Frequency spectrum real-time sensing system enabling edge calculation

Abstract

The invention provides a frequency spectrum real-time sensing system with edge calculation enabling, which relates to the technical field of frequency spectrum sensing, and comprises a generation module used for performing primary processing and analysis on original frequency spectrum signal data, determining a basic characteristic distribution mode of a signal by extracting a main energy region of the signal in time frequency distribution, and generating a frequency spectrum signal; generating a signal analysis correction parameter; the analysis module is used for calibrating and preprocessing the original spectrum signal data by using the signal analysis correction parameters to obtain standardized to-be-analyzed signal data; and based on the standardized to-be-analyzed signal data, respectively performing power spectral density fusion analysis and eigenvalue statistical analysis of the received signal covariance matrix, and generating a first signal existence judgment result and a second signal existence judgment result. According to the invention, the data transmission demand and the back-end processing burden are reduced, and the real-time performance and the response speed of spectrum sensing are improved.

Description

Technical Field

[0001] This invention relates to the field of spectrum sensing technology, and in particular to a real-time spectrum sensing system enabled by edge computing. Background Technology

[0002] In the field of radio spectrum monitoring and sensing, especially for applications in wide-area, dynamic, and complex electromagnetic environments, most existing technical solutions rely on distributed sensor networks for data acquisition. A common approach is to transmit the raw spectrum data or data that has undergone simple preprocessing collected by the front-end sensor nodes back to a remote data center or server for centralized processing and analysis via a communication network. However, this architecture may have some drawbacks in actual deployment.

[0003] For example, in wide-area radio monitoring in border areas, existing systems may deploy multiple sensor nodes along the border. These nodes continuously collect spectrum data and attempt to send the data stream containing a large amount of raw information to the back-end processing center via the backhaul link. Due to the characteristics of electromagnetic signals, such as suddenness, wide bandwidth, or high density, the amount of data generated may be large. This may put pressure on the limited bandwidth of the backhaul link, leading to increased data transmission delays or affecting the integrity of data transmission under unstable communication conditions. At the same time, the centralized processing of all data also places high demands on the real-time computing and analysis capabilities of the back-end server. The process from signal acquisition to the final generation of usable situational information may have a certain time lag, which may make it difficult to fully meet the perception and response requirements with high real-time requirements. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide an edge computing-enabled real-time spectrum sensing system that reduces data transmission requirements and back-end processing burden, and improves the real-time performance and response speed of spectrum sensing.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, an edge computing-enabled real-time spectrum sensing system includes: The storage module is used to receive and store raw spectrum signal data from edge computing nodes at the front end of spectrum sensors equipped with array antennas deployed in the monitoring network; The generation module is used to perform preliminary processing and analysis on the raw spectrum signal data. By extracting the main energy regions of the signal in the time-frequency distribution, it determines the basic characteristic distribution pattern of the signal and generates signal analysis and correction parameters. The analysis module is used to calibrate and preprocess the original spectrum signal data using signal analysis correction parameters to obtain standardized signal data to be analyzed. Based on the standardized signal data to be analyzed, power spectral density fusion analysis and eigenvalue statistical analysis of the received signal covariance matrix are performed to generate the first signal existence decision result and the second signal existence decision result. The fusion module is used to fuse the existence decision results of the first signal and the existence decision results of the second signal to generate a final decision result on whether the target signal exists. The acquisition module is used to call the standardized signal data to be analyzed for spatial spectrum analysis if the final decision result is that the target signal exists. By calculating the array covariance matrix and performing spatial spectrum estimation, the direction of arrival information of the radiation source is obtained. The backhaul module is used to transmit the direction of arrival information, the final decision result, and the compressed signal feature data extracted from the standardized signal data to be analyzed back to the backend server, so as to realize real-time perception and intelligent analysis of the spectrum environment.

[0006] In a second aspect, a computing device includes: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, enable the one or more processors to implement the system.

[0007] Thirdly, a computer-readable storage medium storing a program that, when executed by a processor, implements the system.

[0008] The above-described solution of the present invention has at least the following beneficial effects: Through the collaborative design of architecture optimization and multi-dimensional signal processing mechanisms, the performance of spectrum sensing has been improved. Edge computing nodes are used to directly store, pre-process, and generate correction parameters for the raw signal at the front end of the spectrum sensor, reducing the amount of raw data transmitted between the front end and back end servers, lowering transmission latency, ensuring the real-time performance of spectrum sensing, and avoiding signal distortion during remote transmission. By extracting the main energy regions of the time-frequency distribution to determine the basic characteristic distribution pattern of the signal and generating correction parameters, a basis for signal calibration and preprocessing is provided, improving the accuracy of signal standardization processing. A dual decision mechanism combining power spectral density fusion analysis and statistical analysis of the eigenvalues ​​of the received signal covariance matrix is ​​adopted, and the two types of decision results are then fused for comprehensive judgment, improving the reliability and robustness of the target signal existence determination and reducing the probability of false positives and false negatives. After confirming the existence of the target signal, spatial spectrum analysis is used to accurately obtain the direction of arrival information of the radiation source, realizing the processing from signal detection to radiation source location, providing key support for spectrum resource management. In the backhaul stage, only the direction of arrival information, the final decision result, and compressed signal characteristic data are transmitted, reducing transmission bandwidth usage and energy consumption while ensuring that the back end server obtains core analysis information. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of an edge computing-enabled real-time spectrum sensing system provided by an embodiment of the present invention.

[0010] Figure 2 This is a schematic diagram of an embodiment of the present invention, which shows the fusion determination of the existence of a first signal and the existence of a second signal to generate a final determination of whether a target signal exists. Detailed Implementation

[0011] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0012] like Figure 1 As shown, an embodiment of the present invention proposes an edge computing-enabled real-time spectrum sensing system, comprising: The storage module is used to receive and store raw spectrum signal data from edge computing nodes at the front end of spectrum sensors equipped with array antennas deployed in the monitoring network; The generation module is used to perform preliminary processing and analysis on the raw spectrum signal data. By extracting the main energy regions of the signal in the time-frequency distribution, it determines the basic characteristic distribution pattern of the signal and generates signal analysis and correction parameters. The analysis module is used to calibrate and preprocess the original spectrum signal data using signal analysis correction parameters to obtain standardized signal data to be analyzed. Based on the standardized signal data to be analyzed, power spectral density fusion analysis and eigenvalue statistical analysis of the received signal covariance matrix are performed to generate the first signal existence decision result and the second signal existence decision result. The fusion module is used to fuse the existence decision results of the first signal and the existence decision results of the second signal to generate a final decision result on whether the target signal exists. The acquisition module is used to call the standardized signal data to be analyzed for spatial spectrum analysis if the final decision result is that the target signal exists. By calculating the array covariance matrix and performing spatial spectrum estimation, the direction of arrival information of the radiation source is obtained. The backhaul module is used to transmit the direction of arrival information, the final decision result, and the compressed signal feature data extracted from the standardized signal data to be analyzed back to the backend server, so as to realize real-time perception and intelligent analysis of the spectrum environment.

[0013] In this embodiment of the invention, the performance of spectrum sensing is improved through the collaborative design of architecture optimization and multi-dimensional signal processing mechanisms. Edge computing nodes are used to directly store, preliminarily process, and generate correction parameters for the raw signal at the front end of the spectrum sensor, reducing the amount of raw data transmitted between the front end and the back end server, lowering transmission latency, ensuring the real-time performance of spectrum sensing, and avoiding signal distortion during remote transmission. By extracting the main energy regions of the time-frequency distribution to determine the basic characteristic distribution pattern of the signal and generating correction parameters, a basis for signal calibration and preprocessing is provided, improving the accuracy of signal standardization processing. A dual decision mechanism combining power spectral density fusion analysis and statistical analysis of the eigenvalues ​​of the received signal covariance matrix is ​​adopted, and the two types of decision results are then comprehensively judged through fusion, improving the reliability and robustness of the target signal existence determination and reducing the probability of misjudgment and missed judgment. After confirming the existence of the target signal, spatial spectrum analysis is used to accurately obtain the direction of arrival information of the radiation source, realizing the processing from signal detection to radiation source location, providing key support for the management and control of spectrum resources. In the backhaul stage, only the direction of arrival information, the final decision result, and the compressed signal characteristic data are transmitted, reducing transmission bandwidth usage and energy consumption while ensuring that the back end server obtains core analysis information.

[0014] In a preferred embodiment of the present invention, the receiving and storage of raw spectrum signal data by an edge computing node at the front end of a spectrum sensor deployed in a monitoring network and equipped with an array antenna may include: In this embodiment of the invention, a comprehensive topographic survey and electromagnetic environment assessment of the target monitoring area are first conducted to determine the signal propagation characteristics, potential obstructions, and key monitoring frequency bands within the area. Based on the signal reception range of the spectrum sensors and the coverage angle of the array antennas, the deployment density and specific locations of the sensor front-ends are planned to ensure that the monitoring ranges of adjacent sensors have no blind spots and form a distributed monitoring network covering the entire area. Simultaneously, installation space for edge computing nodes is reserved to ensure that the physical distance between the nodes and the corresponding sensor front-ends is within a reasonable range, reducing signal transmission loss. According to the signal characteristics of the target monitoring frequency band, suitable array antennas are selected, such as long-line antennas for shortwave and log-periodic antennas for VHF, ensuring that the antenna's operating frequency band is perfectly matched to the sensor's receiving frequency band. At each preset deployment point, a stable antenna mounting bracket is erected, and the antenna is positioned at preset angles, such as a 0° elevation angle for horizontal signals and a 30°-60° elevation angle for aerial signals, based on the terrain height difference. Ensure the antenna is fixed at least 3 meters above major obstructions. After installation, use specialized instruments to test the antenna's VSWR and impedance matching. Optimize the antenna's signal reception efficiency by adjusting the installation angle and feeder length to ensure stable capture of electromagnetic signals of different directions and intensities within the area. Finally, connect the antenna to the signal input interface of the spectrum sensor front end via a shielded feeder, ensuring the feeder's outer layer is treated for wear and electromagnetic interference protection. Fix the spectrum sensor front end in the preset installation position, ensuring the equipment is in a dry, shockproof, and electromagnetic interference-free environment. Connect a stable power supply to the sensor front end and start the device for self-testing, checking the sensor's signal receiving module, data transmission interface, and status indicator lights for proper functioning. Set the sensor's operating mode to raw signal acquisition mode via the device's local control panel or temporary debugging terminal, disabling any preset signal preprocessing functions to ensure the sensor only performs pure acquisition and conversion of the signal received by the antenna, without any filtering, amplification, or other modifications.

