Radio signal monitoring and positioning method and system based on big data

Abstract

This invention discloses a method and system for radio signal monitoring and positioning based on big data, belonging to the field of radio spectrum monitoring technology. This invention constructs an integrated air-space-ground multi-source monitoring network, processes massive signal data in a layered manner through edge-cloud collaboration, corrects non-line-of-sight and multipath effects using a spatiotemporal knowledge graph, and combines a fully automatically updated dynamic fingerprint database to achieve TDOA-AOA-RSSI-fingerprint multi-source fusion positioning, complemented by closed-loop incremental learning and full-process data security protection. This invention effectively eliminates monitoring blind spots, significantly improves positioning accuracy and signal processing efficiency in complex environments, reduces fingerprint database maintenance costs, achieves an abnormal signal identification accuracy of ≥98.5%, and provides outdoor positioning accuracy of 8.2m and indoor accuracy of 2.6m. It is applicable to scenarios such as spectrum management, illegal signal crackdown, and civil aviation frequency band protection, demonstrating significant engineering practical value.

Description

Technical Field

[0001] This invention relates to the field of radio spectrum monitoring and signal location technology, and in particular to a radio signal monitoring and location method and system based on big data. Background Technology

[0002] Radio spectrum is a scarce and non-renewable strategic natural resource for the nation. Radio signal monitoring and positioning technology is a core technological support for maintaining order in the airwaves, controlling illegal signal interference, ensuring communication security in key industries, and enabling spectrum-based law enforcement and evidence collection. With the large-scale popularization of technologies such as 5G / 6G mobile communication, the Internet of Things, low-altitude drones, and low-orbit satellites, the space electromagnetic environment is becoming increasingly complex. The number of signal transmission sources is exploding, modulation methods are becoming increasingly complex, and multipath effects in propagation scenarios are significant. This places higher demands on the application of full-area coverage acquisition of radio signals, detection of weak signals with low signal-to-noise ratios, and high-precision and stable positioning in complex environments.

[0003] Existing radio monitoring and positioning technologies can be mainly divided into three core technology routes:

[0004] Firstly, the monitoring network, which is mainly based on fixed monitoring stations and supplemented by mobile monitoring vehicles, has limitations due to terrain and building obstruction. Fixed stations have a large number of monitoring blind spots in densely populated urban areas, enclosed indoor spaces, mountainous areas, tunnels, and other scenarios. Mobile monitoring vehicles, on the other hand, have problems such as high deployment costs, insufficient real-time performance, and inability to achieve 24 / 7 full-area monitoring, making it difficult to achieve continuous signal acquisition without blind spots.

[0005] Secondly, passive positioning schemes based on TDOA (Time Difference of Arrival), AOA (Angle of Arrival), and RSSI (Received Signal Strength) have stringent requirements for clock synchronization accuracy and line-of-sight propagation conditions. In urban scenarios with non-line-of-sight propagation and severe multipath effects, the positioning accuracy drops significantly, making it difficult to meet the application needs of complex environments.

[0006] Third, fingerprint database-based positioning solutions rely on manual road testing for static construction, resulting in high initial deployment costs. Furthermore, fingerprint features are susceptible to environmental changes and become invalid, leading to extremely high maintenance costs and making it difficult to adapt to dynamically changing electromagnetic environments.

[0007] In addition, existing solutions mostly adopt a single-machine serial processing architecture, which has high processing latency for massive amounts of raw IQ data and spectrum data, making it difficult to meet the real-time detection requirements of sudden signals and instantaneous interference signals; signal recognition relies heavily on manually designed features, and the accuracy of recognition of weak signals and complex modulation signals is insufficient; at the same time, the security protection mechanism for monitoring data throughout the entire process of transmission, storage and use is imperfect, making it difficult to meet the application requirements of data compliance and law enforcement evidence collection.

[0008] In summary, existing technologies are insufficient to simultaneously meet the core requirements of radio signal full-area coverage acquisition, efficient real-time processing, high-precision and stable positioning, and low-maintenance-cost deployment in complex electromagnetic environments. There is an urgent need for a radio signal monitoring and positioning solution that is adaptable to multiple scenarios, highly reliable, and highly accurate. Summary of the Invention

[0009] Purpose of the invention

[0010] The purpose of this invention is to overcome the above-mentioned defects of the prior art and provide a radio signal monitoring and positioning method and system based on big data. This invention solves the core problems of large monitoring coverage blind spots, low efficiency of massive signal processing, poor positioning accuracy in complex environments, and high maintenance costs of fingerprint databases in the prior art. At the same time, it improves the intelligence level and data security protection capabilities of the solution, and realizes full-domain monitoring, real-time processing, accurate positioning and compliant management of radio signals.

