CN-121615579-B - Nonlinear modeling and digital predistortion method, system, medium and equipment for radio frequency power amplifier

CN121615579BCN 121615579 BCN121615579 BCN 121615579BCN-121615579-B

Abstract

The invention provides a method, a system, a medium and equipment for nonlinear modeling and digital predistortion of a radio frequency power amplifier, wherein the method comprises the following steps of obtaining a trained nonlinear model of the radio frequency power amplifier; the method comprises the steps of inputting a real-time excitation signal into a radio frequency power amplifier nonlinear model to obtain a corresponding real-time phase response and a real-time group delay response, generating a phase adjustment amount and a group delay adjustment amount based on the real-time phase response and the real-time group delay response, and superposing the phase adjustment amount and the group delay adjustment amount on the real-time excitation signal to eliminate nonlinear distortion of the real-time excitation signal passing through the radio frequency power amplifier. According to the method, the system, the medium and the equipment for nonlinear modeling and digital predistortion of the radio frequency power amplifier, which are disclosed by the invention, the nonlinear modeling and digital predistortion of the radio frequency power amplifier are realized based on the deep learning model of CNN and BiLSTM, and the linearity and the signal quality of a radio frequency transmitting link are effectively improved.

Inventors

ZHENG HANBAI
WU LIANG
QIAN RONG
ZHU HUI

Assignees

中国科学院上海微系统与信息技术研究所

Dates

Publication Date: 20260508
Application Date: 20251121

Claims (6)

1. A method for nonlinear modeling and digital predistortion of a radio frequency power amplifier, the method comprising the steps of: Acquiring a trained radio frequency power amplifier nonlinear model, wherein the radio frequency power amplifier nonlinear model is used for outputting phase response and group delay response of an excitation signal after the radio frequency power amplifier; inputting a real-time excitation signal into the radio frequency power amplifier nonlinear model to obtain a corresponding real-time phase response and a real-time group delay response; generating a phase adjustment amount and a group delay adjustment amount based on the real-time phase response and the real-time group delay response; superposing the phase adjustment amount and the group delay adjustment amount on the real-time excitation signal to eliminate nonlinear distortion of the real-time excitation signal passing through the radio frequency power amplifier; training the radio frequency power amplifier nonlinear model; training the radio frequency power amplifier nonlinear model comprises the following steps: The method comprises the steps of constructing a radio frequency power amplifier nonlinear model, wherein the radio frequency power amplifier nonlinear model comprises a convolutional neural network and a two-way long-short-term memory network which are connected, the convolutional neural network is used for extracting local modulation characteristics and a non-steady change mode of an excitation signal, and the two-way long-term memory network is used for capturing long-term memory effect and dynamic nonlinear effect of the excitation signal; Maintaining the two-way long-short-term memory network unchanged, and training the convolutional neural network; Maintaining the convolutional neural network unchanged, and training the two-way long-short-term memory network; Fine tuning the radio frequency power amplifier nonlinear model to obtain a trained radio frequency power amplifier nonlinear model; generating a phase adjustment amount and a group delay adjustment amount based on the real-time phase response and the real-time group delay response comprises the steps of: Acquiring phase response opposite values and group delay response opposite values corresponding to the real-time phase response and the real-time group delay response; and respectively taking the phase response opposite value and the group delay response opposite value as the phase adjustment amount and the group delay adjustment amount.
2. The method of modeling and digital predistortion of a radio frequency power amplifier according to claim 1, wherein said phase response and said group delay response are a phase shift and a group delay shift, respectively, of said excitation signal relative to an ideal linearity after said excitation signal has passed through said radio frequency power amplifier.
3. The system is characterized by comprising a first acquisition module, a second acquisition module, a generation module and an elimination module; The first acquisition module is used for acquiring a trained radio frequency power amplifier nonlinear model, and the radio frequency power amplifier nonlinear model is used for outputting phase response and group delay response of an excitation signal after the radio frequency power amplifier; the second acquisition module is used for inputting a real-time excitation signal into the radio frequency power amplifier nonlinear model to acquire a corresponding real-time phase response and a real-time group delay response; the generating module is used for generating a phase adjustment amount and a group delay adjustment amount based on the real-time phase response and the real-time group delay response; the elimination module is used for superposing the phase adjustment amount and the group delay adjustment amount on the real-time excitation signal so as to eliminate nonlinear distortion of the real-time excitation signal passing through the radio frequency power amplifier; The system further comprises a training module, wherein the training module is used for training the radio frequency power amplifier nonlinear model; training the radio frequency power amplifier nonlinear model comprises the following steps: The method comprises the steps of constructing a radio frequency power amplifier nonlinear model, wherein the radio frequency power amplifier nonlinear model comprises a convolutional neural network and a two-way long-short-term memory network which are connected, the convolutional neural network is used for extracting local modulation characteristics and a non-steady change mode of an excitation signal, and the two-way long-term memory network is used for capturing long-term memory effect and dynamic nonlinear effect of the excitation signal; Maintaining the two-way long-short-term memory network unchanged, and training the convolutional neural network; Maintaining the convolutional neural network unchanged, and training the two-way long-short-term memory network; Fine tuning the radio frequency power amplifier nonlinear model to obtain a trained radio frequency power amplifier nonlinear model; the generating module generating a phase adjustment amount and a group delay adjustment amount based on the real-time phase response and the real-time group delay response includes the steps of: Acquiring phase response opposite values and group delay response opposite values corresponding to the real-time phase response and the real-time group delay response; and respectively taking the phase response opposite value and the group delay response opposite value as the phase adjustment amount and the group delay adjustment amount.
4. The system of claim 3, wherein the phase response and the group delay response are a phase shift and a group delay shift, respectively, of the excitation signal after passing through the RF power amplifier relative to an ideal linearity.
5. An electronic device, comprising a processor and a memory; the memory is used for storing a computer program; The processor is configured to execute the computer program stored in the memory, so that the electronic device performs the method for nonlinear modeling and digital predistortion of a radio frequency power amplifier according to any one of claims 1 to 2.
6. A computer readable storage medium having stored thereon a computer program, characterized in that the program when executed by an electronic device implements the method of nonlinear modeling and digital predistortion of a radio frequency power amplifier according to any one of claims 1 to 2.