[0015] Industrial-grade edge computing nodes are deployed in the equipment installation area near each spectrum sensor front-end. The connection method is selected based on the site environment. If there is no strong electromagnetic interference, a shielded Ethernet cable is used for wired connection between the node and the sensor front-end, with waterproof connectors at both ends to ensure a stable physical link. If wiring is inconvenient or there is strong electromagnetic interference, a short-range industrial wireless communication link is used. The wireless communication parameters of the node and sensor are adjusted to ensure a signal strength ≥-70dBm with no significant packet loss. After connection, a communication integration test is performed. The sensor front-end sends a test signal, which the edge computing node receives and reports the reception status, confirming that the data transmission format and rate are perfectly matched and that timing synchronization is error-free. After all equipment deployment and connection are completed, the overall system's signal acquisition process is initiated. The spectrum sensor front-end remains in standby receiving mode, with the array antenna scanning the electromagnetic environment within the monitoring area in real time, capturing all raw electromagnetic signals in space that conform to the monitoring frequency band, including various communication signals, radar signals, navigation signals, and other naturally propagated signals without any artificial processing. The antenna transmits the captured electromagnetic signals to the sensor front-end's signal receiving mode via a feeder, where the mode converts the electromagnetic signals into electrical signals recognizable by the sensor. The system receives analog or raw digital electrical signals, preserving their original amplitude, phase, and timing characteristics. These converted signals are then continuously transmitted to edge computing nodes via established wired or wireless links, adhering to real-time transmission principles. The edge computing nodes initiate a data reception process, continuously monitoring the signal data stream from the corresponding sensor front-end. Upon receiving each data segment, real-time reception verification is performed to check the data's timing continuity and amplitude range. If any anomalies are detected, a retransmission request is immediately sent to the sensor front-end to ensure the received data fully restores the original electromagnetic signal characteristics. After successful verification, the edge computing nodes proceed according to a pre-defined procedure. The original data storage rules are followed, and the data is written to local industrial-grade high-speed storage media. During storage, each data segment is uniquely identified by a combination of timestamp, sensor number, and deployment point coordinates. The data format remains unchanged from the original transmission format, without any compression, filtering, feature extraction, or other processing. Simultaneously, storage status monitoring is initiated to check the remaining capacity of the storage media in real time. When the capacity reaches a preset threshold, an alarm is triggered. In addition, a local dual backup mechanism is adopted to synchronously store the same batch of original data to the backup storage media within the node, preventing data loss due to the failure of a single storage media and ensuring the safe and complete preservation of the original spectrum signal data.

[0016] In a preferred embodiment of the present invention, preliminary processing and analysis of the original spectral signal data are performed. By extracting the main energy regions of the signal in the time-frequency distribution, the basic characteristic distribution pattern of the signal is determined, and signal analysis and correction parameters are generated. This may include: In this embodiment of the invention, time-frequency transformation processing is performed on the original spectrum signal data to generate a time-frequency matrix describing the two-dimensional distribution of signal energy with time and frequency. Specifically, the edge computing node first retrieves the original spectrum signal data for the target monitoring period from its local industrial-grade high-speed storage medium in chronological order of acquisition time. During the retrieval process, the integrity of the unique identifier of the data is simultaneously verified, and the continuity of the acquisition timestamps, the matching of sensor numbers with the current deployment points, and the accuracy of coordinate information are checked. If a data segment is found to be missing or mismatched, a local data re-inspection process is immediately triggered to re-retrieve the data for that period, ensuring that the data involved in the processing completely corresponds to the current monitoring area and time period. There is no storage corruption or missing data segments. Considering the suddenness of signals in wide-area monitoring, such as the burst period of shortwave communication signals at borders which is mostly 10-50 milliseconds, the edge computing nodes split the original data according to the principle of adapting to the minimum possible duration of the signal. First, through previous scenario measurements, the minimum effective duration of the target monitoring frequency band signal is statistically determined, and the segment duration is determined based on this. For shortwave scenarios, 10 milliseconds is taken as the single segment duration, and for ultra-shortwave and radar signal scenarios, 1 millisecond is taken as the single segment duration. When splitting, the original data is cut in the order of acquisition time. Each segment is marked with a segment number and start and end timestamps. This avoids the processing delay caused by the large amount of data in a single segment, and can completely capture a single burst signal event.

[0017] For each segment of the original signal data after splitting, the edge computing nodes perform time-frequency transformation processing one by one. The time axis dimension is determined based on the original signal acquisition rate, such as 100 MSps for shortwave and 1 GSps for ultra-shortwave, meaning 100 million and 1 billion data points are acquired per second respectively. Each time segment is divided into several equal time points, each corresponding to a digital sample value of the original signal. The time point interval is the reciprocal of the acquisition rate; for example, 100 MSps corresponds to a 10 nanosecond interval, ensuring no compression or omissions in the timing sequence. The frequency axis dimension is determined based on the preset total monitoring frequency range for the monitoring scenario, such as 2MHz-30MHz for shortwave and 30MHz-6GHz for ultra-shortwave. Continuous frequency points are divided according to the standard of adapting to the signal frequency identification accuracy. For the shortwave scenario, the frequency points are divided at 1kHz intervals, totaling 28,000 frequency points, covering the entire shortwave band without omitting narrowband signals. For the ultra-shortwave scenario, the frequency points are divided at 1MHz intervals, totaling 5,970 frequency points, balancing coverage and processing efficiency. The frequency points are numbered sequentially from low to high to ensure full band coverage. To calculate the energy value at the time-frequency intersection, the amplitude data of the original signal at each time point is first extracted from the digital sampled values ​​after analog-to-digital conversion by the sensor. Each sampled value directly corresponds to the voltage amplitude of the original electromagnetic signal. Then, this amplitude data is multiplied by itself to obtain the squared amplitude value. Finally, the squared amplitude value is divided by the fixed impedance of the signal transmission link, such as 50 ohms, which is the standard impedance of the link consisting of the sensor, antenna, and transmission cable, and remains constant. This yields the actual energy value of the original signal at that time and frequency, ensuring that the energy calculation closely matches the hardware transmission characteristics. A time-frequency matrix is ​​constructed by sequentially filling the energy values ​​of all time-frequency intersections into the corresponding cells of a two-dimensional table, in order from early to late time and from low to high frequency. Each cell uniquely corresponds to a time point and a frequency point, with the row index corresponding to the time point number and the column index corresponding to the frequency point number. During the filling process, the energy value of each cell is checked in real time to ensure that it is within a reasonable range. Finally, a complete time-frequency matrix is ​​formed, which intuitively presents the details of the energy strength distribution of the original spectrum signal at different times and frequencies.

[0018] The energy distribution of the time-frequency matrix is ​​analyzed, an energy segmentation threshold is calculated, and segmentation is performed based on the energy segmentation threshold to extract the main energy regions of the signal in the time-frequency distribution. Specifically, the edge computing node first traverses the entire time-frequency matrix in row-major order, first traversing all frequency point cells of the first time point, and then traversing the cells of subsequent time points in turn. A comprehensive statistical analysis is performed on the energy values ​​of all cells to extract extreme values ​​and the mean. The maximum and minimum values ​​of all energy values ​​are recorded, and all energy values ​​are summed to obtain the total energy. The total energy sum is divided by the total number of cells in the matrix (number of time points multiplied by number of frequency points) to obtain the average energy. The median is calculated by sorting all energy values ​​in ascending order. If the total number is odd, the value in the middle position after sorting is taken as the median; if it is even, the value in the middle two positions is taken as the median. The median is the average of the individual values, reflecting the intermediate level of the energy distribution. The standard deviation is calculated by subtracting the average energy value from each energy value, then multiplying each difference by itself (resulting in a squared difference). The sum of all squared differences is then calculated, divided by the total number of cells, and the result is squared to obtain the standard deviation, reflecting the dispersion of energy values ​​from the average, i.e., the fluctuation of noise. Considering the noise characteristics of the wide-area electromagnetic environment, an energy segmentation threshold is calculated based on the core objective of distinguishing between signal and noise. The basic threshold calculation first calculates the difference between the maximum and average energy values, then multiplies this difference by a preset proportional coefficient. This coefficient is calibrated through prior scene measurements. Border scene noise is complex; in actual measurements, the difference between signal and noise energy in 80% of cases falls within 0.3-0.5 times the average, therefore, a range of 0.3-0.5 is chosen.5. Add the calculated result to the average energy value to obtain the basic segmentation threshold. For threshold correction, optimize the basic threshold by referencing the energy median and standard deviation. If the standard deviation is large, increase the correction by 10% on top of the basic threshold; if the standard deviation is small, decrease the correction by 5%, ultimately obtaining an energy segmentation threshold that balances signal recognition rate and noise filtering effect. Then, initiate the segmentation and effective region extraction of the time-frequency matrix, marking effective energy points. Check the energy value of each cell in the matrix one by one. If the energy value of a cell is greater than or equal to the energy segmentation threshold, mark the cell as an effective energy point, determining that it may contain signal energy, and simultaneously record the cell's time point number and frequency point number. If the energy value is less than the energy segmentation threshold, mark it as an invalid energy point, determining it as noise energy. After marking, perform a second check on all effective energy points, re-verifying the relationship between the energy value and the threshold to avoid misjudgment. For effective energy point region integration, use the eight-neighborhood expansion method to integrate continuous effective energy points. First, select the first unclassified effective energy point as the initial core point, checking its top, bottom, left, right, top left, bottom left, and right... Check if there are any valid energy points in the top, bottom right, and eight adjacent cells. If so, group these valid energy points and the initial core point into the same energy block. Then, use the valid energy points at the edge of this block as the new core point and repeat the eight-neighbor check and merging operation until the block can no longer be expanded. Next, select the next unclassified valid energy point and repeat the above process until all valid energy points are assigned to their corresponding energy blocks. Isolated energy blocks are removed. Set a minimum threshold for the number of valid energy points, based on the division density of frequency points and time points, and the minimum effective signal strength. The time-frequency size is set as follows: for shortwave, 10 effective energy points correspond to a time-frequency range of 10 milliseconds × 1 kHz, which perfectly covers the minimum effective time-frequency size of shortwave signals; for VHF, 5 effective energy points correspond to a time-frequency range of 1 millisecond × 1 MHz, adapting to the characteristics of VHF signals. The number of effective energy points contained in each energy block is counted. If the number is less than a threshold, it is determined to be an isolated block formed by noise and is removed. The energy blocks that are ultimately retained, containing the required number of effective energy points and are continuously distributed, represent the main energy region of the signal in the time-frequency distribution.

[0019] The geometric and statistical characteristics of the main energy regions are analyzed and calculated to determine the basic characteristic distribution pattern of the signal, including the signal's time-frequency clustering, bandwidth occupancy, and duration characteristics. Specifically, edge computing nodes analyze the characteristics of each main energy region in ascending order of region number, ultimately integrating and calculating three core characteristics. The time-frequency clustering is calculated by first summing the energy of each main energy region, then traversing all effective energy points within that region, and accumulating the energy values ​​in cell order to obtain the energy sum of a single region; then, the energy sums of all main energy regions are added together to obtain the total energy value of all main energy regions; simultaneously, the total energy value of the entire time-frequency matrix is ​​calculated by traversing all cells in the matrix, including effective energy points. Accumulate all energy values, including both active and inactive energy points, ensuring no omissions. Divide the total energy value of all major energy regions by the total energy value of the time-frequency matrix; the resulting ratio is the time-frequency concentration. A higher ratio indicates more concentrated signal energy, making it easier to distinguish from dispersed noise. Calculate bandwidth occupancy. For each major energy region, find the maximum and minimum frequencies corresponding to all active energy points within that region. Subtract the minimum frequency from the maximum frequency to obtain the occupied bandwidth of that single major energy region. Combine total occupied bandwidth. If multiple major energy regions exist, first determine if their frequency ranges overlap. If the maximum frequency of region A is greater than the minimum frequency of region B but less than the maximum frequency of region B... If frequency overlap is detected, all frequency ranges are merged into a single continuous frequency range. The minimum frequency value among all regions is taken as the overall minimum, and the maximum frequency value is taken as the overall maximum. If there is no overlap between regions, the overall frequency range is the sum of the frequency ranges of all regions. The bandwidth occupancy rate is calculated by dividing the merged total occupied bandwidth by the preset total monitoring frequency range of the monitoring scenario. For example, the total range for the shortwave scenario is 28MHz, and for the ultra-shortwave scenario it is 5970MHz. The resulting ratio is the bandwidth occupancy rate. The duration characteristic is calculated by calculating the duration of a single region. For each major energy region, the maximum and minimum time values ​​corresponding to all effective energy points within that region are found, and the maximum time value is subtracted from the minimum time value. The minimum value is used to obtain the duration of a single main energy region. The total duration is then combined. If multiple main energy regions exist, the time interval between two adjacent regions is calculated (the minimum time of the latter region minus the maximum time of the former region). If the time interval is less than a preset signal continuity threshold, the durations of the two regions and the interval are combined into a continuous duration. If the interval is greater than the threshold, the durations of each region are calculated separately and then summed to obtain the total signal duration. The time persistence characteristic is calculated by dividing the total duration by the total acquisition time of the original spectrum signal data, i.e., the sum of the durations of all split time segments. The resulting ratio is the time persistence characteristic, clearly reflecting whether the signal exists continuously or intermittently.By integrating the calculated time-frequency clustering, bandwidth occupancy, and time persistence characteristics, and labeling the calculation time and corresponding monitoring area for each characteristic, a basic feature distribution pattern that can comprehensively and accurately reflect the essential time-frequency characteristics of the signal is formed.

