A signal shield management method combining neural networks and blockchain

Abstract

The application provides a signal shielding device management method combining a neural network and a blockchain in the technical field of artificial intelligence and communication, and comprises the following steps: S1, a mobile terminal performs an authentication operation with a server through a preset frequency band, and sends a shielding instruction to a signal shielding device through the server; S2, the signal shielding device analyzes the shielding instruction to obtain real-time shielding parameters, collects real-time wireless communication signals, inputs the real-time wireless communication signals and the real-time shielding parameters into a shielding strategy generation model to obtain a real-time signal shielding strategy, generates an interference signal based on the real-time signal shielding strategy, and transmits the interference signal outward; S3, real-time shielding logs are recorded in real time and input into a risk assessment model to obtain a risk assessment report, so as to dynamically adjust the real-time signal shielding strategy; and S4, the real-time shielding logs and the risk assessment report are uploaded to a blockchain. The application has the advantages that the convenience, reliability and safety of the signal shielding device are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence and communication technology, and in particular to a signal jammer management method that combines neural networks and blockchain. Background Technology

[0002] In the context of the rapid development of modern communication technology, radio signal jammers (hereinafter referred to as jammers) are widely used in various fields as a key security and management device, including sensitive areas such as educational examination centers, prison management, and conference venues, to maintain order and security. Their core function is to prevent unauthorized communication by blocking the transmission of radio signals at specific frequencies, thereby maintaining order and security. For example, in college entrance examination venues, jammers can effectively prevent candidates from using mobile phones to cheat; in prison environments, they can block illegal communications by prisoners, reducing security risks.

[0003] Signal jammers work by emitting a signal with the same frequency but higher intensity as that of electronic devices (such as mobile phones). During operation, this signal scans at a certain speed from the low-frequency band to the high-frequency band of the forward channel (transmitted by the base station and received by the electronic device). This interference prevents the electronic device from detecting normal data transmitted from the base station, thus preventing it from establishing a connection and resulting in a no-signal, no-service state. Although signal jammers play an important role in specific scenarios, traditional signal jammers still have the following drawbacks:

[0004] 1. Limited functionality and lack of intelligent features:

[0005] Traditional signal jammers are fixed-function devices lacking adaptive or intelligent control capabilities. Their operating mode relies on manual adjustment, requiring users to select the target frequency band (e.g., blocking only 4G signals or the entire frequency band) via physical switches or buttons. For example, in conference venues, the jamming range needs to be manually set to avoid interfering with critical communications (such as security walkie-talkies). However, manual operation is inefficient; users may overlook emerging frequency bands (such as the 28GHz millimeter-wave band for 5G) or choose the full-band jamming mode to simplify the process. This crude jamming can lead to serious false jamming problems. For example, in an examination room environment, full-band jamming may disrupt legitimate communications for teachers or invigilators, delaying emergency responses. In urban applications, full-band jamming may interfere with public safety networks (such as police or fire department communications). Furthermore, traditional signal jammers cannot dynamically adjust parameters according to real-time environmental conditions (such as changes in signal strength or device density), resulting in unstable jamming effects. For example, in areas with high-density devices, insufficient scanning speed may allow some devices to "escape" interference.

[0006] 2. Cumbersome operation and lack of convenience:

[0007] Manually adjusting the operating mode is not only time-consuming but also increases the risk of human error. Users need to operate repeatedly to cope with different scenarios (such as switching from an exam scenario to a prison scenario), which involves complex frequency band selection menus, making the process cumbersome and prone to errors. In dynamic environments (such as large-scale events), frequent intervention is required to avoid signal blind spots or excessive blocking. This inefficient operation often forces users to adopt a full-band blocking mode, but this mode consumes a lot of energy and has limited coverage, which cannot meet the needs of large areas. More importantly, traditional signal jammers lack remote management functions. All adjustments must be made locally, and centralized control through networks or cloud platforms is not possible. This limits their application in distributed systems (such as multi-exam-room networking) and increases operation and maintenance costs.

[0008] 3. Security risks and authorization issues:

[0009] Traditional signal jammers are easily abused or used without authorization, posing a significant security risk. Due to their simple design and lack of authentication or access control mechanisms, unauthorized users can easily obtain and deploy them for malicious purposes. For example, in public places, hackers may use jammers to block emergency communications or cooperate with eavesdropping devices to carry out eavesdropping attacks. In commercial espionage, unauthorized jamming may interfere with corporate networks, leading to data leaks or service interruptions.

[0010] In summary, traditional signal jammers have significant shortcomings in terms of convenience, reliability, and security: their manual operation mode leads to inefficiency and false jamming, the lack of intelligence limits their adaptive capabilities, and security vulnerabilities amplify the risk of misuse. Therefore, how to provide a signal jammer management method that combines neural networks and blockchain to improve the convenience, reliability, and security of signal jammers has become an urgent technical problem to be solved. Summary of the Invention

[0011] The technical problem to be solved by this invention is to provide a signal jammer management method that combines neural networks and blockchain, thereby improving the convenience, reliability and security of signal jammers.

[0012] This invention provides a signal jammer management method combining neural networks and blockchain, comprising the following steps:

[0013] Step S10: Create a shielding strategy generation model for outputting a signal shielding strategy based on wireless communication signals and shielding parameters, and a risk assessment model for outputting a risk assessment report based on shielding logs. Set the strategy loss function of the shielding strategy generation model and the risk loss function of the risk assessment model.

[0014] Step S20: Obtain a large number of historical blocking logs and historical risk assessment reports, and construct a dataset after preprocessing and labeling each of the historical blocking logs and historical risk assessment reports; the historical blocking logs include at least historical wireless communication signals, historical blocking parameters, historical signal blocking strategies, historical blocking times, historical operation accounts, and historical blocking locations;

[0015] Step S30: Train and compress the shielding strategy generation model and risk assessment model using the dataset, and deploy the trained shielding strategy generation model and risk assessment model to the signal jammer;

[0016] Step S40: After the signal jammer is powered on, the mobile terminal performs an authentication operation with the server through a preset frequency band. After successful authentication, the server sends a jamming command to the signal jammer. The server generates a jammer operation record based on the jamming command, encrypts the jammer operation record into an encrypted record, and uploads it to the blockchain.

[0017] Step S50: The signal jammer verifies and parses the received jamming command to obtain real-time jamming parameters, collects real-time wireless communication signals, inputs the real-time wireless communication signals and real-time jamming parameters into the deployed jamming strategy generation model to obtain a real-time signal jamming strategy, generates interference signals based on the real-time signal jamming strategy and transmits them to the outside to perform signal jamming operations.

[0018] Step S60: The signal jammer records a real-time jamming log that includes at least the real-time wireless communication signal, real-time jamming parameters, real-time signal jamming strategy, real-time jamming time, real-time operation account, and real-time jamming location. The real-time jamming log is input into the deployed risk assessment model to obtain a real-time risk assessment report. The real-time signal jamming strategy is dynamically adjusted based on the real-time risk assessment report.

[0019] Step S70: The signal jammer encrypts and uploads the real-time jamming log and the real-time risk assessment report to the server. The server encrypts the received real-time jamming log and the real-time risk assessment report into encrypted logs and encrypted reports respectively and uploads them to the blockchain.

[0020] Step S80: The mobile terminal obtains the encrypted records, encrypted logs, and encrypted reports through the blockchain, and manages the signal jammer based on the encrypted records, encrypted logs, and encrypted reports.

[0021] The advantages of this invention are:

[0022] 1. A shielding strategy generation model is created to output a signal shielding strategy based on wireless communication signals and shielding parameters, and a risk assessment model is created to output a risk assessment report based on shielding logs. A large amount of historical shielding logs and historical risk assessment reports are acquired to construct a dataset. The shielding strategy generation model and the risk assessment model are trained and compressed using this dataset, and then deployed to the signal jammer. After the signal jammer is powered on, the mobile terminal performs authentication with the server via a preset frequency band. After successful authentication, the server sends a shielding command to the signal jammer. The server generates a jammer operation record based on the shielding command, encrypts the operation record, and uploads it to the blockchain. The signal jammer verifies and parses the shielding command to obtain real-time shielding parameters, collects real-time wireless communication signals, inputs the real-time wireless communication signals and real-time shielding parameters into the deployed shielding strategy generation model to obtain a real-time signal shielding strategy, generates interference signals based on the real-time signal shielding strategy, and transmits them externally to perform signal shielding operations. The real-time records of the signal jammer include at least the real-time wireless communication signal, real-time shielding parameters, real-time signal shielding strategy, real-time shielding time, and real-time operation account. The real-time blocking logs of the real-time blocking location are input into the risk assessment model to obtain a real-time risk assessment report. Based on the real-time risk assessment report, the real-time signal blocking strategy is dynamically adjusted. The real-time blocking logs and real-time risk assessment reports are encrypted and uploaded to the server. The server encrypts the real-time blocking logs and real-time risk assessment reports into encrypted logs and encrypted reports, respectively, and uploads them to the blockchain. The mobile terminal obtains the encrypted records, encrypted logs, and encrypted reports through the blockchain to manage the signal jammer. That is, the mobile terminal sends blocking commands to the signal jammer through the server, realizing remote management of the signal jammer without local operation. Through a pre-trained and deployed blocking strategy generation model, a real-time signal blocking strategy is automatically generated based on the input real-time wireless communication signal and real-time blocking parameters. Different blocking modes or blocking scenarios can be selected as needed, simplifying the operation process and reducing the risk of human error. Through the authentication operation between the mobile terminal and the server, the risk assessment of the risk assessment model, and the blockchain storage of the jammer operation records, real-time blocking logs, and real-time risk assessment reports, the signal jammer is prevented from being easily abused or used without authorization, and it is easy to trace the source later. Ultimately, it greatly improves the convenience, reliability, and security of the signal jammer.

[0023] 2. The shielding strategy generation model constructed through neural networks can automatically output signal shielding strategies based on real-time wireless communication signals and shielding parameters, reducing manual intervention. The shielding strategy generation model is trained using historical data, which improves the accuracy and adaptability of signal shielding strategy generation. Combined with feedback from real-time shielding logs, the signal shielding strategy is dynamically adjusted to cope with environmental changes (such as changes in signal interference sources), thereby optimizing the shielding effect, reducing the risk of false shielding, and improving the overall system efficiency.

[0024] 3. The risk assessment model outputs a real-time risk assessment report based on real-time shielding logs (including real-time wireless communication signals, real-time shielding parameters, real-time signal shielding strategies, real-time shielding time, real-time operation accounts, and real-time shielding locations). This report is used to dynamically adjust real-time signal shielding strategies, achieving closed-loop control and enabling timely detection of potential problems (such as unauthorized operations or excessive shielding) to proactively reduce security risks. Furthermore, the risk assessment model is trained using historical data, enhancing the reliability and predictive ability of risk assessment and avoiding the lag inherent in traditional methods that rely on post-event analysis.

[0025] 4. All jammer operation records, real-time jamming logs, and real-time risk assessment reports are encrypted and stored using blockchain technology (encrypted records, encrypted logs, and encrypted reports), and access management is achieved through mobile terminals; that is, the distributed ledger characteristics of blockchain are used to ensure that the data is immutable and traceable, preventing unauthorized access or data tampering (e.g., operation account and jamming location records), meeting the needs of high-security scenarios; at the same time, the encrypted upload process enhances the security of data transmission.

[0026] 5. By training and compressing the shielding strategy generation model and risk assessment model before deploying them to the signal jammer, reliance on the cloud (server) is reduced. At the same time, using preset frequency bands for authentication optimizes communication efficiency and allows the signal jammer to communicate with the server while it is working. Model compression reduces hardware resource requirements (such as computing and storage), making it suitable for resource-constrained signal jammers. Local processing reduces network latency, improves response speed, and overall reduces deployment and maintenance costs, while improving system scalability.

[0027] 6. Mobile terminals acquire data (encrypted records, encrypted logs, and encrypted reports) through blockchain, enabling remote monitoring and management of signal jammers. This simplifies user operations, supports managing signal jammers anytime, anywhere, and effectively improves user experience. Combined with server authentication, it ensures that the management process is secure and controllable, making it suitable for distributed or multi-device scenarios.

[0028] 7. A dataset is built based on a large number of historical blocking logs and historical risk assessment reports to train and optimize neural network models (blocking strategy generation model and risk assessment model). The neural network model can learn complex patterns (such as signal characteristics in different environments), improve generalization ability, and ensure robustness in new scenarios (such as different geographical locations or signal types). This is more flexible than rule-based blocking methods and reduces the need for manual configuration.

[0029] 8. The real-time blocking log records cover multi-dimensional parameters (real-time wireless communication signal, real-time blocking parameters, real-time signal blocking strategy, real-time blocking time, real-time operation account, and real-time blocking location) and is uploaded to the blockchain, providing complete operation audit trails, facilitating post-event analysis, report generation, and compliance checks. The transparency of the blockchain ensures that all records are verifiable.

