A digital predistortion training method, device and storage medium

Abstract

The present invention discloses a digital pre-distortion training method, device and storage medium, including: using a first digital pre-distortion training model to perform digital pre-distortion training on a first input parameter to obtain a first training result; using a second digital pre-distortion training model to perform digital pre-distortion training on a second input parameter and the first training result. By adopting the present invention, joint optimization can be used to ensure the accuracy of the trained digital pre-distortion model, while simplifying the network structure and reducing the training weight parameters. Pre-distortion models can be trained for multiple PAs, improving modeling accuracy and the versatility of the model architecture. Different neural network models or combined models can be used according to requirements to achieve flexible structure, ensure model convergence, meet different RRU power amplifier requirements, and achieve resource sharing.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a digital predistortion training method, device and storage medium. Background Art

[0002] The power amplifier is an important component of wireless communication systems. The nonlinear characteristics of the power amplifier lead to in-band nonlinear distortion and out-of-band spectrum leakage, so the linearization technology of the power amplifier has attracted much attention.

[0003] In order to suppress the nonlinear distortion of the power amplifier and maintain high output power to improve the efficiency of the amplifier, digital pre-distortion technology has been widely used. Digital pre-distortion technology estimates the nonlinear model of the power amplifier in the digital domain and generates a pre-distortion signal that is opposite to it. After the pre-distortion signal is nonlinearly amplified by the power amplifier, the nonlinear distortion signal component can be reduced, thereby achieving the linearization of the power amplifier.

[0004] The pre-distortion technology currently mainly includes the traditional digital pre-distortion technology realized by the polynomial method and the pre-distortion technology based on the neural network. The shortcomings are:

[0005] The digital pre-distortion technology implemented by the polynomial method cannot accurately describe the characteristics of the power amplifier under large bandwidth and high PAPR (Peak Average Power Ratio) signal input; the existing neural network-based pre-distortion technology has a single structure and poor flexibility, and cannot meet the needs of future intelligent and open networks. Summary of the invention

[0006] The present invention provides a digital pre-distortion training method, device and storage medium, which are used to solve the problems that the existing neural network-based pre-distortion technology has a single structure, poor flexibility and cannot meet the needs of future intelligent and open networks.

[0007] The present invention provides the following technical solutions:

[0008] A digital pre-distortion training method, comprising:

[0009] Performing digital pre-distortion training on a first input parameter using a first digital pre-distortion training model to obtain a first training result;

[0010] A second digital pre-distortion training model is used to perform digital pre-distortion training on the second input parameter and the first training result.

[0011] In implementation, the first input parameter and / or the second input parameter includes one or a combination of the following parameters:

[0012] PA forward I / Q data, PA feedback I / Q data, corresponding different forward or backward delay I / Q data, polynomial characteristic information, power information, temperature information, PA type, and manufacturer information.

[0013] In implementation, the neural network structure of the first digital pre-distortion training model and / or the second digital pre-distortion training model for digital pre-distortion training is one of the following structures or a combination thereof:

[0014] DNN, RNN, CNN.

[0015] During implementation, it further includes:

[0016] The first digital pre-distortion training model and the second digital pre-distortion training model are deployed according to the real-time transmission and computing storage resources.

[0017] During implementation, it further includes:

[0018] The digital pre-distortion of the first digital pre-distortion training model and / or the second digital pre-distortion training model is trained according to the LOSS function.

[0019] In implementation, the LOSS function is a NMSE function or a MSE function.

[0020] During implementation, it further includes:

[0021] The weights of the LOSS functions in the first digital pre-distortion training model and the second digital pre-distortion training model are allocated as needed.

[0022] A digital predistortion training device, comprising:

[0023] The processor reads the program in the memory and performs the following processes:

[0024] Performing digital pre-distortion training on a first input parameter using a first digital pre-distortion training model to obtain a first training result;

[0025] Performing digital predistortion training on the second input parameter and the first training result using a second digital predistortion training model;

[0026] A transceiver is used to receive and send data under the control of the processor.

[0027] In implementation, the first input parameter and / or the second input parameter includes one or a combination of the following parameters:

[0028] PA forward I / Q data, PA feedback I / Q data, corresponding different forward or backward delay I / Q data, polynomial characteristic information, power information, temperature information, PA type, and manufacturer information.

