CN-122026826-A - Digital predistortion method and system for multi-power working condition unified power amplifier

CN122026826ACN 122026826 ACN122026826 ACN 122026826ACN-122026826-A

Abstract

The invention relates to the technical field of digital information transmission, in particular to a digital predistortion method and a digital predistortion system of a multi-power working condition unified power amplifier, wherein the method comprises the steps of collecting input and output signal sequences of the power amplifier under the multi-power working condition, constructing an original training data set, and preprocessing to obtain a standardized training data set; the method comprises the steps of carrying out feature engineering processing on an input signal sequence in a standardized training data set to construct an enhanced feature vector, training a power amplifier inverse model through a heterogeneous gate control circulation unit network, fusing the enhanced feature vector at an input layer to obtain a trained power amplifier inverse model, assigning parameters of the power amplifier inverse model to a digital predistorter, and realizing unified linearization output under a multi-power working condition. The invention maintains the independent distribution characteristic of each power group while realizing cross-power scale unification, and provides an efficient and unified solution for multi-working condition linearization of the power amplifier.

Inventors

ZHANG XUEBO
ZHOU XUDONG
WANG MENGMENG
WANG RUNHUA

Assignees

南开大学

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1.A digital predistortion method of a multi-power working condition unified power amplifier is characterized by comprising the following steps: S1, acquiring an input signal sequence and an output signal sequence of a power amplifier under a multi-power working condition, constructing an original training data set, preprocessing the input signal sequence and the output signal sequence in the original training data set according to power class characteristics, taking the input signal sequence in the preprocessed original training data set as an output signal sequence in a standardized training data set, and taking the output signal sequence in the preprocessed original training data set as an input signal sequence in the standardized training data set; S2, carrying out feature engineering processing on an input signal sequence in the standardized training data set, calculating signal amplitude of each sampling point and signal characteristics of cubic terms, quintic Fang Xiang and heptatic terms of the signal amplitude, directly splicing the calculated signal characteristics, power grade characteristics and inphase components and orthogonal components of the input signal sequence in the standardized training data set in a channel dimension, and constructing an enhanced feature vector fused with power heterogeneous information; S3, training a power amplifier inverse model through a heterogeneous gate control circulation unit network, and fusing the enhanced feature vectors fused with the power heterogeneous information on an input layer by adopting a direct input strategy to obtain a trained power amplifier inverse model; And S4, assigning the parameters of the trained inverse model of the power amplifier to the digital predistorter, automatically matching the preprocessing parameters of the step S1 according to the current power level characteristics during actual communication, preprocessing the input signals in communication, constructing the enhanced feature vector of the preprocessed signals according to the step S2, inputting the signals into the digital predistorter for digital predistortion, performing inverse standardization operation on the output of the digital predistorter, recovering the signal amplitude, and sending the signals into the power amplifier to realize unified linearization output under the multi-power working condition.
2. The method of claim 1, wherein the multiple power mode in step S1 includes three power level characteristics of-32 dBm, -36dBm, and-38 dBm.
3. The method of claim 1, wherein step S1 is performed on the input signal and the output signal in the original training data set by using a packet Z-Score normalization method according to the power class characteristics.
4. The method for digital predistortion of a multi-power condition unified power amplifier of claim 3 wherein said method for packet Z-Score normalization is as follows: S111, dividing an input signal sequence and an output signal sequence in an original training data set into a plurality of subsets according to power class characteristics; S112, respectively calculating the mean value and standard deviation of the input signal sequence and the output signal sequence in each subset; S113, based on the mean value and standard deviation of the input signal sequence and the output signal sequence in each subset, carrying out standardization processing on the input signal sequence and the output signal sequence according to a formula (1) to obtain a standardized training data set: (1); Wherein: Which is indicative of the power class characteristics, The normalized power class is characterized by Is a subset of the input signal sequence of (a), Characterizing power class as Is a subset of the input signal sequence of (a), Characterizing power class as Is a mean value of the subset of the input signal sequences, Characterizing power class as Is a standard deviation of the subset input signal sequence, The normalized power class is characterized by Is a subset of the output signal sequence of (c), Characterizing power class as Is a subset of the output signal sequence of (c), Characterizing power class as Is a mean value of the subset of the output signal sequences, Characterizing power class as The subset of (2) outputs the standard deviation of the signal sequence.
