Method and device for automatically adjusting temperature compensation parameters of duplexer

Abstract

The invention relates to the technical field of duplexers, and discloses a duplexer temperature compensation parameter automatic adjusting method and device. The method comprises the following steps: performing working temperature sampling on a duplexer to obtain temperature sampling data, and performing distribution feature extraction on the temperature sampling data to obtain temperature feature data; inputting the temperature feature data into a mechanism information neural network for feature reconstruction to obtain temperature compensation features; performing radial basis function neural network mapping on the temperature compensation characteristics to obtain a duplexer frequency drift state quantity; performing sliding window adaptive Kalman filtering operation on the frequency drift state quantity of the duplexer to obtain a compensation parameter; the compensation parameters are input into a feedforward compensation controller, a voltage-controlled resonator of the duplexer is adjusted through a sliding mode control structure, a frequency compensation signal is obtained, and stable compensation of the frequency of the duplexer is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of duplexers, and in particular to a method and a device for automatically adjusting temperature compensation parameters of a duplexer. Background Art

[0002] As a key RF device in mobile communication systems, the frequency characteristics of duplexers are extremely sensitive to temperature changes. With the rapid development of 5G communication technology, the working environment of duplexers has become increasingly complex and the temperature changes dramatically, which puts higher requirements on the temperature compensation of duplexers. Traditional temperature compensation methods mainly rely on fixed compensation strategies, which are difficult to adapt to complex and changing working environments, and the compensation accuracy is limited.

[0003] There are three main problems with the existing duplexer temperature compensation technology: the temperature feature extraction method lacks physical interpretability and it is difficult to accurately describe the impact mechanism of temperature changes on the frequency characteristics of the duplexer; the automatic adjustment algorithm of the compensation parameters is not robust enough and the compensation effect is unstable in the presence of noise interference; the control strategy has weak adaptability and cannot dynamically adjust the compensation parameters according to actual working conditions. Summary of the invention

[0004] The invention provides a method and a device for automatically adjusting the temperature compensation parameters of a duplexer, and the invention realizes the stable compensation of the duplexer frequency.

[0005] In a first aspect, the present invention provides a method for automatically adjusting a temperature compensation parameter of a duplexer, the method comprising: Sampling the working temperature of the duplexer to obtain temperature sampling data, and extracting distribution characteristics of the temperature sampling data to obtain temperature characteristic data; Inputting the temperature characteristic data into the mechanism information neural network for feature reconstruction to obtain temperature compensation characteristics; Performing radial basis function neural network mapping on the temperature compensation characteristics to obtain a frequency drift state quantity of the duplexer; Performing a sliding window adaptive Kalman filter operation on the frequency drift state quantity of the duplexer to obtain a compensation parameter; The compensation parameter is input into a feedforward compensation controller, and the voltage-controlled resonator of the duplexer is adjusted through a sliding mode control structure to obtain a frequency compensation signal.

[0006] In a second aspect, the present invention provides a device for automatically adjusting the temperature compensation parameters of a duplexer, the device comprising: A sampling module is used to sample the working temperature of the duplexer to obtain temperature sampling data, and to extract distribution characteristics of the temperature sampling data to obtain temperature characteristic data; A reconstruction module, used for inputting the temperature characteristic data into the mechanism information neural network for characteristic reconstruction to obtain temperature compensation characteristics; A mapping module, used for performing radial basis function neural network mapping on the temperature compensation characteristics to obtain a frequency drift state quantity of the duplexer; An operation module, used for performing a sliding window adaptive Kalman filter operation on the frequency drift state quantity of the duplexer to obtain a compensation parameter; The adjustment module is used to input the compensation parameter into the feedforward compensation controller, and adjust the voltage-controlled resonator of the duplexer through a sliding mode control structure to obtain a frequency compensation signal.

[0007] The third aspect of the present invention provides a device for automatically adjusting the temperature compensation parameters of a duplexer, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the device for automatically adjusting the temperature compensation parameters of the duplexer executes the above-mentioned method for automatically adjusting the temperature compensation parameters of the duplexer.

[0008] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned method for automatically adjusting the temperature compensation parameters of a duplexer.

[0009] In the technical solution provided by the present invention, the physical mechanism information of temperature compensation is introduced into the feature extraction process through the combination of mechanism information neural network and temperature mechanism dictionary, and a stable state observer is established by adopting radial basis function neural network and Lyapunov stability theory, which can accurately capture the frequency drift characteristics of the duplexer and improve the accuracy of state quantity observation; an adaptive Kalman filtering algorithm is introduced, and the measurement noise is modeled in combination with a Gaussian mixture model, which significantly enhances the compensation robustness of the system in a complex noise environment; a feedforward compensation controller based on sliding mode control is designed, and the control jitter is suppressed by introducing a boundary layer function, so as to achieve smooth compensation of the duplexer frequency; a closed-loop parameter optimization mechanism is constructed, and the neural network is updated online through error feedback, so that the system has continuous learning and self-adaptation capabilities, and the compensation effect of long-term operation is guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0011] Figure 1A schematic diagram of a flow chart of a method for automatically adjusting temperature compensation parameters of a duplexer provided in an embodiment of the present application; Figure 2 A schematic block diagram of the structure of a device for automatically adjusting temperature compensation parameters of a duplexer provided in an embodiment of the present application; Figure 3 A schematic block diagram of the structure of a device for automatically adjusting the temperature compensation parameters of a duplexer provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0013] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change based on actual conditions.

[0014] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0015] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0016] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0017] See also Figure 1 , Figure 1 A flow chart of a method for automatically adjusting the temperature compensation parameters of a duplexer provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method for automatically adjusting the temperature compensation parameters of a duplexer provided in the embodiment of the present application includes steps S100 to S600.

[0018] Step S100, sampling the working temperature of the duplexer to obtain temperature sampling data, and extracting distribution characteristics of the temperature sampling data to obtain temperature characteristic data; It is understandable that the execution subject of the present invention may be a duplexer temperature compensation parameter automatic adjustment device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0019] Specifically, the operating temperature of the duplexer is collected in real time to obtain temperature sampling data. The temperature sampling data is subjected to outliers removal, and the temperature data after outliers removal is normalized to obtain standardized temperature data. The standardized temperature data is input into the Gaussian mixture model, and the temperature change is fitted with probability density using a linear combination of multiple Gaussian components. The Gaussian mixture model is a probability model used to describe data distribution, which adapts to complex temperature changes by linearly combining multiple Gaussian distributions. Through fitting, the basic distribution data of the temperature is obtained, that is, the probability density characteristics of the temperature in different intervals are described. The basic temperature distribution data is calculated by kernel density estimation. Kernel density estimation is a parameter-free method that can more flexibly describe the distribution of data by estimating the probability density using a kernel function. In this process, an appropriate kernel function is selected to perform parameter-free estimation on the probability density of the temperature change to obtain the temperature density distribution data. This step provides a smoother and more detailed probability density description of the temperature data change in a manner that does not rely on specific distribution assumptions. The temperature density distribution data is input into the fuzzy C-means clustering unit based on the Wasserstein distance criterion, and the temperature data is clustered using the fuzzy clustering method. Wasserstein distance is an effective metric for measuring the difference between probability distributions, which can accurately reflect the probability distribution distance between temperature data points. In the fuzzy C-means clustering process, by calculating these distances, the temperature data is effectively clustered, and a temperature data clustering matrix is ​​generated to describe the distribution of temperature data in different feature categories. The temperature data clustering matrix is ​​optimized for the number of components to determine the feature categories of temperature change. The core of the optimization process is to determine the appropriate number of clusters so as to avoid over-clustering or under-clustering while maintaining the clustering precision. Through optimization, reasonable temperature change feature categories are obtained, which can effectively reflect the working status of the duplexer under different temperature conditions. The temperature change feature categories are converted into feature vectors for subsequent numerical processing and analysis. The feature vectors are subjected to dimensionality reduction and orthogonalization. The purpose of dimensionality reduction is to reduce the dimension of the feature vector, remove redundant information, make the data representation more concise, and improve computational efficiency. Orthogonalization ensures that different features are independent of each other and reduces the correlation between features, so as to improve the model's ability to identify data and the analysis effect. By analyzing the temperature change pattern of the orthogonalized feature data, the final temperature feature data is obtained, which contains the key features of the duplexer under different temperature conditions.

