A deep learning driven power amplifier predistortion method

Abstract

The application belongs to the technical field of signal processing, and specifically discloses a power amplifier pre-distortion method driven by deep learning, which comprises the following steps: constructing a signal matrix of a target signal sequence, the target signal sequence being obtained by preprocessing an original time signal sequence collected from a signal source; inputting the signal matrix into a PDTRANS pre-distortion model and a power amplifier in sequence to obtain an output signal sequence of the power amplifier; determining a loss function value based on the output signal sequence and the target signal sequence, and adjusting hyperparameters of the PDTRANS pre-distortion model based on the loss function value; repeating the steps of obtaining the output signal sequence of the power amplifier and adjusting the hyperparameters of the PDTRANS pre-distortion model until the loss function is minimized, so that a trained PDTRANS pre-distortion model matched with the power amplifier is obtained; and performing pre-distortion compensation on the power amplifier based on the trained PDTRANS pre-distortion model. The application realizes accurate pre-distortion modeling of the power amplifier.

Description

Technical Field

[0001] This application belongs to the field of signal processing technology, and more specifically, relates to a deep learning-driven power amplifier predistortion method. Background Technology

[0002] A power amplifier (PA) amplifies the power of an input signal to enhance its strength and coverage. The nonlinear characteristics of a PA cause nonlinear distortion, and the memory effect further degrades its linearity, reducing the quality of the output signal sequence. By constructing an accurate mathematical model using behavioral modeling and digital predistortion techniques to track the PA's nonlinear characteristics and memory effect, and then correcting the signal with a digital predistorter, output distortion can be effectively reduced, and signal quality improved. Traditional digital predistortion methods are mostly based on polynomial models or lookup table models, which often fail to achieve ideal linearization when dealing with PAs exhibiting strong nonlinearity and memory effects.

[0003] In recent years, neural networks have been widely used in various fields due to their powerful nonlinear fitting and global approximation capabilities. One related technique proposes a radio frequency (RF) channel modeling and predistortion method based on Extended Real-Valued Time-Delayed Neural Network (RVTDNN). This method captures the nonlinear characteristics and memory effects in the RF channel through deep neural networks, demonstrating certain modeling accuracy and robustness. Furthermore, a modeling method based on Graph Convolutional Network (GCN) is proposed. This method transforms one-dimensional time series sequences into two-dimensional graph structures for processing, fully utilizing the temporal correlation and local features of the signal, significantly improving the accuracy and performance of the predistorter modeling, and providing a new approach for modeling complex nonlinear systems.

[0004] Although various neural networks have shown certain advantages in power amplifier modeling and predistortion processing, low-complexity neural networks often struggle to achieve ideal results when dealing with complex and high-dimensional data. Summary of the Invention

[0005] To address the aforementioned shortcomings of existing technologies, this application provides a deep learning-driven power amplifier predistortion method, aiming to solve the problem of poor predistortion processing performance of power amplifiers based on neural networks when dealing with complex and high-dimensional data processing.

[0006] Firstly, this application provides a deep learning-driven power amplifier predistortion method, including:

[0007] A signal matrix for the target signal sequence is constructed. The target signal sequence is obtained by preprocessing the original time signal sequence acquired from the signal source.

[0008] The signal matrix is ​​sequentially input into the PDTRANS predistortion model and the power amplifier to obtain the output signal sequence of the power amplifier.

[0009] The loss function value is determined based on the output signal sequence and the target signal sequence, and the hyperparameters of the PDTRANS predistortion model are adjusted based on the loss function value.

[0010] Repeat the steps of acquiring the output signal sequence of the power amplifier and adjusting the hyperparameters of the PDTRANS predistortion model until the loss function is minimized, and obtain a well-trained PDTRANS predistortion model that matches the power amplifier.

[0011] Predistortion compensation for the power amplifier is performed based on the trained PDTRANS predistortion model.

[0012] Secondly, this application also provides a deep learning-driven power amplifier predistortion device, comprising:

[0013] The matrix construction module is used to construct the signal matrix of the target signal sequence, which is obtained by preprocessing the original time signal sequence acquired from the signal source.