[0029] In implementation, the neural network structure of the first digital pre-distortion training model and / or the second digital pre-distortion training model for digital pre-distortion training is one of the following structures or a combination thereof:

[0030] DNN, RNN, CNN.

[0031] During implementation, it further includes:

[0032] The first digital pre-distortion training model and the second digital pre-distortion training model are deployed according to the real-time transmission and computing storage resources.

[0033] During implementation, it further includes:

[0034] The digital pre-distortion of the first digital pre-distortion training model and / or the second digital pre-distortion training model is trained according to the LOSS function.

[0035] In implementation, the LOSS function is a NMSE function or a MSE function.

[0036] During implementation, it further includes:

[0037] The weights of the LOSS functions in the first digital pre-distortion training model and the second digital pre-distortion training model are allocated as needed.

[0038] A digital predistortion training device, comprising:

[0039] A main training module, configured to perform digital pre-distortion training on a first input parameter using a first digital pre-distortion training model to obtain a first training result;

[0040] The auxiliary training module is used to perform digital pre-distortion training on the second input parameter and the first training result by using the second digital pre-distortion training model.

[0041] In implementation, the first input parameter and / or the second input parameter includes one or a combination of the following parameters:

[0042] PA forward I / Q data, PA feedback I / Q data, corresponding different forward or backward delay I / Q data, polynomial characteristic information, power information, temperature information, PA type, and manufacturer information.

[0043] In implementation, the neural network structure of the first digital pre-distortion training model and / or the second digital pre-distortion training model for digital pre-distortion training is one of the following structures or a combination thereof:

[0044] DNN, RNN, CNN.

[0045] During implementation, it further includes:

[0046] The deployment module is used to deploy the first digital pre-distortion training model and the second digital pre-distortion training model according to the transmission real-time performance and computing and storage resource conditions.

[0047] During implementation, it further includes:

[0048] The digital pre-distortion of the first digital pre-distortion training model and / or the second digital pre-distortion training model is trained according to the LOSS function.

[0049] In implementation, the LOSS function is a NMSE function or a MSE function.

[0050] During implementation, it further includes:

[0051] The weight allocation module is used to allocate the weights of the LOSS functions in the first digital pre-distortion training model and the second digital pre-distortion training model as needed.

[0052] A computer-readable storage medium stores a computer program for executing the digital pre-distortion training method.

[0053] The beneficial effects of the present invention are as follows:

[0054] In the technical solution provided by the embodiment of the present invention, since the training consists of a first digital pre-distortion training model and a second digital pre-distortion training model, joint optimization can be performed to ensure the accuracy of the trained digital pre-distortion model, while simplifying the network structure and reducing the training weight parameters.

[0055] Furthermore, since the first input parameter can be multi-feature and multi-dimensional feature information input such as polynomial feature information, power information, temperature information, PA type, etc., a pre-distortion model can be trained for multiple PAs, thereby improving modeling accuracy and the versatility of the model architecture.

[0056] Furthermore, since the neural network structure selected by the first digital pre-distortion training model and / or the second digital pre-distortion training model performs model training based on an intelligent neural network, different neural network models or combined models can be used according to requirements to achieve flexible structure.

[0057] Furthermore, since the weights of the LOSS functions in the first digital pre-distortion training model and the second digital pre-distortion training model can be allocated as needed, multiple LOSS functions are flexibly adjustable and weights can be configured, which effectively ensures model convergence.

[0058] Furthermore, since the first digital pre-distortion training model and the second digital pre-distortion training model can be deployed according to the real-time transmission and computing and storage resource conditions, the first digital pre-distortion training model and the second digital pre-distortion training model can be flexibly configurable and can be trained and deployed on demand to meet different RRU power amplifier requirements and achieve resource sharing. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0060] Figure 1 It is a schematic diagram of the implementation process of the digital pre-distortion training method in an embodiment of the present invention;

[0061] Figure 2 Schematic diagram of the structure of a digital pre-distortion training system in an embodiment of the present invention;

[0062] Figure 3 This is a schematic diagram of the structure of the DPD main training module in an embodiment of the present invention;

[0063] Figure 4 This is a schematic diagram of the structure of a DPD auxiliary training module in an embodiment of the present invention;

[0064] Figure 5 A schematic diagram of a digital pre-distortion training device structure 1 in an embodiment of the present invention;

[0065] Figure 6 Schematic diagram of the second structure of the digital pre-distortion training device in an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The inventors noticed during the invention process that:

[0067] Traditional digital pre-distortion technology is implemented through the polynomial method. The basic implementation principle is to find a polynomial that can approximate the inverse function of the complex gain characteristic curve of the power amplifier. After obtaining the inverse function, the pre-distortion value of the original input signal is obtained by calculation.