[0020] Step S200, inputting temperature characteristic data into the mechanism information neural network for feature reconstruction to obtain temperature compensation characteristics; Specifically, the temperature feature data is input into the encoder for processing. The structure of the encoder consists of four fully connected layers, in which the input dimension of the first fully connected layer is the same as the dimension of the temperature feature data, ensuring that all information of the temperature feature can be fully received. On this basis, the dimensions of the second to fourth fully connected layers in the encoder are reduced to 1 / 2 of the original dimensions in turn, gradually reducing the representation dimension of the data, so that redundant features are compressed. The layer-by-layer dimensionality reduction method can remove unnecessary redundancy while ensuring that important information is retained, thereby improving the generalization ability and computational efficiency of the model. After each fully connected layer, a batch normalization layer and a ReLU activation function are connected. The batch normalization layer normalizes each small batch of data to reduce the internal covariate shift in model training, while the ReLU activation function introduces nonlinear characteristics to improve the model's ability to express complex relationships. Through multi-level encoding operations, intermediate features are obtained. Sparse constraint processing is performed on the intermediate features. The sparse constraint term is constructed using L1 norm regularization, and the sparsity is adjusted by setting the sparse coefficient λ. The L1 norm has the ability to sparse, so that some feature coefficients tend to zero, thereby achieving selective activation in the feature space. Sparse optimization can effectively extract the most critical features for temperature compensation, while suppressing irrelevant or weakly related features, reducing the interference of noise on the model, and obtaining sparse features. At the same time, the physical mechanism information of temperature compensation is constructed as a temperature mechanism dictionary, which contains M atoms, each of which represents a temperature compensation mode, and these modes represent the physical process of compensating the duplexer at different temperatures. In order to ensure the independence of the temperature mechanism dictionary, it is orthogonalized to ensure that each atom in the dictionary is orthogonal to each other, that is, different temperature compensation modes are independent of each other, thereby enhancing the expression ability of the dictionary and obtaining the mechanism dictionary matrix. The sparse features and the mechanism dictionary matrix are feature fused, and the physical mechanism information is embedded in the feature space through the matrix inner product operation to construct the mechanism information retention item. The physical mechanism information is introduced into the feature representation, so that the reconstructed features are not only based on the statistical characteristics of the data, but also contain the physical mechanism of temperature compensation, thereby improving the ability of the features to interpret physical reality and obtaining the first fused features. The first fused features are input into the decoder for processing. The decoder consists of four fully connected layers, of which the dimensions of the first three fully connected layers gradually increase, expanding to twice the original dimensions, in order to gradually restore the information compressed in the previous encoding process. Through layer-by-layer expansion, the expressiveness of the features is restored. At the same time, a batch normalization layer and a ReLU activation function are connected after each fully connected layer. The batch normalization layer also maintains the stability of the data during transmission, while the ReLU activation function introduces nonlinearity so that the decoder can restore more complex data. In the fourth fully connected layer, the output dimension is restored to the same dimension as the temperature feature data to obtain a temperature compensation feature that matches the input.

[0021] Step S300, performing radial basis function neural network mapping on the temperature compensation characteristics to obtain a duplexer frequency drift state quantity; Specifically, a radial basis function neural network structure is constructed, and the number of nodes in the input layer, hidden layer, and output layer of the neural network is set. Among them, the number of nodes in the input layer is consistent with the dimension of the temperature compensation feature, and the number of nodes in the hidden layer is set to N, so as to ensure the expressive power of the network and balance the computational complexity and generalization performance. The number of nodes in the output layer corresponds to the dimension of the frequency drift state of the duplexer to ensure that the network can accurately describe and predict the frequency drift state. After the temperature compensation feature is input into the input layer of the radial basis function neural network, these features are mapped to the hidden layer space by linear mapping. The input data is transformed by using the first weight matrix to map the original temperature compensation feature to a new feature space to obtain the initial hidden layer feature. The Gaussian radial basis function is calculated for the initial hidden layer feature. The Gaussian radial basis function can capture the complex nonlinear relationship between the data by performing nonlinear transformation on the input feature, and obtain a richer feature representation, namely the radial basis function feature. Through this step, the neurons in the hidden layer can form multiple "activation centers" in the feature space, and these activation centers are used to describe the mapping relationship between the temperature compensation feature and the frequency drift of the duplexer. In order to make the neural network adapt to the dynamic changes of the frequency drift of the duplexer under different temperature conditions, the weight matrix of the radial basis function neural network is updated online. An adaptive law is constructed based on the Lyapunov stability criterion. Through this adaptive law, the first weight matrix and the second weight matrix from the hidden layer to the output layer are adjusted in real time. The introduction of the Lyapunov stability criterion ensures that the entire network can maintain stability during the training process and will not diverge, thereby ensuring that the weight update of the network can be carried out in the correct direction at any time, making the model more flexible and accurate in responding to temperature changes, and obtaining the updated network weights. The radial basis function features are linearly combined using the second weight matrix, and the hidden layer features are mapped to the output space by weighted summation to obtain the output layer features. In order to improve the model's ability to describe frequency drift, the output layer features need to be processed by a nonlinear activation function so that the network fits the complex frequency drift law and obtains the activation features. The state observation equation is constructed based on the activation features, and the activation features are linked to the dynamic characteristics of the duplexer. The dynamic behavior of the duplexer frequency drift is described by the state space model. The activation features are input as observation values ​​into the state observation equation, and the frequency drift of the duplexer is modeled and analyzed from the perspective of the dynamic system to obtain the state observation results. The introduction of the state space model makes the impact of temperature changes not only limited to the current moment, but also reflects its impact on future states, thereby providing a comprehensive description of the frequency drift. The state quantity of the state observation results is solved to obtain the state quantity of the frequency drift of the duplexer. The solution process estimates the actual value of the duplexer frequency drift based on the output of the observation equation. This state quantity directly reflects the frequency offset of the duplexer under the current temperature conditions.

[0022] Step S400, performing a sliding window adaptive Kalman filter operation on the duplexer frequency drift state quantity to obtain a compensation parameter; Specifically, the frequency drift state quantity of the duplexer is divided into sliding windows, and the entire frequency drift data sequence is divided into multiple windows. A continuous frequency drift state sequence is extracted in each window to obtain a segmented state sequence. The design of the sliding window helps to process dynamically changing data in real time, ensure the continuity and correlation of the filtering process in a local period, and enable the filtering model to more accurately track the changing trend of the frequency drift under complex temperature change conditions. The segmented state sequence is input into the Gaussian mixture model, and the probability density function of the noise is fitted by the expectation maximization (EM) algorithm to obtain the distribution model of the measurement noise. The Gaussian mixture model is a commonly used tool for describing the statistical characteristics of data, and the EM algorithm can continuously approach the optimal solution in the process of modeling the multi-peak characteristics of the measurement noise through an iterative method, so as to better fit the actual distribution of the noise. Through this step, a Gaussian mixture model reflecting the statistical characteristics of the measurement noise is obtained, which describes the probability characteristics of various noises existing in the frequency drift process. In order to optimize the noise model, the kernel density estimation is performed on the measurement noise distribution model to obtain the noise density function. Kernel density estimation is a parameter-free probability density estimation method that can smooth and refine the distribution of noise through appropriate kernel functions to obtain a more accurate description of noise density, effectively supplement the estimation of noise distribution by Gaussian mixture model, and make the noise model more accurate. Based on the noise density function, the target noise model is obtained by calculating the probability distribution distance between different Gaussian components and judging whether to merge the corresponding components according to the set threshold. A Kalman filter is constructed based on the target noise model, in which the state dimension is set to 4 and the observation dimension is set to 2. Parameter setting helps to ensure that the filter can fully capture the dynamic characteristics of frequency drift and use relevant information for state estimation. At the same time, the state covariance matrix and the observation covariance matrix are initialized to ensure that the filter model has appropriate distribution characteristics in the initial state. The filter model is predicted and updated. The state transfer matrix, observation matrix, process noise matrix and observation noise matrix are iteratively solved. Through the calculation of Kalman gain, the filter can dynamically correct the state so that the filter state quantity gradually approaches the real frequency drift state. The role of the Kalman gain is to adjust the direction and amplitude of the state estimation by balancing the deviation between the observed value and the predicted value, thereby reducing the estimation error and increasing the response speed and accuracy of the filter. The residual covariance is calculated based on the filter state quantity, and the weight coefficient of the state estimation is dynamically adjusted based on the size of the residual to obtain the weighted state quantity. The residual covariance reflects the deviation between the filter state estimate and the actual observation. By dynamically adjusting it, the accuracy of the state estimation is effectively improved. The weighted state quantity is transformed to obtain the frequency compensation value and the phase compensation value, and the final compensation parameter is obtained.

[0023] Step S500: input the compensation parameter into the feedforward compensation controller, adjust the voltage-controlled resonator of the duplexer through the sliding mode control structure, and obtain a frequency compensation signal.

