Environment Adaptive Radio Measurement and Calibration Method and System

Abstract

The present invention relates to an environment - adaptive radio metrology calibration method and system. The method: synchronously samples the temperature, humidity, air pressure, and electromagnetic interference around the radio device to obtain an environmental parameter vector; constructs a linear model and a non - linear target model with a three - layer neural network structure according to the environmental parameter vector; uses the linear model and the non - linear target model to solve the environmental parameter separation vector set; performs a one - dimensional iterative search optimization on the environmental parameter separation vector set to obtain a target environmental parameter vector; inputs the target environmental parameter vector into the non - linear target model for calibration compensation calculation to obtain dynamic calibration compensation data. The present invention reduces the update frequency when the environment is stable, which not only ensures the calibration accuracy but also optimizes the use of system resources, and is particularly suitable for resource - constrained wireless devices.

Description

Technical Field

[0001] The present invention relates to the technical field of radio metrology and calibration, and particularly to an environment-adaptive radio metrology and calibration method and system. Background Art

[0002] Radio metrology and calibration is a key technical link to ensure the stable performance of devices in the next-generation wireless communication system. However, factors such as temperature, humidity, air pressure, and electromagnetic interference in the actual environment have a significant impact on the calibration accuracy. With the continuous improvement of performance requirements for 5G and future communication technologies, traditional calibration technologies are difficult to adapt to the challenges brought by the changing environment. Especially in complex scenarios, the coupling effect and non-linear influence among environmental factors cause the calibration error rate to increase sharply, resulting in a decline in communication quality and a reduction in system reliability.

[0003] Traditional radio metrology and calibration methods mainly adopt three-dimensional (3D) multi-parameter compensation technology to perform compensation by establishing a mapping relationship between environmental factors and calibration parameters. However, this method has extremely high computational complexity, reaching the level of O(N 3 ), which is difficult to implement in resource-constrained real-time systems. Although the recently proposed two-dimensional (2D) search alternative reduces the computational burden to a certain extent, in scenarios where environmental factors are interrelated, it still faces the problem of relatively high computational complexity, and the accuracy often cannot meet the requirements. Summary of the Invention

[0004] The main object of the present invention is to provide an environment-adaptive radio metrology and calibration method and system. When the environment is stable, the present invention reduces the update frequency, which not only ensures the calibration accuracy but also optimizes the use of system resources, and is particularly suitable for resource-constrained wireless devices.

[0005] To achieve the above object, the present invention provides an environment-adaptive radio metrology and calibration method, including the following steps:

[0006] Synchronously sample the temperature, humidity, air pressure, and electromagnetic interference around the radio device to obtain an environmental parameter vector;

[0007] Construct a linear model and a non-linear target model with a three-layer neural network structure according to the environmental parameter vector;

[0008] Use the linear model and the non-linear target model to solve the environmental parameter separation vector set;

[0009] Perform one-dimensional iterative search optimization on the environmental parameter separation vector set to obtain a target environmental parameter vector;

[0010] Input the target environmental parameter vector into the non-linear target model for calibration compensation calculation to obtain dynamic calibration compensation data.

[0011] The present invention also provides an environment - adaptive radio metrology calibration system, including:

[0012] A synchronous sampling module, configured to synchronously sample the temperature, humidity, air pressure, and electromagnetic interference around the radio device to obtain an environmental parameter vector;

[0013] A construction module, configured to construct a linear model and a non - linear target model with a three - layer neural network structure according to the environmental parameter vector;

[0014] A solution module, configured to solve an environmental parameter separation vector set by using the linear model and the non - linear target model;

[0015] A search and optimization module, configured to perform one - dimensional iterative search and optimization on the environmental parameter separation vector set to obtain a target environmental parameter vector;

[0016] A calculation module, configured to input the target environmental parameter vector into the non - linear target model for calibration compensation calculation to obtain dynamic calibration compensation data.

[0017] In summary, the technical solution provided by the present invention, through the oblique projection operator and one - dimensional iterative search technology, reduces the calibration calculation complexity from O(N 3 ) of the traditional 3D method to O(K·n·N), where K is the number of iterations, n is the number of environmental parameters, and N is the number of sampling points for each - dimensional search, significantly reducing the system resource consumption. By using the oblique projection operator technology, the accurate separation of the influences of different environmental factors is realized. Especially for the interactive influence of temperature and humidity, a special processing unit is designed, so that the independent influences of each parameter in a complex environment are accurately quantified, and the problem that traditional methods are difficult to handle the coupling of environmental factors is solved. Through the dual - model cooperation mechanism of providing an initial estimate by the linear model and performing accurate calculation by the non - linear target model, both the calculation speed and the calibration accuracy are ensured. While maintaining similar performance to the state - of - the - art methods, the calculation resource consumption is significantly reduced. Based on the dynamic calibration update period adjustment mechanism of the environmental change rate, the update frequency is automatically increased when the environment changes rapidly and decreased when the environment is stable, which not only ensures the calibration accuracy but also optimizes the system resource usage, and is especially suitable for resource - constrained wireless devices. Through the environmental parameter subset selection technology of sensitivity matrix analysis and information entropy criterion, the environmental parameters that have the most significant impact on the calibration accuracy are automatically identified and focused on, further reducing the calculation burden while maintaining the calibration accuracy of key parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a schematic diagram of the steps of the environment - adaptive radio metrology calibration method in an embodiment of the present invention;

[0019] Figure 2It is a structural block diagram of an environment - adaptive radio metrology calibration system in an embodiment of the present invention.

[0020] The realization of the object of the present invention, functional features and advantages will be further described with reference to the embodiments and the accompanying drawings. Specific embodiments

[0021] In order to make the object, technical solution and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0022] Referring to Figure 1 , this embodiment provides an environment - adaptive radio metrology calibration method, including the following steps:

[0023] S1, synchronously sample the temperature, humidity, air pressure and electromagnetic interference around the radio device to obtain an environmental parameter vector;

[0024] Among them, temperature sensors, humidity sensors, barometric pressure sensors, and electromagnetic interference sensors are arranged at different positions of the radio device to construct an environmental parameter acquisition network. The layout of the sensors needs to consider the structural characteristics of the device and environmental impacts to ensure that the collected data can comprehensively and accurately reflect the environmental state around the device, while avoiding local measurement errors caused by unreasonable sensor positions. A unified clock signal is set for the environmental parameter acquisition network, so that all sensors collect data at a fixed sampling frequency to ensure the synchronization and time alignment characteristics of the data. To meet the real-time calibration requirements of the radio device, the setting of the sampling frequency needs to balance data accuracy and system resource consumption, and select an appropriate fixed sampling frequency, such as ten times per second, to ensure accurate capture of the dynamic changes of environmental parameters and reduce calibration errors caused by asynchronous sampling. The collected original environmental parameter data contains noise and outliers and needs to be preprocessed to improve data quality. For temperature data, since the environmental temperature is affected by sudden fluctuations in a short period of time, low-pass filtering technology is used to remove high-frequency noise interference to ensure the smoothness and accuracy of temperature data. Select an appropriate low-pass filter and set an appropriate cut-off frequency so that it can effectively filter out irrelevant high-frequency components while retaining the main trend of temperature changes. For humidity data and barometric pressure data, since these two types of environmental parameters are affected by factors such as air flow and device heat dissipation, filtering algorithms are applied respectively to remove interference signals. Specific filtering methods are used for humidity data so that it can accurately reflect the changes in environmental humidity, while mean filtering is used for the processing of barometric pressure data to reduce errors caused by short-term fluctuations and make the data more stable and reliable. The processing of electromagnetic interference data is more complex because there are various types of electromagnetic interference signals around the radio device, including periodic interference, transient interference, and sudden abnormal signals. When processing electromagnetic interference data, a filtering method that can retain the main energy components is used to remove outliers and retain useful interference information, thus avoiding unnecessary impacts on radio calibration. After the filtering processing of different types of environmental parameters is completed, all data is normalized to eliminate the differences in the numerical ranges between various parameters so that it can be used for subsequent environmental impact modeling and compensation calculations on a unified scale. Map all parameters to the same numerical range to make it more suitable as input data. After normalization, temperature, humidity, barometric pressure, and electromagnetic interference data form a standardized environmental parameter vector.