[0030] 9. By integrating neural network and blockchain technologies, intelligent, secure, and efficient management of signal jammers is achieved: Signal jamming strategies are dynamically generated and optimized through neural network models (jamming strategy generation model and risk assessment model), significantly improving the accuracy and environmental adaptability of signal jamming; blockchain is used to encrypt and distribute the storage of jammer operation records, real-time jamming logs, and real-time risk assessment reports, ensuring data immutability and full traceability of operations, greatly enhancing system security and audit compliance; simultaneously, localized model deployment reduces latency and resource consumption, while remote management via mobile terminals combined with blockchain improves operational convenience. Ultimately, this results in comprehensive advantages in reducing reliance on manual labor, preventing security risks, extending equipment lifespan, and supporting multi-device collaboration, comprehensively improving the automation level and reliability of signal jamming management.

[0031] 10. The signal feature extraction module adopts a progressive design of a first fully connected layer, a ReLU activation unit, and a second fully connected layer, realizing the transformation from raw signal features to high-level abstract features. The first fully connected layer initially extracts seven signal features, including frequency, bandwidth, and intensity, and maps them to a high-dimensional space. The ReLU activation unit introduces nonlinear features to enhance the model's ability to express complex signal patterns. The second fully connected layer abstracts high-level signal features through dimensionality reduction, denoising, and retaining key information. This design avoids the limitations of relying on manual feature engineering in traditional methods, improves the efficiency of automatic learning and representation of wireless communication signal features, and enhances the model's generalization ability through the nonlinear processing of the ReLU activation unit, enabling it to better handle signal noise and dynamic changes, thereby improving the accuracy of subsequent policy prediction.

[0032] 11. The parameter feature extraction module combines the embedding unit, the connection unit, and the third fully connected layer to intelligently process shielding parameters (such as shielding mode, scene, frequency range, etc.). The embedding unit encodes categorical parameters (such as mode, scene) into vectors, and the connection unit vectorizes and concatenates continuous parameters (such as frequency range, shielding strength). The third fully connected layer integrates and normalizes these vectors to extract high-level parameter features. This design uniformly handles heterogeneous parameters (categorical and continuous), reducing the dimensionality and noise of the input data. The embedding unit uses vector representation for text parameters (such as scene descriptions), which facilitates the model's understanding of semantics. The normalization step ensures the consistency and comparability of parameter features, which optimizes the adaptability of the shielding strategy generation model to user input, supports diverse shielding needs (such as different scenes or device densities), and improves the robustness and versatility of signal shielding strategy generation.

[0033] 12. The feature fusion module simply connects high-level signal features and high-level parameter features through a splicing unit, and then performs deep fusion through the fourth fully connected layer to generate a comprehensive feature vector. This avoids the simple stacking of traditional methods and instead uses a fully connected layer for nonlinear integration. The spliced ​​features retain the independent information of the signal and parameters, while the deep fusion of the fourth fully connected layer captures the interaction between the signal and parameters (such as how signal strength affects shielding strength), ensuring that the signal shielding strategy generation is based on comprehensive and complementary information and avoiding the problem of information silos. In practical applications, this design improves the predictive coordination efficiency of the shielding strategy generation model, especially in high-density wireless environments, where it can more effectively balance shielding effect and resource consumption.

[0034] 13. The prediction output module contains eight independent units (such as target masking prediction and frequency band coverage prediction). Each prediction unit is constructed from a fifth fully connected layer and an activation function layer. Specific loss functions are used for different prediction tasks (such as classification cross-entropy for target prediction and mean squared error for coverage prediction), and the signal masking strategy is finally output uniformly through the joint output unit. This multi-task design allows the model to optimize multiple output targets in parallel, with each prediction unit focusing on a single parameter (such as masking priority or beam direction), avoiding the overfitting risk of single-task models. At the same time, the differentiated design of the loss function (classification loss for discrete output and regression loss for continuous output) matches the parameter characteristics and improves prediction accuracy. For example, the target masking prediction unit is more accurate in handling classification problems, while the scanning velocity prediction unit is more suitable for regression analysis, which makes the generated signal masking strategy more reliable and flexible in actual deployment.

[0035] 14. The policy loss function is a weighted combination of the sub-losses of each prediction unit (w1 to w8 are weight coefficients), including cross-entropy loss and mean squared error loss. The policy loss function customizes weights for different output tasks (e.g., the w1 weight is used for the cross-entropy loss of target prediction) and dynamically adjusts them during training. The weighted combination of the policy loss function balances the priority of each task (e.g., increasing the target prediction weight to improve the weight of key decisions), avoids loss conflicts, and improves the overall convergence speed and stability of the masked policy generation model. In actual training, the masked policy generation model is allowed to efficiently learn complex relationships through backpropagation, reducing training time and resource overhead. This design enhances the versatility of the masked policy generation model and can adapt to the optimization needs of different masking scenarios.

[0036] 15. By employing a phased fully connected layer and nonlinear activation units, this method automatically extracts seven types of features, including frequency and bandwidth, from wireless communication signals and achieves high-order abstraction, overcoming the limitations of traditional methods that rely on manual feature engineering. Through embedding and connection units, parameters such as shielding mode, scene, and frequency range are uniformly processed, and normalized integration is achieved in conjunction with the third fully connected layer, significantly improving the collaborative efficiency of multi-source parameters. The prediction module independently constructs units for eight output tasks, including shielding target (cross-entropy loss) and frequency band coverage (mean square error), and designs a weighted strategy loss function to achieve efficient balance of differentiated training objectives. The prediction output module synchronously generates eight key strategy parameters, including shielding target and beam direction, covering the complex decision-making process that traditionally requires multi-system collaboration with a single model, significantly improving response speed and strategy consistency. The modular design supports flexible expansion (such as adding new signal feature types), and the deep learning framework is suitable for real-time environmental changes (such as equipment density fluctuations), effectively improving the shielding success rate.

[0037] 16. The risk assessment model innovatively integrates four completely different types of data sources (heterogeneous data) from wireless communication signals, shielding parameters, shielding strategies, and spatiotemporal information (time, location, account). This solves the problems of insufficient information, one-sided perspective, and inability to fully reflect complex risk scenarios from a single data source in risk assessment. It reflects a profound understanding of the nature of risk, namely that risk is the result of the combined effect of multiple factors, and significantly improves the perception capability and information completeness of the risk assessment model.

[0038] 17. Wireless communication signal features, combined with 1D CNN (capturing local spectral patterns) and BiGRU (capturing long-term temporal dependencies), can comprehensively and accurately characterize the dynamic characteristics of wireless communication signals and effectively identify abnormal signal patterns. Masked parameter features use an embedding layer to process class parameters (efficient dimensionality reduction, capturing semantic relationships) and fuse them with numerical parameters (FC layer), solving the problem of effective representation of mixed data types (numerical + class). Policy features adopt a multi-head self-attention mechanism, which can automatically learn the complex dependencies and importance weights between policy parameters, surpassing simple rule matching or shallow models, and better understanding the intent of the policy and potential conflicts. Spatiotemporal features use a dedicated time encoder and spatial location encoder to explicitly extract time periodicity (such as weekday / weekend, holiday effect) and geographical location features (such as regional risk differences, proximity effect), integrating key contextual information into the model. In other words, for the characteristics of each data source, the most advanced or most suitable deep learning technology is used for feature extraction, ensuring that the most discriminative information is extracted from the original data, laying a solid foundation for subsequent fusion and evaluation.

[0039] 18. By using gated attention units (GAU) for fusion, the attention weights of different feature sources (wireless signals, parameters, policies, spatiotemporal factors) can be dynamically calculated and weighted fused according to the specific circumstances of the current input sample, thus overcoming the limitations of traditional static fusion (such as simple splicing or fixed weighting). Dynamic fusion means that the risk assessment model can adaptively "focus" on the feature sources most relevant to the current risk, significantly improving the flexibility, adaptability, and accuracy of the risk assessment model. For example, some risks may be mainly triggered by abnormal signals, while others may be more affected by policy conflicts or spatiotemporal location. This mechanism effectively improves the intelligence level of the risk assessment model.

[0040] 19. By setting up a risk assessment model, not only are risk assessment results (level, score, type) output, but also policy adjustment content (S_adjusted) is directly generated using Conditional Generative Adversarial Network (CGAN), forming a closed loop of "risk identification -> policy optimization suggestion". This goes beyond simple risk detection and provides an actionable solution (policy adjustment), greatly improving the system's practical value and decision support capabilities. The application of CGAN enables it to learn the real policy distribution and generate more realistic and effective adjustment schemes.

[0041] 20. By integrating the losses of three key tasks through a risk loss function, the risk assessment model is forced to simultaneously consider the accuracy of risk assessment, the rationality and feasibility of strategy adjustment, and the stability of the integration mechanism during the optimization process. The weight coefficients (λ1, λ2, λ3) can be used to flexibly adjust the emphasis of different tasks. This joint optimization strategy is an important guarantee for the overall superior performance of the risk assessment model.

[0042] 21. For four types of heterogeneous data—wireless signals (1D CNN+BiGRU), masking parameters (embedding layer+fully connected), policies (multi-head self-attention), and spatiotemporal information (periodic / position encoder)—a customized module is used to extract high-discriminative features. Based on gated attention units (GAU), the weights of each feature source are dynamically calculated to achieve context-aware fusion, overcoming the limitations of static fusion. Risk assessment results (level / score / type) and policy adjustments based on conditional generative adversarial networks (CGAN) are jointly output, forming a "perception-assessment-optimization" closed loop. An innovative composite loss function (L2) is used to jointly optimize the accuracy of risk assessment, the rationality of policy adjustments, and the stability of attention, improving the system's generalization ability. A modular architecture ensures scalability, ultimately achieving a comprehensive breakthrough in reducing blind spots in risk perception, improving assessment accuracy, enhancing decision-making intelligence, and improving scenario adaptability.

[0043] 22. By systematically acquiring and integrating historical shielding logs (including signal parameters, shielding parameters, strategies, etc.) and historical risk assessment reports (including risk level, score, type, etc.), the core elements of signal shielding (such as signal frequency, strength, modulation method, shielding mode, scenario, etc.) are covered. This multi-dimensional data integration provides rich contextual information, far exceeding the one-sided data that existing systems usually rely on (such as only recording basic shielding parameters), and can more comprehensively capture the complexity of actual scenarios. By covering multiple shielding modes (such as full frequency band, specified frequency band, adaptive), the solution can adapt to different application scenarios (such as public safety), improve the pertinence and success rate of shielding decisions, and reduce the risks or failures caused by data omissions.

[0044] 23. Through a series of preprocessing operations (such as deduplication, missing value handling, error correction, data format unification, normalization, standardization, key feature extraction, and derivative feature creation), the data cleaning and transformation process is systematized, solving common data quality problems in existing technologies (such as noise, inconsistency, or messy formats), ensuring data consistency and comparability; high-quality preprocessing reduces the risk of bias in subsequent model training and improves prediction accuracy (such as masking effect evaluation or risk prediction); for example, normalization and standardization facilitate the convergence of machine learning algorithms, while derivative feature creation (such as generating new indicators based on signal strength and time combinations) can uncover potential patterns and support more efficient AI application development.

[0045] 24. By dividing the dataset into two different preset ratios (first ratio and second ratio) for the two models (masking strategy generation model and risk assessment model), data redundancy and overfitting risks are avoided. This customized division ensures that each model obtains targeted training and validation data, improves the generalization performance of the model, and reduces cross-interference between models, thereby enhancing the robustness of the overall system.

[0046] 25. By employing first-level validation (including hyperparameter optimization) and second-level validation (re-evaluating metrics after compression) after training, combined with final testing (calculating metrics such as F1 score, generalization ability, and robustness), a rigorous iterative mechanism is formed, significantly improving model reliability (e.g., preventing model failure after deployment) and inference efficiency (compression reduces latency). Specific advantages include: integrated compression techniques (weight pruning, channel pruning, weight quantization, activation quantization) to achieve model lightweighting (reducing model size and computational overhead), while second-level validation ensures that compressed performance metrics (such as accuracy, recall, and inference time) remain within acceptable ranges, reducing hardware resource requirements (such as the processing burden of signal jammers) and supporting deployment in resource-constrained environments; comprehensive performance evaluation, using multiple metrics (accuracy, recall, precision, F1 score, generalization ability, robustness) for evaluation, providing multi-dimensional monitoring (e.g., F1 score balances precision and recall), reducing model blind spots, and enhancing the system's robustness in noisy or adversarial environments.

[0047] 26. By separating the shielding strategy generation model and the risk assessment model (using their respective datasets for training, validation, and testing), parallel development and validation are supported. This independence reduces dependency errors between models (such as data leakage or hyperparameter conflicts) and speeds up iteration (e.g., two models can be optimized simultaneously), thereby improving the overall system's R&D efficiency and fault isolation capabilities.

