An AI agent takes over the integration system of spectrum monitoring
Patent Information
- Application Number
- CN202610786496.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-02
- Publication Date
- 2026-10-30
AI Technical Summary
[0004]基于此，有必要针对现有技术在频谱监测软件中缺乏统一的智能调度与协同分析机制，无法实现多波段同步监测、联动分析及无缝衔接的问题，提供一种AI智能体接管频谱监测一体化系统，能在频谱监测软件中进行统一的智能调度与协同分析，实现多波段同步监测、联动分析及衔接
本发明提出一种AI智能体接管频谱监测一体化系统，所述系统包括外部采集设备、频谱监测平台、AI智能体和用户前端；
Smart Images
Figure CN123019912A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of software automated scheduling technology, and in particular to an integrated system for AI agent takeover of spectrum monitoring. Background Technology
[0002] With the rapid development of information technologies such as 5G, 6G, and the Internet of Things, the electromagnetic spectrum environment is becoming increasingly complex, and the contradiction between the scarcity of spectrum resources and their underutilization is becoming more and more prominent. Spectrum monitoring has become a key means to ensure the rational use of spectrum resources.
[0003] Existing spectrum monitoring platforms and software have many limitations: they rely on manual operation, resulting in low human efficiency and only enabling single-threaded interaction, making it impossible to handle multi-band and multi-task monitoring needs simultaneously; they have low levels of intelligence, requiring professional technicians to operate them, leading to long training cycles and high labor costs; if AI is used directly to process raw monitoring data, the real-time performance of AI signal processing is difficult to meet the requirements due to the massive amount of data, and AI development costs are also high; although some software has functions such as multi-band scanning, fluorescence spectrum analysis, and anomaly detection, each functional module operates independently, lacking a unified intelligent scheduling and collaborative analysis mechanism, making it impossible to achieve multi-band synchronous monitoring, linkage analysis, and seamless integration. Summary of the Invention
[0004] Therefore, it is necessary to address the problem that existing technologies lack a unified intelligent scheduling and collaborative analysis mechanism in spectrum monitoring software, which makes it impossible to achieve multi-band synchronous monitoring, linkage analysis, and seamless connection. To provide an integrated system for taking over spectrum monitoring with an AI intelligent agent, a unified intelligent scheduling and collaborative analysis system can be performed in spectrum monitoring software to achieve multi-band synchronous monitoring, linkage analysis, and connection.
[0005] To achieve the above-mentioned technical effects, the technical solution of the present invention is as follows: An integrated system for managing spectrum monitoring via an AI agent, the system comprising external acquisition equipment, a spectrum monitoring platform, an AI agent, and a user front end; The output of the external acquisition device is connected to the first input of the spectrum monitoring platform. The first output of the spectrum monitoring platform is connected to the input of the user front end. The second output of the spectrum monitoring platform is connected to the input of the AI agent. The output of the AI agent is connected to the second input of the spectrum monitoring platform. The output of the user front end is connected to the input of the AI agent.
[0006] This invention also provides an integrated method for AI agent takeover spectrum monitoring, which is applied to the aforementioned integrated system for AI agent takeover spectrum monitoring and includes the following steps: S1. The external acquisition device acquires raw data and transmits the raw data to the spectrum monitoring platform; S2. The spectrum monitoring platform receives the raw data, processes and monitors the raw data to obtain service data, and outputs the service data to the user front-end and the AI agent respectively; S3. The user front-end is used to receive the business data and input task content to the AI agent. S4. The AI agent receives the service data, performs state machine scheduling based on the service data and task content, generates scheduling output results, and sends them back to the spectrum monitoring platform.
[0007] Preferably, the collection of raw data includes: Configure acquisition parameters for the external acquisition device, and acquire radio frequency signals and in-phase / quadrature data based on the acquisition parameters; The radio frequency signal and in-phase / quadrature data are preprocessed to obtain preprocessed data, which is then used as the original data.
[0008] Preferably, the preprocessing of the radio frequency signal and in-phase / quadrature data includes: S111. Perform data slicing processing on the radio frequency signal and in-phase / quadrature data to split it into segments of fixed frame length; S112. Apply window function weighting to the segmented data after slicing to obtain the weighted data; S113. Perform FFT calculation on the weighted data and output frequency domain data; S114. Based on the frequency domain data obtained by FFT calculation, generate spectral features and candidate signals as the original data.
