LSTM-GO-based class-F power amplifier matching parameter optimization method and device

Abstract

The invention discloses an LSTM-GO-based class-F power amplifier matching parameter optimization method and device, and the method comprises the steps: obtaining a node LC matching network according to a class-F power amplifier; obtaining a node LC matching network structure, parameters and a parameter value range according to the node LC matching network; processing the node LC matching network parameters and the parameter value range by adopting a preset optimization model to obtain optimal node LC matching network parameters and a circuit power additional efficiency peak value; wherein the preset optimization model comprises a trained LSTM neural network behavior level model and a GO algorithm embedded into the trained LSTM neural network behavior level model, and the trained LSTM neural network behavior level model is obtained by taking data of a preset category as a training data set and training an initial LSTM neural network behavior level model. According to the invention, the design accuracy of the class-F power amplifier can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of radio frequency circuit automated design, and specifically relates to a method and device for optimizing matching parameters of a class F power amplifier based on LSTM-GO. Background Art

[0002] As a crucial component of the RF transmitter in a communication system, the efficiency of the RF power amplifier (RFPA) significantly impacts the overall system performance. With the rapid development of integrated electronic systems, further improving the overall efficiency of the power amplifier within the operating frequency band has become a key research topic in RF power amplifier design in recent years. However, the interplay between the input parameters of Class F PAs, as well as between input parameters and output characteristics, hinders rapid improvement in power-added efficiency (PAE) during circuit design, significantly increasing circuit design cycles. Therefore, an optimization method is needed to rapidly optimize PAE.

[0003] In the prior art, power amplifier matching network design is primarily based on computer-aided design (CAD). In CAD, the physical model-based parameter association mechanism results in redundant simulation process parameters due to the mutual independence of circuit device model parameters. This increases computational complexity, requires the design process to consider more physical effects, and increases design costs. With the increasing demand for wireless communication technology, neural networks (NNs) are being used to simulate complex power amplifier (PA) characteristics to improve the nonlinear distortion of PAs. However, it is not possible to determine whether a PA without nonlinear distortion processing has optimal circuit performance. Therefore, there is an urgent need to improve the above-mentioned defects in the prior art. Summary of the Invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a method and device for optimizing matching parameters of a Class F power amplifier based on LSTM-GO. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0005] In a first aspect, the present invention provides a method for optimizing matching parameters of a class F power amplifier based on LSTM-GO, comprising:

[0006] According to the class F power amplifier, a joint LC matching network is obtained; according to the joint LC matching network, a joint LC matching network structure, parameters and parameter value ranges are obtained;

[0007] The preset optimization model is used to process the parameters of the joint LC matching network and the parameter value range to obtain the optimal joint LC matching network parameters and the peak power added efficiency of the circuit;

[0008] Among them, the preset optimization model includes a trained LSTM neural network behavioral level model and a GO algorithm embedded in the trained LSTM neural network behavioral level model. The trained LSTM neural network behavioral level model uses preset category data as a training data set to train the initial LSTM neural network behavioral level model.

[0009] In a second aspect, the present invention further provides a Class F power amplifier matching parameter optimization device based on LSTM-GO, comprising:

[0010] The data acquisition module is used to obtain a section-connected LC matching network according to the class F power amplifier; and obtain a section-connected LC matching network structure, parameters, and parameter value ranges according to the section-connected LC matching network;

[0011] A data processing module is used to process the parameters of the joint LC matching network and the parameter value range using a preset optimization model to obtain the optimal joint LC matching network parameters and the peak power added efficiency of the circuit;

[0012] Among them, the preset optimization model includes a trained LSTM neural network behavioral level model and a GO algorithm embedded in the trained LSTM neural network behavioral level model. The trained LSTM neural network behavioral level model uses preset category data as a training data set to train the initial LSTM neural network behavioral level model.

