method for optimizing wireless network neighboring area interference based on self-supervised learning according to claim 1, characterized in that: The specific operations of deep reinforcement learning spectrum allocation decision-making include: Definition of state space and action space: The system maps the network state to the state space of reinforcement learning, including channel utilization, neighboring interference level and user demand. The action space is defined as various decisions on spectrum resource allocation, such as adjusting frequency bands, borrowing spectrum, and sharing resource operations. Deep reinforcement learning model training: Use the deep reinforcement learning DRL model to learn the optimal spectrum allocation strategy with network performance as the reward function, and optimize the decision-making process through the DQN, PPO or A3C algorithm to ensure that the spectrum sharing strategy selected by the model improves network performance while reducing interference to neighboring cells; Real-time dynamic scheduling: The DRL model automatically generates spectrum allocation strategies under real-time network status changes. When high load or interference is detected, the system will flexibly allocate resources between cells. In the case of insufficient spectrum resources, the model also supports a soft spectrum sharing mechanism, that is, under certain conditions, a cell temporarily borrows idle spectrum from neighboring cells.

5. The method for optimizing wireless network neighboring area interference based on self-supervised learning according to claim 1, characterized in that: Specific operations of advanced wireless communication technology applications include: Massive MIMO and beamforming: beamforming is performed through massive MIMO technology to transmit signal energy to user terminals in a directionally controlled manner, reduce interference to neighboring cells, dynamically adjust beam direction and strength, adapt to user location changes and traffic requirements in real time, and improve isolation between cells; cognitive radio spectrum sensing and allocation: the cognitive radio module automatically detects the occupancy of spectrum resources and identifies available idle frequency bands. When the system detects that the spectrum is idle, it automatically enables dynamic spectrum sharing and temporarily borrows idle frequency bands to make more efficient use of spectrum resources; millimeter wave and visible light communication support: millimeter wave or visible light communication modules are deployed in high-traffic areas to provide additional bandwidth to divert traditional frequency band traffic. Millimeter wave and visible light communications have strong anti-interference characteristics and are suitable for high-demand short-distance data transmission; The specific operations of blockchain security and smart contract execution include: blockchain records the spectrum sharing process: the resource allocation, borrowing and release of the spectrum sharing process are recorded on the blockchain to ensure data transparency and non-tamperability. The time, frequency band and cell information of each resource allocation are all on the chain to facilitate subsequent review and traceability; smart contracts automatically execute sharing agreements: using blockchain smart contracts, the protocol parameters and conditions of spectrum sharing between cells are set in the contract. When the spectrum sharing conditions are met, the smart contract automatically executes resource allocation to ensure that all parties execute the spectrum sharing agreement fairly; breach detection and penalty mechanism: the smart contract sets a detection mechanism for breach of contract in the sharing process. Once a breach of contract occurs, the system will trigger automatic penalties and record the breach of contract; The specific operations of closed-loop feedback optimization driven by user experience include: real-time collection and analysis of QoE data: real-time collection of QoE data from user terminal devices or applications, including download speed, latency, video fluency and call quality. The collected QoE data will be uploaded to edge nodes regularly as a basis for spectrum sharing and interference optimization; QoE-based optimization feedback mechanism: QoE data is fed back to edge computing nodes, and spectrum allocation strategies are dynamically adjusted through self-supervised learning and deep reinforcement learning models. When it is detected that the user experience in a certain area has declined, the system will prioritize allocating more resources to that area; closed-loop feedback iterative optimization: after each adjustment of the spectrum allocation strategy, the system monitors user QoE data in real time to verify the optimization effect. If the user experience improvement is not significant, the system will make a secondary adjustment to form a closed-loop feedback mechanism to ensure continuous optimization with user experience as the core.

6. A wireless network neighboring cell interference optimization system based on self-supervised learning, applied to the wireless network neighboring cell interference optimization method based on self-supervised learning as described in any one of claims 1 to 5, characterized in that: The system comprises: Edge computing layer, used for data collection and local preprocessing; Federated self-supervised learning module, used for federated learning and global model update; Deep reinforcement learning spectrum sharing module, used for deep reinforcement learning spectrum allocation decision; Advanced wireless communication modules, used for advanced wireless communication technology applications; Security and fairness guarantee module, used for blockchain security and smart contract execution; The user experience-driven closed-loop feedback module is used for user experience-driven closed-loop feedback optimization.

7. The wireless network neighboring area interference optimization system based on self-supervised learning according to claim 6, characterized in that: The edge computing layer includes: Local data collection and preprocessing: Edge computing nodes are deployed at each cell base station to collect local network data in real time and preprocess the data to reduce the transmission burden; Edge reasoning and response: Perform reasoning directly on edge nodes based on pre-trained models, identify high-interference periods and traffic peak network status, make quick resource allocation decisions, and improve system response speed.

8. The wireless network neighboring area interference optimization system based on self-supervised learning according to claim 6, characterized in that: The federated self-supervised learning module includes: Local model training: Each edge node independently trains a self-supervised learning model, uses local data to learn the interference pattern and traffic characteristics of the cell, and automatically extracts key network features through the unlabeled data learning ability of the self-supervised learning model; Model aggregation and global update: Each node regularly uploads model parameters to the central server, and aggregates the model through federated learning to form a global model, which improves prediction accuracy and ensures user data privacy.

9. The wireless network neighboring area interference optimization system based on self-supervised learning according to claim 6, characterized in that: The deep reinforcement learning spectrum sharing module includes: Spectrum allocation strategy optimization: Based on deep reinforcement learning (DRL) technology, the optimal spectrum allocation strategy is dynamically learned using network performance indicators as reward functions. Dynamic resource scheduling: The DRL model adjusts spectrum sharing and dynamic allocation between cells according to the real-time network status, flexibly allocates resources, reduces interference between cells, and maximizes spectrum resource utilization.

10. The wireless network neighboring area interference optimization system based on self-supervised learning according to claim 6, characterized in that: The advanced wireless communication modules include: Massive MIMO and beamforming: using massive MIMO technology to isolate signals in the airspace, and using beamforming to transmit signal energy in a directional manner, reducing interference to neighboring areas and improving signal coverage quality; cognitive radio technology: the equipment has spectrum sensing capabilities, can automatically detect the occupancy of spectrum resources, and dynamically select idle frequency bands to efficiently utilize spectrum resources; millimeter wave and visible light communication support: deploying millimeter wave or visible light communication in high-traffic areas, providing additional bandwidth support, and diverting the traffic pressure of traditional frequency bands; The security and fairness guarantee module includes: Blockchain technology application: Through the decentralized characteristics of blockchain, the resource allocation and transaction information in the spectrum sharing process are recorded to ensure that the data is transparent and cannot be tampered with, and to provide reliable records of the sharing process; Smart contract execution: Smart contracts automatically execute spectrum sharing agreements to ensure that communities comply with spectrum sharing rules and maintain fairness and cooperation; The user experience-driven closed-loop feedback module includes: QoE data collection and feedback: collecting user experience quality QoE data from user terminals, including download speed, signal strength, video fluency, and call quality, as feedback data for optimizing the system; feedback optimization strategy: dynamically adjusting spectrum allocation and interference control strategies based on QoE data to form a user experience-driven closed-loop feedback mechanism to ensure that the system performs adaptive optimization with user experience as the core.

Images