Intelligent frequency spectrum monitoring terminal

Abstract

The invention discloses an intelligent frequency spectrum monitoring terminal, and relates to the technical field of electromagnetic frequency spectrums, the intelligent frequency spectrum monitoring terminal comprises a frequency spectrum acquisition access module, an intelligent identification processing module and an edge cooperation management and control module, the frequency spectrum acquisition access module is electrically connected with a plurality of frequency spectrum terminals and is used for completing terminal access, time reference establishment and link asymmetry correction; the intelligent identification processing module is electrically connected with the frequency spectrum acquisition access module and is used for carrying out multi-resolution time-frequency feature construction, target and interference type identification and online learning self-adaption on acquired data, and the edge collaborative management and control module is electrically connected with the intelligent identification processing module. The frequency spectrum acquisition access module is used for performing measurement fusion positioning of time difference of arrival, angle of arrival, frequency difference of arrival and received signal strength, and performing task arrangement, block verification retransmission and result consistency control, the frequency spectrum acquisition access module comprises a time service module, a loopback correction module, a common visual calibration module and a clock monitoring module, and the frequency spectrum acquisition access module has the characteristic of stable operation.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic spectrum technology, specifically to an intelligent spectrum monitoring terminal. Background Technology

[0002] Existing spectrum monitoring terminals mostly employ a distributed deployment of fixed and portable stations. The terminal side performs radio frequency sampling and time-frequency analysis, while the central side provides graphical display and alarms. Some solutions use TDOA (Time Difference of Arrival) to estimate the location of interference sources and suspicious drones. Due to the complex deployment environment, large number of terminals, and poor link conditions, common problems in engineering implementation include: insufficient time synchronization accuracy of multiple stations, TDOA baseline bias caused by link asymmetry, location divergence when using TDOA alone in NLOS scenarios, terminal-side model aging and recognition drift, insufficient consistency of multi-station results and retransmission verification, and lack of unified orchestration and traceability mechanisms for grid-based operation and maintenance.

[0003] Traditional solutions often employ a centralized data processing model, resulting in uncorrected clock and link asymmetry across multiple stations, uncontrollable TDOA errors, instability of a single TDOA system under low-altitude multipath and obstruction conditions, difficulty in maintaining accuracy of terminal-side identification models over long periods, lack of consistency control, block verification, and retransmission for multi-terminal collaboration, and deficiencies in task orchestration, bandwidth management, and reliable log recording at the operation and maintenance level, making large-scale stable operation difficult. Therefore, designing a stable and operational intelligent spectrum monitoring terminal is essential. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent spectrum monitoring terminal to solve the problems mentioned in the background art.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: an intelligent spectrum monitoring terminal, comprising a spectrum acquisition and access module, an intelligent identification and processing module, and an edge collaborative management and control module. The spectrum acquisition and access module is electrically connected to multiple spectrum terminals and is used to complete terminal access, time reference establishment, and link asymmetry correction. The intelligent identification and processing module is electrically connected to the spectrum acquisition and access module and is used to construct multi-resolution time-frequency features, identify target and interference types, and perform online learning and adaptive processing on the acquired data. The edge collaborative management and control module is electrically connected to the intelligent identification and processing module and is used to perform measurement, fusion, and positioning of time difference of arrival, angle of arrival, frequency difference of arrival, and received signal strength, as well as to perform task orchestration, block verification and retransmission, and result consistency control.

[0006] According to the above technical solution, the spectrum acquisition and access module includes a timing module, a loopback correction module, a common visual calibration module, and a clock monitoring module. The timing module is used to establish a unified time reference with hardware timestamps in an Ethernet environment. The loopback correction module is used to measure the uplink and downlink delay difference and perform online correction of asymmetric offset. The common visual calibration module is used to calculate the position and time deviation of multiple stations based on a common visible signal source and generate a baseline matrix. The clock monitoring module is used to perform threshold detection on the timing residual and trigger recalibration and alarm when abnormal.

[0007] The intelligent recognition processing module includes a preprocessing module, a texture extraction module, a lightweight inference module, and an online learning module. The preprocessing module is used to perform bandpass, denoising, and short-time Fourier wavelet transform on the radio frequency signal data to generate a multi-resolution time-frequency map. The texture extraction module is used to extract radio frequency texture features such as frequency offset, radio frequency signal asymmetry, transient texture, and phase noise. The lightweight inference module is used to run a quantized lightweight model on the terminal side to output the target category and confidence level. The online learning module is used to perform small-step incremental fine-tuning on high-confidence samples and calibrate and align them with the central analysis model.

[0008] The edge collaborative management and control module includes a fusion calculation module, a trajectory tracking module, a task orchestration module, an edge computing verification module, and a consistency tracking module. The fusion calculation module is used to construct a cost function by weighting the arrival time difference, angle of arrival, frequency difference of arrival, and received signal strength observations according to confidence level and to calculate the target position. The trajectory tracking module is used to filter and predict the target state on the time series and output an error ellipse. The task orchestration module is used to generate a scheduling strategy based on terminal load, link occupancy, and model temperature and limit the concurrency. The edge computing verification module is used to perform edge computing verification on the sent and returned data and automatically retransmit in case of anomalies. The consistency tracking module is used to determine the consistency of multi-station positioning results and generate a signed log to achieve end-to-end traceability.