[0009] Preferably, the task content includes a natural language task or a structured task template.
[0010] Preferably, the service data includes: spectrum data, direction finding results, signal identification results, and analysis results; the scheduling output results include: tool call instructions, acquisition parameter adjustment instructions, fusion judgment results, manual confirmation requests, and rollback instructions.
[0011] Preferably, the spectrum monitoring platform includes a spectrum monitoring module, an interference direction finding module, an interference identification module, and a signal analysis module; the spectrum monitoring platform receives the raw data, processes and monitors the raw data, including: The spectrum monitoring module performs power spectrum estimation and signal threshold decision on the raw data to generate spectrum data of the raw data. The interference direction finding module performs signal direction finding on the raw data and outputs the direction finding result of the raw data. The interference identification module detects and identifies the original data to obtain the signal identification result of the original data; The signal analysis module demodulates and analyzes the original data based on the signal recognition result of the original data, and outputs the analysis result of the original data.
[0012] Preferably, the AI agent includes a state machine scheduling unit, which performs state machine scheduling based on the business data and the task content to generate the scheduling output result.
[0013] Preferably, the states in which the state machine scheduling unit performs state machine scheduling include: input monitoring state, task takeover state, tool mapping state, candidate action generation state, parameter verification state, parameter transaction issuance state, scheduling execution state, result fusion state, decision output state, and re-collection / confirmation / rollback state.
[0014] Preferably, the process of state machine scheduling based on the business data and task content includes: When the state machine scheduling unit is in the input listening state, it continuously receives the task content and the service data. If a new natural language task or structured task template, abnormal spectrum data, or low-confidence direction finding results, signal recognition results, or parsing results are detected, it switches to the task takeover state. When the state machine scheduling unit is in the task takeover state, it generates a structured task target based on the new natural language task or structured task template, and switches to the tool mapping state. When the state machine scheduling unit is in the tool mapping state, it splits the structured task target into atomic tasks and maps them into tool call sequences, and then switches to the candidate action generation state. When the state machine scheduling unit is in the candidate action generation state, it generates candidate actions by combining spectrum data and direction finding results, and then switches to the parameter verification state. When the state machine scheduling unit is in the parameter verification state, it verifies the candidate action. If the verification fails, it returns to the candidate action generation state to regenerate the candidate action. If the verification passes, it switches to the parameter transaction issuance state. When the state machine scheduling unit is in the parameter transaction issuance state, it records the parameters before adjustment, the parameters after adjustment, the verification indicators, and the rollback parameters to obtain the parameter transaction record; it converts the candidate action into a parameter adjustment instruction or a tool call instruction, and then issues the collected parameter adjustment instruction or tool call instruction to the spectrum monitoring platform and switches to the scheduling execution state; When the state machine scheduling unit is in the scheduling execution state, it calls the corresponding software tool to execute the monitoring business according to the tool call sequence, and then switches to the result fusion state; When the state machine scheduling unit is in the result fusion state, it merges the business data into a unified event object and switches to the decision output state; When the state machine scheduling unit is in the decision output state, it evaluates whether the unified event object meets the task requirements. If the task requirements are met, it outputs the fusion judgment result to the spectrum monitoring platform and returns to the input monitoring state. If the decision conditions are not met, it switches to the re-sampling / confirmation / rollback state. When the state machine scheduling unit is in the re-acquisition / confirmation / rollback state, if abnormal spectrum data or low-confidence direction finding results, signal recognition results or analysis results are detected in the unified event, the service data is re-acquisitioned and the candidate action generation state is returned. If an abnormal signal recognition result is detected in the unified event, a manual confirmation request is generated and forwarded to the user front end by the spectrum monitoring platform; after receiving the manual confirmation result fed back by the user front end and forwarded by the spectrum monitoring platform, the AI agent returns the result fusion status. If the tool fails to execute, the parameter adjustment effect deteriorates, or the resource limit is exceeded, the spectrum monitoring platform is restored to the previous stable parameters according to the parameter transaction record, and the input monitoring state is returned.