[0024] Specifically, the compensation parameters are input into the feedforward compensation controller, and the compensation parameters are decomposed into frequency compensation coefficients and phase compensation coefficients, which are used to perform corresponding compensation operations on the frequency and phase respectively. The frequency compensation coefficients and phase compensation coefficients are mapped and transformed through the temperature-frequency response model of the duplexer to obtain the feedforward compensation component used for compensation control, reflecting the adjustment information required by the duplexer under different temperature conditions. The sliding surface is designed for the feedforward compensation component, and a sliding mode switching function is formed by constructing a linear combination of the frequency deviation state quantity and the frequency deviation change rate. Sliding mode control is a control method used to deal with uncertainty and external interference. By constructing a sliding surface, the state of the system tends to a stable state along the surface. The sliding mode switching function plays a core role in the entire sliding mode control structure. Its value is used to reflect the degree of deviation of the system and provide an input basis for the calculation of the control law. The sliding mode switching function is input into the control law calculation unit, and the control law expression is constructed based on the Lyapunov function to obtain the sliding mode control quantity. The Lyapunov function ensures that the system can remain stable during the control process, that is, the motion state on the sliding surface can eventually converge to the expected value. However, the high-frequency chattering phenomenon in the sliding mode control process will have an adverse effect on the stability of the system. In order to eliminate this chattering, the sliding mode control quantity is subjected to chattering suppression processing. By introducing the boundary layer function to replace the sign function, a smooth control signal is obtained, which makes the sliding mode control signal more continuous and soft, and effectively reduces the fluctuation caused by the chattering. Based on the smooth control signal and the feedforward compensation component, a composite control law is constructed. The composite control law combines feedforward compensation and sliding mode control together, so that the control system has the ability to respond quickly to the current state and can effectively compensate for disturbances and uncertainties. The combination of these two control components obtains the total control quantity, which combines the advantages of feedforward and feedback control and has a high response speed and robustness. When the total control quantity is converted into the driving voltage of the voltage-controlled resonator, the total control quantity is accurately mapped by the digital-to-analog conversion circuit, so that the final compensation voltage signal can fall within the target voltage range of the voltage-controlled resonator, ensuring accurate control of the resonant frequency. The compensation voltage signal is driven and amplified. In order to meet the working requirements of the voltage-controlled resonator, an operational amplifier is used to gain adjust and level convert the voltage signal. The compensation voltage is adjusted to a suitable working range for the voltage-controlled resonator, which not only ensures the stability of the voltage signal, but also provides sufficient driving force to obtain the final driving voltage. The driving voltage is loaded to the voltage-controlled resonator, and the capacitance effect of the voltage-controlled resonator is used to adjust the resonant frequency, so that the center frequency of the duplexer can be dynamically compensated and adjusted. Generate a frequency compensation signal for duplexer frequency adjustment. The compensation signal can respond to the influence of temperature changes in real time. Through the close cooperation of feedforward compensation, sliding mode control, digital-to-analog conversion, and amplification drive, it ensures that the voltage-controlled resonator can still maintain a stable working state under temperature fluctuations and effectively compensate for frequency drift.

[0025] The deviation between the frequency compensation signal generated by the sliding mode control structure and the target frequency response is calculated to obtain the compensation error. The compensation error is the frequency offset caused by the failure of the current compensation signal to fully match the target frequency response. The compensation error is combined with the temperature feature data to form a training sample pair, thereby constructing the network update data for optimizing the neural network. The network update data is input into the encoder of the mechanism information neural network for feature extraction. The encoder contains 4 fully connected layers, which are used to perform multi-level feature encoding on the input data, thereby gradually extracting feature information that has a greater impact on the compensation effect. In each layer, the data is linearly transformed and nonlinearly processed by batch normalization and ReLU activation function in turn to increase the nonlinear ability and stability of the feature representation. After multiple layers of encoding processing, the updated features are obtained. The encoded features are sparsely constrained. By performing sparse operations on the features, important feature information is retained, redundancy is reduced, and the network can effectively learn the key compensation mode. The updated features are reconstructed using the temperature mechanism dictionary. The temperature mechanism dictionary contains M atoms, each of which represents a temperature compensation mode. In order to enable the network to better adapt to the latest temperature features, these atoms are updated based on the physical compensation mechanism. The temperature mechanism dictionary needs to adapt to changes in the environment so that each atom in the dictionary better matches the current physical compensation requirements. In order to keep the dictionary standardized and independent, the Schmidt orthogonalization method is used to normalize the dictionary to obtain an updated dictionary, ensuring that the atoms in the dictionary remain orthogonal in the feature space, reducing mutual interference between features, and increasing the expression ability and robustness of the dictionary. The updated features are fused with the updated dictionary, and the temperature compensation mechanism information is embedded into the feature space by means of matrix inner product. The mechanism information preservation loss term is constructed to obtain the second fused feature. The second fused feature is input into the decoder of the mechanism information neural network. The decoder also contains 4 fully connected layers. Through multi-layer mapping operations, the fused features are remapped back to the original space. In this process, batch normalization and ReLU activation functions are used for nonlinear transformation to ensure the nonlinear expression ability of the features. After reconstruction by the decoder, the reconstructed features are obtained. These features are used to evaluate the learning effect of the network and to construct the overall loss function. The overall loss function consists of the reconstruction error term, the sparse constraint term, and the mechanism information preservation term. The optimization target is constructed by weighted combination of these three parts. The reconstruction error term is used to measure the difference between input and output features, the sparse constraint term ensures that the features learned by the network are sufficiently sparse, and the mechanism information retention term ensures that the network retains the physical compensation mechanism. The weights of the mechanism information neural network are updated by back-propagating the optimization target, and the Adam optimizer is used to update the weights of the mechanism information neural network. The adaptive learning rate feature of the Adam optimizer enables it to show good convergence speed and stability when processing complex data.The updated parameters are imported into the mechanism information neural network and trained iteratively for multiple rounds. In each round of training, the loss is recalculated and the network weights are updated until the training error is less than the preset target value, and the iteration is stopped to finally obtain the target neural network model.

[0026] In an embodiment of the present invention, the physical mechanism information of temperature compensation is introduced into the feature extraction process by combining a mechanism information neural network with a temperature mechanism dictionary, and a stable state observer is established by using a radial basis function neural network and Lyapunov stability theory, which can accurately capture the frequency drift characteristics of the duplexer and improve the accuracy of state quantity observation; an adaptive Kalman filtering algorithm is introduced, and the measurement noise is modeled in combination with a Gaussian mixture model, which significantly enhances the compensation robustness of the system in a complex noise environment; a feedforward compensation controller based on sliding mode control is designed, and control jitter is suppressed by introducing a boundary layer function, thereby achieving smooth compensation of the duplexer frequency; a closed-loop parameter optimization mechanism is constructed, and the neural network is updated online through error feedback, so that the system has continuous learning and adaptive capabilities, thereby ensuring the compensation effect of long-term operation.

[0027] In a specific embodiment, the process of executing step S100 may specifically include the following steps: The working temperature of the duplexer is collected in real time to obtain temperature sampling data, and the temperature sampling data is subjected to outlier removal and normalization processing to obtain standardized temperature data; The standardized temperature data is input into the Gaussian mixture model, and the temperature change is fitted by the probability density through the linear combination of multiple Gaussian components to obtain the basic temperature distribution data; Perform kernel density estimation calculation on the basic temperature distribution data, use the kernel function to perform parameter-free estimation on the probability density of temperature change, and obtain the temperature density distribution data; The temperature density distribution data is input into the fuzzy C-means clustering unit of the Wasserstein distance criterion, and the temperature data clustering matrix is ​​obtained by calculating the probability distribution distance between the temperature data points; The temperature data clustering matrix is ​​optimized and the temperature change feature categories are obtained. The temperature change feature category is converted into a feature vector form, the feature vector is subjected to dimension reduction and orthogonalization processing to obtain orthogonalized feature data, and the temperature change pattern of the orthogonalized feature data is analyzed to obtain temperature feature data.