[0025] S2. Construct a linear model and a non-linear target model with a three-layer neural network structure according to the environmental parameter vector;

[0026] Specifically, perform a linear transformation on the environmental parameter vector to construct a linear model structure and establish a linear mapping relationship between environmental parameters and calibration compensation. Define an environmental impact coefficient matrix and a bias vector. By performing a linear operation on the environmental parameter vector, the model can preliminarily estimate the impact of different environmental parameters on the calibration compensation value. Each element of the environmental impact coefficient matrix corresponds to the degree of influence of a certain environmental parameter on a specific calibration compensation value, while the bias vector is used to adjust the reference value of the model output to compensate for existing systematic errors. Optimize and train the environmental impact coefficient matrix and the bias vector of the linear model framework by the least squares method to make the output of the model as close as possible to the actual compensation value of the historical calibration data. The optimization process of the least squares method minimizes the sum of the squares of the calibration errors to solve the optimal coefficient matrix and bias vector, obtaining a linear model. Construct a non-linear target model with a three-layer neural network architecture according to the dimension of the environmental parameter vector to capture the complex non-linear relationship between environmental parameters and calibration deviations. To ensure that the architecture of the neural network adapts to the characteristics of environmental parameters, the number of input layer nodes of the neural network is set to the number of environmental parameters, so that each environmental parameter can be transmitted as an input signal to the subsequent layers of the network. The number of nodes in the hidden layer is set to twice the number of environmental parameters plus one to provide sufficient computing power while avoiding waste of computing resources caused by excessive complexity, and the number of nodes in the output layer is equal to the number of calibration parameters, so that the final output of the network can be directly used for calibration compensation of radio equipment. During the construction of the non-linear target model, consider the interactive effects between environmental parameters, especially the coupling effect of temperature and humidity on calibration accuracy. Therefore, additional interactive effect processing units are introduced into the three-layer neural network architecture. By adding cross terms to the hidden layer, the model captures the non-linear correlation between temperature and humidity, improving the adaptability to complex environmental conditions. The introduction of cross terms enables the network to not only learn the influence of individual environmental parameters but also identify the trend of calibration deviation changes under the interaction of multiple parameters, thereby improving the accuracy of overall calibration compensation. Optimize the non-linear target model framework by setting a fixed learning rate and training termination conditions so that the parameters of the model can accurately fit the historical calibration data of the radio equipment. During the optimization process, set a fixed learning rate to ensure that the model can converge stably during training, avoiding oscillations caused by too high a learning rate and preventing too slow a convergence rate caused by too low a learning rate. At the same time, set appropriate training termination conditions to prevent overfitting or underfitting. The training termination conditions include that the change rate of the loss function is lower than a certain threshold or the number of training epochs reaches the set maximum value. During the training process, adjust the weights and biases of the neural network by the batch gradient descent algorithm to make the output of the network match the compensation value of the historical calibration data as much as possible, and finally obtain an optimized non-linear target model.

[0027] S3. Solve the environmental parameter separation vector set by using the linear model and the non-linear target model;

[0028] It should be noted that the linear model is used to conduct preliminary mapping analysis on the environmental parameter vector, generate initial independent impact estimates of the environmental parameters, and construct the basic structure of the oblique projection operator based on this. The linear model provides a quick preliminary estimate of the impact of environmental parameters to reduce the computational complexity and accelerate the convergence rate of subsequent iterative optimization. After obtaining the preliminary estimate, combined with the mapping characteristics of the nonlinear target model, an expression of the oblique projection operator with a residual term is constructed to accurately characterize the nonlinear impact of environmental parameters. The oblique projection operator needs to use the output of the linear model as the initial value and correct it through nonlinear calculations to obtain a complete projection operator calculation framework, enabling it to more accurately describe the individual impact of environmental parameters. To optimize the computational accuracy of the oblique projection operator, the mapping difference of the environmental parameters before and after projection is calculated using the nonlinear target model, and the Lagrange multiplier method and conjugate gradient descent algorithm are used to minimize this difference to obtain the optimized residual term. The core of the optimization process is to ensure that the projection operator can accurately separate the impacts of different environmental parameters on calibration compensation, so that even when multiple factors such as temperature, humidity, air pressure, and electromagnetic interference are coupled with each other, the independent impact of a single environmental parameter can still be obtained. Since the nonlinear target model has a stronger expression ability, it can capture the changing trends of more complex environmental factors during the optimization process and make necessary corrections to the initial estimate of the linear model to improve the accuracy of the final result. Based on the optimized residual term, corresponding projection operators are constructed for temperature, humidity, air pressure, and electromagnetic interference respectively based on the nonlinear target model and the complete projection operator calculation framework. Since the linear model provides a preliminary estimate of the environmental parameters, the output of the linear model is used as the initial value in this process to accelerate the convergence process of the nonlinear projection operator. In this way, the computational complexity is reduced, and at the same time, the optimization process of the projection operator is ensured to be more stable. As the projection operators for different environmental parameters are constructed, an initial set of projection operators is formed. This set contains a separate description of the impacts of different environmental parameters and can reduce the mutual coupling impact between environmental parameters to a certain extent. The improved Gram-Schmidt orthogonalization process is applied to the initial set of projection operators to eliminate the mutual interference between different projection operators. The nonlinear target model is used to verify the effect of orthogonalization to ensure that all projection operators can maintain orthogonality and will not affect each other during application. Through the orthogonalization process, the cross-impact between environmental parameters is reduced, so that each projection operator only retains the independent impact of a specific environmental parameter and will not be interfered by other parameters, improving the accuracy of the final separation result. After completing the orthogonalization, an orthogonal set of oblique projection operators is obtained. The environmental parameter vector is input into the orthogonal set of oblique projection operators, and the nonlinear target model is applied for projection mapping verification to ensure that the obtained separated vector accurately reflects the impact of a single environmental parameter on calibration compensation.In this process, the non-linear target model further corrects the result of the projection mapping through the feedback of the calibration data, and ensures that each separated vector only contains the influence of the corresponding environmental parameter without being interfered by other parameters. The independent influences of each environmental parameter are extracted, and a set of separated vectors of environmental parameters with high precision is obtained.

[0029] S4. Perform one-dimensional iterative search optimization on the set of separated vectors of environmental parameters to obtain the target environmental parameter vector.

