Radio frequency power amplifier behavior modeling method based on CG-BPNN technology

Abstract

The invention discloses a radio frequency power amplifier behavior modeling method based on a conjugate gradient back propagation neural network (CG-BPNN) technology, and the method comprises the following three steps: data preparation, CG-BPNN model prediction, and model determination. According to the method, the CG-BPNN modeling method is adopted as the core, the conjugate gradient optimization algorithm is used for updating the weight, falling into a local optimal solution can be effectively avoided, and a global optimal solution can be achieved within a relatively small number of iterations. According to the method, accurate performance prediction and behavior characterization of the radio frequency power amplifier can be realized in combination with limited measurement or simulation data, so that the design efficiency and precision of a radio frequency microwave circuit are further improved, and a more efficient and convenient technical means is provided for microwave engineers.

Description

Technical Field

[0001] The present invention relates to the research field of nonlinear modeling of radio frequency power amplifiers, and more specifically, to a behavior modeling method of radio frequency power amplifiers based on CG-BPNN technology. Background Art

[0002] For mobile communication systems, the power amplifier is a key component of the RF front end. It can increase the transmission power level of the radio signal and is the main source of signal distortion, spurious, harmonics and intermodulation products. However, the nonlinearity of the power amplifier will interfere with the signal transmission quality, so an accurate power amplifier model is needed to quantify and predict its nonlinear characteristics.

[0003] As an important means to solve the nonlinear problem of power amplifiers, behavioral models have attracted widespread attention because they only need input and output data to accurately establish the behavioral model of power amplifiers. It is mainly divided into series models and artificial neural network models, among which series models are only suitable for modeling weak nonlinearities, and as the nonlinearity of power amplifiers increases, the computational complexity will increase. Subsequently, artificial neural networks can fit any nonlinear relationship by borrowing the powerful computing power of computers, and have played a huge advantage in the behavioral modeling of power amplifiers. Rumelhart and other scholars first proposed the BPNN model, which uses the gradient descent method to minimize the error. However, the training efficiency and convergence speed of BPNN will decrease with the increase in the number of training times, especially when the objective function has multiple minima, the error function will usually fall into an indistinguishable minimum point and stop optimizing, resulting in the inability of training to converge. At the same time, the generalization ability of BPNN is not stable enough in repeated experiments. Therefore, in the practical application of some BPNN regression problems, in order to obtain ideal accuracy, some auxiliary algorithms are needed. In summary, the core of the present invention uses a conjugate gradient algorithm to train BPNN to characterize the nonlinear relationship between the performance parameters of RF power amplifiers, providing microwave engineers with a more efficient and convenient modeling technology. Summary of the invention

[0004] The purpose of the present invention is to establish an accurate nonlinear model of a radio frequency power amplifier and propose a radio frequency power amplifier behavior modeling method based on CG-BPNN technology, which provides important guidance for improving the design efficiency and performance of communication systems.

[0005] The technical method and solution of the present invention to solve the above-mentioned RF power amplifier behavior modeling are as follows: a RF power amplifier behavior modeling method based on CG-BPNN technology, including three steps of data preparation, CG-BPNN model prediction and model determination;

[0006] Step 1 Data preparation is to set the characteristic parameters that affect the performance of the RF power amplifier X = [X1, X2, ..., X n ] as input variables, and set the step size of each input variable. The behavior parameters composed of [Pout, PAE, Gain] are used as output variables, and multiple groups of performance indicators are obtained based on measurement or simulation. After importing them into the modeling software, select the appropriate method for data preprocessing;

[0007] The CG-BPNN model prediction in step 2 is to randomly divide the data into a training set and a test set according to a certain ratio, use a function to build a CG-BPNN model, and set the model's input layer number, hidden layer number and number, output layer number, maximum number of iterations, optimal learning rate, optimal error tolerance and other hyperparameters, and then start training the CG-BPNN model. If each hyperparameter reaches the preset requirements or the number of iterations reaches the maximum value, the training ends, and then the test set is used to verify the trained model;

[0008] The model is determined in step 3. If the mean square error (MSE) between the expected value and the predicted value is greater than 0.05, return to step 2 to adjust the hyperparameters and retrain the CG-BPNN model until the requirements are met. When the mean square error (MSE) is less than 0.05, the model is determined, and the RF power amplifier behavior modeling is completed.

[0009] The mean square error MSE is calculated as follows:

[0010]

[0011] Among them, MSE is the result of mean square error, n is the number of prediction samples, P i is the i-th predicted value, T i is the corresponding i-th expected value.

[0012] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0013] The core of the present invention adopts the CG-BPNN modeling method, which uses the conjugate gradient method as an optimization algorithm to update the weights, can effectively avoid falling into the local optimal solution, and can reach the global optimal solution within a relatively small number of iterations. The present invention can accurately characterize the performance parameters of the RF power amplifier using limited measurement or simulation data, thereby further improving the design efficiency and performance of RF microwave circuits, and providing microwave engineers with a more efficient and convenient technical means. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Modeling process for the overall behavior of the present invention;

[0015] Figure 2The GaN HEMTPA and GaN DohertyPA performance parameter prediction model of the present invention;

[0016] Figure 3 The GaN HEMTPA performance parameter fitting result when the frequency is 2.2 GHz in the present invention;

[0017] Figure 4 The GaN HEMTPA performance parameter fitting result when the frequency is 2.3 GHz in the present invention;

[0018] Figure 5 The GaN HEMTPA performance parameter fitting result when the frequency is 2.4 GHz in the present invention;

[0019] Figure 6 The performance parameter fitting result of GaN Doherty PA of the present invention when the frequency is 2.2 GHz;