[0013] Beneficial effects of the present invention:

[0014] This invention provides a method and device for optimizing matching parameters for a Class F power amplifier based on LSTM-GO. By training an LSTM neural network behavioral model using a simulated dataset and combining this trained LSTM neural network behavioral model with the GO algorithm, optimal matching network parameters can be directly obtained to achieve the maximum peak power-added efficiency of the circuit. This method avoids the redundant design parameters encountered in traditional analytical and CAD methods, obviating the need to consider multiple physical issues during the design process and improving design accuracy.

[0015] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of a method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO provided in an embodiment of the present invention;

[0017] Figure 2 1 is a schematic diagram of a method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO provided in an embodiment of the present invention;

[0018] Figure 3 This is a schematic diagram of the loss value when training the LSTM neural network behavior level model provided by an embodiment of the present invention;

[0019] Figure 4 This is a schematic diagram of the PAE optimization results of an experimental Class F power amplifier using the LSTM neural network behavioral level model combined with the GO algorithm provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The present invention will be further described in detail below with reference to specific examples, but the embodiments of the present invention are not limited thereto.

[0021] See Figure 1 and Figure 2 , Figure 1 This is a flow chart of a method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO provided by an embodiment of the present invention. Figure 2 1 is a schematic diagram of a method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO provided by an embodiment of the present invention. The method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO provided by the present invention includes:

[0022] S101. Obtain a joint LC matching network according to a class F power amplifier; and obtain a joint LC matching network structure, parameters, and parameter value ranges according to the joint LC matching network.

[0023] Specifically, in this embodiment, based on an existing semiconductor device model, a conventional design method is used to design a Class F power amplifier, further obtaining a bias circuit, a transistor parallel circuit, and a joint LC matching circuit. Optionally, the joint LC matching circuit is implemented as a microstrip impedance matching network, and the network parameters of the joint LC matching circuit include the WL values ​​of each node of the microstrip impedance matching network. It is understood that the joint LC network is implemented through a microstrip impedance network, and the WL values ​​of each node of its microstrip line can fully reflect the characteristics of the impedance network.

[0024] It should be noted that the form of the training data set must reflect the parameters of the matching network characteristics. According to the actual process library, the microstrip impedance matching network parameters WL are set as parameter input.

[0025] S102, using a preset optimization model to process the parameters of the joint LC matching network and the parameter value range to obtain the optimal joint LC matching network parameters and the peak power added efficiency of the circuit;

[0026] The preset optimization model includes a trained LSTM neural network behavioral model and a GO algorithm embedded in the trained LSTM neural network behavioral model. The trained LSTM neural network behavioral model is obtained by training the initial LSTM neural network behavioral model using data from preset categories as a training dataset. Optionally, the trained LSTM neural network behavioral model and the GO algorithm are connected by passing matching network parameters and circuit PAE of a specific matching structure. The input of the GO algorithm and the input of the LSTM behavioral model are both the WL values ​​of the specific microstrip impedance matching network to fully reflect the matching network characteristics. The optimization goal of the GO algorithm is to adjust the WL values ​​of each microstrip matching network to optimize the PAE output by the LSTM neural network behavioral model.

[0027] Specifically, in this embodiment, the trained LSTM neural network behavior-level model includes a plurality of sequentially connected trained LSTM units; a preset optimization model is used to process the node-connected LC matching network parameters and parameter value ranges, including:

[0028] The trained LSTM units connected in sequence are used to process the parameters of the LC matching network to obtain the trained LSTM unit state C t and the hidden state vector h t ;

[0029] For the trained LSTM unit state C t and the hidden state vector h t After processing, the peak power added efficiency of the circuit is obtained.

