A radio spectrum measurement system based on mobile crowdsourcing

Abstract

The present invention relates to the field of network monitoring technology, and specifically to a radio spectrum measurement system based on mobile crowdsourcing. The system includes: a crowdsourcing acquisition module, a timing processing module, a spatial optimization module, and a dynamic fusion module. In the present invention, by periodic scanning combined with a multi-source heterogeneous data encapsulation protocol, timestamps, geographic coordinates, signal-to-noise ratios, and device voltages are integrated to achieve standardized acquisition and dynamic association of multidimensional data. A time series alignment algorithm calculates timing weight parameters, enhances the distinction between high- and low-frequency events, and suppresses temporal noise. A DBSCAN algorithm calculates spatial weight coefficients, linear weighting is used to generate comprehensive credibility, improves spatiotemporal coupling accuracy, evidence theory ranks credibility and analyzes its correlation with signal-to-noise ratios, and a Kalman filter compensates for voltage fluctuations. This achieves multi-source data fusion and error suppression, enhances the anti-interference capability of spectrum features, reduces positioning deviations, and improves spectrum allocation efficiency and monitoring system stability.

Description

Technical Field

[0001] The present invention relates to the technical field of network monitoring, and in particular to a radio spectrum measurement system based on mobile crowdsourcing. Background Art

[0002] The technical field of radio spectrum measurement systems encompasses radio spectrum measurement, signal capture and identification, and spectrum resource management for network monitoring. The core of this technical field is the comprehensive detection and analysis of radio spectrum occupancy, signal characteristics, frequency distribution, and transmission power through various wireless measurement devices and techniques, thereby understanding spectrum usage and dynamic changes. Radio spectrum measurement technology is widely used in scenarios such as wireless communication network security monitoring, interference source location, spectrum planning, and radio regulation. The technical system in this field typically covers multiple aspects, including RF signal acquisition, spectrum analysis, signal identification, data recording, and visualization, forming a complete technical chain from signal acquisition to monitoring and analysis.

[0003] Among them, the radio spectrum measurement system refers to a dedicated system for capturing, analyzing, and storing radio signals within a specific spectrum range. The technical matters targeted by this patent subject cover real-time acquisition of radio signals, spectrum energy detection of radio frequency signals, time-frequency characteristic analysis of signals, and local storage and remote transmission of measurement data. Specifically, the system receives wireless signals via an antenna, performs signal conditioning via a radio frequency front-end circuit, converts the analog signal into a digital signal using a high-speed analog-to-digital conversion circuit, and then uses a spectrum analysis method to extract spectrum characteristics from the digital signal. Finally, the analysis results are stored or transmitted via a data interface.

[0004] Existing technologies rely on single-point data collection using fixed equipment, resulting in low data coverage density and poor scalability, making them difficult to adapt to the needs of highly dynamic spectrum monitoring in complex environments. Temporal dimension processing uses static timestamp recording and lacks a dynamic weight allocation mechanism, leading to confusion between the priorities of burst signals and regular signals, affecting the timeliness of interference event identification. Spatial dimension processing relies on manual partitioning or simple distance threshold division, failing to consider the spatial continuity of signal propagation and prone to positioning errors in reflection or multipath effect scenarios. Data fusion uses a single model or linear weighting, failing to address the heterogeneity and contradictions of multi-source data such as signal-to-noise ratio and device voltage, resulting in a decrease in the credibility of feature vectors. Data encapsulation protocols lack standardized design, resulting in poor compatibility between different device data formats, increasing cross-platform integration costs and hindering the efficiency of large-scale monitoring network deployment. Summary of the Invention

[0005] The purpose of the present invention is to solve the problem that the existing technology relies on single-point collection of fixed equipment, has low data coverage density and poor scalability, and is difficult to adapt to the high-dynamic spectrum monitoring needs of complex environments. The time dimension processing uses static timestamp recording and lacks a dynamic weight allocation mechanism, which leads to confusion between the priority of burst signals and regular signals, affecting the timeliness of interference event identification. The spatial dimension relies on artificial partitioning or simple distance threshold division, and does not consider the spatial continuity of signal propagation, which is prone to positioning deviation in reflection or multipath effect scenarios. The data fusion link uses a single model or linear weighting, which does not solve the heterogeneity and contradiction of multi-source data such as signal-to-noise ratio and device voltage, resulting in a decrease in the credibility of the feature vector. The data encapsulation protocol lacks a standardized design, and the compatibility of data formats of different devices is poor, which increases the cost of cross-platform integration and restricts the efficiency of large-scale monitoring network deployment. Therefore, a radio spectrum measurement system based on mobile crowdsourcing is proposed.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A radio spectrum measurement system based on mobile crowdsourcing includes:

[0007] The crowdsourcing acquisition module is used to periodically scan radio spectrum signals through terminal devices to obtain timestamps, geographic coordinates, signal-to-noise ratios, and device operating voltages, perform structured encapsulation using a multi-source heterogeneous data encapsulation protocol, transmit the timestamps and geographic coordinates to the timing processing module, and output the signal-to-noise ratios and device operating voltages to the dynamic fusion module;

[0008] A time series processing module is used to call a time series alignment algorithm to perform absolute difference calculation on the timestamps, generate time series weight parameters in ascending order, and pass the time series weight parameters and the geographic coordinates to a spatial optimization module;

[0009] The spatial optimization module is used to calculate the Euclidean distance of the target area through the geographic coordinates using the DBSCAN clustering algorithm, generate a normalized spatial weight coefficient, determine the weighted ratio of the temporal weight parameter and the spatial weight coefficient based on the historical data training regression model to generate a comprehensive credibility index, and output the comprehensive credibility index to the dynamic fusion module.

[0010] As a further solution of the present invention, the multi-source heterogeneous data encapsulation protocol includes field mapping rules and a CRC check mechanism;

[0011] The time series alignment algorithm uses a dynamic time warping algorithm to achieve multi-device time axis calibration;

[0012] Before calculating the Euclidean distance, the geographic coordinates are converted into the UTM coordinate system to achieve dimension unification.

[0013] As a further solution of the present invention, the crowdsourcing collection module includes:

[0014] The spectrum scanning submodule periodically scans radio spectrum signals through the terminal device, detects the carrier frequency and signal strength, records the timestamp of the scanning moment, synchronously obtains the geographic coordinates output by the device's built-in positioning system, and converts the electromagnetic wave signal strength continuously collected during the scanning period into a spectrum scanning value;

[0015] The voltage monitoring submodule uses the raw voltage data output by the device power monitoring circuit, calculates the standard deviation of the voltage sampling sequence using a sliding window algorithm, eliminates transient interference pulses, and generates the device operating voltage.

[0016] The signal analysis submodule separates the energy distribution of the signal frequency bands through fast Fourier transform based on the spectrum scan value, calculates the ratio of the main frequency band power to the noise frequency band power, generates a signal-to-noise ratio, performs piecewise linear interpolation operation on the signal-to-noise ratio and the device operating voltage based on the device power consumption curve. The device power consumption curve is constructed based on the voltage-power consumption correspondence table provided by the device manufacturer, and outputs a signal quality coefficient;

[0017] The data encapsulation submodule calls the timestamp and the geographic coordinates, constructs a time series data frame according to a multi-source heterogeneous protocol, performs structured encoding on the signal quality coefficient and the device operating voltage according to a dynamic fusion protocol, generates a time series transmission frame, and outputs a fusion data packet.

[0018] As a further solution of the present invention, the electromagnetic wave signal strength conversion formula is:

[0019] ;

[0020] in, Represents the spectrum scan value, is the signal power spectral density at frequency f, and are the lower and upper limits of the scanning frequency band, respectively. df is a small frequency increment, which comes from the integral operation and is a standard notation in calculus.

[0021] The sliding window length L is based on the device sampling rate according to Dynamic adjustment, where Indicates the device sampling rate, is the ceiling operator.