[0015] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: This invention proposes an integrated system for AI agent-controlled spectrum monitoring, the system comprising external acquisition equipment, a spectrum monitoring platform, an AI agent, and a user front end; The output of the external acquisition device is connected to the first input of the spectrum monitoring platform. The first output of the spectrum monitoring platform is connected to the input of the user front-end. The second output of the spectrum monitoring platform is connected to the input of the AI agent. The output of the AI agent is connected to the second input of the spectrum monitoring platform. The output of the user front-end is connected to the input of the AI agent. Using this system, the present invention enables unified intelligent scheduling and collaborative analysis in spectrum monitoring software, achieving multi-band synchronous monitoring, linked analysis, and seamless integration. Attached Figure Description
[0016] Figure 1 This is a system block diagram of the integrated intelligent agent takeover spectrum monitoring system in an embodiment of the present invention; Figure 2 This is a schematic diagram of the functional module structure of the spectrum monitoring software in an embodiment of the present invention; Figure 3 This is a schematic diagram of the collaborative decision-making process of AI agents in an embodiment of the present invention; Figure 4 This is a schematic diagram of the AI agent scheduling state machine workflow in this application example. Detailed Implementation
[0017] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. It is understandable to those skilled in the art that some well-known details may be omitted from the accompanying drawings; To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.
[0018] Example 1 like Figure 1 As shown in the figure, this embodiment provides an integrated system for intelligent agent-controlled spectrum monitoring. The system includes an external acquisition device, a spectrum monitoring platform, an AI agent, and a user front end. The spectrum monitoring platform includes a spectrum monitoring module, an interference direction finding module, an interference identification module, and a signal analysis module. The output of the external acquisition device is connected to the input of the spectrum monitoring module, the interference direction finding module, and the interference identification module, respectively. The output of the interference identification module is connected to the input of the signal analysis module. The AI agent is bidirectionally connected to the spectrum monitoring platform; The input terminal of the user front-end is connected to the output terminal of the spectrum monitoring platform, and the output terminal of the user front-end is connected to the input terminal of the AI agent.
[0019] This embodiment also provides an integrated method for AI agent takeover of spectrum monitoring, which is implemented using the aforementioned integrated AI agent takeover spectrum monitoring system, and includes the following steps: S1. The external acquisition device acquires raw data and transmits the raw data to the spectrum monitoring platform; S2. The spectrum monitoring platform receives the raw data, processes and monitors the raw data to obtain service data, and outputs the service data to the user front-end and the AI agent respectively; S3. The user front-end is used to receive the business data and input task content to the AI agent. S4. The AI agent receives the service data, performs state machine scheduling based on the service data and task content, generates scheduling output results, and sends them back to the spectrum monitoring platform.
[0020] Specifically, external acquisition devices are used to provide I / Q data, spectrum characteristics, device status, and adjustable parameter interfaces; the monitoring software platform is used to complete acquisition access, spectrum monitoring, intelligent identification, demodulation and parsing, storage, and logging; the user front end is used for task configuration, result viewing, and manual confirmation; and the AI agent is used for task takeover, parameter adjustment, scheduling fusion, and policy write-back.
[0021] like Figure 2 As shown, in this embodiment, the collection of raw data includes: Configure acquisition parameters for the external acquisition device, and acquire radio frequency signals and in-phase / quadrature data based on the acquisition parameters; The radio frequency signal and in-phase / quadrature data are preprocessed to obtain preprocessed data, which is used as the original data. The preprocessing of the radio frequency signal and in-phase / quadrature data includes: S111. Perform data slicing processing on the radio frequency signal and in-phase / quadrature data to split it into segments of fixed frame length; S112. Apply window function weighting to the segmented data after slicing to obtain the weighted data; S113. Perform FFT calculation on the weighted data and output frequency domain data; S114. Based on the frequency domain data obtained by FFT calculation, generate spectral features and candidate signals as the original data.
[0022] Specifically, fixed frequency measurement, spectrum scanning, and I / Q segment acquisition parameters are configured for the external acquisition device. The external acquisition device is used to receive radio frequency signals and offline I / Q data to complete channel access, sampling, and data buffering. Then, the radio frequency signals and offline I / Q data are processed by performing data slicing, window function weighting, and FFT calculation to generate spectral features and candidate signals.
[0023] External acquisition devices serve only as data sources and execution objects. The improvement of this invention focuses on the takeover, orchestration, verification, fusion, and rollback mechanisms of AI agents within the monitoring software platform.
[0024] The AI agent includes a state machine scheduling unit, which performs state machine scheduling based on the business data and the task content to generate the scheduling output result.