[0048] 27. By dividing the dataset into independent parts according to a preset ratio, and combining multi-level validation, model compression, and containerized deployment, efficient and reliable closed-loop model management is achieved: its dual-ratio data division ensures the independence of model training and generalization ability; the strict two-level validation process (including hyperparameter optimization and pruning, quantization compression) significantly improves inference efficiency while ensuring core indicators such as accuracy and recall; containerization of the model and environment and setting up a scheduling mechanism not only achieves resource isolation and rapid deployment, but also enhances the system's robustness and real-time response capability on edge devices such as signal jammers, ultimately achieving a lightweight, low-latency, and easy-to-maintain industrial-grade AI deployment effect.

[0049] 28. By employing multi-layered encryption mechanisms (such as RSA and ECDH for asymmetric encryption, KDF for key derivation, and HMAC for message authentication), the confidentiality and integrity of data transmission are ensured. For example, mobile terminals use the device's private key to encrypt authentication data, while the server uses the device's public key to decrypt and perform hash verification, effectively preventing man-in-the-middle attacks and data tampering. Timestamp verification (such as the timeliness verification of the first and second timestamps) further resists replay attacks, ensuring the real-time nature of requests. This hybrid encryption strategy is more robust than a single algorithm, reduces security vulnerabilities, and improves the overall anti-attack capability of the system.

[0050] 29. Session keys are derived using KDF functions, based on the root public key, PIN code, and dynamically generated random numbers (such as numbers extracted from timestamps), ensuring the uniqueness and temporariness of session keys and avoiding the risk of static key leakage, while simplifying the key negotiation process; the server and mobile terminal independently derive session keys (first session key and second session key), and ensure consistency through verification, improving the efficiency of end-to-end encryption; the session key derivation process is efficient and secure, reducing computational overhead, supporting high-frequency communication (such as heartbeat connections and command issuance), while ensuring the transience of session keys, conforming to the zero-trust security model.

[0051] 30. Generating random numbers (such as the second random number) by extracting numbers from timestamps (e.g., extracting the second to last digit from the first timestamp) increases the source of randomness and unpredictability. Combined with hash calculation, it enhances the security of the authentication process. In other words, it innovatively combines the characteristics of timestamps, avoids the weaknesses of traditional pseudo-random number generators, and improves the system's dynamic anti-cracking capabilities.

[0052] 31. By integrating multi-layered encryption mechanisms (RSA, ECDH, KDF, HMAC), dynamic timestamp verification, and blockchain technology, the security and reliability of the signal jamming system are significantly improved: it employs asymmetric encryption and end-to-end session key negotiation (such as based on PIN codes and random number-derived keys) to ensure the confidentiality of authentication and command transmission; it utilizes timestamps to dynamically generate random numbers and combines them with hash verification to effectively defend against replay attacks and data tampering; at the same time, it enhances audit traceability by storing operation records immutably through blockchain; the entire process relies on the TLS protocol to achieve efficient communication, supports users to remotely customize jamming parameters, and balances security protection with flexible control, making it suitable for real-time jamming needs in highly sensitive scenarios.

[0053] 32. By employing multi-layered encryption and decryption and dual verification mechanisms (MAC value integrity verification + timestamp validity verification), the absolute security and reliability of the transmission of blocking commands are ensured. At the same time, relying on a locally deployed intelligent model (optimized by hardware acceleration), the wireless communication signals are analyzed in real time and a precise signal blocking strategy is dynamically generated. This enables adaptive interference signal transmission with millisecond-level response, significantly improving blocking efficiency and targeting while greatly reducing server load, network dependence, and the risk of false interference from non-target signals. Furthermore, the operation authorization traceability is strengthened through account / device ID binding. Ultimately, a modern signal blocking system that is secure, reliable, resource-efficient, intelligent, precise, and remotely dynamically controllable is constructed.

[0054] 33. By collecting multi-dimensional real-time shielding logs in real time and using hardware acceleration technology for dynamic risk assessment, adaptive closed-loop optimization of signal shielding strategies was achieved, significantly improving the system's response speed and shielding accuracy. At the same time, based on incremental data, the model was continuously optimized, enhancing the robustness and anti-interference capability against new signal threats.

[0055] 34. Through multi-layered encryption mechanisms (session key and root private key working together), immutable blockchain storage, and real-time key management processes, the security, integrity, and traceability of signal jamming data are significantly improved: dynamic session keys are used to encrypt and upload real-time jamming logs and real-time risk assessment reports, and the server encrypts the data again using the root private key before uploading it to the blockchain, ensuring that the transmission and storage process is leak-proof and tamper-proof throughout; combined with the design of optimizing transmission efficiency in preset frequency bands and sharing block information with mobile terminals, high-reliability real-time monitoring, remote verifiable auditing, and user operation convenience are simultaneously achieved, forming a closed-loop solution that takes into account data protection, system performance, and compliance requirements.

[0056] 35. By leveraging blockchain technology to ensure the immutability and traceability of signal jammer operation data, and utilizing a pre-set root public key to securely decrypt encrypted records, logs, and reports, the integrity and access security of the data are significantly improved. Mobile terminals, based on decrypted jammer operation records, real-time jamming logs, and real-time risk assessment reports, enable dynamic monitoring and rapid response to signal jammers, effectively reducing maintenance delays and manual intervention costs. The server's pre-shared block information mechanism optimizes data transmission efficiency, while the centralized management model, combined with the encryption features of blockchain, provides an efficient and reliable equipment management solution for critical sectors while ensuring user privacy and compliance.

[0057] 36. By dynamically generating precise signal shielding strategies through deep neural networks, and combining multi-source feature fusion with a real-time risk assessment closed loop, the adaptability and security of signal shielding are significantly improved. Blockchain-encrypted evidence storage of shielding device operation records, real-time shielding logs, and real-time risk assessment reports ensures data immutability. Lightweight model compression, hardware acceleration, and containerized deployment enable low-latency edge computing. At the same time, a multi-layered dynamic key derivation and spatiotemporal random number authentication mechanism is designed to strengthen end-to-end security. Ultimately, without relying on the cloud, a comprehensive breakthrough is achieved in shielding accuracy, risk resistance, resource efficiency, and audit credibility. Attached Figure Description

[0058] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0059] Figure 1 This is a flowchart of a signal jammer management method that combines neural networks and blockchain according to the present invention. Detailed Implementation

[0060] The overall concept of the technical solution in this application is as follows: The mobile terminal sends a blocking command to the signal jammer through the server, realizing remote management of the signal jammer without the need for local operation of the signal jammer; Through a pre-trained and deployed blocking strategy generation model, a real-time signal jamming strategy is automatically generated based on the input real-time wireless communication signal and real-time blocking parameters. Different blocking modes or blocking scenarios can be selected as needed, simplifying the operation process and reducing the risk of human error; Through the authentication operation between the mobile terminal and the server, the risk assessment of the risk assessment model, and the blockchain storage of the jammer operation records, real-time blocking logs, and real-time risk assessment reports, the signal jammer is prevented from being easily abused or used without authorization, and it is easy to trace the source later, thereby improving the convenience, reliability, and security of the signal jammer.

[0061] This embodiment provides a signal jammer management method that combines neural networks and blockchain, such as... Figure 1 As shown, it includes the following steps:

[0062] Step S10: Create a shielding strategy generation model for outputting a signal shielding strategy based on wireless communication signals and shielding parameters, and a risk assessment model for outputting a risk assessment report based on shielding logs. Set the strategy loss function of the shielding strategy generation model and the risk loss function of the risk assessment model.

[0063] Step S20: Obtain a large number of historical blocking logs and historical risk assessment reports, and construct a dataset after preprocessing and labeling each of the historical blocking logs and historical risk assessment reports; the historical blocking logs include at least historical wireless communication signals, historical blocking parameters, historical signal blocking strategies, historical blocking times, historical operation accounts, and historical blocking locations;

[0064] A dataset was built based on a large number of historical blocking logs and historical risk assessment reports to train and optimize neural network models (blocking strategy generation model and risk assessment model). The neural network model can learn complex patterns (such as signal characteristics in different environments), improve generalization ability, and ensure robustness in new scenarios (such as different geographical locations or signal types). This is more flexible than rule-based blocking methods and reduces the need for manual configuration.

[0065] Step S30: Train and compress the shielding strategy generation model and risk assessment model using the dataset, and deploy the trained shielding strategy generation model and risk assessment model to the signal jammer;

[0066] By training and compressing the shielding strategy generation model and risk assessment model before deploying them to the signal jammer, reliance on the cloud (server) is reduced. At the same time, using preset frequency bands for authentication optimizes communication efficiency and allows the signal jammer to communicate with the server while it is operating. Model compression reduces hardware resource requirements (such as computing and storage), making it suitable for resource-constrained signal jammers. Local processing reduces network latency, improves response speed, and overall reduces deployment and maintenance costs, while improving system scalability.

[0067] Step S40: After the signal jammer is powered on, the mobile terminal performs an authentication operation with the server through a preset frequency band. After successful authentication, the server sends a jamming command to the signal jammer. The server generates a jammer operation record based on the jamming command, encrypts the jammer operation record into an encrypted record, and uploads it to the blockchain.

[0068] Step S50: The signal jammer verifies and parses the received jamming command to obtain real-time jamming parameters, collects real-time wireless communication signals, inputs the real-time wireless communication signals and real-time jamming parameters into the deployed jamming strategy generation model to obtain a real-time signal jamming strategy, generates interference signals based on the real-time signal jamming strategy and transmits them to the outside to perform signal jamming operations.

[0069] Step S60: The signal jammer records a real-time jamming log that includes at least the real-time wireless communication signal, real-time jamming parameters, real-time signal jamming strategy, real-time jamming time, real-time operation account, and real-time jamming location. The real-time jamming log is input into the deployed risk assessment model to obtain a real-time risk assessment report. The real-time signal jamming strategy is dynamically adjusted based on the real-time risk assessment report.

[0070] The shielding strategy generation model constructed through neural networks can automatically output signal shielding strategies based on real-time wireless communication signals and shielding parameters, reducing manual intervention. The shielding strategy generation model is trained using historical data, which improves the accuracy and adaptability of signal shielding strategy generation. Combined with feedback from real-time shielding logs, the signal shielding strategy is dynamically adjusted to cope with environmental changes (such as changes in signal interference sources), thereby optimizing the shielding effect, reducing the risk of false shielding, and improving the overall system efficiency.

[0071] The risk assessment model outputs a real-time risk assessment report based on real-time shielding logs (including real-time wireless communication signals, real-time shielding parameters, real-time signal shielding strategies, real-time shielding time, real-time operation accounts, and real-time shielding locations). This report is used to dynamically adjust real-time signal shielding strategies, achieving closed-loop control and enabling timely detection of potential problems (such as unauthorized operations or excessive shielding) to proactively reduce security risks. Furthermore, the risk assessment model is trained using historical data, enhancing the reliability and predictive ability of risk assessment and avoiding the lag inherent in traditional methods that rely on post-hoc analysis.

[0072] The real-time blocking log records multi-dimensional parameters (real-time wireless communication signal, real-time blocking parameters, real-time signal blocking strategy, real-time blocking time, real-time operation account, and real-time blocking location), and uploads them to the blockchain to provide complete operation audit trails, facilitating post-event analysis, report generation, and compliance checks. The transparency of the blockchain ensures that all records are verifiable.

[0073] Step S70: The signal jammer encrypts and uploads the real-time jamming log and the real-time risk assessment report to the server. The server encrypts the received real-time jamming log and the real-time risk assessment report into encrypted logs and encrypted reports respectively and uploads them to the blockchain.

[0074] Step S80: The mobile terminal obtains the encrypted records, encrypted logs, and encrypted reports through the blockchain, and manages the signal jammer based on the encrypted records, encrypted logs, and encrypted reports.

[0075] By using blockchain technology, all jammer operation records, real-time jamming logs, and real-time risk assessment reports are encrypted and stored (encrypted records, encrypted logs, and encrypted reports), and access management is achieved through mobile terminals. In other words, the distributed ledger characteristics of blockchain ensure that the data is immutable and traceable, preventing unauthorized access or data tampering (e.g., operation account and jamming location records), meeting the needs of high-security scenarios; at the same time, the encrypted upload process enhances the security of data transmission.

[0076] Mobile terminals acquire data (encrypted records, encrypted logs, and encrypted reports) through blockchain, enabling remote monitoring and management of signal jammers. This simplifies user operations, supports managing signal jammers anytime, anywhere, and effectively improves user experience. Combined with server authentication, it ensures that the management process is secure and controllable, making it suitable for distributed or multi-device scenarios.