[0014] The model training module is used to input the signal matrix sequentially into the PDTRANS predistortion model and the power amplifier to obtain the output signal sequence of the power amplifier; determine the loss function value based on the output signal sequence and the target signal sequence; adjust the hyperparameters of the PDTRANS predistortion model based on the loss function value; repeat the steps of obtaining the output signal sequence of the power amplifier and adjusting the hyperparameters of the PDTRANS predistortion model until the loss function is minimized, and obtain a trained PDTRANS predistortion model that matches the power amplifier.

[0015] The predistortion compensation module is used to perform predistortion compensation on the power amplifier based on the trained PDTRANS predistortion model.

[0016] Thirdly, this application also provides an electronic device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the method described in the first aspect or any possible implementation thereof.

[0017] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0018] Fifthly, this application also provides a computer program product that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0019] This application provides a deep learning-driven power amplifier predistortion method that converts a one-dimensional time-series signal into a two-dimensional signal matrix as model input, enabling the model to capture the dynamic changes of the time signal sequence and interpret the complex interactions within the signal. It utilizes a Transformer-based predistortion model for power amplifier predistortion modeling, employing a self-attention mechanism to independently analyze the input signal and calculate the correlations between different features, thereby deeply exploring the complex dynamic behavior in the signal and more accurately modeling the nonlinear characteristics and memory effects of the power amplifier. The model training adopts a direct learning approach, which can quickly adapt to the actual signal environment and shorten the power amplifier predistortion modeling process. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is one of the flowcharts of the deep learning-driven power amplifier predistortion method provided in the embodiments of this application;

[0022] Figure 2 This is the second flowchart of the deep learning-driven power amplifier predistortion method provided in the embodiments of this application;

[0023] Figure 3 This is a schematic diagram of the process for training a PDTRANS predistortion model based on a direct learning method, provided in an embodiment of this application.

[0024] Figure 4 This is a schematic diagram of the architecture of the PDTRANS predistortion model provided in the embodiments of this application;

[0025] Figure 5 This is a schematic diagram of the structure of the deep learning-driven power amplifier predistortion device provided in the embodiments of this application;

[0026] Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0028] When faced with complex and high-dimensional data processing, the Transformer model has shown significant advantages in processing sequential data, self-attention mechanism and parallel computing. Therefore, this application considers introducing the Transformer into power amplifier pre-distortion modeling and proposes a Transformer-based pre-distortion model (PDTRANS).

[0029] Figure 1 This is one of the flowcharts illustrating the deep learning-driven power amplifier predistortion method provided in this application embodiment. Figure 2 This is the second flowchart illustrating the deep learning-driven power amplifier predistortion method provided in this application embodiment, combined with... Figure 1 and Figure 2 The method includes at least the following steps:

[0030] S101. Construct the signal matrix of the target signal sequence. The target signal sequence is obtained by preprocessing the original time signal sequence collected from the signal source.

[0031] S102. Input the signal matrix into the PDTRANS predistortion model and the power amplifier in sequence to obtain the output signal sequence of the power amplifier.

[0032] S103. Determine the loss function value based on the output signal sequence and the target signal sequence, and adjust the hyperparameters of the PDTRANS predistortion model based on the loss function value.

[0033] S104. Repeat the steps of acquiring the output signal sequence of the power amplifier and adjusting the parameters of the PDTRANS predistortion model until the loss function is minimized, and obtain a well-trained PDTRANS predistortion model that matches the power amplifier.

[0034] S105. Perform predistortion compensation on the power amplifier based on the trained PDTRANS predistortion model.

[0035] The deep learning-driven power amplifier predistortion method provided in this application converts a one-dimensional time-series signal into a two-dimensional signal matrix as model input, enabling the model to capture the dynamic changes of the time signal sequence and interpret the complex interactions within the signal. It utilizes a Transformer-based predistortion model for power amplifier predistortion modeling, independently analyzing the input signal through a self-attention mechanism and calculating the correlation between different features. This allows for in-depth analysis of the complex dynamic behavior in the signal, more accurately modeling the nonlinear characteristics and memory effects of the power amplifier. The model training employs a direct learning approach, enabling rapid adaptation to the actual signal environment and shortening the power amplifier predistortion modeling process.

[0036] For S101, construct the signal matrix of the target signal sequence as the input to the subsequent power amplifier predistortion model.