[0011] Core technology concept

[0012] The core concept of this invention lies in addressing the critical pain points of radio monitoring and positioning in complex electromagnetic environments by constructing a complete technical system encompassing "integrated air-space-ground multi-source acquisition, edge-cloud collaborative hierarchical processing, spatiotemporal knowledge graph-assisted analysis, dynamic fingerprint and multi-source algorithm fusion positioning, and data closed-loop autonomous optimization." On one hand, a comprehensive complementary acquisition network is built using four types of monitoring nodes: fixed, mobile, UAV, and crowdsourced. Combined with an edge-cloud collaborative hierarchical preprocessing architecture, this achieves blind-spot-free signal acquisition and efficient, low-latency processing, solving the problems of insufficient coverage and low processing efficiency in traditional solutions. On the other hand, by constructing a spatiotemporal knowledge graph that integrates geographical environment and electromagnetic propagation characteristics, combined with a dynamic fingerprint database automatic update mechanism that eliminates the need for manual road testing, multi-source fusion positioning (TDOA-AOA-RSSI-fingerprint) is achieved, effectively overcoming the impact of non-line-of-sight and multipath effects on positioning, addressing the core pain points of poor positioning accuracy and high maintenance costs in traditional solutions. Simultaneously, an incremental learning mechanism for data closed-loop and a full-process data security protection system are constructed to improve the system's intelligence and compliance, adapting to the engineering application needs of various scenarios.

[0013] Core improvements

[0014] The core improvement of this invention compared to the prior art lies in:

[0015] 1. An integrated air-space-ground multi-source monitoring network and a big data processing architecture with edge-cloud collaboration were constructed. The complementary collection of four types of nodes achieved full coverage without blind spots. Data diversion was achieved through layered preprocessing at the edge and in the cloud, which greatly reduced the transmission bandwidth pressure and improved the processing efficiency and real-time performance of massive signals.

[0016] 2. A multi-source fusion positioning method based on spatiotemporal knowledge graph correction is proposed. Combined with a fully automatically updated dynamic fingerprint database mechanism, it effectively overcomes the negative impact of non-line-of-sight propagation and multipath effects on positioning, significantly improves positioning accuracy and stability in complex environments, and eliminates the need for manual road testing, thus significantly reducing the deployment and maintenance costs of the fingerprint database.

[0017] 3. A closed-loop autonomous evolution system for signal recognition and positioning has been constructed. Combined with a full-process data security protection mechanism, it can improve the intelligence level of the system while ensuring the integrity, confidentiality and compliance of monitoring data, and meet the engineering application needs of multiple scenarios.

[0018] Radio signal monitoring and location methods based on big data

[0019] The monitoring and positioning method of the present invention includes the following six core steps, and the entire process can be automated, as detailed below:

[0020] Step S1: Synchronous acquisition of multi-source signals from air, space, and ground.

[0021] A distributed integrated air-space-ground monitoring network was constructed, comprising four types of data acquisition units: fixed monitoring nodes, mobile monitoring nodes, UAV monitoring nodes, and crowdsourced terminal nodes. All nodes uniformly adopted GPS / BeiDou dual-mode time synchronization technology, and the time synchronization accuracy of the entire network after calibration was ≤1μs, providing a precise and unified time reference for subsequent positioning algorithms.

[0022] Fixed monitoring nodes are deployed at urban high points, communication towers, airport perimeters, and key control areas, responsible for 24 / 7 uninterrupted full-band signal acquisition in the commonly used 10MHz-6GHz frequency band. Mobile monitoring nodes are mounted on mobile platforms such as monitoring vehicles and ships, responsible for dynamic patrol and data collection in monitoring blind spots, urban edge areas, and temporary support areas, supporting on-the-go testing and real-time data transmission. Drone monitoring nodes are equipped with lightweight radio frequency monitoring terminals, responsible for filling blind spots in ground-obstructed areas, complex terrain areas, and areas difficult for personnel to reach, eliminating propagation blind spots in ground monitoring. Crowdsourced terminal nodes, with user authorization, collect lightweight RSSI and CSI feature data in indoor and densely populated street scenarios through ubiquitous terminals such as smartphones, IoT gateways, and vehicle-mounted T-BOX.

[0023] Each monitoring node receives radio signals from the air through a high-sensitivity radio frequency unit. After bandpass filtering, low-noise amplification, and high-speed ADC analog-to-digital conversion, multi-dimensional data such as IQ raw data, center frequency, signal amplitude, phase, RSSI, AOA, CSI, and time-spectrum characteristics are extracted. After being doubly encrypted using the AES-256 commercial encryption algorithm and the national cryptographic SM4 algorithm, the data is transmitted hierarchically to edge computing nodes and cloud big data centers through fiber optic, 5G, and satellite communication links.

