Machine Learning-Based Radio Metrology Calibration Method and System

Abstract

The present application relates to the field of machine learning technology, and discloses a radio metrology calibration method and system based on machine learning. The method includes: measuring parameters of a radio signal source to obtain an original measurement data set; extracting frequency accuracy features, power flatness features, phase noise features, harmonic ratio features, and environmental change features based on the original measurement data set to obtain a feature data set; dividing the feature data set into a training data subset, a validation data subset, and a test data subset, and constructing a first neural network model including a dual-branch neural network based on the training data subset; inputting the validation data subset into the first neural network model for model parameter optimization to obtain a second neural network model, thereby improving the calculation accuracy of calibration parameters and the accuracy of environmental compensation.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, and particularly to a radio metrology calibration method and system based on machine learning. Background Art

[0002] With the rapid development of radio measurement technology, the calibration accuracy and stability requirements for signal sources, as key measurement devices, are constantly increasing. Traditional signal source calibration methods mainly rely on manual experience and fixed calibration parameters, and it is difficult to cope with measurement errors and drift problems under complex environmental conditions. Especially in high-frequency band measurements, the influence of environmental factors such as ambient temperature and humidity on measurement results is more significant, and traditional calibration methods cannot effectively compensate for the influence brought by these environmental factors.

[0003] Existing calibration technologies generally have problems such as long calibration cycles, much manual intervention, and low calibration efficiency. At the same time, due to the lack of in-depth mining and utilization of historical calibration data, it is impossible to establish a mapping relationship between environmental factors and measurement errors, resulting in the reliability and stability of calibration results being difficult to guarantee. In addition, existing calibration methods lack the ability of adaptive optimization and cannot dynamically adjust calibration parameters according to the device usage status and environmental changes. In large-scale measurement application scenarios, the performance of signal sources will drift over time, and frequent calibration and maintenance are required. However, traditional calibration methods cannot achieve online real-time calibration and lack the ability to predict and analyze the long-term performance change trend of devices, which severely restricts the usage efficiency of measurement devices and the reliability of measurement results. Summary of the Invention

[0004] This application provides a radio metrology calibration method and system based on machine learning, thereby improving the calculation accuracy of calibration parameters and the accuracy of environmental compensation.

[0005] In the first aspect of this application, a radio metrology calibration method based on machine learning is provided. The radio metrology calibration method based on machine learning includes:

[0006] Performing parameter measurement on a signal source of radio to obtain an original measurement data set;

[0007] Based on the original measurement data set, extracting frequency accuracy features, power flatness features, phase noise features, harmonic ratio features, and environmental change features to obtain a feature data set;

[0008] Dividing the feature data set into a training data subset, a validation data subset, and a test data subset, and constructing a first neural network model including a dual-branch neural network based on the training data subset;

[0009] Inputting the validation data subset into the first neural network model for model parameter optimization to obtain a second neural network model.

[0010] The second aspect of the present application provides a radio metrology calibration system based on machine learning. The radio metrology calibration system based on machine learning includes:

[0011] A measurement module for measuring parameters of a radio signal source to obtain an original measurement data set;

[0012] An extraction module for extracting frequency accuracy features, power flatness features, phase noise features, harmonic ratio features, and environmental change features based on the original measurement data set to obtain a feature data set;

[0013] A construction module for dividing the feature data set into a training data subset, a validation data subset, and a test data subset, and constructing a first neural network model including a double-branch neural network based on the training data subset;

[0014] An optimization module for inputting the validation data subset into the first neural network model for model parameter optimization to obtain a second neural network model.

[0015] Compared with the prior art, the present application has the following beneficial effects: By constructing a double-branch neural network structure, parameter calibration and environmental compensation are modeled separately, effectively reducing the complexity of the model, while improving the calculation accuracy of calibration parameters and the accuracy of environmental compensation. The network structure design adopting the self-attention mechanism and residual connection enhances the feature extraction ability of the model for measurement data at different frequency points and different power points, while improving the convergence performance of the model. The LSTM network structure is introduced to model environmental parameters, and combined with the temporal attention mechanism, the dynamic influence of environmental parameter changes on measurement results is accurately captured, realizing precise control of environmental compensation. A loss function and an optimization strategy based on physical constraints are designed to ensure that the calibration results meet the physical characteristics requirements of the signal source and avoid unreasonable calibration parameters output by the model. Through the incremental learning method and the hierarchical parameter update strategy, online optimization of the model is achieved, maintaining the model's memory ability for the learned features, while improving the model's adaptability to new data. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] The structures, proportions, sizes, etc. shown in the drawings of this specification are only used to cooperate with the content disclosed in the specification for those skilled in this technology to understand and read, and are not used to limit the conditions under which the present invention can be implemented. Therefore, they do not have substantial technical significance. Any modification of the structure, change in the proportional relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope covered by the technical content disclosed in the present invention.

[0018] Figure 1 is a schematic flowchart of a radio metrology calibration method based on machine learning provided by an embodiment of the present invention;

[0019] Figure 2 is a schematic block diagram of the structure of a radio metrology calibration system based on machine learning provided by an embodiment of the present invention. Detailed implementation manners

[0020] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0021] The flowcharts shown in the drawings are only illustrative examples, and do not necessarily include all the content and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged. Therefore, the actual execution order may change according to the actual situation.

[0022] It should also be understood that the terms used in this specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this specification of the present application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0023] It should be further understood that the term "and / or" used in this specification of the present application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Please refer to Figure 1 , an embodiment of the radio metrology calibration method based on machine learning in the embodiments of the present application includes:

[0024] Step 100: Measure the parameters of the radio signal source to obtain an original measurement data set;

[0025] It can be understood that the execution entity of this application can be a radio measurement and calibration system based on machine learning, or it can also be a terminal or a server. Specifically, it is not limited here. In the embodiments of this application, the server is taken as an example of the execution entity for illustration.

[0026] Specifically, a series of settings are made for the radio signal source, including selecting multiple discrete frequency points to ensure that the measured data can comprehensively cover the frequency response range of the signal source. For discrete frequency points, multiple repeated frequency measurements are required for each frequency point to reduce measurement errors and enhance the stability and reliability of the data. While the frequency measurement is completed, the output power of the signal source is scanned to obtain signal characteristic data at different power levels. During the power scan process, repeated measurements are made for each power point to ensure the accuracy of the measurement data and reduce errors caused by instantaneous fluctuations or environmental changes. At the same time, a phase noise analyzer is used to measure the phase noise of the signal source, and repeated measurements are made for preset frequency offset points to obtain the phase noise characteristic data of the signal source. The harmonic characteristics of the signal source are measured. Harmonic measurement can reveal the harmonic distribution of the signal source in different frequency ranges, thereby helping to evaluate its signal quality. The parameters such as environmental temperature, humidity, and atmospheric pressure are synchronously recorded to construct an environmental parameter dataset. The acquisition of environmental data depends on high-precision temperature and humidity sensors and pressure sensors. These sensors monitor the changes in environmental parameters in real time and record them in the measurement database for analysis and compensation in the subsequent data processing and model training stages. A unified timestamp format is used to synchronize the time of all measurement data. The frequency measurement data, power measurement data, phase noise measurement data, harmonic measurement data, and environmental parameter data are aligned according to the same time reference to form a complete and time-synchronized original measurement dataset. The realization of time synchronization depends on a high-precision time synchronization protocol or a highly stable clock source, such as a GPS synchronization clock or a high-precision local time reference. By aligning all the data in time, it is ensured that during data analysis and model training, various measurement data can correctly correspond to the same time window, thereby improving the timeliness and accuracy of the data.