[0022] As a further solution of the present invention, the timing processing module includes:

[0023] The time series alignment submodule calls the time series alignment algorithm to perform time axis calibration and sequence length adjustment on the timestamps, identify the start and end point deviations of the differentiated time series, map multiple groups of time series to a unified time base, and generate an aligned time series sequence;

[0024] The difference calculation submodule establishes a data window with a fixed step size based on the aligned time series, accumulates and sums the continuous differences in the window, calculates the arithmetic mean of the absolute differences in each window, and generates a window difference mean;

[0025] The weight generation submodule inputs the window difference mean into a nonlinear transformation function, sorts the transformation results in ascending order, converts the accumulated value into a weight coefficient through linear transformation, and generates a time series weight parameter corresponding to the spatial dimension of the geographic coordinate.

[0026] As a further solution of the present invention, the space optimization module includes:

[0027] The spatial clustering submodule detects the geographic coordinates, calculates the Euclidean distance between the coordinates using the DBSCAN clustering algorithm, divides high-density clusters and noise points according to density differences, calculates the sample distribution density within the cluster and the distance between adjacent clusters, determines the boundary range of the core point through density accessibility, and generates a spatial clustering density value;

[0028] The weight normalization submodule calculates the density difference range between clusters based on the spatial clustering density value, maps the density value to the [0,1] interval through the maximum and minimum value normalization method, constructs a weight distribution function based on the mean distance between clusters, and generates a normalized weight coefficient;

[0029] The credibility fusion submodule calls the time series weight parameter, performs weighted summation with the normalized weight coefficient, allocates the time and space dimension weight ratio, uses the standard deviation normalization method to eliminate the dimensional difference of the weighted result, and generates a comprehensive credibility index.

[0030] The maximum and minimum values ​​are respectively taken from the density extremes of historical data in the same geographical area;

[0031] The standard deviation calculation is based on the weighted sum dataset of all devices within the same time window.

[0032] As a further solution of the present invention, the dynamic fusion module is used to sort the comprehensive credibility indicators using an evidence theory decision model, perform correlation analysis on the sorting results and the signal-to-noise ratio, input the device operating voltage into a Kalman filter algorithm for compensation, and output a fused spectrum feature vector;

[0033] The value of the process noise covariance matrix Q in the Kalman filter algorithm is determined based on the equipment voltage fluctuation variance.

[0034] As a further solution of the present invention, the dynamic fusion module includes:

[0035] The evidence decision submodule calls the comprehensive credibility index, defines the focal element set using the evidence theory decision model, calculates the confidence intervals and plausibility intervals of multiple indicators, merges conflicting evidences using the Dempster synthesis rule, and generates a credibility ranking sequence;

[0036] The correlation analysis submodule calculates the sum of squares of rank differences based on the credibility ranking sequence and the signal-to-noise ratio data, substitutes the sum of squares into the Spearman formula to obtain the correlation coefficient, analyzes the consistency of the ranking and the direction of change of the signal-to-noise ratio, and generates a weighted Spearman correlation coefficient;

[0037] The filter compensation submodule constructs a Kalman filter state transfer matrix and an observation matrix based on the operating voltage of the device, iteratively calculates a priori estimated covariance matrix, adjusts the Kalman gain in combination with the weighted Spearman correlation coefficient, and generates a fusion spectrum feature vector.

[0038] As a further solution of the present invention, the focus element set is divided into three levels: high confidence / medium confidence / low confidence based on the working status of the device, and the working status of the device is determined by the voltage stability and signal quality coefficient;

[0039] The device weight factor is introduced into the Spearman formula Perform weighted calculation, and the formula is improved to:

[0040] ;

[0041] in, represents the weighted Spearman correlation coefficient, The weight factor of the i-th device is calculated by the comprehensive credibility index. represents the rank difference of the i-th device, and n represents the total number of devices participating in the calculation;

[0042] The Kalman gain and weighted Spearman correlation coefficient satisfy,

[0043] Relationship, where Represents the prior estimate covariance matrix, H is the observation matrix, R is the observation noise covariance matrix, the superscript T represents the matrix transpose operation, and the subscript k represents the current time.

[0044] Compared with the prior art, the advantages and positive effects of the present invention are:

[0045] In the present invention, by periodic scanning combined with a multi-source heterogeneous data encapsulation protocol, timestamps, geographic coordinates, signal-to-noise ratios, and device voltages are integrated to achieve standardized collection and dynamic association of multi-dimensional data. The time series alignment algorithm generates timing weight parameters through absolute difference calculation, strengthens the timing distinction between high-frequency events and low-frequency events, and reduces noise interference in the time dimension. The DBSCAN clustering algorithm generates spatial weight coefficients based on Euclidean distance calculated from geographic coordinates, and generates a comprehensive credibility index by linear weighting of timing parameters to improve the coupling accuracy of spatial distribution and temporal dynamics. The evidence theory decision model sorts the credibility indicators and performs correlation analysis with the signal-to-noise ratio. The Kalman filter algorithm compensates for device voltage fluctuations to achieve dynamic fusion and error suppression of multi-source data. Through time-space collaborative optimization and multi-source heterogeneous data fusion, the anti-interference ability of the spectrum feature vector is enhanced, the positioning deviation of the interference source in complex environments is reduced, and the efficiency of spectrum resource allocation and the long-term stability of the monitoring system are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a system flow chart of the present invention;

[0047] Figure 2 This is a flow chart of the crowdsourcing acquisition module of the present invention;

[0048] Figure 3 This is a flow chart of the timing processing module of the present invention;

[0049] Figure 4 This is a flow chart of the space optimization module of the present invention;

[0050] Figure 5 This is the flow chart of the dynamic fusion module of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0052] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0053] Example 1:

[0054] See also Figure 1 , a radio spectrum measurement system based on mobile crowdsourcing includes:

[0055] The crowdsourcing acquisition module is used to periodically scan radio spectrum signals through terminal devices to obtain timestamps, geographic coordinates, signal-to-noise ratios, and device operating voltages. It uses a multi-source heterogeneous data encapsulation protocol for structured encapsulation, transmits timestamps and geographic coordinates to the timing processing module, and outputs signal-to-noise ratios and device operating voltages to the dynamic fusion module.

[0056] The time series processing module is used to call the time series alignment algorithm to calculate the absolute difference of the timestamps, generate the time series weight parameters in ascending order, and pass the time series weight parameters and geographic coordinates to the spatial optimization module;

[0057] The spatial optimization module is used to calculate the Euclidean distance of the target area using the DBSCAN clustering algorithm based on geographic coordinates, generate a normalized spatial weight coefficient, and determine the weighted ratio of the temporal weight parameter and the spatial weight coefficient based on the historical data training regression model to generate a comprehensive credibility index, and output the comprehensive credibility index to the dynamic fusion module;

[0058] The dynamic fusion module is used to sort the comprehensive credibility indicators using the evidence theory decision model, perform correlation analysis on the sorting results and the signal-to-noise ratio, input the equipment working voltage into the Kalman filter algorithm for compensation, and output the fusion spectrum feature vector.

[0059] The multi-source heterogeneous data encapsulation protocol includes field mapping rules and CRC check mechanism;

[0060] The time series alignment algorithm uses a dynamic time warping algorithm to achieve multi-device time axis calibration;

[0061] Before calculating the Euclidean distance, the geographic coordinates are converted to the UTM coordinate system to achieve dimension unification;

[0062] The value of the process noise covariance matrix Q in the Kalman filter algorithm is determined based on the equipment voltage fluctuation variance.

[0063] See also Figure 2 , the crowdsourcing acquisition module includes:

[0064] The spectrum scanning submodule periodically scans radio spectrum signals through the terminal device, detects the carrier frequency and signal strength, records the timestamp of the scanning moment, synchronously obtains the geographic coordinates output by the device's built-in positioning system, and converts the electromagnetic wave signal strength continuously collected during the scanning period into a spectrum scanning value;

[0065] The electromagnetic wave signal strength conversion formula is:

[0066] ;

[0067] in, Represents the spectrum scan value, is the signal power spectral density at frequency f, and are the lower and upper limits of the scanning frequency band, respectively. df is a small frequency increment, which comes from the integral operation and is a standard notation in calculus.