[0030] Specifically, set the calculation result of the linear model as the initial environmental parameter vector, and set a reasonable search range for each parameter in this initial vector to ensure that the search range covers the possible true values of the environmental parameters. After setting the search range, uniformly generate a fixed number of candidate values within this range to form a set of parameter candidate values, so that each candidate value represents a possible state of the current environmental parameter. To ensure the uniformity of the search, the candidate values are evenly distributed within the set search range to cover a sufficient number of possible values, while avoiding being too sparse resulting in insufficient search accuracy or being too dense resulting in excessive computational burden. After completing the construction of the set of parameter candidate values, combine each candidate value with other unchanged parameters and input them into the non-linear target model to calculate the corresponding calibration compensation value and obtain the candidate calibration result. Compare the candidate calibration result with the calibration deviation measured actually to evaluate the quality of each candidate value. Measure the degree of deviation by calculating the mean square error corresponding to each candidate value to obtain the error evaluation result. The candidate value with a smaller mean square error means that the corresponding environmental parameter value is closer to the real situation. Therefore, in the next optimization process, select the parameter candidate value with the smallest error from the error evaluation result and update the current environmental parameter with it to obtain the optimized environmental parameter vector after the current iteration. Since only one environmental parameter is optimized in each iteration while keeping other parameters unchanged, the stability of the search process is ensured and the optimal solution is gradually approached. After completing the update of the environmental parameters, perform a termination judgment on the optimized environmental parameter vector of the current iteration to decide whether to continue the iterative optimization. The termination judgment conditions include two aspects. One is whether the difference between the optimization result of the current iteration and that of the previous iteration is less than the preset threshold. If the optimized environmental parameter vector changes very little compared with the previous result, it is determined that it is close to the convergence state and the iteration is terminated. The other is to judge whether the maximum number of iterations has been reached. If the number of iterations has reached the set upper limit, the search is terminated to prevent an infinite loop. During the iteration process, gradually reduce the search step size to improve the search accuracy, so that a more refined search can be performed when approaching the optimal solution to obtain the target environmental parameter vector.

[0031] S5. Input the target environmental parameter vector into the non-linear target model for calibration compensation calculation to obtain the dynamic calibration compensation data.

[0032] Among them, the target environmental parameter vector is input into the non-linear target model to calculate the compensation amount for the calibration deviation of the radio device. The non-linear target model calculates the compensation values of multiple calibration parameters, including frequency compensation value, power compensation value, and phase compensation value, according to the different influence degrees of environmental parameters, combined with historical calibration data and optimized model weights, so as to form a calibration compensation vector. Since the error characteristics of radio devices are different under different environmental conditions, when calculating the calibration compensation vector, corresponding weight coefficients are assigned according to the importance of different calibration parameters, so that the compensation values output by the model can accurately match the actual working state of the device, and ensure that the adjustment ranges of different calibration parameters meet the device requirements, thus achieving precise compensation. To ensure that the compensation data can adapt to the dynamic changes of the environment in real time, each parameter in the environmental parameter vector is continuously sampled, and its time derivative is calculated to obtain the change rate of the environmental parameters. By analyzing the time series data of the target environmental parameters, the temperature change rate, humidity change rate, air pressure change rate, and electromagnetic interference change rate are calculated to obtain the environmental change rate vector, which reflects the instantaneous change trend of the environmental parameters and provides a basis for the dynamic adjustment of the calibration compensation values, so that the compensation strategy can be adaptively adjusted according to the environmental changes. The environmental change rate vector is compared with a preset change rate threshold to determine whether the calibration compensation value update condition is triggered, and an update trigger signal is generated. To avoid frequent adjustment of the calibration compensation values in a relatively stable environment, a reasonable change rate threshold is set, so that the compensation update is only triggered when the environmental change exceeds the set range, thus ensuring the stability and computational efficiency of the system. If any item in the environmental change rate vector exceeds the threshold, the calibration update signal is triggered, so that the compensation data can be adjusted according to the latest environmental parameters to ensure that the calibration accuracy of the radio device is not affected by environmental fluctuations. During the update process of the calibration compensation values, the maximum relative change rate is calculated according to the environmental change rate vector, and an adaptive adjustment formula is constructed using an exponential decay function to determine the dynamic calibration update period. Since different environmental parameters have different change rates, when calculating the dynamic calibration update period, the maximum relative change rates of each parameter are comprehensively considered, and the compensation update frequency is adjusted in an exponential decay manner, so that the update frequency is increased when the environment changes rapidly, and the update frequency is decreased when the environment is stable, thus optimizing the allocation of the system's computational resources and avoiding unnecessary compensation adjustments. At the same time, through the construction of the adaptive adjustment formula, it is ensured that the update frequency of the calibration compensation data is always consistent with the environmental change rate, so that the adjustment of the compensation values is neither too frequent nor lag behind the changes of the actual environment. After determining the dynamic calibration update period, the calibration compensation vector is updated at a fixed time according to this period, and the update frequency is automatically increased when the environment changes rapidly and appropriately decreased when the environment is stable, so as to obtain calibration compensation data with optimized time series.Apply the calibration compensation data with timing optimization to the calibration parameters of the radio device, and perform real-time compensation adjustments on frequency, power, and phase to ensure that the device always maintains the optimal working state under different environmental conditions. Through the dynamic calibration compensation mechanism, improve the adaptability of the radio device in complex environments, reduce measurement errors caused by environmental changes, enhance the accuracy and reliability of radio metrology, enable the device to always maintain a high-precision calibration state during long-term operation, and obtain dynamic calibration compensation data.

[0033] Normalize and compare the compensated radio parameter vector with the standard parameter vector, and calculate the ratio of the Euclidean distance between the two to the norm of the standard vector, which is used as the measurement standard for the calibration error rate. The purpose of this step is to quantify the current calibration compensation effect, enabling subsequent optimization adjustments to be optimized based on specific error feedback, thereby enhancing the overall system's adaptability. After calculating the calibration error rate, calculate the change gradient of each calibration parameter in the calibration compensation vector with respect to each environmental parameter, and construct an absolute value matrix of partial derivatives based on this to quantify the impact of environmental factors on calibration accuracy. This matrix is the sensitivity matrix. The construction of the sensitivity matrix can provide an analysis of the impact of environmental parameters, enabling the system to accurately identify which environmental factors have a greater impact on radio calibration results and how different calibration parameters change under the influence of environmental factors. After the sensitivity matrix calculation is completed, perform a weighted summation operation on the importance weights of the calibration parameters and the corresponding elements in the sensitivity matrix to calculate the contribution degree of each environmental parameter to the overall calibration accuracy and obtain the environmental parameter impact factor. Since different calibration parameters have different importance for the stability and accuracy of the radio device, when calculating the environmental parameter impact factor, consider the relative weights of the calibration parameters to ensure that the final calculation result can accurately reflect the impact of environmental factors on the overall calibration accuracy. Sort the environmental parameter impact factors in descending order and select the environmental parameters with the top rankings to form a candidate subset, obtaining the preliminary selection result of environmental parameters. Based on the preliminary selection result of environmental parameters, perform information entropy analysis to measure the amount of information contained in the candidate subset, and construct an optimization objective function in combination with a penalty term for the subset size to reduce unnecessary computational complexity while ensuring calibration accuracy. The role of information entropy analysis is to evaluate the effective information content of environmental parameters and eliminate redundant information, while the introduction of the subset size penalty term is to prevent the selection of too many environmental parameters, resulting in an increase in computational complexity. After constructing the optimization objective function, determine the size of the optimal subset by minimizing the overall evaluation index to obtain the optimal environmental parameter subset. Re-execute the design of the oblique projection operator and one-dimensional iterative search optimization for the optimal environmental parameter subset to construct the target calibration system of the radio device. By re-designing the oblique projection operator, ensure that the impact of each environmental parameter is accurately separated in the non-linear target model, and at the same time, combine the one-dimensional iterative search optimization method to improve calibration accuracy and reduce computational complexity.

