Digital Predistortion Method, Device and Electronic Equipment for Multi-Antenna Satellite System

Abstract

A digital predistortion method, device and electronic device for a multi-antenna satellite system provided by the present application relate to the field of satellite communication. The present application is based on a multi-gated mixture-of-experts network, uses an expert network and a private network to learn common features and private features, and uses a gating network to adaptively select features for different tasks. By automatically adjusting the parameterization between the common features and the private features, the predistortion model can effectively learn complex data distributions, improve linearization performance, and reduce cross-channel nonlinear crosstalk. Compared with the existing neural network-based predistortion models, the predistortion model of the present application can obtain a more accurate power amplifier inverse model under the same number of training iterations to effectively compensate for multi-channel power amplifier distortion and crosstalk. Compared with the existing polynomial machine learning methods, the predistortion model has superior performance in both nonlinear distortion compensation and crosstalk suppression.

Description

Technical Field

[0001] The present application relates to the field of satellite communication. Specifically, it relates to a digital predistortion method, device, and electronic device for a multi-antenna satellite system. Background Art

[0002] In a satellite communication network, with the continuous growth of the demand for spectral efficiency and the number of access devices, multiple antennas are deployed at the communication transceiver to improve the capacity, coverage, and data transmission rate of the satellite communication system. However, like traditional communication systems, due to the presence of non-linear devices such as power amplifiers in the channel, multi-antenna satellite systems also face the problem of linearization. The non-linear distortion of the power amplifier will reduce the signal quality and the capacity of the system. Especially when the power amplifier operates in the non-linear region, its saturation characteristic will cause the signal distortion to further intensify. Therefore, to ensure both the efficiency and linearity of the power amplifier, an efficient power amplifier linearization technology is also required in multi-antenna satellite systems. Digital predistortion technology has become one of the most widely used linearization technologies due to its low cost, simple implementation, and excellent performance. However, due to antenna coupling or leakage of the shared local oscillator, the non-linear behavior of each power amplifier may be affected by the behavior of adjacent channels. Therefore, traditional one-to-one individual compensation linearization technologies are not sufficient to meet the linearization requirements of multi-antenna systems with strong crosstalk. Summary of the Invention

[0003] The purpose of the embodiments of the present application is to provide a digital predistortion method, device, and electronic device for a multi-antenna satellite system, which solves the above problems existing in the prior art and can correct the non-linear distortion of the signal corresponding to the current antenna and eliminate the crosstalk from other antennas.

[0004] In a first aspect, a digital predistortion method for a multi-antenna satellite system is provided. The method may include:

[0005] Obtain the input data of multiple antennas to be processed; the input data includes the I / Q components and signal amplitude of the input signal;

[0006] Concatenate the I / Q components and signal amplitude of multiple antennas to obtain a concatenated vector;

[0007] Use the trained predistortion model to process the concatenated vector to obtain the target I / Q components of each antenna, and convert each target I / Q component into a predistorted signal, so that the power amplifier corresponding to each antenna processes the corresponding predistorted signal and then transmits it; wherein, the predistortion model includes a first number of expert networks, a second number of gating networks, and a second number of private networks.

[0008] In a possible implementation, for any expert network, the expert network has a fully connected network structure, including an expert input layer, an expert hidden layer, and an expert output layer;

[0009] The activation functions of the expert hidden layer and the expert output layer are both Tanh functions.

[0010] In a possible implementation, for any gating network, the gating network only includes a network layer;

[0011] The activation function of the network layer is the Softmax function.

[0012] In a possible implementation, the private network has a fully connected network structure, including a private input layer, a private hidden layer, and a private output layer;

[0013] The activation functions of the private hidden layer and the private output layer are both Leaky ReLU functions.

[0014] In a possible implementation, using a trained predistortion model to process the concatenated vector to obtain the target I / Q components of each antenna, including:

[0015] Using the first number of expert networks in the predistortion model to process the concatenated vector respectively to obtain the common feature vectors corresponding to each expert network;

[0016] And using the second number of gating networks in the predistortion model to process the concatenated vector to obtain the weights of each expert network;

[0017] Using the second number of private networks in the predistortion model to process the weights of each expert network and the common feature vectors corresponding to each expert network to obtain the target I / Q components of each antenna.