[0027] Step 200: Based on the original measurement dataset, extract frequency accuracy features, power flatness features, phase noise features, harmonic ratio features, and environmental change features to obtain a feature dataset;

[0028] Specifically, the original measurement data is cleaned and preprocessed to ensure the quality and reliability of the data. Since the radio measurement process is affected by equipment errors, environmental interference or instantaneous fluctuations, the 3σ criterion is used to detect anomalies in the original measurement data. By calculating the mean and standard deviation of the data, the data that exceeds the range of three times the standard deviation is marked as anomalies to obtain the anomaly marking results. After removing the outliers, the remaining non-abnormal data is normalized to eliminate the differences in the dimensions of different measurement parameters, so that different features can be learned on the same scale to obtain normalized measurement data. Frequency features are extracted from the normalized measurement data to characterize the frequency stability and accuracy of the radio signal source. The frequency accuracy is calculated, that is, the deviation between the measured frequency and the standard frequency. At the same time, the frequency stability is calculated to evaluate the stability of the signal source on different time scales, and the frequency resolution is analyzed to determine the measurement system's perception of small frequency changes. The long-term drift trend analysis is performed on each frequency measurement point to examine the changes in the signal frequency on a longer time scale, and the short-term jitter characteristics are extracted to reflect the stability changes of the signal in a short time. Through the above calculations, complete frequency feature data is obtained. At the same time, the phase noise data in the normalized measurement data is analyzed, the noise values ​​at multiple frequency deviation points are extracted, and the phase noise characteristic data is constructed. The phase noise values ​​are calculated at multiple frequency deviation points (such as 1Hz, 10Hz, 100Hz, 1kHz, etc.) to fully characterize the phase noise characteristics of the signal source. The harmonic data in the normalized measurement data is processed, and the harmonic levels of different harmonics and their ratios to the fundamental wave are calculated to obtain harmonic characteristic data. The environmental parameters in the normalized measurement data are analyzed in time and frequency using wavelet transform to reveal the pattern of environmental parameters changing over time. The temperature change rate, humidity change rate and their correlation coefficients with the measurement parameters are calculated to quantify the impact of environmental factors on the stability of the radio signal. Through this step, the key environmental factors that can affect the measurement accuracy are extracted and used as part of the characteristic data to enhance the environmental adaptability and robustness of the model. The frequency characteristic data, phase noise characteristic data, harmonic characteristic data and environmental characteristic data are combined to form a characteristic data set.

[0029] Step 300: Divide the feature data set into a training data subset, a verification data subset, and a test data subset, and construct a first neural network model including a two-branch neural network based on the training data subset;

[0030] It should be noted that the feature dataset is divided to ensure that the training, validation, and testing of the model can cover the diversity of the data and have good generalization ability. The feature dataset is divided into a training data subset, a validation data subset, and a testing data subset in a ratio of 7:2:1. Among them, the training data is used for learning model parameters, the validation data is used to adjust hyperparameters and prevent overfitting, while the testing data is used for the final model performance evaluation. After the division, in order to enhance the robustness of the model and avoid biases caused by the data order, the training data subset is randomly shuffled to ensure that the model does not depend on the input order of the data during the training process, avoid the occurrence of local optimal solutions, and obtain the preprocessed training data. When constructing the neural network model, since radio metrology calibration involves multiple physical characteristic data, a dual-branch neural network architecture is adopted. The first branch is used to process the calibration parameters of the radio signal itself, while the second branch is used to process the compensation coefficients of environmental factors. In the first branch, an eight-layer fully connected neural network, namely the parameter calibration network, is designed. The first three layers respectively contain 1024, 512, and 256 neurons for feature extraction and non-linear mapping. These three layers use the ReLU activation function to effectively avoid the problem of gradient disappearance and improve the learning ability of the model. The self-attention mechanism is introduced in the fourth and fifth layers, where the dimension of the weight matrix is set to 256×256, so as to enhance the model's ability to capture long-range dependencies and enable it to more accurately learn the relationships between different frequency points, power levels, and other calibration parameters. A residual connection is set between these two self-attention layers to ensure the effective propagation of gradients and improve the training stability of the model. In the sixth to eighth layers, 256, 128, and 64 neurons are set respectively for feature compression and the generation of parameter calibration outputs. The last layer uses a linear activation function to ensure that the calibration parameters output by the model are optimized and applied in a continuous numerical space. At the same time, in the second branch, a three-layer LSTM network, namely the environmental compensation network, is constructed to process environmental feature data and calculate environmental compensation coefficients. Since environmental factors (such as temperature, humidity, and atmospheric pressure) have temporal correlations, LSTM (Long Short-Term Memory network) is a suitable structural choice. Each layer of LSTM contains 128 hidden units to effectively capture long-term dependencies in the time series. A dropout layer is added between the LSTM units to prevent overfitting and ensure that the model still has good generalization ability when facing complex environmental data. In order to enable the model to focus on the environmental data features at different time points and assign higher weights to the time windows with greater influence on the measurement, a temporal attention mechanism is introduced in the output part of the LSTM. Through an adaptive weighting method, the environmental feature data with greater influence on the measurement accuracy is emphasized, and the environmental compensation output is obtained. The calibration parameter output and the environmental compensation output are fused to obtain the final model output.Adopt weighted summation operation. By setting a set of weight parameters, linearly weight the outputs of the two branches so that environmental compensation can reasonably correct the calibration parameters. To ensure the dynamic adaptability of the fusion process, these weight coefficients are adaptively updated through backpropagation, ensuring that the model is continuously optimized during training, so that the influence of environmental factors is reasonably compensated without causing excessive interference to the final calibration result. After the above modeling process, a complete first neural network model is obtained. Its dual-branch structure can effectively consider the physical measurement characteristics of radio signals and external environmental factors simultaneously, and achieve high-precision metrological calibration through adaptive optimization.

[0031] Step 400: Input the validation data subset into the first neural network model for model parameter optimization to obtain the second neural network model.

[0032] Specifically, the validation data subset is input into the first neural network model to calculate the relative frequency deviation of the frequency measurement data, that is, the ratio difference between the measured frequency and the reference frequency, to measure the accuracy of the model in frequency calibration. Calculate the absolute deviation of the power measurement data, that is, the direct difference between the measured power and the standard power, to evaluate the accuracy of power calibration. Calculate the root mean square error of each frequency offset point for the phase noise measurement data. By statistically averaging the squared errors between the model prediction values and the true values at different frequency offset points, to quantify the performance of the model in phase noise compensation, and obtain the parameter error values, including frequency error, power error, phase noise error, and harmonic error. Construct a loss function based on the output results of the first neural network model to clarify the optimization direction and objective. Since radio signal metrology involves multiple important physical characteristics, different weight coefficients are assigned to different error terms when constructing the loss function, so as to reflect their impact on the final measurement accuracy. Assign a first weight coefficient to the frequency error term to ensure that the model can achieve high precision in frequency calibration. At the same time, assign a second weight coefficient to the power error term to ensure the accuracy of power calibration. Since phase noise has a greater impact on the stability of radio signals, assign a third weight coefficient to the phase noise error term to enhance the model's ability to calibrate phase noise. Assign a fourth weight coefficient to the harmonic error term to ensure that the model can effectively calibrate the harmonic components and make the measurement results more accurate. Through the weighted loss function, the model pays attention to the importance of different error terms during the optimization process and ensures that the optimization objective meets the actual requirements of radio metrology. Add physical constraint terms to the weighted loss function to construct a constrained loss function that better conforms to the measurement standards. These physical constraints include: ensuring that the frequency error does not exceed the inherent resolution of the measurement device within a certain range to prevent meaningless small deviations from interfering with the optimization results; restricting the upper and lower bounds of the power error to ensure that the calibrated power output conforms to the dynamic range of the measurement device; constraining the optimization range of phase noise to ensure that the optimization process does not affect the essential characteristics of the signal; and setting reasonable constraints on the harmonic error to prevent over-calibration from causing excessive harmonic suppression and thus affecting the normal transmission characteristics of the signal. By introducing these physical constraint terms, the optimization process can better meet the needs of the real measurement system, making the final calibration model have higher physical credibility. During the optimization process, the Adam optimization algorithm is adopted. The Adam algorithm combines the advantages of momentum gradient descent and adaptive learning rate, enabling the model to converge quickly during training and avoid falling into local optimal solutions. Based on the constrained loss function and the parameter error values, the Adam optimization algorithm calculates the gradients and updates the model weights, enabling the model to gradually reduce the errors and improve the calibration accuracy. During the optimization process, the first-order moment estimate and second-order moment estimate of the Adam algorithm can ensure the adaptive adjustment of the learning rate, making the optimization more stable, avoiding unstable oscillations caused by too high a learning rate, and preventing the convergence speed from being too slow due to too low a learning rate.As the optimization progresses, the model parameters are continuously adjusted. Therefore, iterative training is performed on the first neural network model. Through multiple rounds of optimization, the loss function is continuously reduced, and the errors of each parameter gradually converge to a smaller range. In each iterative training, the model is updated according to the optimized parameters, and a new error calculation is performed using the validation dataset to ensure that the optimization direction is correct. When the loss function converges within the preset threshold range, that is, when each error term meets the requirements of the radio metrology standard, the trained model parameters are obtained. Based on the trained model parameters, the first neural network model is finally updated to form the second neural network model.