[0009] According to the above technical solution, the working steps of the monitoring terminal include:

[0010] S1. The spectrum terminal powers on, completes self-test, and connects to the monitoring network via Ethernet. It performs hardware time synchronization to establish a unified time reference, performs link back-and-forth loop measurement to correct uplink and downlink asymmetry, completes common visual signal source calibration to generate baseline matrix and site calibration parameters, and the management terminal issues sampling plan, center frequency point bandwidth gain trigger threshold and buffering strategy.

[0011] S2. The front-end tuning completes the downconversion and gain settings, the analog-to-digital conversion obtains the spectrum signal data and buffers it frame by frame, performs DC frequency offset correction and windowing segmentation, executes short-time Fourier transform and wavelet multi-resolution transform to generate time-frequency waterfall plot, extracts the frequency offset spectrum signal asymmetric transient texture phase noise RF texture summary, and hashes the data blocks and summaries by block and writes them into the transmission queue.

[0012] S3. Load a lightweight quantization model on the terminal side to identify target categories and interference types, output category confidence and key frequency band labels, associate and bind the identification results with the site timestamp baseline parameters and radio frequency texture, high confidence samples enter the online fine-tuning process to complete small step incremental updates, low confidence samples enter the retry and manual review queue, and abnormal frames trigger local buffering and rate limiting reporting.

[0013] S4: The management terminal gathers observations from multiple stations and calculates the time difference of arrival based on a unified time reference. It constructs a weighted cost function by combining the angle of arrival, frequency difference of arrival, and received signal strength to solve the position. It uses time series filtering to estimate the trajectory and outputs the error ellipse and target number. It fuses and associates the positioning results with the identification results of S3 to generate spatiotemporal linked target entries, supporting the dynamic addition and removal of mobile stations and temporarily added stations.

[0014] S5. Task orchestration uses terminal load, link occupancy model, and temperature for concurrent control and bandwidth management. Edge computing verification drives retransmission mechanisms to ensure data consistency. Disconnection and calibration anomalies trigger automatic reconnection and recalibration. Signature logs and status snapshots are generated and archived in the audit archive. Location alarm identification tags and spectrum waterfall are visualized in real time on the PC and management platform, supporting remote collaborative annotation, playback, and report export.

[0015] According to the above technical solution, S1 specifically refers to:

[0016] S1-1. After the terminal powers on and completes its self-test, it connects to the Ethernet. The hardware timestamp unit marks the transmitted and received packets at the nanosecond level in the network card and programmable logic layer. The management end periodically sends a time synchronization sequence to each terminal. The terminal compares its local clock with the received timestamp to obtain the instantaneous deviation. The system simultaneously performs back-and-forth loop measurement to separate the uplink and downlink delay difference. The uplink delay is denoted as Δu and the downlink delay is denoted as Δd. The link offset δ and the local clock drift rate γ are estimated from this. The time correction is updated in real time on the terminal using a sliding window recursive method according to t′=t-γt-δ. The management end synchronously maintains the time health index of each station. When the time health is lower than the threshold, local re-synchronization and loop retest are triggered. After completion, the new time base is written to the terminal time base register area and an acknowledgment frame is returned.

[0017] S1-2. The management terminal selects reference signal sources that can be simultaneously received by each station and issues calibration instructions. All stations synchronously collect reference signals at a unified trigger time and report local timestamps and station location parameters. Based on this, the management terminal estimates the fine-grained time deviation and geometric fine-tuning amount of each station, generates a baseline parameter set for collaborative positioning, and writes it back to the calibration area of ​​the terminal and management terminal as input for subsequent calculations. After entering online inspection, reference signals are sampled according to short cycles and baseline residuals are calculated. When the residual is below the threshold, only the health index is updated. When it approaches the threshold, rapid retesting and fine-tuning are triggered and a transition baseline is generated. When the threshold is exceeded or a topology change is detected, full recalibration is performed and the old and new baseline parameters are saved in a versioned manner to support rollback, ensuring that subsequent positioning continues to use a unified time reference and reliable baseline. The formula is: Station i time deviation Δt i =t si -t c -δ, where t si For the reference signal timestamp recorded at station i, t c To standardize the triggering time, the baseline residual r ij =|(t) si -Δt i )-(t sj -Δt j )-t ij |, where t sj The timestamp of the reference signal recorded at station j, Δt j For the time deviation of station j, t ij The theoretical time difference of arrival is calculated between the site geometry and the reference source location.

[0018] According to the above technical solution, S2 specifically refers to:

[0019] S2-1: The front end is tuned to the target center frequency and the bandwidth gain and trigger threshold are set. The analog-to-digital conversion continuously outputs the spectrum signal data and buffers it frame by frame. DC removal and frequency offset correction, windowing segmentation and overlapping splicing are performed. Multi-resolution processing generates short-time Fourier time-frequency diagram and wavelet time-frequency diagram. Transient suppression and out-of-band suppression are performed on strong pulses and broadband interference. An index is established for each frame according to the timestamp and station identifier and written into the acquisition buffer.