[0020] Based on the basic feature distribution pattern, a comprehensive parameterization transformation is performed on the time-frequency clustering, bandwidth occupancy, and time duration characteristics to generate signal analysis correction parameters for signal feature extraction and correction. Specifically, this includes: edge computing nodes first determining the core types of signal analysis correction parameters, then combining these with the three core features in the basic feature distribution pattern to perform targeted parameterization transformations, ensuring a high degree of fit between the parameters and the actual signal characteristics; generating time window correction coefficients to adapt to the signal's time dimension characteristics, ensuring the processing can completely capture the signal's timing information; and setting the basic time window length based on the acquisition rate of the monitoring frequency band and the typical signal time span. For shortwave scenarios with an acquisition rate of 100 MSps, 500 sampling points are required to ensure timing accuracy. The result is calculated by dividing 500 sampling points by 100 MSps. A base time window of 5 milliseconds is established. For ultra-shortwave scenarios with a sampling rate of 1 GSps, 1000 sampling points are required. Dividing 1000 sampling points by 1 GSps yields a base time window of 1 millisecond. The base window is dynamically adjusted: if the proportion corresponding to the duration characteristic is high (60%-70%), a value of 1.2 is used; for 70%-80%, 1.5 is used; and for above 80%, 2.0 is used, indicating a long signal duration, requiring an extended time window for complete coverage. If the proportion is low (10%-20%), 0.5 is used; for 20%-30%, 0.7 is used; and for below 30%, 0.9 is used, indicating a short-duration burst signal, requiring a shortened time window for accurate capture of key segments. Finally, the time window correction coefficient is determined by multiplying the base time window length by the corresponding adjustment coefficient, with the result rounded to four decimal places to ensure adjustment accuracy.

[0021] Generate frequency resolution adjustment parameters to adapt to the signal frequency dimension characteristics, ensuring accurate identification of signal frequency details in subsequent processing. Set the base frequency resolution based on the density of the monitored frequency range division and the requirements for identifying typical signal frequencies. For shortwave scenarios, divide the frequency range into 1kHz intervals and set the base resolution to 1kHz; for VHF scenarios, divide the frequency range into 1MHz intervals and set the base resolution to 1MHz. Dynamically adjust the base resolution. If the bandwidth occupancy is high (40%-50%), use 0.9; for 50%-60%, use 0.7; and for above 60%, use 0.5. This indicates a wide signal bandwidth, requiring a reduction in resolution to expand the coverage of a single analysis and avoid missing wideband signals. If the bandwidth occupancy is low (10%-20%), use 2.0; for 20%-30%, use 1.5; and for below 30%, use 1.2. This indicates a narrowband signal, requiring an increase in resolution to accurately identify frequency details. Finally, determine the frequency resolution adjustment parameters: multiply the base frequency resolution by the corresponding adjustment coefficient, and retain the result to four decimal places to ensure adjustment accuracy.

[0022] An energy normalization benchmark is generated to eliminate energy fluctuations caused by signal strength differences and environmental interference, providing a unified calibration standard. Effective energy points are screened, and the energy values ​​of all effective energy points within all major energy regions are statistically analyzed. The average energy value and standard deviation of these energy values ​​are calculated. Energy points exceeding the average energy value plus three times the standard deviation are identified as abnormally high energy points and removed. The corrected average energy value is calculated, and the energy values ​​of the remaining effective energy points are summed. The sum is divided by the number of remaining effective energy points to obtain the corrected average energy value. The energy normalization benchmark is determined, and the corrected average energy value is retained to six decimal places as the energy normalization benchmark to ensure calibration accuracy. Finally, the time window correction coefficient, frequency resolution adjustment parameters, and energy normalization benchmark are integrated and packaged in a format of parameter type-value-generation timestamp-sensor identifier to form a complete set of signal analysis and correction parameters adapted to the current signal characteristics. Key intermediate data from the parameter generation process is also stored, ultimately providing a core basis for the accurate calibration and standardized preprocessing of the original spectrum signal.

[0023] The calculation of time-frequency clustering, bandwidth occupancy, and time persistence characteristics takes into account details such as multi-region merging, interval determination, and outlier removal, comprehensively capturing the essential characteristics of the signal, making the generated correction parameters match the actual signal situation, reducing the deviation of signal calibration and analysis from the source, and improving the processing accuracy.

[0024] In a preferred embodiment of the present invention, the original spectral signal data is calibrated and preprocessed using signal analysis correction parameters to obtain standardized signal data to be analyzed. Based on the standardized signal data to be analyzed, power spectral density fusion analysis and eigenvalue statistical analysis of the received signal covariance matrix are performed respectively to generate a first signal existence decision result and a second signal existence decision result, which may include: In this embodiment of the invention, based on signal analysis and correction parameters, the original spectrum signal data is calibrated and preprocessed to obtain standardized signal data to be analyzed. Specifically, the edge computing node first retrieves the previously generated signal analysis and correction parameters from its local parameter storage area, verifies the three-dimensional identifier of the parameter's monitoring period, deployment location, and monitoring frequency band one by one, and then retrieves the corresponding original spectrum signal data from the original data storage area according to the completely consistent three-dimensional identifier, ensuring that the parameters and data belong to the same monitoring scenario and preventing misuse across scenarios and time periods. Subsequently, the entire process of four-dimensional fine calibration and three rounds of preprocessing are performed locally on the edge computing node to avoid the increase of original data backhaul. To monitor link pressure in a wide area, the specific operation is as follows: For precise time window calibration, first read the individual time segments already divided from the original spectrum signal data and record their original duration (e.g., 10 milliseconds for shortwave and 1 millisecond for VHF). Then, retrieve the corresponding time window correction coefficient (e.g., 1.8 for long-duration boundary signals, 10 milliseconds × 1.8 = 18 milliseconds). If the correction coefficient is greater than 1, extract the missing time point data from the subsequent adjacent time segments according to the acquisition sequence. For example, a 10-millisecond segment needs 8 milliseconds of data, i.e., 8 × acquisition rate time points. A 100 MSps acquisition rate corresponds to 800 time points. Add this data to the end of the current segment to ensure accuracy. To ensure the supplementary data is sequentially continuous and non-repeating with the original segment, for example, if the original segment's timestamp is 10:00:00.000-10:00:00.010, the supplemented data extends to 10:00:00.018. If the correction coefficient is less than 1, the energy value of each time point within the segment is calculated, sorted from highest to lowest energy value, and the data from the top-ranked time points is selected. The number of selections is calculated as target duration × acquisition rate. A acquisition rate of 0.6 milliseconds × 1 GSps corresponds to 600 time points. This data is then extracted to form the core time period, preserving key timing information while eliminating redundant low-energy data. Fine-tuning of the frequency resolution is performed for the time-calibrated signal segment, first determining... The current frequency point interval is used to determine the adjustment method based on the frequency resolution adjustment parameters. If the parameter is less than 1, the merging ratio is calculated as the reciprocal of the parameter. Every two adjacent frequency points are grouped together, and the energy values ​​of all frequency points in each group are accumulated. Then, the energy value of each group is divided by the number of frequency points in the group to obtain the energy value of a single frequency point after merging. The original frequency points are replaced sequentially from low to high frequency to ensure that no frequency is missed in the broadband signal. If the parameter is greater than 1, the number of subdivisions is calculated as the integer part of the parameter. Each original frequency point is divided into two subdivision frequency points, and the interval between the subdivision frequency points is calculated as the original interval ÷ the number of subdivisions. The energy value of the original frequency point is evenly distributed to the subdivision points to accurately capture the frequency details of the narrowband signal.

[0025] Energy normalization uses a unified standard. It iterates through the energy values ​​of all cells in the time- and frequency-calibrated data, retrieves the previously generated energy normalization benchmark, and divides each energy value by this benchmark to obtain the normalized energy value. This operation completely eliminates energy fluctuations caused by differences in signal strength and environmental interference across different time periods, ensuring that the signal energy collected by different nodes and at different times within the same monitoring network is on the same quantitative dimension, providing a unified basis for cross-node collaborative analysis. Three rounds of refined data cleaning are performed. The first round removes outliers by calculating the average energy value of all normalized energy values ​​and setting a reasonable fluctuation range of 0.1 to 10 times the average energy value (0.2-20.0). Each energy value is checked individually; if it exceeds this range, it is considered an extreme outlier. The result of adding the two adjacent energy values ​​before and after the outlier and dividing by 2 is used to replace the outlier; the second round of invalid segment removal counts the number of consecutive zero energy value time points, sets a threshold of 5 for shortwave scenarios and 3 for ultra-shortwave scenarios, adapts to the minimum duration of the signal, the 5 time points for shortwave correspond to 50 nanoseconds and a 100 MSps acquisition rate; the 3 time points for ultra-shortwave correspond to 3 nanoseconds and a 1 GSps acquisition rate. If consecutive zero values ​​reach the threshold, it is determined to be an invalid period with no signal and is directly removed; the third round of format standardization organizes the processed signal data into a fixed format of time point number-frequency point number-normalized energy value, stores it in the standardized data area of ​​the edge computing node, and marks it with calibration preprocessing completion identifier, processing timestamp and parameter association ID for easy traceability and retrieval.