[0020] Figure 7 The performance parameter fitting result of GaN Doherty PA of the present invention when the frequency is 2.3 GHz;

[0021] Figure 8 This is the GaN Doherty PA performance parameter fitting result when the frequency is 2.4 GHz. DETAILED DESCRIPTION

[0022] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0023] like Figure 1 As shown, a RF power amplifier behavior modeling method based on CG-BPNN technology includes three steps: data preparation, CG-BPNN model prediction, and model determination;

[0024] Step 1 Data preparation is to set the characteristic parameters that affect the performance of the RF power amplifier X = [X1, X2, ..., X n ] as input variables, and set the step size of each input variable. The behavior parameters composed of [Pout, PAE, Gain] are used as output variables, and multiple groups of performance indicators are obtained based on measurement or simulation. After importing them into the modeling software, select the appropriate method for data preprocessing;

[0025] The CG-BPNN model prediction in step 2 is to randomly divide the data into a training set and a test set according to a certain ratio, use a function to build a CG-BPNN model, and set the model's input layer number, hidden layer number and number, output layer number, maximum number of iterations, optimal learning rate, optimal error tolerance and other hyperparameters, and then start training the CG-BPNN model. If each hyperparameter reaches the preset requirements or the number of iterations reaches the maximum value, the training ends, and then the test set is used to verify the trained model;

[0026] The model is determined in step 3. If the mean square error (MSE) between the expected value and the predicted value is greater than 0.05, return to step 2 to adjust the hyperparameters and retrain the CG-BPNN model until the requirements are met. When the mean square error (MSE) is less than 0.05, the model is determined, and the RF power amplifier behavior modeling is completed.

[0027] In step 3, the mean square error (MSE) is calculated as follows:

[0028]

[0029] Among them, MSE is the result of mean square error, n is the number of prediction samples, P i is the i-th predicted value, T i is the corresponding i-th expected value.

[0030] Embodiments of the present invention.

[0031] Based on the performance parameters of a 2.2-3 GHz high-efficiency broadband gallium nitride high electron mobility transistor power amplifier (GaN HEMTPA) and a 2.2-2.4 GHz high-efficiency broadband GaN Doherty PA, a RF power amplifier behavior modeling method based on CG-BPNN technology proposed in the present invention is described in detail.

[0032] Step 1: Data preparation: [f, Pin] is used as input variables, and the frequency f is set to 2.2GHz-2.4GHz with a step size of 0.1GHz; Pin is set to 0dBm-40dBm with a step size of 0.1dBm. The behavior parameters composed of [Pout, PAE, Gain] are used as output variables. Finally, 1200 sets of performance parameters are extracted, which are imported into the modeling software and the minimum-maximum normalization method is selected for data preprocessing.

[0033] The CG-BPNN model prediction in step 2 is as follows: after randomly dividing the data into a training set and a test set in a ratio of 7:3, the CG-BPNN model is constructed using the newff() function, and the number of input layers is set to 2, the number of hidden layers and the initial value of the number are 2 and 10, the number of output layers is 3, the maximum number of iterations is 1000, the optimal learning rate is 0.01, and the optimal error tolerance is 0.001. Then the CG-BPNN model starts training. If each hyperparameter meets the preset requirements or the number of iterations reaches the maximum value, the training ends, and the trained model is verified using the test set;

[0034] The model of step three is determined as follows: if the mean square error MSE between the expected value and the predicted value is greater than 0.05, then return to the second step to adjust the hyperparameters and retrain the CG-BPNN model until the requirements are met. When the mean square error MSE is less than 0.05, the model is determined, that is, the CG-BPNN model predicts that the mean square error MSE of the GaNHEMTPA performance parameters is 0.035425, and the mean square error MSE of the GaNDohertyPA performance parameters is 0.020093, and the RF power amplifier behavior modeling is completed at this time. It can be seen that this method accurately characterizes the performance parameters of GaNHEMTPA and GaNDohertyPA.

[0035] In summary, the RF power amplifier performance parameter modeling method based on CG-BPNN technology provides an important reference for the behavioral modeling research and circuit design of RF power amplifiers.

[0036] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A RF power amplifier behavior modeling method based on CG-BPNN technology, including three steps: data preparation, CG-BPNN model prediction and model determination. The data preparation in step 1 is to prepare the characteristic parameters X that affect the performance of the RF power amplifier = [X1, X2, ..., X n ] as input variables, and set the step size of each input variable. The behavior parameters composed of [Pout, PAE, Gain] are used as output variables, and multiple groups of performance indicators are obtained based on measurement or simulation. After importing them into the modeling software, select the appropriate method for data preprocessing; The CG-BPNN model prediction in step 2 is to randomly divide the data into a training set and a test set according to a certain ratio, use a function to build a CG-BPNN model, and set the model's input layer number, hidden layer number and number, output layer number, maximum number of iterations, optimal learning rate, optimal error tolerance and other hyperparameters, and then start training the CG-BPNN model. If each hyperparameter reaches the preset requirements or the number of iterations reaches the maximum value, the training ends, and then the test set is used to verify the trained model; The model is determined in step 3. If the mean square error (MSE) between the expected value and the predicted value is greater than 0.05, return to step 2 to adjust the hyperparameters and retrain the CG-BPNN model until the requirements are met. When the mean square error (MSE) is less than 0.05, the model is determined, and the RF power amplifier behavior modeling is completed.

2. The RF power amplifier behavior modeling method based on CG-BPNN technology according to claim 1 is characterized in that: The mean square error MSE of step 4 can be obtained as follows: Among them, MSE is the result of mean square error, n is the number of prediction samples, P i is the i-th predicted value, T i is the corresponding i-th expected value.

Images