[0025] Specifically, the states in which the state machine scheduling unit performs state machine scheduling include: input monitoring state, task takeover state, tool mapping state, candidate action generation state, parameter verification state, parameter transaction issuance state, scheduling execution state, result fusion state, decision output state, and re-collection / confirmation / rollback state.
[0026] The monitoring software platform reads multi-channel I / Q data streams in real time through the PCIE 4.0 interface. After signal acquisition and preprocessing, the data is sent to the core software function, which then uniformly schedules four business function modules: spectrum monitoring, interference direction finding, interference identification, and signal analysis.
[0027] The spectrum monitoring module performs power spectrum estimation and signal threshold decision on the raw data to generate spectrum data of the raw data. Specifically, the raw data includes full-band data, and the spectrum data includes: two-dimensional spectrograms, waterfall plots, and digital fluorescence spectra; the spectrum monitoring module performs power spectrum estimation, local tracking, and signal threshold decision on the full-band data, and generates two-dimensional spectrograms, waterfall plots, and digital fluorescence spectra in real time, supports full probability interception of burst pulse signals and switching of multiple analysis bandwidths, and provides multiple measurement modes such as multi-channel parallel demodulation, discrete scanning, frequency band scanning, and digital scanning, supporting operators to intuitively observe and manually monitor the electromagnetic situation across the entire frequency band.
[0028] The interference direction finding module performs signal direction finding on the raw data and outputs the direction finding result of the raw data. Specifically, the raw data includes multi-channel coherent I / Q data, and the direction finding results of the raw data include the direction of arrival and confidence level. The interference direction finding module uses phase interferometry to complete intermediate frequency direction finding and scanning direction finding based on the multi-channel coherent I / Q data. It can support the separation and azimuth estimation of multiple co-frequency signals and output the direction of arrival and confidence level for subsequent positioning and correlation analysis.
[0029] The interference identification module detects and identifies the original data to obtain the signal identification result of the original data; Specifically, the signal identification results of the raw data include signal detection results, technical system identification results, and modulation pattern identification results; the interference identification module is equipped with a lightweight CNN classifier, which performs signal detection, technical system identification, and modulation pattern identification on the signal segments output by monitoring and direction finding, and obtains signal detection results, technical system identification results, and modulation pattern identification results; it supports automatic discrimination of technical systems such as GSM, 5G NR, Bluetooth, WiFi, LTE, PDT, and modulation patterns such as AM, FM, CW, ASK, PSK, DPSK, QAM, FSK, and MSK, distinguishes between legitimate signals and interference signals, and supports automatic training and dynamic upgrading of localized models.
[0030] The signal analysis module demodulates and analyzes the original data based on the signal recognition result of the original data, and outputs the analysis result of the original data.
[0031] Specifically, the signal parsing module selects the corresponding demodulation algorithm based on the signal detection results, technical system identification results, and modulation pattern identification results, completes carrier synchronization, symbol timing recovery, and channel equalization, and outputs digital audio, protocol content, and data payload. At the same time, it performs high-precision measurement of parameters such as signal carrier frequency, bandwidth, and symbol rate, and supports analog demodulation such as AM, FM, CW, DSB, and SSB, as well as digital protocol parsing such as LoRaWAN, NB-IoT, ADS-B, and AIS.
[0032] Example 2 This embodiment further illustrates the AI-agent in the above embodiments.
[0033] In this embodiment, the AI agent is not a standalone signal classification model, but rather a takeover and orchestration unit positioned between the user front-end, the task management module, and the monitoring software platform. The spectrum monitoring software is responsible for real-time signal processing such as acquisition, spectrum analysis, direction finding, identification, and parsing, while the AI agent is responsible for task understanding, context assembly, tool orchestration, parameter adjustment, result fusion, and strategy write-back based on user tasks and software operating status.
[0034] Through the state machine scheduling unit, the AI agent organizes the software function modules that originally relied on manual operation item by item into a closed-loop monitoring process of "task understanding, state perception, tool execution, result fusion, review and rollback, and feedback optimization". In high-throughput I / Q data and multi-task spectrum monitoring scenarios, it avoids the real-time pressure caused by the AI directly processing all raw data, and highlights the technical features that distinguish this invention from ordinary AI classification models or simple software module combinations.