[0020] S2-2. Extract the radio frequency texture summary from the preprocessing results, including the transient texture index and phase noise spectrum of the frequency offset trajectory asymmetry. Combine the time-frequency map to generate a multi-channel feature tensor and bind it with the time reference baseline parameters. Divide the features and original segments into blocks according to a fixed data block size and calculate the hash check value. Establish a sending queue and a feedback queue shared by sending and receiving. Record the timestamp, frame number, station number, hash value and priority of each data block for subsequent edge recognition and cooperative positioning steady-state input and consistency verification.

[0021] According to the above technical solution, S3 specifically refers to:

[0022] S3-1. Load the quantization lightweight model on the terminal, read the multi-channel feature tensor and corresponding timestamp output by S2, perform forward inference to obtain the target category, interference type and confidence level, generate a unique identifier for each result by combining the station number and baseline parameters, extract key frequency bands and time-frequency positions and generate annotation masks, write them to the recognition result buffer and event trigger queue according to the result level, and at the same time establish a mapping index for the hash value, frame number and result number of the corresponding data block for subsequent localization fusion and consistency verification.

[0023] S3-2. For high-confidence samples, a small-step incremental update process is initiated, freezing the backbone layer and adjusting only the shallow layer and normalization parameters. A sliding window is used to control the update frequency and parameter drift amplitude. The model version number, checksum, and effect summary before and after the update are written to the local version record. For low-confidence and conflicting samples, a retry and feedback queue is initiated, triggering secondary inference and manual review strategies. If necessary, the model is rolled back to the previous stable version. Periodically, the model is calibrated and aligned with the central analysis model to correct inter-class boundary and domain biases, ensuring the recognition stability of the terminal under long-term operation. The status logs and indicators generated during the learning process are reported to the management terminal for global scheduling and strategy adjustment.

[0024] According to the above technical solution, S4 specifically refers to:

[0025] S4-1. The management end aggregates observations from multiple stations according to a unified time base, constructs a weighted cost function with the target position vector x as the independent variable, and minimizes the residuals of the time difference of arrival, angle of arrival, frequency difference of arrival, and received signal strength. The formula is J(x)=∑ k (w t t t 2 +w a t a 2 +w f r f 2 +w p r p 2 ), where r t To achieve the difference between time difference observations and geometric predictions, r a r is the difference between the observed angle of arrival and the geometric prediction. f The difference between the observed arrival frequency and the geometric prediction, r p The difference between the received signal strength and the path loss model, w t w a w f w pTo adaptively update the weight coefficients based on the inverse of the online confidence level or variance, iteratively minimize J(x) to obtain the current position estimate and residual statistics, generate the position estimate, error radius and target number, and write them into the positioning result buffer;

[0026] S4-2: Perform state prediction at fixed time steps, then correct with new positioning observations to generate the trajectory points and confidence intervals at that time; perform residual threshold judgment for each batch of observations, reduce weights and trigger resampling or relocation when anomalies occur, and mark continuous anomalies as unstable targets; maintain the target lifecycle and reassociation, handle the matching and anti-exchange of multiple target intersections, associate the trajectory with the identification results of S3 by timestamp and station number, and complete the category label and key frequency band annotation; perform consistency verification between multi-station results, remove abnormal observations of time reference or baseline, and trigger rapid recalibration and rollback when necessary; finally output trajectory polylines, error ellipses, quality scores and signature logs, and push them to the visualization and alarm channels.

[0027] According to the above technical solution, S5 specifically refers to:

[0028] S5-1: The management terminal summarizes the load, link occupancy and model temperature of each site, generates concurrency and priority strategies, and distributes the collection, identification, location and backhaul tasks in segments according to the queue. Dynamic rate limiting and segmented scheduling are used to avoid congestion. When a node goes offline, it triggers automatic reconnection and fast recalibration. Failed tasks are written back to the queue according to the number of retries and the cooldown time, and the reason code is recorded for subsequent parameter tuning.

[0029] S5-2 performs hash verification and retransmission control on uplink and downlink block data, performs cross-consistency judgment on cross-station positioning results and removes abnormal observations of time base or baseline, generates signature logs and status snapshots for successful results and stores them in the database, automatically generates handling orders for abnormal results and links alarm channels, and pushes trajectory polylines, error ellipses, category labels and quality scores to the visualization interface, supporting remote collaborative annotation, playback and report export.

[0030] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: While maintaining the theme of distributed spectrum monitoring + edge intelligence + cooperative positioning, this invention supplements time synchronization and link asymmetry correction, constructs measurement fusion of time difference of arrival + angle of arrival + frequency difference of arrival + received signal strength, and provides edge self-evolution identification and consistency control, thus achieving the following effects:

[0031] Baseline calibrable and clock accurate: Hardware timestamps are used for timing and loopback link correction, significantly reducing time difference of arrival (TDOA) system errors; more stable positioning: Based on TDOA, angle of arrival (ADR), frequency difference of arrival (FDR), and received signal strength are fused synchronously, maintaining convergence under non-line-of-sight conditions; more accurate identification: Multi-resolution time-frequency features and radio frequency textures are introduced, supporting online fine-tuning and calibration alignment, reducing long-term drift; stronger collaboration: Consistency control and edge computing verification are provided, automatic retransmission of anomalies is enabled, and results are traceable; manageable operation: Task orchestration, bandwidth management, and log signing are available to support grid-scale operation and maintenance. Attached Figure Description