[0030] In this embodiment, the trained LSTM unit includes a forget gate, an input gate, an output gate, and a nonlinear output; the parameters of the node-connected LC matching network are processed using the trained LSTM units connected in sequence, including:

[0031] The node-connected LC matching network parameter x is processed for the t-th trained LSTM unit, including:

[0032] The forget gate multiplies the input node-connected LC matching network parameter x with the corresponding forget gate weight matrix to obtain the result 1, and the previous hidden state vector h t-1 Multiply it with the corresponding forget gate weight matrix to get result 2, linearly add result 1 and result 2 to get the first linear combination value; after processing the first linear combination value with the sigma function, the forget gate output f is obtained. t ;

[0033] The input gate multiplies the node LC matching network parameter x with the corresponding input gate weight matrix to obtain the result three, and the previous hidden state vector h t-1Multiply it with the corresponding input gate weight matrix to get result 4, linearly add result 3 and result 4 to get the second linear combination value; after processing the second linear combination value with the sigma function, the input gate output i is obtained. t ;

[0034] The output gate multiplies the node LC matching network parameter x with the corresponding output gate weight matrix to obtain the result five, and the previous hidden state vector h t-1 Multiply it with the corresponding output gate weight matrix to get the result six, linearly add the result five and the result six to get the third linear combination value; after the third linear combination value is processed by the sigma function, the output gate output o is obtained. t ;

[0035] The nonlinear output multiplies the node LC matching network parameter x with the corresponding nonlinear weight matrix to obtain the result seven, and the previous hidden state vector h t-1 Multiply it with the corresponding nonlinear weight matrix to get result eight, linearly add result seven and result eight to get the fourth linear combination value; after processing the fourth linear combination value with the tanh function, the nonlinear output g is obtained. t ;

[0036] The output of the forget gate is f t and the previously trained LSTM unit state C t-1 Multiply, get the first multiplication result, and output the input gate i t With nonlinear output g t Multiply them together to get the second multiplication result, add the first multiplication result to the second multiplication result to get the t-th trained LSTM unit state C t ;

[0037] The t-th trained LSTM unit state C t After being processed by the tanh function, the output gate outputs o t Multiply them together to get the hidden state vector h of the t-th trained LSTM unit t .

[0038] In this embodiment, the GO algorithm includes an optimization target determination phase, a population initialization phase, and a loop iteration phase. The algorithm uses a preset optimization model to process the parameters of the node-connected LC matching network and the parameter value range, and also includes:

[0039] In the optimization target determination stage, the optimal parameters of the LC matching network and the peak power added efficiency of the circuit are found as the optimization target, and the relevant parameters of the GO algorithm and the value range of the LC matching network parameters are initialized;

[0040] In the population initialization phase, the population is initialized; wherein the population is N rows and D columns of data consisting of the parameters of the node-connected LC matching network and its corresponding circuit power added efficiency peak, N represents the set number of iterations, D represents the number of columns of the population individual data, and the population includes an upper bound ub and a lower bound lb;

[0041] In the loop iteration stage, the positions of the individuals in the population are updated by loop iteration, and the optimal positions of the individuals in the population are found, thereby obtaining the optimal parameters of the joint LC matching network. Based on the optimal parameters of the joint LC matching network, the optimal peak value of the circuit power added efficiency is obtained.

[0042] In this embodiment, the iterative cycle phase includes a learning phase and a reflection phase; wherein,

[0043] In the learning phase, the gap Gap1 between the better individual and the best individual, the gap Gap2 between the worse individual and the best individual, the gap Gap3 between the better individual and the worse individual, and the gap Gap4 between two random individuals in the population are calculated, which can be expressed as:

[0044]

[0045] Normalize Gap1, Gap2, Gap3 and Gap4 to get the normalized result LF k , used to characterize the degree of improvement of individuals in the current population during learning, expressed as:

[0046]

[0047] According to the growth resistance GR of the individuals in the population i and the maximum growth resistance GR in individuals of the population max , get the growth resistance SF of the individual in the population i , expressed as:

[0048]

[0049] LF k , SF i and Gap k Multiply them together to get the knowledge KA that the current population individual gets from the kth group k , expressed as:

[0050] KA k =SF i ·LF k Gap k ,(k=1,2,3,4);

[0051] According to the knowledge KA1, KA2, KA3 and KA4 of the current population individuals, the growth of the current population individuals after learning is obtained, which is expressed as:

[0052]