[0068] The higher the order of the digital pre-distortion model of the polynomial method, the higher the modeling accuracy, but the amount of parameter estimation calculation increases exponentially with the increase of the polynomial order. Most polynomial models only consider the amplitude model and ignore the phase compensation of the signal, which affects the accuracy of the digital pre-distortion model. In addition, the polynomial method uniformly establishes a digital pre-distortion model based on polynomials. The model is fixed and single, and cannot accurately describe the characteristics of the power amplifier under large bandwidth and high PAPR signal input. At the same time, the input feature information that can be processed is limited. At the same time, the current base station uses polynomial digital pre-distortion. The algorithm is implemented locally in the RRU (Radio Remote Unit). The DPD (Digital Pre-Distortion) polynomial is trained through the computing resources and storage resources inside the RRU to perform PA (power amplifier) ​​nonlinear distortion compensation.

[0069] At present, pre-distortion based on neural networks is still in the primary research and exploration stage. There are simple memoryless digital pre-distortion models and memory digital pre-distortion models. The memoryless digital pre-distortion model directly fits the input signal through the input-output neural network to produce the output result. The memory digital pre-distortion model further considers the delayed signal on the basis of the memoryless digital pre-distortion model.

[0070] The existing neural network-based digital pre-distortion model has a simple implementation structure, but has few input parameter types, relatively simple input features, and poor universality. At the same time, it adopts a fully connected approach, resulting in a large network scale, many weight parameters that need to be trained and optimized, and a large amount of training. The current neural network algorithms are all trained with unified input and output connections, and use a single Loss function, which has the problems of gradient dispersion and model non-convergence. In the face of future intelligent and open network needs, it is necessary to further explore high-performance intelligent digital pre-distortion DPD solutions.

[0071] Based on this, an embodiment of the present invention provides a digital pre-distortion solution based on machine learning. With the help of a DPD first digital pre-distortion training model and a second digital pre-distortion training model, through feature information input and auxiliary feature information input, and using an intelligent neural network algorithm, a digital pre-distortion model can be trained for multiple PAs, thereby improving the universality of the model architecture, and with the help of additional feature information input, the accuracy of the model is improved, and out-of-band interference is suppressed.

[0072] The digital pre-distortion model is optimized by using multiple loss functions. Even if the gradient of the loss function of the first digital pre-distortion training model is diffused during the training process, the second digital pre-distortion training model can still train a digital pre-distortion model.

[0073] The first digital pre-distortion training model and the second digital pre-distortion training model can also be flexibly deployed and configured to meet different real-time and computing storage resource requirements.

[0074] The specific implementation of the present invention will be described below in conjunction with the accompanying drawings.

[0075] Figure 1 A flowchart for implementing a digital pre-distortion training method is shown in the figure, including:

[0076] Step 101: Perform digital predistortion training on a first input parameter using a first digital predistortion training model to obtain a first training result;

[0077] Step 102: Use a second digital pre-distortion training model to perform digital pre-distortion training on the second input parameter and the first training result.

[0078] In implementation, the first input parameter and / or the second input parameter includes one or a combination of the following parameters:

[0079] PA forward I / Q data, PA feedback I / Q data, corresponding different forward or backward delay I / Q data, polynomial characteristic information, power information, temperature information, PA type, and manufacturer information.

[0080] In implementation, the neural network structure of the first digital pre-distortion training model and / or the second digital pre-distortion training model for digital pre-distortion training is one of the following structures or a combination thereof:

[0081] DNN, RNN, CNN.

[0082] During implementation, the method may further include:

[0083] The digital pre-distortion of the first digital pre-distortion training model and / or the second digital pre-distortion training model is trained according to the LOSS function.

[0084] In a specific implementation, the LOSS function is an NMSE function or an MSE function.

[0085] In the following implementation, the implementation method of the functional module will be used for explanation. For ease of understanding, in the embodiment, the functional module using the first digital pre-distortion training model will be called a DPD main training module, and the functional module using the second digital pre-distortion training model will be called a DPD auxiliary training module.