[0028] Specifically, the operating temperature of the duplexer is collected in real time to obtain temperature sampling data. By installing a temperature sensor, the temperature of the duplexer operating environment is continuously collected to obtain a set of continuous temperature data, which is recorded as ,in Indicated in The temperature value collected at a time point, is the total amount of data collected. The collected temperature data is subjected to outlier removal, and the data points that are far from the overall trend of the data set are removed using the 3-times standard deviation method or the box-and-whisker plot-based method. For example, if the mean of the data is and the standard deviation is , then those that satisfy The data points are considered as outliers and removed. After removing the outliers, the temperature data is normalized to eliminate the dimensional differences between different temperatures. Scaling the temperature data to a uniform range, usually [0,1] or [-1,1], improves the stability and convergence speed of subsequent model training. Normalization methods include minimum-maximum normalization and standardization. and are the minimum and maximum values ​​of the temperature data, respectively. For each data point , normalized using the following formula: ; The normalized temperature data set is recorded as ,in Represents the total amount of data after removing outliers. The standardized temperature data is input into the Gaussian mixture model, and the temperature change is fitted by the linear combination of multiple Gaussian components. The Gaussian mixture model is a probability density model that approximates complex distributions using the linear combination of multiple Gaussian distributions. Assume that the Gaussian mixture model contains Gaussian components, the probability density function of the Gaussian mixture model is expressed as: ; in, It is The weights of the Gaussian components satisfy and They are The mean and covariance matrix of the Gaussian components, Represents a mean value of , the covariance is The model parameters are maximised by the expectation maximisation (EM) algorithm. Perform iterative solution to obtain the probability density function describing the temperature change, that is, the basic temperature distribution data. Perform kernel density estimation calculation on the basic temperature distribution data. Kernel density estimation is a parameter-free probability density estimation method that can smooth the distribution of data through the kernel function to make the estimation result smoother and more continuous. The formula for kernel density estimation is: ; in, is the sample size, is the bandwidth parameter, is a kernel function, and commonly used kernel functions include Gaussian kernel functions, etc. Kernel density estimation can smooth the basic temperature distribution data so that the probability density of temperature changes can better reflect the actual situation and obtain temperature density distribution data. The temperature density distribution data is input into the fuzzy C-means clustering unit based on the Wasserstein distance criterion for clustering analysis. Wasserstein distance is an effective metric for measuring the difference between two probability distributions and is suitable for changes in the probability distribution of data. Through fuzzy C-means clustering, temperature data points are assigned to multiple clusters, each cluster representing a temperature change pattern. The goal of the clustering process is to minimize the weighted sum of the distances within the cluster. The clustering result is expressed as a temperature data clustering matrix, where the elements of the matrix represent the degree of membership of each data point to a cluster. The temperature data clustering matrix is ​​optimized by calculating the number of components to determine the optimal temperature change feature category. The purpose of component number optimization is to find the optimal number of clusters. , so as to accurately describe the change pattern of temperature data. Use the silhouette coefficient or information criterion (such as AIC, BIC) to evaluate the clustering effect under different numbers of clusters, and select the number of clusters with the best evaluation value as the final feature category. Convert the temperature change feature category into a feature vector form to form a vectorized representation of the temperature change. Perform dimensionality reduction and orthogonalization on the feature vector to reduce the dimension of the data and remove the correlation between features. Common methods of dimensionality reduction include principal component analysis, which projects high-dimensional features into a low-dimensional space through linear transformation while retaining as much original data information as possible. Assume that the eigenvector is , principal component analysis obtains the dimension reduction matrix through eigendecomposition , the eigenvector after dimensionality reduction is: ; Orthogonalization ensures that the features after dimensionality reduction are independent of each other, and orthogonalized feature data is obtained. The temperature change pattern of the orthogonalized feature data is analyzed to extract the main feature information reflecting the duplexer at different temperatures. By analyzing these features, the temperature feature data obtained is used for frequency compensation control of the duplexer.

[0029] In a specific embodiment, the process of executing step S200 may specifically include the following steps: The mechanism information neural network includes two sub-networks, the encoder and the decoder, both of which adopt the autoencoder structure; The temperature feature data is processed by an encoder. The encoder contains 4 fully connected layers. The input dimension of the first fully connected layer in the encoder is the same as the dimension of the temperature feature data. The dimensions of the second to fourth fully connected layers in the encoder are reduced to 1 / 2 of the original dimensions respectively. The fully connected layer in each encoder is connected to a batch normalization layer and a ReLU activation function to obtain intermediate features. Perform sparse constraint processing on the intermediate features, construct sparse constraint terms through L1 norm regularization, set the sparse coefficient λ, perform sparse optimization on the intermediate features, and obtain sparse features; The physical mechanism information of temperature compensation is constructed as a temperature mechanism dictionary, which contains M atoms, each atom represents a temperature compensation mode, and the temperature mechanism dictionary is orthogonalized to obtain a mechanism dictionary matrix; Perform feature fusion on the sparse features and the mechanism dictionary matrix, embed the physical mechanism information into the feature space through matrix inner product operation, construct the mechanism information preservation item, and obtain the first fusion feature; The first fusion feature is processed by a decoder. The decoder contains 4 fully connected layers. The dimensions of the first to third fully connected layers in the decoder are increased to twice the original dimensions respectively. The output dimension of the fourth fully connected layer in the decoder is the same as the dimension of the temperature feature data. The fully connected layer in each decoder is connected to a batch normalization layer and a ReLU activation function to obtain the temperature compensation feature.

[0030] Specifically, an autoencoder structure is constructed that includes two sub-networks, an encoder and a decoder. The design of the network structure helps to achieve effective compression and reconstruction of the temperature feature data to extract feature information that is meaningful for the temperature compensation of the duplexer. Both the encoder and the decoder use an autoencoder structure. The autoencoder compresses the input data into a low-dimensional feature representation, and then restores it to the original space through the decoder, thereby learning the intrinsic structure and important features of the data. The temperature feature data is processed by an encoder, and the encoder contains four fully connected layers. The input dimension of the first fully connected layer in the encoder is the same as the dimension of the temperature feature data. Assume that the dimension of the temperature feature data is , the input dimension of the first fully connected layer is , and use linear mapping to map the input data to a new feature space to obtain the output features. For temperature feature data, the output of the first layer is: ; in, is the weight matrix of the first layer, is the bias vector, Represents the features after the first layer. The output is passed through the batch normalization layer and the ReLU activation function. The batch normalization layer is used to standardize each small batch of data, reduce the impact of internal covariate shift, thereby speeding up training and improving stability. The formula of the ReLU activation function is: ; The introduction of nonlinear characteristics enables the neural network to fit complex nonlinear relationships. The features processed by the first layer are input into the second to fourth fully connected layers, and the dimensions of the second to fourth fully connected layers are reduced to 1 / 2 of the original dimensions. The output dimension of the second layer is , the output dimension of the third layer is , the output dimension of the fourth layer is The output of each layer is expressed as: ; in, Indicates The weight matrix of the layer, is a bias term. The output after each layer is processed by a batch normalization layer and a ReLU activation function. By gradually reducing the dimension, the encoder extracts the low-dimensional feature representation of the data and obtains the intermediate features. The intermediate features are sparsely constrained to ensure that the extracted features have strong representation capabilities while eliminating redundant information. The sparse constraint is achieved by introducing L1 norm regularization, and the formula is: ; in, is the sparse coefficient, which is used to control the strength of the sparseness. Indicates The value of the intermediate feature. Through the sparse processing, unnecessary features are reduced, and the features that play an important role in temperature compensation are retained to obtain sparse features. The physical mechanism information of temperature compensation is constructed as a temperature mechanism dictionary, which contains atoms, each atom represents a temperature compensation mode. The temperature mechanism dictionary is understood as a set of basic elements used to describe the relationship between temperature and compensation. These elements cover the compensation requirements of the duplexer under different temperature changes. In order to make these atoms independent of each other in the feature space, the temperature mechanism dictionary is orthogonalized. The Schmidt orthogonalization method is used to process the dictionary so that each atom in the dictionary maintains orthogonality in the feature space. Assume that the temperature mechanism dictionary is , through the Schmidt orthogonalization method, the orthogonalized mechanism dictionary matrix is ​​obtained ,satisfy: ; Orthogonalization helps to improve the independence of the temperature mechanism dictionary and enhance the effect of feature fusion. On this basis, the sparse features and mechanism dictionary matrix are fused. The physical mechanism information is embedded into the feature space through matrix inner product operation to construct the mechanism information preservation item. Assume that the sparse feature is represented as S and the mechanism dictionary matrix is , then the feature fusion process is expressed as: ; in, Represents the first fused feature. The feature fusion process ensures that the physical mechanism information is preserved in the feature space, so that the neural network not only relies on the statistical characteristics of the data, but also retains the physical interpretation of the compensation process. The first fused feature is input to the decoder for processing. The decoder also consists of four fully connected layers. In the decoder, the dimensions of the first to third fully connected layers are increased to twice the original dimensions, and the output dimension of the fourth fully connected layer is the same as the dimension of the temperature feature data. The first layer expands the input feature dimension to , the second layer extends to , the third layer extends to , the output dimension of the fourth layer is The output of each layer is expressed as: ; After each layer output, a batch normalization layer and a ReLU activation function are connected to perform nonlinear transformation. The layer-by-layer expansion operation restores the encoded low-dimensional features back to the original space, and the output is the temperature compensation feature. .

[0031] In a specific embodiment, the process of executing step S300 may specifically include the following steps: A radial basis function neural network structure is constructed, the number of input layer nodes of the radial basis function neural network is set to the dimension of the temperature compensation feature, the number of hidden layer nodes is set to N, and the number of output layer nodes is set to the dimension of the duplexer frequency drift state quantity; Input the temperature compensation feature into the input layer of the radial basis function neural network, map the temperature compensation feature to the hidden layer space through linear mapping, and transform the linear mapping using the first weight matrix to obtain the initial hidden layer feature; Perform Gaussian radial basis function calculation on the initial hidden layer features to obtain radial basis function features; According to the Lyapunov stability criterion, an adaptive law is constructed to update the first weight matrix and the second weight matrix of the radial basis function neural network online, where the second weight matrix is ​​the weight matrix from the hidden layer to the output layer, to obtain the updated network weights; Perform a linear combination of the second weight matrix on the radial basis function features, map the hidden layer features to the output space through a weighted sum operation, and obtain the output layer features; Perform nonlinear activation function processing on the output layer features to obtain activation features; The state observation equation is constructed based on the activation feature, and the dynamic characteristics of the duplexer frequency drift are described by the state space model. The activation feature is input into the state observation equation as the observation value to obtain the state observation result. The state quantity of the state observation result is solved to obtain the state quantity of the duplexer frequency drift.