[0053] According to the growth of individual learning in the current population, the current iteration result is obtained, which is expressed as:

[0054]

[0055] Among them, r1 represents a random number in the range [0,1]. Represents the result after N iterations of the i-th population individual;

[0056] In the reflection stage, for the j-th dimension data of the i-th population individual Expressed as:

[0057]

[0058] Among them, r2, r3, r4 and r5 are uniformly distributed random numbers in the range of [0,1], lb represents the lower limit of the search domain, ub represents the upper limit of the search domain, AF represents the attenuation coefficient, which will gradually converge to 0.01 with the iteration, R j represents the j aspects of an ordered individual found in the previous sort, Represents the current population individual The jth aspect of .

[0059] In this embodiment, the process of constructing the training data set includes:

[0060] Simulating a class F power amplifier using a preset method to obtain matching circuit parameters and a peak power added efficiency of the class F power amplifier; optionally, the preset method may be an ADS simulation method;

[0061] The parameters of the class F power amplifier matching circuit are used as samples in the training data set, and the peak power added efficiency of the class F power amplifier matching circuit is used as the true label corresponding to the sample.

[0062] In this embodiment, the matching network component parameters used in each simulation are treated as an n-dimensional array and arranged in a certain order as a data sample, and the simulation result PAE of this data is used as the true label of the sample to obtain a constructed training data set.

[0063] It can be understood that ADS (Advanced Design System) is an advanced design system and a leading electronic design automation software suitable for RF, microwave and signal integrity applications.

[0064] In this embodiment, the initial LSTM neural network behavior-level model is trained, including:

[0065] Input some samples in the training data set into the j-th LSTM neural network behavior-level model to be trained for training, and obtain the prediction results output during the j-th training process; optionally, the LSTM neural network behavior-level model to be trained is implemented by calling the lstmLayer function in MATLAB;

[0066] Calculate the classification loss based on the prediction results output during the j-th training process and the true labels of the samples of the j-th LSTM neural network behavior-level model to be trained, and use it as the classification loss of the j-th training process;

[0067] Backpropagation is performed based on the classification loss of the j-th training process to update the network parameters of the j-th LSTM neural network behavioral level model to be trained, and the j+1-th LSTM neural network behavioral level model to be trained is obtained; this is iterated until the number of training times or the degree of convergence meets the preset conditions, and a trained LSTM neural network behavioral level model is obtained.

[0068] In this embodiment, it also includes:

[0069] When the classification loss of the j-th training process is greater than the preset loss value, increase the samples of the j-th training process, adjust the network parameters of the j-th LSTM neural network behavioral level model to be trained, and continue training until the classification loss of the j-th training process is less than the preset loss value.

[0070] In addition, this embodiment also includes a test dataset. The training dataset is used to train the LSTM neural network behavioral model, and the test dataset is used to verify the reliability of the trained LSTM neural network behavioral model. It should be noted that the number of samples in the training dataset and the number of samples in the test dataset can be determined based on actual conditions.

[0071] When the trained LSTM neural network behavioral level model is tested using the test data set, if the test fails, the system returns to the training stage, increases the sample capacity of the training data set, and continues to train the model; when the test passes, a power amplifier LSTM neural network behavioral level model with a certain process, matching structure, and bias structure is obtained.

[0072] In summary, the present invention provides a method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO. First, a training data set is constructed through radio frequency circuit simulation, and a Class F power amplifier LSTM neural network behavioral level model of a specific process is trained using the training data set. The trained LSTM neural network behavioral level model can accurately predict the circuit PAE; secondly, the trained LSTM neural network behavioral level model is used in combination with the GO optimization algorithm to optimize the WL value of each node in the matching network so that the Class F power amplifier obtains the maximum PAE. Compared with the existing CAD method, it is only necessary to determine the initial parameters of the matching network under the process and circuit structure to optimize the matching network parameters that maximize the circuit PAE. The prediction method of the present invention is not affected by the parameter association mechanism based on physics, which makes the process of finding the maximum PAE more standardized and more accurate.