[0033] After the model is deployed, based on the signal source performance data in the actual usage process, the neural network model is dynamically updated to enable it to adapt to the long-term changing measurement environment and maintain a high calibration accuracy. During the actual usage process, the performance of the signal source changes over time. Therefore, long-term stability data, environmental adaptability data, and reliability data are continuously monitored and collected to ensure the effectiveness of the model. When the cumulative new data volume exceeds 10% of the initial training data subset, the system triggers a model update signal to determine whether the model needs to be adjusted and optimized, so as to ensure the long-term effectiveness and adaptability of the radio metrology calibration model. After triggering the model update, in order to ensure the stability of the neural network and make full use of the existing training results, the model is optimized by fixing some parameters and updating some parameters. In the second neural network model, the parameters of the first five layers of the parameter calibration network will be fixed, including 1024, 512, and 256 neurons in the first to third layers, and the self-attention layer parameters of the 256×256 dimension in the fourth to fifth layers. These layers are mainly responsible for low-level feature extraction and global feature attention calculation, and have learned the basic feature patterns of signal source calibration. Therefore, they do not need to be adjusted frequently to ensure the stability of the basic structure of the model and avoid overfitting or forgetting of historical data. After fixing these parameters, the fixed parameter layer is obtained. While ensuring the stability of the fixed parameter layer, the last three layers of the parameter calibration network are updated to enhance the model's adaptability to new data. The parameters of 256, 128, and 64 neurons in the sixth to eighth layers will be retrained so that the model can better learn the new signal source characteristics and improve its generalization ability in different measurement environments. At the same time, in order to ensure the stability of the environmental compensation mechanism, the parameters of the three-layer LSTM network in the environmental compensation network remain unchanged because the LSTM part has learned the long-term impact of environmental factors on radio signal metrology and will not deviate significantly with the change of short-term measurement data. By fixing some parameters and updating some parameters, the adaptability of the model is effectively improved, and the learning results of the original network are retained to obtain the updatable parameter layer to adapt to the latest data distribution. After the model is updated, its calibration accuracy is verified. The test data subset is input into the updated neural network model, and the calibration error and measurement deviation are calculated to evaluate the calibration accuracy of the new model. By comparing with the performance of the old model, it is judged whether the updated neural network model improves the accuracy of radio metrology and ensures its applicability to the latest signal source data distribution. The calibration accuracy evaluation results are used for model fine-tuning to optimize the application effect of the model in the actual measurement environment. On this basis, in order to optimize the performance of the model and prevent the forgetting of existing knowledge due to the addition of new data, an incremental learning method is used to fine-tune the updatable parameter layer. Incremental learning can enable the model to adapt to new data without affecting the learned knowledge and ensure that the model does not completely forget the previous measurement characteristics due to the change of training data.During the fine-tuning process, based on the calibration accuracy evaluation results, the last three layers of the parameter calibration network are adjusted, and the parameters of the model are gradually optimized through the backpropagation algorithm while keeping the feature extraction ability of the fixed parameter layer unchanged. Thus, while continuously learning new data, the measurement features of the old data are retained, enabling the neural network to operate stably for a long time and adapt to the changing radio measurement environment. After the fine-tuning is completed, the fixed parameter layer and the adjusted updatable parameter layer are combined to form the final target calibration model.

[0034] Perform Fourier transform on the frequency measurement data in the test data subset. Through the transform, the signal is converted from the time domain to the frequency domain to more effectively analyze its frequency characteristics. In the frequency domain space, extract each frequency component and calculate the power spectral density of the frequency component to quantify the stability of the radio signal. In this way, the frequency stability evaluation data is obtained. Based on the output of the parameter calibration network of the first neural network model, perform self-attention weighting on the frequency stability evaluation data to enhance the model's perception ability of key frequency features. Calculate the weight distribution of different frequency components on the weight matrix of 256×256 dimensions, so that the model can reasonably adjust the contribution degree of the calibration parameters according to the influence degree of different frequency components on the measurement error. In this way, the frequency accuracy score is obtained, and the frequency calibration evaluation parameters are formed to quantify the optimization effect of the model on the frequency characteristics and ensure its high-precision calibration ability during long-term operation. At the same time, analyze the phase noise measurement data in the test data subset to evaluate the model's phase noise compensation ability. Calculate the root mean square value of the noise for multiple different frequency offset points to quantify the phase jitter of the signal source under different frequency offset conditions. Based on the prediction result of the second neural network model, calculate the phase noise compensation coefficient to correct the phase noise deviation caused by environmental or equipment changes during the measurement process. Through these calculations, the phase noise evaluation parameters are obtained, which are used to measure the optimization effect of the model on the phase noise and ensure that the calibration result can effectively reduce the phase jitter of the radio signal. To evaluate the model's adaptability to environmental changes, perform time series correlation analysis on the output results of the three-layer LSTM unit of the environmental compensation network and the environmental parameter data in the test data subset to calculate the influence of environmental factors on the radio signal measurement accuracy. During this process, calculate environmental variables such as the temperature change rate and humidity change rate, and analyze their influence coefficients on the calibration result. Through this step, the environmental compensation evaluation parameters are obtained, which are used to measure the model's adaptability under different environmental conditions and ensure that it can still maintain high measurement accuracy in a complex external environment. Perform adaptive weighting on the frequency calibration evaluation parameters, phase noise evaluation parameters, and environmental compensation evaluation parameters to obtain the initial evaluation score. Based on the learning result of the neural network model, adaptively adjust the weights of each parameter in the calibration accuracy evaluation so that the final evaluation result can fully consider the combined effects of multiple factors such as frequency characteristics, phase noise, and environmental impact. Based on the output of the fusion layer of the dual-branch neural network model, perform residual compensation on the initial evaluation score, that is, use the output error of the fusion layer to dynamically adjust the evaluation score to ensure that it is closer to the distribution of the actual measurement data. Through the residual compensation mechanism, the compensated evaluation score is obtained, and its physical constraint verification is performed to ensure that the evaluation result conforms to the physical laws and measurement standards of radio measurement, and finally the calibration accuracy evaluation result is obtained.