[0068] The spectrum scanning submodule uses a terminal device, such as a smartphone model ABC-S1, to periodically scan the radio spectrum signals in the city center, at the specific geographical coordinates of longitude 116.397128° and latitude 39.916527°. The scanning operation is performed every 500 milliseconds. The built-in RF front-end module of the device is responsible for receiving signals within the specified frequency band, such as the authorized frequency band from 800 MHz to 900 MHz. Multiple carrier signals are detected, one of which has a signal strength of -75dBm at 850.5 MHz and another has a signal strength of -82dBm at 860.2 MHz. At the same time, the signal strength is recorded. The scanning time before, for example, 10:00:00:350 on May 30, 2025, is used, and the built-in global positioning system (GPS) module of the device is synchronously called to obtain the current geographic coordinates output by the module. The coordinate data is 116.397128 degrees east longitude and 39.916527 degrees north latitude. The electromagnetic wave signal strength data continuously collected during the scanning period, that is, from 10:00:00:00:00:00 milliseconds on May 30, 2025 to 10:00:00:00:500 milliseconds on May 30, 2025, are converted into a series of power values ​​that vary with frequency. The specific conversion is based on the formula Calculate the spectrum scan value, where Represents the signal power spectral density at frequency f, in watts per hertz (W / Hz). Represents the lower limit frequency of the scanning frequency band, which is set to 800 MHz (8×10 8 hertz), Represents the upper limit frequency of the scanning band, which is set to 900 MHz (i.e. 9×10 8 Hertz), df is a small increment of frequency, the integral operation logic is to specify the frequency interval Divide into several extremely small frequency subintervals and calculate the center frequency point of each subinterval The power spectral density at With the width of the subinterval The product of, and then add up all these products, when When it approaches 0, the accumulated sum is the integral value. This process is intended to quantify the total signal energy in a specified frequency band. Assume that in a simplified scenario, the entire frequency band is divided into 1000 0.1 MHz (1×10 5 Hz) and measure the average power spectral density of each sub-band. For example, in the first sub-band (800.0MHz-800.1MHz), the measured , then the power of this sub-band is , accumulate the power of all 1000 sub-bands to get the total spectrum scan value. If the value obtained by detailed integration is The value is Watt, the spectrum scanning value represents the comprehensive intensity of electromagnetic energy in a specific frequency band at the location of the device under the set scanning parameters, and outputs a fused data packet. Formula The benefit of this is that by integrating the signal power spectrum density within a specified frequency range, the total signal energy within the frequency band can be accurately quantified, overcoming the limitations of single-point frequency intensity measurement and providing more comprehensive raw data input for subsequent signal quality evaluation. =800MHz to = Signal power spectral density in the 810MHz band It can be approximated as a constant whose value is , which is a typical reading of a weak broadband signal by a calibrated measuring device in a specific quiet environment, then the spectrum scan value The calculation process is:

[0069] ;

[0070] ;

[0071] The results show that in the 800MHz to 810MHz frequency band, the spectrum scanning value is Watt, this value will be used as one of the inputs of the subsequent signal analysis submodule to evaluate the signal quality.

[0072] The voltage monitoring submodule uses the raw voltage data output by the device power monitoring circuit, calculates the standard deviation of the voltage sampling sequence using a sliding window algorithm, eliminates transient interference pulses, and generates the device operating voltage.

[0073] The sliding window length L is based on the device sampling rate according to Dynamic adjustment, where Indicates the device sampling rate, is the ceiling operator;

[0074] The voltage monitoring submodule calls the raw voltage data output by the power monitoring circuit provided by the power management unit (PMU) of the aforementioned smartphone model ABC-S1. The raw voltage data is a series of instantaneous voltage values ​​collected at a specific sampling rate, such as a sampling rate of =50 Hz, that is, 50 voltage samples are collected per second. These samples may contain instantaneous fluctuations caused by circuit noise or external electromagnetic interference. For example, the voltage sequence collected in a certain second is [3.75V, 3.76V, 3.74V, 3.95V, 3.75V, ..., 3.77V], where 3.95V may be an interference pulse. The sliding window algorithm is used to calculate the standard deviation of the voltage sampling sequence. The length of the sliding window L is based on the device sampling rate. according to Dynamic adjustment, where is the ceiling operator, for =50 Hz, then the window length The window length of 0.2 seconds is selected based on experience. This time is long enough to smooth out most short-term voltage spikes and respond quickly to the actual voltage trend changes. The specific calculation process is to take 10 consecutive voltage sampling values, for example [3.75, 3.76, 3.74, 3.78, 3.75, 3.77, 3.76, 3.79, 3.75, 3.77] volts, calculate the average of these 10 values, and get Volts, then calculate the square of the difference between each sample value and the average value, calculate the average of these squared differences, and finally take the square root to get the standard deviation ,For example , if the calculated standard deviation =0.015 volts. This standard deviation value is used to judge the stability of the voltage. A smaller standard deviation indicates that the voltage is relatively stable, while a larger standard deviation indicates that there are significant fluctuations or interference. By setting a standard deviation threshold, such as 0.05 volts, if the standard deviation calculated in the current window is less than the threshold, it is considered that the voltage data in the window is less affected by the instantaneous interference pulse, and its average value can represent the current device operating voltage. If the standard deviation is greater than the threshold, the maximum or minimum value in the window may be eliminated and recalculated, or more complex filtering methods may be used to eliminate the instantaneous interference pulse and generate the device operating voltage. For example, after processing, the stable device operating voltage is 3.76 volts. Sliding window length In the setting, the coefficient 0.2 is an empirical value obtained from a large amount of experimental data. The experimental process includes using different types of mobile terminal devices in a variety of typical electromagnetic environments, collecting voltage data at different sampling rates (for example, from 10Hz to 100Hz), and artificially introducing interference pulses of different strengths and widths. By comparing the suppression effect of interference pulses with different window lengths (corresponding to time windows such as 0.1 seconds, 0.2 seconds, and 0.3 seconds) and the response sensitivity to real voltage changes, it is found that the 0.2-second time window can achieve a good balance between effectively filtering out common transient interference (such as short voltage drops caused by GSM communications) and maintaining a rapid response to voltage trend changes. For example, for the sampling rate = 100 Hz device, its sliding window length sampling points, if the sampling rate is =20 Hz, then However, to ensure statistical significance, a minimum window length is usually set, for example, 5 sampling points. Therefore, L will be 5 in this case. This dynamic adjustment ensures that similar filtering time constants can be achieved on devices with different sampling capabilities.

[0075] The signal analysis submodule uses fast Fourier transform (FFT) to separate the energy distribution of signal frequency bands based on the spectrum scan values. It then calculates the ratio of the power in the main frequency band to the power in the noise frequency band to generate a signal-to-noise ratio (SNR). It then performs piecewise linear interpolation on the SNR and the device operating voltage based on the device power consumption curve, which is constructed using a voltage-power consumption table provided by the device manufacturer. The module then outputs the signal quality factor.