[0032] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0033] Figure 1 This is a schematic diagram of the overall modular structure of the present invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Please see Figure 1 The present invention provides a technical solution: an intelligent spectrum monitoring terminal, including a spectrum acquisition and access module, an intelligent identification and processing module, and an edge collaborative management and control module. The spectrum acquisition and access module is electrically connected to multiple spectrum terminals and is used to complete terminal access, time reference establishment, and link asymmetry correction. The intelligent identification and processing module is electrically connected to the spectrum acquisition and access module and is used to construct multi-resolution time-frequency features, identify target and interference types, and perform online learning and adaptation on the acquired data. The edge collaborative management and control module is electrically connected to the intelligent identification and processing module and is used to perform measurement, fusion, and positioning of time difference of arrival, angle of arrival, frequency difference of arrival, and received signal strength, as well as to perform task orchestration, block verification and retransmission, and result consistency control.

[0036] The spectrum acquisition and access module includes a timing module, a loopback correction module, a common visual calibration module, and a clock monitoring module. The timing module is used to establish a unified time reference with hardware timestamps in an Ethernet environment. The loopback correction module is used to measure the uplink and downlink delay difference and perform online correction of asymmetric offset. The common visual calibration module is used to calculate the position and time deviation of multiple stations based on a common visible signal source and generate a baseline matrix. The clock monitoring module is used to perform threshold detection on the timing residual and trigger recalibration and alarm when abnormal.

[0037] The intelligent recognition and processing module includes a preprocessing module, a texture extraction module, a lightweight inference module, and an online learning module. The preprocessing module is used to perform bandpass, denoising, and short-time Fourier wavelet transform on the radio frequency signal data to generate multi-resolution time-frequency maps. The texture extraction module is used to extract radio frequency texture features such as frequency offset, radio frequency signal asymmetry, transient texture, and phase noise. The lightweight inference module is used to run a quantized lightweight model on the terminal side to output the target category and confidence level. The online learning module is used to perform small-step incremental fine-tuning on high-confidence samples and calibrate and align them with the central analysis model.

[0038] The edge collaborative management and control module includes a fusion calculation module, a trajectory tracking module, a task orchestration module, an edge computing verification module, and a consistency tracking module. The fusion calculation module is used to construct a cost function by weighting the arrival time difference, angle of arrival, frequency difference of arrival, and received signal strength observations according to confidence level and to calculate the target position. The trajectory tracking module is used to filter and predict the target state on the time series and output the error ellipse. The task orchestration module is used to generate a scheduling strategy based on terminal load, link occupancy, and model temperature and limit the concurrency. The edge computing verification module is used to perform edge computing verification on the sent and returned data and automatically retransmit in case of anomalies. The consistency tracking module is used to determine the consistency of multi-station positioning results and generate a signed log to achieve full-link traceability.

[0039] The working steps of the monitoring terminal include:

[0040] S1. The spectrum terminal powers on, completes self-test, and connects to the monitoring network via Ethernet. It performs hardware time synchronization to establish a unified time reference, performs link back-and-forth loop measurement to correct uplink and downlink asymmetry, completes common visual signal source calibration to generate baseline matrix and site calibration parameters, and the management terminal issues sampling plan, center frequency point bandwidth gain trigger threshold and buffering strategy.

[0041] S2. The front-end tuning completes the downconversion and gain settings, the analog-to-digital conversion obtains the spectrum signal data and buffers it frame by frame, performs DC frequency offset correction and windowing segmentation, executes short-time Fourier transform and wavelet multi-resolution transform to generate time-frequency waterfall plot, extracts the frequency offset spectrum signal asymmetric transient texture phase noise RF texture summary, and hashes the data blocks and summaries by block and writes them into the transmission queue.

[0042] S3. Load a lightweight quantization model on the terminal side to identify target categories and interference types, output category confidence and key frequency band labels, associate and bind the identification results with the site timestamp baseline parameters and radio frequency texture, high confidence samples enter the online fine-tuning process to complete small step incremental updates, low confidence samples enter the retry and manual review queue, and abnormal frames trigger local buffering and rate limiting reporting.

[0043] S4: The management terminal gathers observations from multiple stations and calculates the time difference of arrival based on a unified time reference. It constructs a weighted cost function by combining the angle of arrival, frequency difference of arrival, and received signal strength to solve the position. It uses time series filtering to estimate the trajectory and outputs the error ellipse and target number. It fuses and associates the positioning results with the identification results of S3 to generate spatiotemporal linked target entries, supporting the dynamic addition and removal of mobile stations and temporarily added stations.

[0044] S5. Task orchestration performs concurrent control and bandwidth management based on terminal load, link occupancy model, and temperature. Edge computing verification drives the retransmission mechanism to ensure data consistency. Disconnection and calibration anomalies trigger automatic reconnection and recalibration. Signature logs and status snapshots are generated and archived in the audit archive. Location alarm identification tags and spectrum waterfall are visualized in real time on the PC and management platform, supporting remote collaborative annotation, playback, and report export.