[0077] By integrating neural network and blockchain technologies, intelligent, secure, and efficient management of signal jammers has been achieved. The neural network model (which generates jamming strategies and assesses risks) dynamically generates and optimizes jamming strategies, significantly improving the accuracy and environmental adaptability of jamming. Blockchain is used to encrypt and distribute the storage of jammer operation records, real-time jamming logs, and real-time risk assessment reports, ensuring data immutability and full traceability of operations, greatly enhancing system security and audit compliance. Simultaneously, localized model deployment reduces latency and resource consumption, while remote management via mobile terminals combined with blockchain improves operational convenience. Ultimately, this approach offers comprehensive advantages in reducing reliance on manual labor, preventing security risks, extending equipment lifespan, and supporting multi-device collaboration, thus comprehensively improving the automation level and reliability of signal jamming management.

[0078] In step S10, the shielding strategy generation model is constructed based on a signal feature extraction module, a parameter feature extraction module, a feature fusion module, and a prediction output module.

[0079] The signal feature extraction module is constructed based on a first fully connected layer, a second fully connected layer, and a ReLU activation unit. The first fully connected layer is used to extract initial signal features from the wireless communication signal, including at least frequency features, bandwidth features, intensity features, modulation features, phase features, signal-to-noise ratio features, and device density features, and to map each of the initial signal features to a high-dimensional space. The ReLU activation unit is used to introduce nonlinearity into the initial signal features in the high-dimensional space to obtain nonlinear features. The second fully connected layer is used to reduce the dimensionality and abstract the nonlinear features to obtain high-level signal features.

[0080] The first fully connected layer has an output dimension of 128 and its function is to map the initial signal features (such as the frequency and intensity features of continuous values) to a high-dimensional space to capture nonlinear relationships. The ReLU activation unit is used to introduce nonlinearity and enhance the model's expressive power. The second fully connected layer has an input dimension of 128 and an output dimension of 64. It is used to reduce the dimensionality and abstract the nonlinear features and extract high-level signal features (deep feature representation / high-level dynamic features), such as the stability features of signal bandwidth or the dynamic change features of signal-to-noise ratio. The signal feature extraction module is used to extract high-level signal features of wireless communication signals, supporting the model to adapt to different signal types (such as high-frequency broadband signals or low signal-to-noise ratio narrowband signals) and reducing the interference of signal noise on prediction.

[0081] The signal feature extraction module adopts a progressive design consisting of a first fully connected layer, a ReLU activation unit, and a second fully connected layer, realizing the transformation from raw signal features to high-level abstract features. The first fully connected layer initially extracts seven signal features, including frequency, bandwidth, and intensity, and maps them to a high-dimensional space. The ReLU activation unit introduces nonlinear features to enhance the model's ability to express complex signal patterns. The second fully connected layer abstracts high-level signal features through dimensionality reduction, denoising, and retaining key information. This design avoids the limitations of relying on manual feature engineering in traditional methods, improves the efficiency of automatic learning and representation of wireless communication signal features, and enhances the model's generalization ability through the nonlinear processing of the ReLU activation unit, enabling it to better handle signal noise and dynamic changes, thereby improving the accuracy of subsequent policy prediction.

[0082] The parameter feature extraction module is constructed based on an embedding unit, a connection unit, and a third fully connected layer. The embedding unit is used to encode the shielding mode and shielding scenario in the shielding parameters to obtain a mode vector and a scenario vector. The connection unit is used to convert the shielding frequency range, shielding intensity, shielding area size, shielding duration, and shielding position in the shielding parameters into a frequency range vector, a shielding intensity vector, an area size vector, a shielding duration vector, and a shielding position vector, and concatenates the mode vector, scenario vector, frequency range vector, shielding intensity vector, area size vector, shielding duration vector, and shielding position vector into a concatenated vector. The third fully connected layer is used to integrate and normalize the concatenated vector and extract high-level parameter features.

[0083] The input to the embedding unit is a categorical variable, such as a full-band shielding mode, a specified-band shielding mode, a scene-adaptive shielding mode, and a custom shielding mode. This variable maps the categorical variable to a 32-dimensional vector, encoding discrete modes into continuous representations and capturing the similarity between modes / scenes (e.g., the distance between a full-band mode and a specified-band mode). The connection unit concatenates the output of the embedding unit with other shielding parameters (continuous values: shielding frequency range, shielding intensity, shielding area size, shielding duration, and shielding position) into a single vector (the concatenated vector). The output dimension of the third fully connected layer is 64, used to integrate and normalize the concatenated vector (e.g., scene-based constraints). The parameter feature extraction module extracts high-level parameter features (high-level semantic features) of the shielding parameters, emphasizing the adaptability of the mode (e.g., adjusting intensity based on area size in scene-adaptive mode), ensuring the consistency and robustness of the signal shielding strategy in the operating environment and configuration.

[0084] The parameter feature extraction module combines an embedding unit, a connection unit, and a third fully connected layer to intelligently process shielding parameters (such as shielding mode, scene, frequency range, etc.). The embedding unit encodes categorical parameters (such as mode, scene) into vectors, and the connection unit vectorizes and concatenates continuous parameters (such as frequency range, shielding strength). The third fully connected layer integrates and normalizes these vectors to extract high-level parameter features. This design uniformly handles heterogeneous parameters (categorical and continuous), reducing the dimensionality and noise of the input data. The embedding unit uses vector representation for textual parameters (such as scene descriptions), facilitating the model's understanding of semantics. The normalization step ensures the consistency and comparability of parameter features, which optimizes the adaptability of the shielding strategy generation model to user input, supports diverse shielding needs (such as different scenes or device densities), and improves the robustness and versatility of signal shielding strategy generation.

[0085] The feature fusion module is constructed based on a splicing unit and a fourth fully connected layer; the splicing unit is used to splice high-level signal features and high-level parameter features to obtain spliced ​​features; the fourth fully connected layer is used to perform deep fusion of spliced ​​features to generate a comprehensive feature vector;

[0086] The splicing unit is used to splice the outputs (64-dimensional) of the signal feature extraction module and the parameter feature extraction module (64-dimensional) along the feature dimensions into a 128-dimensional vector (spliced ​​feature); the fourth fully connected layer is used to perform deep fusion on the 128-dimensional vector (such as the collaborative optimization of signal strength and shielding strength) to generate a 128-dimensional comprehensive feature vector; the feature fusion module is used to integrate the features of wireless communication signals and shielding parameters, enhance feature correlation (such as frequency coverage decision based on channel state information), and provide a unified input for the next module; for scenario optimization, the feature fusion module prioritizes high-priority features (such as strengthening the scene information weight when the shielding mode is scene adaptive) to improve decision accuracy.

[0087] The feature fusion module simply connects high-level signal features and high-level parameter features through a splicing unit, and then performs deep fusion through a fourth fully connected layer to generate a comprehensive feature vector. This avoids the simple stacking of traditional methods and instead utilizes a fully connected layer for nonlinear integration. The spliced ​​features retain the independent information of the signal and parameters, while the deep fusion of the fourth fully connected layer captures the interaction between the signal and parameters (such as how signal strength affects shielding strength), ensuring that the signal shielding strategy generation is based on comprehensive and complementary information and avoiding the problem of information silos. In practical applications, this design improves the predictive coordination efficiency of the shielding strategy generation model, especially in high-density wireless environments, where it can more effectively balance shielding effectiveness and resource consumption.

[0088] The prediction output module is constructed based on a shielding target prediction unit, a frequency band coverage prediction unit, a scanning speed prediction unit, a transmit power prediction unit, a beam direction prediction unit, a beamwidth prediction unit, a sidelobe suppression prediction unit, a shielding priority prediction unit, and a joint output unit; each of the shielding target prediction unit, frequency band coverage prediction unit, scanning speed prediction unit, transmit power prediction unit, beam direction prediction unit, beamwidth prediction unit, sidelobe suppression prediction unit, and shielding priority prediction unit is constructed from a fifth fully connected layer and an activation function layer;

[0089] The prediction output module comprises eight independent units (such as target masking prediction and frequency band coverage prediction), each constructed from a fifth fully connected layer and an activation function layer. Specific loss functions are used for different prediction tasks (e.g., classification cross-entropy for target prediction and mean squared error for coverage prediction), and the signal masking strategy is ultimately output uniformly through a joint output unit. This multi-task design allows the model to optimize multiple output targets in parallel, with each prediction unit focusing on a single parameter (e.g., masking priority or beam direction), avoiding the overfitting risk of single-task models. Simultaneously, the differentiated design of the loss functions (classification loss for discrete output and regression loss for continuous output) matches the parameter characteristics, improving prediction accuracy. For example, the target masking prediction unit is more accurate in handling classification problems, while the scanning velocity prediction unit is more suitable for regression analysis, making the generated signal masking strategy more reliable and flexible in practical deployment.

[0090] The shielding target prediction unit is used to infer the shielding target from the integrated feature vector; the frequency band coverage prediction unit is used to infer the frequency band coverage from the integrated feature vector; the scanning speed prediction unit is used to infer the scanning speed from the integrated feature vector; the transmit power prediction unit is used to infer the transmit power from the integrated feature vector; the beam direction prediction unit is used to infer the beam direction from the integrated feature vector; the beamwidth prediction unit is used to infer the beamwidth from the integrated feature vector; the sidelobe suppression prediction unit is used to infer the sidelobe suppression from the integrated feature vector; the shielding priority prediction unit is used to infer the shielding priority from the integrated feature vector; and the joint output unit is used to output a signal shielding strategy carrying the shielding target, frequency band coverage, scanning speed, transmit power, beam direction, beamwidth, sidelobe suppression, and shielding priority.

[0091] The fifth fully connected layer has an input dimension of 128, and its output dimension is adjusted according to the type of prediction unit. The activation function layer selects an activation function based on the output data type (ReLU for continuous output, Softmax for classification output). The output dimension of the shielded target prediction unit is equal to the number of potential target categories, and its activation function is Softmax, which is used to classify and output shielded targets (such as target identifiers based on modulation methods). The output dimension 2 of the frequency band coverage prediction unit (min and max frequency bands) uses ReLU as its activation function and is used to predict the continuous frequency band range (unit: Hz). The output dimension 1 of the scan speed prediction unit uses ReLU as its activation function and is used to predict the continuous scan speed (unit: Hz / s). The output dimension 1 of the transmit power prediction unit uses ReLU as its activation function and is used to predict the continuous transmit power (unit: d). The output dimension 2 (azimuth and elevation) of the beam direction prediction unit is activated by ReLU and used to predict continuous beam directions (in degrees); the output dimension 1 of the beamwidth prediction unit is activated by ReLU and used to predict continuous beamwidth (in degrees); the output dimension 1 of the sidelobe suppression prediction unit is activated by ReLU and used to predict continuous sidelobe suppression / suppression value (in dB); the output dimension of the masking priority prediction unit is equal to the number of priority levels, and the activation function is Softmax, used to classify and output priority weights; the prediction output module is used to map the comprehensive feature vector to the multi-task strategy output, ensuring that each prediction unit optimizes independently (e.g., prioritizing the frequency band coverage of high-priority targets), and different prediction units share the underlying features (comprehensive feature vector) to reduce overfitting and improve generalization ability.

[0092] The formula for the policy loss function is:

[0093] L1=w1*L target +w2*L coverage +w3*L speed +w4*L power +w5*L direction +w6*L width +w7*L suppression +w8*L priority ;

[0094] Where L1 represents the loss value of the policy loss function; L target The sub-loss representing the masked target prediction unit is the classification cross-entropy loss, which measures the difference between the predicted probability distribution and the actual label distribution, meaning minimizing the probability of misclassification; L coverage This represents the sub-loss of the frequency band coverage prediction unit, using the mean square error loss, which is calculated as the average of the squared differences between the predicted and actual values, meaning minimizing the magnitude of the prediction error; L speedL represents the sub-loss of the scan rate prediction unit, using mean squared error loss; power The sub-loss of the transmit power prediction unit is represented by the mean square error loss; L direction The sub-loss of the beam direction prediction unit is represented by the mean square error loss; L width The sub-loss of the beamwidth prediction unit is represented by the mean square error loss; L suppression The sub-loss of the sidelobe suppression prediction unit is represented by the mean squared error loss; L priority The sub-loss of the masked priority prediction unit is represented by the classification cross-entropy loss; w1, w2, w3, w4, w5, w6, w7, and w8 all represent weight coefficients.