[0076] The signal analysis submodule receives the spectrum scanning value output by the aforementioned spectrum scanning submodule, for example Watts, and the device operating voltage output by the voltage monitoring submodule, for example, 3.76 volts. First, based on the spectrum scan value, the signal data collected in the time domain (its integral form is expressed as the spectrum scan value, and in actual operation, FFT may be performed on the original time domain sampling data) is converted to the frequency domain by performing a fast Fourier transform (FFT) operation, thereby obtaining the energy distribution of the signal at different frequency points. For example, the analysis results show that the cumulative power in the frequency band of 850.0 MHz to 851.0 MHz (the main signal band) is Watts, while the total noise power in the adjacent 849.0 MHz to 850.0 MHz and 851.0 MHz to 852.0 MHz bands (assumed to be the main noise source bands) is Watts, and then the ratio of the main frequency band power to the noise frequency band power is calculated to generate the signal-to-noise ratio (SNR), which is calculated as , or usually expressed in decibels (dB) , combining this signal-to-noise ratio (e.g., a value of 5) with the device operating voltage (e.g., 3.76 volts), and performing piecewise linear interpolation on the device power consumption curve constructed by referring to the voltage-power consumption correspondence table provided by the device manufacturer. The device power consumption curve is a pre-stored lookup table or function that describes the standard power consumption values ​​of the device at different operating voltages. For example, the following correspondence may exist:

[0077] Table 1: Device voltage-power consumption correspondence table

[0078] Voltage (V) Power consumption reference value (mW) 3.60 500 3.70 550 3.80 620 3.90 700

[0079] As shown in Table 1, the table lists the reference power consumption values ​​of specific devices at different operating voltages. The specific process of piecewise linear interpolation operation is as follows: first, based on the current device operating voltage of 3.76 volts, the corresponding standard power consumption is found on the power consumption curve. Since 3.76 volts is between 3.70 volts and 3.80 volts, the following is calculated by linear interpolation:

[0080] ;

[0081] This value represents a baseline performance state at the current voltage. The calculated signal-to-noise ratio value (e.g., 5) is then comprehensively evaluated with this standard power consumption value (or a performance benchmark directly associated with the voltage). This evaluation can be a preset mapping function, such as the signal quality coefficient. ,in is the operating voltage. A simple implementation method is to normalize the SNR to a predetermined range (for example, 0 to 1, where an SNR of 0 corresponds to 0, and an SNR greater than or equal to 10 corresponds to 1, with a linear mapping in between), and then multiply it by a voltage-related adjustment factor. This adjustment factor reflects whether the voltage is within the optimal operating range. For example, if the voltage of 3.76V is in the optimal range, the adjustment factor is 1. If the voltage is too low (for example, below 3.65V) or too high (for example, above 3.85V), the adjustment factor is less than 1. Assuming an SNR of 5, it is 0.5 after normalization, and the voltage adjustment factor is 0.95 (because 3.76V deviates slightly from the ideal 3.8V). The signal quality coefficient is 0.5×0.95=0.475, and the output signal quality coefficient is 0.475.

[0082] The data encapsulation submodule calls the timestamp and geographic coordinates, constructs a time series data frame according to the multi-source heterogeneous protocol, performs structured encoding on the signal quality coefficient and the equipment operating voltage according to the dynamic fusion protocol, generates a time series transmission frame, and outputs a fused data packet.

[0083] The data encapsulation submodule calls the timestamp recorded by the spectrum scanning submodule, such as 10:00:00 350 milliseconds on May 30, 2025, and the geographic coordinates, such as 116.397128° east longitude and 39.916527° north latitude, to construct a time series data frame according to the predefined multi-source heterogeneous protocol. The protocol specifies the structure of the data frame, such as the frame header, data type identifier, timestamp field, coordinate field, data payload field, and checksum field. The specific construction process is as follows: first, a fixed-length frame header is generated to identify the start of the data frame and the protocol version, then the data type identifier is filled in to indicate that this is a frame containing spectrum sensing data, then the timestamp is converted to a standard format, such as Unix timestamp (millisecond level), and filled in the corresponding field, and the latitude and longitude values ​​of the geographic coordinates are multiplied by a fixed scaling factor (for example, 10 7 ) and rounded, stored in integer form to save space and maintain accuracy, for example, longitude becomes 116.3971280°, latitude becomes 39.9165270°, and then the signal quality coefficient output by the signal analysis submodule, for example, 0.475, and the device operating voltage output by the voltage monitoring submodule, for example, 3.76 volts, are structured and encoded according to the dynamic fusion protocol. The dynamic fusion protocol may stipulate that these two floating-point values ​​are converted into fixed-point numbers of a specific precision or compressed in binary format. For example, the signal quality coefficient 0.475 is multiplied by 1000 and rounded to 475, and the device operating voltage is 3.76 volts. 3.76 volts is multiplied by 100 and rounded to 376. These two encoded values ​​are then arranged in sequence to form a data payload. Finally, a checksum, such as CRC16, is calculated based on all contents from the frame header to the data payload and appended to the end of the frame to generate a complete timing transmission frame. The timing transmission frame is a fused data packet. For example, its binary content may be: [frame header byte sequence][type identification byte][timestamp byte sequence][longitude byte sequence][latitude byte sequence][encoded signal quality coefficient byte sequence][encoded operating voltage byte sequence][checksum byte sequence], and the fused data packet is output.

[0084] See also Figure 3 , the timing processing module includes:

[0085] The time series alignment submodule calls the time series alignment algorithm to calibrate the time axis and adjust the sequence length of the timestamps, identify the start and end point deviations of the differentiated time series, map multiple groups of time series to a unified time base, and generate aligned time series.

[0086] The timing alignment submodule receives fused data packets collected and encapsulated at different times from multiple different terminal devices (such as device A, device B, and device C). Each data packet contains its own independently recorded timestamp. For example, the data packet timestamp of device A is 10:00:00.350, that of device B is 10:00:00.420, and that of device C is 10:00:00.385. The time series alignment process is called to first calibrate the time axis of the timestamp in each data packet. The global positioning system (GPS) timing signal is used as a unified synchronization source. The standard time provided by this synchronization source has a small The synchronization error is about 1 microsecond. The calibration process compares the timestamp reported by each device with the closest GPS second pulse signal to calculate the deviation between the local clock of the device and the GPS standard time. For example, if the local clock of device A is 2 milliseconds fast, its timestamp after calibration is 10:00:00.348, and device B is 1 millisecond slow, so the timestamp after calibration is 10:00:00.421. Device C is accurate and still 10:00:00.385. The sequence length is adjusted to ensure that all time series involved in the analysis have the same number of sampling points or time span in the time period of interest. If some devices are in a specific time period, the time span will be the same. If there is no data at a time point, interpolation methods (such as linear interpolation or previous value filling) are used to fill the missing data, or the data is marked as invalid in subsequent processing. Then, the starting endpoint deviation of the differentiated time series is identified. For example, if the analysis window is 10:00:00 to 10:00:05, the data of device A starts at 10:00:00.348, device B starts at 10:00:00.421, and device C starts at 10:00:00.385. The system will determine a common analysis starting point, such as selecting the earliest valid data point or a fixed grid time point 10:00:00.500. , map multiple sets of calibrated and length-adjusted time series (such as the signal quality coefficient value series of each device) to this unified time base. For example, if data was originally collected every 500ms, the unified time base is also set to 500ms intervals, starting from 10:00:00.500, 10:00:01.000, 10:00:01.500, etc. The data values ​​of each device at these reference time points are obtained by selecting the nearest original data point or interpolation to generate an aligned time series. At this time, all devices have a corresponding signal quality coefficient value at each unified time point.

[0087] The difference calculation submodule establishes a fixed-step data window based on the aligned time series, accumulates and sums the continuous differences within the window, calculates the arithmetic mean of the absolute differences within each window, and generates the window difference mean;

[0088] The difference calculation submodule is based on the alignment timing sequence generated by the timing alignment submodule. For example, for the signal quality coefficient indicator, the difference calculation submodule obtains the alignment timing sequence of devices A, B, and C at the same time point. Aligned sequences on , , First, a fixed step size data window is established. The fixed step size ranges from 5 to 10 sampling points, and a prime number length is preferred, such as 5 or 7. In this example, the step size is W = 5 sampling points. The purpose of selecting the prime number length is to reduce harmonic interference in the subsequent possible frequency domain analysis. For the sequence of device A, the data contained in the first window is , the second window is (if sliding window) or (If it is a non-overlapping window), it is assumed here that it is a sliding window, and then the continuous differences in the window are accumulated and summed. Specifically, for the first window , calculate the successive differences: , , , , these differences reflect the fluctuation of the signal quality coefficient in a short period of time, and then calculate the arithmetic mean of these absolute differences in each window. For the above example, the window difference mean is , (Note: for a window of length W, there are W-1 differences), perform this operation on the time series of each device to obtain a series of window difference means, for example, the window difference mean sequence of device A is , device B is , device C is , generate the window difference mean.