[0035] In the embodiments of the present application, by constructing a dual-branch neural network structure, parameter calibration and environmental compensation are modeled separately, effectively reducing the complexity of the model while improving the calculation accuracy of the calibration parameters and the accuracy of environmental compensation. The network structure design adopting the self-attention mechanism and residual connection enhances the feature extraction ability of the model for measurement data at different frequency points and different power points, and at the same time improves the convergence performance of the model. The LSTM network structure is introduced to model environmental parameters, and combined with the temporal attention mechanism, the dynamic influence of environmental parameter changes on measurement results is accurately captured, realizing precise control of environmental compensation. A loss function and optimization strategy based on physical constraints are designed to ensure that the calibration results meet the physical property requirements of the signal source and avoid unreasonable calibration parameters output by the model. Through the incremental learning method and hierarchical parameter update strategy, online optimization of the model is achieved, maintaining the model's memory ability for the learned features while improving the model's adaptability to new data.

[0036] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0037] Set multiple discrete frequency points for the radio signal source, and perform repeated frequency measurements on each discrete frequency point to obtain frequency measurement data, and scan the output power of the signal source, and perform repeated measurements on each power point to obtain power measurement data;

[0038] Based on a phase noise analyzer, perform repeated measurements on the signal source at preset frequency offset points to obtain phase noise measurement data, and perform harmonic measurements on the signal source to obtain harmonic measurement data;

[0039] Record the environmental temperature, humidity, and atmospheric pressure parameters to obtain environmental parameter data;

[0040] Based on a unified timestamp format, synchronize the time of the frequency measurement data, power measurement data, phase noise measurement data, harmonic measurement data, and environmental parameter data to obtain the original measurement dataset.

[0041] Specifically, multiple discrete frequency points are set for the radio signal source to ensure that the collected data can fully cover the working range of the signal source. These discrete frequency points are selected as equally spaced frequency points or specific key frequency points according to experimental requirements. For example, for a signal source operating in the range of 1 GHz to 10 GHz, a set of discrete frequency points ) is selected, where each represents a selected discrete frequency point. For example, if a 100 MHz interval is taken, then there are 1 GHz, GHz GHz. For each discrete frequency point, multiple repeated frequency measurements are carried out to reduce measurement errors and improve the reliability of the data. For each frequency point , perform repeated measurements to obtain a set of measurement data ( ), where represents the frequency value measured at the -th measurement. Subsequently, the average measurement value and the measurement standard deviation are calculated, and the formulas are as follows:

[0042] ;

[0043] ;

[0044] where represents the average frequency of the measurement, and reflects the degree of dispersion of the measurement values, that is, the measurement stability. At the same time, the output power of the signal source is scanned to analyze the output characteristics at different power levels. Set the output power range to , and perform a power scan within this range with a step value , that is, a series of power points are obtained:

[0045] ;

[0046] where represents the -th power measurement point, and is the total number of power measurement points. For each power point , multiple repeated measurements are carried out to obtain a set of measurement values and calculate their average value and standard deviation:

[0047] ;

[0048] ;

[0049] where represents the average power measured, and reflects the uncertainty of the power measurement. To analyze the phase noise of the signal, a phase noise analyzer is used to measure the signal source at different preset frequency offset points. Set multiple frequency offset points ( ), such as 1 Hz, 10 Hz, 100 Hz, 1 kHz, etc., and measure the phase noise power spectral density at these frequency offset points, and its definition is as follows:

[0050] ;

[0051] Among them, represents the phase noise power spectral density at the frequency offset . This data reflects the phase jitter of the radio signal at different frequency offsets. To evaluate the quality of the signal source, the harmonics are measured. Set the fundamental frequency , and measure its higher harmonic components , and calculate the harmonic ratio:

[0052] ;

[0053] Among them, is the power of the th harmonic, is the fundamental power. A lower indicates less nonlinear distortion of the signal source and higher signal quality. At the same time, to study the influence of environmental factors on the measurement results, record the environmental temperature, humidity, and atmospheric pressure, that is:

[0054] ;

[0055] Among them, represents the temperature varying with time, represents the humidity, represents the air pressure. Calculate the temperature change rate and the humidity change rate :

[0056] ;

[0057] The environmental data is helpful for the subsequent training of the machine learning calibration model. Use a unified timestamp format to synchronize the time of all measurement data. For example, the frequency measurement data , the power measurement data , the phase noise measurement data , the harmonic measurement data , and the environmental parameter data are recorded according to the same time reference for subsequent data fusion and feature extraction, and finally a complete original measurement data set is obtained.

[0058] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0059] Perform anomaly detection on the original measurement data set based on the 3σ criterion to obtain the anomaly value marking result, and normalize the non-anomaly data in the original measurement data set to obtain the normalized measurement data;

[0060] Extract the frequency characteristics of the normalized measurement data, calculate the frequency accuracy, frequency stability and frequency resolution, analyze the long-term drift trend and short-term jitter characteristics for each frequency measurement point, and obtain the frequency characteristic data;

[0061] Extract the noise values at multiple frequency offset points from the phase noise data in the normalized measurement data to obtain the phase noise characteristic data, and calculate the harmonic levels and their ratios for the harmonic data in the normalized measurement data to obtain the harmonic characteristic data;

[0062] Perform time-frequency analysis on the environmental parameters in the normalized measurement data through wavelet transform, calculate the correlation coefficients between the temperature change rate, humidity change rate and the measurement parameters, obtain the environmental characteristic data, and combine the frequency characteristic data, phase noise characteristic data, harmonic characteristic data and environmental characteristic data into a characteristic data set.

[0063] Specifically, perform statistical analysis on the original measurement data set, and calculate the mean and standard deviation of each measurement parameter. Assume that the observation value set of a certain measurement parameter is , where represents the value of the th measurement, and is the total number of measurement samples. The mean is calculated as:

[0064] ;

[0065] where represents the mean of this parameter. At the same time, the standard deviation is calculated by the following formula:

[0066] ;

[0067] where represents the standard deviation of this parameter. According to the 3σ criterion, if a certain measurement value satisfies:

[0068] ;

[0069] then this measurement value is regarded as an outlier and marked in the outlier marking result. All non-outlier data will be retained and normalized. The normalization method adopts min-max normalization:

[0070] ;

[0071] where and are the minimum and maximum values in the data set respectively, and the normalized data It is limited within the interval [0, 1] to make the numerical scales of different physical quantities consistent and improve the stability of neural network training. Frequency feature extraction is performed on the normalized measurement data to calculate frequency accuracy, frequency stability, and frequency resolution. For the frequency measurement data set , the frequency accuracy is defined as:

[0072] ;

[0073] where is the measured average frequency value:

[0074] ;

[0075] while is the reference frequency. The frequency stability is calculated using the Allan variance:

[0076] ;

[0077] where represents the sampling time interval, reflects the frequency stability of the signal within the time scale . The frequency resolution is defined as:

[0078] ;

[0079] where reflects the ability of the measurement device to distinguish small frequency changes. For each measurement frequency point, the long-term drift trend is calculated, that is, solved based on linear regression:

[0080] ;

[0081] where represents the drift rate, measured by least squares fitting. The short-term jitter characteristics are extracted by the fast Fourier transform to obtain the spectral components of the frequency jitter, thereby quantifying the influence of high-frequency noise components. For phase noise feature extraction, the noise values at multiple frequency offset points in the normalized measurement data are calculated. The phase noise is in units and is defined as:

[0082] ;

[0083] where represents the phase noise at the offset frequency , represents the phase noise power spectral density. The phase noise data at different frequency offset points will be recorded and used for subsequent model training. For harmonic feature extraction, the powers of different harmonic frequency components are calculated, and the harmonic level is calculated:

[0084] ;

[0085] wherein, is the power of the th harmonic, is the fundamental power. To quantify the relative energy ratio of the harmonic to the fundamental wave, the total harmonic distortion is calculated:

[0086] ;

[0087] This index is used to evaluate the degree of nonlinear distortion of radio signals. In terms of environmental feature extraction, time-frequency analysis is performed on parameters such as temperature, humidity, and atmospheric pressure through wavelet transform. Set the signal as the temperature measurement value sequence, then the wavelet transform is defined as:

[0088] ;

[0089] wherein, is the scale parameter, is the time shift, is the wavelet basis function. Through wavelet transform, the temperature change rate and the humidity change rate are calculated:

[0090] ;

[0091] Perform correlation analysis on the environmental parameter data and the radio measurement data. Let be the measurement parameter, be the environmental variable, then the correlation coefficient is calculated as follows:

[0092] ;

[0093] If is large, it indicates that the environmental parameter has a greater impact on the measurement data. Combine the frequency feature data, phase noise feature data, harmonic feature data, and environmental feature data to form a complete feature dataset.