[0086] First, the overall architecture is described.

[0087] Figure 2This is a structural diagram of a digital pre-distortion training system. As shown in the figure, the functional architecture of the digital pre-distortion method mainly consists of a DPD main training module and a DPD auxiliary training module as well as a connection interface between the two.

[0088] The implementation of the DPD main training module is described below.

[0089] Figure 3 This is a structural diagram of the DPD main training module. As shown in the figure, the DPD main training module mainly completes the training of the DPD basic model. The module implementation includes a feature information input layer, a neural network training layer and an output prediction layer. The main training module trains the input feature information through the neural network training layer and outputs the trained output prediction value after digital pre-distortion.

[0090] The feature information input by the feature information input layer of the DPD main training module may include but is not limited to the following information:

[0091] PA forward I / Q data, PA feedback I / Q data, corresponding forward or backward delay I / Q data, polynomial characteristic information, power information, temperature information, PA type, manufacturer information and other characteristic information, one or more combinations thereof. Among them, I / Q is a radio frequency signal, which can be represented by amplitude and phase in polar coordinates and by X and Y values ​​in rectangular coordinates. However, in digital communication systems, X is generally replaced by I to indicate in-phase, and Y is replaced by Q to indicate a 90° phase.

[0092] Among them, various forms of neural network structures can be used for feature information processing, such as DNN (Deep Neural Network), RNN (Recurrent Neural Network), CNN (Convolutional Neural Network) or any hybrid architecture of the three networks.

[0093] Among them, the LOSS function used in the main training module can minimize the error between the predicted value and the target value, and can adopt error functions such as NMSE (normalized mean squared error) or MSE (mean squared error).

[0094] The implementation of the DPD auxiliary training module is explained below.

[0095] Figure 4The DPD auxiliary training module is a structural diagram. As shown in the figure, the DPD auxiliary training module further optimizes the training of the digital pre-distortion model based on the main training module. The DPD auxiliary training module completes the training output of the final digital pre-distortion model through auxiliary feature information processing and input, and uses all or part of the neural network training layer output results of the main input module. The auxiliary training module includes an auxiliary feature information input layer, a main module information input layer, an input merging layer, a neural network training layer, an output prediction layer and a LOSS function.

[0096] Among them, the auxiliary feature information is the auxiliary feature information for completing digital pre-distortion, wherein the auxiliary feature information includes but is not limited to one or more combinations of other feature information except the feature information of the DPD main training module mentioned above. Feature processing and screening of corresponding feature information are completed through the auxiliary feature information input layer. It is easy to understand that the auxiliary feature information can be selected from PA forward I / Q data, PA feedback I / Q data, corresponding different forward or backward delay I / Q data, polynomial feature information, power information, temperature information, PA type, and manufacturer information according to needs and purposes, and can be different from or partially the same as the main input module.

[0097] Among them, the main module information output layer is the output result of all or part of the neural network training layer of the main input module, which is used to complete the final digital pre-distortion model training together with the auxiliary feature information input layer.

[0098] Among them, the merging layer completes the normalization and merging of the auxiliary feature information and the main module information, which is used for the neural network model training of the auxiliary training module.

[0099] Among them, various forms of neural network structures can be used for auxiliary feature information and main module information processing training, such as DNN, RNN, CNN or any hybrid architecture of the three networks.

[0100] Among them, the LOSS function used in the auxiliary training module is to minimize the error between the predicted value and the target value. Error functions such as NMSE or MSE can be adopted, which can be the same as or different from the LOSS function of the main training module.

[0101] The connection interface is used for information and data exchange between the main training module and the auxiliary training module, such as the main module information input layer for the auxiliary training module, the main module adjusting the weights of the neural network training layer of the main training module according to the training status of the auxiliary training module LOSS function, the transmission of parameter configuration information for the auxiliary module and auxiliary module training feedback information.

[0102] During implementation, the method may further include:

[0103] The weights of the LOSS functions in the first digital pre-distortion training model and the second digital pre-distortion training model are allocated as needed.

[0104] The LOSS functions of the main training module and the auxiliary training module can be dynamically adjusted according to the model training situation and can be different LOSS functions.

[0105] Different weights can be assigned to make the digital pre-distortion training model more optimal, such as the main module LOSS weight is 0.6, the auxiliary module LOSS weight is 0.4, etc.