[0026] The process involves estimating the power spectral density of standardized signal data to obtain the initial spectral energy distribution for each node or time period, and then fusing these initial spectral energy distributions to generate a comprehensive spectral energy distribution. Specifically, this includes: edge computing nodes determining the data processing dimensions based on the monitoring network deployment mode; single-node multi-time periods or multi-node simultaneous operation to ensure the processing dimensions adapt to the collaborative needs of wide-area monitoring; data grouping preprocessing; for single-node multi-time periods, dividing the signal into groups of 10 consecutive standardized signal segments based on signal temporal continuity, with each group labeled with node ID-group number-time period range; for multi-node simultaneous operation, synchronously retrieving standardized data from the same monitoring time period from each node, verifying timestamp accuracy, and if discrepancies exist, aligning all node numbers using the earliest node's timestamp as the benchmark. To ensure complete consistency in the data acquisition timeframe, power spectral density is estimated group by group and segment by segment. For each data group, the standardized signal segments within the group are processed one by one. Taking a single segment as an example, the calibrated duration of the segment is read first. The normalized energy values ​​of all frequency points within the segment are traversed. The energy value of each frequency point is divided by the calibrated duration of the segment to obtain the power spectral density of that frequency point, reflecting the energy intensity of that frequency per unit time. For example, if the energy value of a certain frequency point is 16.0, 16.0 ÷ 8 = 2.0. This operation is repeated to complete the power spectral density calculation for all signal segments within the group. An initial spectral energy distribution is generated for each segment or each node. A continuous distribution curve is formed with frequency as the horizontal axis and power spectral density as the vertical axis, which intuitively presents the spectral energy state of that time period.

[0027] The first step of the multi-source initial distribution weighted fusion process involves assigning weight coefficients based on data quality assessment results. Weights range from 0.05 to 0.9, with a total of 1.0. For example, data with over 98% integrity and normalized energy fluctuation less than 20% of the average energy value receives a weight of 0.9; 90%-98% integrity with fluctuation of 20%-30% receives a weight of 0.6; 80%-90% integrity with fluctuation of 30%-50% receives a weight of 0.3; and integrity below 80% receives a weight of 0.05. For instance, three nodes might have weights of 0.4, 0.3, and 0.3, totaling 1.0. The second step involves frequency point-by-frequency point fusion, iterating through all frequency points and extracting time-related parameters for each frequency point. The power spectral density value corresponding to each segment is multiplied by its respective weighting coefficient, and then all the product results are added together to obtain the fused power spectral density value of that frequency point. For example, at a frequency point of 15MHz, the density of node 1 is 3.0×0.4=1.2, the density of node 2 is 2.5×0.3=0.75, the density of node 3 is 2.8×0.3=0.84, and the fused value is 1.2+0.75+0.84=2.79. The third step is to generate a comprehensive distribution. The fused power spectral density values ​​of all frequency points are arranged in order from low to high frequency to form a comprehensive spectrum energy distribution covering the entire monitoring frequency band, which effectively offsets the interference caused by random noise and equipment errors in a single time period.

[0028] The comprehensive spectrum energy distribution is analyzed, and the extracted energy features are compared with a preset first decision threshold. Based on the comparison result, a first signal existence decision result is generated. Specifically, the process includes: edge computing nodes focusing on the comprehensive spectrum energy distribution, combining the noise characteristics of wide-area monitoring, and carrying out a process of precise feature extraction, scenario-based threshold retrieval, and multi-condition comparison and judgment. The core energy feature is extracted in three dimensions, and the first feature is the peak power spectral density. The power spectral density values ​​of all frequency points in the comprehensive spectrum energy distribution are traversed, the maximum value is recorded, and the frequency point corresponding to the maximum value and the distribution of adjacent frequency points are marked. Feature 2: Effective frequency band average power spectral density. First, calculate the average power spectral density of all frequency points in the comprehensive distribution. Select frequency points with a power spectral density greater than 1.2 times the average value (i.e., >2.4) and determine them as effective frequency bands. Accumulate the power spectral density values ​​of all frequency points within these effective frequency bands and then divide by the number of frequency points in the effective frequency bands to obtain the effective frequency band average power spectral density. Feature 3: Energy proportion. Accumulate the sum of the power spectral densities of all frequency points within the effective frequency bands and then divide by the sum of the power spectral densities of all frequency points in the entire monitoring frequency band to obtain the energy proportion, which reflects the degree of concentration of the signal in the frequency domain.

[0029] The first decision threshold retrieval and explanation for the scenario: The threshold generation logic is based on historical noise data statistics of the monitoring scenario. Taking the border scenario as an example, noise data from the past 30 days of no-signal periods is collected. 100 time segments are selected each day to generate 3000 comprehensive spectral energy distributions of noise. The peak value, effective average value, and energy percentage of each distribution are extracted. The maximum value of these three indicators is taken and multiplied by 1.5 (adapting to the large fluctuations in border noise) to obtain the first decision threshold for the border scenario: peak threshold 4.5, average threshold 3.75, and percentage threshold 0.45. For the 30 days of no-signal period noise data, the maximum value of each indicator is multiplied by 1.2 to generate corresponding thresholds, such as peak threshold 3.0, average threshold 2.4, and percentage threshold 0.36. Multi-condition comparison and judgment: Four judgment conditions are set, including peak power spectral density ≥ peak... The criteria are as follows: Threshold values ​​are defined as follows: effective average power spectral density ≥ average threshold; energy percentage ≥ percentage threshold; at least three adjacent frequency points corresponding to the peak value have power spectral density values ​​greater than the average threshold. Conditions are verified one by one. Taking a border scenario as an example, if the peak value is 4.5 ≥ 4.5 (satisfied) and the effective average power spectral density is 3.125 < 3.75 (not satisfied), then the target signal is directly determined not to exist. If the peak value is 4.6 ≥ 4.5, the effective average power spectral density is 3.8 ≥ 3.75, the percentage power spectral density is 0.6 ≥ 0.45, and the densities of the three adjacent frequency points (3.9, 4.0, 3.8) are all > 3.75 (all four conditions are satisfied), then the target signal is determined to exist. A decision result is generated: if all conditions are met, the first signal existence decision result is "existent"; otherwise, it is "non-existent". All feature values, threshold values, and specific items where conditions are not met are recorded to provide a basis for threshold optimization.

[0030] Using the signal data to be analyzed, a received signal covariance matrix is ​​constructed, and eigenvalue decomposition is performed on the received signal covariance matrix to obtain an eigenvalue sequence. Specifically, this includes: edge computing nodes retrieving locally standardized signal data to be analyzed; combining this with the multi-channel reception characteristics of an array antenna (e.g., deploying an 8-channel array antenna, with each channel receiving signals independently, adapting to the multi-directional acquisition requirements of wide-area signals); and performing three steps: data processing and adaptation, covariance matrix construction, and eigenvalue decomposition. For multi-channel data processing and adaptation, standardized data is classified according to the receiving channel number (1-8), with each channel corresponding to a set of time-series data. The length of each channel's data is calculated; if channel 1 has 1... With 200 time points and 1180 channels in total, using the shortest channel of 1180 as the standard, the first 1180 time points of data from all channels are extracted to ensure that the data length of all 8 channels is completely consistent. Using the time points as rows and the receiving channels as columns, the timing data of all channels are organized into a two-dimensional data table. Each row represents the signal values ​​of the 8 channels at a time point. For example, row 1 is [2.1, 2.0, 1.9, 2.2, 2.0, 1.8, 2.3, 2.1], and row 1180 is [1.7, 1.8, 1.6, 1.9, 1.7, 1.5, 1.8, 1.6], forming a two-dimensional data structure of 1180 rows × 8 columns.

[0031] The received signal covariance matrix is ​​constructed in 8×8 dimensions. Each element in the matrix, at row i and column j (i, j = 1-8), corresponds to the covariance value between channel i and channel j. The calculation method is as follows: Calculate the average value of channel i by summing the signal values ​​at all 1180 time points of channel i and then dividing by 1180. For example, the total sum for channel 1 is 2360, so the average value is 2360 ÷ 1180 = 2.0. Similarly, calculate the average value of channel j by summing all signal values ​​at channel j and then dividing by 1180. For example, the total sum for channel 2 is 2360, so the average value is 2360 ÷ 1180 = 2.0. Given 2320, the average value is 2320 ÷ 1180 ≈ 1.966. Calculate the covariance contribution value at each time point: extract the signal value of channel i at that time point, subtract the average value of channel i to obtain the channel i difference; extract the signal value of channel j at that time point, subtract the average value of channel j to obtain the channel j difference; multiply the two differences to obtain the contribution value at that time point. For example, at time point 1, the channel 1 difference is 2.1 - 2.0 = 0.1, and the channel 2 difference is 2.0 - 1.966 = 0. 034, contribution value 0.1×0.034=0.0034; calculate the covariance value, sum the contribution values ​​of all 1180 time points, such as the total of 23.58, then divide by (1180-1)=1179 to get the covariance value of channel i and j, 23.58÷1179≈0.020; according to the above method, calculate the covariance value of 8×8=64 matrix elements one by one, fill in the corresponding matrix positions to form the received signal covariance matrix. The channel correlation of signal components is strong and the covariance value is large; the channel correlation of noise components is weak and the covariance value is small; eigenvalue decomposition and sequence generation, decompose the constructed 8×8 received signal covariance matrix, extract all 8 eigenvalues ​​of the matrix, sort them in descending order of eigenvalue value to form an eigenvalue sequence, among which the first 3 larger eigenvalues ​​mainly correspond to the energy contribution of the signal, and the last 5 smaller eigenvalues ​​mainly correspond to the energy contribution of the noise, clearly reflecting the difference between the signal and noise in channel correlation.

[0032] Statistical analysis is performed on the eigenvalue sequence to calculate the eigenvalue statistics. These statistics are then compared with a preset second decision threshold. Based on the comparison result, a second signal existence decision is generated. Specifically, the process involves edge computing nodes performing precise statistic calculation, dynamic threshold retrieval, and comparison / determination on the eigenvalue sequence, taking into account the eigenvalue distribution patterns in a pure noise scenario within a wide-area electromagnetic environment. The precise calculation of the eigenvalue statistics begins by determining the selection quantity K. The total length of the eigenvalue sequence is equal to the number of array antenna channels (8), and K is taken as 1 / 3 to 1 / 2 of the total length. (8 / 3 ≈ 2.67, take 3), adapting to the characteristic that the signal energy is concentrated in the first few eigenvalues; the second step is to calculate the sum of the two classes, accumulating the first 3 largest eigenvalues ​​in the sequence to obtain the sum of the first K eigenvalues ​​(6.8 + 5.5 + 4.2 = 16.5); accumulating all 8 eigenvalues ​​in the sequence to obtain the sum of the eigenvalues ​​of the entire sequence (6.8 + 5.5 + 4.2 + 1.3 + 1.1 + 0.9 + 0.8 + 0.7 = 21.3); the third step is to calculate the statistic, dividing the sum of the first K eigenvalues ​​by the sum of the eigenvalues ​​of the entire sequence to obtain the cumulative proportion of eigenvalues. The measurement (16.5 ÷ 21.3 ≈ 0.775) shows a significantly higher proportion when a signal is present and a lower proportion when there is only noise. The dynamic second decision threshold retrieval and update logic is based on measured data from a pure noise scenario. Noise data is collected during signal-free periods in the monitoring area, generating the corresponding covariance matrix and eigenvalue sequence, and calculating the cumulative proportion of eigenvalues. This operation is repeated to collect 100 sets of noise data, resulting in 100 statistics. For example, if the average is 0.45, this average is multiplied by 1.5 to obtain the second decision threshold: 0.45 × 1.5. =0.675; The threshold is dynamically updated. The noise characteristics of the border scene are less affected by the season, and the threshold is updated once a quarter; The statistical quantity is compared with the threshold. The statistical quantity is directly compared with the threshold. If the statistical quantity is greater than or equal to the second decision threshold, it means that the energy proportion of the first K large eigenvalues ​​is significant and there are obvious signal correlation components in the matrix. The target signal is determined to exist, and the second signal existence decision result is generated as "existence". If the statistical quantity is less than the threshold, it means that the energy distribution of the eigenvalues ​​is uniform and the matrix is ​​dominated by noise correlation components. The target signal is determined to not exist, and the decision result is generated as "not existence".