[0035] like Figure 3 and Figure 4As shown, the AI agent closed-loop scheduling state machine adopts a progressive hierarchical design to unify the AI agent takeover process, parameter adjustment mechanism, closed-loop scheduling, and result fusion mechanism into an executable software scheduling process. This state machine includes input monitoring state, task takeover state, tool mapping state, candidate action generation state, parameter verification state, parameter transaction issuance state, scheduling execution state, result fusion state, decision output state, and re-acquisition / confirmation / rollback state. The AI agent scheduling module is as follows: When the state machine scheduling unit is in the input listening state, it continuously receives the task content and the service data. If a new natural language task or structured task template, abnormal spectrum data, or low-confidence direction finding results, signal recognition results, or parsing results are detected, it switches to the task takeover state. Specifically, as the starting point of the state machine, it continuously receives user tasks and software status. The task content includes natural language tasks, structured task templates, target frequency bands, and priorities from user input. The business data includes monitoring results and operational status information. The monitoring results include spectrum data, direction finding results, signal recognition results, and analysis results. The spectrum data, direction finding results, signal recognition results, and analysis results include: spectrum peak value, I / Q segment index, and recognition confidence. The operational status information includes: historical comparison results, device status, resource load, and abnormal alarms. The AI agent's output includes task plans, tool call instructions, acquisition parameter adjustment instructions, fusion judgment results, alarm suggestions, manual confirmation requests, and rollback instructions. All of the above inputs and outputs are bound to task ID, timestamp, and tracking ID. If a new natural language task or structured task template, abnormal spectrum data, or low-confidence direction finding results, signal recognition results, or analysis results are detected, the system switches to the task takeover state. When the state machine scheduling unit is in the task takeover state, it generates a structured task target based on the new natural language task or structured task template, and switches to the tool mapping state. Specifically, the state machine scheduling unit performs intent parsing and entity extraction on user tasks, combines knowledge and rule bases to form structured task objectives, and establishes task numbers, target frequency bands, constraints, and priorities. This state is used to convert manually proposed monitoring intents into task descriptions executable by the monitoring software platform. The knowledge and rule base includes frequency planning, station records, radio system characteristics, modulation style parameters, historical task records, anomaly judgment rules, parameter boundaries, permission levels, and resource thresholds, used to constrain the task splitting and parameter adjustment actions of the AI agent. The tool registry registers the input parameters, output results, resource usage, and failure handling strategies of tools such as spectrum scanning, I / Q segment acquisition, direction finding task issuance, signal detection and classification, demodulation parsing, historical comparison, alarm tables, re-acquisition, and acquisition parameter configuration.
[0036] When the state machine scheduling unit is in the tool mapping state, it splits the structured task target into atomic tasks and maps them into tool call sequences, and then switches to the candidate action generation state. Specifically, the state machine scheduling unit breaks down structured task objectives into atomic tasks and maps these atomic tasks to specific tools in the tool registry. The structured task objectives are mapped to tool call sequences such as spectrum scanning, I / Q acquisition, intelligent identification, direction finding, demodulation analysis, historical comparison, and re-acquisition. Each tool registers its input parameters, output results, parameter boundaries, resource usage, and failure handling strategies. The AI agent calls these tools through the task management module, directly controlling the underlying hardware without bypassing the software platform. For example, when the software status indicates an abnormal peak at a certain frequency and the identification confidence is low, the AI agent can generate a toolchain that includes "narrowing the scanning range, extending the dwell time, acquiring I / Q segments, calling the identification tool, performing historical comparison, and triggering re-acquisition or manual confirmation if necessary." The AI agent manages the task queue, acquisition strategy, tool call sequence, result fusion strategy, and alarm handling process within the monitoring software, rather than directly controlling the underlying hardware without boundaries.
[0037] When the state machine scheduling unit is in the candidate action generation state, it generates candidate actions by combining spectrum data and direction finding results, and then switches to the parameter verification state. Specifically, the state machine scheduling unit generates acquisition parameters or scheduling parameters to adjust actions based on task objectives, real-time spectrum status, identification confidence, historical records, and resource load. Candidate actions include adjusting the center frequency, scanning range, frequency step, analysis bandwidth, dwell time, sampling rate, detection threshold, I / Q recording duration, tool call order, or identification model version.
[0038] When the state machine scheduling unit is in the parameter verification state, it verifies the candidate action. If the verification fails, it returns to the candidate action generation state to regenerate the candidate action. If the verification passes, it switches to the parameter transaction issuance state.