[0106] The auxiliary training module is optional. When the weight of the LOSS function of the auxiliary training module is 0 or no auxiliary feature information input is required or resources are insufficient, only the DPD main training module is executed, and the DPD auxiliary training module is not executed to obtain a digital pre-distortion model.

[0107] During implementation, the method may further include:

[0108] The first digital pre-distortion training model and the second digital pre-distortion training model are deployed according to the real-time transmission and computing storage resources.

[0109] The main training module and the auxiliary training module can also be deployed flexibly. The main training module and the auxiliary training module can be deployed at the remote end or the near end according to the real-time transmission and computing and storage resources. For example, when local computing resources are limited, the main module can be deployed at the remote end, such as CU (Central Unit) or DU (Distributed Unit), and the auxiliary module can be deployed at the near end, such as RRU.

[0110] The main training module and auxiliary training module adopt full or partial training according to the computing resource situation. When the computing resources meet the demand, the main training module and the auxiliary training module are trained and updated at the same time. When the computing resources do not meet the demand, the main training module and the auxiliary training module can be trained in time-sharing or only the auxiliary module is trained. The weight of the main module is fixed, and only a small part of the weight is updated.

[0111] The feature information of the main module and the feature information of the auxiliary training module can be flexibly adjusted and configured according to the computing storage resources and real-time requirements, and the corresponding feature input items can be added or deleted. For example, when the computing storage resources are limited, only the PA forward I / Q data, PA feedback I / Q data, and the corresponding different forward or backward delay I / Q data are selected as feature input items for digital pre-distortion model training.

[0112] Based on the same inventive concept, the embodiments of the present invention further provide a digital pre-distortion training device and a computer-readable storage medium. Since the principles of solving the problem by these devices are similar to those of the digital pre-distortion training method, the implementation of these devices can refer to the implementation of the method, and the repeated parts will not be repeated.

[0113] When implementing the technical solution provided by the embodiment of the present invention, it can be implemented as follows.

[0114] Figure 5 This is a schematic diagram of the structure of a digital pre-distortion training device. As shown in the figure, the device includes:

[0115] The processor 500 is used to read the program in the memory 520 and execute the following process:

[0116] Performing digital pre-distortion training on a first input parameter using a first digital pre-distortion training model to obtain a first training result;

[0117] Performing digital predistortion training on the second input parameter and the first training result using a second digital predistortion training model;

[0118] The transceiver 510 is configured to receive and send data under the control of the processor 500 .

[0119] In implementation, the first input parameter and / or the second input parameter includes one or a combination of the following parameters:

[0120] PA forward I / Q data, PA feedback I / Q data, corresponding different forward or backward delay I / Q data, polynomial characteristic information, power information, temperature information, PA type, and manufacturer information.

[0121] In implementation, the neural network structure of the first digital pre-distortion training model and / or the second digital pre-distortion training model for digital pre-distortion training is one of the following structures or a combination thereof:

[0122] DNN, RNN, CNN.

[0123] During implementation, it further includes:

[0124] The first digital pre-distortion training model and the second digital pre-distortion training model are deployed according to the real-time transmission and computing storage resources.

[0125] During implementation, it further includes:

[0126] The digital pre-distortion of the first digital pre-distortion training model and / or the second digital pre-distortion training model is trained according to the LOSS function.

[0127] In implementation, the LOSS function is a NMSE function or a MSE function.

[0128] During implementation, it further includes:

[0129] The weights of the LOSS functions in the first digital pre-distortion training model and the second digital pre-distortion training model are allocated as needed.

[0130] Among them, Figure 5In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically linking together various circuits of one or more processors represented by processor 500 and memory represented by memory 520. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface. The transceiver 510 may be a plurality of components, namely, a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium. The processor 500 is responsible for managing the bus architecture and general processing, and the memory 520 may store data used by the processor 500 when performing operations.

[0131] Figure 6 This is a schematic diagram of the structure of the digital pre-distortion training device. As shown in the figure, the device includes:

[0132] A main training module 601 is used to obtain a first training result after performing digital pre-distortion training on a first input parameter using a first digital pre-distortion training model;

[0133] The auxiliary training module 602 is used to perform digital pre-distortion training on the second input parameter and the first training result by using the second digital pre-distortion training model.