[0034] In one example, the temperature, humidity, air pressure, and electromagnetic interference around the radio device are synchronously sampled to obtain an environmental parameter vector, including:

[0035] Temperature sensors, humidity sensors, air pressure sensors, and electromagnetic interference sensors are arranged at different positions of the radio device to form an environmental parameter acquisition network;

[0036] A unified clock signal is set for the environmental parameter acquisition network and synchronous data acquisition is performed at a fixed sampling frequency to obtain the original environmental parameter data;

[0037] For the temperature data in the original environmental parameter data, a low-pass filter is applied to remove high-frequency noise interference to obtain temperature filtered data;

[0038] For the humidity data and air pressure data in the original environmental parameter data, a filtering algorithm is respectively executed to remove interference signals to obtain humidity filtered data and air pressure filtered data;

[0039] For the electromagnetic interference data in the original environmental parameter data, an energy component retention type filtering is applied to remove outliers to obtain electromagnetic interference filtered data;

[0040] The temperature filtered data, humidity filtered data, air pressure filtered data, and electromagnetic interference filtered data are normalized to obtain an environmental parameter vector.

[0041] In this example, an environmental parameter acquisition network is established. The temperature sensor, humidity sensor, air pressure sensor, and electromagnetic interference sensor are reasonably arranged according to the structural characteristics, operating characteristics of the radio device, and the spatial distribution characteristics of the environmental conditions. For example, the temperature sensor is arranged near the device's power amplifier to capture the heat changes generated during the operation of the device in real time; the humidity sensor is placed near the device's ventilation opening to monitor the humidity change trend inside the device; the air pressure sensor is usually arranged at the top of the device or in the air intake area to monitor the instantaneous fluctuations of air pressure; and the electromagnetic interference sensor is arranged near the signal input and output ports or sensitive circuit areas to capture the changes in the electromagnetic environment that cause interference to the device in real time. A unified high-precision clock signal source is introduced into the acquisition network, and this signal source is distributed to all sensor nodes in the network simultaneously, enabling all sensors to perform precise synchronous data acquisition according to the same clock signal and at a uniformly set fixed sampling frequency, ensuring the data time sequence alignment and consistency among the environmental parameters, and obtaining the original environmental parameter data. Different parameter characteristics in the original environmental parameter data are processed separately. For example, for temperature data, which is affected by high-frequency interference (such as wind speed fluctuations or local instantaneous heat generation inside the device), a high-order Butterworth low-pass filter algorithm is used to suppress the high-frequency noise components in the original sensor data that exceed a specific cut-off frequency, and extract the temperature filtered data that can truly reflect the environmental temperature change trend. Similarly, for humidity and air pressure data, due to their susceptibility to external random interference and the instability of the sensors themselves, different algorithms such as Chebyshev filtering and sliding window filtering are used respectively to retain the low-frequency change trend of the data itself and remove the high-frequency random interference, obtaining stable humidity and air pressure filtered data. For electromagnetic interference data, due to its inherent complexity, which manifests as irregular and sudden abnormal pulses or continuous periodic interference, traditional filtering methods are difficult to effectively remove these abnormal interference signals. For such data, an energy component retention type filtering algorithm is adopted, such as decomposition and reconstruction through an improved wavelet transform method:

[0042]

[0043] where represents the data reconstruction result after electromagnetic interference filtering, represents the wavelet coefficient of the th layer after wavelet transform, represents the energy level of the th layer coefficient, represents the scale range mainly containing energy after decomposition, The energy threshold is obtained by training based on historical data and is used to screen out abnormal wavelet components. This method can effectively extract the main electromagnetic interference signals and exclude occasional outliers, obtaining pure electromagnetic interference filtered data. Normalize the temperature filtered data, humidity filtered data, air pressure filtered data, and electromagnetic interference filtered data to unify the dimension and eliminate the difference in numerical scales between different data. Use the dynamic adaptive normalization formula:

[0044]

[0045] where, represents the th environmental parameter after normalization, is the original value of the th environmental parameter after filtering, and are respectively the long-term average value and variance of the th parameter in the historical data, is a small value to prevent division by zero, is an adaptive adjustment factor determined by training with historical data and is used to adjust the sensitivity during normalization. The processed and normalized temperature parameter , humidity parameter , air pressure parameter and electromagnetic interference parameter are combined to form the final environmental parameter vector .

[0046] In an example, construct a linear model and a non-linear target model with a three-layer neural network structure based on the environmental parameter vector, including:

[0047] Perform a linear transformation on the environmental parameter vector and construct a linear model structure. Establish a linear mapping relationship between the environmental parameters and the calibration compensation by setting the environmental impact coefficient matrix and the bias vector to obtain the linear model framework;

[0048] Optimize and train the environmental impact coefficient matrix and the bias vector of the linear model framework by the least squares method to obtain the linear model;

[0049] Construct a three-layer neural network architecture according to the dimension of the environmental parameter vector, with the structure of setting the number of input layer nodes equal to the number of environmental parameters, the number of hidden layer nodes equal to twice the number of environmental parameters plus one, and the number of output layer nodes equal to the number of calibration parameters;

[0050] Add an interaction impact processing unit for temperature and humidity to the three-layer neural network architecture. Capture the coupling effect between environmental parameters by additionally setting cross terms in the three-layer neural network architecture to obtain the non-linear target model framework;

[0051] The nonlinear target model framework is optimized by setting a fixed learning rate and training termination conditions to obtain the nonlinear target model.

[0052] In this example, a linear transformation method is adopted to establish a basic linear model framework. Assuming the environmental parameter vector consists of environmental parameters, such as temperature, humidity, air pressure, and electromagnetic interference, etc., then an environmental impact coefficient matrix (with dimension ) and a bias vector (with dimension ) are constructed to establish the calculation formula for the calibration parameter vector :

[0053]

[0054] where represents the calibration compensation vector, and each of its elements corresponds to different calibration parameters of the radio device, such as frequency compensation, power compensation, and phase compensation. The element of the matrix represents the influence degree of the th environmental parameter on the th calibration compensation parameter, and the bias vector is used to correct the system error not considered due to environmental factors. To optimize this linear model framework so that it can accurately reflect the influence of environmental parameters on calibration compensation, optimization training is carried out based on the least squares method. Assuming that the historical measurement data provides a set of known environmental parameters and their corresponding true calibration compensations , then the optimization goal is to minimize the error:

[0055]

[0056] where represents the number of samples of the historical data, and represents the Euclidean norm. By solving this optimization goal, the optimal environmental impact coefficient matrix and the bias vector are obtained to form the optimized linear model. After completing the construction of the linear model, considering that the influence of environmental parameters on calibration compensation has significant nonlinear characteristics, for example, the coupling effect of temperature and humidity will cause complex changes in the parameters of the radio device, so a three-layer neural network architecture is constructed to establish the nonlinear target model. The number of input layer nodes of this neural network is equal to the number of environmental parameters , the number of output layer nodes is equal to the number of calibration parameters , and the number of nodes in the hidden layer is set to , to provide sufficient non - linear fitting ability while avoiding the increase in computational cost caused by over - complication of the model. To capture the interaction effects between environmental parameters such as temperature and humidity, cross - terms are introduced in the hidden layer. Assuming that the temperature parameter and the humidity parameter have a synergistic effect, an additional cross - term is introduced in the calculation formula of the hidden layer:

[0057]

[0058] where, represents the output of the th node in the hidden layer, is the activation function (such as ReLU), is the weight vector, is the weight factor of the temperature - humidity interaction, is the bias term. This design can enhance the model's ability to express the non - linear coupling effect of environmental factors, making the model's prediction more accurate under different environmental conditions. When training the neural network, to ensure the stability and convergence of the model, a fixed learning rate is set and a training termination condition is introduced, for example:

[0059]

[0060] where, is the calibration compensation value predicted by the model, is the set error tolerance. By setting a reasonable learning rate and termination conditions, gradient oscillation or over - fitting problems during the training process are prevented, so as to ensure that the finally obtained non - linear target model can accurately describe the impact of environmental parameters on the calibration compensation of radio equipment.

