Parameter measurement method for FM monitoring antenna

Abstract

The present application relates to the field of signal parameter measurement technology, and provides a parameter measurement method for frequency modulation monitoring antennas. Through multi-dimensional data acquisition, intelligent analysis and dynamic optimization, adaptive monitoring and precise measurement of complex electromagnetic environments are achieved. The method deeply integrates environmental perception, signal processing and parameter optimization to form a complete technical chain, so that the parameter measurement method can have self-learning and self-adjustment capabilities, can respond to environmental changes and interference challenges in real time, significantly improve monitoring efficiency and measurement accuracy, and provide a systematic solution for the application of frequency modulation monitoring antennas in complex scenarios.

Description

Technical Field

[0001] The present application relates to the technical field of signal parameter measurement, and in particular to a parameter measurement method for a frequency modulation monitoring antenna. Background Art

[0002] In the field of frequency modulation monitoring, traditional antenna parameter measurement methods mainly rely on single-dimensional data collection and analysis. Early technologies manually set fixed monitoring frequencies and beam directions to collect signals in the electromagnetic environment, and used simple signal processing algorithms to analyze basic parameters such as signal strength and frequency. With the development of technology, sensor networks have been gradually introduced to realize the automatic collection of electromagnetic data, and algorithms such as adaptive filtering have been used to suppress some interference.

[0003] With the rise of machine learning in recent years, some solutions have attempted to use convolutional neural networks to perform preliminary pattern recognition on electromagnetic signals to distinguish common interference types. These solutions also employ optimization techniques such as genetic algorithms to adjust antenna scanning frequencies, improving monitoring efficiency to some extent. Regarding environmental factors, some research has also incorporated temperature and humidity data, using empirical formulas or simple linear models to correct for environmental influences on signal propagation. Furthermore, some solutions have begun to use spatial geometric models to describe the effects of reflectors on signals and employ ray tracing algorithms to simulate multipath propagation paths. However, most approaches have not addressed how to generate signals from multi-source heterogeneous data to adapt to complex monitoring environments or adaptively optimize monitoring parameters. Summary of the Invention

[0004] To address the shortcomings of the prior art, the present application provides a parameter measurement method for an FM monitoring antenna. The method comprises: acquiring interference data of a monitoring area in real time, the interference data including environmental electromagnetic data, temperature and humidity data, and reflector coordinates; performing pattern recognition on the interference data using a convolutional neural network to output interference features, the interference features including interference type and spatiotemporal intensity distribution; and constructing a multi-objective optimization function to process the interference features to generate a test signal.

[0005] The test signal is injected into the FM monitoring antenna, and the backscattered signal is received and monitored at the same time. The scattering parameters, energy distribution and phase delay of the backscattered signal are obtained in real time. The signal error between the test signal and the backscattered signal is comprehensively determined based on the scattering parameters, energy distribution and phase delay, and the backscattered signal is dynamically compensated according to the signal error.

[0006] The beam pointing and scanning frequency of the monitoring area are optimized according to the compensated backscattered signal. The interference data of the monitoring area are reacquired based on the optimized beam pointing and scanning frequency of the monitoring area. The interference characteristics are updated at the same time, and the parameter combination of the test signal is updated through the genetic algorithm.

[0007] As an optional implementation manner, the test signal generation strategy includes:

[0008] A generative adversarial network is constructed. The generator processes the interference features to output candidate parameter combinations of the test signal. The discriminator evaluates the candidate parameter combinations of the test signal through a multi-objective optimization function.

[0009] A physical constraint layer is configured according to the properties of the FM monitoring antenna, and candidate parameter combinations of the test signal are screened according to the physical constraint layer;

[0010] The candidate parameter combinations of the test signal are optimized by the simulated annealing algorithm. After multiple iterative training of the generator and the discriminator, the test signal is generated. The parameter combination of the test signal includes the frequency step amount, the modulation index and the duty cycle.

[0011] As an optional implementation manner, the output sub-strategy of the interference feature includes:

[0012] Obtain interference data in the monitoring area in real time, including environmental electromagnetic data, temperature and humidity data, and reflector coordinates, and align the interference data by timestamp;

[0013] The environmental electromagnetic data is converted into a time-frequency matrix through short-time Fourier transform, the temperature and humidity data are processed by Kriging interpolation to generate the temperature and humidity field distribution, and the reflector coordinates are constructed into a spatial topology through the octree algorithm;

[0014] A three-branch convolutional neural network consisting of an electromagnetic branch, an environmental branch, and a spatial branch is constructed. The three-branch convolutional neural network extracts the features of the interference data separately and performs weighted fusion through an attention mechanism to generate interference features including interference type and spatiotemporal intensity distribution.

[0015] The interference features are compared and analyzed with the historical feature data to determine whether to adjust the parameters and attention weights of the three-branch convolutional neural network.

[0016] As an optional implementation, the electromagnetic branch identifies the interference pattern of the time-frequency graph matrix through a residual network, the environmental branch extracts the gradient changes of the temperature and humidity field distribution through a two-dimensional convolutional layer, and the spatial branch analyzes the spatial topology and extracts the path complexity of multipath reflection through a graph convolutional network.

[0017] As an optional implementation, the sub-strategy for constructing the multi-objective optimization function includes:

[0018] Interference suppression, antenna matching, and propagation efficiency are used as optimization objectives, and the target weights of the optimization objectives are dynamically adjusted according to the interference type through a Bayesian network.

[0019] The multi-objective optimization function is constructed by taking the optimization objectives and objective weights into account through weighted summation.

[0020] As an optional implementation manner, the backscatter signal compensation strategy includes:

[0021] extracting the dominant component of the signal error and determining the compensation direction of the backscattered signal according to the dominant component of the signal error;

[0022] Based on the compensation direction of the backscattered signal, the parameter combination of the test signal is collaboratively adjusted to compensate for the backscattered signal;

[0023] The signal quality of the backscattered signal after compensation is monitored in real time. The signal quality includes the convergence of the scattering parameters and the energy concentration. The adjustment step size of the parameter combination of the test signal and the compensation direction of the backscattered signal are updated based on the signal quality.

[0024] As an optional implementation manner, the sub-strategy for determining the signal error between the test signal and the backscattered signal includes:

[0025] Inject the test signal into the FM monitoring antenna, receive the backscattered signal, and obtain the scattering parameters, energy distribution and phase delay of the backscattered signal in real time;

[0026] Performing cross-correlation analysis on the scattering parameters, energy distribution and phase delay of the backscattered signal to obtain a correlation matrix, and determining the error weights of the scattering parameters, energy distribution and phase delay according to the correlation matrix;

[0027] The signal error between the test signal and the backscattered signal is determined by weighted summation according to the error weight.

[0028] As an optional implementation method, the monitoring area is scanned by a field strength probe, and the backscattered signal is sampled in three dimensions. The energy distribution of the backscattered signal is obtained by obtaining the energy value of each sampling point. The carrier phase of the test signal and the backscattered signal is compared through coherent detection technology to obtain the phase delay of the backscattered signal.

[0029] As an optional implementation manner, the optimization strategy for beam pointing in the monitoring area includes:

[0030] The energy distribution and phase delay of the compensated backscattered signal are mapped to the direction of arrival space through Fourier transform to construct an energy density map. The energy concentration area is identified through the peak detection algorithm and marked as the target signal direction.

[0031] The propagation feasibility of beam pointing is evaluated by combining the temperature and humidity field distribution in the monitoring area and the spatial topology of the reflectors;

[0032] Through the reinforcement learning algorithm, the beam pointing of the monitoring area is dynamically optimized with the received power of the target signal and the interference energy as the reward function.

[0033] As an optional implementation, the optimization strategy for the scanning frequency of the monitoring area includes:

[0034] The scattering parameters of the compensated backscattered signal are divided into multiple frequency bands according to frequency. Combined with the monitoring area after beam pointing optimization, the signal attenuation and interference intensity distribution of each frequency band in the monitoring area are analyzed to identify the scanning frequency band and frequency step size.

[0035] Spectrum sensing technology is used to obtain the spectrum occupancy status of the monitoring area in real time. The spectrum change trend is determined based on the interference type, and risky and idle frequency bands are marked.

[0036] The constraints are determined by combining the scanning frequency band, frequency step size and spectrum change trend, and the scanning frequency of the monitoring area is optimized through genetic algorithm.

[0037] Compared with the existing technology, the beneficial effects of this application are: through multi-dimensional data acquisition, intelligent analysis and dynamic optimization, adaptive monitoring and precise measurement of complex electromagnetic environments are realized. This method deeply integrates environmental perception, signal processing and parameter optimization to form a complete technical chain, so that the parameter measurement method can have self-learning and self-adjustment capabilities, and can respond to environmental changes and interference challenges in real time, significantly improving monitoring efficiency and measurement accuracy, and providing a systematic solution for the application of FM monitoring antennas in complex scenarios.