[0073] Furthermore, the present invention requires computer resources to perform circuit simulation of a Class F power amplifier with a microstrip impedance matching network. This method uses the simulated dataset and trains an LSTM neural network behavioral model. Combining this behavioral model with the GO algorithm directly yields the optimal matching network parameters for maximum PAE. This approach avoids the complexity of design parameters in traditional analytical and CAD methods, obviating the need to consider multiple physical considerations during the design process and improving design accuracy.

[0074] Based on the same inventive concept, the present invention further provides a device for optimizing matching parameters of a Class F power amplifier based on LSTM-GO, which is used to implement the method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO provided in the above embodiment of the present invention. The embodiment of the method can be referred to above and will not be described in detail here. The device includes:

[0075] A data acquisition module is used to obtain a joint LC matching network based on a class F power amplifier; and obtain a joint LC matching network structure, parameters, and parameter value ranges based on the joint LC matching network;

[0076] A data processing module is used to process the parameters of the joint LC matching network and the parameter value range using a preset optimization model to obtain the optimal joint LC matching network parameters and the peak power added efficiency of the circuit;

[0077] Among them, the preset optimization model includes a trained LSTM neural network behavioral level model and a GO algorithm embedded in the trained LSTM neural network behavioral level model. The trained LSTM neural network behavioral level model uses preset category data as a training data set to train the initial LSTM neural network behavioral level model.

[0078] In an optional embodiment of the present invention, the effect of the LSTM-GO based class F power amplifier matching parameter optimization method provided in the above embodiment is verified through simulation experiments, specifically:

[0079] See Figure 3 and Figure 4 , Figure 3 This is a schematic diagram of the loss value when training the LSTM neural network behavior level model provided by an embodiment of the present invention. Figure 4 This is a schematic diagram of the PAE optimization results of the experimental Class F power amplifier using the LSTM neural network behavior level model combined with the GO algorithm provided by the embodiment of the present invention; Figure 3 It can be seen that in the embodiment provided by the present invention, after training the LSTM neural network behavior-level model for multiple rounds, the loss value is close to 0, and the difference between its output value and the true label is close to 0, indicating that the trained LSTM neural network behavior-level model has higher reliability; Figure 4 This shows that the embodiment provided by the present invention optimizes the circuit matching network parameters and finds the optimal circuit matching network parameter values ​​with good convergence. The Fitness value is the reciprocal of the PAE peak value of the current iteration result under the number of iterations.

[0080] It should be noted that, in this document, relational terms such as first and second are used solely to distinguish one entity or operation from another, and do not necessarily require or imply any actual relationship or order between these entities or operations. Furthermore, the terms "comprise," "include," or any other variations thereof are intended to encompass non-exclusive inclusion, such that an article or device comprising a list of elements includes not only those elements but also other elements not explicitly listed. Without further limitation, an element defined by the phrase "comprising a..." does not preclude the presence of additional identical elements in the article or device comprising the element. Terms such as "connected" or "connected" are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. References to orientations or positional relationships, such as "upper," "lower," "left," and "right," are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate description and simplify the description of the present invention. They do not indicate or imply that the device or element referred to must have, be constructed, or operate in a specific orientation, and are therefore not to be construed as limiting the present invention.

[0081] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described can be combined in any suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification.

[0082] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO, characterized in that: include: According to the class F power amplifier, a joint LC matching network is obtained; according to the joint LC matching network, a structure, parameters and parameter value ranges of the joint LC matching network are obtained; Using a preset optimization model to process the parameters of the joint LC matching network and the parameter value range, the optimal parameters of the joint LC matching network and the peak power added efficiency of the circuit are obtained; Among them, the preset optimization model includes a trained LSTM neural network behavior-level model and a GO algorithm embedded in the trained LSTM neural network behavior-level model. The trained LSTM neural network behavior-level model uses preset category data as a training data set to train the initial LSTM neural network behavior-level model.