[0039] Specifically, the state machine scheduling unit performs boundary checks, permission checks, conflict checks, and resource impact checks on candidate actions. If a candidate action has out-of-bounds parameters, insufficient permissions, task conflicts, or resource overruns, it proceeds to the verification failure branch and regenerates the candidate action; if the verification passes, it enters the parameter transaction issuance state.
[0040] When the state machine scheduling unit is in the parameter transaction issuance state, it records the parameters before adjustment, the parameters after adjustment, the verification indicators, and the rollback parameters to obtain the parameter transaction record; it converts the candidate action into a parameter adjustment instruction or a tool call instruction, and then issues the collected parameter adjustment instruction or tool call instruction to the spectrum monitoring platform and switches to the scheduling execution state; Specifically, through parameter transaction records, the system can track the reason for each AI parameter tuning, the execution result, and the previous stable parameter that can be recovered.
[0041] When the state machine scheduling unit is in the scheduling execution state, it calls the corresponding software tool to execute the monitoring business according to the tool call sequence, and then switches to the result fusion state; Specifically, the state machine scheduling unit invokes the corresponding software tools according to the tool invocation sequence to perform spectrum scanning, I / Q acquisition, intelligent identification, direction finding, demodulation analysis, or historical comparison. After execution, each software tool publishes a standardized event to the event bus, which includes the task number, source tool, frequency, time, confidence level, quality factor, and trace number.
[0042] When the state machine scheduling unit is in the result fusion state, it merges the business data into a unified event object and switches to the decision output state; Specifically, the state machine scheduling unit unifies the spectrum peak, identification result, direction finding result, demodulation analysis result, historical comparison result and manual confirmation result into evidence objects, and merges them into unified event objects according to time overlap, frequency proximity, spatial consistency and confidence relationship.
[0043] When the state machine scheduling unit is in the decision output state, it evaluates whether the unified event object meets the task requirements. If the task requirements are met, it outputs the fusion judgment result to the spectrum monitoring platform and returns to the input monitoring state. If the decision conditions are not met, it switches to the re-sampling / confirmation / rollback state.
[0044] Specifically, the state machine scheduling unit makes judgments based on fusion confidence, evidence integrity, result conflict, and resource consumption. If the fusion result meets the task requirements, it outputs alarms, reports, evidence chains, and policy write-back results; if the fusion result does not meet the judgment conditions, it enters the re-collection / confirmation / rollback state.
[0045] When the state machine scheduling unit is in the re-acquisition / confirmation / rollback state, if abnormal spectrum data or low-confidence direction finding results, signal recognition results or analysis results are detected in the unified event, the service data is re-acquisitioned and the candidate action generation state is returned. If an abnormal signal recognition result is detected in the unified event, a manual confirmation request is generated and forwarded to the user front end by the spectrum monitoring platform; after receiving the manual confirmation result fed back by the user front end and forwarded by the spectrum monitoring platform, the AI agent returns the result fusion status. If the tool fails to execute, the parameter adjustment effect deteriorates, or the resource limit is exceeded, the spectrum monitoring platform is restored to the previous stable parameters according to the parameter transaction record, and the input monitoring state is returned.
[0046] Through the aforementioned state machine, the AI agent no longer merely outputs signal categories or recognition confidence levels, but is able to complete task takeover, tool mapping, parameter generation, legality verification, parameter transaction distribution, result fusion, decision output, and anomaly rollback within the monitoring software platform, thereby forming a traceable, verifiable, and recoverable closed-loop monitoring process.
[0047] The system provided by this invention aims to enable the AI agent to not perform signal classification alone, but to take over the task, call tools, adjust parameters, fuse results, and trigger re-sampling or manual confirmation within the spectrum monitoring software.
[0048] This invention addresses three key issues: First, how AI can invoke processing tasks through software tools instead of directly processing all the raw data in scenarios with I / Q data access speeds of several GB per second; second, how AI can generate acquisition parameters and adjust actions based on abnormal peak values, signal-to-noise ratio, recognition confidence, historical records, and resource load; and third, how the system can complete closed-loop scheduling and rollback when there is low confidence, conflicting results, or multi-task competition.
[0049] Example 3 This embodiment demonstrates a specific scenario of the system described in the above embodiments.