[0018] In a second aspect, a digital predistortion device for a multi-antenna satellite system is provided, and the device may include:

[0019] An acquisition unit, configured to acquire input data of multiple antennas to be processed; the input data includes the I / Q components and the signal amplitude of the input signal;

[0020] A splicing unit, configured to splice the I / Q components and the signal amplitude of multiple antennas to obtain a concatenated vector;

[0021] A processing unit, configured to process the spliced vector by using a trained predistortion model to obtain target I / Q components of each antenna, and convert the target I / Q components into predistortion signals, so that power amplifiers corresponding to the antennas process the corresponding predistortion signals and then transmit them; wherein, the predistortion model includes a first number of expert networks, a second number of gating networks, and a second number of private networks.

[0022] In a possible implementation, for any one of the expert networks, the expert network has a fully connected network structure, including an expert input layer, an expert hidden layer, and an expert output layer;

[0023] The activation functions of the expert hidden layer and the expert output layer are both Tanh functions.

[0024] In a possible implementation, for any one of the gating networks, the gating network only includes a network layer;

[0025] The activation function of the network layer is a Softmax function.

[0026] In a third aspect, an electronic device is provided. The electronic device includes a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus;

[0027] The memory is used to store a computer program;

[0028] The processor is configured to implement the method steps described in any one of the first aspects when executing the program stored on the memory.

[0029] In a fourth aspect, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and the computer program implements the method steps described in any one of the first aspects when executed by a processor.

[0030] A digital predistortion method for a multi-antenna satellite system provided in this application is based on a multi-feedback loop architecture, and uses a multi-gated mixture of experts network to model the predistortion model of the multi-antenna system. Among them, the multi-gated mixture of experts network uses expert networks and private networks to learn common features and private features, and uses gating networks to adaptively select features for different tasks. By automatically adjusting the parameterization between common features and private features, the predistortion model can effectively learn complex data distributions, improve linearization performance, and reduce cross-channel nonlinear crosstalk. Compared with existing neural network-based predistortion models, the predistortion model of this application can obtain a more accurate power amplifier inverse model under the same number of training iterations to effectively compensate for multi-channel power amplifier distortion and crosstalk. Compared with existing polynomial machine learning methods, the predistortion model has better performance in both nonlinear distortion compensation and crosstalk suppression. Brief Description of the Drawings

[0031] To more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present application, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other relevant drawings can also be obtained based on these drawings.

[0032] Figure 1 It is a system architecture diagram of a digital predistortion method applied to a multi-antenna satellite system provided by an embodiment of the present application;

[0033] Figure 2 It is a schematic flowchart of a digital predistortion method for a multi-antenna satellite system provided by an embodiment of the present application;

[0034] Figure 3 It is a schematic diagram of a single DPD module provided by an embodiment of the present application;

[0035] Figure 4 It is a schematic diagram of the application process of a digital predistortion method for a multi-antenna satellite system provided by an embodiment of the present application;

[0036] Figure 5 It is a schematic structural diagram of a digital predistortion device for a multi-antenna satellite system provided by an embodiment of the present application;

[0037] Figure 6 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed Embodiments

[0038] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0039] The digital predistortion method for a multi-antenna satellite system provided by the embodiments of the present application can be applied in Figure 1 the system architecture shown in Figure 1As shown in the figure, the system may include: a server and a multi-antenna satellite system. The server may be a physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms. The multi-antenna satellite system and the server may be directly or indirectly connected through wired or wireless communication methods, which are not limited in this application.

[0040] A multi-antenna satellite system, configured to send the input data of each antenna during a period to be processed to the server;

[0041] The server is configured to obtain the input data of each antenna to execute a digital predistortion method for a multi-antenna satellite system provided in this application.