[0033] The four-dimensional operation of calibration and preprocessing, combined with scenario-based parameter adaptation, provides a highly unified and standardized data foundation for collaborative analysis of wide-area distributed monitoring networks, reduces analysis bias, improves the accuracy and stability of signal recognition in complex noise environments, and reduces data transmission delay.

[0034] like Figure 2 As shown, in a preferred embodiment of the present invention, fusing the existence determination results of the first signal and the existence determination results of the second signal to generate a final determination result of whether the target signal exists may include: In this embodiment of the invention, the existence determination results of the first and second signals are uniformly converted into standardized determination identifiers conforming to a preset data format. Specifically, this includes: firstly, determining that the source of the first signal existence determination result is the conclusion output by the front-end edge node after performing power spectral density fusion analysis on the standardized signal data to be analyzed; and secondly, determining that the conclusion output by the front-end edge node after performing statistical analysis of the eigenvalues ​​of the received signal covariance matrix on the standardized signal data to be analyzed. Both original outputs are stored in the temporary data unit of the front-end edge node as textual determination records, containing only three qualitative descriptions: presence of a target signal, absence of a target signal, and indeterminability to determine the presence of a target signal. Secondly, a unified preset data format is pre-defined as a single numerical identifier, and the numerical range is limited to three fixed values: 0, 0.5, and 1. Simultaneously, a unique mapping rule is established, mapping the qualitative description result of the presence of a target signal to standardized identifier 1, the absence of a target signal to standardized identifier 0, and the indeterminability to determine the presence of a target signal to standardized identifier 0. The standardization identifier is set to 0.5. This mapping rule is pre-stored in the configuration file of the front-end edge node to ensure that the conversion standard of all judgment results remains unchanged. Next, the original text records of the first signal existence judgment result and the second signal existence judgment result are extracted from the temporary data unit of the front-end edge node. The specific description of each record is checked one by one to confirm that it belongs to one of the three categories: the target signal exists, the target signal does not exist, or the existence of the target signal cannot be determined. During the check, any text redundancy or formatting errors in the original records must be eliminated to ensure that the extracted judgment information is accurate. Finally, according to the mapping rule preset in the configuration file, the original qualitative description of the first signal existence judgment result is converted into the corresponding numerical identifier. At the same time, the original qualitative description of the second signal existence judgment result is also converted into a numerical identifier of the same data type and the same numerical range of 0-1. After the conversion is completed, the format of the two standardized identifiers is checked to confirm that their values ​​are only one of 0, 0.5, and 1, with no other invalid values, ensuring that the standardized identifiers of the two judgment results are completely consistent in data format.

[0035] Based on the judgment type represented by the standardized judgment identifier and the preset perception strategy, the corresponding judgment fusion rule is selected from the predefined rule set. Specifically, this includes: First, the original judgment type is deduced by the specific value of the standardized identifier, i.e., based on the mapping relationship that a value of 1 corresponds to the existence of a target signal, a value of 0 corresponds to the absence of a target signal, and a value of 0.5 corresponds to the inability to determine whether a target signal exists, the original judgment conclusions represented by the first and second standardized judgment identifiers are determined respectively, and then the combination type of the two judgment identifiers is determined. All possible combination types include six categories: existence + existence, existence + uncertainty, existence + absence, uncertainty + uncertainty, uncertainty + absence, and absence + absence, with no other additional combination cases. Second, the preset perception strategy is determined based on the actual monitoring needs of wide-area dynamic complex electromagnetic environments, mainly including two core strategies. One is the wide-area monitoring real-time response priority strategy, which is suitable for scenarios such as border wide-area radio monitoring that require rapid output of judgment results. The core requirement is to shorten the judgment time while taking into account the basic judgment accuracy. The other is the complex electromagnetic environment accurate judgment priority strategy, which is suitable for scenarios with many electromagnetic signal interferences and complex signal characteristics. The core requirement is to improve the judgment accuracy and allow for The computation time is increased appropriately. Then, a predefined rule set is consulted. This rule set is pre-stored in the rule database of the front-end edge nodes and is a complete rule system based on the matching relationship between all six types of decision combinations and the two types of perception strategies. The specific matching logic is as follows: for the "existence + existence" combination, it corresponds to a rule that directly determines the existence of the target signal under both types of perception strategies; for the "non-existence + non-existence" combination, it corresponds to a rule that directly determines the non-existence of the target signal under both types of perception strategies; for the "existence + non-existence" combination, it corresponds to a rule that compares the weighted operation with the threshold under both types of perception strategies, and the weight allocation is as follows: The differences are as follows: For the three combinations of existence + uncertainty, uncertainty + uncertainty, and uncertainty + non-existence, the simplified weighted calculation and threshold comparison rules are used under the wide-area monitoring real-time response priority strategy, while the complete weighted calculation and threshold comparison rules are used under the complex electromagnetic environment precise judgment priority strategy. Finally, based on the preset perception strategy corresponding to the current monitoring task, such as the current border wide-area monitoring task, the preset strategy is the combination type of wide-area monitoring real-time response priority and two determined standardized decision identifiers. The rule database is quickly located and the perfectly matching decision fusion rules are selected through the dual index of perception strategy and combination type.

[0036] Based on the decision fusion rules, the standardized decision identifiers are processed to obtain preliminary fusion judgment values. These preliminary fusion judgment values ​​are then compared with a preset final decision threshold to generate a final judgment result indicating whether the target signal exists. Specifically, this involves: firstly, using an elliptical axis alignment transformation algorithm to correct the feature offset of the two standardized decision identifiers. Since the first judgment result comes from power spectral density fusion analysis, the spectral energy characteristics of the core focused signal are susceptible to energy fluctuations from sudden signals in a wide-area environment. The second judgment result comes from statistical analysis of the eigenvalues ​​of the received signal covariance matrix, the correlation characteristics of the core focused signal are susceptible to correlation interference from noise signals in a complex electromagnetic environment. This results in a shift in the feature dimension between the two judgment results. The elliptical axis alignment transformation algorithm dynamically adjusts the basic weights of both to align the two types of judgment features on the same analytical dimension.

[0037] The specific correction calculation method is as follows: First, obtain the basic adjustment coefficient for judging similar signals in the current monitoring scenario. The coefficient is determined by statistically analyzing 1000 accurate judgment cases of similar signals in the same monitoring scenario in the past, and dividing the number of cases judged accurately by power spectral density fusion analysis by the total number of accurate cases to obtain the basic adjustment coefficient. Then, multiply the value of the first standardized decision identifier by the basic adjustment coefficient to obtain the alignment adjustment value of the first identifier. At the same time, multiply the value of the second standardized decision identifier by (1 minus the basic adjustment coefficient) to obtain the alignment adjustment value of the second identifier. This step completes the alignment correction of the elliptical axis vectors of the two standardized identifiers, eliminating the judgment deviation caused by feature offset.

[0038] Next, the calculation is performed according to the selected fusion rule. If the selected rule is the simplified weighted operation and threshold comparison rule corresponding to the "existence + uncertainty" combination under the wide-area monitoring real-time response priority strategy, then the weight ratio is first determined according to the strategy. It is preset that the weight of the corresponding decision result under this strategy is higher than that of uncertainty, and the weight ratio of the two is fixed at 7:3, that is, the weight ratio of the first decision result is 0.7, the weight ratio of the second decision result is 0.3, and the sum of the two is equal to 1; then the first identifier value (0.65) after alignment adjustment is multiplied by its corresponding weight ratio (0.7) to obtain the first weighted value ( 0.65 multiplied by 0.7 results in 0.455; simultaneously, the aligned and adjusted second identifier value (0.175) is multiplied by its corresponding weight ratio (0.3) to obtain the second weighted value (0.175 multiplied by 0.3, resulting in 0.0525); finally, the first weighted value (0.455) and the second weighted value (0.0525) are added together to obtain the preliminary fusion judgment value (0.455 plus 0.0525, resulting in 0.5075); if the rule under the priority strategy for accurate judgment in complex electromagnetic environments is selected, the weight ratio will be adjusted to 6:4, and the first decision... The first result has a weight of 0.6, and the second result has a weight of 0.4. The calculation process is as follows: the aligned first identifier value (0.65) is multiplied by 0.6 to get 0.39, and the aligned second identifier value (0.175) is multiplied by 0.4 to get 0.07. The two are added together to get a preliminary fusion judgment value of 0.46, ensuring the accuracy of judgment in complex scenarios. Finally, the preliminary fusion judgment value is compared with the preset final judgment threshold. This final judgment threshold is determined based on the statistical data of judgments in historical no-signal scenarios, that is, the fusion judgment values ​​when there was indeed no target signal in the monitoring area in the past 500 times. The maximum value among them is taken as the threshold. For example, the maximum fusion judgment value when there is no signal in the past is 0.3, so the final judgment threshold is set to 0.3. If the preliminary fusion judgment value, such as 0.5075, is greater than the preset final judgment threshold of 0.3, it is determined that the target signal exists in the current monitoring area, and a final judgment result of the existence of the target signal is generated. If the preliminary fusion judgment value is less than the final judgment threshold, it is determined that the target signal does not exist in the current monitoring area, and a final judgment result of the non-existence of the target signal is generated. The whole process does not require additional calls to backend server resources, and all calculations are completed on the front-end edge node.

[0039] Data accuracy is ensured through extraction and format verification, avoiding fusion deviations caused by differences in the format of the original judgment results. Fusion rules are dynamically selected based on the perception strategy to adapt to the core needs of different application scenarios such as wide-area real-time monitoring and precise monitoring in complex environments, thereby improving the real-time response capability of spectrum sensing and meeting the needs for real-time performance and data transmission efficiency in wide-area, dynamic, and complex electromagnetic environments.

[0040] In a preferred embodiment of the present invention, if the final decision result is that the target signal exists, then standardized signal data to be analyzed is used for spatial spectrum analysis. By calculating the array covariance matrix and performing spatial spectrum estimation, the direction of arrival information of the radiation source is obtained, which may include: In this embodiment of the invention, if the final judgment result indicates the existence of the target signal, a call instruction for the standardized signal data to be analyzed is generated. Specifically, this includes: first, the front-end edge computing unit continuously receives and verifies the final judgment result obtained through fusion judgment. This result is a definite conclusion formed by fusing signal feature analysis and dual existence judgment in a wide-area monitoring scenario, providing a definite indication of the presence or absence of the target signal only within the monitoring area; when the front-end edge computing unit verifies and confirms that the final judgment result indicates the existence of the target signal, the call instruction generation process is immediately initiated, and this process must adapt to the real-time requirements of front-end data processing in a wide-area dynamic electromagnetic environment; next, the front-end edge computing unit will construct a complete call instruction around the standardized signal data to be analyzed. The instruction must contain three types of core identification information: first, the data storage path, determining the location of the standardized signal data to be analyzed in the front-end edge computing unit; second, the data storage path, determining the location of the standardized signal data to be analyzed in the front-end edge computing unit. The system employs several key parameters: First, it specifies the exact folder location, file name, and data block number within the node storage unit to ensure precise data location and prevent errors caused by scattered data storage. Second, it defines the data time range, accurately marking the start and end times of the target signal's capture by the sensor array to ensure only relevant and valid data is retrieved, eliminating interference from irrelevant time periods and reducing subsequent computational redundancy. Third, it prioritizes data retrieval, assigning the highest priority to this data retrieval, thus maximizing the use of processing resources from the front-end edge computing unit and preventing delays due to multi-task parallelism, ensuring smooth progress of subsequent direction-of-arrival (DOA) calculations. Finally, the generated retrieval instructions must strictly adhere to the data communication standards between front-end devices in wide-area monitoring scenarios, ensuring that the instruction format and encoding are consistent with the unit storing standardized signal data to be analyzed, preventing data loss or misreading due to format incompatibility during instruction transmission.