[0045] S1 specifically refers to:

[0046] S1-1. After the terminal powers on and completes its self-test, it connects to the Ethernet. The hardware timestamp unit marks the transmitted and received packets at the nanosecond level in the network card and programmable logic layer. The management end periodically sends a time synchronization sequence to each terminal. The terminal compares its local clock with the received timestamp to obtain the instantaneous deviation. The system simultaneously performs back-and-forth loop measurement to separate the uplink and downlink delay difference. The uplink delay is denoted as Δu and the downlink delay is denoted as Δd. The link offset δ and the local clock drift rate γ are estimated from this. The time correction is updated in real time on the terminal using a sliding window recursive method according to t′=t-γt-δ. The management end synchronously maintains the time health index of each station. When the time health is lower than the threshold, local re-synchronization and loop retest are triggered. After completion, the new time base is written to the terminal time base register area and an acknowledgment frame is returned.

[0047] S1-2. The management terminal selects reference signal sources that can be simultaneously received by each station and issues calibration instructions. All stations synchronously collect reference signals at a unified trigger time and report local timestamps and station location parameters. Based on this, the management terminal estimates the fine-grained time deviation and geometric fine-tuning amount of each station, generates a baseline parameter set for collaborative positioning, and writes it back to the calibration area of ​​the terminal and management terminal as input for subsequent calculations. After entering online inspection, reference signals are sampled according to short cycles and baseline residuals are calculated. When the residual is below the threshold, only the health index is updated. When it approaches the threshold, rapid retesting and fine-tuning are triggered and a transition baseline is generated. When the threshold is exceeded or a topology change is detected, full recalibration is performed and the old and new baseline parameters are saved in a versioned manner to support rollback, ensuring that subsequent positioning continues to use a unified time reference and reliable baseline. The formula is: Station i time deviation Δt i =t si -t c -δ, where t si For the reference signal timestamp recorded at station i, t c To standardize the triggering time, the baseline residual rij =|(t) si -Δt i )-(t sj -Δt j )-t ij |, where t sj The timestamp of the reference signal recorded at station j, Δt j For the time deviation of station j, t ij The theoretical time difference of arrival is calculated between the site geometry and the reference source location;

[0048] S2 specifically refers to:

[0049] S2-1: The front end is tuned to the target center frequency and the bandwidth gain and trigger threshold are set. The analog-to-digital conversion continuously outputs the spectrum signal data and buffers it frame by frame. DC removal and frequency offset correction, windowing segmentation and overlapping splicing are performed. Multi-resolution processing generates short-time Fourier time-frequency diagram and wavelet time-frequency diagram. Transient suppression and out-of-band suppression are performed on strong pulses and broadband interference. An index is established for each frame according to the timestamp and station identifier and written into the acquisition buffer.

[0050] S2-2. Extract the radio frequency texture summary from the preprocessing results, including the transient texture index and phase noise spectrum of the frequency offset trajectory asymmetry. Combine the time-frequency map to generate a multi-channel feature tensor and bind it with the time reference baseline parameters. Divide the features and original segments into blocks according to a fixed data block size and calculate the hash check value. Establish a sending queue and a feedback queue shared by sending and receiving. Record the timestamp, frame number, station number, hash value and priority of each data block for subsequent edge recognition and cooperative positioning steady-state input and consistency verification.

[0051] S3 specifically refers to:

[0052] S3-1. Load the quantization lightweight model on the terminal, read the multi-channel feature tensor and corresponding timestamp output by S2, perform forward inference to obtain the target category, interference type and confidence level, generate a unique identifier for each result by combining the station number and baseline parameters, extract key frequency bands and time-frequency positions and generate annotation masks, write them to the recognition result buffer and event trigger queue according to the result level, and at the same time establish a mapping index for the hash value, frame number and result number of the corresponding data block for subsequent localization fusion and consistency verification.

[0053] S3-2. For high-confidence samples, a small-step incremental update process is initiated, freezing the backbone layer and adjusting only the shallow layer and normalization parameters. A sliding window is used to control the update frequency and parameter drift amplitude. The model version number, checksum, and effect summary before and after the update are written to the local version record. For low-confidence and conflicting samples, a retry and feedback queue is initiated, triggering secondary inference and manual review strategies. If necessary, the model is rolled back to the previous stable version. The model is periodically calibrated and aligned with the central analysis model to correct inter-class boundary and domain biases, ensuring the recognition stability of the terminal under long-term operation. The status logs and indicators generated during the learning process are reported to the management terminal for global scheduling and strategy adjustment.