[0095] Since the masking policy generation model simultaneously predicts multiple related but heterogeneous outputs (classification and regression tasks), the policy loss function, through weighted and integrated sub-losses, allows the masking policy generation model to share knowledge on the comprehensive feature vector generated by the feature fusion module, improving overall generalization ability. The weight coefficients are used to adjust the relative importance of each task (e.g., if the masking priority has a greater impact on the signal masking policy, its weight can be increased). The classification task (masking target and masking priority) uses the classification cross-entropy loss because it is suitable for multi-class classification problems and can effectively handle probabilistic outputs (through activation function layers in the prediction unit, such as Softmax). The regression task (the remaining outputs) uses the mean squared error loss because it is sensitive to the error in predicting continuous values, promoting model convergence to accurate values. Minimizing L1 means optimizing all prediction units simultaneously, ensuring that the generated signal masking policy (including masking target, frequency band coverage, etc.) is close to the real policy overall. During training, the loss value of the policy loss function updates the model parameters (such as the weights of fully connected layers) through backpropagation.

[0096] The policy loss function is a weighted combination of the sub-losses of each prediction unit (w1 to w8 are weight coefficients), including cross-entropy loss and mean squared error loss. The policy loss function customizes weights for different output tasks (e.g., the w1 weight is used for the cross-entropy loss of target prediction) and dynamically adjusts them during training. The weighted combination of the policy loss function balances the priority of each task (e.g., increasing the target prediction weight to improve the weight of key decisions), avoids loss conflicts, and improves the overall convergence speed and stability of the masked policy generation model. In actual training, the masked policy generation model is allowed to efficiently learn complex relationships through backpropagation, reducing training time and resource overhead. This design enhances the versatility of the masked policy generation model and can adapt to the optimization needs of different masking scenarios.

[0097] By employing a phased fully connected layer and nonlinear activation units, this method automatically extracts seven types of features, including frequency and bandwidth, from wireless communication signals and achieves high-order abstraction, overcoming the limitations of traditional methods that rely on manual feature engineering. Through embedding and connection units, parameters such as shielding mode, scene, and frequency range are uniformly processed, and a third fully connected layer is used for normalization integration, significantly improving the collaborative efficiency of multi-source parameters. The prediction module independently constructs units for eight output tasks, including shielding target (cross-entropy loss) and frequency band coverage (mean square error), and designs a weighted strategy loss function to achieve an efficient balance of differentiated training objectives. The prediction output module simultaneously generates eight key strategy parameters, including shielding target and beam direction, covering the complex decision-making process that traditionally requires multi-system collaboration with a single model, significantly improving response speed and strategy consistency. The modular design supports flexible expansion (such as adding new signal feature types), and the deep learning framework is suitable for real-time environmental changes (such as fluctuations in device density), effectively improving the shielding success rate.

[0098] In step S10, the risk assessment model is constructed based on a multi-source feature extraction layer, a dynamic fusion layer, and an assessment report output layer.

[0099] The multi-source feature extraction layer is constructed based on a wireless communication signal feature extraction module, a masking parameter feature extraction module, a policy feature extraction module, and a spatiotemporal feature extraction module. The wireless communication signal feature extraction module extracts local spectral features from the wireless communication signal using a 1D convolutional network (Conv1D) and extracts signal temporal dependency features using a bidirectional gated recurrent unit (GRU), outputting wireless communication signal features including the local spectral features and the signal temporal dependency features. The masking parameter feature extraction module converts categorical parameters (such as masking patterns) in the masking parameters into embedding vectors using an embedding layer, and fuses numerical parameters in the masking parameters with the embedding vectors using a sixth fully connected layer (FCN) to obtain the masking parameter features. The policy feature extraction module uses a multi-head self-attention mechanism. Self-Attention extracts the dependencies between policy parameters from the signal masking strategy (emphasizing the weight of high-priority policies) to obtain policy features; the spatiotemporal feature extraction module is used to extract spatiotemporal features, including time periodic features and geographical location features, from the masking time, operation account, and masking location using a time encoder (LSTM) and a spatial location encoder (GeoHash).

[0100] The risk assessment model innovatively integrates four completely different types of data sources (heterogeneous data) from wireless communication signals, shielding parameters, shielding strategies, and spatiotemporal information (time, location, account). This solves the problems of insufficient information, one-sided perspective, and inability to fully reflect complex risk scenarios from a single data source in risk assessment. It reflects a profound understanding of the nature of risk, namely that risk is the result of the combined effect of multiple factors, and significantly improves the perception capability and information completeness of the risk assessment model.

[0101] The wireless communication signal features, combined with 1D CNN (capturing local spectral patterns) and BiGRU (capturing long-term temporal dependencies), can comprehensively and accurately characterize the dynamic characteristics of wireless communication signals and effectively identify abnormal signal patterns. Masking parameter features use an embedding layer to process class parameters (efficient dimensionality reduction, capturing semantic relationships) and fuse them with numerical parameters (FC layer), solving the problem of effective representation of mixed data types (numerical + class). Policy features employ a multi-head self-attention mechanism, automatically learning the complex dependencies and importance weights between policy parameters, surpassing simple rule matching or shallow models, and better understanding the policy's intent and potential conflicts. Spatiotemporal features, through specialized time encoders and spatial location encoders, explicitly extract time periodicity (e.g., weekday / weekend, holiday effects) and geographical location features (e.g., regional risk differences, proximity effects), integrating key contextual information into the model. In short, for each data source, the most advanced or suitable deep learning techniques are used for feature extraction, ensuring that the most discriminative information is extracted from the original data, laying a solid foundation for subsequent fusion and evaluation.

[0102] The dynamic fusion layer is used to calculate the attention weights of wireless communication signal features, shielding parameter features, policy features, and spatiotemporal features through a gated attention unit, and then weighted and fused to obtain fused features.

[0103] In practice, the weights of key features in risk scenarios can be adaptively learned, such as strengthening strategy features in full-band shielding mode and strengthening spatiotemporal features in geographically sensitive scenarios.

[0104] By using gated attention units (GAUs) for fusion, the attention weights of different feature sources (wireless signals, parameters, policies, spatiotemporal factors) can be dynamically calculated and weighted according to the specific circumstances of the current input sample, thus overcoming the limitations of traditional static fusion (such as simple splicing or fixed weighting). Dynamic fusion means that the risk assessment model can adaptively "focus" on the feature sources most relevant to the current risk, significantly improving the flexibility, adaptability, and accuracy of the risk assessment model. For example, some risks may be mainly triggered by abnormal signals, while others may be more affected by policy conflicts or spatiotemporal location. This mechanism effectively improves the intelligence level of the risk assessment model.

[0105] The assessment report output layer is constructed based on a risk assessment module, a strategy generation module, and a report output module. The risk assessment module is used to infer the fusion features through a seventh fully connected layer and a ReLU activation function to obtain the risk level, risk score, and risk type. The strategy generation module is used to infer the fusion features and signal masking strategies through a conditional generative adversarial network (CGAN) to obtain the strategy adjustment content. The report output module is used to output a risk assessment report carrying the risk level, risk score, risk type, and strategy adjustment content.

[0106] The risk level is categorized as low, medium, or high risk, and is based on softmax classification, meaning the feature vectors output by the seventh fully connected layer and the ReLU activation function are mapped onto a probability distribution using the softmax function. The risk score is a 0-1 regression value, based on sigmoid mapping, meaning the feature vectors output by the seventh fully connected layer and the ReLU activation function are mapped onto a 0-1 regression value using the sigmoid function. The risk type is defined as interference risk, security risk, or compliance risk, and is based on multi-label sigmoid mapping.

[0107] By setting up a risk assessment model, not only are risk assessment results (level, score, type) output, but also policy adjustment content (S_adjusted) is directly generated using Conditional Generative Adversarial Network (CGAN), forming a closed loop of "risk identification -> policy optimization suggestion". This goes beyond simple risk detection and provides an actionable solution (policy adjustment), greatly improving the system's practical value and decision support capabilities. The application of CGAN enables it to learn the real policy distribution and generate more realistic and effective adjustment schemes.

[0108] The formula for the risk loss function is: L2 = λ1 * L risk +λ2*L strategy +λ3*L GA ;

[0109] Where L2 represents the loss value of the risk loss function; L risk Indicates risk assessment loss; L strategy Indicates the loss due to strategy adjustment; L GA λ1, λ2, and λ3 represent the gated attention regularization loss; λ1, λ2, and λ3 all represent weight coefficients. ;

[0110] in, Represents the cross-entropy loss function; Indicates the true risk level label; This represents the probability distribution of the predicted risk level. Represents the mean square error function; Indicates the true risk score label; Indicates the predicted risk score; Represents the binary cross-entropy function; Indicates the true risk type label; This indicates the probability of predicting a risk type; ;

[0111] in, This represents the original signal shielding strategy; This indicates the adjusted signal shielding strategy; Represents the JavaScript divergence function; This represents the distribution of real historical signal masking strategies; This represents the distribution of signal shielding strategies synthesized by a conditional generative adversarial network. ;

[0112] in, Indicates the attention weights of the dynamic fusion layer; This represents the L2 norm, used to penalize excessive weight concentration.

[0113] By integrating the losses of the three key tasks through a risk loss function, the risk assessment model is forced to simultaneously consider the accuracy of risk assessment, the rationality and feasibility of strategy adjustment, and the stability of the integration mechanism during the optimization process. The weight coefficients (λ1, λ2, λ3) can be used to flexibly adjust the emphasis of different tasks. This joint optimization strategy is an important guarantee for the overall superior performance of the risk assessment model.

[0114] For four types of heterogeneous data—wireless signals (1D CNN+BiGRU), masking parameters (embedding layer+fully connected), policies (multi-head self-attention), and spatiotemporal information (periodic / positional encoders)—a customized module is used to extract high-discriminative features. Based on gated attention units (GAU), the weights of each feature source are dynamically calculated to achieve context-aware fusion, overcoming the limitations of static fusion. Risk assessment results (level / score / type) and policy adjustments based on conditional generative adversarial networks (CGAN) are jointly output, forming a "perception-assessment-optimization" closed loop. An innovative composite loss function (L2) is used to jointly optimize the accuracy of risk assessment, the rationality of policy adjustments, and the stability of attention, improving the system's generalization ability. A modular architecture ensures scalability, ultimately achieving a comprehensive breakthrough in reducing blind spots in risk perception, improving assessment accuracy, enhancing decision-making intelligence, and increasing scenario adaptability.

[0115] Step S20 specifically involves:

[0116] Obtain a large amount of historical blocking logs and historical risk assessment reports; the historical blocking logs include at least historical wireless communication signals, historical blocking parameters, historical signal blocking strategies, historical blocking times, historical operation accounts, and historical blocking locations; the historical risk assessment reports include at least historical risk levels, historical risk scores, historical risk types, and historical strategy adjustment content;

[0117] The historical wireless communication signals include at least signal frequency, signal bandwidth, signal strength, modulation scheme, phase, signal-to-noise ratio, and channel state information; the modulation schemes include at least amplitude modulation, frequency modulation, phase modulation, and digital modulation; in specific implementations, device density can be analyzed through the channel state information (CSI); the historical shielding parameters include at least shielding mode, shielding scenario, shielding frequency range, shielding strength, shielding area size, shielding duration, and shielding location; the shielding modes include at least full-band shielding mode, specified frequency band shielding mode, scenario-adaptive shielding mode, and custom shielding mode; the shielding scenario is triggered when the shielding mode is scenario-adaptive shielding mode, and the shielding scenario can be set as needed in actual applications; when the shielding mode is full-band shielding mode or scenario-adaptive shielding mode, the shielding frequency range is empty; the shielding parameters can be used to control the operation of the signal jammer, such as signal shielding. If the device determines that it is not currently in a preset shielded position using a locator, it cannot initiate the transmission of interference signals. The historical signal shielding strategy includes at least the shielding target, frequency band coverage, scanning speed, transmission power, beam direction, beamwidth, sidelobe suppression, and shielding priority. The shielding target, i.e., the object to be shielded, includes at least mobile phone signals (2G, 3G, 4G, 5G), Wi-Fi signals, Bluetooth signals, and walkie-talkie signals. The transmission power is the output power after adjustment by a power amplifier. The beam direction, beamwidth, and sidelobe suppression are all beamforming parameters used to focus the interference signal to a specific direction or area to enhance the transmission efficiency and quality of the interference signal.

[0118] The historical blocking logs and historical risk assessment reports mentioned above undergo preprocessing including at least the following: deduplication, missing value handling, error correction, data format unification, data type conversion, key feature extraction, derived feature creation, normalization, and standardization.

[0119] Duplicate data removal involves checking for identical records in the historical shielding log. If duplicates are found, they are deleted, leaving only one copy. Missing value handling involves checking each field in the historical shielding log for missing values. For critical fields (such as signal frequency and shielding mode), if there are many missing values, these records may be deleted. If there are few missing values, they can be filled using other data (such as the average, median, or mode). Error data correction involves checking whether the data in the historical shielding log conforms to logical and format requirements, such as whether the unit of signal frequency is correct and whether the shielding intensity is within a reasonable range. Data that does not meet the requirements is corrected or deleted. Data format consistency involves ensuring that all data fields in the historical shielding log have the same format, for example, unifying all time fields to "YYYY-MM-DD". The "HH:MM:SS" format unifies the signal frequency to MHz units, etc.; data type conversion converts data into a data type suitable for subsequent processing, such as converting signal strength from string to numeric type; key feature extraction extracts features related to shielding effect and shielding efficiency, such as the matching degree between signal frequency and shielding frequency range, and the comparison between signal strength and shielding strength; derived feature creation generates new features based on existing data, such as calculating the shielding duration based on shielding time and shielding duration, and calculating the shielding coverage based on shielding location and shielding area size; normalization normalizes numerical data to between 0 and 1; standardization makes the mean of the data 0 and the standard deviation 1 to eliminate the dimensional differences between different features.