[0041] The process involves executing a call command and calculating the array covariance matrix corresponding to the sensor array based on standardized signal data to be analyzed. Specifically, the front-end edge computing unit first receives the call command and decomposes it according to the data parsing specifications for wide-area monitoring scenarios, extracting the storage path, time range, and call priority information. Then, based on the storage path, it locates the unit storing the standardized signal data to be analyzed. According to the time range, it selects standardized signal data collected by all receiving channels of the sensor array within that time period and pre-processed after calibration. This data has a unified format and standard, eliminating errors between different channels and fully covering the target signal information captured by each receiving channel at the same time point, ensuring data consistency and validity. Next, the array covariance matrix calculation process is initiated. The first step is to determine the total number of receiving channels in the sensor array, for example, the array contains N receiving channels, where N is the actual number of channels deployed in the sensor array, adapting to the coverage requirements of wide-area monitoring. The second step is to synchronously extract all signal data points for each receiving channel within the selected time range, ensuring that the number of data points for each channel is completely consistent and that each data point corresponds to a specific time point, avoiding calculation errors due to time asynchrony. The third step involves calculating the covariance between any two receiving channels, assuming they are channel X and channel Y, where X and Y are any integers from 1 to N, and X can be equal to Y. First, the first signal data point is taken from channel X, and then the first signal data point corresponding to the same time node is taken from channel Y. These two data points are multiplied to obtain the first product result. Then, the second, third, and so on, corresponding to the same time node, are sequentially taken from channels X and Y, up to the last signal data point, and each pairwise multiplied to obtain the product results for all time nodes. Finally, all these product results are summed. First, calculate a total sum. Then, divide this total sum by the total number of signal data points within the time range to obtain the covariance value between channel X and channel Y. Next, following the same calculation method, calculate the covariance values ​​for all channel combinations sequentially. First, using channel 1 as the reference, calculate the covariance values ​​between channel 1 and channel 1, channel 1 and channel 2, channel 1 and channel 3... channel 1 and channel N. Then, using channel 2 as the reference, calculate the covariance values ​​between channel 2 and channel 1, channel 2 and channel 2, channel 2 and channel 3... channel 2 and channel N. Continue this process until the covariance values ​​between channel N and all other channels are calculated.The fifth step involves arranging all calculated covariance values ​​in an ordered manner, with rows corresponding to the first channel and columns corresponding to the second channel. Specifically, the first row lists the covariance values ​​of channel 1 with each channel, the second row lists the covariance values ​​of channel 2 with each channel, and so on, until the Nth row lists the covariance values ​​of channel N with each channel. This results in a square matrix with both rows and columns equal to the total number of receiving channels, N. This matrix is ​​the array covariance matrix corresponding to the sensor array adapted for wide-area monitoring, thus yielding the final calculated covariance matrix.

[0042] Spatial spectrum analysis is performed on the covariance matrix calculation results to generate spatial spectrum estimation results characterizing the energy distribution of the signal in different directions. Specifically, after obtaining the array covariance matrix calculation results from the front-end edge computing unit, the spatial direction range for this spatial spectrum analysis is first determined based on the coverage requirements of wide-area monitoring. This range needs to cover all directions where radiation sources may appear within the monitoring area, typically a full directional range from 0 degrees to 360 degrees, ensuring that no potential signal source direction is missed. Subsequently, this full directional range is divided into multiple continuous and uniform preset directions. During the division, the interval between two adjacent preset directions must be sufficiently small, for example, 0.1 degrees, to ensure the accuracy of subsequent energy analysis and avoid errors due to spatial distribution. The excessive spacing caused the signal to miss the true radiation source direction. Next, for each preset spatial direction, the signal energy value was calculated one by one. The first step involved determining the array response vector corresponding to that preset direction based on the current preset spatial direction and the actual physical structure of the sensor array, including the antenna placement, the overall shape of the array, and the spacing between antennas. These parameters are fixed parameters during sensor array deployment and are adapted to the signal reception requirements of wide-area monitoring. The number of elements in this vector is consistent with the total number of receiving channels in the sensor array, and each element corresponds to the signal response characteristics of a receiving channel in that preset direction. The second step involved performing a conjugate transpose on the array response vector, first transposing each element of the vector... The first step involves performing a conjugate transformation, where the real part remains unchanged while the imaginary part is inverted. If the element is a real number, the conjugate transformation remains unchanged. The transformed vector is then adjusted to row vector form (if the original vector was a column vector) or column vector form (if the original vector was a row vector), forming the conjugate transpose vector. The second step multiplies the processed conjugate transpose vector with the calculated covariance matrix. Each element of the conjugate transpose vector is multiplied by the corresponding element of the covariance matrix, and the results are summed to obtain an intermediate result. The third step multiplies this intermediate result with the original array response vector (the vector without conjugate transpose processing). Each element of the intermediate result is multiplied by the corresponding element of the covariance matrix. The first step involves multiplying each element by the corresponding element of the original array response vector, and then summing all the multiplication results to obtain the signal energy value corresponding to the preset spatial direction. The second step is to calculate the signal energy value for each preset spatial direction one by one according to the same calculation logic, ensuring that each preset direction can obtain the corresponding energy value. Finally, the angle value of each preset spatial direction is correlated with its corresponding signal energy value to form a complete dataset containing the direction angle and signal energy. This dataset is the spatial spectrum estimation result, which can intuitively and clearly characterize the energy distribution of the target signal in different spatial directions in the monitoring area and accurately reflect the strength differences of the signal energy in each direction.

[0043] Based on the spatial spectrum estimation results, peak search and angle calibration processes are used to determine the specific direction-of-arrival (DOA) angle value corresponding to the target radiation source, which serves as the final DOA information. Specifically, the front-end edge computing unit first initiates peak search processing on the generated spatial spectrum estimation results. The first step involves determining a reasonable energy intensity threshold value based on the background noise characteristics of a wide-area monitoring scenario. This is achieved by first statistically analyzing the average background noise energy when there is no target signal within the monitoring area, and then adding a certain percentage margin based on the typical energy intensity range of the target signal. This percentage is set according to the noise fluctuations in the actual monitoring environment to ensure effective differentiation between the target signal and background noise, ultimately determining the energy intensity threshold value. The second step involves processing the peak search results using a pre-set spatial spectrum estimation method. The process involves sequentially checking the signal energy values ​​corresponding to each preset direction in the spatial spectrum estimation results, identifying all signal energy values ​​with energy intensities greater than a set threshold as peak points, and meticulously recording the angle values ​​of the preset spatial directions corresponding to each peak point. Directions with energy intensities less than or equal to the threshold are excluded to avoid background noise interference. Next, angle calibration is performed. Firstly, based on the fixed physical parameters of the sensor array, including antenna spacing, array arrangement, and antenna pointing angle, a correspondence table between the preset spatial directions and actual spatial angles of the sensor array is established through actual site testing. During testing, a standard signal is transmitted at a fixed position with a known actual angle, and the time it takes for the sensor array to capture this signal is recorded. The first step involves repeatedly testing the corresponding preset spatial directions to form a stable and accurate correspondence table, ensuring the mapping accuracy between the preset directions and the actual angles. The second step involves comparing each preset spatial direction angle value corresponding to all previously recorded peak points with the corresponding correspondence table, converting each preset direction angle value into its corresponding actual spatial angle value, and collecting all converted actual spatial angle values ​​to form a candidate direction of arrival set. The candidate direction of arrival set is then filtered and refined. The first step involves retrieving the previously generated signal basic feature distribution pattern, including core features such as signal time-frequency clustering, bandwidth occupancy, and duration. These features are inherent attributes of the target signal derived from the analysis of the original spectrum signal. The second step involves targeting the candidate direction of arrival set... The process involves four steps: First, for each actual spatial angle value in the set of directions of arrival, extract the signal features corresponding to that angle value and obtain the signal energy change and duration characteristics in that direction from the spatial spectrum estimation results. Second, compare and match the extracted signal features with the basic characteristic distribution patterns that the target radiation source should possess. If the signal features corresponding to a certain actual angle value completely match the characteristics of the target radiation source, such as consistent time-frequency concentration, matching bandwidth occupancy, and consistent time duration, then retain that angle value. If there are significant differences in the features, such as excessively low time-frequency concentration or mismatched bandwidth occupancy, then discard that angle value to avoid interference from false peak points. Third, after matching and filtering, retain all valid candidate directions that match the characteristics of the target radiation source.Finally, the valid candidate directions are confirmed. If there is only one valid candidate direction, the actual spatial angle value corresponding to that direction is directly determined as the direction of arrival angle of the target radiation source. If there are multiple valid candidate directions, the direction with the highest energy intensity and the most accurate feature matching is selected based on the signal energy intensity ranking. This actual spatial angle value is then determined as the direction of arrival information of the target radiation source, thus completing the precise location of the direction of arrival.

[0044] The entire calculation of direction of arrival information is completed in the front-end edge computing unit, eliminating the need to send massive amounts of raw signal data or intermediate calculation data back to the back-end server. This reduces the amount of data transmitted through the backhaul link in wide-area monitoring scenarios. The front end independently completes all core processes such as data retrieval, matrix calculation, spatial spectrum analysis, and angle calibration locally, reducing the computational load on the back-end server and improving the overall operating efficiency of the spectrum sensing system.

[0045] In a preferred embodiment of the present invention, the direction of arrival information, the final decision result, and the compressed signal feature data extracted from the standardized signal data to be analyzed are transmitted back to the backend server to realize real-time perception and intelligent analysis of the spectrum environment, which may include: In this embodiment of the invention, the final judgment result, direction of arrival information, and extracted and compressed signal feature data are integrated to form a data set to be uploaded. Specifically, this includes: firstly, ensuring that the integrated core data accurately matches the actual needs of the wide-area spectrum monitoring scenario, and strictly locking in three types of key information: firstly, the final judgment result of whether the target signal exists, obtained after multi-dimensional judgment fusion; secondly, the direction of arrival information of the radiation source determined through spatial spectrum estimation, peak screening, and feature matching, which must be presented as a two-dimensional angle value of azimuth and elevation, both accurate to two decimal places, to ensure the accuracy of direction positioning; and thirdly, the signal feature data extracted from the standardized signal data to be analyzed and processed by lossless compression, whereby the extracted features must cover key signal identification dimensions, including time-frequency clustering and bandwidth occupancy. The compression process, which considers time duration and modulation characteristics, removes redundant bytes without altering core feature values. Each type of compressed feature data is recorded in the format of feature name-feature quantization value-data precision. A multi-level data format verification process is then initiated. The first level verifies data integrity, confirming that none of the three types of data are missing. The second level verifies data standardization, requiring the final judgment to strictly match the text format indicating presence or absence, disallowing expressions such as "may exist" or "suspected non-existence." The direction-of-arrival (DOA) angle value must be within a preset range; data exceeding this range is considered invalid and must be retrieved again from the front end. The compressed signal feature data must ensure that each feature name corresponds one-to-one with its quantization value and data precision, without formatting errors. The third level verifies data consistency, checking the logical match between the DOA information and the final judgment result.