[0134] In implementation, the first input parameter and / or the second input parameter includes one or a combination of the following parameters:

[0135] PA forward I / Q data, PA feedback I / Q data, corresponding different forward or backward delay I / Q data, polynomial characteristic information, power information, temperature information, PA type, and manufacturer information.

[0136] In implementation, the neural network structure of the first digital pre-distortion training model and / or the second digital pre-distortion training model for digital pre-distortion training is one of the following structures or a combination thereof:

[0137] DNN, RNN, CNN.

[0138] During implementation, it further includes:

[0139] The deployment module is used to deploy the first digital pre-distortion training model and the second digital pre-distortion training model according to the transmission real-time performance and computing and storage resource conditions.

[0140] During implementation, it further includes:

[0141] The digital pre-distortion of the first digital pre-distortion training model and / or the second digital pre-distortion training model is trained according to the LOSS function.

[0142] In implementation, the LOSS function is a NMSE function or a MSE function.

[0143] During implementation, it further includes:

[0144] The weight allocation module is used to allocate the weights of the LOSS functions in the first digital pre-distortion training model and the second digital pre-distortion training model as needed.

[0145] For the convenience of description, the various parts of the above-mentioned device are divided into various modules or units according to their functions and described separately. Of course, when implementing the present invention, the functions of each module or unit can be implemented in the same or multiple software or hardware.

[0146] A computer-readable storage medium is also provided in an embodiment of the present invention. The computer-readable storage medium stores a computer program for executing the above digital pre-distortion training method.

[0147] For specific implementation, please refer to the implementation of the digital pre-distortion training method.

[0148] In summary, the technical solution provided by the implementation of the present invention is mainly composed of a DPD main training module, a DPD auxiliary training module and a connection interface between the two.

[0149] The DPD main training module mainly completes the input and output training processing of feature information processing. The module implementation includes a feature information input layer, a neural network training layer, and an output prediction layer. The main training module trains the input feature information through the neural network training layer and outputs the trained output prediction value.

[0150] The DPD auxiliary training module completes the auxiliary feature information processing and input based on the main training module, and uses all or part of the neural network training layer output results of the main input module to complete the training output of the final digital pre-distortion model. The auxiliary training module includes the auxiliary feature information input layer, the main module information input layer, the input merging layer, the neural network training layer, the output prediction layer and the LOSS function.

[0151] The connection interface is used for information and data exchange between the main and auxiliary modules.

[0152] Auxiliary modules are flexibly optional and can be flexibly selected according to the type of input feature information or resource configuration.

[0153] The main LOSS and auxiliary LOSS are flexibly configurable, and the weights can be adjusted as needed to train and optimize the digital pre-distortion model.

[0154] The DPD main training module and auxiliary training module training mechanisms are optional, and training can be done simultaneously or in different time periods depending on resource conditions.

[0155] The DPD main training module and auxiliary training module can be flexibly deployed at the near end or far end according to the real-time requirements and resource conditions of the RRU.

[0156] The feature information of the main module and the feature information of the auxiliary training module can be flexibly adjusted and configured according to the computing storage resources and real-time requirements, and the corresponding feature input items can be added or deleted.

[0157] It can be seen that the training consists of a main training module and an auxiliary training module, which can jointly optimize to ensure the accuracy of the trained digital pre-distortion model, while simplifying the network structure and reducing the training weight parameters.

[0158] The multi-feature and multi-dimensional feature information input, such as polynomial feature information, power information, temperature information, PA type, etc., can train pre-distortion models for multiple PAs, improve the modeling accuracy, and enhance the versatility of the model architecture.

[0159] Model training is based on intelligent neural networks, and different neural network models or combined models can be used according to needs to achieve flexible structure.

[0160] The multi-LOSS function is flexible and adjustable, with configurable weights, which effectively ensures the convergence of the model.

[0161] The main and auxiliary modules are flexibly configurable and can be trained and deployed on demand to meet different RRU power amplifier requirements and achieve resource sharing.