[0120] The preprocessed historical blocking logs are annotated with at least the actual signal blocking strategy and the actual blocking effect. The preprocessed historical risk assessment reports are annotated with the actual risk assessment reports. A dataset is constructed based on the annotated historical blocking logs and historical risk assessment reports.

[0121] By systematically acquiring and integrating historical shielding logs (including signal parameters, shielding parameters, strategies, etc.) and historical risk assessment reports (including risk level, score, type, etc.), the core elements of signal shielding (such as signal frequency, strength, modulation method, shielding mode, scenario, etc.) are covered. This multi-dimensional data integration provides rich contextual information, far exceeding the one-sided data that existing systems usually rely on (such as only recording basic shielding parameters), and can more comprehensively capture the complexity of actual scenarios. By covering multiple shielding modes (such as full frequency band, specified frequency band, adaptive), the solution can adapt to different application scenarios (such as public safety), improve the pertinence and success rate of shielding decisions, and reduce the risks or failures caused by data omissions.

[0122] Through a series of preprocessing operations (such as deduplication, missing value handling, error correction, data format unification, normalization, standardization, key feature extraction, and derivative feature creation), the data cleaning and transformation process is systematized, solving common data quality problems in existing technologies (such as noise, inconsistency, or messy formatting), ensuring data consistency and comparability. High-quality preprocessing reduces the risk of bias in subsequent model training and improves prediction accuracy (such as masking effect evaluation or risk prediction). For example, normalization and standardization facilitate the convergence of machine learning algorithms, while derivative feature creation (such as generating new indicators based on signal strength and time combinations) can uncover potential patterns and support more efficient AI application development.

[0123] Step S30 specifically includes:

[0124] Step S31: Divide the dataset into a first training set, a first validation set, and a first test set based on a preset first ratio, and divide the dataset into a second training set, a second validation set, and a second test set based on a preset second ratio;

[0125] By dividing the dataset into two different preset ratios (first ratio and second ratio) for the two models (masking strategy generation model and risk assessment model), data redundancy and overfitting risks are avoided. This customized division ensures that each model obtains targeted training and validation data, improves the generalization performance of the models, and reduces cross-interference between models, thereby enhancing the robustness of the overall system.

[0126] Step S32: Train the masking strategy generation model using the first training set until the loss value of the strategy loss function is less than a preset first loss threshold; calculate the accuracy, recall, precision, and inference time using the first validation set to perform first-level validation on the trained masking strategy generation model. During the validation process, continuously optimize the hyperparameters of the masking strategy generation model, and then combine weight pruning, channel pruning, weight quantization, and activation quantization to compress the masking strategy generation model that has passed the first-level validation. Then, calculate the accuracy, recall, precision, and inference time again to perform second-level validation on the compressed masking strategy generation model; calculate the F1 score, generalization ability index, and robustness index using the first test set to test the masking strategy generation model that has passed the second-level validation.

[0127] The risk assessment model is trained using the second training set until the loss value of the risk loss function is less than a preset second loss threshold. Accuracy, recall, precision, and inference time are calculated using the second validation set to perform a first-level validation of the trained risk assessment model. During the validation process, the hyperparameters of the risk assessment model are continuously optimized. Weight pruning, channel pruning, weight quantization, and activation quantization are then combined to compress the risk assessment model that passes the first-level validation. Accuracy, recall, precision, and inference time are then calculated again to perform a second-level validation of the compressed risk assessment model. The F1 score, generalization ability index, and robustness index are calculated using the second test set to test the risk assessment model that passes the second-level validation.

[0128] Parameter pruning reduces the number of parameters in a model by removing less important parameters. Weight pruning removes weights with smaller absolute values ​​based on their magnitude or importance; channel pruning removes convolutional layer channels that contribute little to the output, reducing computational cost. Quantization converts the model's weights and activation functions from floating-point numbers (such as 32-bit floating-point numbers) to low-precision representations (such as 8-bit integers or 16-bit floating-point numbers) to significantly reduce storage requirements and computational complexity. In practice, weight quantization converts weights from 32-bit floating-point to 8-bit integers; activation quantization converts the output of the activation function from floating-point to low-precision representation.

[0129] If the verification or testing of the shielding strategy generation model and risk assessment model fails, the corresponding training set is expanded and training continues.

[0130] By employing first-level validation (including hyperparameter optimization) and second-level validation (re-evaluating metrics after compression) after training, combined with final testing (calculating metrics such as F1 score, generalization ability, and robustness), a rigorous iterative mechanism is formed, significantly improving model reliability (e.g., preventing model failure after deployment) and inference efficiency (compression reduces latency). Specific advantages include: integrated compression techniques (weight pruning, channel pruning, weight quantization, activation quantization) to achieve model lightweighting (reducing model size and computational overhead), while second-level validation ensures that compressed performance metrics (such as accuracy, recall, and inference time) remain within acceptable ranges, reducing hardware resource requirements (such as the processing burden of signal jammers) and supporting deployment in resource-constrained environments; comprehensive performance evaluation, using multiple metrics (accuracy, recall, precision, F1 score, generalization ability, robustness) for evaluation, providing multi-dimensional monitoring (e.g., F1 score balances precision and recall), reducing model blind spots, and enhancing the system's robustness in noisy or adversarial environments.

[0131] By separating the shielding strategy generation model and the risk assessment model (using their respective datasets for training, validation, and testing), parallel development and validation are supported. This independence reduces dependency errors between models (such as data leaks or hyperparameter conflicts) and accelerates iteration speed (e.g., two models can be optimized simultaneously), thereby improving the overall system's R&D efficiency and fault isolation capabilities.

[0132] Step S33: Using containerization technology, the tested shielding strategy generation model and the first runtime environment of the shielding strategy generation model are encapsulated in a first container, and the tested risk assessment model and the second runtime environment of the risk assessment model are encapsulated in a second container. The independent first container and the second container are deployed to the signal jammer, and the scheduling priority and communication mechanism of the shielding strategy generation model and the risk assessment model are set.

[0133] In real-time, the shielding strategy generation model and risk assessment model, isolated by containerization technology, can communicate through container networks, host networks, port mapping, shared storage, message queues, API calls, and other methods.

[0134] By dividing the dataset into independent parts according to a preset ratio, and combining multi-level validation, model compression, and containerized deployment, efficient and reliable closed-loop model management is achieved. The dual-ratio data division ensures the independence of model training and generalization ability. The rigorous two-level validation process (including hyperparameter optimization and pruning, quantization compression) significantly improves inference efficiency while ensuring core indicators such as accuracy and recall. Containerization of the model and environment and setting up a scheduling mechanism not only achieves resource isolation and rapid deployment, but also enhances the system's robustness and real-time response capability on edge devices such as signal jammers, ultimately achieving a lightweight, low-latency, and easy-to-maintain industrial-grade AI deployment effect.

[0135] Step S40 specifically includes:

[0136] Step S41: After the signal jammer is powered on, it establishes a heartbeat connection with the server through the preset frequency band and sends the power-on time, device ID and real-time location of the device to the server.

[0137] Step S42: The mobile terminal pre-creates a pair of device public keys and device private keys based on the RSA algorithm, and pre-installs the device public key into the server; the server pre-creates a pair of root public keys and root private keys based on the ECDH algorithm, and pre-installs the root public key into the mobile terminal and the signal jammer.

[0138] The mobile terminal obtains the current first timestamp, account, password, and PIN code, generates a first random number, performs a hash calculation on the first timestamp, account, password, and first random number to obtain a first hash value, encrypts the first timestamp, account, password, first random number, and first hash value using the device private key to obtain authentication encrypted data, generates an authentication request based on the authentication encrypted data, and uploads it to the server through the preset frequency band based on the TLS protocol.

[0139] Step S43: The server parses the received authentication request to obtain authentication encrypted data, decrypts the authentication encrypted data using the device public key to obtain a first timestamp, account, password, first random number and first hash value, performs integrity verification on the first timestamp, account, password and first random number using the first hash value, performs timeliness verification on the first timestamp, and then performs legality verification on the account and password.

[0140] If the verification passes, the current second timestamp is obtained. The second random number is obtained by extracting the second to last digit from the first timestamp and adding the first random number. The second hash value is obtained by hashing the second timestamp and the second random number. The second timestamp, the second random number, and the second hash value are encrypted with the root private key to obtain the response encrypted data. The response feedback is generated based on the response encrypted data and sent to the mobile terminal through the preset frequency band based on the TLS protocol.

[0141] Step S44: The mobile terminal parses the received response feedback to obtain response encrypted data, decrypts the response encrypted data with the root public key to obtain a second timestamp, a second random number and a second hash value, performs integrity verification on the second timestamp and the second random number with the second hash value, performs timeliness verification with the second timestamp, and then verifies the second random number by adding the second to last digit of the first timestamp to the first random number.

[0142] If the verification passes, the third random number is obtained by extracting the third digit from the second timestamp and adding it to the second random number. The first session key is then derived from the root public key, PIN code and the third random number using the KDF function. The first session key is then uploaded to the server.

[0143] Step S45: The server matches the corresponding PIN code from the preset user management table based on the account, extracts the third-to-last digit from the second timestamp and adds it to the second random number to obtain a third random number, and derives a second session key based on the root public key, PIN code, and third random number using the KDF function. The server then verifies the first session key using the second session key and sends the key negotiation result back to the mobile terminal to complete the authentication operation. When the first session key and the second session key are the same, authentication is successful.

[0144] Step S46: The mobile terminal obtains real-time blocking parameters, including at least the blocking mode, blocking scenario, blocking frequency range, blocking strength, blocking area size, blocking duration, and blocking location. It obtains the current third timestamp, calculates the first MAC value of the real-time blocking parameters, account, device ID, and third timestamp using the HMAC algorithm, encrypts the real-time blocking parameters, account, device ID, third timestamp, and first MAC value into session encrypted content using the first session key, and generates a blocking command based on the session encrypted content and sends it to the server.

[0145] Step S47: The server decrypts the encrypted session content carried by the blocking instruction using the second session key to obtain the real-time blocking parameters, account, device ID, third timestamp, and first MAC value. After performing integrity and timeliness checks using the first MAC value and the third timestamp, the server generates a blocking operation record based on the account, device ID, and third timestamp. The server encrypts the blocking operation record into an encrypted record using the root private key and uploads it to the blockchain. The server obtains the first block information stored in the encrypted record from the blockchain and shares the first block information with the mobile terminal using the second session key.

[0146] Step S48: The server encrypts the second session key using the root private key to obtain an encryption key, and sends the encryption key and the blocking instruction to the corresponding signal jammer through a preset frequency band based on the device ID.

[0147] By employing multi-layered encryption mechanisms (such as RSA and ECDH for asymmetric encryption, KDF for key derivation, and HMAC for message authentication), the confidentiality and integrity of data transmission are ensured. For example, mobile terminals use their device private keys to encrypt authentication data, while servers use their device public keys to decrypt and perform hash verification, effectively preventing man-in-the-middle attacks and data tampering. Timestamp verification (such as the timeliness verification of the first and second timestamps) further resists replay attacks, ensuring the real-time nature of requests. This hybrid encryption strategy is more robust than a single algorithm, reduces security vulnerabilities, and improves the overall system's resistance to attacks.

[0148] Session keys are derived using KDF functions, based on the root public key, PIN code, and dynamically generated random numbers (such as numbers extracted from timestamps), ensuring the uniqueness and temporality of session keys and avoiding the risk of static key leakage, while simplifying the key negotiation process. The server and mobile terminal independently derive session keys (first session key and second session key), and ensure consistency through verification, improving the efficiency of end-to-end encryption. The session key derivation process is efficient and secure, reducing computational overhead, supporting high-frequency communication (such as heartbeat connections and command issuance), while ensuring the transience of session keys, conforming to the zero-trust security model.

[0149] By extracting numbers from timestamps (e.g., extracting the second to last digit from the first timestamp) to generate random numbers (e.g., the second random number), the source of randomness and unpredictability are increased. Combined with hash calculation, the security of the authentication process is enhanced. In other words, the innovative combination of timestamp characteristics avoids the weaknesses of traditional pseudo-random number generators and improves the system's dynamic anti-cracking capabilities.