[0042] In a satellite communication network, with the continuous growth of the demand for spectral efficiency and the number of access devices, multiple antennas are deployed at the communication transceiver to improve the capacity, coverage, and data transmission rate of the satellite communication system. However, like traditional communication systems, due to the presence of non-linear devices such as power amplifiers in the channel, the multi-antenna satellite system also faces the problem of linearization. The non-linear distortion of the power amplifier will reduce the signal quality and the capacity of the system. Especially when the power amplifier operates in the non-linear region, its saturation characteristic will cause the signal distortion to further intensify. Therefore, to ensure both the efficiency and linearity of the power amplifier, an efficient power amplifier linearization technology is also required in the multi-antenna satellite system. Digital predistortion technology has become the most widely used linearization technology at present due to its advantages such as low cost, simple implementation, and excellent performance. However, due to antenna coupling or leakage of the shared local oscillator, the non-linear behavior of each power amplifier may be affected by the behavior of adjacent channels. Therefore, the traditional one-to-one independent compensation linearization technology is not sufficient to meet the linearization requirements of a multi-antenna system with strong crosstalk.

[0043] Currently, multi-antenna digital predistortion architectures are mainly divided into two types: single DPD architecture and multiple DPD architectures. Among them, the single DPD architecture only uses one digital predistortion module to linearly process the main lobe beam direction of the combined signal of each power amplifier, and its implementation complexity is greatly reduced. However, this method cannot guarantee the linearity of all power amplifiers, resulting in distortion in directions other than the main beam direction. The multiple DPD architecture deploys a dedicated DPD module on each radio frequency link, that is, each DPD module performs one-to-one individual compensation for each power amplifier. This method can guarantee the linearity of all power amplifiers and can compensate for signal distortion in each direction. However, the complexity of the system will increase exponentially with the increase in the number of antennas and is only applicable to all-digital architectures.

[0044] Therefore, the present application provides a digital predistortion method for a multi-antenna satellite system to solve the above problems existing in the prior art, which can correct the non-linear distortion of the signal corresponding to the current antenna and eliminate the crosstalk from other antennas.

[0045] The preferred embodiments of the present application will be described below in conjunction with the accompanying drawings of the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application and are not used to limit the present application. And without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0046] Figure 2 It is a schematic flow chart of a digital predistortion method for a multi-antenna satellite system provided by an embodiment of the present application. As Figure 2 shown, the method may include:

[0047] Step S210, obtain the input data of multiple antennas to be processed.

[0048] Specifically, the multi-antenna satellite system collects the input data during the period to be processed and sends the input data to the server.

[0049] The input data includes the I / Q components and the signal amplitude of the input signal; among them, the I / Q components are the I component (in-phase component) and the Q component (quadrature component).

[0050] Step S220, splice the I / Q components and the signal amplitude of multiple antennas to obtain a spliced vector.

[0051] Step S230, use the trained predistortion model to process the spliced vector to obtain the target I / Q components of each antenna, and convert the obtained target I / Q components into predistorted signals, so that the power amplifiers corresponding to each antenna process the corresponding predistorted signals and then transmit them.

[0052] It can be understood that each predistortion signal is a complex signal obtained by performing conversion processing on the corresponding target I / Q component.

[0053] Before performing step S230, the training process of the predistortion model includes:

[0054] A. Configure a pre-trained predistortion model; the pre-trained predistortion model includes a first number of expert networks, a second number of gating networks, and a second number of private networks. Among them, the first number is determined artificially, and the second number is determined according to the number of antennas in the multi-antenna satellite system.

[0055] That is to say, the pre-trained predistortion model of this application is a complex neural network structure composed of multiple expert networks and gating networks. Expert network: Shares the underlying structure, each expert network is relatively independent, and can capture unique relevant features in a single task (the processing task corresponding to each antenna channel). Gating network: Adaptively selects the contributions of expert networks for different tasks. Private network: Each task has a private network, which is specifically used to learn the private features of each power amplifier.

[0056] B. Construct a training set. Since this solution only uses a single DPD module to linearize the power amplifiers of all channels (antennas) in the multi-antenna satellite system, and at the same time needs to consider the non-linear crosstalk problem between multiple antennas, and since the DPD module simulates the reverse characteristics of the power amplifier, it is necessary to reverse the input signal and output signal of the power amplifier during the DPD coefficient extraction process; therefore, the input of the pre-trained predistortion model is the I / Q components and signal amplitudes (signal amplitude-related terms) of the output signals of the multi-antenna power amplifiers at the first moment and multiple second moments within the historical period; where the second moment is the historical moment of the first moment.