[0094] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0095] Divide the feature dataset into a training data subset, a validation data subset, and a test data subset according to the ratio of 7:2:1, and perform random shuffling on the training data subset to obtain the preprocessed training data;

[0096] Construct an eight-layer fully-connected network structure for the first branch, where the number of neurons in the 1st - 3rd layers are 1024, 512, and 256 respectively, the 4th - 5th layers adopt the self-attention mechanism with a weight matrix dimension of 256×256, and the number of neurons in the 6th - 8th layers are 256, 128, and 64 respectively, to obtain the parameter calibration network;

[0097] Construct a three-layer LSTM network structure for the second branch, with each layer containing 128 hidden units. The input is environmental feature data and the output is the environmental compensation coefficient, to obtain the environmental compensation network;

[0098] Adopt the ReLU activation function in the 1st - 3rd layers of the parameter calibration network, set a residual connection between the self-attention layers in the 4th - 5th layers, and adopt the linear activation function in the last layer to obtain the calibrated parameter output;

[0099] Set a dropout layer between the three LSTM units of the environmental compensation network, and weight the environmental features at different times through the temporal attention mechanism to obtain the environmental compensation output;

[0100] Fuse the calibrated parameter output and the environmental compensation output based on the weighted sum operation, and adaptively update the weight coefficients through backpropagation to obtain the first neural network model.

[0101] Specifically, divide the feature data set into a training data subset, a validation data subset, and a test data subset according to the ratio of 7:2:1. The training data subset is used for parameter optimization of the model, the validation data subset is used to adjust hyperparameters and prevent overfitting, and the test data subset is used to evaluate the generalization performance of the final model. Assume the feature data set is , where represents the th input feature, represents the corresponding calibration label. Then the training data subset contains 70% of the data, the validation data subset contains 20% of the data, and the test data subset contains 10% of the data, that is:

[0102] ;

[0103] To prevent the data from having sequential dependence during training, randomly shuffle the training data subset to ensure that the model does not generate learning biases due to the data input order. Use the random permutation method for Shuffle to make the data sequence randomly distributed and obtain the preprocessed training data. Construct the first neural network model, where the first branch uses an eight-layer fully connected neural network, namely the parameter calibration network. The first three layers of this network perform feature extraction and dimensionality reduction. The first layer contains 1024 neurons, the second layer contains 512 neurons, and the third layer contains 256 neurons. Its input is feature data , and map it to the hidden feature space through linear transformation and non-linear activation function:

[0104] ;

[0105] ;

[0106] ;

[0107] Among them, are the hidden layer outputs of the first three layers respectively, are the weight matrices respectively, are the bias terms respectively. ReLU is used as the non-linear activation function, which helps to improve the expression ability of the model and prevent gradient disappearance. In the fourth and fifth layers, the self-attention mechanism is introduced to calculate the weighted relationship matrix between the input features and enhance the model's ability to focus on key features. For the input matrix (output from the third layer), define the query matrix , key matrix , and value matrix :

[0108] ;

[0109] Among them, is the trainable parameter matrix, calculate the attention weight matrix:

[0110] ;

[0111] Among them, is the dimension of the key matrix, and finally calculate the weighted output:

[0112] ;

[0113] And set a residual connection between the fourth and fifth layers to ensure gradient flow and improve the training stability of the model. The sixth to eighth layers are used for further feature extraction, containing 256, 128, and 64 neurons respectively, and finally obtain the calibrated parameter output:

[0114] ;

[0115] Among them, the last layer uses a linear activation function to ensure that the output can be adaptively adjusted within a continuous numerical range. At the same time, the second branch constructs a three-layer LSTM network, namely the environmental compensation network, which is used to process environmental feature data. LSTM is a neural network suitable for time series data. Assuming that the environmental feature input is , where represents the environmental data at the -th time step, the update method of the hidden state of the LSTM is as follows:

[0116] ;

[0117] ;

[0118] ;

[0119] ;

[0120] ;

[0121] Among them, are the activation values of the forget gate, input gate and output gate respectively, is the cell state, is the hidden state. Each layer of LSTM contains 128 hidden units, and a dropout layer is added between layers to prevent overfitting. Finally, the environmental features at different times are weighted through a temporal attention mechanism:

[0122] ;

[0123] ;

[0124] to obtain the environmental compensation output. In the final fusion stage, a weighted sum operation is used to fuse the calibration parameter output and the environmental compensation output :

[0125] ;

[0126] Among them, is a trainable weight parameter, which is adaptively updated through backpropagation to minimize the loss function:

[0127] ;

[0128] Finally, the first neural network model is obtained. Its dual-branch structure can comprehensively consider the calibration parameters of radio signals and environmental compensation factors, thereby improving the measurement accuracy and adapting to different measurement environments.

[0129] Among them, an eight-layer fully-connected network structure of the first branch is constructed, including: performing collaborative domain alignment on the preprocessed training data, calculating the first-order moment and second-order moment statistical features of the data to obtain the source domain feature distribution; calculating the marginal probability density of the source domain feature distribution based on the kernel density estimation method to obtain the marginal distribution matrix; performing conditional probability calculation on the marginal distribution matrix, using the KL divergence to measure the distribution difference between different domains to obtain the conditional distribution matrix; inputting the marginal distribution matrix and the conditional distribution matrix into the joint optimizer, updating the network parameters by minimizing the distribution difference to obtain the domain alignment parameters; adding a partial shrinkage layer to the 4th - 5th layers of the eight-layer fully-connected network, calculating the Jacobian matrix of the hidden layer activation value with respect to the input to obtain the feature sensitivity matrix; calculating the Frobenius norm based on the feature sensitivity matrix and adding it as a regularization term to the loss function to obtain the shrinkage constraint term; according to the domain alignment parameters and the shrinkage constraint term, performing staged optimization on the weights and biases of the eight-layer fully-connected network, fixing the parameters of the partial shrinkage layer in the pre-training stage, and unfreezing and updating all parameters in the fine-tuning stage to obtain the optimized parameter configuration; applying the optimized parameter configuration to the eight-layer fully-connected network to reconstruct the network to obtain a parameter calibration network with domain migration ability.

[0130] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0131] Input the validation data subset into the first neural network model, calculate the relative frequency deviation for the frequency measurement data, calculate the absolute deviation for the power measurement data, and calculate the root mean square error of each frequency offset point for the phase noise measurement data to obtain the error values of each parameter;

[0132] Construct a loss function according to the output result of the first neural network model, assign a first weight coefficient to the frequency error term, assign a second weight coefficient to the power error term, assign a third weight coefficient to the phase noise error term, and assign a fourth weight coefficient to the harmonic error term to obtain a weighted loss function;

[0133] Add a physical constraint term to the weighted loss function to obtain a constrained loss function;

[0134] Based on the constrained loss function and the error values of each parameter, use the Adam optimization algorithm for parameter optimization to obtain the optimized parameter configuration;

[0135] Perform iterative training on the first neural network model to obtain the trained model parameters, and update the first neural network model based on the trained model parameters to obtain the second neural network model.