[0024] Step S2: End-Edge-Cloud Collaborative Big Data Preprocessing

[0025] A layered collaborative processing model, employing real-time preprocessing at the edge and deep preprocessing in the cloud, is adopted to remove data noise, outliers, and redundant information, standardize data format, and generate a high-quality, standardized signal dataset. Specifically, this includes:

[0026] Real-time preprocessing at the edge: After decrypting the received encrypted data, the edge computing node first uses a 3-Sigma outlier detection algorithm to remove outlier data with amplitude and phase exceeding three times the standard deviation, as well as invalid data frames with a packet loss rate exceeding 5%. Then, it uses an adaptive Kalman filter algorithm to filter out noise from the raw IQ data, suppressing the influence of environmental thermal noise and equipment noise floor. After data cleaning, it extracts the basic time-domain and frequency-domain features of the signal, such as center frequency, bandwidth, power, and higher-order cumulative quantities. Only the cleaned valid data and feature data are encrypted and uploaded to the cloud, while invalid data is directly discarded at the edge, which can reduce the backhaul bandwidth pressure by more than 60%, and the single-node processing latency is ≤10ms.

[0027] Cloud-based deep preprocessing: Based on the Spark distributed computing framework, multi-source heterogeneous data is processed in parallel. First, data from different nodes and in different formats are standardized in terms of format, normalized in terms of values, and unified in terms of dimensions, converting structured signal parameters and unstructured spectrogram data into a unified computable format. Then, pre-stored environmental context data such as terrain elevation data, 3D building models, real-time meteorological data, and electromagnetic propagation loss models are fused to pre-correct for signal propagation attenuation and occlusion effects. Finally, a missing value imputation algorithm based on random forest is used to predict and impute missing feature data, providing highly reliable data support for subsequent signal analysis and positioning calculations.

[0028] Step S3: Spatiotemporal knowledge graph fusion and signal fingerprint construction

[0029] Spatiotemporal Knowledge Graph Construction: A spatiotemporal knowledge graph is constructed based on the Neo4j graph database, defining six categories of entities: monitoring nodes, buildings, topography, signal sources, electromagnetic propagation models, and environmental impact factors, as well as the spatial location relationships, occlusion relationships, and propagation impact relationships between entities. Among them, building entities store 3D coordinates, height, wall material, and electromagnetic attenuation coefficient attributes; topography entities store elevation data, vegetation cover type, and attenuation coefficient attributes; and electromagnetic propagation models store propagation loss calculation formulas and correction parameters for corresponding scenarios. During signal analysis and positioning calculations, prior knowledge is invoked based on the signal acquisition location and frequency band, and the weights of different features are dynamically adjusted through an attention mechanism to reduce the negative impacts of non-line-of-sight propagation and multipath effects.

[0030] Multi-dimensional signal feature extraction: For the preprocessed signal data, three types of core features are extracted: First, signal statistical features, including signal bandwidth, power spectral density, higher-order cumulants, and cyclic spectrum features; second, transform domain features, including time-frequency spectrum features obtained from short-time Fourier transform and time-frequency domain detail features extracted from wavelet packet transform; and third, contextual features, including signal acquisition time, acquisition node location, background noise level, and channel state information (CSI).

[0031] Robust signal fingerprint generation and anomaly detection: A stacked denoising autoencoder network is constructed, consisting of a 3-layer encoder and a 2-layer decoder. The encoder compresses the multi-dimensional input features layer by layer, and the decoder reconstructs the compressed features, aiming to minimize the reconstruction error to complete the network pre-training. The extracted multi-dimensional features are input into the pre-trained encoder, which outputs a 128-dimensional feature vector as the unique signal fingerprint of the signal. During the training process, random Gaussian noise is added to the input features of the stacked denoising autoencoder to improve the stability of the fingerprint in low signal-to-noise ratio and multipath interference scenarios, achieving accurate identification and repetitive recognition of the signal source. A baseline feature model of normal signals is established based on massive historical data statistics. Combined with the isolated forest anomaly detection algorithm, the deviation value between the current signal fingerprint and the baseline features is compared in real time. When the deviation exceeds a preset threshold, it is automatically judged as an abnormal signal. For sudden instantaneous signals, amplitude change threshold, frequency change threshold, and duration threshold are set. The response time of the abnormal signal is ≤50ms, and the recognition accuracy is ≥98%.