[0032] Specifically, a radial basis function neural network structure is constructed. The radial basis function neural network is a three-layer neural network structure consisting of an input layer, a hidden layer, and an output layer. The number of input layer nodes is the same as the dimension of the temperature compensation feature, which is set to , represents the dimension of temperature compensation feature. The number of hidden layer nodes is set to , indicating that the network hidden layer contains neurons, used to perform nonlinear mapping on the input features. The number of output layer nodes is set to the dimension of the duplexer frequency drift state quantity, denoted as , which is used to represent the final estimated frequency drift state. The temperature compensation feature is input into the input layer of the radial basis function neural network, and linear mapping is performed to map the input data to the hidden layer space. Assume that the temperature compensation feature vector is , the initial features of the hidden layer are obtained through a linear transformation, and the linear mapping formula used is: ; in, is the weight matrix from the input layer to the hidden layer, is the bias vector, Represents the initial features of the hidden layer. Through this process, the input feature vector is mapped to a new feature space. The Gaussian radial basis function is calculated for the initial hidden layer features to obtain the radial basis function features. The Gaussian radial basis function is an activation function used to perform nonlinear mapping on the initial features. The calculation formula of the radial basis function is: ; in, Indicates The output of hidden layer nodes, It is The center vector of hidden layer nodes, is the bandwidth parameter of the Gaussian kernel function, which is used to control the expansion range of the Gaussian function. Through nonlinear mapping, the initial hidden layer features are converted into radial basis function features, recorded as This nonlinear transformation enables the network to handle complex feature relationships. In order to make the neural network adaptive, an adaptive law is constructed based on the Lyapunov stability criterion to update the network weights online. For the radial basis function neural network, the first weight matrix from the input layer to the hidden layer needs to be updated. , and the second weight matrix from hidden layer to output layer The second weight matrix It is used to map the nonlinear features of the hidden layer to the output space to obtain the final frequency drift state. The Lyapunov criterion ensures that the entire neural network remains stable during training and operation. By constructing the Lyapunov function To evaluate the stability of the system, the general form of the Lyapunov function is: ; in, Represents the estimation error of the network, and the goal is to make Gradually decreases with time until it approaches zero. Based on the Lyapunov function, an adaptive update rule is constructed for the first and second weight matrices to ensure the stability of the system and continuously improve the estimation performance of the network. After obtaining the updated weight matrix, the radial basis function characteristics are With the second weight matrix Perform linear combination to obtain the output layer features. The calculation formula is: ; in, is the bias vector of the output layer, Represents the output layer feature. This feature is obtained by weighted summation of hidden layer features and represents the network's preliminary estimate of the input feature. In order to enhance the nonlinear fitting ability of the network, the output layer feature is processed by a nonlinear activation function to obtain an activated feature. Activation functions include ReLU, Sigmoid, or Tanh. Assuming the Tanh activation function is used, its formula is: ; The nonlinear activation function can effectively compress the range of feature values ​​and introduce nonlinear characteristics to improve the generalization ability of the network. After being processed by the activation function, the activation feature obtained is recorded as Based on the activation characteristics obtained, the state observation equation is constructed. The state observation equation is used to describe the dynamic characteristics of the duplexer frequency drift and can model the frequency drift from the system perspective. Assume that the state space model is: ; ; in, represents the state variables of the system, is the input, is the output, are the state transfer matrix, input matrix, output matrix and direct transfer matrix of the system respectively. By inputting the state observation equation as the observation value, the frequency drift of the duplexer can be fully described from the perspective of the state space, and the state observation result can be obtained. The state quantity of the state observation result is solved to obtain the frequency drift state quantity of the duplexer. The solution process is to estimate the real state of the frequency drift based on the observation equation, which minimizes the observation error and makes the estimated frequency drift state quantity This is as close as possible to the real situation, thus providing accurate compensation for the duplexer frequency.

[0033] In a specific embodiment, the process of executing step S400 may specifically include the following steps: The frequency drift state quantity of the duplexer is divided into sliding windows, and a continuous frequency drift state sequence is extracted in each window to obtain a segmented state sequence; The segmented state sequence is input into the Gaussian mixture model, and the probability density function of the noise is fitted by the EM algorithm to obtain the measurement noise distribution model; Perform kernel density estimation on the measurement noise distribution model to obtain the noise density function; Based on the noise density function, the probability distribution distance between different Gaussian components is calculated, and the corresponding components are merged through threshold judgment to obtain the target noise model; Construct a Kalman filter based on the target noise model, set the state dimension to 4, the observation dimension to 2, initialize the state covariance matrix and the observation covariance matrix, and obtain the filter model; The filter model is predicted and updated, the state transfer matrix, observation matrix, process noise matrix and observation noise matrix are iteratively solved, the Kalman gain is used for state correction, and the filter state quantity is obtained; The residual covariance is calculated based on the filtering state quantity, the weight coefficient of the state estimation is dynamically adjusted to obtain the weighted state quantity, and the weighted state quantity is transformed into a frequency compensation value and a phase compensation value to obtain the compensation parameter.

[0034] Specifically, the frequency drift state quantity of the duplexer is divided into sliding windows to capture the local characteristics of the changes in the time series. By setting a time window of fixed length and sliding it with a certain step size, a continuous frequency drift state sequence is extracted in each window. Assume that the frequency drift state quantity is the sequence , the length of the sliding window is , the step length is , then the state sequence is divided into multiple sliding windows, for example, the first window is , the second window is , and so on, multiple segmented state sequences are obtained, representing the frequency drift of the duplexer in different time periods. The segmented state sequence is input into the Gaussian mixture model, and the probability density function of the noise is fitted by the expectation maximization (EM) algorithm to obtain the distribution model of the measurement noise. The Gaussian mixture model describes the distribution characteristics of the data through the linear combination of multiple Gaussian distributions, which is suitable for capturing the complex distribution form of noise. Assume that the Gaussian mixture model contains Gaussian components, then its probability density function is expressed as: ; in, Indicates The weights of the Gaussian components satisfy , and Respectively The mean and covariance matrix of the Gaussian components, It means the mean , the covariance is Gaussian distribution. The parameters of the model are solved iteratively through the EM algorithm, so that the model fits the noise data optimally and obtains the distribution model of the measurement noise. Kernel density estimation is performed on the measurement noise distribution model. Kernel density estimation is a parameter-free probability density estimation method that can smooth the distribution of noise and obtain a more detailed description of noise density. The formula for kernel density estimation is: ; in, is the sample size, is the bandwidth parameter, is the kernel function, and commonly used kernel functions include Gaussian kernel function, etc. Through kernel density estimation, the probability density of the measurement noise is smoothed to obtain the noise density function. Based on the noise density function, the probability distribution distance between different Gaussian components is calculated, and the corresponding components are determined by setting a threshold to obtain the target noise model. The probability distribution distance between Gaussian components is measured by Wasserstein distance or Kullback-Leibler (KL) divergence. Assume that the probability density functions of the two Gaussian distributions are and , then the KL divergence is expressed as: ; By calculating the KL divergence and comparing it with the set threshold, we can determine whether the two components are similar. If they are similar, we merge them to obtain a more simplified target noise model. Based on the target noise model, we build a Kalman filter. The Kalman filter is a recursive algorithm for state estimation of dynamic systems, which is used to filter out noise in frequency drift. The state dimension of the filter is set to 4 and the observation dimension is set to 2, which means that the state variables of the system are 4-dimensional and the observation variables are 2-dimensional. When initializing the filter, we set the state covariance matrix and the observation covariance matrix The state covariance matrix is ​​used to describe the uncertainty of the state variables. The initial state covariance matrix is ​​set to the identity matrix multiplied by a scaling factor, such as ,in represents the initial uncertainty of the state, is a 4-dimensional identity matrix. The observation covariance matrix The variance of the measurement noise is described by the noise model. During the filtering process, the Kalman filter performs prediction and update calculations. In the prediction step, according to the state transfer matrix and the process noise covariance matrix , predict the next state of the system. The prediction formula is: ; ; in, Indicates The predicted value of the state at a moment, represents the predicted state covariance matrix, is the control input matrix, is the control input. In the update step, based on the observation matrix and the observation noise covariance matrix , correct the predicted value. Calculate the Kalman gain ; Then use the observed values Correct the status: ; ; in, is the Kalman gain, is the corrected state estimate, is the corrected state covariance matrix. Based on the state quantity obtained by filtering, the residual covariance is calculated to dynamically adjust the weight coefficient of the state estimation to obtain the weighted state quantity. The residual covariance is used to measure the difference between the observed value and the predicted value, and its formula is: ; Through the residual covariance Dynamically adjust the weight coefficient of the state estimation to make the filter respond more flexibly and accurately to the new observation value. The obtained weighted state quantity is used to compensate for the frequency drift and phase drift. The weighted state quantity is converted into the frequency compensation value and the phase compensation value to obtain the compensation parameter. The frequency compensation value and the phase compensation value are obtained through a mapping function. Assuming that the mapping function is a linear transformation, the frequency compensation value and the phase compensation value are expressed as ; ; in, and are mapping weight matrices for frequency compensation and phase compensation, respectively. In this way, compensation parameters for compensating the duplexer frequency and phase are obtained.