[0162] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0163] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0164] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0165] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0166] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A digital predistortion training method, characterized in that: include: Performing digital pre-distortion training on a first input parameter using a first digital pre-distortion training model to obtain a first training result, wherein the first digital pre-distortion training model includes a feature information input layer, a neural network training layer, an output prediction layer, and a loss LOSS function; Performing digital predistortion training on the second input parameter and the first training result using a second digital predistortion training model to obtain a final digital predistortion model, wherein the second digital predistortion training model includes an auxiliary feature information input layer, a main module information input layer, an input merging layer, a neural network training layer, an output prediction layer, and a LOSS function, wherein the LOSS function of the first digital predistortion training model is the same as or different from the LOSS function of the second digital predistortion training model; The first input parameter includes: forward RF signal I / Q data of a power amplifier PA, PA feedback I / Q data, and corresponding different forward or backward delay I / Q data; The second input parameter includes one or a combination of the following parameters: corresponding different forward or backward delay I / Q data, polynomial characteristic information, power information, temperature information, PA type, and manufacturer information; The neural network structure of the first digital pre-distortion training model and / or the second digital pre-distortion training model for digital pre-distortion training is one of the following structures or a combination thereof: Deep neural network DNN, recurrent neural network RNN, convolutional neural network CNN.

2. The method according to claim 1, characterized in that Further including: The first digital pre-distortion training model and the second digital pre-distortion training model are deployed according to the real-time transmission and computing storage resources.

3. The method according to any one of claims 1 to 2, characterized in that: Further including: The digital pre-distortion of the first digital pre-distortion training model and / or the second digital pre-distortion training model is trained according to the LOSS function.

4. The method according to claim 3, characterized in that The LOSS function is a normalized mean square error (NMSE) function or a normalized mean square error (MSE) function.

5. The method according to claim 3, characterized in that Further including: The weights of the LOSS functions in the first digital pre-distortion training model and the second digital pre-distortion training model are allocated as needed.

6. A digital predistortion training device, characterized in that: include: The processor reads the program in the memory and performs the following processes: Performing digital pre-distortion training on a first input parameter using a first digital pre-distortion training model to obtain a first training result, wherein the first digital pre-distortion training model includes a feature information input layer, a neural network training layer, an output prediction layer, and a loss LOSS function; Performing digital predistortion training on the second input parameter and the first training result using a second digital predistortion training model to obtain a final digital predistortion model, wherein the second digital predistortion training model includes an auxiliary feature information input layer, a main module information input layer, an input merging layer, a neural network training layer, an output prediction layer, and a LOSS function, wherein the LOSS function of the first digital predistortion training model is the same as or different from the LOSS function of the second digital predistortion training model; a transceiver for receiving and transmitting data under the control of the processor; The first input parameter includes: forward RF signal I / Q data of a power amplifier PA, PA feedback I / Q data, and corresponding different forward or backward delay I / Q data; The second input parameter includes one or a combination of the following parameters: corresponding different forward or backward delay I / Q data, polynomial characteristic information, power information, temperature information, PA type, and manufacturer information; The neural network structure of the first digital pre-distortion training model and / or the second digital pre-distortion training model for digital pre-distortion training is one of the following structures or a combination thereof: Deep neural network DNN, recurrent neural network RNN, convolutional neural network CNN.

7. A digital predistortion training device, characterized in that: include: A main training module, used to obtain a first training result after performing digital pre-distortion training on a first input parameter using a first digital pre-distortion training model, wherein the first digital pre-distortion training model includes a feature information input layer, a neural network training layer, an output prediction layer, and a loss LOSS function; an auxiliary training module, configured to perform digital pre-distortion training on the second input parameter and the first training result using a second digital pre-distortion training model to obtain a final digital pre-distortion model, wherein the second digital pre-distortion training model comprises an auxiliary feature information input layer, a main module information input layer, an input merging layer, a neural network training layer, an output prediction layer and a LOSS function, wherein the LOSS function of the first digital pre-distortion training model is the same as or different from the LOSS function of the second digital pre-distortion training model; in, The first input parameter includes: power amplifier PA forward RF signal I / Q data, PA feedback I / Q data, and corresponding different forward or backward delay I / Q data; The second input parameter includes one or a combination of the following parameters: corresponding different forward or backward delay I / Q data, polynomial characteristic information, power information, temperature information, PA type, and manufacturer information; The neural network structure of the first digital pre-distortion training model and / or the second digital pre-distortion training model for digital pre-distortion training is one of the following structures or a combination thereof: Deep neural network DNN, recurrent neural network RNN, convolutional neural network CNN.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program for executing the method according to any one of claims 1 to 5.

Images