[0032] Step S4: Dynamic fingerprint database update and multi-source hybrid localization calculation

[0033] Monitoring area grid division and fingerprint database initialization: The monitoring area is finely divided into grids according to scene type. Outdoor open areas are divided into 50m×50m grid units, dense urban blocks are divided into 20m×20m grid units, and complex indoor areas are divided into 5m×5m grid units. Each grid unit stores the mean, variance, number of samples, and update time attributes of the signal fingerprint features at that location, thus completing the fingerprint database initialization.

[0034] The dynamic fingerprint database is automatically updated: Based on valid signal data and location tags uploaded by crowdsourced terminal nodes and monitoring nodes, a weighted average algorithm with a forgetting factor is used to update the fingerprint feature values ​​of the corresponding grid. The update formula is: F_new=α×F_old+(1-α)×F_new_sample, where F_new is the updated grid fingerprint feature value, F_old is the feature value before the update, F_new_sample is the newly collected sample feature value, and α is the forgetting factor, which ranges from 0.7 to 0.95 and is dynamically adjusted according to the sample update frequency and environmental change degree within the grid. At the same time, based on weather, population density, and building change information in the spatiotemporal knowledge graph, an environmental correction factor is introduced to correct the fingerprint feature values ​​in real time. When the number of valid samples in the grid exceeds the preset update threshold, the fingerprint database version is automatically updated without manual road testing.

[0035] Multi-source hybrid high-precision positioning calculation: For the target signal to be located, the monitoring node data that received the signal and whose signal-to-noise ratio was higher than a preset threshold were first selected. Multi-algorithm positioning calculation and error correction were then performed: Firstly, based on the selected node data, the time difference of signal arrival at different nodes was calculated using the TDOA algorithm, and the initial positioning coordinates of the target were obtained using the Chan algorithm; the angle of arrival of the signal at each node was calculated using the AOA algorithm, and the initial positioning coordinates of the target were obtained using the least squares method; based on the spatiotemporal knowledge graph, the non-line-of-sight propagation and occlusion between the target and the monitoring nodes were determined, and the initial positioning results of TDOA and AOA were corrected to reduce the impact of multipath effects; Secondly, the signal of the target signal to be located was extracted. The fingerprint is matched with the grid fingerprint features in the dynamic fingerprint database, and the K-nearest neighbor algorithm is used to obtain the fingerprint localization result. At the same time, based on the RSSI data of each node, combined with the corrected electromagnetic propagation loss model, the RSSI localization result is calculated. Thirdly, according to the confidence of the results of different localization algorithms, a confidence-weighted fusion algorithm is used to fuse the localization results of TDOA, AOA, fingerprint, and RSSI. Among them, the weight of TDOA and AOA results is increased in the line-of-sight propagation scenario, and the weight of fingerprint matching results is increased in the non-line-of-sight and dense occlusion scenarios, and finally the accurate localization coordinates of the target are output. For the moving target signal, the unscented Kalman filter algorithm is used to smooth the localization coordinate sequence to achieve continuous and stable tracking.

[0036] Step S5: Visualize the results and provide tiered early warnings

[0037] The system uses a 3D GIS electronic map visualization platform to display real-time electromagnetic situation information across the entire region, including signal spectrum waveforms, spectrum waterfall plots, signal types, modulation methods, positioning coordinates, moving target trajectories, and monitoring node status. Based on the intensity, impact range, and severity of abnormal signals, the system categorizes them into three levels: general anomalies, relatively severe anomalies, and serious anomalies. It triggers alerts through various methods, such as optical alarms, SMS push notifications, APP warnings, and platform pop-ups, to notify management personnel for appropriate action. The system also supports functions such as historical data querying, statistical report generation, data export, and remote parameter configuration.

[0038] Step S6: Closed-Loop Intelligent Evolution and Security Management

[0039] The location results obtained from on-site verification, law enforcement investigations, and manual verification, as well as abnormal signal confirmation data and environmental change information, are uniformly fed back to the big data center as incremental training samples. This allows for incremental updates and fine-tuning of the CNN-LSTM deep learning model, signal fingerprint database, and spatiotemporal knowledge graph, enabling the system to autonomously evolve to adapt to changes in the electromagnetic environment and continuously improve performance. All raw data, feature data, location results, and operation logs are backed up in a distributed, off-site manner. Crowdsourced data undergoes hash desensitization and differential privacy processing to strictly prevent user privacy leaks and meet the requirements of relevant data security laws and regulations.