[0037] The original time signal sequence is preprocessed to obtain the target signal sequence. The time signal sequence, i.e., the signal arranged in a time-ordered manner, provides the foundation for subsequently capturing the memory effect and nonlinear characteristics of the power amplifier. The main purpose of preprocessing is to remove unnecessary noise and interference, ensuring the validity and stability of the signal, and providing a high-quality data foundation for subsequent power amplifier predistortion modeling and applications.

[0038] Optionally, preprocessing methods include filtering, for example... Figure 2 The diagram illustrates filtering the original time signal sequence using a bandpass filter. Optionally, the preprocessing method also includes normalization to ensure that all sample signals in the constructed signal matrix have the same scale, and to ensure that the signal amplitude is within a dynamic range suitable for the subsequent predistortion model.

[0039] Optionally, the original time signal sequence is sampled using a sliding window method, where the sequence length is the sliding window length. Sliding window sampling can convert long time series signals into short sequences, which facilitates subsequent analysis, feature extraction, and model training, enabling the model to be trained quickly under conditions of small datasets and limited hardware resources.

[0040] Furthermore, a signal matrix of the target signal sequence is constructed. Since time-series signals are one-dimensional data, constructing the signal matrix of the target signal sequence transforms the one-dimensional time-series signal into a two-dimensional format, converting the time-series signal into a form that is easy to process and analyze. This allows subsequent predistortion models to capture the dynamic changes of the time-series signal and reveal the complex interactions within the signal.

[0041] Optionally, S101 specifically includes:

[0042] Extract the I / Q component features and amplitude envelope features of the sample signal at different times in the target signal sequence;

[0043] A signal matrix of the target signal sequence is constructed using the number of extracted features and the sequence length as dimensions.

[0044] Specifically, constructing a signal matrix for the target signal sequence transforms the sequence into a form that is easy to process and analyze. The dimensions of the signal matrix are the sequence length and the number of features. Features are pre-extracted, such as the frequency, phase, and amplitude of the signal. Specifically, this can be the I / Q component features and amplitude envelope features of the sample signal at different times.

[0045] The I / Q components refer to the signal's behavior in terms of its in-phase and quadrature components. By decomposing the signal into complex form in a Cartesian coordinate system, the real part is the I component, and the imaginary part is the Q component; in polar coordinates, they can be represented by amplitude and phase. The amplitude envelope is the curve of the signal amplitude changing over time, reflecting the signal's trend and fluctuation range.

[0046] For example, constructing the signal matrix of the target signal sequence. as follows:

[0047]

[0048] in, This represents the I component of the sample signal at time n. This represents the Q component of the sample signal at time n. This represents the amplitude envelope of the sample signal at time n. Indicates memory depth, The length of the target signal sequence, which is also the number of sample signals.

[0049] For S102 to S104, a direct learning method is used for power amplifier predistortion modeling and training.

[0050] Figure 3 This is a schematic diagram of the process for training a PDTRANS predistortion model based on a direct learning method, as provided in the embodiments of this application. Figure 3 As shown, at time n, the input to the PDTRANS predistortion model is the preprocessed target signal. The output is a predistorted signal. The predistortion signal output by the PDTRANS predistortion model. Predistortion compensation is performed on the input power amplifier to obtain the output signal of the power amplifier. .

[0051] The loss function value at time n is calculated using the output signal of the power amplifier and the target signal input to the PDTRANS predistortion model. The loss function characterizes the difference between the output signal sequence of the power amplifier and the target signal sequence, and can specifically employ mean squared error (MSE), root mean squared error (RMSE), or mean absolute error (MAE). The hyperparameters of the PDTRANS predistortion model are then adjusted using the loss function value.

[0052] The hyperparameters of the PDTRANS predistortion model are continuously adjusted using the loss function value until the loss function is minimized (e.g., the loss function value is 0 or reaches a preset threshold close to 0), thereby obtaining a well-trained PDTRANS predistortion model that matches the power amplifier. After the PDTRANS predistortion model converges, the input signal of the power amplifier should be as close as possible to the target signal input to the PDTRANS predistortion model, thereby achieving predistortion compensation for the power amplifier.

[0053] The direct learning mode is simple and efficient, and can quickly adapt to the actual signal environment, shortening the predistortion modeling process of the power amplifier.