[0038] Multi-source data fusion constructs a multi-dimensional feature space to comprehensively characterize the physical characteristics, spatial distribution and environmental correlation of the interference, providing a complete information basis for subsequent analysis. The deep feature extraction capability of the convolutional neural network automatically identifies complex interference patterns, avoiding the limitations of manual feature engineering, and improving the accuracy of interference type judgment and the positioning accuracy of spatiotemporal intensity distribution. The multi-objective optimization function makes the generated test signal more suitable for actual monitoring scenarios through dynamic weight adjustment.

[0039] The scattering parameters, energy distribution and phase delay of the backscattered signal are monitored synchronously to construct a complete signal characteristic vector, comprehensively evaluate the signal transmission status, avoid the one-sidedness of single parameter evaluation, and comprehensively judge the signal error between the test signal and the backscattered signal. The backscattered signal is dynamically compensated according to the signal error, effectively suppressing environmental interference and improving signal quality and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be derived from these drawings without inventive work. Among them:

[0041] Figure 1 A flow chart of a parameter measurement method for an FM monitoring antenna provided in an embodiment of the present application;

[0042] Figure 2 An output sub-strategy diagram of interference characteristics of a parameter measurement method for a frequency modulation monitoring antenna provided in an embodiment of the present application;

[0043] Figure 3 This is a diagram of an optimization strategy for beam pointing in a monitoring area of a parameter measurement method for an FM monitoring antenna provided in an embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to make the objectives, technical solutions and advantages of the embodiments of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application are clearly and completely described below in conjunction with the drawings in the specification. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0045] Example

[0046] like Figure 1 As shown, a method flow chart of a parameter measurement method for a frequency modulation monitoring antenna is provided in an embodiment of the present application, and the method includes:

[0047] S1. Obtain interference data of the monitoring area in real time. The interference data includes environmental electromagnetic data, temperature and humidity data, and reflector coordinates. Perform pattern recognition on the interference data through a convolutional neural network to output interference features. The interference features include interference type and spatiotemporal intensity distribution. A multi-objective optimization function is constructed to process the interference features to generate a test signal.

[0048] like Figure 2 As shown, the output sub-strategy of the interference feature includes:

[0049] Obtain interference data in the monitoring area in real time, including environmental electromagnetic data, temperature and humidity data, and reflector coordinates, and align the interference data by timestamp;

[0050] The environmental electromagnetic data is converted into a time-frequency matrix through short-time Fourier transform, the temperature and humidity data are processed by Kriging interpolation to generate the temperature and humidity field distribution, and the reflector coordinates are constructed into a spatial topology through the octree algorithm;

[0051] A three-branch convolutional neural network consisting of an electromagnetic branch, an environmental branch, and a spatial branch is constructed. The three-branch convolutional neural network extracts the features of the interference data separately and performs weighted fusion through an attention mechanism to generate interference features including interference type and spatiotemporal intensity distribution.

[0052] The interference features are compared and analyzed with the historical feature data to determine whether to adjust the parameters and attention weights of the three-branch convolutional neural network.

[0053] The environmental electromagnetic data, temperature and humidity data, and reflector coordinates in the monitoring area differ in data type, acquisition frequency, and format, and need to be processed uniformly to provide standardized input for subsequent feature extraction. At the same time, the impact of different data on signal propagation and interference is correlated in the time dimension, and timestamp alignment can accurately reflect the dynamic changes of interference characteristics. A spectrum analyzer is used to obtain environmental electromagnetic data and signal strength distribution in different frequency bands. Temperature and humidity data are obtained through temperature and humidity sensors, and a lidar is used to scan the monitoring area in real time to obtain three-dimensional coordinate data of reflectors. The timestamp matching algorithm is used to integrate the environmental electromagnetic data, temperature and humidity data, and reflector coordinates at the same time point into interference data. This achieves standardized processing and time synchronization of interference data, avoiding analysis deviations caused by data asynchrony. The unified data format and structure provide a good foundation for the subsequent efficient processing of the three-branch convolutional neural network, improving the accuracy and efficiency of interference feature extraction. The preprocessed and fused interference data are used as the input of the three-branch convolutional neural network, ensuring that the three-branch convolutional neural network can process data features of different dimensions in parallel, laying the foundation for accurate extraction of interference features.

[0054] Environmental electromagnetic data has time-varying characteristics, among which pulse interference manifests as sudden spikes in the time domain and has wide spectrum characteristics in the frequency domain. A single time domain or frequency domain analysis cannot fully capture the interference characteristics; the continuous environmental electromagnetic data is divided into multiple short time periods of fixed length, and a Hanning window function is applied to each time period to reduce spectrum leakage. A fast Fourier transform is performed on each windowed time period, and the time domain signal is converted into a frequency domain representation to obtain the spectrum distribution within the time period. The spectra of all time periods are arranged in chronological order to form a two-dimensional matrix, with the horizontal axis representing time and the vertical axis representing frequency. The matrix element value is the signal strength at the corresponding time and frequency point, thereby converting the environmental electromagnetic data into a visual time-frequency diagram matrix; the time-frequency diagram matrix intuitively displays the dynamic changing characteristics of electromagnetic interference. Preferably, narrowband interference is manifested as a continuous high-energy line at a fixed frequency in the time-frequency diagram, while pulse interference is manifested as a wide spectrum burst in a short time, which enables subsequent neural networks to more accurately identify different types of interference patterns.

[0055] Temperature and humidity sensors within a monitoring area are typically distributed discretely, and signal propagation is affected by temperature and humidity variations in continuous space. Therefore, interpolation methods are needed to convert discrete sensor data into a continuous temperature and humidity field distribution to quantify the impact of environmental factors on signal propagation. Multiple temperature and humidity sensors are deployed within the monitoring area at a certain grid density, and temperature and humidity values are regularly acquired at each point. The spatial correlation between the sensor data is analyzed, indicating that temperature and humidity values at points with close proximity are more similar, while correlation decreases with greater distance. Based on this spatial correlation, for any point within the monitoring area without a sensor installed, an estimated temperature and humidity value is calculated based on the known sensor data surrounding it and its spatial positional relationship. Through repeated iterative calculations, the temperature and humidity field distribution for the entire monitoring area is ultimately generated. The Kriging interpolation method fully accounts for the spatial correlation of temperature and humidity data. Compared with simple linear interpolation methods, it can more accurately restore the actual distribution of the temperature and humidity field, especially in areas with sparse sensor distribution, and can provide more reliable estimates.

[0056] Reflectors within the monitoring area are irregularly distributed and numerous, including buildings and metal structures. Data structures are needed to represent their spatial relationships for subsequent analysis of the multipath propagation effects of signals. The octree recursively partitions the three-dimensional space, simplifying complex spatial scenes into a hierarchical structure. This facilitates rapid querying and analysis of the topological relationships between reflectors. The entire monitoring area is defined as an initial cube space, which is then recursively divided into eight equal sub-cubes until the number of reflectors contained in each sub-cube falls below a preset threshold. Each cube node contains a pointer to the sub-cube and a list of reflectors within the cube. In this way, the reflectors in the three-dimensional space are organized into a hierarchical tree structure, with each node representing a specific spatial region. Based on the octree structure, the connectivity between adjacent spatial regions and the reflector distribution density are analyzed, and characteristic parameters reflecting spatial topological characteristics, such as reflector clustering and spatial connectivity, are extracted. The octree structure greatly improves the efficiency of spatial query and analysis, enabling rapid location and analysis of the spatial characteristics of specific regions when processing large amounts of reflector data. This hierarchical structure also facilitates the subsequent extraction of features such as the path complexity of multipath reflections by graph convolutional networks.

[0057] Environmental electromagnetic data, temperature and humidity data, and reflector coordinates contain different information characteristics, making it difficult for a single network structure to effectively extract their key features. A three-branch convolutional neural network uses different network architectures to extract features based on the data characteristics. Combined with the attention mechanism, it can further highlight important features and achieve effective fusion of multi-dimensional features. A three-branch convolutional neural network is constructed, consisting of an electromagnetic branch, an environmental branch, and a spatial branch. The electromagnetic branch performs multi-layer convolution operations on the time-frequency map matrix through a residual network. Residual connections are used to solve the gradient vanishing problem caused by increased network depth and extract interference patterns in the time-frequency map matrix, including spectral peaks of narrowband interference and continuous spectral distribution of broadband interference. The environmental branch processes the temperature and humidity field distribution through a two-dimensional convolution layer and extracts the gradient variation characteristics of the temperature and humidity field distribution through the convolution kernel, capturing the spatial variation trends of temperature and humidity and quantifying the impact of environmental factors on signal propagation. The spatial branch uses the spatial topological structure of the reflector as input through a graph convolutional network. Through information transmission and aggregation between nodes, it analyzes the connection relationship between reflectors and extracts the path complexity of multipath reflection.