[0040] Radio signal monitoring and positioning system based on big data

[0041] The system of this invention adopts a three-layer collaborative computing architecture of end-edge-cloud, including a distributed air-space-ground monitoring network on the end side, an edge computing node cluster, and a cloud service platform. Each part realizes data interaction and collaborative work through a high-speed communication network, as detailed below:

[0042] 1. Distributed air-space-ground monitoring network

[0043] The distributed air-space-ground monitoring network is the core data acquisition unit of the system, including four types of acquisition units: fixed monitoring nodes, mobile monitoring nodes, UAV monitoring nodes, and crowdsourced terminal nodes. Each monitoring node integrates a high-sensitivity radio frequency receiving unit, a high-speed analog-to-digital conversion unit, a multi-mode time synchronization unit, an encrypted data transmission unit, and a local cache unit.

[0044] The radio frequency receiving unit adopts the AD9361 high-performance radio frequency front-end, covering the 10MHz-6GHz frequency band, with a receiving sensitivity of ≤-120dBm and equipped with a low-noise amplifier to improve the reception capability of weak signals; the high-speed analog-to-digital conversion unit has a sampling rate of ≥100MSps, a resolution of ≥12 bits, and supports orthogonal sampling to complete high-precision conversion of analog signals to digital signals; the multi-mode time synchronization unit supports GPS / BeiDou dual-mode time synchronization with a synchronization accuracy of ≤1μs to ensure time uniformity across all network nodes; the encrypted data transmission unit supports fiber optic + 5G + satellite tri-mode transmission, supports local caching during network outages and automatic resumption of transmission after network recovery, and has a data transmission success rate of ≥99.8%.

[0045] 2. Edge computing node cluster

[0046] The edge computing node cluster is deployed on the base station side and in the regional equipment room. It includes multiple lightweight edge computing units. Each edge computing unit corresponds to a monitoring node in the access area and is responsible for real-time decryption, data cleaning, noise filtering, preliminary feature extraction and abnormal signal detection of the received signal data. The processed effective feature data is uploaded to the cloud service platform to realize the hierarchical and distributed processing of data, reduce the computing and bandwidth pressure on the cloud, and the single node processing latency is ≤10ms. It can process more than 100 signal data at the same time.

[0047] 3. Cloud service platform

[0048] The cloud service platform is the core computing, storage, analysis and interaction unit of the system, including a big data processing center, a spatiotemporal knowledge graph module, a dynamic fingerprint management module, a location computing module, a big data storage center, a visualization interaction platform and a privacy and security module.

[0049] Big Data Processing Center: Adopting the Spark+Hadoop distributed computing framework, it integrates a data preprocessing module, a CNN-LSTM deep learning module, and a signal classification and anomaly recognition module. Among them, the data preprocessing module completes the standardization, normalization, fusion processing, and missing value imputation of multi-source data; the deep learning module automatically extracts deep features of signals in the time domain, frequency domain, and transform domain to achieve intelligent recognition of signal type and modulation mode; the anomaly recognition module combines the isolated forest algorithm to achieve accurate detection of weak signals, burst signals, and illegal interference signals.

[0050] Spatiotemporal Knowledge Graph Module: Built and operated based on the Neo4j graph database, it stores prior knowledge such as geographic information, 3D building models, electromagnetic propagation characteristics, environmental context data, and historical propagation patterns, providing environmental constraints and parameter corrections for signal recognition, anomaly detection, and location calculation.

[0051] Dynamic fingerprint management module: responsible for the grid division of the monitoring area, generation and storage of signal fingerprints, fully automatic dynamic update and environmental correction of the fingerprint database, and version management. It adopts a weighted fusion algorithm with a forgetting factor to achieve autonomous update of the fingerprint database without human intervention.

[0052] The positioning calculation module includes a TDOA positioning unit, an AOA positioning unit, an RSSI positioning unit, a fingerprint matching unit, a multi-source fusion unit, and a trajectory smoothing unit. It adopts a confidence-weighted fusion algorithm combined with Kalman filtering to achieve high-precision indoor and outdoor positioning and continuous tracking of moving targets.

[0053] Big Data Storage Center: Adopts HDFS distributed storage architecture and SSD+HDD tiered storage strategy. Hot real-time data is stored in high-speed SSD storage devices, while historical archived data is stored in high-capacity HDD storage devices. It supports automatic data backup, fault recovery, and lifecycle management. Storage capacity can be smoothly expanded from 100TB to PB level.

[0054] Visual interactive platform: Developed based on WebGIS and 3D map technology, it supports access from both PC and mobile devices, and provides functions such as real-time spectrum display, location trajectory annotation, anomaly classification and early warning, historical data query, statistical report generation, and remote node control.

[0055] Privacy and security module: Responsible for data security and privacy protection throughout the entire system process. It adopts AES-256 / SM4 dual encryption transmission, data hash desensitization, differential privacy processing, RBAC role-based access control, and blockchain evidence storage technology to ensure data security and compliance and meet the needs of law enforcement evidence collection.