[0046] Subsequently, unique and cross-scenario traceable association identifiers are generated for the three types of data. The generation logic of the association identifiers needs to be adapted to multi-node deployment scenarios. The specific rules are: region code + sensor node number + collection timestamp + data check bit. The region code is preset according to the deployment scenario; the sensor node number is a unique sequence code within the region; the collection timestamp adopts the complete format of year-month-day-hour-minute-second-millisecond to ensure the distinction between different millisecond-level collection data from the same node; the data check bit is obtained by adding the preset value corresponding to the region code, the node number value, and the time unit values ​​in the timestamp in sequence, and taking the last two digits of the sum as the check bit. The final association identifier format is region code-node number-timestamp-check bit, ensuring that each piece of data can be traced back to a specific region, specific node, and specific collection time, avoiding data confusion between multiple regions and multiple nodes; finally, the data is spliced ​​and integrated. A hierarchical delimiter mechanism is adopted to avoid data confusion. The first-level delimiter, such as "&&", is used to separate the three types of core data. The second-level delimiter, such as "||", is used to separate each type of data from its associated identifier. The third-level delimiter, such as ",", is used to separate different fields in the feature data. The splicing order strictly follows the final judgment result || associated identifier && direction of arrival information || associated identifier && compressed signal feature data || associated identifier. The compressed signal feature data is arranged in a fixed order: time-frequency aggregation degree - feature quantization value - data precision, bandwidth occupancy rate - feature quantization value - data precision, time persistence - feature quantization value - data precision, and modulation mode feature - feature quantization value - data precision. After splicing, the data structure is verified again to ensure that the delimiters at each level are used correctly, without omissions or redundancies, and finally form a rigorous, traceable, and unconfused data set to be uploaded.

[0047] Based on the data set to be uploaded, the data is encapsulated according to a preset communication protocol to generate a transmission data packet. Specifically, this includes: first, determining the design logic of the preset communication protocol to adapt to scenarios with limited bandwidth and unstable signals in border communication links. The protocol must balance data transmission efficiency and integrity, and stipulate that the transmission data packet consists of three parts: a data header, a data body, and a checksum. The field definitions, data types, and arrangement order of each part are preset fixed values ​​to ensure consistency between front-end sending and back-end parsing. The first step is to finely construct the data header, which consists of 6 fields. Field 1 is the regional code identifier. The regional code in the identifier associated with the data set to be uploaded is converted into a preset binary value, and the node number is converted... The data is converted to binary; field 3 is the compressed data collection timestamp value, which is compressed according to preset rules in the format of year-month-day-hour-minute-second-millisecond, and then converted to binary; field 4 is the total data length, which is the total number of bytes in the data set to be uploaded and converted to binary; field 5 is the data type identifier, which assigns a fixed binary identifier to the three types of core data and combines them in the order of judgment result-direction of arrival-feature data; field 6 is a reserved field, filled with 00000000 for future function expansion; the binary data of the 6 fields are concatenated in the order of field 1-field 2-field 3-field 4-field 5-field 6 to ensure that the total length is 20 bytes, forming a standard data header.

[0048] The second step is to construct the data body, directly incorporating the verified data set to be uploaded completely. The length of the data body strictly matches the value of the total data length field in the data header. The data arrangement order and delimiters used in the data body are completely consistent with the data set to be uploaded, without any modifications, ensuring that the backend can accurately separate various types of data according to preset rules. If the total number of bytes in the data set to be uploaded exceeds the single transmission limit of the communication link, the data body is split into multiple sub-data bodies. The length of each sub-data body does not exceed the transmission limit, and a sub-data body number minus the total number of sub-data bodies is added to the end of each sub-data body to ensure that the backend can completely reassemble it. The third step is to calculate the checksum with high precision. First, all data in the data header and data body are numerically converted. The conversion rules are completely fixed and consistent between the front end and the back end: text is converted to the value 1 if it exists, to the value 0 if it does not exist, and to the value 0 if it is not present. Angle values, feature quantization values, and data precision are directly retained as original values. Region codes and node numbers are converted according to preset values ​​in the association identifier. Timestamp compression values ​​are directly taken according to the values ​​of each time unit. Delimiters at all levels are converted according to preset values. The data is first converted to decimal values. Then, all the converted values ​​are added sequentially in the order of data header first, data body last. During the accumulation process, the intermediate result is recorded every 10 values ​​to avoid calculation errors due to too many values. Finally, the total sum of all values ​​is obtained. Then, a preset check radix of 256 is selected. The total sum is divided by 256, and all remainder digits are retained during the calculation. The remainder is then converted into a 4-byte binary value as the final check code, which is used by the backend to accurately verify whether the data transmission has been lost or tampered with. Finally, the data packets are assembled. If the data body is not split, it is directly concatenated in the order of data header (20 bytes) - data body (variable bytes) - check code (4 bytes), ensuring that no bytes are lost or redundant during the concatenation process. If the data body has been split, each sub-data body is equipped with the same data header and an independent check code. Then, each sub-data packet is assembled in the order of sub-data body number - data header - sub-data body - check code. Finally, a transmission data packet that conforms to the preset communication protocol and is adapted to low-bandwidth and unstable links is formed, ensuring that it can be accurately identified by the backend in the complex communication environment of the border.

[0049] Data packets are transmitted to the backend server via the communication network. The backend server parses and reassembles the received data packets to restore the final decision result, direction of arrival information, and compressed signal characteristic data. Specifically, this includes: first, executing the data packet transmission operation. The front-end spectrum sensor node selects the corresponding communication mode to start the transmission process based on the communication link type of the deployment scenario. Before transmission, the front-end node first obtains the real-time status parameters of the current communication link, including link bandwidth, signal strength, and packet loss rate. It determines the initial transmission rate through a built-in link adaptation algorithm. For example, when the link bandwidth is 1Mbps, the signal strength is ≥-70dBm, and the packet loss rate is ≤3%, the initial transmission rate is... The initial transmission rate is set to 500Kbps. With a link bandwidth of 500Kbps, signal strength of -80dBm to -70dBm, and a packet loss rate of 3%-5%, the initial transmission rate is set to 200Kbps to ensure the transmission rate matches the link's carrying capacity. During transmission, the front-end node's transmission status monitoring mode monitors the link status every 100 milliseconds. If a drop in link bandwidth exceeding 30% is detected (e.g., from 1Mbps to below 700Kbps), signal strength below -80dBm, or packet loss rate above 5%), the rate adjustment mechanism is immediately activated, reducing the current transmission rate by 50% (e.g., from 500Kbps to 250Kbps), while simultaneously extending the interval between data packet transmissions. The interval can be extended from sending one packet every 10 milliseconds to sending one packet every 20 milliseconds to avoid exacerbating link congestion. If no acknowledgment signal is received from the backend server after a data packet is sent, the acknowledgment signal must be returned within 500 milliseconds. If the timeout occurs, it is considered unacknowledged, and a single retransmission is initiated. If no acknowledgment signal is received after the single retransmission, a second retransmission is performed after a 300-millisecond interval. If the second retransmission fails, a third retransmission is performed after a 500-millisecond interval. If all three retransmissions fail, the associated identifier, sending time, link status, and other information of the data packet are recorded, a transmission error log is generated and uploaded to the regional control center in real time, and the system switches to a backup communication link to attempt retransmission. After receiving the data packet, the end server immediately initiates the parsing and reassembly process. The first step is to extract the checksum. According to the preset communication protocol, the checksum is located in the last 4 bytes of the data packet. This part of the data is directly truncated and converted into a decimal value. The second step is to reconstruct the checksum. Following the exact same process of numerical conversion, accumulation, and modulo, the checksum in the data header and data body of the data packet is recalculated. It is ensured that the conversion rules, accumulation order, and checksum base are consistent with the front end. For example, if the front end converts a value to 1, the back end must also maintain consistency and must not convert it to other values. The third step is to compare the checksum. The extracted checksum is compared bit by bit with the reconstructed checksum. If they match, the data packet is determined to be transmitted completely without errors.If there is a discrepancy, it is determined that the data is lost or tampered with. The backend server immediately generates a retransmission request containing the data packet association identifier, sub-data body number, and error type, and sends it to the corresponding front-end sensor node through the original communication link. After receiving the request, the front-end node quickly locates the target data packet based on the association identifier and sub-data body number, and retransmits it at the adjusted sending rate until the backend receives and verifies it.

[0050] After confirming the data packet's integrity, the backend server initiates data splitting and reassembly. If the data packet is complete, the header is first truncated to 20 bytes. The six fields of the header are then parsed one by one: the binary region code identifier is converted to the corresponding region code, the node number identifier is converted to the node number, the timestamp compressed value is restored to a complete timestamp, and the total data length is converted to a decimal value. Next, based on the total data length in the header, data of the corresponding length is truncated from the header to form the data body. The data body is then split according to the delimiter rules to separate the final decision result, direction of arrival information, and compressed signal characteristic data. Simultaneously, the total length of the split data is verified against the total length of the data in the header. If the lengths are inconsistent, it is considered an abnormal split, and parsing is restarted. Finally, the three types of split data are bound to the region code, node number, and timestamp in the data header using the association identifier. If it is a split sub-data packet, it is first sorted by sub-data body number, and then the data header and sub-data body of each sub-data packet are parsed one by one. The sub-data bodies are then concatenated in sequence according to the sub-data body number to form a complete data body. The subsequent splitting process is the same as that of the complete data packet. After concatenation, the length of the complete data body is checked to see if it is consistent with the total length of the data in the data header. If they are consistent, the splitting continues. If they are inconsistent, the missing sub-data packets are retrieved again to ensure that the finally reconstructed data is complete and without omissions.