[0136] Specifically, the validation data subset is input into the first neural network model to evaluate its actual error situation in radio metrology calibration. The frequency measurement data is analyzed, and the relative frequency deviation is calculated to quantify the error of the model in frequency calibration. Assuming the true reference frequency is , and the frequency value predicted by the neural network is , then the relative frequency deviation is defined as:

[0137] ;

[0138] This value is used to measure the deviation ratio between the predicted frequency and the true frequency. If is too large, it indicates that there are significant errors in the model's frequency calibration and optimization is required. At the same time, the power measurement data is analyzed, and the absolute deviation is calculated to measure the power calibration ability of the model. Assuming the true power value is , and the predicted power of the neural network is , then the absolute deviation is calculated as follows:

[0139] ;

[0140] This index reflects the absolute error between the predicted power and the true power. A smaller indicates that the power prediction of the model is more accurate, while a larger means that the power compensation ability of the calibration network needs to be improved. In terms of phase noise measurement, the root mean square error is calculated for multiple frequency offset points to measure the phase noise compensation accuracy of the model. For the frequency offset set , if the predicted phase noise of the neural network is , and the true measurement value is , then the root mean square error is defined as follows:

[0141] ;

[0142] This error reflects the phase noise compensation ability of the model at different frequency offset points. A lower RMSE value means a better calibration effect. After obtaining the error values of each parameter, a loss function is constructed to guide the optimization process of the neural network. Since different error terms have different importance in the calibration process, corresponding weights are assigned to different error terms. Let the weight of the frequency error term be , the weight of the power error term be , the weight of the phase noise error term be , and the weight of the harmonic error term be , then the weighted loss function is defined as follows:

[0143] ;

[0144] Among them, represents the order harmonic power, is the true measured value, which is used to measure the optimization ability of the model for harmonic error. After constructing the weighted loss function, in order to ensure that the optimization process conforms to the physical laws of radio measurement, a physical constraint term is added to construct a constrained loss function. For example, in order to ensure that the calibration result does not exceed the capabilities of the measurement device, a frequency error constraint is added:

[0145] ;

[0146] where is the maximum allowable relative frequency error of the device. The power error also needs to be physically constrained:

[0147] ;

[0148] where is the maximum allowable error of power measurement. The phase noise compensation is also restricted within an acceptable range:

[0149] ;

[0150] The constrained loss function is expressed as:

[0151] , ;

[0152] Among them, is the constraint penalty coefficient. If the error exceeds the allowable range, the value of the loss function will increase significantly to prompt the optimization process to meet the physical constraint conditions. During the optimization process, the Adam optimization algorithm is used to update the neural network parameters :

[0153] ;

[0154] ;

[0155] ;

[0156] Among them, is the first-order momentum estimate, is the second-order momentum estimate, is the learning rate, is a stable term. The Adam optimization algorithm can effectively accelerate convergence and adapt to optimization problems with different gradient change rates. Through multiple iterative trainings, the loss function of the neural network gradually converges, and finally the trained model parameters are obtained. The optimized parameters are used to update the first neural network model so that it can perform radio metrology calibration more precisely, and a second neural network model is obtained.

[0157] In a specific embodiment, performing the radio metrology calibration method based on machine learning further includes the following steps:

[0158] Based on the signal source performance data during actual use, the signal source performance data includes long-term stability data, environmental adaptability data, and reliability data. When the cumulative new data volume exceeds 10% of the training data subset, a model update trigger signal is obtained;

[0159] Fix the parameters of the first five layers of the parameter calibration network in the second neural network model, including 1024, 512, 256 neurons in the first to third layers and the parameters of the self-attention layer with a dimension of 256×256 in the fourth to fifth layers, to obtain a fixed parameter layer;

[0160] Update the parameters of the last three layers of the parameter calibration network, including the parameters of 256, 128, 64 neurons in the sixth to eighth layers, and at the same time keep the parameters of the three-layer LSTM network in the environmental compensation network unchanged, to obtain an updatable parameter layer;

[0161] Input the test data subset into the updated neural network model for calibration accuracy evaluation to obtain a calibration accuracy evaluation result;

[0162] Adjust the updatable parameter layer based on the calibration accuracy evaluation result, and use the incremental learning method to finely tune the parameters, while maintaining the feature extraction ability learned in the fixed parameter layer, to obtain an adjusted neural network model;

[0163] Combine the fixed parameter layer and the adjusted updatable parameter layer to obtain a target calibration model.

[0164] Specifically, optimize the neural network model based on the signal source performance data during actual use, and ensure the long-term stability and environmental adaptability of radio metrology calibration. Continuously monitor the long-term stability data, environmental adaptability data, and reliability data of the signal source to update the model when the data distribution changes. Set the initial size of the training data subset to , when the new data volume satisfies:

[0165] ;

[0166] That is, when the cumulative newly added data volume exceeds 10% of the training data subset, a model update signal is triggered to ensure that the model can continuously adapt to the new measurement environment and improve the calibration accuracy. After the model update is triggered, partial updates are performed on the parameter calibration network in the second neural network model. To avoid affecting the original feature extraction ability, the parameters of the first five layers of the parameter calibration network are fixed, including 1024, 512, and 256 neurons in the first to third layers, and the self-attention layer parameters of 256×256 dimensions in the fourth to fifth layers. These layers are used for basic feature extraction and global attention calculation, and their parameters have learned effective feature mappings on large-scale training data, so they do not need to be adjusted frequently. Let the parameters of the original model be , and the parameter set of the first five layers is:

[0167] ;

[0168] During the model update process, these parameters are frozen, that is:

[0169] ;

[0170] Ensure that gradient updates are not performed on these parameters during training. At the same time, the model still needs to adapt to the new data distribution, so the parameters of the last three layers of the parameter calibration network are updated, that is, the parameters of 256, 128, and 64 neurons in the sixth to eighth layers. These layers are responsible for the final calibration parameter prediction. Let the parameter set of the last three layers be:

[0171] ;

[0172] During the update process, these parameters participate in gradient calculation and weight adjustment to adapt to the new measurement data distribution. At the same time, to ensure that the compensation ability of environmental factors is not affected, the parameters of the three-layer LSTM network in the environmental compensation network remain unchanged, that is:

[0173] ;

[0174] Among them, represents the parameter set of the LSTM network. After the parameter update is completed, the calibration accuracy of the model is evaluated. The test data subset is input into the updated neural network model, and the calibration error is calculated. For frequency measurement data, calculate the relative frequency deviation:

[0175] ;

[0176] Among them, is the predicted frequency of the neural network, is the reference measurement frequency. For power measurement data, calculate the absolute deviation:

[0177] ;

[0178] Among them, is the power value predicted by the model, while is the true power measurement value. For phase noise, calculate the root mean square error at multiple frequency offset points:

[0179] ;

[0180] Among them, is the predicted phase noise, is the true phase noise measurement value. Take the error calculation result as the basis for calibrating the accuracy evaluation. After obtaining the calibration accuracy evaluation result, adjust the updatable parameter layer and use the incremental learning method to fine-tune the parameters to ensure that the model can adapt to new data without forgetting the existing knowledge. Incremental learning optimizes through:

[0181] ;

[0182] Among them, is the learning rate, is the loss function, is the gradient of the updatable parameter. To prevent the model from forgetting old data, the incremental learning process uses a regularization method, such as:

[0183] ;

[0184] Among them, is the parameter before update, is the regularization coefficient, and this regularization term is used to maintain the consistency between the old and new models. Combine the fixed parameter layer and the adjusted updatable parameter layer to form a new target calibration model:

[0185] ;

[0186] This target calibration model not only inherits the effective feature extraction ability of the previous training, but also adapts to the new data distribution through incremental learning, enabling the model to still maintain a high measurement accuracy and environmental adaptability during long-term use, and can further optimize its calibration ability through a continuous data update mechanism in the future.