[0089] The weight generation submodule inputs the window difference mean into the nonlinear transformation function, sorts the transformation results in ascending order, converts the accumulated value into a weight coefficient through linear transformation, and generates a time series weight parameter corresponding to the spatial dimension of the geographic coordinate;

[0090] The nonlinear transformation function uses the sigmoid function to realize probability mapping, and its expression is , where k is the curve steepness coefficient and its value is 0.5, The mean shift is the median of the mean of the window differences.

[0091] The weight generation submodule receives the window difference mean sequence of each device generated by the difference calculation submodule, such as device A , device B , device C , these window difference means are input into the nonlinear transformation function one by one, which uses the sigmoid function to realize the probability mapping, and its expression is , where k is the curve steepness coefficient, set to 0.5, is the mean shift, which is set to the median of the mean window difference of all devices in the current batch. The setting of the steepness coefficient k=0.5 is based on simulation experiments. We tested multiple k values ​​(such as 0.1, 0.5, 1.0, and 2.0) on simulated data sequences with different degrees of fluctuation. It was found that k=0.5 can distinguish the difference in data stationarity (i.e., the difference in the mean size of the window difference) while avoiding the output saturation to 0 or 1 due to too large a k value, or the discrimination due to too small a k value. Insufficient problem, it provides a relatively smooth transition. Assume that the mean data of the window difference currently processed is [0.015, 0.018, 0.012, 0.010, 0.011, 0.014, 0.008, 0.009, 0.007], and after sorting it is [0.007, 0.008, 0.009, 0.010, 0.011, 0.012, 0.014, 0.015, 0.018], the median of which is , for the first window difference mean of device A , its sigmoid transformation value is , Since the smaller the mean of the window difference is, the more stable the data is, we expect its weight to be higher, so we usually use , or adjust the sign of k. If you use this directly And small differences correspond to high "original" weights, then the subsequent sorting sorts the transformation results in ascending order, for example, the sigmoid transformation results of the current window for all devices Then, these sorted values ​​(or some accumulated form thereof) are converted into the final weight coefficients through linear transformation. For example, if there are three devices A, B, and C, their transformation values ​​are 0.5005, 0.4750 (from ),0.4256(from ), after ascending sorting, it is [0.4256, 0.4750, 0.5005]. If these values ​​or their normalized forms are used directly as weights (note that a smaller value means a smaller original difference and is more credible), or weights are assigned according to the sorting position, for example, the weight of the first place is 0.5, the second is 0.3, and the third is 0.2 (the total is 1), these weight parameters are associated with the spatial dimension information of the geographic coordinates, because each time series data itself is bound to the geographic coordinates, and finally a time series weight parameter corresponding to the spatial dimension of the geographic coordinates is generated. For example, the time series weight of device C in this time window is 0.5, device B is 0.3, and device A is 0.2. Formula In the formula, x is the mean of the input window difference, which represents the degree of fluctuation of the data in the short term; k is the steepness coefficient, which controls the slope of the function curve and affects the sensitivity of the output value to the change of the input value; is the mean shift, usually set to the mean or median of the input data set, as the symmetrical center point of the sigmoid function curve; e is the base of the natural logarithm, approximately 2.71828, exponential operation The scaled exponent of the deviation of the input value from the center point is calculated, 1 is added, and the inverse is taken to map the result to the (0,1) interval. The logic of this formula is to convert the input value x to a value between 0 and 1. When x is much smaller than hour, is a negative number with a large absolute value. is a large positive number. Close to 0, (here we assume , if the window difference is as small as possible, it may be used or or subsequent sorting). and As an example, the mean of the following three window differences is calculated:

[0092] Device C: ;

[0093] Device B: ;

[0094] Device A: ;

[0095] These results show that when the window difference mean x is less than the center point hour, Slightly less than 0.5, when x is greater than hour, Slightly greater than 0.5, if the goal is to have smaller differences and larger weights, you can use As a weight basis, or set k to a negative value, such as ,but:

[0096] Device C: ;

[0097] Device B: ;

[0098] Device A: ;

[0099] At this time, the smaller the value, The larger (when ), the larger the value, The smaller (when ), : is the sign-changed form of the steepness coefficient k in the original formula, that is , the article takes This is a strategy for mapping smaller input x to larger sigmoid output values ​​to make the relationship of “smaller differences, higher weights” more intuitive and direct. :Use negative steepness The result after introducing the sigmoid function: , which is a modified sigmoid function used to implement reverse mapping. This conforms to the intuitive understanding that small differences are associated with high weights, and these values ​​or their normalized forms can be used directly in the future. The benefit of this formula is that, through nonlinear mapping, the mean of window differences, which can vary widely, can be compressed into a fixed small range (such as 0 to 1). The center point and steepness can also be adjusted to achieve good sensitivity and discrimination within the range of differences of interest, facilitating subsequent weight assignment.

[0100] The unified time base uses GPS timing signals as the synchronization source, and its time synchronization error is less than 1 microsecond;

[0101] The fixed step size ranges from 5 to 10 sampling points and preferably takes a prime number length, which includes 5, 7, and 11.

[0102] See also Figure 4 , space optimization modules include:

[0103] The spatial clustering submodule detects geographic coordinates, uses the DBSCAN clustering algorithm to calculate the Euclidean distance between coordinates, divides high-density clusters and noise points based on density differences, calculates the sample distribution density within the cluster and the distance between adjacent clusters, determines the boundary range of the core point through density accessibility, and generates a spatial cluster density value;

[0104] The spatial clustering submodule processes the geographic coordinates of each terminal device when reporting data. For example, in a certain area, 100 devices uploaded data at the same time, and their geographic coordinates are First, the validity of these geographic coordinates is tested, and obviously wrong coordinates (such as longitude and latitude out of range) are eliminated. Then, a density-based spatial clustering method is used. The specific steps are as follows: for any two device coordinate points and , calculate the Euclidean distance between them (In actual application, it will be converted to metric units), set two key parameters: neighborhood radius and the minimum number of neighborhood samples MinPts required for the core object, The setting of refers to the scale of the study area and the expected distribution density of data points, for example, for vehicle or pedestrian data in an urban environment, It can be set to 50 meters. The MinPts setting is usually greater than or equal to 3, for example, it is set to 5. The experimental verification process of this parameter includes: selecting a historical geographic location dataset, trying different The rationality of the clustering results is observed by adjusting the values ​​of the distance (such as 20 meters, 50 meters, 100 meters) and the MinPts value (such as 3, 5, 10). For example, whether the high-density areas such as main roads and points of interest can be effectively distinguished, while avoiding the incorrect merging of sparse areas or over-segmentation of dense areas. By comparing clustering evaluation indicators such as the silhouette coefficient or DB index under different parameter combinations, the optimal parameter combination is selected. For each data point P, check its Neighborhood (i.e., with P as the center, The number of data points contained in the circular area with radius ,like , then point P is marked as a core point if But P is located at a core point Q If P is neither a core point nor a boundary point (i.e. And there is no core point in its neighborhood), then P is marked as a noise point, and high-density clusters (composed of core points and their density-reachable boundary points) and noise points are divided according to density differences. The sample distribution density within each formed cluster is counted, such as cluster Include points, and their average internal distance is , then the density can be approximated as , and calculate the distance between adjacent clusters, such as cluster and clusters The distance between them can be the distance between their nearest boundary points or the distance between their centroids, which is obtained by density accessibility (i.e. starting from a core point, through a series of The neighborhood can reach another point) determines the boundary range that the core point can extend to, and finally generates a spatial clustering density value for each data point or each cluster. For example, for a data point belonging to a high-density cluster The density value of the point can be the average point density of the cluster, or the local density of the point itself (its The density value of noise points is low, such as 0 or a preset small value, to generate a spatial clustering density value.