[0051] The backend server utilizes the reconstructed data to perform data fusion and correlation analysis, enabling real-time perception and intelligent situational analysis of spectrum occupancy status and signal source direction throughout the entire monitoring area. Specifically, this includes: firstly, data preprocessing and partitioning / classification. After receiving the reconstructed data from all front-end sensor nodes, the backend server first removes invalid data, then classifies it according to three dimensions: geographical partition, time granularity, and scene type. Geographical partitioning is based on the actual control requirements of the deployment scenario; border scenarios are partitioned according to border line segments and terrain types, with each partition covering a 5-kilometer range. The time granularity is set to 1 minute according to the real-time requirements of spectrum monitoring, meaning each minute is an analysis period to ensure rapid capture of dynamic signal changes. Scene types are classified as fixed monitoring and mobile monitoring to avoid cross-interference between data from different scenarios. Next, multi-dimensional data fusion processing is performed, including radiation source direction fusion. For all reconstructed data within the same geographical partition, time granularity, and scene type, the same radiation source is first matched using modulation mode features and time-frequency aggregation features in the compressed signal feature data; that is, features with a difference in quantization values ​​within ±0.05 are considered the same radiation source. Then, all waves corresponding to that radiation source are collected. The azimuth angle values ​​are calculated; the average azimuth angle is calculated by adding all azimuth angle values ​​and dividing by the number of azimuth angles. For example, if there are three azimuth angles of 45.32°, 45.40°, and 45.36°, the average value is (45.32 + 45.40 + 45.36) ÷ 3 = 45.36°. An outlier threshold is set at ±5° of the average value. Each azimuth angle is compared to the average value; values ​​exceeding the threshold are considered outlier. For example, if the average value is 45.36° and a certain azimuth angle is 51.00°, the difference is 5.64°, which is considered outlier and removed. The average of the remaining valid azimuth angles is then calculated again. The value is used as the final azimuth angle of the radiation source; the elevation angle is processed in the same way to finally form the accurate direction of arrival information of the radiation source in this region, time period, and scene; signal feature fusion is performed on compressed signal feature data under the same geographical region, the same time granularity, and the same scene type, and the data is classified and statistically analyzed according to feature type. The weighted average of the effective values ​​is calculated by time-frequency clustering, and weights are assigned according to the signal reception strength of the sensor nodes. The weight of signal strength ≥-80dBm is 0.8, the weight of -90dBm~-80dBm is 0.5, and the weight of ≤-90dBm is 0.2. Weighted average = (Eigenvalue 1 × Weight 1 + Eigenvalue 2 × Weight 2 + ... + Eigenvalue n × Weight n) ÷ (Weight 1 + Weight 2 + ... + Weight n), reflecting the energy concentration level of the signal in this area; Band occupancy is calculated based on the maximum, minimum, and median values, presenting the fluctuation range of band occupancy; Time duration is categorized by continuous duration ≥30 seconds, 10-30 seconds, and ≤10 seconds to determine the signal stability; Modulation mode characteristics are determined by the frequency of occurrence of the same feature quantization value to identify the dominant signal modulation type in this area; Spectral activity is determined and fused. For the final decision results within the same geographical region, time granularity, and scenario type, a majority voting + abnormal node exclusion rule is adopted. The number of decisions regarding presence and absence in the data set is counted, while node data with signal received strength ≤ -100dBm is excluded. If the percentage of occurrences of occurrences in the remaining data is ≥60%, then spectral activity is determined to exist in that region, time period, and scenario; if the percentage of occurrences of absences is ≥60%, then no significant spectral activity is determined; if the percentages are both between 40% and 60%, then the spectral activity status is finally determined by combining the signal feature fusion results.

[0052] Then, deep correlation analysis is performed, including historical data correlation, comparing the current fusion result with historical data stored on the backend server. Comparison dimensions include the same partition, same time period, different dates, same partition, different time periods, same date, different partitions, same time period, and same date. This analyzes the differences between the current spectrum activity status and historical status, the changing trends of signal characteristics, and the movement trajectory of the radiation source. Geographic information correlation involves overlaying the fused radiation source direction-of-arrival information with an electronic map of the monitoring area. Using the sensor node deployment coordinates as a reference point, combined with the radiation source's azimuth and elevation angles, the approximate geographical range of the radiation source is calculated. Simultaneously, the control information for this area is correlated to determine whether the radiation source is located in a sensitive control area. Multi-source data correlation involves linking the spectrum data with other relevant data. If spectrum activity is determined to exist and the radiation source is located in a sensitive area, this correlation is applied. The system simultaneously retrieves video data or ship positioning data for the region to assist in determining the carrier type of the radiation source and improve the comprehensiveness of the situation analysis. Finally, it generates a real-time situation report and application output. Based on the data fusion and correlation analysis results, the backend server generates a real-time spectrum environment situation report in a structure of region-time period-spectrum status-radiation source information-risk level. The report identifies the spectrum occupancy status, dominant signal characteristics, direction of arrival and approximate geographical range of the radiation source, signal activity trajectory and trend, and risk level for each geographical region. At the same time, the situation report is pushed to the regional control center and the front-end sensor node control platform in real time, enabling managers to intuitively grasp the dynamic spectrum environment of the monitored area, providing accurate basis for rapid response, and realizing real-time spectrum perception and intelligent analysis in a wide-area, dynamic, and complex electromagnetic environment.

[0053] By standardizing data formats, splitting sub-data packets, and adapting transmission links, the amount of data transmitted and transmission conflicts are reduced, thereby alleviating the transmission pressure on low-bandwidth and unstable links at the border and avoiding delays caused by excessive data volume. At the same time, by using mechanisms such as checksum verification, multiple rounds of retransmission, and backup link switching, the integrity and stability of data transmission can be ensured even in scenarios with weak communication signals and easily interrupted links.

[0054] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0055] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0056] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A real-time spectrum sensing system enabled by edge computing, characterized in that, include: The storage module is used to receive and store raw spectrum signal data from edge computing nodes at the front end of spectrum sensors equipped with array antennas deployed in the monitoring network; The generation module is used to perform preliminary processing and analysis on the raw spectrum signal data. By extracting the main energy regions of the signal in the time-frequency distribution, it determines the basic characteristic distribution pattern of the signal and generates signal analysis and correction parameters. The analysis module is used to calibrate and preprocess the original spectrum signal data using signal analysis correction parameters to obtain standardized signal data to be analyzed. Based on the standardized signal data to be analyzed, power spectral density fusion analysis and eigenvalue statistical analysis of the received signal covariance matrix are performed to generate the first signal existence decision result and the second signal existence decision result. The fusion module is used to fuse the existence decision results of the first signal and the existence decision results of the second signal to generate a final decision result on whether the target signal exists. The acquisition module is used to call the standardized signal data to be analyzed for spatial spectrum analysis if the final decision result is that the target signal exists. By calculating the array covariance matrix and performing spatial spectrum estimation, the direction of arrival information of the radiation source is obtained. The backhaul module is used to transmit the direction of arrival information, the final decision result, and the compressed signal feature data extracted from the standardized signal data to be analyzed back to the backend server, so as to realize real-time perception and intelligent analysis of the spectrum environment.

2. The edge computing-enabled real-time spectrum sensing system according to claim 1, characterized in that, The raw spectral signal data undergoes preliminary processing and analysis. By extracting the main energy regions in the time-frequency distribution of the signal, the basic characteristic distribution pattern of the signal is determined, and signal analysis and correction parameters are generated, including: Perform time-frequency transformation processing on the original spectrum signal data to generate a time-frequency matrix describing the two-dimensional distribution of signal energy with time and frequency; The energy distribution of the time-frequency matrix is ​​analyzed, the energy segmentation threshold is calculated, and the signal is segmented based on the energy segmentation threshold to extract the main energy regions in the time-frequency distribution. The geometric and statistical characteristics of the main energy regions are analyzed and calculated to determine the basic characteristic distribution pattern of the signal, including the signal’s time-frequency concentration, bandwidth occupancy, and time duration characteristics. Based on the basic feature distribution pattern, the time-frequency clustering, bandwidth occupancy and time duration features are comprehensively parameterized and transformed to generate signal analysis correction parameters for signal feature extraction and correction.

3. The edge computing-enabled real-time spectrum sensing system according to claim 2, characterized in that, The signal analysis and correction parameters include at least the time window correction coefficient, the frequency resolution adjustment parameter, and the energy normalization reference.

4. The edge computing-enabled real-time spectrum sensing system according to claim 3, characterized in that, Using signal analysis and correction parameters, the original spectral signal data is calibrated and preprocessed to obtain standardized signal data to be analyzed. Based on the standardized signal data, power spectral density fusion analysis and eigenvalue statistical analysis of the received signal covariance matrix are performed to generate a first signal existence decision result and a second signal existence decision result, including: Based on signal analysis and correction parameters, the original spectrum signal data is calibrated and preprocessed to obtain standardized signal data to be analyzed. Power spectral density is estimated for standardized signal data to be analyzed, the initial spectral energy distribution of each node or time period is obtained, and the initial spectral energy distribution is fused to generate a comprehensive spectral energy distribution; The comprehensive spectrum energy distribution is analyzed, the extracted energy features are compared with the preset first decision threshold, and the existence of the first signal is determined and generated based on the comparison result. Using the signal data to be analyzed, a received signal covariance matrix is ​​constructed, and eigenvalue decomposition is performed on the received signal covariance matrix to obtain an eigenvalue sequence. Statistical analysis is performed on the eigenvalue sequence to calculate the eigenvalue statistics. The eigenvalue statistics are compared with a preset second decision threshold. Based on the comparison result, the existence of the second signal is determined and generated.

5. The edge computing-enabled real-time spectrum sensing system according to claim 4, characterized in that, The existence determination results of the first signal and the second signal are fused together to generate a final determination result regarding the existence of the target signal, including: The existence determination results of the first signal and the existence determination results of the second signal are uniformly converted into standardized determination identifiers that conform to the preset data format; Based on the judgment type represented by the standardized judgment identifier and the preset perception strategy, the corresponding judgment fusion rule is selected from the predefined rule set; Based on the decision fusion rules, the standardized decision identifier is processed to obtain a preliminary fusion decision value. The preliminary fusion decision value is then compared with the preset final decision threshold to generate a final decision result on whether the target signal exists.

6. The edge computing-enabled real-time spectrum sensing system according to claim 5, characterized in that, If the final determination is that the target signal exists, then standardized signal data to be analyzed is used for spatial spectrum analysis. By calculating the array covariance matrix and performing spatial spectrum estimation, the direction of arrival information of the radiation source is obtained, including: If the final judgment indicates that the target signal exists, a call instruction for the standardized signal data to be analyzed is generated. Execute the call command, calculate the array covariance matrix corresponding to the sensor array based on the standardized signal data to be analyzed, and obtain the covariance matrix calculation result; Spatial spectrum analysis is performed on the covariance matrix calculation results to generate spatial spectrum estimation results that characterize the energy distribution of the signal in different directions; Based on the spatial spectrum estimation results, the specific direction-of-arrival angle value corresponding to the target radiation source is determined through peak search and angle calibration, which serves as the final direction-of-arrival information.

7. The edge computing-enabled real-time spectrum sensing system according to claim 6, characterized in that, Based on the spatial spectrum estimation results, peak search and angle calibration processes are used to determine the specific direction-of-arrival (DOA) angle value corresponding to the target radiation source, which serves as the final DOA information, including: Peak search processing is performed on the spatial spectrum estimation results to identify peak points with intensity greater than a preset threshold; The peak points are converted into corresponding candidate directions of arrival based on the preset correspondence between the direction of arrival and spatial angle of the array antenna, forming a set of candidate directions; For the candidate direction set, a matching screening is performed based on the distribution pattern of the basic characteristics of the signal to select the effective candidate direction that matches the characteristics of the target radiation source; The specific angle values ​​corresponding to the valid candidate directions are determined as the final direction of arrival information.

8. The edge computing-enabled real-time spectrum sensing system according to claim 7, characterized in that, The direction-of-arrival (DOA) information, the final decision result, and the compressed signal feature data extracted from the standardized signal data to be analyzed are transmitted back to the backend server to achieve real-time perception and intelligent analysis of the spectrum environment, including: The final judgment result, direction of arrival information, and extracted and compressed signal feature data are integrated to form a data set to be uploaded; Based on the data set to be uploaded, it is encapsulated according to a preset communication protocol to generate a transmission data packet; The data packets are transmitted to the backend server through the communication network. The backend server parses and reassembles the received data packets to restore the final judgment result, direction of arrival information and compressed signal characteristic data. The backend server uses the restored data to perform data fusion and correlation analysis, enabling real-time perception and intelligent situational analysis of the spectrum occupancy status and signal source direction throughout the entire monitoring area.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the system as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the system as described in any one of claims 1 to 8.

Images