[0035] In a specific embodiment, the process of executing step S500 may specifically include the following steps: Input the compensation parameter into the feedforward compensation controller, decompose the compensation parameter into a frequency compensation coefficient and a phase compensation coefficient, map and transform the frequency compensation coefficient and the phase compensation coefficient through a duplexer temperature-frequency response model, and obtain a feedforward compensation component; The sliding surface is designed for the feedforward compensation component, and the sliding mode switching function is obtained by constructing a linear combination of the frequency deviation state quantity and the frequency deviation change rate. The sliding mode switching function is input into the control law calculation unit, and the control law expression is constructed based on the Lyapunov function to obtain the sliding mode control quantity, and the sliding mode control quantity is subjected to chattering suppression processing. By introducing the boundary layer function to replace the sign function, a smooth control signal is obtained; A composite control law is constructed based on a smooth control signal and a feedforward compensation component, the two control components are combined to obtain a total control quantity, and the total control quantity is converted into a driving voltage of a voltage-controlled resonator, and the total control quantity is mapped to a target voltage range through a digital-to-analog conversion circuit to obtain a compensation voltage signal; The compensation voltage signal is driven and amplified, and an operational amplifier is used to perform gain adjustment and level conversion on the voltage signal. The voltage value is adjusted to the working range of the voltage-controlled resonator to obtain a driving voltage, and the driving voltage is loaded to the voltage-controlled resonator. The resonant frequency is adjusted through the capacitor effect, and the center frequency of the duplexer is dynamically compensated to obtain a frequency compensation signal.

[0036] Specifically, the compensation parameter is input into the feedforward compensation controller and decomposed into a frequency compensation coefficient and a phase compensation coefficient. Assume that the compensation parameter is a vector ,in is the frequency compensation coefficient, is the phase compensation coefficient. These two parameters are input into the temperature-frequency response model of the duplexer. The model maps and transforms these two coefficients according to different temperature conditions to obtain the feedforward compensation component. The frequency compensation and phase compensation are converted through the temperature-frequency response model. This conversion is regarded as a mapping function ,Right now: ; in, represents the feedforward compensation component, which contains the feedforward adjustment information of frequency and phase to adapt to the influence of temperature on the duplexer frequency. The sliding surface is designed for the feedforward compensation component to construct the sliding mode control part. Sliding mode control is a control method that is robust to system parameter uncertainty and external disturbances. The key lies in designing the sliding surface and making the system state tend to be stable along the sliding surface. By constructing the frequency deviation state quantity and frequency deviation change rate , design a sliding surface , the design form of the sliding surface is: ; in, Indicates the current frequency deviation, represents the rate of change of frequency deviation, is a positive parameter in the sliding surface design, which is used to adjust the response speed of the sliding surface. Through this sliding surface, the sliding mode switching function is obtained , reflecting the changing trend of frequency deviation over time. The sliding mode switching function is input into the control law calculation unit, and the control law expression is constructed based on the Lyapunov function to ensure that the system tends to be stable on the sliding mode surface. It is used to describe the energy of the system state and make it It decreases monotonically over time, thus ensuring the stability of the system. One form of the Lyapunov function is: ; In order to make the Lyapunov function decrease, the time derivative is taken and the sliding mode control quantity is designed So that its time derivative is negative, that is: ; in, is the control gain, is a sign function, which is used to indicate the direction of sliding mode control. However, the sign function may cause chattering of the control signal because the sign function will produce discontinuous changes when switching. In order to suppress the chattering phenomenon, the boundary layer function is used. Alternative sign function, where is the boundary layer thickness, represents the saturation function, which is defined as: ; By introducing the boundary layer function, the sliding mode control quantity is smoothed to obtain a smooth control signal Based on the smooth control signal and the feedforward compensation component, a composite control law is constructed. The composite control law combines feedforward control and sliding mode control, making use of both the rapid response capability of feedforward control to the system and the robustness of sliding mode control to uncertainty. The composite control law is expressed as: ; in, represents the feedforward compensation component, Represents the smooth signal of sliding mode control. By combining these two control components, the total control quantity is obtained , which is used to accurately adjust the duplexer frequency. When the total control amount is converted into the driving voltage of the voltage-controlled resonator, the total control amount is mapped to the target voltage range through the digital-to-analog conversion circuit (DAC) so that the voltage-controlled resonator can receive and adjust it. Assume that the output voltage range of the DAC is , and the control quantity The range is , then it is converted into a compensation voltage signal through linear mapping : ; This mapping can ensure that the control signal is within the voltage range after digital-to-analog conversion and has a good linear relationship. The compensation voltage signal is driven and amplified to ensure that it can drive the voltage-controlled resonator. An operational amplifier is used to adjust the gain and level of the voltage signal. Assume that the compensation voltage signal Need to be boosted The amplified driving voltage It is expressed as: ; The voltage value is adjusted to the operating range of the voltage-controlled resonator through the gain adjustment of the operational amplifier, ensuring that the amplitude of the drive signal is sufficient to effectively adjust the resonant frequency. The drive voltage is applied to the voltage-controlled resonator, and the resonant frequency is adjusted through the capacitance effect. The resonant frequency of the voltage-controlled resonator is affected by its capacitance value, which in turn changes with the change of the applied voltage. Assume that the relationship between the capacitance of the resonator and the drive voltage is , then the resonant frequency The expression is: ; in, is the inductance of the resonator, Indicates driving voltage By adjusting the driving voltage, the capacitance value is dynamically changed, thereby adjusting the resonant frequency of the resonator, so that the center frequency of the duplexer can achieve accurate dynamic compensation.

[0037] In a specific embodiment, the method for automatically adjusting the temperature compensation parameters of a duplexer further includes the following steps: Calculate the deviation between the frequency compensation signal and the target frequency response to obtain the compensation error, and combine the compensation error with the temperature characteristic data to construct a training sample pair to obtain the network update data; The network update data is input into the encoder of the mechanism information neural network, and feature encoding is performed through the four fully connected layers in the encoder. The encoded features are subjected to sparse constraint processing to obtain updated features. Reconstruct the temperature mechanism dictionary for the updated features, update the M atoms of the temperature mechanism dictionary based on the physical compensation mechanism, and normalize the dictionary using the Schmidt orthogonalization method to obtain the updated dictionary; The updated features are fused with the updated dictionary, the temperature compensation mechanism information is embedded in the feature space through the matrix inner product, and the mechanism information preservation loss term is constructed to obtain the second fused feature; The decoder reconstructs the second fusion feature, maps the feature back to the original space through the four fully connected layers in the decoder, and uses batch normalization and ReLU activation function for nonlinear transformation to obtain the reconstructed feature; Based on the reconstruction features, the overall loss function is constructed, and the reconstruction error term, sparse constraint term and mechanism information retention term are weightedly combined to obtain the optimization target. The optimization target is back-propagated and the weights of the mechanism information neural network are updated using the Adam optimizer to obtain the updated parameters. The updated parameters are imported into the mechanism information neural network. After multiple rounds of iterative training, the iteration is stopped when the training error is less than the target value to obtain the target neural network model.