[0061] In an example, solving the environmental parameter separation vector set using a linear model and a non - linear target model includes:

[0062] Using the linear model to conduct a preliminary mapping analysis of the environmental parameter vector, generating an initial independent influence estimate of the environmental parameters, and obtaining the basic structure of the oblique projection operator;

[0063] Combining the mapping characteristics of the non - linear target model to construct an expression of the oblique projection operator with a residual term, introducing the output of the linear model as the initial value into the non - linear calculation, and obtaining a complete projection operator calculation framework;

[0064] Using the non - linear target model to calculate the mapping difference before and after the projection of the environmental parameters, and using the Lagrange multiplier method and the conjugate gradient descent algorithm to minimize the difference to obtain an optimized residual term;

[0065] Based on a non - linear target model and a complete projection operator calculation framework, corresponding projection operators are constructed for temperature, humidity, air pressure, and electromagnetic interference respectively. The initial estimation provided by the linear model is used to accelerate the convergence process, and an initial set of projection operators is obtained;

[0066] The initial set of projection operators is subjected to improved Gram - Schmidt orthogonalization. The non - linear target model is used to verify the orthogonal effect, eliminating the mutual interference between projection operators, and an orthogonalized set of oblique projection operators is obtained;

[0067] The environmental parameter vector is input into the orthogonalized set of oblique projection operators, and the non - linear target model is applied for projection mapping verification to extract the independent effects of each environmental parameter, obtaining an environmental parameter separation vector set.

[0068] In this example, a linear model is used to perform a preliminary mapping analysis on the environmental parameter vector to generate an initial estimate of the independent effects of environmental parameters, and based on this, the basic structure of the oblique projection operator is established. The environmental parameter vector is set as , where represents temperature, represents humidity, represents air pressure, represents electromagnetic interference, and the calibration compensation vector represents the compensation values that the radio device needs to adjust, such as frequency, power, and phase, etc. The linear mapping model is used:

[0069]

[0070] where, is the environmental impact coefficient matrix, indicating the linear impact of each environmental parameter on calibration compensation, is the bias vector, representing the compensation term for the inherent error of the system. Through this linear model, a preliminary impact estimate of each environmental parameter is obtained, forming a linear decomposition of the environmental parameter impact. However, due to the non - linear characteristics of the impact of environmental parameters on radio calibration compensation, combined with the mapping characteristics of the non - linear target model, an expression of the oblique projection operator with a residual term is constructed to optimize the estimation of the independent effects of environmental parameters. The non - linear target model is defined as:

[0071]

[0072] where, is the non - linear mapping function, indicating the non - linear relationship between environmental parameters and calibration compensation, and is the residual term, representing the non - linear error part that the linear model fails to accurately describe. In order to introduce non - linear calculation using the output of the linear model as the initial value, the oblique projection operator is constructed:

[0073]

[0074] Among them, represents the projection operator of the th environmental parameter, is the independent influence value of the environmental parameter estimated by the linear model, while represents the residual term of the environmental parameter in the non-linear mapping, is the residual adjustment factor, indicating the magnitude of non-linear correction. In order to optimize the oblique projection operator so that it can more accurately separate the independent influence of environmental parameters, calculate the mapping difference before and after the projection of environmental parameters, and optimize it through the Lagrange multiplier method and the conjugate gradient descent algorithm to minimize the projection error. Define the optimization objective:

[0075]

[0076] Among them, is the th historical environmental parameter sample, represents the calculation result of the non-linear target model after applying the projection operator, while represents the calculation result under the original environmental parameter input. Solve this optimization problem through the Lagrange multiplier method, and combine the conjugate gradient descent algorithm to adjust to minimize the error, obtain the optimized residual term, and further improve the calculation framework of the projection operator. Based on the optimized non-linear projection operator, construct corresponding projection operator sets for temperature, humidity, air pressure, and electromagnetic interference respectively. Since the linear model provides a preliminary estimate of environmental parameters, the output of the linear model is used as the initial projection value to accelerate the convergence of the non-linear optimization process. Thereby effectively reducing the computational complexity, enabling the projection operator to converge to the optimal solution faster, and forming the initial projection operator set. Implement the improved Gram-Schmidt orthogonalization process on the initial projection operator set to eliminate the mutual influence between projection operators. Assume that the initial projection operator set is , where represents the projection operator of a certain environmental parameter, then process it through the following orthogonalization formula:

[0077]

[0078] Among them, represents the orthogonalized projection operator, while represents the inner product relationship between two projection operators. In this way, ensure the orthogonality between different projection operators, thereby avoiding cross-interference of environmental parameter influences and improving the separation accuracy. After completing the orthogonalization, the environmental parameter vector Input the set of orthonormalized oblique projection operators, and apply the non - linear target model for projection mapping verification to ensure that the independent effects of each environmental parameter are correctly extracted. The independent effect values of the extracted environmental parameters form the environmental parameter separation vector set:

[0079]

[0080] Among them, respectively represent the independent effect values of temperature, humidity, air pressure, and electromagnetic interference, without the interference of other parameters.

[0081] In an example, perform a one - dimensional iterative search optimization on the environmental parameter separation vector set to obtain the target environmental parameter vector, including:

[0082] Set the calculation result of the linear model as the initial environmental parameter vector, and set a search interval for each parameter in the initial environmental parameter vector. Generate a fixed number of uniformly distributed candidate values within the search interval to obtain the parameter candidate value set;

[0083] Combine each candidate value in the parameter candidate value set with other unchanged parameters and input them into the non - linear target model to calculate the corresponding calibration compensation value, obtaining the candidate calibration results;

[0084] Compare the candidate calibration results with the calibration deviation of the actual measurement, calculate the mean square error of each candidate value, and obtain the error evaluation result;

[0085] Select the parameter candidate value corresponding to the minimum error from the error evaluation result, update the current environmental parameter, and obtain the optimized environmental parameter vector of the current iteration;

[0086] Perform an iteration termination judgment on the optimized environmental parameter vector of the current iteration. When the difference between the results of two adjacent iterations is less than the preset threshold or the maximum number of iterations is reached, end the iteration. At the same time, gradually reduce the search step size to obtain the target environmental parameter vector.

