CN-122021278-A - Radio frequency power amplifier design method and system based on neural network and multi-algorithm cooperation

CN122021278ACN 122021278 ACN122021278 ACN 122021278ACN-122021278-A

Abstract

The invention discloses a method and a system for designing a radio frequency power amplifier based on cooperation of a neural network and multiple algorithms, and relates to the field of radio frequency power amplifier design, comprising the following steps of constructing a performance prediction model of a power amplifier, presetting a performance target of a power amplifier matching network, and carrying out cooperative iterative optimization on passive element parameters through at least two global optimization algorithms to generate candidate passive element parameter combinations; in each iterative optimization process, the currently generated candidate passive element parameter combination is input into a performance prediction model to obtain a corresponding prediction performance index, and according to the difference between the prediction performance index and a performance target, the passive element parameter is optimized by adopting a corresponding global optimization algorithm in different optimization stages until a termination condition is met to obtain an optimal passive element parameter combination, so that the power amplifier design is realized. The method and the device can solve the problem that the prior art cannot dynamically schedule different algorithms according to the optimization process.

Inventors

QUAN XING
CHEN CHUANYU
WU XINYAN
JIANG HAO
WU YANHUI
GAO XIAOQIANG
ZHAN JINSONG

Assignees

西安电子科技大学

Dates

Publication Date: 20260512
Application Date: 20260120

Claims (10)

1. The design method of the radio frequency power amplifier based on the cooperation of the neural network and the multiple algorithms is characterized by comprising the following steps: constructing a performance prediction model of the power amplifier, wherein the input of the performance prediction model is the passive element parameter of a matching network of the power amplifier, and the output is a prediction performance index; Presetting a performance target of a power amplifier matching network, and performing collaborative iterative optimization on the passive element parameters through at least two global optimization algorithms to generate candidate passive element parameter combinations; and in each iterative optimization process, inputting the currently generated candidate passive element parameter combination into the performance prediction model to obtain a corresponding prediction performance index, and optimizing the passive element parameter by adopting a corresponding global optimization algorithm in different optimization stages according to the difference between the prediction performance index and a performance target until a termination condition is met to obtain an optimal passive element parameter combination, thereby realizing the design of the power amplifier.
2. The method for designing a radio frequency power amplifier based on cooperation of a neural network and a plurality of algorithms according to claim 1, wherein the constructing a performance prediction model of the power amplifier specifically comprises: Acquiring passive element parameters of a power amplifier matching network, performing simulation according to the passive element parameters to obtain corresponding key performance indexes, and taking the passive element parameters and the corresponding key performance indexes as a training data set; and training the convolutional neural network by using the training data set to obtain a performance prediction model.
3. The method for designing the radio frequency power amplifier based on the cooperation of the neural network and the multiple algorithms according to claim 1, wherein the convolutional neural network comprises an input layer, a feature extraction module, a pooling layer, a regression output module and an output layer which are sequentially connected; The feature extraction module is formed by sequentially connecting a plurality of convolution layers, wherein each layer of convolution layer is connected with a batch normalization layer and a ReLU activation function in series; the regression output module is composed of a plurality of full-connection layers in sequence, wherein the rear of each full-connection layer is connected with a Dropout regularization layer.
4. The method for designing a radio frequency power amplifier based on cooperation of a neural network and multiple algorithms according to claim 1, wherein the passive element parameters are optimized by adopting corresponding global optimization algorithms in different optimization stages according to the difference between the predicted performance index and the performance target, specifically comprising: when the performance target of the preset power amplifier matching network is a single target, the passive element parameters are firstly subjected to extensive search in a global exploration stage through a genetic algorithm, and then are subjected to focusing convergence in a local convergence stage through a particle swarm algorithm, so that the optimal passive element parameter combination is obtained.
5. The method for designing a radio frequency power amplifier based on cooperation of a neural network and multiple algorithms according to claim 1, wherein the passive element parameters are optimized by adopting corresponding global optimization algorithms in different optimization stages according to the difference between the predicted performance index and the performance target, and specifically further comprising: when the performance targets of the preset power amplifier matching network are a plurality of targets, parameter searching is carried out in a global exploration stage and a local convergence stage respectively through a genetic algorithm and a particle swarm algorithm to obtain a pareto non-inferior solution; and sequencing and screening the pareto non-inferior solutions in a multi-objective optimization stage through an NSGA-II algorithm to obtain the optimal passive element parameter combination.
6. The method for designing a radio frequency power amplifier based on neural network and multi-algorithm cooperation according to claim 5, wherein the candidate passive element parameter combinations and the corresponding prediction performance indexes generated by the genetic algorithm, the particle swarm optimization and the NSGA-II algorithm in the corresponding optimization stage are stored in a global shared solution pool, Extracting candidate passive element parameter combinations with preset proportions from the global shared solution pool according to the current optimization stage, and continuing to optimize the candidate passive element parameter combinations serving as initial populations corresponding to a genetic algorithm, a particle swarm algorithm or an NSGA-II algorithm; the optimization stage comprises a global exploration stage, a local convergence stage and a multi-objective optimization stage, wherein the global exploration stage adopts a genetic algorithm for optimization, the local convergence stage adopts a particle swarm algorithm for optimization, and the multi-objective optimization stage adopts an NSGA-II algorithm for optimization.
7. A radio frequency power amplifier design system based on neural network and multi-algorithm cooperation, comprising: The model construction module is used for constructing a performance prediction model of the power amplifier, wherein the input of the performance prediction model is the passive element parameter of the power amplifier matching network, and the output is a predicted performance index; And in each iterative optimization process, inputting the currently generated candidate passive element parameter combination into the performance prediction model to obtain a corresponding prediction performance index, and optimizing the passive element parameter by adopting a corresponding global optimization algorithm in different optimization stages according to the difference between the prediction performance index and the performance target until a termination condition is met to obtain an optimal passive element parameter combination, thereby realizing the design of the power amplifier.
8. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable in the processor, the processor implementing the steps of a method for designing a radio frequency power amplifier based on neural network and multi-algorithm cooperation as claimed in any one of claims 1 to 6 when the computer program is executed by the processor.
9. A computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program when executed by a processor implements the steps of a method for designing a radio frequency power amplifier based on cooperation of a neural network and multiple algorithms according to any one of claims 1 to 6.
10. A computer program product comprising a computer program, characterized in that the computer program when executed by a processor implements the steps of a method for designing a radio frequency power amplifier based on cooperation of a neural network and multiple algorithms according to any one of claims 1 to 6.