[0038] Specifically, the deviation between the frequency compensation signal and the target frequency response is calculated to obtain the compensation error. Assume that the frequency compensation signal is , the target frequency response is , the compensation error is expressed as the difference between the two: ; in, Indicates at time The compensation error is an important indicator for evaluating the current compensation effect, through which the shortcomings of the system in the current environment can be found. The compensation error is combined with the current temperature characteristic data to form a new data pair. ,in Represents temperature feature data. This combination constructs new training sample pairs to form a training data set for neural network update, which is recorded as The network update data is input into the encoder part of the mechanism information neural network for feature extraction. The encoder contains four fully connected layers, which gradually map and compress the input data to extract the core features of the input data. Suppose the input is ,in is the dimension of the temperature feature data, and the weight matrix of the first fully connected layer is , the bias is , then the output of the first layer is: ; in, is the activation function, which is defined as , which is used to introduce nonlinear characteristics. The feature dimension of the first layer output is half of the original dimension. By analogy, the output dimensions of the second, third, and fourth fully connected layers are one-quarter, one-eighth, and one-sixteenth of the original dimension, respectively. After the step-by-step encoding process of four fully connected layers, a low-dimensional feature representation is obtained, which contains the key information of the input data. The encoded features are sparsely constrained to ensure that the features extracted by the model are sparse and improve the generalization ability of the model. The sparse constraint is implemented through L1 norm regularization. The expression of the sparse constraint term is: ; in, is the sparse coefficient, which is used to control the degree of sparseness. represents the first The temperature mechanism dictionary is reconstructed for the updated features. The temperature mechanism dictionary is a set of basic elements used to describe the temperature compensation mechanism, denoted as ,in is the number of atoms in the dictionary, each atom Represents a temperature compensation mode. In order to ensure that the temperature mechanism dictionary can be well matched with the current update features, the dictionary is updated. The atoms are adjusted, and the dictionary is normalized by the Schmidt orthogonalization method to ensure that the atoms in the dictionary are orthogonal to each other, and the updated dictionary matrix is ​​obtained. The goal of the Schmidt orthogonalization method is to project each atom in the original dictionary to a direction orthogonal to the previous atom, thereby ensuring the orthogonality of the dictionary matrix. The updated features are fused with the updated dictionary matrix, and the temperature compensation mechanism information is embedded into the feature space through the matrix inner product, and the mechanism information preservation loss term is constructed. Let the updated feature be , update the dictionary matrix to , the result of feature fusion is: ; in, is the second fusion feature. The second fusion feature is input into the decoder for reconstruction. The decoder also contains four fully connected layers, which remap the low-dimensional features back to the original space by layer-by-layer mapping. Let the first layer weight matrix of the decoder be , the bias is , then the output of the first layer is: ; Similarly, the subsequent three layers of the decoder gradually map the features back to the original dimensions. The output of each layer is processed by batch normalization and ReLU activation function. The purpose of batch normalization is to reduce the internal covariate shift so that the features of each small batch have the same distribution, thereby accelerating the convergence of the network. After four layers of decoding, the reconstructed features are , which is a reconstruction of the original input temperature feature. In order to optimize the mechanism information neural network, an overall loss function is constructed. The overall loss function contains three main parts: reconstruction error term, sparse constraint term and mechanism information preservation term. The reconstruction error term is used to measure the difference between the reconstructed feature and the input feature, and is expressed as mean square error: ; Sparse Constraints Used to control the sparsity of features. Mechanism information preservation term It is used to ensure the characteristics of the updated features after embedding the physical mechanism information. The overall loss function is the weighted sum of these three parts: ; in, is the weight coefficient of the loss term, which is used to adjust the contribution of each loss term to the overall loss. The overall loss function is back-propagated and calculated, with the goal of minimizing the loss function. The Adam optimizer is used to update the weights of the mechanism information neural network. The Adam optimizer is a gradient-based optimization algorithm that uses momentum and adaptive learning rate to update parameters, making the network converge faster and more stable. The weight update formula of Adam is: ; in, Indicates The network parameters at the iteration, is the learning rate, and They are the momentum estimate of the gradient and the momentum estimate of the square of the gradient, To prevent the denominator from being a decimal of zero, the updated parameters are imported into the mechanism information neural network, and multiple rounds of iterative training are continued. In each round of training, the network is gradually optimized by continuously calculating the loss and updating the weights. When the training error is less than the preset target value, the iteration is stopped, and finally the optimized target neural network model is obtained.

[0039] See also Figure 2 , Figure 2 A schematic block diagram of the structure of the automatic adjustment device 200 for the temperature compensation parameters of a duplexer provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the automatic adjustment device 200 for the temperature compensation parameters of a duplexer includes: The sampling module 210 is used to sample the working temperature of the duplexer to obtain temperature sampling data, and to extract distribution characteristics of the temperature sampling data to obtain temperature characteristic data; Reconstruction module 220, used to input temperature characteristic data into the mechanism information neural network for characteristic reconstruction to obtain temperature compensation characteristics; A mapping module 230 is used to perform radial basis function neural network mapping on the temperature compensation characteristics to obtain a duplexer frequency drift state quantity; An operation module 240 is used to perform a sliding window adaptive Kalman filter operation on the duplexer frequency drift state quantity to obtain a compensation parameter; The adjustment module 250 is used to input the compensation parameter into the feedforward compensation controller, and adjust the voltage-controlled resonator of the duplexer through the sliding mode control structure to obtain a frequency compensation signal.

[0040] Through the collaborative cooperation of the above-mentioned components, the physical mechanism information of temperature compensation is introduced into the feature extraction process through the combination of mechanism information neural network and temperature mechanism dictionary. A stable state observer is established by using radial basis function neural network and Lyapunov stability theory, which can accurately capture the frequency drift characteristics of the duplexer and improve the accuracy of state quantity observation; an adaptive Kalman filter algorithm is introduced, and the measurement noise is modeled in combination with a Gaussian mixture model, which significantly enhances the compensation robustness of the system in complex noise environments; a feedforward compensation controller based on sliding mode control is designed, and the control jitter is suppressed by introducing a boundary layer function, so as to achieve smooth compensation of the duplexer frequency; a closed-loop parameter optimization mechanism is constructed, and the neural network is updated online through error feedback, so that the system has continuous learning and adaptive capabilities, ensuring the compensation effect of long-term operation.

[0041] See also Figure 3 , Figure 3This is a schematic block diagram of the structure of a duplexer temperature compensation parameter automatic adjustment device 300 provided in an embodiment of the present application. The duplexer temperature compensation parameter automatic adjustment device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected via a device bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0042] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned methods for automatically adjusting the temperature compensation parameters of a duplexer.

[0043] The processor 301 is used to provide computing and control capabilities to support the operation of the entire duplexer temperature compensation parameter automatic adjustment device 300 .

[0044] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned methods for automatically adjusting the temperature compensation parameters of the duplexer.

[0045] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the automatic adjustment device 300 for the temperature compensation parameters of the duplexer involved in the scheme of the present application. The specific automatic adjustment device 300 for the temperature compensation parameters of the duplexer may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0046] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0047] It should be noted that, technical personnel in the relevant field can clearly understand that, for the convenience and simplicity of description, the specific working process of the duplexer temperature compensation parameter automatic adjustment device 300 described above can refer to the corresponding process of the aforementioned duplexer temperature compensation parameter automatic adjustment method, and will not be repeated here.

[0048] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the method for automatically adjusting the temperature compensation parameters of a duplexer as provided in the embodiment of the present application.

[0049] The computer-readable storage medium may be an internal storage unit of the automatic adjustment device 300 for temperature compensation parameters of a duplexer in the aforementioned embodiment, such as a hard disk or memory of the automatic adjustment device 300 for temperature compensation parameters of a duplexer. The computer-readable storage medium may also be an external storage device of the automatic adjustment device 300 for temperature compensation parameters of a duplexer, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped with the automatic adjustment device 300 for temperature compensation parameters of a duplexer.

[0050] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0051] If 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 the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0052] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for automatically adjusting the temperature compensation parameters of a duplexer, characterized in that: include: Sampling the working temperature of the duplexer to obtain temperature sampling data, and extracting distribution characteristics of the temperature sampling data to obtain temperature characteristic data; Inputting the temperature characteristic data into the mechanism information neural network for feature reconstruction to obtain temperature compensation characteristics; Performing radial basis function neural network mapping on the temperature compensation characteristics to obtain a frequency drift state quantity of the duplexer; Performing a sliding window adaptive Kalman filter operation on the frequency drift state quantity of the duplexer to obtain a compensation parameter; The compensation parameter is input into a feedforward compensation controller, and the voltage-controlled resonator of the duplexer is adjusted through a sliding mode control structure to obtain a frequency compensation signal.

2. The method for automatically adjusting the temperature compensation parameters of a duplexer according to claim 1, characterized in that: The operating temperature of the duplexer is sampled to obtain temperature sampling data, and distribution feature extraction is performed on the temperature sampling data to obtain temperature feature data, including: The operating temperature of the duplexer is collected in real time to obtain temperature sampling data, and the temperature sampling data is subjected to outlier removal and normalization processing to obtain standardized temperature data; The standardized temperature data is input into a Gaussian mixture model, and a probability density fitting is performed on the temperature change through a linear combination of multiple Gaussian components to obtain basic temperature distribution data; Performing kernel density estimation calculation on the basic temperature distribution data, using the kernel function to perform parameter-free estimation on the probability density of temperature change, and obtaining temperature density distribution data; The temperature density distribution data is input into the fuzzy C-means clustering unit of the Wasserstein distance criterion, and the temperature data clustering matrix is ​​obtained by calculating the probability distribution distance between the temperature data points; Optimizing the number of components of the temperature data clustering matrix to obtain temperature change feature categories; The temperature change feature category is converted into a feature vector form, the feature vector is subjected to dimension reduction and orthogonalization processing to obtain orthogonalized feature data, and the orthogonalized feature data is subjected to temperature change pattern analysis to obtain temperature feature data.