[0105] The weight normalization submodule calculates the density difference range between clusters based on the spatial clustering density value, maps the density value to the [0,1] interval through the maximum and minimum value normalization method, constructs a weight distribution function based on the mean distance between clusters, and generates a normalized weight coefficient;

[0106] The weight normalization submodule is based on the spatial clustering density value generated by the spatial clustering submodule, for example, for different clusters within the region , and their corresponding average density values ​​are , and at the same time obtain the density extreme value of historical data in the geographical area, that is, the historical maximum density and historical minimum density (Usually the density of noise points can be regarded as Or a very small positive number), these historical extreme values ​​are obtained by performing the same cluster analysis on all data points collected in the area over the past period of time (e.g., one month), and recording the maximum and minimum average densities of the clusters formed. For example, after statistical analysis, the historical maximum density , the lowest density in history First, calculate the density difference range between the current clusters. For example, the current cluster density value is [10, 25, 5, 30] points / km 2 , and then the density value of each current cluster is normalized by the maximum and minimum value method Map to The normalized formula is , for a cluster density of 25 points / km 2 The cluster with normalized density is , if the density value of the current cluster Beyond the historical scope, e.g. , then its normalized value is set to 1, if , then set it to 0, and then combine the average distance between each cluster (for example, cluster With all other clusters The average distance ) to construct a weight distribution function. One way is that the higher the density of the cluster, the higher its weight. At the same time, if a high-density cluster has a large average distance from other clusters (meaning it is relatively isolated or unique), its weight may also be improved. On the contrary, if it is surrounded by other high-density clusters (with a small average distance), its contribution may need to be adjusted. A simple weight distribution function can directly use the normalized density value as a base weight, or multiplied by an adjustment factor related to the inter-cluster distance, e.g. ,in It is a function that varies with distance. For example, if the average distance between clusters is larger, the adjustment factor is smaller to reduce the excessive impact of isolated high-density clusters, or vice versa, depending on the preference of the application scenario for spatial distribution. Finally, a normalized weight coefficient is generated, for example, cluster The normalized weight coefficient is 0.489.

[0107] The credibility fusion submodule calls the time series weight parameter, performs weighted summation with the normalized weight coefficient, allocates the weight ratio of the time and space dimensions, uses the standard deviation normalization method to eliminate the dimensional difference of the weighted result, and generates a comprehensive credibility index.

[0108] The credibility fusion submodule calls the timing weight parameters generated by the timing processing module, for example, the timing weight of device i in time window t , and combined with the normalized weight coefficient generated by the spatial optimization module, the normalized weight coefficient is for the spatial cluster where the device i is located For example, , if device i, if device , then its spatial weight is , perform weighted summation of these two types of weights and assign weight proportions to the spatiotemporal dimensions. For example, the importance of the time dimension is set to =0.7, the importance of spatial dimension is =0.3 (where These proportions are determined by expert experience or cross-validation experiments based on the emphasis of the application scenario on time and space factors. For example, in a rapidly changing environment, the time weight may be higher). The preliminary weight value formula for device i is: , The series of weighted values ​​calculated for all devices in the same time window (for example, if there are N devices, N value: ), the standard deviation normalization method (also known as Z-score normalization) is used to eliminate dimensional differences. The specific calculation process is: first calculate this group Mean of the values and standard deviation , such as the mean = 0.55 and standard deviation = 0.05, and then for each device To do the conversion: , for device i, 0.5667, 0.5667 may be a calculation result or data value, which involves some standardized calculation or the output of a specific algorithm, such as the value generated by the weighted algorithm, time alignment or space optimization module. Its standard deviation normalization value is ,These normalized values ​​usually have the characteristics of a mean of 0 and a standard deviation of 1, ,generating a comprehensive credibility index. For example, the comprehensive credibility index of device i is 0.334.

[0109] The maximum and minimum values ​​are taken from the density extremes of historical data in the same geographical area;

[0110] The standard deviation calculation is based on the weighted sum dataset of all devices within the same time window.

[0111] See also Figure 5 , the dynamic fusion module includes:

[0112] The evidence decision submodule calls the comprehensive credibility index, adopts the evidence theory decision model to define the focal element set, calculates the confidence intervals and plausibility intervals of multiple indicators, merges the conflicting evidence through the Dempster synthesis rule, and generates a credibility ranking sequence;

[0113] The evidence-based decision-making submodule uses the comprehensive credibility index of each device output by the credibility fusion submodule. For example, the index for device A is 0.334, for device B is 1.250, and for device C is -0.860. The evidence-based decision-making model is used to define the focal element set. First, the potential credibility level of each device is divided into three focal elements: high credibility, medium credibility, and low credibility based on the device's operating status. The device's operating status is determined by the voltage stability output by the voltage monitoring submodule (for example, a voltage fluctuation standard deviation less than 0.02V is considered stable) and the signal quality coefficient output by the signal analysis submodule (for example, an SQC greater than 0.6 is considered high quality). For example, if the voltage is stable and the SQC is high, the device data is initially considered to be more likely to belong to the "high credibility" focal element. If the voltage is unstable or the SQC is low, it is more likely to belong to the "low credibility" focal element. Other cases are classified as "medium credibility." Then, the comprehensive credibility index (for example, 0.334) is mapped to the basic probability allocation (BPA) value for these three focal elements through a conversion function (such as a piecewise function or probability distribution function), or called a quality function. For example, for an index of 0.334, one might assign (in Represents uncertainty), perform this operation on each device to obtain a set of BPA functions, and then calculate the interval of belief (Bel) and plausibility (Pls) of each device on each focal element. is the sum of all evidence that fully supports hypothesis A, plausibility is the sum of all evidence that does not conflict with hypothesis A, e.g. (According to the specific definition of focal elements), the BPA functions of these focal elements from different information sources (here, it can be understood as the comprehensive credibility index of each device as one evidence source, or multiple different evaluation dimensions as different evidence sources) are combined through the Dempster synthesis rule. The formula of the Dempster synthesis rule is , in It is a conflict factor. 、 Two Basic Probability Assignment (BPA) functions, each derived from two different information sources (e.g., two devices, or two evaluation dimensions of a device); A represents a focal element, i.e., a set of hypotheses, such as {high confidence}, {high confidence, medium confidence}, etc. The target focal element to be calculated after synthesis; B and C are BPA functions. and The focal element defined in 、 A collection of For the intersection of focal elements B and C, Dempster's rule only performs cumulative processing on combinations whose intersection is not empty; Sum all combinations that satisfy the intersection of B and C is exactly equal to A and is not empty; K conflict factor (conflict coefficient), which represents all conflicting combinations (i.e. ) is the key factor for subsequent normalization. If K = 1, it means complete conflict and synthesis fails. The normalization coefficient is used to adjust the results so that the sum of all synthesized quality distributions is still 1. This process can integrate multiple uncertain and imprecise evidence, resolve conflicts between evidence, and obtain a combined BPA. Finally, based on the combined BPA, the trust or plausibility of each device is calculated as high, medium, or low credibility level, and the devices are sorted accordingly to generate a credibility ranking sequence. For example, the ranking result is device B (highest credibility), device A (medium credibility), and device C (lowest credibility).