Description

Radio frequency power amplifier design method and system based on neural network and multi-algorithm cooperation Technical Field The invention relates to the field of radio frequency power amplifier design, in particular to a method and a system for designing a radio frequency power amplifier based on cooperation of a neural network and multiple algorithms. Background A Power Amplifier (PA) is a key module in a modern wireless communication system, and its performance directly affects the transmission quality, coverage and system Power consumption of a signal. In many Power amplifier structures, the class E Power amplifier is paid attention to by virtue of its high efficiency characteristic under high frequency condition, and the core principle of its efficient operation is that by precisely controlling the waveform phase of voltage and current by using zero voltage switching (Zero Voltage Switching, ZVS) technology, the voltage drop of the transistor at the moment of conduction is zero, so as to remarkably reduce the switching loss, and improve the overall energy conversion efficiency, however, the excellent performance of the class E Power amplifier is highly dependent on the precise design of its output matching circuit, in the actual design, on-chip parasitic parameters (such as wiring inductance, substrate capacitance), nonlinear characteristics of the transistor, and device loss, etc., so that the matching circuit parameters derived based on the ideal formula deviate seriously from expectations, and the actual Power added efficiency (Power ADDED EFFICIENCY, PAE) and the output Power are difficult to reach theoretical optimal values. The traditional design of the class-E power amplifier, in particular to the parameter determination of an output matching circuit thereof, generally depends on preliminary calculation based on an analysis expression deduced under the assumption of ideal devices and zero parasitic parameters, and is combined with circuit simulation software to carry out iterative adjustment on the basis, however, in the actual integrated circuit process, parasitic parameter effects such as wiring inductance, substrate capacitance and the like are not negligible. This results in a huge deviation between the initial circuit parameters calculated based on the ideal formula and the actual simulation results, and thus, it is difficult to achieve accurate impedance transformation when designing for the first time, and it is difficult to achieve optimal performance (e.g., high efficiency, high output power) of the booster power amplifier. Secondly, in order to compensate the deviation and meet the requirement of multi-target performance, the design flow is seriously dependent on manual experience to perform the cyclic trial and error of 'simulation-fine tuning-re-simulation'. The designer needs to search manually in a multidimensional and strongly coupled circuit parameter space, the searching mode is extremely low in efficiency and extremely easy to sink into a local optimal solution, and a globally optimal parameter combination is difficult to find, so that the design period is long, and the quality of a final design result cannot be guaranteed. Therefore, the conventional design method severely depends on experience of a designer, and reduces the searching range of parameters by experience, but determines an optimal parameter combination meeting the requirement of multi-objective performance within the range, and still needs to perform a large number of 'simulation-evaluation-adjustment' iterations. In recent years, with the application of machine learning technology in the field of circuit design, a technical solution for predicting circuit performance by constructing a data set and training a neural network model, and further assisting parameter optimization, for example, a general machine learning driving design framework for multiple power amplifier topologies is provided in reference Xuzhe Zhao,"Machine Learning Assisted Design of mmWave Wireless Transceiver Circuits",pp.31-35,Jul.2024., the solution aims at processing multiple topologies, and is a "general-purpose" tool, and the depth optimization is not performed for the parameter coupling characteristic and performance sensitivity of a specific task of an E-type power amplifier matching circuit, so that the accuracy and efficiency of the E-type power amplifier matching circuit under the specific scene are not optimal, and the parameter recovery is completely dependent on a differentiable model and a gradient-based optimization algorithm. In a circuit parameter space with high nonlinearity and multimodal characteristics, the method is extremely sensitive to initial values, is easy to sink into a local optimal solution, and is difficult to realize global-range parameter search, which is the key to finding the optimal working point of the class E power amplifier. In addition, the scheme adopts a single optimization strategy, does not discu