3. The method for automatically adjusting the temperature compensation parameters of a duplexer according to claim 2, characterized in that: The step of inputting the temperature characteristic data into the mechanism information neural network for feature reconstruction to obtain the temperature compensation feature comprises: The mechanism information neural network includes two sub-networks, an encoder and a decoder, and both the encoder and the decoder adopt an autoencoder structure; Performing encoder processing on the temperature feature data, the encoder comprises 4 fully connected layers, the input dimension of the first fully connected layer in the encoder is the same as the dimension of the temperature feature data, the dimensions of the second to fourth fully connected layers in the encoder are successively reduced to 1 / 2 of the original dimensions, and each fully connected layer in the encoder is connected to a batch normalization layer and a ReLU activation function to obtain intermediate features; Performing sparse constraint processing on the intermediate features, constructing sparse constraint terms through L1 norm regularization, setting sparse coefficients λ, performing sparse optimization on the intermediate features, and obtaining sparse features; The physical mechanism information of temperature compensation is constructed into a temperature mechanism dictionary, wherein the temperature mechanism dictionary contains M atoms, each atom represents a temperature compensation mode, and the temperature mechanism dictionary is orthogonalized to obtain a mechanism dictionary matrix; Performing feature fusion on the sparse features and the mechanism dictionary matrix, embedding the physical mechanism information into the feature space through matrix inner product operation, constructing a mechanism information retention item, and obtaining a first fused feature; The first fusion feature is processed by a decoder, and the decoder includes 4 fully connected layers. The dimensions of the first to third fully connected layers in the decoder are increased to twice the original dimensions respectively. The output dimension of the fourth fully connected layer in the decoder is the same as the dimension of the temperature feature data. Each fully connected layer in the decoder is connected to a batch normalization layer and a ReLU activation function to obtain a temperature compensation feature.

4. The method for automatically adjusting the temperature compensation parameters of a duplexer according to claim 3, characterized in that: The step of performing radial basis function neural network mapping on the temperature compensation feature to obtain a duplexer frequency drift state quantity includes: Constructing a radial basis function neural network structure, setting the number of input layer nodes of the radial basis function neural network to the dimension of the temperature compensation feature, setting the number of hidden layer nodes to N, and setting the number of output layer nodes to the dimension of the duplexer frequency drift state quantity; Inputting the temperature compensation feature into the input layer of the radial basis function neural network, mapping the temperature compensation feature to the hidden layer space through linear mapping, wherein the linear mapping is transformed using a first weight matrix to obtain an initial hidden layer feature; Performing Gaussian radial basis function calculation on the initial hidden layer features to obtain radial basis function features; An adaptive law is constructed according to the Lyapunov stability criterion, and a first weight matrix and a second weight matrix of the radial basis function neural network are updated online, where the second weight matrix is ​​a weight matrix from a hidden layer to an output layer, to obtain an updated network weight; Performing a linear combination of a second weight matrix on the radial basis function features, mapping the hidden layer features to the output space through a weighted sum operation, and obtaining output layer features; Performing nonlinear activation function processing on the output layer features to obtain activation features; A state observation equation is constructed based on the activation feature, the dynamic characteristics of the duplexer frequency drift are described by a state space model, and the activation feature is input as an observation value into the state observation equation to obtain a state observation result; The state observation result is subjected to state quantity calculation to obtain the duplexer frequency drift state quantity.

5. The method for automatically adjusting the temperature compensation parameters of a duplexer according to claim 4, characterized in that: The step of performing a sliding window adaptive Kalman filter operation on the duplexer frequency drift state quantity to obtain a compensation parameter includes: Dividing the frequency drift state quantity of the duplexer into sliding windows, and extracting a continuous frequency drift state sequence in each window to obtain a segmented state sequence; The segmented state sequence is input into a Gaussian mixture model, and the probability density function of the noise is fitted by an EM algorithm to obtain a measurement noise distribution model; Performing kernel density estimation on the measurement noise distribution model to obtain a noise density function; Based on the noise density function, the probability distribution distance between different Gaussian components is calculated, and the corresponding components are merged through threshold judgment to obtain a target noise model; Construct a Kalman filter based on the target noise model, set the state dimension to 4, set the observation dimension to 2, initialize the state covariance matrix and the observation covariance matrix, and obtain a filter model; Performing prediction and update calculation on the filter model, iteratively solving the state transfer matrix, observation matrix, process noise matrix and observation noise matrix, using Kalman gain to perform state correction, and obtaining the filter state quantity; The residual covariance is calculated based on the filtering state quantity, the weight coefficient of the state estimation is dynamically adjusted to obtain the weighted state quantity, and the weighted state quantity is transformed into a frequency compensation value and a phase compensation value to obtain a compensation parameter.

6. The method for automatically adjusting the temperature compensation parameters of a duplexer according to claim 5, characterized in that: The step of inputting the compensation parameter into a feedforward compensation controller and adjusting the voltage-controlled resonator of the duplexer through a sliding mode control structure to obtain a frequency compensation signal includes: Inputting the compensation parameter into a feedforward compensation controller, decomposing the compensation parameter into a frequency compensation coefficient and a phase compensation coefficient, mapping and transforming the frequency compensation coefficient and the phase compensation coefficient through a duplexer temperature-frequency response model, and obtaining a feedforward compensation component; Designing a sliding mode surface for the feedforward compensation component, and obtaining a sliding mode switching function by constructing a linear combination of a frequency deviation state quantity and a frequency deviation change rate; The sliding mode switching function is input into a control law calculation unit, a control law expression is constructed based on a Lyapunov function, a sliding mode control quantity is obtained, and a chattering suppression process is performed on the sliding mode control quantity, and a smooth control signal is obtained by introducing a boundary layer function to replace a sign function; A composite control law is constructed based on the smooth control signal and the feedforward compensation component, the two control components are combined to obtain a total control amount, and the total control amount is converted into a driving voltage of a voltage-controlled resonator, and the total control amount is mapped to a target voltage range through a digital-to-analog conversion circuit to obtain a compensation voltage signal; The compensation voltage signal is driven and amplified, and an operational amplifier is used to perform gain adjustment and level conversion on the voltage signal, and the voltage value is adjusted to the working range of the voltage-controlled resonator to obtain a driving voltage, and the driving voltage is loaded to the voltage-controlled resonator, and the resonant frequency is adjusted through the capacitance effect, and the center frequency of the duplexer is dynamically compensated to obtain a frequency compensation signal.

7. The method for automatically adjusting the temperature compensation parameters of a duplexer according to claim 6, characterized in that: The method for automatically adjusting the temperature compensation parameters of the duplexer also includes: Calculating the deviation between the frequency compensation signal and the target frequency response to obtain a compensation error, and combining the compensation error with the temperature characteristic data to construct a training sample pair to obtain network update data; Inputting the network update data into the encoder of the mechanism information neural network, performing feature encoding through the four fully connected layers in the encoder, and performing sparse constraint processing on the encoded features to obtain updated features; Reconstructing the temperature mechanism dictionary for the updated features, updating M atoms of the temperature mechanism dictionary based on a physical compensation mechanism, and normalizing the dictionary using a Schmidt orthogonalization method to obtain an updated dictionary; Performing a feature fusion operation on the updated feature and the updated dictionary, embedding the temperature compensation mechanism information into the feature space through matrix inner product, constructing a mechanism information preservation loss term, and obtaining a second fused feature; Performing decoder reconstruction on the second fusion feature, mapping the feature back to the original space through 4 fully connected layers in the decoder, and performing nonlinear transformation using batch normalization and ReLU activation function to obtain reconstructed features; Based on the reconstruction feature, an overall loss function is constructed, and the reconstruction error term, the sparse constraint term and the mechanism information retention term are weightedly combined to obtain an optimization target, and a back-propagation calculation is performed on the optimization target. The weight of the mechanism information neural network is updated using the Adam optimizer to obtain an updated parameter; The update parameters are imported into the mechanism information neural network, and after multiple rounds of iterative training, the iteration is stopped when the training error is less than the target value to obtain the target neural network model.

8. A device for automatically adjusting the temperature compensation parameters of a duplexer, characterized in that: The method for automatically adjusting the temperature compensation parameters of a duplexer according to any one of claims 1 to 7 comprises: A sampling module is used to sample the working temperature of the duplexer to obtain temperature sampling data, and to extract distribution characteristics of the temperature sampling data to obtain temperature characteristic data; A reconstruction module, used for inputting the temperature characteristic data into the mechanism information neural network for characteristic reconstruction to obtain temperature compensation characteristics; A mapping module, used for performing radial basis function neural network mapping on the temperature compensation characteristics to obtain a frequency drift state quantity of the duplexer; An operation module, used for performing a sliding window adaptive Kalman filter operation on the frequency drift state quantity of the duplexer to obtain a compensation parameter; The adjustment module is used to input the compensation parameter into a feedforward compensation controller, and adjust the voltage-controlled resonator of the duplexer through a sliding mode control structure to obtain a frequency compensation signal.

9. A device for automatically adjusting the temperature compensation parameters of a duplexer, characterized in that: The automatic adjustment device for the temperature compensation parameters of a duplexer comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instruction in the memory to enable the device for automatically adjusting the temperature compensation parameters of a duplexer to perform the method for automatically adjusting the temperature compensation parameters of a duplexer according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by the processor, the method for automatically adjusting the temperature compensation parameters of a duplexer according to any one of claims 1 to 7 is implemented.

Images