[0187] In a specific embodiment, the process of inputting the test data subset into the updated neural network model for calibration accuracy evaluation and obtaining the calibration accuracy evaluation result may specifically include the following steps:

[0188] Perform Fourier transform on the frequency measurement data in the test data subset, extract frequency components in the frequency domain space, calculate the power spectral density for each frequency component, and obtain the frequency stability evaluation data;

[0189] Based on the output of the parameter calibration network of the first neural network model, self-attention weighting is performed on the frequency stability evaluation data, and the frequency accuracy score is calculated through a weight matrix of 256×256 dimensions to obtain the frequency calibration evaluation parameter;

[0190] Calculate the root mean square value of noise at multiple frequency offset points for the phase noise measurement data in the test data subset, and calculate the phase noise compensation coefficient according to the prediction result of the second neural network model to obtain the phase noise evaluation parameter;

[0191] Perform time series correlation analysis on the output results of the three-layer LSTM unit of the environmental compensation network and the environmental parameter data in the test data subset, and calculate the influence coefficients of the temperature change rate and humidity change rate on the calibration result to obtain the environmental compensation evaluation parameter;

[0192] Perform adaptive weighting on the frequency calibration evaluation parameter, phase noise evaluation parameter, and environmental compensation evaluation parameter to obtain the initial evaluation score;

[0193] Based on the output of the fusion layer of the dual-branch neural network model, perform residual compensation on the initial evaluation score to obtain the compensated evaluation score, and perform physical constraint verification on the compensated evaluation score to obtain the calibration accuracy evaluation result.

[0194] Specifically, perform Fourier transform on the frequency measurement data in the test data subset, and extract frequency components in the frequency domain space to calculate the power spectral density and obtain the frequency stability evaluation data. Obtain the time-domain measurement data, assuming the frequency measurement signal is , where is the time variable, and this signal is transformed to the frequency domain through the discrete Fourier transform (DFT):

[0195] ;

[0196] Among them, represents the th discrete frequency component, is the signal length, is the sampling interval. Calculate the power spectral density (PSD) of each frequency component:

[0197] ;

[0198] Among them, represents the power density of the th frequency component, which is used to quantify the energy distribution of the signal at different frequency points. By analyzing , evaluate the stability of the frequency measurement data and obtain the frequency stability evaluation data. Based on the output of the parameter calibration network of the first neural network model, perform self-attention weighting on the frequency stability evaluation data to calculate the frequency accuracy score. Let the parameter calibration output of the neural network be , where is the feature dimension, calculate the query matrix , key matrix and value matrix :

[0199] ;

[0200] Among them, is the trainable parameter matrix, and then calculate the attention weight matrix:

[0201] ;

[0202] Among them, is the dimension of the key matrix, and finally calculate the weighted frequency accuracy score:

[0203] ;

[0204] Among them, represents the frequency calibration evaluation parameter, which is used to quantify the optimization effect of the neural network on the frequency stability. At the same time, calculate the root mean square value of the noise at multiple frequency offset points for the phase noise measurement data in the test data subset, and calculate the phase noise compensation coefficient according to the prediction result of the second neural network model to obtain the phase noise evaluation parameter. Let the phase noise measurement data be , where is the set of frequency offset points:

[0205] ;

[0206] Among them, is the phase noise value predicted by the neural network model, and the phase noise compensation coefficient is calculated by the neural network:

[0207] ;

[0208] The phase noise evaluation parameter is expressed as:

[0209] ;

[0210] Among them, It reflects the optimization effect of the calibration model on phase noise. To quantify the influence of environmental factors on measurement accuracy, the output results of the three-layer LSTM units of the environmental compensation network are analyzed for temporal correlation with the environmental parameter data in the test data subset, and the influence coefficients of the temperature change rate and humidity change rate on the calibration results are calculated to obtain the environmental compensation evaluation parameters. Let the environmental characteristic data be , and the LSTM calculation is as follows:

[0211] ;

[0212] where is the hidden state of the LSTM unit, and the influence coefficients of temperature and humidity on the calibration results are calculated using temporal correlation analysis:

[0213] ;

[0214] ;

[0215] where and reflect the influence of temperature and humidity on the measurement parameter , thus forming the environmental compensation evaluation parameters. Adaptive weighting is performed on the frequency calibration evaluation parameters, phase noise evaluation parameters, and environmental compensation evaluation parameters to obtain the initial evaluation score. Let the weighting parameter be , then:

[0216] ;

[0217] where represents the initial evaluation score. After obtaining the initial evaluation score, based on the output of the fusion layer of the dual-branch neural network model, residual compensation is performed on the initial evaluation score, and finally the compensated evaluation score is calculated:

[0218] ;

[0219] where is the residual compensation weight, and is the residual term calculated by the neural network. Physical constraint verification is performed on the compensated evaluation score to ensure that the error meets the radio measurement standard:

[0220] ;

[0221] where is the allowable measurement error threshold to ensure that the final calibration accuracy evaluation result meets the accuracy requirements of the measurement device.

[0222] The above described the radio metrology calibration method based on machine learning in the embodiments of the present application. Next, the radio metrology calibration system 10 based on machine learning in the embodiments of the present application will be described. Please refer to Figure 2 An embodiment of the radio metrology calibration system 10 based on machine learning in the embodiments of the present application includes:

[0223] A measurement module 11, configured to measure parameters of a radio signal source to obtain an original measurement data set;

[0224] An extraction module 12, configured to extract frequency accuracy features, power flatness features, phase noise features, harmonic ratio features, and environmental change features based on the original measurement data set to obtain a feature data set;

[0225] A construction module 13, configured to divide the feature data set into a training data subset, a validation data subset, and a test data subset, and construct a first neural network model including a dual-branch neural network based on the training data subset;

[0226] An optimization module 14, configured to input the validation data subset into the first neural network model for model parameter optimization to obtain a second neural network model.

[0227] Through the collaborative cooperation of the above-mentioned various components, by constructing a dual-branch neural network structure, parameter calibration and environmental compensation are respectively modeled, effectively reducing the complexity of the model, while improving the calculation accuracy of calibration parameters and the accuracy of environmental compensation. The network structure design adopting the self-attention mechanism and residual connection enhances the model's feature extraction ability for measurement data at different frequency points and different power points, while improving the model's convergence performance. The LSTM network structure is introduced to model environmental parameters, combined with the temporal attention mechanism, accurately capturing the dynamic impact of environmental parameter changes on measurement results, and realizing precise control of environmental compensation. A loss function and optimization strategy based on physical constraints are designed to ensure that the calibration results meet the physical property requirements of the signal source, and to avoid the model outputting unreasonable calibration parameters. Through the incremental learning method and hierarchical parameter update strategy, online optimization of the model is achieved, maintaining the model's memory ability for the learned features, while improving the model's adaptability to new data.

[0228] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described systems, systems, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0229] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0230] As described above, the above embodiments are only used to illustrate the technical solution of this application, rather than to limit it; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of various embodiments of this application.