[0054] S4 specifically refers to:

[0055] S4-1. The management end aggregates observations from multiple stations according to a unified time base, constructs a weighted cost function with the target position vector x as the independent variable, and minimizes the residuals of the time difference of arrival, angle of arrival, frequency difference of arrival, and received signal strength. The formula is J(x)=∑ k (w t r t 2 +w a r a 2 +w f r f 2 +w p r p 2 ), where r t To achieve the difference between time difference observations and geometric predictions, r a r is the difference between the observed angle of arrival and the geometric prediction. f The difference between the observed arrival frequency and the geometric prediction, r p The difference between the received signal strength and the path loss model, w t w a w f w p To adaptively update the weight coefficients based on the inverse of the online confidence level or variance, iteratively minimize J(x) to obtain the current position estimate and residual statistics, generate the position estimate, error radius and target number, and write them into the positioning result buffer;

[0056] S4-2: Perform state prediction at fixed time steps, then correct with new positioning observations to generate the trajectory points and confidence intervals at that time; perform residual threshold judgment for each batch of observations, reduce weights and trigger resampling or repositioning when anomalies occur, and mark continuous anomalies as unstable targets; maintain target lifecycle and reassociation, handle matching and anti-exchange of multiple targets, associate the trajectory with the identification results of S3 by timestamp and station number, and complete the category label and key frequency band annotation; perform consistency verification between multi-station results, remove abnormal observations of time reference or baseline, and trigger rapid recalibration and rollback when necessary; finally output trajectory polyline, error ellipse, quality score and signature log, and push to the visualization and alarm channels;

[0057] S5 specifically refers to:

[0058] S5-1: The management terminal summarizes the load, link occupancy and model temperature of each site, generates concurrency and priority strategies, and distributes the collection, identification, location and backhaul tasks in segments according to the queue. Dynamic rate limiting and segmented scheduling are used to avoid congestion. When a node goes offline, it triggers automatic reconnection and fast recalibration. Failed tasks are written back to the queue according to the number of retries and the cooldown time, and the reason code is recorded for subsequent parameter tuning.

[0059] S5-2 performs hash verification and retransmission control on uplink and downlink block data, performs cross-consistency judgment on cross-station positioning results and removes abnormal observations of time base or baseline, generates signature logs and status snapshots for successful results and stores them in the database, automatically generates handling orders for abnormal results and links alarm channels, and pushes trajectory polylines, error ellipses, category labels and quality scores to the visualization interface, supporting remote collaborative annotation, playback and report export.

[0060] While maintaining the themes of distributed spectrum monitoring, edge intelligence, and cooperative positioning, this invention supplements time synchronization and link asymmetry correction, constructs a measurement fusion of time difference of arrival, angle of arrival, frequency difference of arrival, and received signal strength, and provides edge self-evolution identification and consistency control, achieving the following effects:

[0061] Baseline calibrable and clock accurate: Hardware timestamps are used for timing and loopback link correction, significantly reducing time difference of arrival (TDOA) system errors; more stable positioning: Based on TDOA, angle of arrival (ADR), frequency difference of arrival (FDR), and received signal strength are fused synchronously, maintaining convergence under non-line-of-sight conditions; more accurate identification: Multi-resolution time-frequency features and radio frequency textures are introduced, supporting online fine-tuning and calibration alignment, reducing long-term drift; stronger collaboration: Consistency control and edge computing verification are provided, automatic retransmission of anomalies is enabled, and results are traceable; manageable operation: Task orchestration, bandwidth management, and log signing are available to support grid-scale operation and maintenance.

[0062] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another entity or operation, and do not necessarily imply any such actual relationship or order between these entities and operations. Furthermore, the terms "comprising," "including," and any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, and includes elements inherent to such a process, method, article, or apparatus.

[0063] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments and make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent spectrum monitoring terminal, characterized in that: The system includes a spectrum acquisition and access module, an intelligent identification and processing module, and an edge collaborative management and control module. The spectrum acquisition and access module is electrically connected to multiple spectrum terminals and is used to complete terminal access, time reference establishment, and link asymmetry correction. The intelligent identification and processing module is electrically connected to the spectrum acquisition and access module and is used to construct multi-resolution time-frequency features, identify target and interference types, and perform online learning and adaptive processing on the acquired data. The edge collaborative management and control module is electrically connected to the intelligent identification and processing module and is used to perform measurement, fusion, and positioning of time difference of arrival, angle of arrival, frequency difference of arrival, and received signal strength, as well as to perform task orchestration, block verification and retransmission, and result consistency control.

2. The intelligent spectrum monitoring terminal according to claim 1, characterized in that: The spectrum acquisition and access module includes a timing module, a loopback correction module, a common visual calibration module, and a clock monitoring module. The timing module is used to establish a unified time reference with hardware timestamps in an Ethernet environment. The loopback correction module is used to measure the uplink and downlink delay difference and perform online correction of asymmetric offset. The common visual calibration module is used to calculate the position and time deviation of multiple stations based on a common visible signal source and generate a baseline matrix. The clock monitoring module is used to perform threshold detection on the timing residual and trigger recalibration and alarm when abnormal. The intelligent recognition processing module includes a preprocessing module, a texture extraction module, a lightweight inference module, and an online learning module. The preprocessing module is used to perform bandpass, denoising, and short-time Fourier wavelet transform on the radio frequency signal data to generate a multi-resolution time-frequency map. The texture extraction module is used to extract radio frequency texture features such as frequency offset, radio frequency signal asymmetry, transient texture, and phase noise. The lightweight inference module is used to run a quantized lightweight model on the terminal side to output the target category and confidence level. The online learning module is used to perform small-step incremental fine-tuning on high-confidence samples and calibrate and align them with the central analysis model. The edge collaborative management and control module includes a fusion calculation module, a trajectory tracking module, a task orchestration module, an edge computing verification module, and a consistency tracking module. The fusion calculation module is used to construct a cost function by weighting the arrival time difference, angle of arrival, frequency difference of arrival, and received signal strength observations according to confidence level and to calculate the target position. The trajectory tracking module is used to filter and predict the target state on the time series and output an error ellipse. The task orchestration module is used to generate a scheduling strategy based on terminal load, link occupancy, and model temperature and limit the concurrency. The edge computing verification module is used to perform edge computing verification on the sent and returned data and automatically retransmit in case of anomalies. The consistency tracking module is used to determine the consistency of multi-station positioning results and generate a signed log to achieve end-to-end traceability.