[0050] This invention is specifically implemented on an integrated spectrum monitoring and sensing software platform, focusing on an AI agent takeover and scheduling unit implemented within the platform. This unit monitors the software's operating status, receives user tasks, and invokes software tool interfaces and hardware / software configuration parameter interfaces. When abnormal spectrum events, low-confidence identification results, or historical comparison conflicts occur, the AI agent automatically organizes rescanning, re-sampling, identification, direction finding, analysis, or historical comparison.
[0051] This system was implemented for spectrum monitoring and support in the venue and surrounding areas of the 15th National Games. The monitoring equipment covers the 30 MHz to 18 GHz frequency band, with a real-time intermediate frequency bandwidth of up to 1 GHz, a scanning speed of up to 110 GHz / s, and a direction-finding response time of up to 0.94 ms. This invention focuses on the 2.4 GHz and 5.8 GHz frequency bands related to UAV remote control or image transmission, triggering resampling, identification, and direction finding for sudden, frequency-hopping, bandwidth-occupied, or non-whitelisted signals. It assists in distinguishing normal venue communication signals from suspected "black flight" UAV signals and generates event logs for on-site confirmation.
[0052] This invention reduces reliance on manual monitoring, analysis, and configuration during spectrum monitoring, enabling the system to automatically proceed to subsequent review and handling processes when abnormal signals are detected, insufficient confidence levels are identified, or conflicts arise between multi-module results. The AI agent can call software tool interfaces and hardware / software configuration parameter interfaces based on software operating status and monitoring results to controllably adjust parameters such as scanning range, dwell time, detection threshold, I / Q recording duration, and tool call order, thereby improving the continuity of the monitoring process and the efficiency of abnormal signal handling. Through parameter verification, parameter transactions, and rollback mechanisms, this invention reduces the risk of frame drops, resource congestion, or task conflicts caused by improper parameter adjustments. Through a unified event fusion and manual confirmation mechanism, monitoring tasks, parameter changes, fusion results, and rollback processes are continuously recorded, facilitating post-event review and manual confirmation.
[0053] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is impossible to exhaustively list all the implementation methods here. All obvious variations or modifications derived from the technical solutions of the present invention are still within the protection scope of the present invention.
Claims
1. An integrated system for AI-powered intelligent agent-controlled spectrum monitoring, characterized in that, The system includes external acquisition devices, a spectrum monitoring platform, an AI agent, and a user front-end; The output of the external acquisition device is connected to the first input of the spectrum monitoring platform. The first output of the spectrum monitoring platform is connected to the input of the user front end. The second output of the spectrum monitoring platform is connected to the input of the AI agent. The output of the AI agent is connected to the second input of the spectrum monitoring platform. The output of the user front end is connected to the input of the AI agent.
2. A method for integrating AI agent takeover of spectrum monitoring, wherein the method is implemented using the integrated AI agent takeover of spectrum monitoring system as described in claim 1, characterized in that, Includes the following steps: S1. The external acquisition device acquires raw data and transmits the raw data to the spectrum monitoring platform; S2. The spectrum monitoring platform receives the raw data, processes and monitors the raw data to obtain service data, and outputs the service data to the user front-end and the AI agent respectively; S3. The user front-end is used to receive the business data and input task content to the AI agent. S4. The AI agent receives the service data, performs state machine scheduling based on the service data and task content, generates scheduling output results, and sends them back to the spectrum monitoring platform.
3. The integrated method for AI agent takeover of spectrum monitoring according to claim 2, characterized in that, The collected raw data includes: Configure acquisition parameters for the external acquisition device, and acquire radio frequency signals and in-phase / quadrature data based on the acquisition parameters; The radio frequency signal and in-phase / quadrature data are preprocessed to obtain preprocessed data, which is then used as the original data.
4. The integrated method for AI agent takeover of spectrum monitoring according to claim 3, characterized in that, The preprocessing of the radio frequency signal and in-phase / quadrature data includes: S111. Perform data slicing processing on the radio frequency signal and in-phase / quadrature data to split it into segments of fixed frame length; S112. Apply window function weighting to the segmented data after slicing to obtain the weighted data; S113. Perform FFT calculation on the weighted data and output frequency domain data; S114. Based on the frequency domain data obtained by FFT calculation, generate spectral features and candidate signals as the original data.
5. The integrated method for AI agent takeover of spectrum monitoring according to claim 4, characterized in that, The task content includes natural language tasks or structured task templates.