[0058] The features extracted by the three branches are used as input. A multi-head attention mechanism is used to automatically assign weights to different features based on their importance to interference identification. The weighted fusion generates interference features that include interference type and spatiotemporal intensity distribution. The three-branch convolutional neural network extracts features based on different data characteristics, fully leveraging the advantages of each network architecture. Compared with a single network structure, it can extract interference features more comprehensively and accurately. The features output by the three branches are spliced into a total feature. The total feature is processed by a fully connected network to output the weight values of the three branch networks. The output weight values are normalized to ensure that the sum of all weights is 1. The features of each branch are multiplied by the corresponding weight value and then summed to obtain the final interference feature. The fused interference feature is then mapped to the interference type and spatiotemporal intensity distribution through a fully connected layer. The interference type is the interference pattern extracted from the electromagnetic branch, and the spatiotemporal intensity distribution includes the time series intensity and spatial region intensity. The attention mechanism further enhances the weights of key features, effectively suppressing redundant information and improving the quality and recognition of the interference feature. The generated interference feature contains rich interference information and serves as the input to the multi-objective optimization function, providing a key basis for constructing the objective function and generating test signal parameters.

[0059] The interference environment in the monitoring area is dynamically changing, including the movement of interference sources and fluctuations in ambient temperature and humidity. Fixed network parameters cannot adapt to these changes. By comparing the current interference features with historical feature data and dynamically adjusting the parameters and attention weights of the three-branch convolutional neural network, the three-branch convolutional neural network can continuously maintain its ability to accurately extract interference features. The currently generated interference features are compared with the historical feature data, and the similarity between the features is calculated using the cosine similarity algorithm. A similarity threshold is set. If the average similarity between the current interference features and the historical feature data is less than the similarity threshold, it indicates that the interference environment has changed significantly. At this time, the backpropagation algorithm is triggered. Using the current interference data and the expected output as training samples, the parameters of the three-branch convolutional neural network are fine-tuned. At the same time, the weight parameters of the attention mechanism are updated to enhance the responsiveness to newly emerging interference features. This achieves the adaptive adjustment of the three-branch convolutional neural network to the dynamic interference environment, can timely capture changes in interference features, avoid feature extraction failure caused by environmental changes, and ensure the accuracy and stability of the interference feature output. The calibrated interference features better reflect the current actual interference situation, provide more accurate data support for the subsequent construction of multi-objective optimization functions, and help generate more effective test signal parameters.

[0060] Furthermore, the electromagnetic branch identifies the interference pattern of the time-frequency graph matrix through a residual network, the environmental branch extracts the gradient changes of the temperature and humidity field distribution through a two-dimensional convolutional layer, and the spatial branch analyzes the spatial topology and extracts the path complexity of multipath reflection through a graph convolutional network.

[0061] The time-frequency map matrix contains a large amount of distribution information of electromagnetic signals in the time and frequency dimensions. Traditional convolutional networks find it difficult to effectively extract deep features when the number of layers is increased, and the gradient vanishing problem is prone to occur. The residual network, through the residual connection structure, can mine complex interference patterns without losing information. A residual network is constructed here to identify the time-frequency map matrix, and a convolution layer is set at the front end of the residual network for preliminary feature extraction, wherein each residual block includes a convolution layer and a jump connection structure for learning the detailed features of the interference pattern, and captures local features in the time-frequency map matrix through convolution operations, wherein the local features include spectral spikes and time domain pulses, and uses jump connections to fuse shallow features with deep features, thereby enhancing the residual network's ability to express interference patterns. Preferably, for narrowband interference, the residual network can learn the continuous signal strength changes at a fixed frequency through the residual block, and for pulse interference, it can identify the burst energy characteristics in the time domain.

[0062] At the end of the residual network, a global average pooling layer is used to compress the local features of the time-frequency map matrix into a one-dimensional vector, and then the classification results of the interference pattern are output through the fully connected layer. The classification results of the interference pattern include narrowband interference, broadband interference, and pulse interference. Compared with ordinary convolutional networks, residual networks can extract interference features at a deeper level, avoid gradient vanishing, and improve the recognition accuracy of complex time-varying interference. At the same time, they can automatically learn the feature combinations of different interference patterns without the need for manual design of feature extraction rules. The output interference pattern serves as the core part of the interference feature and provides key input for multi-objective optimization functions, such as guiding the selection of test signal frequencies to avoid interference frequency bands.

[0063] The gradient change of the temperature and humidity field distribution directly affects the signal propagation path and quality. The signal refraction caused by the temperature gradient and the signal attenuation caused by the humidity difference require a two-dimensional convolution layer to capture the temperature and humidity change trend in the spatial dimension and quantify the impact of environmental factors on the signal. The temperature and humidity field distribution generated by Kriging interpolation is sliced into multiple two-dimensional matrices according to the height dimension, where each matrix represents the temperature and humidity distribution of a certain height plane. As the input of the two-dimensional convolution layer, the receptive field is gradually expanded through convolution kernels of different sizes, and the local gradient change characteristics of the temperature and humidity field are extracted through the convolution layer. The local gradient change characteristics include the temperature gradient direction and the humidity mutation area. The activation function is used to enhance the nonlinear expression of the features. The output of each convolution layer is channel-spliced, and the most significant gradient features are extracted through the global maximum pooling layer. The output is the feature vector of the temperature and humidity field gradient, which includes parameters such as the temperature gradient intensity and the humidity change rate. The two-dimensional convolution layer can efficiently capture the spatial change pattern of the temperature and humidity field, convert the impact of environmental factors on the signal into a quantifiable feature vector, and improve the efficiency of feature extraction.

[0064] The spatial topology of the reflector determines the multipath propagation path of the signal. Traditional methods have difficulty handling irregular spatial relationships. Graph convolutional networks can effectively analyze the connection relationship between reflectors and extract the path complexity of multipath reflections through information transmission between nodes. Path complexity includes the number of paths and the number of reflections. The spatial topology of the reflector generated by the octree algorithm is converted into a graph structure. Each reflector is used as a node, and the node features include position coordinates and reflection coefficients. The edges between nodes represent the spatial proximity relationship between reflectors. The edge weights are set according to the distance and signal reflection probability. The node status is updated by aggregating the feature information of adjacent nodes. Through multi-layer graph convolution operations, node features gradually integrate global spatial information. A fully connected layer is set at the end of the graph convolution network to map node features to the path complexity of multipath reflection, including the average path length, the distribution of reflection times, and the direction of the strongest reflection path. The graph convolution network breaks through the limitations of traditional grid data and can accurately analyze the multipath effect in irregular spatial topology to reduce the path complexity evaluation error. The output multipath reflection path complexity is used to optimize the parameters of the test signal. Preferably, the modulation index is adjusted to reduce the inter-symbol interference caused by multipath interference, and the beam pointing optimization is assisted to avoid the direction of the strong reflection path.

[0065] The sub-strategies for constructing multi-objective optimization functions include:

[0066] Interference suppression, antenna matching, and propagation efficiency are used as optimization objectives, and the target weights of the optimization objectives are dynamically adjusted according to the interference type through a Bayesian network.

[0067] The multi-objective optimization function is constructed by taking the optimization objectives and objective weights into account through weighted summation.

[0068] During FM monitoring, interference signals can affect measurement accuracy. The degree of match between the test signal and the antenna determines signal transmission efficiency, and complex environmental factors can affect signal propagation. Therefore, optimization objectives must be set across three dimensions: interference suppression, antenna matching, and propagation efficiency to improve overall monitoring performance. Interference suppression aims to reduce the impact of interference signals on the test signal, minimize spectral overlap between the interference signal and the test signal, and improve the signal-to-noise ratio (SNR) of the received signal. Antenna matching ensures that parameters such as the test signal's frequency and polarization match the frequency response and polarization characteristics of the FM monitoring antenna, maximizing the antenna's reception efficiency for the test signal. Propagation efficiency considers environmental factors such as temperature, humidity, and reflectors in the monitoring area, optimizing the test signal's parameters to minimize attenuation and multipath effects during signal propagation and improve the integrity and accuracy of the signal at the receiving end. This clarifies the direction and focus of optimization and provides a clear framework for subsequent weight allocation and objective function construction, enabling test signal generation to comprehensively consider multiple factors. Based on these three core objectives, the Bayesian network guides weight allocation, which in turn influences the construction of the multi-objective optimization function and the test signal parameter generation strategy.