Claims

1. A radio metrology calibration method based on machine learning, characterized in that: The method comprises: Perform parameter measurement on the radio signal source to obtain an original measurement data set; specifically, the method includes: setting multiple discrete frequency points for the radio signal source, and performing repeated frequency measurement on each discrete frequency point to obtain frequency measurement data, and scanning the output power of the signal source, performing repeated measurement on each power point to obtain power measurement data; performing repeated measurement on the signal source at a preset frequency deviation point based on a phase noise analyzer to obtain phase noise measurement data, and performing harmonic measurement on the signal source to obtain harmonic measurement data; recording ambient temperature, humidity, and atmospheric pressure parameters to obtain ambient parameter data; and performing time synchronization on the frequency measurement data, the power measurement data, the phase noise measurement data, the harmonic measurement data, and the ambient parameter data based on a unified timestamp format to obtain an original measurement data set; Based on the original measurement data set, frequency accuracy features, power flatness features, phase noise features, harmonic ratio features and environmental change features are extracted to obtain a feature data set; The feature data set is divided into a training data subset, a verification data subset and a test data subset, and a first neural network model including a dual-branch neural network is constructed based on the training data subset; specifically comprising: dividing the feature data set into a training data subset, a verification data subset and a test data subset in a ratio of 7:2:1, randomly shuffling the training data subset to obtain preprocessed training data; constructing an eight-layer fully connected network structure of the first branch, wherein the number of neurons in layers 1-3 are 1024, 512, and 256, respectively, and layers 4-5 use a self-attention mechanism, the weight matrix dimension is 256×256, and the number of neurons in layers 6-8 are 256, 128, and 64, respectively, to obtain a parameter calibration network; constructing A three-layer LSTM network structure of the second branch is constructed, each layer contains 128 hidden units, the input is environmental feature data, and the output is an environmental compensation coefficient, so as to obtain an environmental compensation network; a ReLU activation function is used in the 1st to 3rd layers of the parameter calibration network, a residual connection is set between the self-attention layers of the 4th and 5th layers, and a linear activation function is used in the last layer to obtain a calibration parameter output; a dropout layer is set between the three-layer LSTM units of the environmental compensation network, and environmental features at different times are weighted by a temporal attention mechanism to obtain an environmental compensation output; the calibration parameter output is fused with the environmental compensation output based on a weighted sum operation, and the weight coefficient is adaptively updated by back propagation to obtain a first neural network model; The verification data subset is input into the first neural network model to optimize model parameters to obtain a second neural network model.

2. The radio metrology calibration method based on machine learning according to claim 1, characterized in that: The extracting frequency accuracy features, power flatness features, phase noise features, harmonic ratio features and environmental change features based on the original measurement data set to obtain a feature data set includes: Performing anomaly detection on the original measurement data set based on the 3σ criterion to obtain an outlier labeling result, and performing normalization processing on non-abnormal data in the original measurement data set to obtain normalized measurement data; Extracting frequency characteristics from the normalized measurement data, calculating frequency accuracy, frequency stability and frequency resolution, analyzing long-term drift trends and short-term jitter characteristics for each frequency measurement point, and obtaining frequency characteristic data; Extracting multiple frequency deviation point noise values ​​from the phase noise data in the normalized measurement data to obtain phase noise characteristic data, and calculating harmonic levels and ratios thereof from the harmonic data in the normalized measurement data to obtain harmonic characteristic data; The environmental parameters in the normalized measurement data are subjected to time-frequency analysis by wavelet transform, the correlation coefficients between the temperature change rate, the humidity change rate and the measurement parameters are calculated to obtain environmental characteristic data, and the frequency characteristic data, the phase noise characteristic data, the harmonic characteristic data and the environmental characteristic data are combined into a characteristic data set.

3. The radio metrology calibration method based on machine learning according to claim 2, characterized in that: The step of inputting the verification data subset into the first neural network model to optimize model parameters to obtain a second neural network model comprises: Inputting the verification data subset into the first neural network model, calculating the relative frequency deviation of the frequency measurement data, calculating the absolute deviation of the power measurement data, and calculating the root mean square error of each frequency deviation point of the phase noise measurement data to obtain the error value of each parameter; Constructing a loss function according to the output result of the first neural network model, assigning a first weight coefficient to the frequency error term, assigning a second weight coefficient to the power error term, assigning a third weight coefficient to the phase noise error term, and assigning a fourth weight coefficient to the harmonic error term, to obtain a weighted loss function; Adding a physical constraint term to the weighted loss function to obtain a constrained loss function; Based on the constrained loss function and the error values ​​of each parameter, the Adam optimization algorithm is used to perform parameter optimization to obtain an optimized parameter configuration; The first neural network model is iteratively trained to obtain trained model parameters, and the first neural network model is updated based on the trained model parameters to obtain a second neural network model.

4. The radio metrology calibration method based on machine learning according to claim 3 is characterized in that: The radio metrology calibration method based on machine learning also includes: Based on the signal source performance data in actual use, the signal source performance data includes long-term stability data, environmental adaptability data and reliability data, when the cumulative amount of new data exceeds 10% of the training data subset, a model update trigger signal is obtained; Fixing the parameters of the first five layers of the parameter calibration network in the second neural network model, including 1024, 512, and 256 neurons in the 1st to 3rd layers and the self-attention layer parameters of the 4th to 5th layers with a dimension of 256×256, to obtain a fixed parameter layer; The parameters of the last three layers of the parameter calibration network are updated, including the 256, 128, and 64 neuron parameters of the 6th to 8th layers, while the parameters of the three-layer LSTM network of the environmental compensation network are kept unchanged to obtain an updateable parameter layer; Inputting the test data subset into the updated neural network model to perform calibration accuracy evaluation to obtain a calibration accuracy evaluation result; The updateable parameter layer is adjusted based on the calibration accuracy evaluation result, and the parameters are fine-tuned using an incremental learning method to maintain the learned feature extraction capability in the fixed parameter layer, thereby obtaining an adjusted neural network model; The fixed parameter layer and the adjusted updateable parameter layer are combined to obtain a target calibration model.

5. The radio metrology calibration method based on machine learning according to claim 4, characterized in that: The step of inputting the test data subset into the updated neural network model to perform calibration accuracy evaluation to obtain a calibration accuracy evaluation result includes: Performing Fourier transformation on the frequency measurement data in the test data subset, extracting frequency components in the frequency domain space, calculating the power spectrum density for each frequency component, and obtaining frequency stability evaluation data; Based on the parameter calibration network output of the first neural network model, self-attention weighting is performed on the frequency stability evaluation data, and a frequency accuracy score is calculated through a weight matrix of 256×256 dimensions to obtain a frequency calibration evaluation parameter; Calculating the noise root mean square values ​​of multiple frequency deviation points for the phase noise measurement data in the test data subset, and calculating the phase noise compensation coefficient according to the prediction result of the second neural network model to obtain the phase noise evaluation parameter; Perform time series correlation analysis on the output results of the three-layer LSTM unit of the environmental compensation network and the environmental parameter data in the test data subset, calculate the influence coefficients of the temperature change rate and the humidity change rate on the calibration result, and obtain the environmental compensation evaluation parameters; Adaptively weighting the frequency calibration evaluation parameter, the phase noise evaluation parameter, and the environment compensation evaluation parameter to obtain an initial evaluation score; Based on the fusion layer output of the dual-branch neural network model, residual compensation is performed on the initial evaluation score to obtain a compensated evaluation score, and physical constraint verification is performed on the compensated evaluation score to obtain a calibration accuracy evaluation result.

6. A radio metrology calibration system based on machine learning, characterized in that: For executing the radio metrology calibration method based on machine learning as described in any one of claims 1 to 5, the radio metrology calibration system based on machine learning comprises: A measurement module, used to measure parameters of a radio signal source to obtain an original measurement data set; An extraction module, used to extract frequency accuracy features, power flatness features, phase noise features, harmonic ratio features and environmental change features based on the original measurement data set to obtain a feature data set; A construction module, used for dividing the feature data set into a training data subset, a verification data subset and a test data subset, and constructing a first neural network model including a two-branch neural network based on the training data subset; The optimization module is used to input the verification data subset into the first neural network model to optimize the model parameters to obtain a second neural network model.

Images