Description

Nonlinear modeling and digital predistortion method, system, medium and equipment for radio frequency power amplifier Technical Field The invention belongs to the technical field of radio frequency circuit modeling and wireless communication signal processing, and particularly relates to a method, a system, a medium and equipment for radio frequency Amplifier (PA) nonlinear modeling and digital predistortion (DIGITAL PRE-Distortion). Background As wireless communication advances into the 6G era, the nonlinear distortion and memory effect of the radio frequency power amplifier under the condition of a high-frequency broadband signal have increasingly significant influence on the communication performance. Particularly in millimeter wave and sub-millimeter wave frequency bands, if time delay and phase offset introduced by the PA are not compensated, spectrum spreading and waveform distortion are caused, so that communication quality and positioning accuracy are affected. The existing PA modeling and predistortion method mainly comprises the following steps: (1) The nonlinear behavior model based on the Volterra series or the memory polynomial has good fitting effect under the condition of low bandwidth, but is limited by parameter redundancy and strong correlation under the condition of high bandwidth, the fitting precision is reduced, and the strong nonlinear and deep memory effect is hard to characterize; (2) Modeling methods based on neural networks, such as feedforward neural networks, long Short-Term Memory networks (LSTM), convolutional neural networks (Convolutional Neural Network, CNN), or the like. Although the method has modeling capability to a certain extent, the method has the problems of high network complexity, unstable training or insufficient memory modeling capability. Therefore, how to construct an efficient PA model that combines local feature extraction with long-term memory modeling is a key challenge in current depth predistortion techniques. Disclosure of Invention In view of the above-mentioned drawbacks of the prior art, the present invention aims to provide a method, a system, a medium, and a device for nonlinear modeling and digital predistortion of a radio frequency power amplifier, which are based on a deep learning model of a CNN and a two-way long and short Term Memory network (Bidirectional Long Short-Term Memory, biLSTM), so as to effectively improve the linearity and signal quality of a radio frequency transmitting link. The invention provides a radio frequency power amplifier nonlinear modeling and digital predistortion method, which comprises the following steps of obtaining a trained radio frequency power amplifier nonlinear model, inputting a real-time excitation signal into the radio frequency power amplifier nonlinear model to obtain a corresponding real-time phase response and a corresponding real-time group delay response, generating a phase adjustment amount and a group delay adjustment amount based on the real-time phase response and the real-time group delay response, and superposing the phase adjustment amount and the group delay adjustment amount on the real-time excitation signal to eliminate nonlinear distortion of the real-time excitation signal passing through the radio frequency power amplifier. In an implementation manner of the first aspect, training the radio frequency power amplifier nonlinear model is further included; training the radio frequency power amplifier nonlinear model comprises the following steps: The method comprises the steps of constructing a radio frequency power amplifier nonlinear model, wherein the radio frequency power amplifier nonlinear model comprises a convolutional neural network and a two-way long-short-term memory network which are connected, the convolutional neural network is used for extracting local modulation characteristics and a non-steady change mode of an excitation signal, and the two-way long-term memory network is used for capturing long-term memory effect and dynamic nonlinear effect of the excitation signal; Maintaining the two-way long-short-term memory network unchanged, and training the convolutional neural network; Maintaining the convolutional neural network unchanged, and training the two-way long-short-term memory network; and fine tuning the radio frequency power amplifier nonlinear model to obtain a trained radio frequency power amplifier nonlinear model. In an implementation manner of the first aspect, the phase response and the group delay response are a phase offset and a group delay offset of the excitation signal relative to ideal linearity after passing through the radio frequency power amplifier, respectively. In one implementation of the first aspect, generating the phase adjustment amount and the group delay adjustment amount based on the real-time phase response and the real-time group delay response comprises the steps of: Acquiring phase response opposite values and group delay response oppo