3. The intelligent spectrum monitoring terminal according to claim 2, characterized in that: The working steps of the monitoring terminal include: S1. The spectrum terminal powers on, completes self-test, and connects to the monitoring network via Ethernet. It performs hardware time synchronization to establish a unified time reference, performs link back-and-forth loop measurement to correct uplink and downlink asymmetry, completes common visual signal source calibration to generate baseline matrix and site calibration parameters, and the management terminal issues sampling plan, center frequency point bandwidth gain trigger threshold and buffering strategy. S2. The front-end tuning completes the downconversion and gain settings, the analog-to-digital conversion obtains the spectrum signal data and buffers it frame by frame, performs DC frequency offset correction and windowing segmentation, executes short-time Fourier transform and wavelet multi-resolution transform to generate time-frequency waterfall plot, extracts the frequency offset spectrum signal asymmetric transient texture phase noise RF texture summary, and hashes the data blocks and summaries by block and writes them into the transmission queue. S3. Load a lightweight quantization model on the terminal side to identify target categories and interference types, output category confidence and key frequency band labels, associate and bind the identification results with the site timestamp baseline parameters and radio frequency texture, high confidence samples enter the online fine-tuning process to complete small step incremental updates, low confidence samples enter the retry and manual review queue, and abnormal frames trigger local buffering and rate limiting reporting. S4: The management terminal gathers observations from multiple stations and calculates the time difference of arrival based on a unified time reference. It constructs a weighted cost function by combining the angle of arrival, frequency difference of arrival, and received signal strength to solve the position. It uses time series filtering to estimate the trajectory and outputs the error ellipse and target number. It fuses and associates the positioning results with the identification results of S3 to generate spatiotemporal linked target entries, supporting the dynamic addition and removal of mobile stations and temporarily added stations. S5. Task orchestration uses terminal load, link occupancy model, and temperature for concurrent control and bandwidth management. Edge computing verification drives retransmission mechanisms to ensure data consistency. Disconnection and calibration anomalies trigger automatic reconnection and recalibration. Signature logs and status snapshots are generated and archived in the audit archive. Location alarm identification tags and spectrum waterfall are visualized in real time on the PC and management platform, supporting remote collaborative annotation, playback, and report export.

4. The intelligent spectrum monitoring terminal according to claim 3, characterized in that: Specifically, S1 is: S1-1. After the terminal powers on and completes its self-test, it connects to the Ethernet. The hardware timestamp unit marks the transmitted and received packets at the nanosecond level in the network card and programmable logic layer. The management end periodically sends a time synchronization sequence to each terminal. The terminal compares its local clock with the received timestamp to obtain the instantaneous deviation. The system simultaneously performs back-and-forth loop measurement to separate the uplink and downlink delay difference. The uplink delay is denoted as Δu and the downlink delay is denoted as Δd. The link offset δ and the local clock drift rate γ are estimated from this. The time correction is updated in real time on the terminal using a sliding window recursive method according to t′=t-γt-δ. The management end synchronously maintains the time health index of each station. When the time health is lower than the threshold, local re-synchronization and loop retest are triggered. After completion, the new time base is written to the terminal time base register area and an acknowledgment frame is returned. S1-2. The management terminal selects reference signal sources that can be simultaneously received by each station and issues calibration instructions. All stations synchronously collect reference signals at a unified trigger time and report local timestamps and station location parameters. Based on this, the management terminal estimates the fine-grained time deviation and geometric fine-tuning amount of each station, generates a baseline parameter set for collaborative positioning, and writes it back to the calibration area of ​​the terminal and management terminal as input for subsequent calculations. After entering online inspection, reference signals are sampled according to short cycles and baseline residuals are calculated. When the residual is below the threshold, only the health index is updated. When it approaches the threshold, rapid retesting and fine-tuning are triggered and a transition baseline is generated. When the threshold is exceeded or a topology change is detected, full recalibration is performed and the old and new baseline parameters are saved in a versioned manner to support rollback, ensuring that subsequent positioning continues to use a unified time reference and reliable baseline. The formula is: Station i time deviation Δt i =t si -t c -δ, where t si For the reference signal timestamp recorded at station i, t c To standardize the triggering time, the baseline residual r ij =|(t) si -Δt i )-(t sj -Δt j )-t ij |, where t sj The timestamp of the reference signal recorded at station j, Δt j For the time deviation of station j, t ij The theoretical time difference of arrival is calculated between the site geometry and the reference source location.