[0087] In this example, use the linear model to calculate the initial environmental parameter vector to provide a starting point for the optimization search. Assume the environmental parameter vector is , where represents temperature, represents humidity, represents air pressure, represents electromagnetic interference, then calculate its initial estimate through the linear model:

[0088]

[0089] Among them, is the environmental impact coefficient matrix, representing the influence relationship of environmental parameters on calibration compensation, is the bias term used to compensate for system errors. Through this linear model, an initial environmental parameter vector is obtained , which is used as the starting point for the optimization iteration. After obtaining the initial environmental parameter vector, a search interval is set for each parameter in the vector to ensure that the optimization process is carried out within a reasonable range. The way to set the search interval is based on the distribution characteristics of historical data and combined with the allowable range of environmental parameter changes in the system. For example, if the temperature parameter has a small variation range in the past data, its search interval is set narrower, while if the electromagnetic interference parameter varies greatly, a wider search interval is set. To generate candidate values uniformly distributed within the search interval, the following formula is used:

[0090]

[0091] where represents the th candidate value of the th environmental parameter, is the estimated value of this parameter in the previous iteration, is the search step size, is the number of candidate values. This formula ensures that the candidate values are uniformly distributed within the set interval to provide comprehensive optimization options. After generating the set of parameter candidate values, each candidate value in it is combined with other unchanged parameters and input into the non-linear objective model to calculate the corresponding calibration compensation value. The non-linear objective model is expressed as:

[0092]

[0093] where is the calibration compensation vector, represents the non-linear objective model, which can accurately describe the influence of environmental parameters on radio calibration compensation. After calculating each group of candidate parameters , a set of candidate calibration results is obtained. To evaluate the quality of these candidate calibration results, the mean square error between them and the actual measured calibration deviation is calculated. The mean square error is defined as:

[0094]

[0095] where is the actually measured calibration compensation value, is the candidate calibration value calculated by the non-linear objective model, is the number of calibration parameters. The mean square error It reflects the fitting degree of candidate environmental parameters to calibration compensation. The smaller the error, the closer the environmental parameter value is to the true situation. After completing the error evaluation, select the candidate value with the smallest mean square error from the error evaluation results and update the current environmental parameter to converge it to the optimal solution. The update formula is:

[0096]

[0097] where, is the environmental parameter value after the current iterative update, and this parameter value corresponds to the candidate value with the smallest mean square error. After each iteration, perform a termination judgment on the optimization result to decide whether to continue the optimization. The termination conditions include two aspects. One is whether the difference between the environmental parameter vector of the current iteration and the result of the previous iteration is less than the preset threshold:

[0098]

[0099] where, is the termination threshold, which is used to control the optimization accuracy. If this condition is met, stop the iteration and output the final optimized environmental parameter vector. The other termination condition is that the number of iterations reaches the maximum value , to prevent the optimization process from falling into an infinite loop. To improve the optimization accuracy, gradually reduce the search step size in each iteration, so that the optimization process conducts a large-scale search in the initial stage and more refined fine-tuning in the later stage. The update strategy of the search step size adopts exponential decay:

[0100]

[0101] where, is the search step size at the th iteration, is the initial search step size, is the decay factor. This method ensures that in the later stage of optimization, the search step size gradually converges, thereby improving the accuracy of the final environmental parameter vector. Through the above steps, the target environmental parameter vector is obtained, which can accurately reflect the true environmental state around the radio device and provide the optimal compensation parameters for radio metrology calibration.

[0102] In an example, input the target environmental parameter vector into the non-linear target model for calibration compensation calculation to obtain dynamic calibration compensation data, including:

[0103] Input the target environmental parameter vector into the non-linear target model for compensation calculation, generate frequency compensation values, power compensation values and phase compensation values according to the different weight coefficients of multiple calibration parameters, and obtain the calibration compensation vector;

[0104] Calculate the time derivative of the continuous sampling value of each parameter in the environmental parameter vector, generate the temperature change rate, humidity change rate, air pressure change rate and electromagnetic interference change rate, and obtain the environmental change rate vector;

[0105] Compare the environmental change rate vector with a preset change rate threshold to determine whether a calibration compensation value update condition is triggered, and obtain an update trigger signal;

[0106] The maximum relative change rate is calculated based on the environmental change rate vector, and an adaptive adjustment formula is constructed using an exponential decay function to obtain a dynamic calibration update cycle;

[0107] The calibration compensation vector is updated regularly according to the dynamic calibration update cycle, the update frequency is increased when the environment changes rapidly, and the update frequency is reduced when the environment is stable, so as to obtain the calibration compensation data with optimized timing;

[0108] The timing-optimized calibration compensation data is applied to the calibration parameters of the radio equipment, and the frequency, power and phase are compensated and adjusted in real time to obtain dynamic calibration compensation data.

[0109] In this example, the target environment parameter vector is input into the nonlinear target model to calculate the calibration compensation value and obtain the calibration compensation vector for radio equipment calibration. The optimized target environment parameter vector is set to ,in represents the optimized temperature, Represents the optimized humidity, Indicates the optimized air pressure, Represents the optimized electromagnetic interference. Nonlinear target model The calibration compensation value is calculated by a deep neural network or a nonlinear mapping method, where the different weight coefficients of each calibration parameter determine its contribution to the final compensation value. The calculation formula is:

[0110]

[0111] in, Represents the calibration compensation vector, including frequency compensation , Power compensation and phase compensation is the calibration weight matrix whose elements Indicates The environmental parameters The degree of influence of the calibration compensation value; is a bias vector used to correct systematic errors. Since the relationship between environmental parameters and calibration compensation is usually non - linear, using a non - linear model for compensation calculation can improve compensation accuracy and ensure the stability of radio equipment under complex environmental conditions. After calculating the calibration compensation vector, in order to ensure that the calibration compensation can respond to environmental changes in a timely manner, the environmental parameter vector is dynamically monitored, and its time derivative is calculated to obtain the environmental change rate vector . By performing time differentiation on the continuous sampled values of environmental parameters, we get:

[0112]

[0113] where, represents the change rate of temperature, represents the change rate of humidity, represents the change rate of air pressure, represents the change rate of electromagnetic interference. The calculation of the environmental change rate uses a sliding window method to reduce the impact of short - term fluctuations on the calculation results. The environmental change rate vector is compared with a preset change rate threshold vector to determine whether to trigger an update of the calibration compensation value. The trigger condition is:

[0114]

[0115] When this condition is met, an update trigger signal is generated, indicating that the current environmental change has exceeded the system - set threshold, meaning that the current calibration compensation is no longer applicable to the current environment and a new calibration calculation is required. After determining the calibration compensation update trigger signal, in order to optimize the timing of calibration updates, the maximum relative change rate needs to be calculated, and an adaptive adjustment formula is constructed based on an exponential decay function to calculate the dynamic calibration update period. The maximum relative change rate is defined as:

[0116]

[0117] Based on this, the dynamic calibration update period is calculated:

[0118]

[0119] where, is the maximum update period, representing the longest calibration interval when the environmental change is extremely slow; is the minimum update period, representing the shortest calibration interval when the environment changes violently; As an adjustment factor, it controls the change speed of the calibration update frequency. This adaptive update strategy ensures that the calibration compensation is updated frequently when the environment changes rapidly, while reducing unnecessary consumption of computing resources when the environment is stable. For example, if a sudden change in temperature and humidity is detected, the calibration compensation value should be adjusted quickly, while when the environment is in a stable state, the calibration update frequency is reduced to reduce energy consumption and computational burden. After determining the dynamic calibration update period, the calibration compensation vector is updated at regular intervals according to this period to optimize the calibration timing. When the environment changes rapidly, the system automatically increases the update frequency of the calibration compensation, enabling the radio device to quickly adapt to the environmental changes; while when the environment is stable, the system reduces the update frequency to reduce unnecessary compensation adjustments and ensure the stable operation of the system. The optimized calibration compensation data is applied to the calibration parameters of the radio device for real-time compensation adjustment of frequency, power, and phase. For frequency compensation, the following dynamic calibration formula is used:

[0120]

[0121] Wherein, is the compensated frequency, is the original frequency, is the temperature-humidity interaction coefficient, is the compensation adjustment factor. Similarly, power compensation and phase compensation are adjusted using corresponding non-linear calculation formulas to ensure that the radio device can maintain high-precision measurement performance under different environmental conditions.