[0054] Figure 4 This is a schematic diagram of the architecture of the PDTRANS predistortion model provided in the embodiments of this application, as shown below. Figure 4 As shown, the PDTRANS predistortion model includes:

[0055] (1) Input module

[0056] The input module includes convolutional (Conv) layers, designed to perform preliminary feature extraction on the input signal matrix. The input signal matrix undergoes a 3×3 convolution operation to extract local time-series features, including I / Q component features and amplitude envelope features. The convolution operation enhances the model's understanding of instantaneous characteristics and local patterns, laying the foundation for subsequent modules to capture the nonlinear features of the signal.

[0057] (2) Transformer module

[0058] The Transformer module is the core of the PDTRANS predistortion model. Through self-attention mechanism and parallel computing capabilities, it is responsible for modeling the complex nonlinear characteristics and memory effects of signals, adapting to diverse signal characteristics and environmental requirements.

[0059] The Transformer module specifically includes an improved normalization layer, a parallel convolutional layer, a self-attention layer, a short-circuit projection layer, and an output projection layer.

[0060] The Improved Normalization Layer (Revised Layer Norm) normalizes the input features (extracted features from the input module output) to reduce the impact of signal amplitude variations on model performance and improve training stability.

[0061] Three parallel convolutional layers generate key features, query features, and value features based on the normalized input features, respectively:

[0062]

[0063] in, These represent query, key, and value characteristics, respectively. , , These are used to generate learnable weight matrices for the three features, respectively. This represents the signal matrix after normalization. Query features are used to calculate attention weights, key features are used together with query features to calculate attention weights, and value features are used to generate the weighted attention output.

[0064] The self-attention layer models the dynamic dependencies of signals using a self-attention mechanism. It determines the self-attention result based on key features, query features, and value features, satisfying the following conditions:

[0065]

[0066] in, Indicates the scaling factor. These represent query, key, and value features, respectively. The self-attention mechanism captures the complex relationships between different parts of the input signal by dynamically assigning weights.

[0067] The short-circuit projection layer extracts the depth features of the value features through depthwise convolution, adds them to the self-attention results of the self-attention layer, and then performs convolutional layer operations and normalization to further enhance the model's feature extraction capabilities.

[0068] The output projection layer comprises two merging layers and a deep convolutional layer. The first merging layer aggregates the outputs of the short-circuit projection layer and the self-attention layer, outputting the aggregated result, emphasizing the combination of attention features and local features. The aggregated result from the first merging layer is then passed through a deep convolutional layer, followed by a second merging layer for further deep feature aggregation, outputting a comprehensive feature and enhancing the expressive power of the final feature. This staged feature fusion strategy enables the model to more comprehensively model the dynamic characteristics and nonlinear behavior of complex signals, while balancing global and local features.

[0069] (3) Feature Dimensionality Reduction Module

[0070] The high dimensionality of features processed by the Transformer module is detrimental to subsequent processing. Therefore, a feature dimensionality reduction module is used to reduce feature dimensionality. This module includes fully connected layers; optionally, two fully connected layers are used to further improve the model's nonlinear modeling capability and the expressive power of the feature maps. The two fully connected layers use a nonlinear activation function to increase the model's ability to express the nonlinear characteristics of the signal while reducing subsequent computational complexity. Optionally, the nonlinear activation function is the Tanh activation function.

[0071] Fully connected layers perform feature dimensionality reduction through a non-linear transformation of the Tanh activation function, specifically satisfying:

[0072]

[0073] in, This represents the feature matrix after dimensionality reduction. This represents the weight matrix of the fully connected layer. This represents the bias vector of the fully connected layer. This represents the overall characteristics output by the Transformer module.

[0074] (4) Output module

[0075] The output module further processes the dimensionality-reduced feature matrix and outputs the final predistortion signal, specifically satisfying the following:

[0076]

[0077] in, Indicates the predistortion signal. This represents the weight matrix of the output layer. This represents the bias vector of the output layer. This represents the feature matrix after dimensionality reduction.

[0078] The predistortion signal output by the model is opposite to the nonlinear characteristics of the power amplifier. When input to the power amplifier, it can compensate for the nonlinear distortion of the power amplifier and achieve linear amplification.

[0079] Optionally, in S103, the hyperparameters of the PDTRANS predistortion model are adjusted, specifically including:

[0080] Adjust the learning rate, batch size, and number of network layers of the PDTRANS predistortion model.