[0069] Different types of interference affect parameter monitoring in different ways and degrees. Narrowband interference mainly affects the spectral purity of the signal, while broadband pulse interference will damage the time domain characteristics of the signal. Therefore, it is necessary to dynamically adjust the weights of each optimization target according to the interference type to achieve more effective optimization; a mapping relationship table between interference type and target weight is pre-constructed. Preferably, for narrowband interference, the target weight of interference suppression is set to 0.6, the target weight of antenna matching is set to 0.3, and the target weight of propagation efficiency is set to 0.1. For broadband pulse interference, the target weights are adjusted to 0.3, 0.2 and 0.5 respectively. The Bayesian inference algorithm is used according to the real-time interference characteristics and combined with historical Data and prior probability are used to dynamically calculate and update the target weights of each optimization target. Preferably, when it is detected that the interference type changes from narrowband interference to broadband pulse interference, the Bayesian network automatically adjusts the target weights and enhances the target weights of propagation efficiency to cope with problems such as multipath effects caused by broadband pulse interference; thereby, the multi-objective optimization function can adaptively adjust the optimization focus according to changes in the interference type. Compared with the fixed weight scheme, it can more accurately optimize different interference scenarios. The dynamically adjusted target weights directly participate in the calculation of the multi-objective optimization function, determine the importance of each optimization target in the multi-objective optimization function, and thus affect the generation direction of the parameter combination of the test signal.

[0070] It is necessary to combine the dynamic weight with the core optimization goal to form a quantitative evaluation standard to judge the advantages and disadvantages of the parameter combination of different test signals, so as to guide the generation of test signals; a multi-objective optimization function is constructed by weighted summation, and the three goals of interference suppression, antenna matching and propagation efficiency are multiplied by the corresponding target weights respectively, and then added to obtain the objective function value. At the same time, physical constraints are built into the multi-objective optimization function. According to the hardware characteristics of the FM monitoring antenna such as the working frequency band, bandwidth and maximum transmission power, as well as the physical laws of electromagnetic propagation, that is, the relationship between signal frequency and wavelength, the value range of the test signal parameters is limited. Preferably, it is stipulated that The frequency step must be within the antenna operating frequency band and meet certain frequency resolution requirements, while the modulation index cannot exceed the maximum nonlinear distortion range that the antenna can handle. The constructed multi-objective optimization function can comprehensively consider multiple optimization objectives and physical constraints, and intuitively evaluate the pros and cons of parameter combinations of different test signals through quantified objective function values, providing clear optimization directions and screening criteria for the generation of test signals. The multi-objective optimization function serves as the basis for the discriminator to evaluate candidate parameter combinations of test signals. During the test signal generation process, it guides the generator and discriminator to perform iterative training and parameter optimization until a parameter combination of the test signal that meets the requirements is generated.

[0071] The test signal generation strategy includes:

[0072] A generative adversarial network is constructed. The generator processes the interference features to output candidate parameter combinations of the test signal. The discriminator evaluates the candidate parameter combinations of the test signal through a multi-objective optimization function.

[0073] A physical constraint layer is configured according to the properties of the FM monitoring antenna, and candidate parameter combinations of the test signal are screened according to the physical constraint layer;

[0074] The candidate parameter combinations of the test signal are optimized by the simulated annealing algorithm. After multiple iterative training of the generator and the discriminator, the test signal is generated. The parameter combination of the test signal includes the frequency step amount, the modulation index and the duty cycle.

[0075] There are many parameter combinations for test signals. Traditional exhaustive or random search methods are inefficient and difficult to find the optimal solution. Generative adversarial networks have powerful generation and learning capabilities and can quickly search for potential high-quality parameter combinations in high-dimensional parameter space. A generative adversarial network is constructed. The generative adversarial network includes a generator and a discriminator. A multi-layer fully connected neural network is designed as the generator. The generator generates candidate parameter combinations, namely frequency step size, modulation index and duty cycle, by learning the mapping relationship between interference characteristics and parameter combinations of test signals. The discriminator also uses a multi-layer fully connected neural network. The discriminator compares the candidate parameter combinations output by the generator with the actual parameter combinations according to a multi-objective optimization function for evaluation, and outputs a probability value to characterize the degree to which the candidate parameter combination is close to the optimal solution.

[0076] The generator and discriminator are optimized through adversarial training. The generator attempts to generate parameter combinations that can deceive the discriminator, while the discriminator continuously improves its ability to distinguish the generated parameter combinations. During the training process, the generator adjusts its own parameters according to the feedback from the discriminator, and gradually generates better candidate parameter combinations. Through the powerful generation ability of the generative adversarial network, a large number of high-quality candidate parameter combinations of test signals can be generated in a short time. Compared with traditional search methods, the efficiency and coverage of parameter search are greatly improved, and rich candidate resources are provided for subsequent screening of the optimal parameter combination. The generated candidate parameter combinations are used as preliminary results and enter the physical constraint layer for screening. Combinations that do not meet the actual physical conditions are eliminated, and feasible parameter combinations are retained for the next step of optimization.

[0077] The candidate parameter combinations generated by the generator may not conform to the physical characteristics and electromagnetic propagation laws of the FM monitoring antenna, including the frequency exceeding the antenna's operating frequency band and the modulation index being too large, resulting in signal distortion. Therefore, it is necessary to screen the candidate parameter combinations through the physical constraint layer to ensure that the generated test signal can effectively operate in the actual device; according to the properties of the FM monitoring antenna, including technical parameters and the basic principles of electromagnetic propagation, physical constraints are set, where the technical parameters include the operating frequency band, bandwidth and maximum modulation index settings, and the basic principles of electromagnetic propagation include the relationship between signal frequency and wavelength, and the restrictions on duty cycle and pulse repetition frequency; the candidate parameter combinations output by the generator are compared with the physical constraints one by one For frequency steps, check whether they are within the antenna operating frequency band. For modulation index, determine whether it exceeds the maximum nonlinear distortion range that the antenna can withstand. For duty cycle, ensure that it meets the timing requirements of signal transmission. Eliminate parameter combinations that do not meet any constraints, and retain candidate parameter combinations that meet all physical constraints for optimization. Through physical constraint layer filtering, infeasible parameter combinations are eliminated to ensure that the parameter combinations used in subsequent optimization processes are feasible in actual applications, reduce the risk of invalid calculations and engineering implementation in test signal generation, and use the simulated annealing algorithm to perform global optimization as a valid solution to obtain the optimal parameter combination for the test signal.

[0078] Although the candidate parameter combination screened by the physical constraint layer is feasible, it is not necessarily the optimal solution. The simulated annealing algorithm has a global search capability and can search for the globally optimal or approximately globally optimal parameter combination in the feasible solution space, thereby generating the test signal with the best performance. The screened candidate parameter combination is used as the initial solution of the simulated annealing algorithm, and the algorithm parameters such as the initial temperature, temperature drop rate and number of iterations are set. In each iteration, a neighborhood solution is generated based on the current solution. The neighborhood solution can be obtained by making small random adjustments to the parameters. The objective function values of the current solution and the neighborhood solution are calculated. If the objective function value of the neighborhood solution is better, the neighborhood solution is accepted as the new current solution. If the neighborhood solution is worse, the neighborhood solution is accepted with a certain probability, and the probability decreases as the temperature decreases. , through continuous iteration, gradually lowering the temperature, so that the algorithm converges to the global optimal or approximate global optimal solution; when the preset number of iterations is reached or the temperature drops to a certain level, the algorithm stops, and the final optimal solution is used as the parameter combination of the test signal to generate the final test signal; the simulated annealing algorithm can perform global search in a complex parameter space, effectively avoiding falling into the local optimal solution. Compared with other local search algorithms, it can find a better test signal parameter combination, significantly improve the performance of the test signal, and thus improve the parameter measurement accuracy and reliability of the FM monitoring antenna. The generated test signal is used to inject into the FM monitoring antenna to drive subsequent backscattered signal reception, signal error judgment and dynamic compensation processes. It is the key input signal of the entire FM monitoring method.

[0079] S2. Inject the test signal into the FM monitoring antenna, and simultaneously receive and monitor the backscattered signal, obtain the scattering parameters, energy distribution and phase delay of the backscattered signal in real time, and comprehensively judge the signal error between the test signal and the backscattered signal based on the scattering parameters, energy distribution and phase delay, and dynamically compensate the backscattered signal according to the signal error.

[0080] The sub-strategies for determining the signal error between the test signal and the backscattered signal include:

[0081] Inject the test signal into the FM monitoring antenna, receive the backscattered signal, and obtain the scattering parameters, energy distribution and phase delay of the backscattered signal in real time;

[0082] Performing cross-correlation analysis on the scattering parameters, energy distribution and phase delay of the backscattered signal to obtain a correlation matrix, and determining the error weights of the scattering parameters, energy distribution and phase delay according to the correlation matrix;

[0083] The signal error between the test signal and the backscattered signal is determined by weighted summation according to the error weight.