[0150] By integrating multi-layered encryption mechanisms (RSA, ECDH, KDF, HMAC), dynamic timestamp verification, and blockchain technology, the security and reliability of the signal jamming system are significantly improved. It employs asymmetric encryption and end-to-end session key negotiation (such as based on PIN codes and random number-derived keys) to ensure the confidentiality of authentication and command transmission. It utilizes timestamps to dynamically generate random numbers and combines them with hash verification to effectively defend against replay attacks and data tampering. Simultaneously, it enhances audit tracing capabilities by using blockchain to immutably store operation records. The entire process relies on the TLS protocol for efficient communication, supports remote customization of jamming parameters by users, and balances security protection with flexible control, making it suitable for real-time jamming needs in highly sensitive scenarios.

[0151] Step S50 specifically involves:

[0152] The signal jammer receives the jamming command and encryption key sent by the server, decrypts the encryption key with the preset root public key to obtain the second session key, and decrypts the session encryption content carried by the jamming command with the second session key to obtain the real-time jamming parameters, account, device ID, third timestamp and first MAC value. Integrity verification and timeliness verification are performed by the first MAC value and the third timestamp respectively.

[0153] The signal jammer acquires real-time wireless communication signals, preprocesses the real-time wireless communication signals, and then inputs the real-time wireless communication signals and real-time jamming parameters into a deployed jamming strategy generation model. The jamming strategy generation model uses hardware acceleration technology to infer a real-time signal jamming strategy, generates an interference signal based on the real-time signal jamming strategy, and transmits it externally through a power amplifier and a radio frequency antenna to perform signal jamming operations.

[0154] By employing multi-layered encryption and decryption and dual verification mechanisms (MAC value integrity verification + timestamp validity verification), the absolute security and reliability of the shielding command transmission are ensured. At the same time, relying on a locally deployed intelligent model (optimized by hardware acceleration), the wireless communication signal is analyzed in real time and a precise signal shielding strategy is dynamically generated, achieving millisecond-level adaptive interference signal transmission. This significantly improves shielding efficiency and targeting while greatly reducing server load, network dependence, and the risk of false interference from non-target signals. Furthermore, the operation authorization traceability is strengthened through account / device ID binding. Ultimately, a modern signal shielding system that is secure, reliable, resource-efficient, intelligent, precise, and remotely dynamically controllable is constructed.

[0155] Step S60 specifically involves:

[0156] The signal jammer records real-time blocking logs, including at least real-time wireless communication signals, real-time blocking parameters, real-time signal blocking strategies, real-time blocking time, real-time operation accounts, and real-time blocking locations. The real-time blocking logs are input into a deployed risk assessment model. The risk assessment model uses hardware acceleration technology to infer and obtain a real-time risk assessment report. It communicates with the blocking strategy generation model through a set communication mechanism, and then dynamically adjusts the real-time signal blocking strategy based on the real-time risk assessment report.

[0157] The signal jammer constructs an incremental dataset based on the real-time jamming logs and real-time risk assessment reports, and optimizes the jamming strategy generation model and risk assessment model using the incremental dataset.

[0158] By collecting multi-dimensional real-time blocking logs and using hardware acceleration technology for dynamic risk assessment, adaptive closed-loop optimization of signal blocking strategies was achieved, significantly improving the system's response speed and blocking accuracy. At the same time, based on incremental data, the model was continuously optimized, enhancing its robustness and anti-interference capabilities against new signal threats.

[0159] Step S70 specifically involves:

[0160] After completing the blocking work, the signal jammer decrypts the second session key obtained by decrypting the encryption key issued by the server, and encrypts the real-time blocking log and real-time risk assessment report into log ciphertext and report ciphertext respectively. The log ciphertext and report ciphertext are then uploaded to the server through a preset frequency band. The server decrypts the log ciphertext and report ciphertext using the second session key generated during the key negotiation process to obtain the real-time blocking log and real-time risk assessment report.

[0161] The server uses a pre-created root private key to encrypt the real-time blocking log and the real-time risk assessment report into encrypted logs and encrypted reports, respectively, and uploads them to the blockchain. It then obtains the second block information stored in the encrypted logs and the third block information stored in the encrypted reports from the blockchain feedback, and shares the second block information and the third block information with the mobile terminal through the second session key.

[0162] By employing a multi-layered encryption mechanism (session key and root private key working together), immutable blockchain storage, and a real-time key management process, the security, integrity, and traceability of signal shielding data are significantly improved. Dynamic session keys are used to encrypt and upload real-time shielding logs and real-time risk assessment reports. The server then encrypts the data a second time using the root private key before uploading it to the blockchain, ensuring that transmission and storage are leak-proof and tamper-proof throughout the entire process. Combined with the design of optimizing transmission efficiency in preset frequency bands and sharing block information with mobile terminals, highly reliable real-time monitoring, remotely verifiable auditing, and user-friendly operation are simultaneously achieved, forming a closed-loop solution that balances data protection, system performance, and compliance requirements.

[0163] Step S80 specifically involves:

[0164] The mobile terminal obtains the encrypted records, encrypted logs, and encrypted reports from the blockchain using the first, second, and third block information shared by the server. It then decrypts these encrypted records, logs, and reports using a pre-set root public key to obtain the signal jammer operation record, real-time jamming log, and real-time risk assessment report. Based on these records, the mobile terminal manages the signal jammer. Each of these documents carries the signal jammer's device ID.

[0165] Blockchain ensures the immutability and traceability of signal jammer operation data, while a pre-set root public key securely decrypts encrypted records, logs, and reports, significantly improving data integrity and access security. Mobile terminals, based on decrypted jammer operation records, real-time jamming logs, and real-time risk assessment reports, enable dynamic monitoring and rapid response to signal jammers, effectively reducing maintenance delays and manual intervention costs. The server's pre-shared block information mechanism optimizes data transmission efficiency, and the centralized management model combined with blockchain's encryption features provides an efficient and reliable equipment management solution for critical sectors, ensuring user privacy and compliance.

[0166] By dynamically generating precise signal shielding strategies through deep neural networks and combining multi-source feature fusion with a real-time risk assessment closed loop, the adaptability and security of signal shielding are significantly improved. Blockchain-encrypted evidence storage of shielding device operation records, real-time shielding logs, and real-time risk assessment reports ensures data immutability. Lightweight model compression, hardware acceleration, and containerized deployment enable low-latency edge computing. At the same time, a multi-layered dynamic key derivation and spatiotemporal random number authentication mechanism is designed to strengthen end-to-end security. Ultimately, without relying on the cloud, a comprehensive breakthrough is achieved in shielding accuracy, risk resistance, resource efficiency, and audit credibility.

[0167] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for managing signal blockers in combination with neural networks and blockchain, the method comprising: Includes the following steps: Step S10: Create a shielding strategy generation model for outputting a signal shielding strategy based on wireless communication signals and shielding parameters, and a risk assessment model for outputting a risk assessment report based on shielding logs. Set the strategy loss function of the shielding strategy generation model and the risk loss function of the risk assessment model. Step S20: Obtain a large number of historical blocking logs and historical risk assessment reports, and construct a dataset after preprocessing and labeling the historical blocking logs and historical risk assessment reports; the historical blocking logs include at least historical wireless communication signals, historical blocking parameters, historical signal blocking strategies, historical blocking times, historical operation accounts, and historical blocking locations; Step S30: Train and compress the shielding strategy generation model and risk assessment model using the dataset, and deploy the trained shielding strategy generation model and risk assessment model to the signal jammer; Step S40: After the signal jammer is powered on, the mobile terminal performs an authentication operation with the server through a preset frequency band. After successful authentication, the server sends a jamming command to the signal jammer. The server generates a jammer operation record based on the jamming command, encrypts the jammer operation record into an encrypted record, and uploads it to the blockchain. Step S50: The signal jammer verifies and parses the received jamming command to obtain real-time jamming parameters, collects real-time wireless communication signals, inputs the real-time wireless communication signals and real-time jamming parameters into the deployed jamming strategy generation model to obtain a real-time signal jamming strategy, generates interference signals based on the real-time signal jamming strategy and transmits them to the outside to perform signal jamming operations. Step S60: The signal jammer records a real-time jamming log that includes at least the real-time wireless communication signal, real-time jamming parameters, real-time signal jamming strategy, real-time jamming time, real-time operation account, and real-time jamming location. The real-time jamming log is input into the deployed risk assessment model to obtain a real-time risk assessment report. The real-time signal jamming strategy is dynamically adjusted based on the real-time risk assessment report. Step S70: The signal jammer encrypts and uploads the real-time jamming log and the real-time risk assessment report to the server. The server encrypts the received real-time jamming log and the real-time risk assessment report into encrypted logs and encrypted reports respectively and uploads them to the blockchain. Step S80: The mobile terminal obtains the encrypted records, encrypted logs, and encrypted reports through the blockchain, and manages the signal jammer based on the encrypted records, encrypted logs, and encrypted reports; Step S80 specifically involves: The mobile terminal obtains the encrypted records, encrypted logs, and encrypted reports from the blockchain through the first block information, second block information, and third block information shared by the server. It decrypts the encrypted records, encrypted logs, and encrypted reports using a preset root public key to obtain the jammer operation record, real-time jamming log, and real-time risk assessment report. Based on the jammer operation record, real-time jamming log, and real-time risk assessment report, the mobile terminal manages the signal jammer. 2.The signal shield management method combining neural network and blockchain of claim 1, wherein: In step S10, the shielding strategy generation model is constructed based on a signal feature extraction module, a parameter feature extraction module, a feature fusion module, and a prediction output module. The signal feature extraction module is constructed based on a first fully connected layer, a second fully connected layer, and a ReLU activation unit. The first fully connected layer is used to extract initial signal features from the wireless communication signal, including at least frequency features, bandwidth features, intensity features, modulation features, phase features, signal-to-noise ratio features, and device density features, and to map each of the initial signal features to a high-dimensional space. The ReLU activation unit is used to introduce nonlinearity into the initial signal features in the high-dimensional space to obtain nonlinear features. The second fully connected layer is used to reduce the dimensionality and abstract the nonlinear features to obtain high-level signal features; The parameter feature extraction module is constructed based on an embedding unit, a connection unit, and a third fully connected layer. The embedding unit is used to encode the shielding mode and shielding scenario in the shielding parameters to obtain a mode vector and a scenario vector. The connection unit is used to convert the shielding frequency range, shielding intensity, shielding area size, shielding duration, and shielding position in the shielding parameters into a frequency range vector, a shielding intensity vector, an area size vector, a shielding duration vector, and a shielding position vector, and concatenates the mode vector, scenario vector, frequency range vector, shielding intensity vector, area size vector, shielding duration vector, and shielding position vector into a concatenated vector. The third fully connected layer is used to integrate and normalize the concatenated vector and extract high-level parameter features. The feature fusion module is constructed based on a splicing unit and a fourth fully connected layer; the splicing unit is used to splice high-level signal features and high-level parameter features to obtain spliced ​​features; the fourth fully connected layer is used to perform deep fusion of spliced ​​features to generate a comprehensive feature vector; The prediction output module is constructed based on a shielding target prediction unit, a frequency band coverage prediction unit, a scanning speed prediction unit, a transmit power prediction unit, a beam direction prediction unit, a beamwidth prediction unit, a sidelobe suppression prediction unit, a shielding priority prediction unit, and a joint output unit; each of the shielding target prediction unit, frequency band coverage prediction unit, scanning speed prediction unit, transmit power prediction unit, beam direction prediction unit, beamwidth prediction unit, sidelobe suppression prediction unit, and shielding priority prediction unit is constructed from a fifth fully connected layer and an activation function layer; The shielding target prediction unit is used to infer the shielding target from the integrated feature vector; the frequency band coverage prediction unit is used to infer the frequency band coverage from the integrated feature vector; the scanning speed prediction unit is used to infer the scanning speed from the integrated feature vector; the transmit power prediction unit is used to infer the transmit power from the integrated feature vector; the beam direction prediction unit is used to infer the beam direction from the integrated feature vector; the beamwidth prediction unit is used to infer the beamwidth from the integrated feature vector; the sidelobe suppression prediction unit is used to infer the sidelobe suppression from the integrated feature vector; the shielding priority prediction unit is used to infer the shielding priority from the integrated feature vector; and the joint output unit is used to output a signal shielding strategy carrying the shielding target, frequency band coverage, scanning speed, transmit power, beam direction, beamwidth, sidelobe suppression, and shielding priority. The formula for the policy loss function is: L1=w1*L target +w2*L coverage +w3*L speed +w4*L power +w5*L direction +w6*L width +w7*L suppression +w8*L priority ; Where L1 represents the loss value of the policy loss function; L target The sub-loss representing the masked target prediction unit is the classification cross-entropy loss; L coverage The sub-loss of the frequency band coverage prediction unit is represented by the mean square error loss; L speed L represents the sub-loss of the scan rate prediction unit, using mean squared error loss; power The sub-loss of the transmit power prediction unit is represented by the mean square error loss; L direction The sub-loss of the beam direction prediction unit is represented by the mean square error loss; L width The sub-loss of the beamwidth prediction unit is represented by the mean square error loss; L suppression The sub-loss of the sidelobe suppression prediction unit is represented by the mean squared error loss; L priority The sub-loss of the masked priority prediction unit is represented by the classification cross-entropy loss; w1, w2, w3, w4, w5, w6, w7, and w8 all represent weight coefficients. 3.The signal shield management method combining neural network and blockchain of claim 1, wherein: In step S10, the risk assessment model is constructed based on a multi-source feature extraction layer, a dynamic fusion layer, and an assessment report output layer. The multi-source feature extraction layer is constructed based on a wireless communication signal feature extraction module, a shielding parameter feature extraction module, a policy feature extraction module, and a spatiotemporal feature extraction module. The wireless communication signal feature extraction module extracts local spectral features from the wireless communication signal using a 1D convolutional network and extracts signal temporal dependency features from the wireless communication signal using a bidirectional gated recurrent unit, outputting wireless communication signal features including the local spectral features and the signal temporal dependency features. The shielding parameter feature extraction module converts categorical parameters in the shielding parameters into embedding vectors using an embedding layer and fuses numerical parameters in the shielding parameters with the embedding vectors using a sixth fully connected layer to obtain shielding parameter features. The policy feature extraction module extracts the dependencies between policy parameters from the signal shielding policy using a multi-head self-attention mechanism to obtain policy features. The spatiotemporal feature extraction module extracts spatiotemporal features, including time periodicity features and geographical location features, from the shielding time, operating account, and shielding location using a time encoder and a spatial location encoder. The dynamic fusion layer is used to calculate the attention weights of wireless communication signal features, shielding parameter features, policy features, and spatiotemporal features through a gated attention unit, and then weighted and fused to obtain fused features. The assessment report output layer is constructed based on the risk assessment module, the strategy generation module, and the report output module; the risk assessment module is used to infer the fused features through the seventh fully connected layer and the ReLU activation function to obtain the risk level, risk score, and risk type. The strategy generation module is used to reason about the fusion features and signal shielding strategies through a conditional generative adversarial network to obtain the strategy adjustment content. The report output module is used to output a risk assessment report that includes risk level, risk score, risk type, and strategy adjustment content. The formula for the risk loss function is: L2 = λ1 * L risk +λ2*L strategy +λ3*L GA ; Where L2 represents the loss value of the risk loss function; L risk Indicates risk assessment loss; L strategy Indicates the loss due to strategy adjustment; L GA λ1, λ2, and λ3 represent the gated attention regularization loss; λ1, λ2, and λ3 all represent weight coefficients. ； in, Represents the cross-entropy loss function; Indicates the true risk level label; This represents the probability distribution of the predicted risk level. Represents the mean square error function; Indicates the true risk score label; Indicates the predicted risk score; Represents the binary cross-entropy function; Indicates the true risk type label; This indicates the probability of predicting a risk type; ； in, This represents the original signal shielding strategy; This indicates the adjusted signal shielding strategy; Represents the JavaScript divergence function; This represents the distribution of real historical signal masking strategies; This represents the distribution of signal shielding strategies synthesized by a conditional generative adversarial network. ； wherein, denotes the attention weight of the dynamic fusion layer; denotes the L2 norm. 4.The method of claim 1, wherein: Step S20 specifically involves: Obtain a large amount of historical blocking logs and historical risk assessment reports; the historical blocking logs include at least historical wireless communication signals, historical blocking parameters, historical signal blocking strategies, historical blocking times, historical operation accounts, and historical blocking locations; the historical risk assessment reports include at least historical risk levels, historical risk scores, historical risk types, and historical strategy adjustment content. The historical wireless communication signals include at least signal frequency, signal bandwidth, signal strength, modulation scheme, phase, signal-to-noise ratio, and channel state information; the historical shielding parameters include at least shielding mode, shielding scenario, shielding frequency range, shielding strength, shielding area size, shielding duration, and shielding location; the shielding modes include at least full-band shielding mode, specified frequency band shielding mode, scenario-adaptive shielding mode, and custom shielding mode; the historical signal shielding strategy includes at least shielding target, frequency band coverage, scanning speed, transmit power, beam direction, beamwidth, sidelobe suppression degree, and shielding priority. The historical blocking logs and historical risk assessment reports mentioned above undergo preprocessing including at least the following: deduplication, missing value handling, error correction, data format unification, data type conversion, key feature extraction, derived feature creation, normalization, and standardization. The preprocessed historical blocking logs are annotated with at least the actual signal blocking strategy and the actual blocking effect. The preprocessed historical risk assessment reports are annotated with the actual risk assessment reports. A dataset is constructed based on the annotated historical blocking logs and historical risk assessment reports.