[0114] The correlation analysis submodule calculates the sum of squares of rank differences based on the credibility ranking sequence and signal-to-noise ratio data, substitutes it into the Spearman formula to obtain the correlation coefficient, analyzes the consistency of the ranking and signal-to-noise ratio change directions, and generates a weighted Spearman correlation coefficient;

[0115] Introducing device weight factor into Spearman formula Perform weighted calculation, and the formula is improved to:

[0116] ;

[0117] in, represents the weighted Spearman correlation coefficient, The weight factor of the i-th device is calculated by the comprehensive credibility index. represents the rank difference of the i-th device, and n represents the total number of devices participating in the calculation;

[0118] The correlation analysis submodule is based on the credibility ranking sequence generated by the evidence decision submodule. For example, the device order is [device B, device A, device C], and its corresponding rank is [1, 2, 3]. The signal-to-noise ratio (SNR) data of the corresponding devices obtained from the signal analysis submodule, for example, the SNR of device B is 8.5dB, device A is 6.99dB, and device C is 4.2dB. First, the SNR data is also sorted and ranked. The SNR is sorted from high to low as [8.5dB, 6.99dB, 4.2dB], and the corresponding SNR rank is also [1, 2, 3]. The difference between the two ranks of each device is calculated. , device B , device A , device C , and then calculate the squares of these rank differences , and introduces the device weight factor calculated by the comprehensive credibility index , the weight factor The setting of the device's comprehensive credibility index refers to the relative size of the device among all devices. For example, the comprehensive credibility index can be normalized to The interval is used as the weight, or it is assigned according to its position in the credibility ranking. For example, assuming that the weight calculated by the comprehensive credibility index is device B , device A , device C , (the sum of weights is 1), substitute these values ​​into the improved Spearman rank correlation coefficient calculation formula The correlation coefficient is obtained from represents the weighted Spearman correlation coefficient, represents the rank difference of the i-th device, n represents the total number of devices involved in the calculation, in this example n=3, then , this value is 1, indicating that the credibility ranking is completely consistent with the SNR ranking. Then, the consistency between the ranking indicated by the correlation coefficient and the direction of change of the signal-to-noise ratio is analyzed. If If it is close to 1, it means a high positive correlation, that is, the higher the device reliability, the higher the SNR it reports. If it is close to -1, it means a high negative correlation. If it is close to 0, there is no obvious correlation. The weighted Spearman correlation coefficient is generated, that is, =1. formula middle, is the final calculated weighted Spearman correlation coefficient; the numbers 1 and 6 are the original constants of the Spearman formula; Indicates the sum of all n devices; is the weight factor of the i-th device, which comes from its comprehensive credibility and reflects the different contributions of devices with different credibility in calculating the correlation; is the square of the difference between the ranks of the i-th device in the two sequences to be compared (here, the credibility ranking and the SNR ranking), which is used to quantify the degree of inconsistency between the two rankings; n is the total number of devices (or data points) participating in the ranking; the denominator is a normalization factor related to the sample size. The operation logic is to first calculate the ranking (rank) of each data point in the two variables, and then calculate the difference between the two ranks. , and then calculate The square of and multiply by the corresponding weight , all data points Add up to get The larger the sum, the greater the difference in the ranking of the two variables. Multiply by 6 and divide by , get a standardized difference measure, and finally subtract this measure from 1 to get the weighted Spearman correlation coefficient , whose value range is -1 to +1. To show the calculation process more clearly, assume the following data:

[0119] Table 2: Device credibility and SNR data and rank calculation example

[0120] equipment Comprehensive credibility index Credibility rank (R1) <![CDATA[w i (Example)]]> SNR (dB) SNR rank (R2) <![CDATA[d i =R1 i −R2 i ]]> <![CDATA[d i 2 ]]> <![CDATA[w i d i 2 ]]> Dev1 0.85 1 0.4 12.0 1 0 0 0 Dev2 0.65 2 0.3 9.5 2 0 0 0 Dev3 0.40 4 0.1 7.0 4 0 0 0 Dev4 0.50 3 0.2 8.0 3 0 0 0

[0121] As shown in Table 2, assuming there are 4 devices, their credibility indicators and SNR data are shown in the table, and their respective ranks and weights have been calculated. (Here the weight Assumption is based on the credibility index in The intervals are normalized and adjusted to sum to 1). In this example, since the two ranks are exactly the same for simplicity, all ,therefore Then n=4, If the data are not completely consistent, for example, Dev3’s credibility rank is 3 and SNR rank is 4; Dev4’s credibility rank is 4 and SNR rank is 3, then:

[0122] Dev3: , , ;

[0123] Dev4: , , ; . The result The weighted Spearman correlation coefficient indicates that devices with higher comprehensive reliability tend to report higher signal-to-noise ratios. This weighted Spearman correlation coefficient will be used to adjust the gain of the subsequent Kalman filter. The benefit of the formula is that by introducing the device weight factor , so that devices with higher comprehensive credibility occupy a larger proportion when calculating correlation, and the consistency or inconsistency of their sorting has a greater impact on the final correlation coefficient, so that the correlation analysis results can better reflect the trend of reliable data.

[0124] The filter compensation submodule constructs the Kalman filter state transfer matrix and observation matrix based on the device operating voltage, iteratively calculates the prior estimated covariance matrix, adjusts the Kalman gain in combination with the weighted Spearman correlation coefficient, and generates a fused spectrum feature vector.

[0125] Kalman gain Weighted Spearman correlation coefficient satisfy Relationship, where represents the prior estimate covariance matrix, H is the observation matrix, R is the observation noise covariance matrix, the superscript T represents the matrix transpose operation, and the subscript k represents the current time;

[0126] The filter compensation submodule is based on the device operating voltage obtained from the voltage monitoring submodule, such as 3.76 volts. This voltage value can be used to evaluate the stability of the device state and indirectly affect the setting of the observation noise to construct the state transfer matrix of the Kalman filter. With the observation matrix , state transition matrix Describes how the spectrum feature vector (such as the signal strength at a specific frequency point) evolves from time k-1 to time k. If it is assumed that the change in signal strength is a random walk process, then It can be an identity matrix, or contain a small decay or growth factor, e.g. (if the state contains a value and a rate of change), or simply (if the state is just value), the observation matrix Describes how the state vector is mapped to the observation value. If the signal strength is directly observed, then , assuming that the state here is a single spectrum eigenvalue, and , iteratively calculate the prior estimated covariance matrix , and its calculation formula is ,in is the posterior estimated covariance at the previous moment, is the process noise covariance matrix, representing the model uncertainty, e.g. ,but , combined with the weighted Spearman correlation coefficient output by the correlation analysis submodule , such as the above calculation , adjust the Kalman gain , and the relationship is ,in is the currently calculated prior estimate covariance matrix (e.g. [0.06]), is the observation matrix (e.g. [1]), is the observation noise covariance matrix, the size of which can be adjusted according to the stability of the device operating voltage, for example, when the voltage is stable Smaller, such as , when the voltage is unstable is larger, the superscript T represents the matrix transpose operation, and the subscript k represents the current moment, then

[0127] ;

[0128] Kalman gain Used to update the state estimate: ,in is the prior state estimate, is the current observation value. Through this iterative filtering process, the spectrum data is smoothed and the noise is compensated to generate a fused spectrum feature vector. For example, if the prior estimate (arbitrary unit), current observation value , then the updated formula is: . formula middle, is the Kalman gain at the current moment k; is the weighted Spearman correlation coefficient introduced as a regulating factor for the standard Kalman gain; is the state covariance matrix of the prior estimate, which represents a measure of the uncertainty of the current state estimate; is the observation matrix, which maps the state space to the observation space; is the transpose of the observation matrix; is the covariance matrix of the observation noise, which represents the uncertainty of the observation value; It is the inverse of the observed residual covariance. The operation logic of the whole formula is: first calculate , which maps the covariance of the state prediction to the observation space and then adds the observation noise covariance , get the covariance of the observed residuals, take its inverse matrix, and then multiply it by , get the standard Kalman gain part, and finally multiply it by the weighted Spearman correlation coefficient ,this The introduction of makes the credibility ranking highly consistent with the physical quantity (such as SNR) ( Close to 1), the current observation is more trusted, and the gain is relatively large (close to the standard Kalman gain). If the consistency is low ( If the value is close to 0, the weight of the observation is reduced, the gain is reduced, and more reliance is placed on the model's prediction. Parameter assignment and calculation example (continued): (Scalar case) (Scalar case, the state is directly observed) (Scalar case, variance of observation noise) calculation .calculate .calculate . Calculate the standard Kalman gain part Finally, multiply by : The result This shows that, in the current state, the weight of the new observation in updating the state estimate is about 72.75% (after After adjustment). If =1, then =0.75, if =0.5, then =0.375, indicating that the weighted Spearman correlation coefficient effectively adjusts the system's confidence in new observations. This Kalman gain value will be used to update the estimated value of the spectral feature vector. The benefit of the formula is that by using the weighted Spearman correlation coefficient The introduction of Kalman gain in the calculation makes the filtering process not only rely on traditional model predictions and observation noise, but also considers the consistency of the current data and other related physical quantities (such as SNR) in credibility ranking. When the consistency is high, the current observation is considered to be more reliable and a larger gain is given. Otherwise, its influence is reduced, thereby improving the robustness and accuracy of the fused spectrum feature vector.