[0084] Based on the field intensity probe scanning the monitoring area, the backscattered signal is sampled in three dimensions. The energy distribution of the backscattered signal is obtained by obtaining the energy value of each sampling point. The carrier phase of the test signal and the backscattered signal is compared through coherent detection technology to obtain the phase delay of the backscattered signal.

[0085] After the test signal is injected into the antenna, the core characteristic parameters of the backscattered signal need to be obtained to evaluate the distortion during signal transmission and reflection. The scattering parameters, energy distribution, and phase delay respectively reflect the reflection loss, spatial distribution, and time delay characteristics of the signal, and are the key basis for judging signal errors. The vector network analyzer is used to scan different frequency bands in segments to obtain the scattering parameters of the backscattered signal, covering the main operating frequency bands of the FM monitoring antenna. The attenuation value of the analyzer is automatically adjusted according to the signal strength to ensure that both weak and strong signals can be accurately acquired. The time domain gate calibration is triggered every 100 sets of data to eliminate the system errors caused by transmission lines and connectors.

[0086] The energy distribution of the backscattered signal reflects the propagation and reflection characteristics of the signal in space. Through three-dimensional spatial sampling, the interference source can be located and the multipath reflection path can be identified, providing a basis for beam optimization. The detection units are arranged in the field strength probe array to cover the monitoring space. A distributed architecture is adopted. Before each scan, a standard signal is transmitted through the built-in calibration source to calibrate the amplitude and phase consistency of all probes. The monitoring area is divided along the height direction, and a spiral scanning path is adopted. Starting from the center of the bottom layer, it expands outward in the form of an Archimedean spiral. After completing one layer, it rises to the next layer. In the area where the signal fluctuates violently, the sampling points are automatically added to reduce the spacing. The area where the signal fluctuates violently is close to the metal reflector, and the standard spacing is maintained in the signal stable area. All probes are synchronized with the GPS clock to ensure that the energy data obtained at the same time corresponds to the same signal state.

[0087] The raw data of each sampling point is digitally down-converted to convert the RF signal into a baseband signal. Square-law detection is then performed to obtain the instantaneous power. A fixed-length sliding window is used to perform average filtering on the power data to eliminate the influence of random noise and improve the stability of energy measurement. Kriging interpolation is used to fill data in sparse sampling areas to generate a continuous energy distribution. This enables energy sampling in the monitoring area, clearly capturing the energy distribution formed by multipath reflections, and detecting potential interference sources in advance. The obtained energy distribution is used to construct an energy density map, providing a spatial distribution basis for beam pointing optimization.

[0088] The phase delay of the backscattered signal contains information such as the length of the signal propagation path and the number of reflections. By accurately measuring the phase delay, the path complexity of the multipath reflection can be calculated to assist in locating the position of the reflector. The carrier synchronization of the test signal and the received signal is achieved through the phase-locked loop circuit, and the received signal is mixed with the locally generated carrier of the same frequency and phase to obtain a zero intermediate frequency signal. Then, the phase delay is directly obtained by measuring the phase of the zero intermediate frequency signal. Parallel coherent detection channels are set up, and the phase of the reflected signal in different directions in each coherent detection channel is measured at the same time. The zero intermediate frequency signal is orthogonally sampled to obtain I / Q two-way data, and the phase is calculated by the inverse tangent function. The phase ambiguity problem is solved by the multi-phase phase-locked loop algorithm. By comparing the phase changes of adjacent sampling points, the phase value is automatically adjusted to make it change continuously. The phase value is processed by Kalman filtering, and the historical measurement value and the current observation value are integrated to suppress random noise interference and improve the phase measurement accuracy. An auxiliary frequency is superimposed on the test signal, and the phase delay of each frequency point is measured respectively. The phase delay of different frequency points is derived to obtain the phase gradient, which is used to judge the refraction and diffraction phenomena in the signal propagation path; thus, high-precision phase delay measurement is achieved. The multi-frequency point analysis capability can effectively identify phase distortion in non-line-of-sight propagation scenarios. The obtained phase delay is used for spatial Fourier transform to convert the time domain phase information into wave arrival direction information, providing an angle domain basis for beam optimization.

[0089] Multi-dimensional parameter acquisition comprehensively covers the key characteristics of signal transmission, avoiding errors and misjudgments caused by single parameter analysis, and providing a complete data basis for subsequent error evaluation. The acquired parameters are directly used as input for cross-correlation analysis, improving the accuracy of error weight calculation, and thus optimizing the comprehensive error judgment results.

[0090] Scattering parameters, energy distribution and phase delay affect each other during signal transmission, that is, reflection loss will change the energy distribution, and multipath effect will cause phase delay and energy dispersion at the same time. Single parameter analysis cannot reflect the essential cause of signal distortion. Through cross-correlation analysis, the potential causal relationship between parameters can be explored, providing a basis for accurate assessment of error weights; the acquired scattering parameters are stored in frequency segments, and the energy distribution is constructed as a spatial grid matrix, where the x, y, and z coordinates correspond to the energy values, and the phase delay is arranged into vectors according to the frequency points. The three types of data are normalized to eliminate dimensional differences, and the energy distribution and phase delay are processed through a time-space sliding window. Preferably, the time window is 500ms and the space window covers 10% of the monitoring area. The correlation coefficient between the energy center of gravity offset and the phase delay change in the time-space sliding window is calculated to identify the correspondence between the energy concentration area and the phase anomaly.

[0091] By comparing the frequency response curves of the test signal and the backscattered signal, the impact of reflection loss on phase linearity is analyzed, where the reflection loss is also the scattering parameter. That is, if a frequency band is too high and causes a phase mutation, it is marked as a strongly correlated frequency band. The Granger causality test algorithm is used to determine the causal directionality between the parameters. If the energy distribution change always precedes the phase delay fluctuation in the time series, it is determined that the energy distribution has a causal effect on the phase delay. The above analysis results are integrated into a correlation matrix. The matrix element value represents the correlation strength between any two parameters in a specific scenario. The correlation strength ranges from 0 to 1. Preferably, when there is a large metal reflector in the monitoring area, the correlation strength between the energy distribution and the scattering parameter can reach 0.8. Through multi-dimensional correlation analysis, the complex causal chain of signal distortion is revealed, avoiding misjudgment caused by parameter fragmentation analysis. Compared with traditional single parameter evaluation, potential signal problems can be identified in advance. The quantized correlation matrix serves as the core input for determining the error weight and directly affects the accuracy of weight distribution.

[0092] In different interference scenarios, the contribution of scattering parameters, energy distribution and phase delay to signal error is different. Multipath interference focuses on energy distribution and phase delay, while impedance mismatch focuses on scattering parameters. Dynamically determining error weights can make the comprehensive error calculation more in line with the actual situation and avoid evaluation bias caused by fixed weights. Initial weights are set for the three types of parameters based on historical data. Preferably, scattering parameters 0.4, energy distribution 0.3 and phase delay 0.3 are used as the basis for adaptive adjustment. The correlation matrix and the current error state are input through a recursive neural network to output the adjusted error weight. The recursive neural network continuously optimizes the weight mapping relationship through online learning and sets the weight adjustment range to prevent excessive weight fluctuations. When the correlation matrix is abnormal, preferably when the correlation strength of a parameter suddenly drops below 0.2, the Bayesian reasoning mechanism is triggered, and the error weight is corrected in combination with prior knowledge. After every 5 adjustments of the error weight are completed, the error assessment accuracy before and after the adjustment is compared through simulation. If the error judgment accuracy drops by more than 5%, the correlation analysis steps are traced back to check whether there are abnormalities in the data or algorithm. The adaptive weight mechanism enables the error assessment to quickly adapt to environmental changes, that is, the movement of interference sources and sudden changes in weather, so as to improve the judgment accuracy of the comprehensive error. The determined error weight is directly used for the comprehensive error calculation to guide the selection of the compensation strategy for the backscattered signal, that is, parameters with high error weights are compensated first.

[0093] The error weight of each parameter is combined with the actual parameter deviation, and the overall signal error between the test signal and the backscattered signal is quantified through weighted summation, providing a clear quantitative indicator for subsequent compensation; the deviation between the actual value and the ideal value of each parameter is calculated, and Z-score normalization is performed to eliminate the dimension effect. The normalized parameter deviation is multiplied by the corresponding weight and added together to obtain the signal error between the test signal and the backscattered signal. The 95% confidence interval of the comprehensive error is calculated. If the error value exceeds this interval, it is marked as an abnormal error, triggering a more refined compensation strategy; the unified quantitative evaluation standard makes the signal errors in different scenarios comparable, which facilitates the rapid location of signal quality problems. Standardization and confidence interval evaluation enhance the robustness of error judgment and reduce the interference of accidental factors. The comprehensive error serves as the core basis for compensation decision-making, triggering the dynamic compensation process of the backscattered signal. Its accuracy directly affects the effectiveness of the compensation strategy.