[0081] Specifically, the learning rate, batch size, and number of network layers directly affect the model's training efficiency, generalization ability, and computational complexity. Therefore, it is necessary to continuously adjust the hyperparameters using the loss function value to find a balance between improving model performance and consuming resources.

[0082] Different learning rates are used to balance the model's convergence speed and stability; batch size adjustment focuses on memory efficiency and the accuracy of model gradient estimation; network layer optimization focuses on the relationship between the model's expressive power and computational complexity. Optionally, adjusting the number of network layers refers to adjusting the number of fully connected layers in the output feature dimensionality reduction module.

[0083] The embodiments of this application, through hyperparameter tuning of the model, can more efficiently learn the nonlinear characteristics and memory effect of the power amplifier, and generate accurate predistortion signals to compensate for the distortion behavior of the power amplifier.

[0084] Optionally, the training of the PDTRANS predistortion model employs standardized procedures and optimization strategies to ensure the stability and reliability of model performance. Specifically, 100,000 sets of signal data were used for modeling during training, and the dataset was divided into training, validation, and test sets in a 3:1:1 ratio. All input signals were normalized before training to unify the data range to the same scale, thereby optimizing the convergence speed and performance of model training. The normalized data was then fed into the PDTRANS predistortion model for learning.

[0085] Optionally, training is performed using the PyTorch framework on a Graphics Processing Unit (GPU) environment to accelerate model computation. The specific hardware configuration includes three NVIDIA GTX 2080 GPUs and an Intel Core i7 CPU. The Adam optimizer is used during training, with an initial learning rate set to [missing value]. The learning rate is dynamically adjusted during training to ensure fast convergence and robust performance.

[0086] Optionally, the mean square error (MSE) is used to measure the difference between the output signal sequence of the power amplifier and the target signal sequence after predistortion compensation using the PDTRANS predistortion model. That is, the loss function of the PDTRANS predistortion model satisfies:

[0087]

[0088] in, Represents the loss function. This represents the nth output signal in the output signal sequence of the power amplifier after predistortion compensation using the PDTRANS predistortion model. This represents the nth target signal in the target signal sequence, where N represents the sequence length.

[0089] Optionally, the gradient is calculated using the backpropagation method, and the model parameters are iteratively optimized using a gradient descent strategy. The training process continues until the performance metrics on the validation set reach their optimum, ensuring that the model can accurately model the nonlinear characteristics of the power amplifier.

[0090] For S105, the predistortion compensation of the power amplifier is performed using the trained PDTRANS predistortion model. Different power amplifiers have different PDTRANS predistortion models obtained through direct learning.

[0091] A PDTRANS predistortion model is introduced at the front end of the power amplifier. The PDTRANS predistortion model is trained by direct learning until the model converges. Then, the target signal sequence containing the signal information at the current moment is input into the PDTRANS predistortion model to obtain the predistortion signal output by the model. This compensates for the nonlinear distortion generated by the power amplifier at the current moment, realizes the linearization of the power amplifier, thereby improving the adjacent channel power ratio of the power amplifier output signal and enhancing the linearity, anti-interference capability and overall performance of the communication system.

[0092] Optionally, S105 specifically includes:

[0093] The time series signals of the current time and the nearest historical time are collected from the signal source, and the input matrix is ​​constructed after preprocessing. The input matrix includes the I / Q components and amplitude envelope of the time series signals of the current time and the nearest historical time.

[0094] The input matrix is ​​fed into the trained PDTRANS predistortion model, and the predistortion signal at the current time is output.

[0095] The predistorted signal is input to the power amplifier for signal linearization compensation at the current moment.

[0096] The PDTRANS predistortion model provided in this application embodiment can achieve real-time, high-precision power amplifier predistortion compensation, significantly improving the linearity and anti-interference capability of the power amplifier, ensuring the stability and efficiency of the system in complex environments, and is suitable for complex power amplifier nonlinear environments. It can effectively cope with dynamically changing signals and operating conditions and has broad application prospects in satellite communication, adaptive modulation, spectrum management and other applications.

[0097] The deep learning-driven power amplifier predistortion device provided in this application is described below. The deep learning-driven power amplifier predistortion device described below can be referred to in correspondence with the deep learning-driven power amplifier predistortion method described above.