[0129] The focal element set is divided into three levels: high confidence, medium confidence, and low confidence based on the working status of the equipment. The working status of the equipment is jointly determined by voltage stability and signal quality coefficient.

[0130] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A radio spectrum measurement system based on mobile crowdsourcing, characterized in that: The system comprises: The crowdsourcing acquisition module is used to periodically scan radio spectrum signals through terminal devices to obtain timestamps, geographic coordinates, signal-to-noise ratios, and device operating voltages, perform structured encapsulation using a multi-source heterogeneous data encapsulation protocol, transmit the timestamps and geographic coordinates to the timing processing module, and output the signal-to-noise ratios and device operating voltages to the dynamic fusion module; A time series processing module is used to call a time series alignment algorithm to perform absolute difference calculation on the timestamps, generate time series weight parameters in ascending order, and pass the time series weight parameters and the geographic coordinates to a spatial optimization module; The timing processing module includes: The time series alignment submodule calls the time series alignment algorithm to perform time axis calibration and sequence length adjustment on the timestamps, identify the start and end point deviations of the differentiated time series, map multiple groups of time series to a unified time base, and generate an aligned time series sequence; The difference calculation submodule establishes a data window with a fixed step size based on the aligned time series, accumulates and sums the continuous differences in the window, calculates the arithmetic mean of the absolute differences in each window, and generates a window difference mean; The weight generation submodule inputs the window difference mean into a nonlinear transformation function, sorts the transformation results in ascending order, converts the accumulated value into a weight coefficient through linear transformation, and generates a time series weight parameter corresponding to the spatial dimension of the geographic coordinate; A spatial optimization module is configured to calculate the Euclidean distance of the target area using the geographic coordinates and the DBSCAN clustering algorithm to generate a normalized spatial weight coefficient, determine a weighted ratio between the temporal weight parameter and the spatial weight coefficient based on a regression model trained with historical data to generate a comprehensive credibility index, and output the comprehensive credibility index to a dynamic fusion module; a dynamic fusion module, configured to sort the comprehensive credibility indicators using an evidence theory decision model, perform correlation analysis between the sorting results and the signal-to-noise ratio, input the device operating voltage into a Kalman filter algorithm for compensation, and output a fused spectrum feature vector; The value of the process noise covariance matrix Q in the Kalman filter algorithm is determined based on the equipment voltage fluctuation variance.

2. The radio spectrum measurement system based on mobile crowdsourcing according to claim 1, characterized in that The multi-source heterogeneous data encapsulation protocol includes field mapping rules and CRC check mechanism; The time series alignment algorithm uses a dynamic time warping algorithm to achieve multi-device time axis calibration; Before calculating the Euclidean distance, the geographic coordinates are converted into the UTM coordinate system to achieve dimension unification.

3. The radio spectrum measurement system based on mobile crowdsourcing according to claim 2, characterized in that: The crowdsourcing collection module includes: The spectrum scanning submodule periodically scans radio spectrum signals through the terminal device, detects the carrier frequency and signal strength, records the timestamp of the scanning moment, synchronously obtains the geographic coordinates output by the device's built-in positioning system, and converts the electromagnetic wave signal strength continuously collected during the scanning period into a spectrum scanning value; The voltage monitoring submodule uses the raw voltage data output by the device power monitoring circuit, calculates the standard deviation of the voltage sampling sequence using a sliding window algorithm, eliminates transient interference pulses, and generates the device operating voltage. The signal analysis submodule separates the energy distribution of the signal frequency bands through fast Fourier transform based on the spectrum scan value, calculates the ratio of the main frequency band power to the noise frequency band power, generates a signal-to-noise ratio, performs piecewise linear interpolation operation on the signal-to-noise ratio and the device operating voltage based on the device power consumption curve. The device power consumption curve is constructed based on the voltage-power consumption correspondence table provided by the device manufacturer, and outputs a signal quality coefficient; The data encapsulation submodule calls the timestamp and the geographic coordinates, constructs a time series data frame according to a multi-source heterogeneous protocol, performs structured encoding on the signal quality coefficient and the device operating voltage according to a dynamic fusion protocol, generates a time series transmission frame, and outputs a fusion data packet.

4. The radio spectrum measurement system based on mobile crowdsourcing according to claim 3, characterized in that The electromagnetic wave signal strength conversion formula is: ； in, Represents the spectrum scan value, is the signal power spectral density at frequency f, and are the lower and upper limits of the scanning frequency band, respectively. df is a small frequency increment, which comes from the integral operation and is a standard notation in calculus. The sliding window length L is based on the device sampling rate according to Dynamic adjustment, where Indicates the device sampling rate, is the ceiling operator.

5. The radio spectrum measurement system based on mobile crowdsourcing according to claim 1, characterized in that: The unified time base uses GPS timing signals as synchronization source, and its time synchronization error is less than 1 microsecond; The fixed step length range is 5-10 sampling points and preferably takes a prime number length, which includes 5, 7, and 11; The nonlinear transformation function uses the sigmoid function to realize probability mapping, and its expression is: , where k is the curve steepness coefficient and its value is 0.5, The mean shift is the median of the mean of the window differences.

6. The radio spectrum measurement system based on mobile crowdsourcing according to claim 5, characterized in that: The space optimization module includes: The spatial clustering submodule detects the geographic coordinates, calculates the Euclidean distance between the coordinates using the DBSCAN clustering algorithm, divides high-density clusters and noise points according to density differences, calculates the sample distribution density within the cluster and the distance between adjacent clusters, determines the boundary range of the core point through density accessibility, and generates a spatial clustering density value; The weight normalization submodule calculates the density difference range between clusters based on the spatial clustering density value, maps the density value to the [0,1] interval through the maximum and minimum value normalization method, constructs a weight distribution function based on the mean distance between clusters, and generates a normalized weight coefficient; The credibility fusion submodule calls the time series weight parameter, performs weighted summation with the normalized weight coefficient, allocates the time and space dimension weight ratio, uses the standard deviation normalization method to eliminate the dimensional difference of the weighted result, and generates a comprehensive credibility index; The maximum and minimum values ​​are respectively taken from the density extremes of historical data in the same geographical area; The standard deviation calculation is based on the weighted sum dataset of all devices within the same time window.

7. The radio spectrum measurement system based on mobile crowdsourcing according to claim 1, characterized in that: The dynamic fusion module includes: The evidence decision submodule calls the comprehensive credibility index, defines the focal element set using the evidence theory decision model, calculates the confidence intervals and plausibility intervals of multiple indicators, merges conflicting evidences using the Dempster synthesis rule, and generates a credibility ranking sequence; The correlation analysis submodule calculates the sum of squares of rank differences based on the credibility ranking sequence and the signal-to-noise ratio data, substitutes the sum of squares into the Spearman formula to obtain the correlation coefficient, analyzes the consistency of the ranking and the direction of change of the signal-to-noise ratio, and generates a weighted Spearman correlation coefficient; The filter compensation submodule constructs a Kalman filter state transfer matrix and an observation matrix based on the operating voltage of the device, iteratively calculates a priori estimated covariance matrix, adjusts the Kalman gain in combination with the weighted Spearman correlation coefficient, and generates a fusion spectrum feature vector.

8. The radio spectrum measurement system based on mobile crowdsourcing according to claim 7, characterized in that: The focal element set is divided into three levels: high confidence, medium confidence, and low confidence based on the working status of the equipment. The working status of the equipment is determined by the voltage stability and signal quality coefficient. The device weight factor is introduced into the Spearman formula Perform weighted calculation, and the formula is improved to: ； in, represents the weighted Spearman correlation coefficient, The weight factor of the i-th device is calculated by the comprehensive credibility index. represents the rank difference of the i-th device, and n represents the total number of devices participating in the calculation.

Images