[0094] Compensation strategies for backscatter signals include:

[0095] extracting the dominant component of the signal error and determining the compensation direction of the backscattered signal according to the dominant component of the signal error;

[0096] Based on the compensation direction of the backscattered signal, collaboratively adjust the parameter combination of the test signal to compensate for the backscattered signal;

[0097] The signal quality of the backscattered signal after compensation is monitored in real time. The signal quality includes the convergence of the scattering parameters and the energy concentration. The adjustment step size of the parameter combination of the test signal and the compensation direction of the backscattered signal are updated based on the signal quality.

[0098] Signal errors can be caused by a variety of factors, such as phase delay, uneven energy distribution, and scattering loss. It is necessary to identify the dominant error source and select a targeted compensation strategy to avoid blind adjustments. Perform principal component analysis on the comprehensive error, project the signal error to different eigenvector directions, and calculate the error contribution in each direction. For example, if the error contribution in the phase delay direction exceeds 50%, the phase error is determined to be the dominant component. If the error caused by uneven energy distribution accounts for the highest proportion, energy optimization is taken as the priority direction. A decision tree with three layers of decision nodes is established. The first layer determines the dominant error type, including scattering parameters, energy distribution, and phase delay. The second layer analyzes the error source, including multipath effects, antenna mismatch, and environment. Interference, the third layer outputs the compensation direction. If the dominant component of the signal error is phase delay and is caused by multipath reflection, the compensation direction is phase correction and beamforming. If the uneven energy distribution is caused by deviation in the antenna radiation direction, the antenna pointing is adjusted and the signal power is optimized. For situations where the signal error components account for a similar proportion, that is, a phase error of 45% and an energy error of 40%, a fuzzy logic algorithm is used to comprehensively consider historical compensation data and current environmental parameters and dynamically adjust the priority of the compensation direction. This allows the root cause of the error to be targeted, the matching degree of the compensation strategy is improved, the number of invalid adjustments is reduced, and compensation efficiency is improved. The clear compensation direction directly guides the test signal parameter adjustment strategy and narrows the parameter search space.

[0099] The parameter combinations of the test signals are interrelated and affect the signal quality. It is necessary to coordinately adjust multiple parameters according to the compensation direction to achieve a comprehensive compensation effect; pre-define parameter adjustment rules corresponding to different compensation directions. For example, if the dominant error is phase delay, give priority to increasing the modulation index to optimize phase linearity, and at the same time fine-tune the frequency step to avoid the phase-sensitive frequency band. If the phase is advanced, reduce the carrier frequency, and if it is lagging, increase the carrier frequency; in the energy compensation scenario, if the energy is unevenly distributed, adjust the phase weighting coefficient of the antenna array to change the beam pointing, and simultaneously increase the duty cycle to enhance the signal energy in the target area. If the energy concentration area is offset, re-plan the scanning path; each After adjusting the parameters for the first time, a feasibility check is performed according to the above-mentioned physical constraint layer. If the parameters are out of range, the adjustment range is proportionally rolled back or the adjustment strategy is changed. The particle swarm optimization algorithm is used to search for the optimal combination of multiple parameters in parallel, and the parameter adjustment range is divided into multiple subspaces. Each particle searches independently, and the optimal solution is exchanged regularly to accelerate the convergence speed. The coordinated adjustment of multiple parameters avoids the degradation of other performance caused by the optimization of a single parameter, and achieves a comprehensive improvement in signal quality. The measured data shows that compared with the adjustment of a single parameter, the comprehensive error is reduced. The adjusted test signal is used for a new round of signal acquisition, and its parameter combination directly affects the monitoring results of the signal quality after compensation.

[0100] A single compensation cannot completely eliminate the error, and environmental changes will also cause the compensation effect to attenuate. It is necessary to monitor the signal quality in real time and dynamically adjust the compensation strategy and parameter step size; continuously monitor the scattering parameters, calculate the Euclidean distance between two adjacent measurements, and if the Euclidean distance is less than the set distance threshold, the scattering parameters are considered to have converged. The energy concentration is evaluated by calculating the entropy value of the energy distribution. The lower the entropy value, the more concentrated the energy. The set entropy value threshold is used to determine whether the energy distribution meets the standard. At the same time, the standard deviation of the phase delay is monitored to determine whether the phase is stable. The adjustment step size of the parameter combination is dynamically adjusted according to the change in signal quality. If the error decreases significantly after compensation, the parameter is increased. A large step size is used to accelerate optimization. If the error fluctuates or increases, the step size is reduced for fine-tuning to avoid over-adjustment. If the signal quality does not improve after three consecutive adjustments, the compensation direction is triggered to be re-evaluated, and the current error component is compared with the historical compensation data. If the original compensation direction is invalid, it is switched to the suboptimal compensation direction, that is, from phase correction to energy optimization, and failure cases are recorded for subsequent strategy optimization. This makes the compensation strategy of the backscattered signal adaptive and maintains stable signal quality in a dynamic environment. The compensated backscattered signal is used to optimize the beam pointing and scanning frequency of the monitoring area, promoting the continuous optimization of the entire measurement method.

[0101] S3. Optimize the beam pointing and scanning frequency of the monitoring area according to the compensated backscatter signal, reacquire the interference data of the monitoring area based on the optimized beam pointing and scanning frequency of the monitoring area, update the interference characteristics, and update the parameter combination of the test signal through the genetic algorithm.

[0102] like Figure 3 As shown in Figure 2, the optimization strategy for beam pointing in the monitoring area includes:

[0103] The energy distribution and phase delay of the compensated backscattered signal are mapped to the direction of arrival space through Fourier transform to construct an energy density map. The energy concentration area is identified through the peak detection algorithm and marked as the target signal direction.

[0104] The propagation feasibility of beam pointing is evaluated by combining the temperature and humidity field distribution in the monitoring area and the spatial topology of the reflectors;

[0105] Through the reinforcement learning algorithm, the beam pointing of the monitoring area is dynamically optimized with the received power of the target signal and the interference energy as the reward function.

[0106] The compensated backscatter signal contains mixed information of the target signal and the interference signal. The direction of the signal source needs to be located through spatial mapping to provide an initial target for beam pointing; the energy distribution and phase delay of the compensated backscatter signal are converted from the time domain signal to the wave arrival direction space through the fast Fourier transform algorithm to generate an energy density map, and the energy density map is subjected to multi-scale Gaussian smoothing to eliminate random noise interference. The non-maximum suppression algorithm is used to retain the energy maximum point in the neighborhood, and an energy threshold is set to filter weak signal points. Preferably, the energy threshold is 1.5 times the global average energy, and finally 3-5 potential target signal directions are identified; the time domain and space domain signals are converted into angle domain expression, which can quickly locate the target signal direction, and the determined target signal direction is used as the initial reference for beam pointing, providing directional constraints for subsequent feasibility evaluation and dynamic optimization.

[0107] There may be strong reflectors or harsh environmental conditions in the direction of the target signal, including high-humidity areas. Direct pointing will cause multipath interference or signal attenuation. The feasibility of pointing needs to be evaluated in combination with environmental factors. The temperature and humidity field distribution and the spatial topology of the reflector in the monitoring area are retrieved, and the environmental characteristics in the target direction are quickly retrieved through the octree spatial index structure. If a high humidity gradient area or a large metal reflector is detected, it is marked as a high-risk area. The judgment of the high humidity gradient area is obtained by comparing the humidity change rate with the change rate threshold, while the judgment of the large metal reflector is obtained by comparing the volume of the reflector with the volume threshold.

[0108] Based on the ray tracing algorithm, signal propagation paths are simulated in parallel on a GPU cluster, and the reflection, refraction, and absorption losses of each path are calculated. If the predicted total loss is greater than the preset loss threshold, the direction is deemed infeasible. For infeasible directions, a greedy search algorithm is used to find feasible alternative directions in their neighborhood, prioritizing paths that pass through low-loss media and avoid large reflectors. Environmental constraint assessment is used to prevent beam pointing from falling into high-loss or interference areas, thereby improving signal reception stability. The selected feasible directions are used as the search space of the reinforcement learning algorithm, narrowing the optimization range and improving the efficiency of beam pointing adjustment.