[0098] Figure 5 This is a schematic diagram of the structure of the deep learning-driven power amplifier predistortion device provided in the embodiments of this application, as shown below. Figure 5 As shown, the device includes at least:

[0099] The matrix construction module 501 is used to construct the signal matrix of the target signal sequence, which is obtained by preprocessing the original time signal sequence acquired from the signal source.

[0100] The model training module 502 is used to input the signal matrix sequentially into the PDTRANS predistortion model and the power amplifier to obtain the output signal sequence of the power amplifier; determine the loss function value based on the output signal sequence and the target signal sequence; adjust the hyperparameters of the PDTRANS predistortion model based on the loss function value; repeat the steps of obtaining the output signal sequence of the power amplifier and adjusting the hyperparameters of the PDTRANS predistortion model until the loss function is minimized, and obtain a trained PDTRANS predistortion model that matches the power amplifier.

[0101] The predistortion compensation module 503 is used to perform predistortion compensation on the power amplifier based on the trained PDTRANS predistortion model.

[0102] Optionally, the matrix construction module 501 is specifically used for:

[0103] Extract the I / Q component features and amplitude envelope features of the sample signal at different times in the target signal sequence;

[0104] A signal matrix of the target signal sequence is constructed using the number of extracted features and the sequence length as dimensions.

[0105] Optionally, the PDTRANS predistortion model includes an input module, a Transformer module, a feature dimensionality reduction module, and an output module;

[0106] The input module includes convolutional layers; the convolutional layers extract local time-series features of the signal matrix, including I / Q component features and amplitude envelope features;

[0107] The Transformer module includes an improved normalization layer, a parallel convolutional layer, a self-attention layer, a short-circuit projection layer, and an output projection layer. The improved normalization layer normalizes the input features. The parallel convolutional layer generates key features, query features, and value features based on the normalized input features. The self-attention layer determines the self-attention result based on the key features, query features, and value features. The short-circuit projection layer extracts the deep features of the value features, adds the deep features and the self-attention result, and then performs convolution and normalization operations. The output projection layer aggregates the output features of the short-circuit projection layer and the self-attention result, performs deep convolution on the aggregated result, and then aggregates again to output a comprehensive feature.

[0108] The feature dimensionality reduction module includes a fully connected layer, which performs feature dimensionality reduction through a non-linear transformation of the Tanh activation function;

[0109] The output module outputs a predistorted signal based on the dimension-reduced feature matrix.

[0110] Optionally, the model training module 502 is specifically used for:

[0111] The learning rate, batch size, and number of network layers of the PDTRANS predistortion model are adjusted based on the loss function value.

[0112] Optionally, the loss function of the PDTRANS predistortion model satisfies:

[0113]

[0114] in, Represents the loss function. This represents the nth output signal in the output signal sequence of the power amplifier after predistortion compensation using the PDTRANS predistortion model. This represents the nth target signal in the target signal sequence, where N represents the sequence length.

[0115] Optionally, the pre-distortion compensation module 503 is specifically used for:

[0116] The time series signals of the current time and the nearest historical time are collected from the signal source, and the input matrix is ​​constructed after preprocessing. The input matrix includes the I / Q components and amplitude envelope of the time series signals of the current time and the nearest historical time.

[0117] The input matrix is ​​fed into the trained PDTRANS predistortion model, and the predistortion signal at the current time is output.

[0118] The predistorted signal is input to the power amplifier for signal linearization compensation at the current moment.

[0119] It is understood that the detailed functional implementation of each of the above units / modules can be found in the description in the aforementioned method embodiments, and will not be repeated here.

[0120] It should be understood that the above-described device is used to execute the methods in the above embodiments. The implementation principle and technical effect of the corresponding program modules in the device are similar to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.

[0121] Based on the methods described in the above embodiments, this application provides an electronic device. The device may include at least one memory for storing a program and at least one processor for executing the program stored in the memory. When the program stored in the memory is executed, the processor performs the methods described in the above embodiments.

[0122] Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application, such as... Figure 6 As shown, the electronic device may include a processor 601, a communications interface 602, a memory 603, and a communication bus 604, wherein the processor 601, the communications interface 602, and the memory 603 communicate with each other via the communication bus 604. The processor 601 can call software instructions in the memory 603 to execute the methods described in the above embodiments.

[0123] Furthermore, the logical instructions in the aforementioned memory 603 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application.