5. The digital predistortion method of a multi-power condition unified power amplifier of claim 1, wherein the method for performing feature engineering processing on the input signal in the standardized training data set in step S2 is as follows: S211, calculating signal amplitude according to a formula (2) for each complex sampling point in the input signal in the standardized training data set: (2); Wherein: Representing the input signal in the standardized training dataset, Representing the signal amplitude of the input signal in the standardized training dataset, Representing the in-phase component of the input signal sequence in the normalized training dataset, Representing orthogonal components of the input signal sequence in the standardized training dataset; s212, calculating a third power term, a fifth power Fang Xiang and a seventh power term of the signal amplitude; S213, directly splicing the signal amplitude of the input signal in the calculated standardized training data set, the cubic term, the penta-cubic Fang Xiang, the hepta-cubic term, the power grade characteristic and the inphase component and the orthogonal component of the input signal sequence in the standardized training data set in the channel dimension to construct the enhanced feature vector fused with the power heterogeneous information.
6. The method for digital predistortion of a multi-power condition unified power amplifier as set out in claim 1, wherein the enhanced eigenvector of the fused power heterogeneous information constructed in step S2 is , Wherein: Representing the input signal in the standardized training dataset, Representing the signal amplitude of the input signal in the standardized training dataset, Representing the in-phase component of the input signal sequence in the normalized training dataset, Representing orthogonal components of the input signal sequence in the normalized training dataset, A third order term representing the signal amplitude of the input signal in the normalized training dataset, A fifth order term representing the signal amplitude of the input signal in the normalized training dataset, A seventh-order term representing the signal amplitude of the input signal in the normalized training dataset, Representing a power class characteristic.
7. The method of claim 1, wherein in step S3, when the inverse model of the power amplifier is trained by the heterogeneous gating loop network, the output signal and the corresponding power class characteristic in the standardized training data set are taken as inputs, the input signal in the standardized training data set is taken as a supervision tag, and the cross-power mapping relation from output to input is learned by minimizing the mean square error.
8. The digital predistortion method of a multi-power condition unified power amplifier of claim 1, wherein in step S4, the preprocessing parameters of step S1 are automatically matched according to the current power class characteristics during actual communication, and the input signals in communication are preprocessed by using the formula (3): (3); Wherein: Representing the current power class characteristics of the pre-processed input signal, Representing an input signal in a current power class feature communication, Which is representative of the current power level characteristics, Representing the mean value corresponding to the current power level characteristic, Representing the standard deviation corresponding to the current power class characteristics.
9. The method for digital predistortion of a multi-power condition unified power amplifier of claim 1 wherein in step S4 the output of the digital predistorter is inverse normalized according to equation (4): (4); Wherein: The output of the digital predistorter is inversely normalized to represent the current power class characteristics, The output of the digital predistorter is characteristic of the current power level, Representing the standard deviation corresponding to the current power class characteristics, Representing the mean value corresponding to the current power class characteristic.
10. A digital predistortion system of a multi-power condition unified power amplifier for executing the digital predistortion method of the multi-power condition unified power amplifier according to any one of claims 1 to 9, which is characterized by comprising a power amplifier, a data acquisition module, a preprocessing module, a characteristic engineering processing module, a heterogeneous gate control circulation unit network and a digital predistorter; The data acquisition module is used for acquiring an input signal sequence and an output signal sequence of the power amplifier under the condition of multiple power working conditions and constructing an original training data set; the preprocessing module is used for preprocessing an input signal sequence and an output signal sequence in the original training data set according to the power grade characteristics to construct a standardized training data set; The characteristic engineering processing module is used for carrying out characteristic engineering processing on an input signal sequence in the standardized training data set and constructing an enhanced characteristic vector fused with power heterogeneous information; the heterogeneous gating circulation unit network is used for training the inverse power amplifier model to obtain the trained inverse power amplifier model; The digital predistorter is used for carrying out digital predistortion processing on the constructed enhanced feature vector in actual communication.