[0109] To monitor dynamic changes in the environment, namely the movement of interference sources and fluctuations in ambient temperature and humidity, the beam pointing needs to be adjusted in real time to maximize the target signal reception power and suppress interference. Through the deep deterministic policy gradient algorithm, the beam pointing angle is used as a continuous action space. The beam pointing angle includes azimuth and elevation angles. The ratio of the target signal reception power to the interference energy is used as the reward function. An experience replay buffer is constructed. After every 50 beam adjustments, samples are randomly sampled to update the network parameters. The target network parameters are updated every 100 steps through the soft update strategy of the target network. Action exploration is achieved through the OU noise process. The exploration randomness is increased in the early stage of training, and the noise intensity is gradually reduced as training progresses. Dynamic optimization enables the beam pointing to track the target signal in real time and avoid interference. The optimized beam pointing is directly applied to the monitoring antenna array, while triggering the re-acquisition of interference data, providing input in the new environment for subsequent scanning frequency optimization and parameter combination update of the test signal.

[0110] Strategies for optimizing the scanning frequency of the monitoring area include:

[0111] The scattering parameters of the compensated backscattered signal are divided into multiple frequency bands according to frequency. Combined with the monitoring area after beam pointing optimization, the signal attenuation and interference intensity distribution of each frequency band in the monitoring area are analyzed to identify the scanning frequency band and frequency step size.

[0112] Spectrum sensing technology is used to obtain the spectrum occupancy status of the monitoring area in real time. The spectrum change trend is determined based on the interference type, and risky and idle frequency bands are marked.

[0113] The constraints are determined by combining the scanning frequency band, frequency step size and spectrum change trend, and the scanning frequency of the monitoring area is optimized through genetic algorithm.

[0114] The attenuation and interference intensity of the compensated backscattered signal in different frequency bands are different, and the signal propagation characteristics of the monitoring area change after the beam pointing is optimized. It is necessary to combine the two for analysis to determine the efficient scanning frequency band and step size to avoid invalid scanning; the scattering parameters of the backscattered signal are divided into multiple continuous frequency bands according to frequency, and the average value of the reflection coefficient and the transmission loss are calculated for each frequency band. At the same time, the monitoring area after the beam pointing is optimized is called, and the propagation paths of signals of different frequency bands in the monitoring area are simulated. Based on the ray tracing algorithm, combined with the coordinates of the reflector and the temperature and humidity field distribution in the monitoring area, the attenuation value of the signal of each frequency band during the propagation process is calculated. By statistically analyzing the historical interference data, the interference intensity distribution of each frequency band is analyzed, and the frequency bands with lower signal attenuation and weaker interference are marked as candidate scanning bands.

[0115] The step size is determined based on the differences in signal characteristics within the candidate frequency bands. If the signal characteristics within the frequency band change dramatically, that is, there are multiple interference peaks, a small step size is used. If the signal characteristics are stable, a large step size is used to ensure that both signal details can be captured and scanning efficiency can be improved. Through comprehensive analysis of multiple factors, cost-effective scanning frequency bands and frequency steps are screened out, which improves scanning efficiency and reduces signal distortion problems caused by improper frequency band selection. The preliminarily determined scanning frequency bands and frequency steps, together with the spectrum occupancy status analysis results, constitute the basic constraints for frequency optimization.

[0116] The spectrum usage in the monitoring area changes in real time. Historical spectrum data alone cannot meet the needs of dynamic monitoring. It is necessary to perceive the spectrum occupancy status in real time and predict trends to avoid interference bands and utilize idle resources. Compressed sensing technology is used to quickly obtain broadband spectrum data in the monitoring area through sparse sampling, and the reconstruction algorithm is used to restore the complete spectrum. At the same time, energy detection method and feature recognition method are combined to distinguish different types of signals, including communication signals and interference signals, and mark the boundaries and intensities of occupied frequency bands. Combined with the above-mentioned identified interference types, including narrowband interference and broadband interference, their distribution patterns on the spectrum are analyzed to obtain spectrum change trends. The occupancy probability of each frequency band in the future period is predicted based on historical spectrum data through a long short-term memory network. Frequency bands with occupancy probability greater than the first threshold are marked as risk bands, and those less than the second threshold are marked as idle bands.

[0117] A visual spectrum map is established to update the labeling information of risky and idle frequency bands in real time. When the occupancy rate of a risky frequency band shows a rapid upward trend, an early warning mechanism is triggered, prompting the need to adjust the scanning strategy first. This enables dynamic monitoring and forward-looking analysis of spectrum resources, allowing scanning frequencies to avoid high-risk frequency bands to reduce the interference collision rate and effectively improve the anti-interference capability of the monitoring antenna. The spectrum occupancy status and trend prediction results are combined with the candidate scanning frequency bands and frequency step amounts to form a complete frequency optimization constraint condition, providing an optimization basis for the genetic algorithm.

[0118] Under complex constraints, traditional optimization algorithms are prone to falling into local optimality. It is necessary to use the global search capability of genetic algorithms to quickly find the optimal scanning frequency combination; the scanning start frequency, end frequency and frequency step are encoded as chromosome genes, and the parameter accuracy is ensured by real number encoding. The initial population is randomly generated, including 100 individuals, and a feasibility check is performed according to the constraints, and individuals that exceed the antenna operating frequency band or violate the frequency step limit are eliminated; a fitness function is constructed, and scanning efficiency, spectrum utilization and interference suppression effect are used as core evaluation indicators, among which scanning efficiency is related to the number of scanning frequency bands and frequency step size, spectrum utilization refers to the utilization degree of idle frequency bands, and interference suppression effect refers to the degree of avoidance of risky frequency bands. The fitness value of each individual is then calculated by weighted summation. Preferably, the weight of scanning efficiency is set to 0.4, the weight of spectrum utilization is set to 0.3, and the weight of interference suppression effect is set to 0.3.

[0119] Through selection, crossover and mutation operations, the population is iteratively updated, and the individuals with the highest fitness are retained in each generation, accelerating the convergence of the genetic algorithm. When the fitness of the optimal individual has not significantly improved over multiple consecutive generations, the iteration is terminated and the optimal individual is decoded as the final scanning frequency. The genetic algorithm efficiently searches for the optimal scanning frequency combination under complex constraints, shortening the calculation time and improving spectrum utilization. The optimized scanning frequency is applied to the monitoring antenna, triggering the re-acquisition of interference data at the same time, pushing the entire monitoring process into a new round of dynamic optimization cycle.

[0120] The optimized beam pointing and scanning frequency change the way the monitoring antenna receives and acquires signals. To accurately reflect the interference situation in the current monitoring environment, interference data needs to be reacquired. Based on the optimized beam pointing, the phase weighting coefficient or mechanical steering device of the monitoring antenna array is adjusted to align the antenna main beam with the target signal. At the same time, according to the optimized scanning frequency setting, the frequency range and frequency step size of the signal acquisition device are configured. The signal acquisition device is started and continuous or timed data acquisition is performed within the monitoring area. During the data acquisition process, interference data such as environmental electromagnetic data, temperature and humidity data, and reflector coordinates are simultaneously acquired. To ensure the accuracy and consistency of the acquired interference data, different types of data are time-synchronized and calibrated. All sensors and signal acquisition devices operate under the same time reference. The acquired data is also calibrated to eliminate the influence of equipment errors and environmental noise. The reacquired interference data can more accurately reflect the actual situation in the current monitoring environment and provide a reliable data foundation for subsequent updates of interference characteristics and test signal parameter combinations. The acquired interference data is directly used to update the interference characteristics and provide a basis for updating the test signal parameter combinations.

[0121] The monitoring environment and signal propagation conditions change over time, and the reacquired interference data will contain new interference patterns and features, so the interference features need to be updated in a timely manner. The reacquired interference data is preprocessed, including data cleaning, denoising, and normalization. For environmental electromagnetic data, high-frequency noise and low-frequency interference are removed through filtering algorithms. For temperature and humidity data, outlier detection and processing are performed, and normalization is performed to keep them within the same numerical range. For reflector coordinates, spatial coordinate transformation and alignment are performed to ensure data consistency. Feature extraction is performed on the preprocessed interference data, and the extracted new features are fused and updated with historical feature data to form the latest interference feature vector.

[0122] The updated interference features are evaluated and verified to ensure their accuracy and reliability. By comparing with known interference types and historical feature data, it is verified whether the new features can accurately identify and classify different types of interference. At the same time, cross-validation and other methods are used to evaluate the stability and generalization ability of the features to ensure their effectiveness in different monitoring environments. Timely updating of interference features can adapt to changes in the monitoring environment and improve the ability to identify and process interference. The updated interference features serve as an important basis for updating the parameter combination of test signals, provide more accurate input for the genetic algorithm, and help generate more optimized parameter combinations of test signals.