6. The integrated method for AI agent takeover of spectrum monitoring according to claim 5, characterized in that, The business data includes: spectrum data, direction finding results, signal identification results, and analysis results; the scheduling output results include: tool call instructions, acquisition parameter adjustment instructions, fusion judgment results, manual confirmation requests, and rollback instructions.
7. The integrated method for AI agent takeover of spectrum monitoring according to claim 6, characterized in that, The spectrum monitoring platform includes a spectrum monitoring module, an interference direction finding module, an interference identification module, and a signal analysis module; the spectrum monitoring platform receives the raw data, processes and monitors the raw data, including: The spectrum monitoring module performs power spectrum estimation and signal threshold decision on the raw data to generate spectrum data of the raw data. The interference direction finding module performs signal direction finding on the raw data and outputs the direction finding result of the raw data. The interference identification module detects and identifies the original data to obtain the signal identification result of the original data; The signal analysis module demodulates and analyzes the original data based on the signal recognition result of the original data, and outputs the analysis result of the original data.
8. The integrated method for AI agent takeover of spectrum monitoring according to claim 6, characterized in that, The AI agent includes a state machine scheduling unit, which performs state machine scheduling based on the business data and the task content to generate the scheduling output result.
9. The integrated method for AI agent takeover of spectrum monitoring according to claim 8, characterized in that, The states in which the state machine scheduling unit performs state machine scheduling include: input monitoring state, task takeover state, tool mapping state, candidate action generation state, parameter verification state, parameter transaction issuance state, scheduling execution state, result fusion state, decision output state, and re-collection / confirmation / rollback state.
10. The integrated method for AI agent takeover of spectrum monitoring according to claim 9, characterized in that, The process of state machine scheduling based on the business data and task content includes: When the state machine scheduling unit is in the input listening state, it continuously receives the task content and the service data. If a new natural language task or structured task template, abnormal spectrum data, or low-confidence direction finding results, signal recognition results, or parsing results are detected, it switches to the task takeover state. When the state machine scheduling unit is in the task takeover state, it generates a structured task target based on the new natural language task or structured task template, and switches to the tool mapping state. When the state machine scheduling unit is in the tool mapping state, it splits the structured task target into atomic tasks and maps them into tool call sequences, and then switches to the candidate action generation state. When the state machine scheduling unit is in the candidate action generation state, it generates candidate actions by combining spectrum data and direction finding results, and then switches to the parameter verification state. When the state machine scheduling unit is in the parameter verification state, it verifies the candidate action. If the verification fails, it returns to the candidate action generation state to regenerate the candidate action. If the verification passes, it switches to the parameter transaction issuance state. When the state machine scheduling unit is in the parameter transaction issuance state, it records the parameters before adjustment, the parameters after adjustment, the verification indicators, and the rollback parameters to obtain the parameter transaction record; it converts the candidate action into a parameter adjustment instruction or a tool call instruction, and then issues the collected parameter adjustment instruction or tool call instruction to the spectrum monitoring platform and switches to the scheduling execution state; When the state machine scheduling unit is in the scheduling execution state, it calls the corresponding software tool to execute the monitoring business according to the tool call sequence, and then switches to the result fusion state; When the state machine scheduling unit is in the result fusion state, it merges the business data into a unified event object and switches to the decision output state; When the state machine scheduling unit is in the decision output state, it evaluates whether the unified event object meets the task requirements. If the task requirements are met, it outputs the fusion judgment result to the spectrum monitoring platform and returns to the input monitoring state. If the decision conditions are not met, it switches to the re-sampling / confirmation / rollback state. When the state machine scheduling unit is in the re-acquisition / confirmation / rollback state, if abnormal spectrum data or low-confidence direction finding results, signal recognition results or analysis results are detected in the unified event, the service data is re-acquisitioned and the candidate action generation state is returned. If an abnormal signal recognition result is detected in the unified event, a manual confirmation request is generated and forwarded to the user front end by the spectrum monitoring platform; after receiving the manual confirmation result fed back by the user front end and forwarded by the spectrum monitoring platform, the AI agent returns the result fusion status. If the tool fails to execute, the parameter adjustment effect deteriorates, or the resource limit is exceeded, the spectrum monitoring platform is restored to the previous stable parameters according to the parameter transaction record, and the input monitoring state is returned.