Description

Digital predistortion method and system for multi-power working condition unified power amplifier Technical Field The invention relates to the technical field of digital information transmission, in particular to a digital predistortion method and a digital predistortion system for a multi-power working condition unified power amplifier. Background With the evolution of modern wireless communication to an intelligent and anti-interference direction, frequency hopping communication has become a core technical means for improving confidentiality and reliability of communication. In a frequency hopping communication system, a power amplifier is required to realize fast frequency switching within a predetermined frequency band, and also to dynamically adjust a transmission level between multiple power levels according to a channel state and a link budget. The multi-power jump scene provides a serious challenge for linearization of the power amplifier, namely under different power class characteristics, the working bias point, junction temperature characteristic and impedance matching state of the power transistor are obviously different, so that the nonlinear distortion curve of the power amplifier is drifted. When power jump is encountered, the pre-correction coefficient of the traditional digital pre-distortion technology optimized for single fixed power is mismatched with the actual distortion characteristic, the adjacent channel leakage ratio is rapidly deteriorated, the error vector amplitude is deteriorated by more than 3%, and the communication quality is seriously damaged. The current multi-working condition digital predistortion technology mainly develops along three technical paths, the first type is a coefficient storage method, predistortion coefficients of all power class characteristics are extracted offline and stored in a lookup table, and corresponding coefficients are called online according to the power class characteristics. The method is simple to realize, but with the subdivision of the power level characteristics and the improvement of the memory effect order, the storage capacity is exponentially increased, the resource consumption is overlarge, and the coefficient switching has millisecond delay, so that the real-time requirement of frequency hopping communication is difficult to meet. The second type is a model reconstruction method, a generalized power amplifier model is built by adopting a polynomial or deep neural network, and power parameters are embedded as input variables to realize multi-working-condition coverage. However, the polynomial model needs to introduce a higher-order cross term when processing strong memory effect, the calculation complexity is high, the existing deep neural network mostly adopts a full-connection architecture, and although power is input as a numerical characteristic, the isomerism of a power parameter and a baseband signal on physical meaning and dimension scale is not fully considered, the model is difficult to effectively learn the internal association of distortion rules under different powers, and the cross-power generalization capability is limited. The third type is a self-adaptive adjustment method, which is used for rapidly adapting to working condition changes through online parameter updating or meta-learning, but online learning relies on real-time feedback, so that the convergence rate is low, and stability risks exist in a rapid frequency hopping scene. Disclosure of Invention The invention aims to solve the technical problem of providing a digital predistortion method and a digital predistortion system for a multi-power working condition unified power amplifier, which are capable of realizing cross-power scale unification while preserving the independent distribution characteristics of each power group, avoiding dimension explosion and structural redundancy and providing a high-efficiency unified solution for multi-working condition linearization of the power amplifier in frequency hopping communication. A digital predistortion method of a multi-power working condition unified power amplifier comprises the following steps: S1, acquiring an input signal sequence and an output signal sequence of a power amplifier under a multi-power working condition, constructing an original training data set, preprocessing the input signal sequence and the output signal sequence in the original training data set according to power class characteristics, taking the input signal sequence in the preprocessed original training data set as an output signal sequence in a standardized training data set, and taking the output signal sequence in the preprocessed original training data set as an input signal sequence in the standardized training data set; S2, carrying out feature engineering processing on an input signal sequence in the standardized training data set, calculating signal amplitude of each sampling point and signal characteristics of cubic terms,