5. The method of claim 1, wherein: Step S30 specifically includes: Step S31: Divide the dataset into a first training set, a first validation set, and a first test set based on a preset first ratio, and divide the dataset into a second training set, a second validation set, and a second test set based on a preset second ratio; Step S32: Train the masking strategy generation model using the first training set until the loss value of the strategy loss function is less than a preset first loss threshold; calculate the accuracy, recall, precision, and inference time using the first validation set to perform first-level validation on the trained masking strategy generation model. During the validation process, continuously optimize the hyperparameters of the masking strategy generation model, and then combine weight pruning, channel pruning, weight quantization, and activation quantization to compress the masking strategy generation model that has passed the first-level validation. Then, calculate the accuracy, recall, precision, and inference time again to perform second-level validation on the compressed masking strategy generation model; calculate the F1 score, generalization ability index, and robustness index using the first test set to test the masking strategy generation model that has passed the second-level validation. The risk assessment model is trained using the second training set until the loss value of the risk loss function is less than a preset second loss threshold. Accuracy, recall, precision, and inference time are calculated using the second validation set to perform a first-level validation of the trained risk assessment model. During the validation process, the hyperparameters of the risk assessment model are continuously optimized. Weight pruning, channel pruning, weight quantization, and activation quantization are then combined to compress the risk assessment model that passes the first-level validation. Accuracy, recall, precision, and inference time are then calculated again to perform a second-level validation of the compressed risk assessment model. The F1 score, generalization ability index, and robustness index are calculated using the second test set to test the risk assessment model that passes the second-level validation. Step S33: Using containerization technology, the tested shielding strategy generation model and the first runtime environment of the shielding strategy generation model are encapsulated in a first container, and the tested risk assessment model and the second runtime environment of the risk assessment model are encapsulated in a second container. The independent first container and the second container are deployed to the signal jammer, and the scheduling priority and communication mechanism of the shielding strategy generation model and the risk assessment model are set.

6. The method of claim 1, wherein: Step S40 specifically includes: Step S41: After the signal jammer is powered on, it establishes a heartbeat connection with the server through the preset frequency band and sends the power-on time, device ID and real-time location of the device to the server. Step S42: The mobile terminal pre-creates a pair of device public keys and device private keys based on the RSA algorithm, and pre-installs the device public key into the server; the server pre-creates a pair of root public keys and root private keys based on the ECDH algorithm, and pre-installs the root public key into the mobile terminal and the signal jammer. The mobile terminal obtains the current first timestamp, account, password, and PIN code, generates a first random number, performs a hash calculation on the first timestamp, account, password, and first random number to obtain a first hash value, encrypts the first timestamp, account, password, first random number, and first hash value using the device private key to obtain authentication encrypted data, generates an authentication request based on the authentication encrypted data, and uploads it to the server through the preset frequency band based on the TLS protocol. Step S43: The server parses the received authentication request to obtain authentication encrypted data, decrypts the authentication encrypted data using the device public key to obtain a first timestamp, account, password, first random number and first hash value, performs integrity verification on the first timestamp, account, password and first random number using the first hash value, performs timeliness verification on the first timestamp, and then performs legality verification on the account and password. If the verification passes, the current second timestamp is obtained. The second random number is obtained by extracting the second to last digit from the first timestamp and adding the first random number. The second hash value is obtained by hashing the second timestamp and the second random number. The second timestamp, the second random number, and the second hash value are encrypted with the root private key to obtain the response encrypted data. The response feedback is generated based on the response encrypted data and sent to the mobile terminal through the preset frequency band based on the TLS protocol. Step S44: The mobile terminal parses the received response feedback to obtain response encrypted data, decrypts the response encrypted data with the root public key to obtain a second timestamp, a second random number and a second hash value, performs integrity verification on the second timestamp and the second random number with the second hash value, performs timeliness verification with the second timestamp, and then verifies the second random number by adding the second to last digit of the first timestamp to the first random number. If the verification passes, the third random number is obtained by extracting the third digit from the second timestamp and adding it to the second random number. The first session key is then derived from the root public key, PIN code and the third random number using the KDF function. The first session key is then uploaded to the server. Step S45: The server matches the corresponding PIN code from the preset user management table based on the account, extracts the third-to-last digit from the second timestamp and adds the second random number to obtain the third random number, derives the second session key based on the root public key, PIN code and the third random number using the KDF function, verifies the first session key using the second session key, and feeds back the key negotiation result to the mobile terminal to complete the authentication operation. Step S46: The mobile terminal obtains real-time blocking parameters, including at least the blocking mode, blocking scenario, blocking frequency range, blocking strength, blocking area size, blocking duration, and blocking location. It obtains the current third timestamp, calculates the first MAC value of the real-time blocking parameters, account, device ID, and third timestamp using the HMAC algorithm, encrypts the real-time blocking parameters, account, device ID, third timestamp, and first MAC value into session encrypted content using the first session key, and generates a blocking command based on the session encrypted content and sends it to the server. Step S47: The server decrypts the encrypted session content carried by the blocking instruction using the second session key to obtain the real-time blocking parameters, account, device ID, third timestamp, and first MAC value. After performing integrity and timeliness checks using the first MAC value and the third timestamp, the server generates a blocking operation record based on the account, device ID, and third timestamp. The server encrypts the blocking operation record into an encrypted record using the root private key and uploads it to the blockchain. The server obtains the first block information stored in the encrypted record from the blockchain and shares the first block information with the mobile terminal using the second session key. Step S48: The server encrypts the second session key using the root private key to obtain an encryption key, and sends the encryption key and the blocking instruction to the corresponding signal jammer through a preset frequency band based on the device ID.

7. The method of claim 1, wherein: Step S50 specifically involves: The signal jammer receives the jamming command and encryption key sent by the server, decrypts the encryption key with the preset root public key to obtain the second session key, and decrypts the session encryption content carried by the jamming command with the second session key to obtain the real-time jamming parameters, account, device ID, third timestamp and first MAC value. Integrity verification and timeliness verification are performed by the first MAC value and the third timestamp respectively. The signal jammer acquires real-time wireless communication signals, preprocesses the real-time wireless communication signals, and then inputs the real-time wireless communication signals and real-time jamming parameters into a deployed jamming strategy generation model. The jamming strategy generation model uses hardware acceleration technology to infer a real-time signal jamming strategy, generates an interference signal based on the real-time signal jamming strategy, and transmits it externally through a power amplifier and a radio frequency antenna to perform signal jamming operations. 8.The method of claim 1, wherein: Step S60 specifically involves: The signal jammer records real-time blocking logs, including at least real-time wireless communication signals, real-time blocking parameters, real-time signal blocking strategies, real-time blocking time, real-time operation accounts, and real-time blocking locations. The real-time blocking logs are input into a deployed risk assessment model. The risk assessment model uses hardware acceleration technology to infer and obtain a real-time risk assessment report. It communicates with the blocking strategy generation model through a set communication mechanism, and then dynamically adjusts the real-time signal blocking strategy based on the real-time risk assessment report. The signal jammer constructs an incremental dataset based on the real-time jamming logs and real-time risk assessment reports, and optimizes the jamming strategy generation model and risk assessment model using the incremental dataset. 9.The method of claim 1, wherein: Step S70 specifically involves: After completing the blocking work, the signal jammer decrypts the second session key obtained by decrypting the encryption key issued by the server, and encrypts the real-time blocking log and real-time risk assessment report into log ciphertext and report ciphertext respectively. The log ciphertext and report ciphertext are then uploaded to the server through a preset frequency band. The server decrypts the log ciphertext and report ciphertext using the second session key generated during the key negotiation process to obtain the real-time blocking log and real-time risk assessment report. The server uses a pre-created root private key to encrypt the real-time blocking log and the real-time risk assessment report into encrypted logs and encrypted reports, respectively, and uploads them to the blockchain. It then obtains the second block information stored in the encrypted logs and the third block information stored in the encrypted reports from the blockchain feedback, and shares the second block information and the third block information with the mobile terminal through the second session key.

Images