2. The LSTM-GO-based class F power amplifier matching parameter optimization method according to claim 1, characterized in that: The trained LSTM neural network behavior-level model includes a plurality of sequentially connected trained LSTM units; the preset optimization model is used to process the parameters of the node-connected LC matching network and the parameter value range, including: The trained LSTM units connected in sequence are used to process the parameters of the node-connected LC matching network to obtain the trained LSTM unit state C t and the hidden state vector h t ; For the trained LSTM unit state C t and the hidden state vector h t After processing, the peak power added efficiency of the circuit is obtained.

3. The method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO according to claim 2, characterized in that: The trained LSTM unit includes a forget gate, an input gate, an output gate, and a nonlinear output; and the sequentially connected trained LSTM units are used to process the parameters of the node-connected LC matching network, including: The node-connected LC matching network parameter x is processed for the t-th trained LSTM unit, including: The forget gate multiplies the input LC matching network parameter x with the corresponding forget gate weight matrix to obtain the result 1, and the previous hidden state vector h t-1 Multiply the result 1 and the result 2 by the corresponding forget gate weight matrix to obtain the second result, linearly add the result 1 and the result 2 to obtain the first linear combination value; process the first linear combination value through the sigma function to obtain the forget gate output f t ; The input gate multiplies the node-connected LC matching network parameter x with the corresponding input gate weight matrix to obtain result three, and the previous hidden state vector h t-1 Multiply the result 3 and the result 4 by the corresponding input gate weight matrix to obtain the result 4, linearly add the result 3 and the result 4 to obtain the second linear combination value; after the second linear combination value is processed by the sigma function, the input gate output i is obtained. t ; The output gate multiplies the node-connected LC matching network parameter x with the corresponding output gate weight matrix to obtain the result five, and the previous hidden state vector h t-1 Multiply the result 5 and the result 6 by the corresponding output gate weight matrix to obtain the result 6, linearly add the result 5 and the result 6 to obtain the third linear combination value; after the third linear combination value is processed by the sigma function, the output gate output o is obtained. t ; The nonlinear output multiplies the node-connected LC matching network parameter x with the corresponding nonlinear weight matrix to obtain the result seven, and the previous hidden state vector h t-1 Multiply the result 7 and the result 8 with the corresponding nonlinear weight matrix to obtain the result 8, linearly add the result 7 and the result 8 to obtain the fourth linear combination value; after the fourth linear combination value is processed by the tanh function, the nonlinear output g is obtained. t ; The forget gate output f t and the previously trained LSTM unit state C t-1 Multiply, get the first multiplication result, and output the input gate i t With nonlinear output g t Multiply them to get the second multiplication result, add the first multiplication result and the second multiplication result to get the t-th trained LSTM unit state C t ; The t-th trained LSTM unit state C t After being processed by the tanh function, the output gate outputs o t Multiply them together to get the hidden state vector h of the t-th trained LSTM unit t .

4. The method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO according to claim 1, wherein: The GO algorithm includes an optimization target determination phase, a population initialization phase, and a loop iteration phase; the preset optimization model is used to process the parameters of the node-connected LC matching network and the parameter value range, and further includes: In the optimization target determination stage, the optimal parameters of the joint LC matching network and the peak power added efficiency of the circuit are found as optimization targets, and the relevant parameters of the GO algorithm and the value range of the joint LC matching network parameters are initialized; In the population initialization stage, the population is initialized; wherein the population is N rows and D columns of data consisting of the parameters of the node-connected LC matching network and the corresponding circuit power added efficiency peak, N represents the set number of iterations, D represents the number of columns of individual population data, and the population includes an upper bound ub and a lower bound lb; In the loop iteration stage, the positions of the population individuals are updated by loop iteration to find the optimal positions of the population individuals, thereby obtaining the optimal parameters of the joint LC matching network. Based on the optimal parameters of the joint LC matching network, the optimal circuit power added efficiency peak is obtained.