[0122] In one example, the environment-adaptive radio metrology calibration method further includes:

[0123] Normalize and compare the measured radio parameter vector after compensation with the standard parameter vector, and calculate the ratio of the Euclidean distance between the two to the norm of the standard vector to obtain the calibration error rate;

[0124] Calculate the change gradient of each calibration parameter in the calibration compensation vector with respect to each environmental parameter, and construct an absolute value matrix of partial derivatives between the calibration parameters and environmental factors to obtain the sensitivity matrix;

[0125] Perform a weighted summation operation on the corresponding elements of the importance weight of the calibration parameter and the sensitivity matrix, and calculate the contribution degree of each environmental parameter to the overall calibration accuracy to obtain the environmental parameter influence factor;

[0126] Sort the environmental parameter influence factors in descending order, select the environmental parameters with higher rankings to form a candidate subset, and obtain the initial selection result of environmental parameters;

[0127] Based on the initial selection result of environmental parameters, perform information entropy analysis, construct an optimization objective function in combination with the subset size penalty term, and determine the optimal subset size by minimizing the overall evaluation index to obtain the optimal environmental parameter subset;

[0128] Re - execute the design of the skew projection operator and the one - dimensional iterative search optimization for the optimal subset of environmental parameters to construct the target calibration system of the radio device.

[0129] In this example, the compensated radio parameter vector and the standard parameter vector are normalized and compared, and the ratio of the Euclidean distance between the two and the norm of the standard vector is calculated to obtain the calibration error rate. Suppose the measured radio parameter vector is , where represents the frequency compensation value, represents the power compensation value, represents the phase compensation value, and the standard parameter vector represents the compensation value that the system should have in an ideal environment. To measure the compensation error, normalization is performed:

[0130]

[0131] where, and represent the mean and standard deviation of the standard parameter vector respectively. Calculate their normalized Euclidean distance:

[0132]

[0133] Calculate the calibration error rate:

[0134]

[0135] where, reflects the deviation degree between the current calibration parameters and the standard parameters. The smaller the value, the higher the calibration accuracy. After calculating the calibration error rate, analyze the influence of environmental parameters on calibration compensation. Therefore, calculate the change gradient of each calibration parameter in the calibration compensation vector with respect to each environmental parameter, and construct the absolute value matrix of partial derivatives between the calibration parameters and environmental factors, that is, the sensitivity matrix . Suppose the environmental parameter vector is , where represents the temperature, represents the humidity, represents the air pressure, represents the electromagnetic interference. Then the elements of the sensitivity matrix are defined as:

[0136]

[0137] where, represents the influence of the th environmental parameter on the The sensitivity of a calibration compensation value. By calculating this matrix, the influence intensity of different environmental parameters on calibration compensation is determined. To analyze the contribution of different environmental parameters to calibration accuracy, the importance weights of calibration parameters and the sensitivity matrix are subjected to a weighted summation operation to calculate the influence factor of each environmental parameter:

[0138]

[0139] where, represents the contribution degree of the th environmental parameter to the overall calibration accuracy. After calculating the environmental parameter influence factors, they are sorted in descending order, and the environmental parameters with the top rankings are selected to form a candidate subset, obtaining the preliminary selection result of environmental parameters. After determining the preliminary environmental parameters, to optimize the subset selection, an optimization objective function is constructed based on information entropy analysis and combined with a subset size penalty term to determine the optimal subset size. Let the candidate subset be , then the information entropy is defined as:

[0140]

[0141] where, represents the contribution probability of the th environmental parameter to the calibration error. The optimization objective is to minimize the overall evaluation index:

[0142]

[0143] where, is the subset size penalty factor, which is used to control the scale of the environmental parameter subset, reducing the computational complexity while ensuring calibration accuracy. By solving this optimization problem, the optimal environmental parameter subset is determined. After determining the optimal environmental parameter subset, the tilt projection operator design and one-dimensional iterative search optimization are re-executed for this subset to construct the final radio device target calibration system. The optimization objective of the tilt projection operator is to ensure the maximization of the independent influence of environmental parameters on calibration compensation, so its expression is:

[0144]

[0145] where, represents the projection operator of the th environmental parameter, is the residual term, is the projection weight. To optimize the oblique projection operator, the mean square error of the residual term is minimized by the Lagrange multiplier method, and the Gram - Schmidt orthogonalization method is combined to make the projection operators with different environmental parameters orthogonal, thereby reducing the mutual interference between parameters. After the projection operator is optimized, a one - dimensional iterative search is used to optimize the target environmental parameter vector , and one parameter is optimized in each iteration while keeping other parameters unchanged, and the optimal solution is selected through the mean square error minimization criterion:

[0146]

[0147] where represents the iteration round, represents the true calibration compensation value. The iteration termination condition is:

[0148]

[0149] or reaching the maximum number of iterations . Through the above optimization process, the target calibration system of the radio device is finally constructed, and its optimal calibration accuracy can be ensured under different environmental conditions.

[0150] Referring to Figure 2 , this embodiment provides an environment - adaptive radio metrology calibration system, including:

[0151] Synchronous sampling module 1, used to synchronously sample the temperature, humidity, air pressure and electromagnetic interference around the radio device to obtain an environmental parameter vector;

[0152] Construction module 2, used to construct a linear model and a non - linear target model with a three - layer neural network structure according to the environmental parameter vector;

[0153] Solution module 3, used to solve the environmental parameter separation vector set by using the linear model and the non - linear target model;

[0154] Search and optimization module 4, used to perform one - dimensional iterative search and optimization on the environmental parameter separation vector set to obtain the target environmental parameter vector;

[0155] Calculation module 5, used to input the target environmental parameter vector into the non - linear target model for calibration compensation calculation to obtain dynamic calibration compensation data.

[0156] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to that described in the above method embodiment, and details are not repeated here.

[0157] It should be noted that in this text, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, system, article or method including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, system, article or method. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, system, article or method including such element.

[0158] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. An environment-adaptive radio metrology calibration method, characterized in that, Including: Synchronously sampling the temperature, humidity, air pressure, and electromagnetic interference around the radio device to obtain an environmental parameter vector; Constructing a linear model and a non-linear target model with a three-layer neural network structure according to the environmental parameter vector; Solving the environmental parameter separation vector set by using the linear model and the non-linear target model; specifically including: performing preliminary mapping analysis on the environmental parameter vector by using the linear model to generate an initial independent influence estimate of the environmental parameters and obtaining the basic structure of the oblique projection operator; constructing an oblique projection operator expression with a residual term in combination with the mapping characteristics of the non-linear target model, introducing the output of the linear model as the initial value into the non-linear calculation to obtain a complete projection operator calculation framework; calculating the mapping difference before and after the projection of the environmental parameters by using the non-linear target model, and minimizing the difference by using the Lagrange multiplier method and the conjugate gradient descent algorithm to obtain an optimized residual term; based on the non-linear target model and the complete projection operator calculation framework, constructing corresponding projection operators for temperature, humidity, air pressure, and electromagnetic interference respectively, using the initial estimates provided by the linear model to accelerate the convergence process to obtain an initial set of projection operators; performing improved Gram-Schmidt orthogonalization processing on the initial set of projection operators, and verifying the orthogonal effect by using the non-linear target model to eliminate the mutual interference between the projection operators to obtain an orthogonal set of oblique projection operators; inputting the environmental parameter vector into the orthogonal set of oblique projection operators, and performing projection mapping verification by using the non-linear target model to extract the independent influence of each environmental parameter to obtain the environmental parameter separation vector set; Performing one-dimensional iterative search optimization on the environmental parameter separation vector set to obtain a target environmental parameter vector; Inputting the target environmental parameter vector into the non-linear target model for calibration compensation calculation to obtain dynamic calibration compensation data.