[0123] As the monitoring environment and interference characteristics change, the original parameter combination of the test signal is no longer optimal and needs to be optimized through a genetic algorithm to generate a parameter combination of the test signal that is more suitable for the current environment; the parameters of the test signal are encoded to form a chromosome, and each parameter is mapped to a real number interval through real number encoding. A certain number of initial chromosomes are randomly generated to form an initial population; the quality of each chromosome is evaluated through a fitness function. The design of the fitness function takes into account multiple factors, including interference suppression effect, antenna matching degree and propagation efficiency. Preferably, the interference suppression effect can be taken as the main goal, and its interference suppression ability is evaluated by calculating the bit error rate or signal-to-noise ratio of the test signal in different interference scenarios. At the same time, the antenna matching degree and propagation efficiency are considered, and its performance is evaluated by calculating the radiation efficiency of the antenna and the coverage range of the signal in the monitoring area. These factors are weighted and summed to obtain the fitness value of each chromosome.

[0124] Genetic operations are performed on the initial population, including selection, crossover, and mutation. The selection operation uses roulette wheel selection or tournament selection methods based on the fitness value of the chromosome to select chromosomes with higher fitness to enter the next generation. The crossover operation randomly selects two chromosomes and exchanges genes between them to generate new chromosomes. The mutation operation randomly mutates the genes of the chromosomes with a certain probability to introduce new genetic diversity. The genetic operation is repeated and the population is continuously iterated and optimized until the preset termination condition is met. The termination condition is that a certain number of iterations is reached, the fitness value no longer increases, or other specific performance indicators are met. During each iteration, the optimal chromosome and its fitness value are recorded and adjusted and optimized as needed. When the termination condition is met, the chromosome with the highest fitness is selected as the optimal solution and decoded into a parameter combination of the test signal. These parameter combinations are the parameter combinations of the test signal optimized by the genetic algorithm and can be directly injected into the FM monitoring antenna.

[0125] Updating the parameter combination of the test signal through the genetic algorithm can make the test signal better adapt to changes in the monitoring environment. Compared with traditional parameter optimization methods, the genetic algorithm has global search capabilities and strong robustness, and can find better solutions in complex parameter spaces. The updated parameter combination of the test signal will be used to form a new test signal and injected into the FM monitoring antenna for a new round of monitoring and data acquisition, forming a closed-loop optimization process.

Claims

1. A method for measuring parameters of an FM monitoring antenna, characterized in that: include: Obtain interference data from the monitoring area in real time. The interference data includes environmental electromagnetic data, temperature and humidity data, and reflector coordinates. Use convolutional neural networks to perform pattern recognition on the interference data to output interference features. The interference features include interference type and spatiotemporal intensity distribution. A multi-objective optimization function is then constructed to process the interference features to generate test signals. The test signal is injected into the FM monitoring antenna, and the backscattered signal is received and monitored at the same time. The scattering parameters, energy distribution and phase delay of the backscattered signal are obtained in real time. The signal error between the test signal and the backscattered signal is comprehensively determined based on the scattering parameters, energy distribution and phase delay, and the backscattered signal is dynamically compensated according to the signal error. Optimize the beam pointing and scanning frequency of the monitoring area based on the compensated backscatter signal, reacquire the interference data of the monitoring area based on the optimized beam pointing and scanning frequency of the monitoring area, update the interference characteristics, and update the parameter combination of the test signal through the genetic algorithm; The performing pattern recognition on the interference data by using a convolutional neural network to output interference features includes: Obtain interference data in the monitoring area in real time, including environmental electromagnetic data, temperature and humidity data, and reflector coordinates, and align the interference data by timestamp; The environmental electromagnetic data is converted into a time-frequency matrix through short-time Fourier transform, the temperature and humidity data are processed by Kriging interpolation to generate the temperature and humidity field distribution, and the reflector coordinates are constructed into a spatial topology through the octree algorithm; A three-branch convolutional neural network consisting of an electromagnetic branch, an environmental branch, and a spatial branch is constructed. The three-branch convolutional neural network extracts the features of the interference data separately and performs weighted fusion through an attention mechanism to generate interference features including interference type and spatiotemporal intensity distribution. The interference features are compared and analyzed with the historical feature data to determine whether to adjust the parameters and attention weights of the three-branch convolutional neural network.

2. The parameter measurement method for an FM monitoring antenna according to claim 1, wherein: The constructing of a multi-objective optimization function to process interference features to generate a test signal comprises: A generative adversarial network is constructed. The generator processes the interference features to output candidate parameter combinations of the test signal. The discriminator evaluates the candidate parameter combinations of the test signal through a multi-objective optimization function. A physical constraint layer is configured according to the properties of the FM monitoring antenna, and candidate parameter combinations of the test signal are screened according to the physical constraint layer; The candidate parameter combinations of the test signal are optimized by the simulated annealing algorithm. After multiple iterative training of the generator and the discriminator, the test signal is generated. The parameter combination of the test signal includes the frequency step amount, the modulation index and the duty cycle.

3. The parameter measurement method for an FM monitoring antenna according to claim 2, wherein: The electromagnetic branch identifies the interference pattern of the time-frequency graph matrix through a residual network, the environmental branch extracts the gradient changes of the temperature and humidity field distribution through a two-dimensional convolutional layer, and the spatial branch analyzes the spatial topology and extracts the path complexity of multipath reflection through a graph convolutional network.

4. The parameter measurement method for an FM monitoring antenna according to claim 3, wherein: The multi-objective optimization function construction includes: Interference suppression, antenna matching, and propagation efficiency are used as optimization objectives, and the target weights of the optimization objectives are dynamically adjusted according to the interference type through a Bayesian network. The multi-objective optimization function is constructed by taking the optimization objectives and objective weights into account through weighted summation.

5. The parameter measurement method for an FM monitoring antenna according to claim 4, wherein: The dynamically compensating the backscattered signal according to the signal error comprises: extracting the dominant component of the signal error and determining the compensation direction of the backscattered signal according to the dominant component of the signal error; Based on the compensation direction of the backscattered signal, collaboratively adjust the parameter combination of the test signal to compensate for the backscattered signal; The signal quality of the backscattered signal after compensation is monitored in real time. The signal quality includes the convergence of the scattering parameters and the energy concentration. The adjustment step size of the parameter combination of the test signal and the compensation direction of the backscattered signal are updated based on the signal quality.

6. The parameter measurement method for an FM monitoring antenna according to claim 5, wherein: The comprehensive determination of the signal error between the test signal and the backscattered signal includes: Inject the test signal into the FM monitoring antenna, receive the backscattered signal, and obtain the scattering parameters, energy distribution and phase delay of the backscattered signal in real time; Performing cross-correlation analysis on the scattering parameters, energy distribution and phase delay of the backscattered signal to obtain a correlation matrix, and determining the error weights of the scattering parameters, energy distribution and phase delay according to the correlation matrix; The signal error between the test signal and the backscattered signal is determined by weighted summation according to the error weight.

7. The parameter measurement method for an FM monitoring antenna according to claim 6, wherein: Based on the field intensity probe scanning the monitoring area, the backscattered signal is sampled in three dimensions. The energy distribution of the backscattered signal is obtained by obtaining the energy value of each sampling point. The carrier phase of the test signal and the backscattered signal is compared through coherent detection technology to obtain the phase delay of the backscattered signal.

8. The parameter measurement method for an FM monitoring antenna according to claim 7, wherein: The beam pointing of the optimized monitoring area includes: The energy distribution and phase delay of the compensated backscattered signal are mapped to the direction of arrival space through Fourier transform to construct an energy density map. The energy concentration area is identified through the peak detection algorithm and marked as the target signal direction. The propagation feasibility of beam pointing is evaluated by combining the temperature and humidity field distribution in the monitoring area and the spatial topology of the reflectors; Through the reinforcement learning algorithm, the beam pointing of the monitoring area is dynamically optimized with the received power of the target signal and the interference energy as the reward function.

9. The parameter measurement method for an FM monitoring antenna according to claim 8, wherein: The scanning frequency of the optimized monitoring area includes: The scattering parameters of the compensated backscattered signal are divided into multiple frequency bands according to frequency. Combined with the monitoring area after beam pointing optimization, the signal attenuation and interference intensity distribution of each frequency band in the monitoring area are analyzed to identify the scanning frequency band and frequency step size. Spectrum sensing technology is used to obtain the spectrum occupancy status of the monitoring area in real time. The spectrum change trend is determined based on the interference type, and risky and idle frequency bands are marked. The constraints are determined by combining the scanning frequency band, frequency step size and spectrum change trend, and the scanning frequency of the monitoring area is optimized through genetic algorithm.

Images