[0057] Specifically: Since the power amplifier has a memory effect, that is, the output signal of the power amplifier at the first moment is related not only to the input signal of the power amplifier at the first moment, but also to the input signal of the power amplifier at the historical moment (the second moment) of the first moment. Therefore, when extracting the characteristics of the power amplifier, it is necessary to consider the input signal at the second moment at the same time. Therefore, the training set is to collect the signals within the historical period, and process the "signals at each moment of this historical period" according to the following expression (here, for the input signal at each moment, it is equivalent to splicing the signals of the memory length of P second moments), and finally use the corresponding historical splicing vectors at all moments as the training set.

[0058] The specific expression can be:

[0059]

[0060] Where is the historical concatenated vector, and are respectively the I component and the Q component of the input signal of the power amplifier in the th antenna at the first moment , is the memory length, is the th antenna's power amplifier's signal amplitude of the input signal at the first moment , is the non - linear order.

[0061] The determination method of the concatenated vector in practical applications is the same as that of the historical concatenated vector in the training process.

[0062] C. In the feed - forward transmission, the underlying structure shared by multiple gates is divided into multiple expert networks, which are relatively independent of each other, so as to capture unique features in a single task. At the same time, each power amplifier learning task (which can also be understood as the task corresponding to each antenna) is configured with a gating network, which outputs a probability distribution (weight) to regulate the contribution of each expert network to each task. Therefore, the outputs of multiple expert networks will be weighted - averaged according to the weights generated by the specific task gating network and then input into the private network related to each task to learn and extract the private features of that task.

[0063] During the training process of the pre - trained pre - distortion model, the expressions of the processing processes of each network can be:

[0064]

[0065]

[0066] Among them, is the historical common feature vector output by the th expert network, is the historical concatenated vector, is the second parameter of the th expert network, is the weight matrix of the gating network, is the th private network's input, is the th antenna's power amplifier's historical target I / Q component at the first moment , is the private network function of the th task, is the fourth parameter; wherein, both the third parameter and the fourth parameter are parameters of the private network function.

[0067] Since the private network at the top layer of the pre-trained pre-distortion model is designed to independently learn the private characteristics of each power amplifier, each private network will output the historical target I / Q components corresponding to the predicted output of each power amplifier, denoted as:

[0068]

[0069] wherein, and are the historical target I component and the historical target Q component output by the pre-trained pre-distortion model at the first moment n in the th antenna respectively, thus determining the historical target I / Q components.

[0070] In some embodiments, during the feedback training process, the pre-trained pre-distortion model is trained using backpropagation and gradient descent to optimize the joint loss function of multiple power amplifier learning tasks. Therefore, in order to achieve accurate learning and feature extraction of the multi-antenna power amplifier inverse model, the loss function of the pre-distortion model proposed in this application is:

[0071]

[0072] wherein, and are the I / Q components of the actual input signal of the kth antenna power amplifier; the actual input signal is the signal input to the corresponding antenna without any crosstalk; since the pre-distortion model needs to obtain the mapping relationship between the output signal and the input signal of the power amplifier in order to extract the inverse characteristics including crosstalk and the superposition of the power amplifier itself, the output signal of the pre-distortion model needs to be as close as possible to the input signal of the power amplifier.

[0073] In order to achieve efficient sharing and independent learning between multiple tasks, the common features of the input signals of multiple power amplifiers are learned through the underlying expert network. At the top layer of the model, the private network corresponds to each specific task, and based on the output of the underlying expert network, the private network learns and identifies the private characteristics of multiple power amplifiers.

[0074] In a specific embodiment, the input signals of multiple antennas are collected. As shown in Figure 3 , since this application only uses a single DPD module to linearize multiple power amplifiers simultaneously, when training the pre-trained pre-distortion model, it is necessary to process the input signals of multiple power amplifiers and combine the historical splicing vectors to form a training set, and train the pre-trained pre-distortion model based on the training set.