[0056] Beneficial effects

[0057] Compared with the prior art, the present invention has achieved the following significant technical effects:

[0058] First, this invention utilizes an integrated air-space-ground multi-source monitoring network, combining four types of acquisition nodes to achieve complementary coverage, effectively reducing coverage blind spots in traditional monitoring schemes. In typical application scenarios, the effective monitoring coverage rate can reach over 95%, significantly improving the deployment flexibility and coverage capability of the monitoring network. Combined with a layered processing architecture that integrates edge and cloud, edge-side preprocessing greatly reduces invalid data backhaul, while cloud-based distributed parallel computing enables efficient processing of massive signals. The average processing latency for a single signal can be as low as 42ms, stably supporting concurrent processing of thousands of signals. Compared to traditional serial processing schemes, the processing efficiency is improved by more than 10 times.

[0059] Secondly, this invention constructs a spatiotemporal knowledge graph that integrates geographical environment and electromagnetic propagation characteristics, providing environmental prior constraints and error correction for positioning calculations. Combined with the TDOA-AOA-RSSI-fingerprint multi-source fusion positioning algorithm, it effectively overcomes the impact of non-line-of-sight propagation and multipath effects on positioning accuracy. In typical scenarios, the average outdoor positioning accuracy can reach 8.2m, and the average indoor positioning accuracy can reach 2.6m, which is more than 80% higher than the traditional single TDOA positioning scheme, significantly improving the accuracy and stability of positioning in complex electromagnetic environments. At the same time, the dynamic fingerprint database fully automatic update mechanism can realize real-time updates and environmental correction of the fingerprint database without manual road testing, reducing fingerprint database maintenance costs by more than 90% and significantly improving the environmental adaptability and engineering feasibility of the solution.

[0060] Furthermore, this invention improves the identification capability of weak and complex modulated signals through a robust signal fingerprint extraction scheme based on deep learning, achieving an abnormal signal identification accuracy of over 98.5% and a false negative rate of ≤1.5%. Through a data-closed-loop incremental learning mechanism, the model, fingerprint database, and knowledge graph are autonomously optimized, allowing system performance to continuously improve with the accumulation of operational data, eliminating the need for frequent manual upgrades and maintenance. Simultaneously, a comprehensive data security and privacy protection mechanism ensures the integrity, confidentiality, and immutability of monitoring data, making it suitable for various application scenarios such as radio spectrum management, combating illegal broadcasting and fake base stations, drone control, and civil aviation navigation frequency band protection, demonstrating significant engineering practicality and promotional value. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating the big data-based radio signal monitoring and positioning method of the present invention.

[0062] Figure 2 This is a schematic diagram of the edge-cloud architecture of the big data-based radio signal monitoring and positioning system of the present invention.

[0063] Figure 3 This is a schematic diagram of the edge-cloud collaborative preprocessing and signal fingerprint construction process in this invention;

[0064] Figure 4 This is a schematic diagram of the process of dynamic fingerprint database update, multi-source fusion localization, and closed-loop evolution in this invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0066] This embodiment takes the radio spectrum management of a prefecture-level city and the security protection of the navigation frequency band of a civil aviation airport as a typical application scenario. The total monitoring area is about 500 km², covering a variety of complex scenarios such as the city's core business district, high-density residential areas, industrial parks, suburban mountainous areas, large indoor shopping malls, and airport airspace protection zones. The core requirement is to achieve normalized monitoring of legitimate communication signals and to achieve intelligent identification, accurate positioning, and rapid alarm handling of illegal broadcasts, fake base stations, drone interference signals, and abnormal interference signals in the civil aviation frequency band. This verifies the feasibility and performance of the technical solution of this invention.

[0067] 1. System deployment and hardware / software configuration

[0068] (1) Deployment of distributed air-space-ground monitoring network

[0069] Ten fixed monitoring nodes are deployed within the monitoring area, including six at communication towers and high-rise buildings in the urban core area, two around the airport airspace protection zone, and two at high points in suburban mountainous areas, achieving 24 / 7 uninterrupted monitoring across the entire 10MHz-6GHz frequency band. Five mobile monitoring nodes are deployed, equipped with dedicated vehicle-mounted monitoring equipment and powered by a 12V vehicle power supply, responsible for dynamic patrol and data collection in urban fringe areas and blind spots, as well as mobile monitoring in temporary emergency support areas. Three drone monitoring nodes are configured, equipped with lightweight portable monitoring terminals, with a high-altitude operating radius of ≥5km, responsible for supplementing data collection in areas obscured by tall buildings and complex terrain in mountainous areas, eliminating blind spots in ground monitoring. With user authorization, 200 crowdsourced terminal nodes are connected, including authorized smartphones, IoT gateways, and vehicle-mounted T-BOX devices, mainly covering indoor areas and densely populated urban blocks, collecting lightweight RSSI and CSI feature data, which are then uploaded after anonymization.