5. The method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO according to claim 4, characterized in that: The iterative cycle phase includes a learning phase and a reflection phase; wherein, In the learning phase, the gap Gap1 between the better individual and the best individual, the gap Gap2 between the worse individual and the best individual, the gap Gap3 between the better individual and the worse individual, and the gap Gap4 between two random individuals in the population are calculated and expressed as: Normalize Gap1, Gap2, Gap3 and Gap4 to get the normalized result LF k , used to characterize the degree of improvement of individuals in the current population during learning, expressed as: According to the growth resistance GR of the individuals in the population i and the maximum growth resistance GR in individuals of the population max , get the growth resistance SF of the individual in the population i , expressed as: LF k , SF i and Gap k Multiply them together to get the knowledge KA that the current population individual gets from the kth group k , expressed as: YOU k =SF i ·LF k ·Gap k ,(k=1,2,3,4); According to the knowledge KA1, KA2, KA3 and KA4 of the current population individuals, the growth of the current population individuals after learning is obtained, which is expressed as: According to the growth of individual learning in the current population, the current iteration result is obtained, which is expressed as: Among them, r1 represents a random number in the range [0,1]. Represents the result after N iterations of the i-th population individual; In the reflection stage, for the j-th dimension data of the i-th population individual Expressed as: Among them, r2, r3, r4 and r5 are uniformly distributed random numbers in the range of [0,1], lb represents the lower limit of the search domain, ub represents the upper limit of the search domain, AF represents the attenuation coefficient, R j represents the j aspects of an ordered individual found in the previous sort, Represents the current population individual The jth aspect of .

6. The method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO according to claim 1, wherein: The process of constructing the training dataset includes: The class F power amplifier is simulated using a preset method to obtain the matching circuit parameters and the peak power added efficiency of the matching circuit of the class F power amplifier; The class F power amplifier matching circuit parameters are used as samples in a training data set, and the class F power amplifier matching circuit power added efficiency peak value is used as a true label corresponding to the sample.

7. The LSTM-GO-based class F power amplifier matching parameter optimization method according to claim 6, characterized in that: Training the initial LSTM neural network behavior-level model includes: Inputting some samples in the training data set into the j-th LSTM neural network behavior-level model to be trained for training, and obtaining the prediction results output during the j-th training process; Calculate the classification loss based on the prediction results output during the j-th training process and the true labels of the samples of the j-th LSTM neural network behavior-level model to be trained, and use it as the classification loss of the j-th training process; Backpropagation is performed based on the classification loss of the j-th training process to update the network parameters of the j-th LSTM neural network behavioral level model to be trained, and the j+1-th LSTM neural network behavioral level model to be trained is obtained; this is iterated until the number of training times or the degree of convergence meets the preset conditions, and the trained LSTM neural network behavioral level model is obtained.

8. The method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO according to claim 7, characterized in that: Also includes: When the classification loss of the j-th training process is greater than the preset loss value, the samples of the j-th training process are increased, the network parameters of the j-th LSTM neural network behavioral level model to be trained are adjusted, and training is continued until the classification loss of the j-th training process is less than the preset loss value.

9. The method for optimizing matching parameters of a Class F power amplifier based on LSTM-GO according to claim 1, wherein: The node-connected LC matching network parameters include WL values ​​of each node in the microstrip impedance matching network.

10. A Class F power amplifier matching parameter optimization device based on LSTM-GO, characterized in that: include: A data acquisition module is used to obtain a joint LC matching network based on a class F power amplifier; and obtain the joint LC matching network structure, parameters and parameter value ranges based on the joint LC matching network; A data processing module is used to process the parameters of the joint LC matching network and the parameter value range using a preset optimization model to obtain the optimal parameters of the joint LC matching network and the peak power added efficiency of the circuit; Among them, the preset optimization model includes a trained LSTM neural network behavior-level model and a GO algorithm embedded in the trained LSTM neural network behavior-level model. The trained LSTM neural network behavior-level model uses preset category data as a training data set to train the initial LSTM neural network behavior-level model.

Images