5. The intelligent spectrum monitoring terminal according to claim 4, characterized in that: Specifically, S2 is: S2-1: The front end is tuned to the target center frequency and the bandwidth gain and trigger threshold are set. The analog-to-digital conversion continuously outputs the spectrum signal data and buffers it frame by frame. DC removal and frequency offset correction, windowing segmentation and overlapping splicing are performed. Multi-resolution processing generates short-time Fourier time-frequency diagram and wavelet time-frequency diagram. Transient suppression and out-of-band suppression are performed on strong pulses and broadband interference. An index is established for each frame according to the timestamp and station identifier and written into the acquisition buffer. S2-2. Extract the radio frequency texture summary from the preprocessing results, including the transient texture index and phase noise spectrum of the frequency offset trajectory asymmetry. Combine the time-frequency map to generate a multi-channel feature tensor and bind it with the time reference baseline parameters. Divide the features and original segments into blocks according to a fixed data block size and calculate the hash check value. Establish a sending queue and a feedback queue shared by sending and receiving. Record the timestamp, frame number, station number, hash value and priority of each data block for subsequent edge recognition and cooperative positioning steady-state input and consistency verification.

6. The intelligent spectrum monitoring terminal according to claim 5, characterized in that: Specifically, S3 is: S3-1. Load the quantization lightweight model on the terminal, read the multi-channel feature tensor and corresponding timestamp output by S2, perform forward inference to obtain the target category, interference type and confidence level, generate a unique identifier for each result by combining the station number and baseline parameters, extract key frequency bands and time-frequency positions and generate annotation masks, write them to the recognition result buffer and event trigger queue according to the result level, and at the same time establish a mapping index for the hash value, frame number and result number of the corresponding data block for subsequent localization fusion and consistency verification. S3-2. For high-confidence samples, a small-step incremental update process is initiated, freezing the backbone layer and adjusting only the shallow layer and normalization parameters. A sliding window is used to control the update frequency and parameter drift amplitude. The model version number, checksum, and effect summary before and after the update are written to the local version record. For low-confidence and conflicting samples, a retry and feedback queue is initiated, triggering secondary inference and manual review strategies. If necessary, the model is rolled back to the previous stable version. The system periodically calibrates and aligns with the central analysis model to correct inter-class boundaries and domain deviations, ensuring the recognition stability of the terminal during long-term operation. The status logs and indicators generated during the learning process are reported to the management terminal for global scheduling and strategy adjustment.

7. The intelligent spectrum monitoring terminal according to claim 6, characterized in that: Specifically, S4 is: S4-1. The management end aggregates observations from multiple stations according to a unified time base, constructs a weighted cost function with the target position vector x as the independent variable, and minimizes the residuals of the time difference of arrival, angle of arrival, frequency difference of arrival, and received signal strength. The formula is J(x)=∑ k (w t r t 2 +w a r a 2 +w f r f 2 +w p r p 2 ), where r t To achieve the difference between time difference observations and geometric predictions, r a r is the difference between the observed angle of arrival and the geometric prediction. f The difference between the observed arrival frequency and the geometric prediction, r p The difference between the received signal strength and the path loss model, w t w a w f w p To adaptively update the weight coefficients based on the inverse of the online confidence level or variance, iteratively minimize J(x) to obtain the current position estimate and residual statistics, generate the position estimate, error radius and target number, and write them into the positioning result buffer; S4-2. Perform state prediction at fixed time steps and then correct with new positioning observations to generate the trajectory points and confidence intervals at that time; perform residual threshold judgment for each batch of observations, reduce the weight and trigger resampling or repositioning when anomalies occur, and mark continuous anomalies as unstable targets; maintain the target life cycle and re-association, handle the matching and anti-exchange of multiple target intersections, associate the trajectory with the identification results of S3 by timestamp and station number, and complete the category label and key frequency band annotation; Perform consistency checks across multiple stations, remove abnormal observations from the time base or baseline, and trigger rapid recalibration and rollback if necessary; finally output trajectory polylines, error ellipses, quality scores and signature logs, and push them to the visualization and alarm channels.

8. The intelligent spectrum monitoring terminal according to claim 7, characterized in that: Specifically, S5 is: S5-1: The management terminal summarizes the load, link occupancy and model temperature of each site, generates concurrency and priority strategies, and distributes the collection, identification, location and backhaul tasks in segments according to the queue. Dynamic rate limiting and segmented scheduling are used to avoid congestion. When a node goes offline, it triggers automatic reconnection and fast recalibration. Failed tasks are written back to the queue according to the number of retries and the cooldown time, and the reason code is recorded for subsequent parameter tuning. S5-2 performs hash verification and retransmission control on uplink and downlink block data, performs cross-consistency judgment on cross-station positioning results and removes abnormal observations of time base or baseline, generates signature logs and status snapshots for successful results and stores them in the database, automatically generates handling orders for abnormal results and links alarm channels, and pushes trajectory polylines, error ellipses, category labels and quality scores to the visualization interface, supporting remote collaborative annotation, playback and report export.

Images