[0070] All of the above monitoring nodes are equipped with GPS / BeiDou dual-mode timing modules, and the time synchronization accuracy of the entire network is calibrated to ≤1μs. All collected data are transmitted after being doubly encrypted with AES-256 and national cryptographic SM4.

[0071] (2) Edge and cloud computing platform configuration

[0072] Edge computing nodes are deployed in the equipment rooms of five core base stations within the monitoring area. Each edge node corresponds to a fixed or crowdsourced monitoring node within the access area, responsible for real-time data preprocessing and initial anomaly detection in the corresponding area. A server cluster is deployed in the cloud, including three master servers and five slave servers. Each server is configured with an Intel Xeon E5 series CPU and more than 32GB of memory, and deploys the Spark 3.3.0+Hadoop 3.3.4 distributed computing framework and the TensorFlow 2.8.0 deep learning framework. The total storage capacity is 100TB, using an SSD+HDD tiered storage strategy. The system software environment uses the Ubuntu Server 20.04 LTS operating system, and is equipped with Neo4j graph database, MySQL 8.0 structured database, and Redis cache database. The visualization and interactive platform is developed based on WebGIS and a 3D map engine.

[0073] 2. Specific implementation process

[0074] The implementation process of radio signal monitoring and positioning in this embodiment is as follows: the monitoring nodes of the entire network perform signal acquisition according to the preset configuration, the edge computing nodes receive the acquired data of the corresponding area, and after completing real-time data cleaning, noise reduction and preliminary feature extraction, the effective feature data is uploaded to the cloud; the cloud completes the standardization, fusion processing and pre-correction of multi-source data to generate a standardized signal dataset.

[0075] Based on the geographic information, 3D building data, and electromagnetic propagation model of the monitoring area, a spatiotemporal knowledge graph of the entire domain is constructed, covering entity and attribute data of 320 high-rise buildings, 12 urban main roads, and 5 types of terrain within the monitoring area. For the preprocessed signal data, multi-dimensional features are extracted and a 128-dimensional robust signal fingerprint is generated. Combined with the normal signal baseline model and the isolated forest algorithm, real-time detection and alarm of abnormal signals are realized.

[0076] The monitoring area is divided into grids according to scenarios. Based on crowdsourced data and real-time monitoring data, the grid fingerprint feature values ​​are automatically updated using a weighted average algorithm with a forgetting factor to complete the real-time maintenance of the dynamic fingerprint database. For detected abnormal target signals, valid monitoring node data is selected to complete the initial positioning and error correction of TDOA and AOA. Combining fingerprint matching and RSSI positioning results, the final positioning coordinates are output through confidence-weighted fusion. The trajectory of moving targets is smoothed by unscented Kalman filtering.

[0077] The system displays the electromagnetic situation and abnormal signal information of the entire domain in real time through a visual interactive platform. It triggers multi-channel hierarchical early warning for serious abnormal signals and notifies management personnel to handle the situation. The signal data and positioning results after on-site verification and law enforcement confirmation are fed back to the big data center to complete the incremental updates and autonomous optimization of the model, fingerprint database and knowledge graph.

[0078] 3. Performance Test Results

[0079] In this embodiment, the system ran stably and continuously for 30 days. The core performance test results are as follows:

[0080] Monitoring coverage performance: Effective monitoring coverage rate within the monitoring area ≥95%, detection rate of weak signals with power ≤-110dBm ≥95%, center frequency measurement error ≤0.1MHz, signal amplitude measurement error ≤0.5dB;

[0081] Signal recognition performance: The accuracy rate of abnormal signal recognition reaches 98.5%, the false negative rate is 1.2%, the false positive rate is 0.7%, and the average response time to sudden instantaneous signals is ≤45ms;

[0082] Positioning accuracy performance: In outdoor scenarios, 97% of the samples had a positioning accuracy of ≤10m, with an average positioning accuracy of 8.2m; in indoor scenarios, 98% of the samples had a positioning accuracy of ≤3m, with an average positioning accuracy of 2.6m; the trajectory fluctuation of moving target tracking was ≤0.3m.

[0083] Data processing performance: The average processing latency for a single signal is 42ms, and it can stably support the concurrent processing of more than 1000 signals. The data query response time is ≤1s.

[0084] Data transmission and security performance: The data transmission success rate reaches 99.8%, and the entire process is encrypted to prevent data leakage and tampering. Blockchain evidence storage can meet the compliance requirements of law enforcement evidence collection.