[0124] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0125] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0126] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0127] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0128] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0129] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0130] Those skilled in the art will readily understand that the above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A deep learning-driven power amplifier predistortion method, characterized in that, include: A signal matrix is ​​constructed for the target signal sequence, which is obtained by preprocessing the original time signal sequence acquired from the signal source; The signal matrix used to construct the target signal sequence includes: Extract the I / Q component features and amplitude envelope features of the sample signal at different times in the target signal sequence; construct the signal matrix of the target signal sequence with the number of extracted features and the sequence length as dimensions; The signal matrix is ​​sequentially input into the PDTRANS predistortion model and the power amplifier to obtain the output signal sequence of the power amplifier. The PDTRANS predistortion model includes an input module, a Transformer module, a feature reduction module, and an output module. The input module includes a convolutional layer; the convolutional layer extracts local time-series features of the signal matrix, including I / Q component features and amplitude envelope features; The Transformer module includes an improved normalization layer, a parallel convolutional layer, a self-attention layer, a short-circuit projection layer, and an output projection layer. The improved normalization layer normalizes the input features. The parallel convolutional layer generates key features, query features, and value features based on the normalized input features. The self-attention layer determines the self-attention result based on the key features, query features, and value features. The short-circuit projection layer extracts the depth features of the value features, adds the depth features and the self-attention result, and then performs convolution and normalization operations. The output projection layer aggregates the output features of the short-circuit projection layer and the self-attention result, performs deep convolution on the aggregated result, and then aggregates again to output a comprehensive feature. The feature dimensionality reduction module includes a fully connected layer, which performs feature dimensionality reduction through a nonlinear transformation of the Tanh activation function; The output module outputs a predistorted signal based on the dimension-reduced feature matrix. The loss function value is determined based on the output signal sequence and the target signal sequence, and the hyperparameters of the PDTRANS predistortion model are adjusted based on the loss function value. The hyperparameters include the learning rate, batch size, and number of network layers. Repeat the steps of obtaining the output signal sequence of the power amplifier and adjusting the hyperparameters of the PDTRANS predistortion model until the loss function is minimized, and obtain a trained PDTRANS predistortion model that matches the power amplifier. Predistortion compensation is performed on the power amplifier based on the trained PDTRANS predistortion model. The predistortion compensation of the power amplifier based on the trained PDTRANS predistortion model includes: The time series signals of the current moment and the nearest historical moment are collected from the signal source, and an input matrix is ​​constructed after preprocessing. The input matrix includes the I / Q components and amplitude envelope of the time series signals of the current moment and the nearest historical moment. The input matrix is ​​input into the trained PDTRANS predistortion model, and the predistortion signal at the current time is output. The predistorted signal is input to the power amplifier for signal linearization compensation at the current moment.

2. The power amplifier predistortion method according to claim 1, characterized in that, The loss function of the PDTRANS predistortion model satisfies: in, Represents the loss function. This represents the nth output signal in the output signal sequence of the power amplifier after predistortion compensation using the PDTRANS predistortion model. This represents the nth target signal in the target signal sequence, where N represents the sequence length.

3. A power amplifier predistortion device for implementing the method of any one of claims 1-2, characterized in that, include: A matrix construction module is used to construct a signal matrix of a target signal sequence, which is obtained by preprocessing the original time signal sequence acquired from the signal source. The model training module is used to sequentially input the signal matrix into the PDTRANS predistortion model and the power amplifier to obtain the output signal sequence of the power amplifier; determine the loss function value based on the output signal sequence and the target signal sequence; adjust the hyperparameters of the PDTRANS predistortion model based on the loss function value; repeat the steps of obtaining the output signal sequence of the power amplifier and adjusting the hyperparameters of the PDTRANS predistortion model until the loss function is minimized, thereby obtaining a trained PDTRANS predistortion model that matches the power amplifier. The predistortion compensation module is used to perform predistortion compensation on the power amplifier based on the trained PDTRANS predistortion model.

4. An electronic device, characterized in that, include: At least one memory for storing computer programs; At least one processor is configured to execute a program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to perform the method as described in any one of claims 1-2.

5. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is run on the processor, it causes the processor to perform the method as described in any one of claims 1-2.

6. A computer program product, characterized in that, When the computer program product is run on a processor, the processor causes the processor to perform the method as described in any one of claims 1-2.

Images