2. The environmental adaptive radio metrology calibration method according to claim 1, wherein, The synchronously sampling the temperature, humidity, air pressure, and electromagnetic interference around the radio device to obtain an environmental parameter vector includes: Arranging temperature sensors, humidity sensors, air pressure sensors, and electromagnetic interference sensors at different positions of the radio device to form an environmental parameter acquisition network; Setting a unified clock signal for the environmental parameter acquisition network and performing synchronous data acquisition at a fixed sampling frequency to obtain raw environmental parameter data; Applying a low-pass filter to the temperature data in the raw environmental parameter data to remove high-frequency noise interference to obtain temperature filtered data; Performing filtering algorithm processing on the humidity data and air pressure data in the raw environmental parameter data respectively to remove interference signals to obtain humidity filtered data and air pressure filtered data; Applying energy component retention type filtering to the electromagnetic interference data in the raw environmental parameter data to remove outliers to obtain electromagnetic interference filtered data; Normalizing the temperature filtered data, the humidity filtered data, the air pressure filtered data, and the electromagnetic interference filtered data to obtain an environmental parameter vector.

3. The environmental adaptive radio metrology calibration method according to claim 1, wherein The constructing a linear model and a non-linear target model with a three-layer neural network structure according to the environmental parameter vector includes: Perform a linear transformation on the environmental parameter vector and construct a linear model structure. Establish a linear mapping relationship between the environmental parameters and the calibration compensation by setting the environmental impact coefficient matrix and the bias vector, and obtain a linear model framework. Optimize and train the environmental impact coefficient matrix and the bias vector of the linear model framework by the least squares method to obtain a linear model. Construct a three-layer neural network architecture according to the dimension of the environmental parameter vector, and set the number of input layer nodes equal to the number of environmental parameters, the number of hidden layer nodes equal to twice the number of environmental parameters plus one, and the number of output layer nodes equal to the number of calibration parameters. Add an interaction effect processing unit for temperature and humidity to the three-layer neural network architecture, and capture the coupling effect between environmental parameters by additionally setting cross terms in the three-layer neural network architecture to obtain a non-linear target model framework. Optimize the non-linear target model framework by setting a fixed learning rate and training termination conditions to obtain a non-linear target model.

4. The environmental adaptive radio metrology calibration method according to claim 1, wherein, The one-dimensional iterative search optimization of the environmental parameter separation vector set to obtain the target environmental parameter vector includes: Set the calculation result of the linear model as the initial environmental parameter vector, set a search interval for each parameter in the initial environmental parameter vector, and generate a fixed number of uniformly distributed candidate values within the search interval to obtain a set of parameter candidate values. Combine each candidate value in the set of parameter candidate values with other unchanged parameters and input them into the non-linear target model to calculate the corresponding calibration compensation value to obtain a candidate calibration result. Compare the candidate calibration result with the actually measured calibration deviation, calculate the mean square error of each candidate value to obtain an error evaluation result. Select the parameter candidate value corresponding to the minimum error from the error evaluation result, update the current environmental parameter to obtain the optimized environmental parameter vector for the current iteration. Perform an iteration termination judgment on the optimized environmental parameter vector for the current iteration. When the difference between the results of two adjacent iterations is less than the preset threshold or the maximum number of iterations is reached, end the iteration, and gradually reduce the search step size at the same time to obtain the target environmental parameter vector.

5. The environmental adaptive radio metrology calibration method according to claim 1, wherein, The input of the target environmental parameter vector into the non-linear target model for calibration compensation calculation to obtain dynamic calibration compensation data includes: Input the target environmental parameter vector into the non-linear target model for compensation calculation, and generate a frequency compensation value, a power compensation value, and a phase compensation value according to the different weight coefficients of multiple calibration parameters to obtain a calibration compensation vector. Calculate the time derivative of the continuous sampling values of each parameter in the environmental parameter vector to generate a temperature change rate, a humidity change rate, a pressure change rate, and an electromagnetic interference change rate to obtain an environmental change rate vector. Compare the environmental change rate vector with a preset change rate threshold to determine whether to trigger the calibration compensation value update condition to obtain an update trigger signal. Calculate the maximum relative change rate based on the environmental change rate vector, and use an exponential decay function to construct an adaptive adjustment formula to obtain a dynamic calibration update period. Timely update the calibration compensation vector according to the dynamic calibration update period, increase the update frequency when the environment changes rapidly, and decrease the update frequency when the environment is stable, so as to obtain calibration compensation data with optimized timing. Apply the calibration compensation data with optimized timing to the calibration parameters of the radio device, and perform real-time compensation adjustment on the frequency, power and phase to obtain dynamic calibration compensation data.

6. The environment adaptive radio metrology calibration method according to claim 5, wherein The environment-adaptive radio metrology calibration method further includes: Normalize and compare the measured radio parameter vector after compensation with the standard parameter vector, and calculate the ratio of the Euclidean distance between the two to the norm of the standard vector to obtain the calibration error rate. Calculate the change gradient of each calibration parameter in the calibration compensation vector with respect to each environmental parameter, and construct an absolute value matrix of partial derivatives between the calibration parameters and the environmental factors to obtain the sensitivity matrix. Perform a weighted summation operation on the importance weights of the calibration parameters and the corresponding elements of the sensitivity matrix, and calculate the contribution degree of each environmental parameter to the overall calibration accuracy to obtain the environmental parameter influence factor. Arrange the environmental parameter influence factors in descending order, select the environmental parameters with higher rankings to form a candidate subset, and obtain the preliminary selection result of environmental parameters. Based on the preliminary selection result of environmental parameters, perform information entropy analysis, construct an optimization objective function in combination with the subset size penalty term, and determine the optimal subset size by minimizing the overall evaluation index to obtain the optimal environmental parameter subset. Re-execute the design of the oblique projection operator and the one-dimensional iterative search optimization for the optimal environmental parameter subset, and construct the target calibration system of the radio device.

7. An environment-adaptive radio metrology calibration system, characterized in that, For implementing the steps of the method according to any one of claims 1 to 6, the environment-adaptive radio metrology calibration system includes: A synchronous sampling module for synchronously sampling the temperature, humidity, air pressure and electromagnetic interference around the radio device to obtain an environmental parameter vector. A construction module for constructing a linear model and a non-linear target model with a three-layer neural network structure according to the environmental parameter vector. A solution module for solving the environmental parameter separation vector set by using the linear model and the non-linear target model. A search optimization module for performing one-dimensional iterative search optimization on the environmental parameter separation vector set to obtain a target environmental parameter vector. A calculation module for inputting the target environmental parameter vector into the non-linear target model for calibration compensation calculation to obtain dynamic calibration compensation data.

Images