[0085] 4. System maintenance and optimization

[0086] Hardware maintenance: Monthly calibration of fixed node antennas and synchronization modules; quarterly testing of power supply, transmission and data acquisition performance of mobile / drone nodes; and semi-annual sampling verification of crowdsourcing terminals.

[0087] Software optimization: The big data framework and deep learning framework are updated quarterly, and the model parameters are fine-tuned monthly based on actual operating data to continuously optimize the interactive experience of the visualization platform;

[0088] Data management: Perform local data backups weekly, off-site backups monthly, and automatically clean up expired archived data every six months to free up storage resources and improve system operating efficiency;

[0089] Evolutionary optimization: Incremental learning is performed monthly to update the knowledge graph and fingerprint database based on changes in the electromagnetic environment, continuously improving the system's recognition rate, positioning accuracy, and environmental adaptability.

[0090] The present invention and its embodiments have been described above, and such description is not restrictive. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for monitoring and locating radio signals based on big data, characterized in that: Includes the following steps: S1. Construct an integrated air-ground monitoring network that includes four types of monitoring nodes: fixed, mobile, drone, and crowdsourced, to achieve synchronous acquisition of multi-source signals with a network-wide time synchronization accuracy of ≤1μs; S2. Adopting an edge-cloud collaborative mode of real-time preprocessing at the edge and deep preprocessing in the cloud to complete data cleaning, noise reduction and feature extraction; S3. Construct a spatiotemporal knowledge graph, extract multi-dimensional signal features to generate robust signal fingerprints, and realize abnormal signal detection; S4. The monitoring area is divided into grids, the dynamic fingerprint database is updated automatically, and a multi-source hybrid algorithm is used to achieve high-precision positioning. S5. 3D GIS visualization and hierarchical early warning; S6. Based on closed-loop incremental learning of verification data, provide full-process data security protection.

2. The method for monitoring and locating radio signals based on big data according to claim 1, characterized in that: The data collected in step S1 is double-encrypted with AES-256 and the national standard SM4, and transmitted in stages through optical fiber, 5G and satellite links, with a signal receiving sensitivity of ≤-120dBm.

3. The method for monitoring and locating radio signals based on big data according to claim 1, characterized in that: In step S2, 3-Sigma anomaly detection and adaptive Kalman filtering are used on the edge side, with a single node processing latency of ≤10ms; the cloud side uses the Spark distributed framework to fuse terrain, building and meteorological data to complete signal propagation pre-correction.

4. The radio signal monitoring and positioning method based on big data according to claim 1, characterized in that: In step S3, the spatiotemporal knowledge graph is constructed based on Neo4j and includes six entities: monitoring nodes, buildings, topography, signal sources, electromagnetic propagation models, and environmental impact factors. The signal fingerprint is generated as a 128-dimensional feature vector by a stacked denoising autoencoder.

5. The method for monitoring and locating radio signals based on big data according to claim 1, characterized in that: In step S3, the isolated forest algorithm is used for anomaly detection. The response time for sudden instantaneous signals is ≤50ms, and the accuracy of anomaly signal identification is ≥98%.

6. The method for monitoring and locating radio signals based on big data according to claim 1, characterized in that: In step S4, the dynamic fingerprint database is updated using a weighted average algorithm with a forgetting factor, and is autonomously corrected by an environmental correction factor, without the need for manual road testing.

7. The method for monitoring and locating radio signals based on big data according to claim 1, characterized in that: In step S4, multi-source hybrid positioning is fused with TDOA, AOA, RSSI, and fingerprint positioning results. Weights are dynamically assigned for line-of-sight / non-line-of-sight scenarios, and the trajectory of the moving target is smoothed by unscented Kalman filtering.

8. A radio signal monitoring and positioning system based on big data, characterized in that, It adopts a three-layer collaborative architecture of end-edge-cloud, including a distributed air-space-ground monitoring network, an edge computing node cluster, and a cloud service platform. The monitoring network contains four types of data collection nodes. The edge nodes are responsible for real-time preprocessing, and the cloud platform integrates big data processing, spatiotemporal knowledge graph, dynamic fingerprint management, positioning computing, storage, visualization, and security protection modules.

9. The system according to claim 8, characterized in that, The monitoring node integrates a GPS / BeiDou dual-mode synchronization unit and an AD9361 RF front-end, with a sampling rate ≥100MSps and a resolution ≥12 bits.

10. The system according to claim 8, characterized in that, The cloud storage adopts an HDFS distributed architecture and a tiered storage strategy of SSD and HDD. The security module uses dual encryption, hash desensitization, differential privacy, and blockchain evidence storage to achieve full-process data protection.

Images