[0075] Collect the input signals and output signals of multiple power amplifiers within a historical period. For each moment within the historical period, shape the output signal of the power amplifier at each moment into a historical concatenated vector formed by concatenating the output signal at the first moment, the historical I / Q components output at multiple second moments, and the signal amplitude (signal amplitude-related terms). Since a single DPD module is used and inter-channel non-linear crosstalk needs to be considered, the historical concatenated vectors shaped by multiple power amplifiers are used as training samples. Correspondingly, shape the input signals of multiple power amplifiers into the I / Q components of the input signal at the first moment, and concatenate and combine them as training labels.

[0076] The input data of the pre-distortion model in the training set includes the cross-distortion terms of other antennas. Therefore, in the offline training, not only the reverse relationship between the input data and output data of each antenna power amplifier is simulated, but also the reverse relationship between the input data of the power amplifier and the cross-distortion terms is simulated.

[0077] The pre-distortion model based on multi-task learning combines the inputs of all antennas and combines the shared learning mode to facilitate extracting the crosstalk characteristics crosstalking to the current antenna for the current antenna. In practical applications, the non-linear distortion of the power amplifier and the non-linear crosstalk between multiple antennas can be suppressed simultaneously. Send the pre-compensated signal into the power amplifiers of each antenna for amplification, and the influence of its non-linear distortion and crosstalk cancels out with the pre-compensation, making the output signal of the power amplifier of each antenna close to a linear relationship with the original input signal.

[0078] Furthermore, crosstalk means that the input signal of the current antenna is affected by the input signals on adjacent antennas, so that the input signal of the power amplifier of the current antenna contains signal components from other antennas, and further distortion occurs after passing through the power amplifier. Therefore, in order to more accurately obtain the characteristics of the power amplifier and compensate for crosstalk, the input signals of other antennas are concatenated in the training set. In short, not only the characteristics of the output and input of the current antenna are extracted, but also the characteristics of the output of the current antenna and other antennas are extracted (equivalent to extracting the characteristics of the crosstalk part).

[0079] Train the pre-trained pre-distortion model through the training samples and training labels, that is, learn and extract the non-linear characteristics of multiple power amplifiers.

[0080] Combined Figure 4 As shown, the pre-trained pre-distortion model includes multiple expert networks, multiple gating networks, and multiple private networks.

[0081] Expert Networks: The underlying structure shared in the pre-trained pre-distortion model is divided into multiple expert networks, which are relatively independent of each other. Multiple expert networks all adopt a fully-connected network structure, including an expert input layer, an expert hidden layer, and an expert output layer. Among them, the activation functions of the expert hidden layer and the expert output layer both adopt the Tanh function. Multiple expert networks have the same input data and the same structure, but during training, the network parameters of each expert network are randomly initialized, so that the results cannot approach a single result, and different data views of the input data can be learned, solving the problem that a single network of the same scale cannot effectively capture the common features shared in all tasks. By dividing the structure into multiple expert networks, each expert network can learn the representation of a specific task, thereby capturing the unique relevant features in a single task.

[0082] Gating Networks: There is a gating network corresponding to each power amplifier learning task. Therefore, the number of gating networks is equal to the number of tasks. Among them, the gating network is usually lightweight and only contains one network layer, and its activation function is the Softmax function. The input data is the same as that of the expert network, and the input data is the constructed training set. The output data is a probability distribution, which represents the weights of each expert network in a certain power amplifier learning task.

[0083] Private Networks: The output data of the expert network and the output data of the gating network are weighted and used as the input of the private network. Since the private network is used to extract the private features of each power amplifier learning task, the number of private networks is equal to the number of tasks, and all adopt a fully-connected architecture, including a private input layer, a private hidden layer, and a private output layer. Among them, the activation functions of the private hidden layer and the private output layer both adopt the Leaky ReLU function. The private network combines the common features of multiple tasks and the private features obtained for specific tasks to obtain the model output, and its output corresponds to the historical target I / Q components of the predicted power amplifier input.

[0084] Step S230 specifically includes: integrating the trained pre-distortion model into a multi-antenna system, and its output is connected to the power amplifiers of multiple antennas.

[0085] The trained pre-distortion model is combined with the power amplifier in a multi-antenna satellite communication system to pre-compensate the nonlinearity of multiple power amplifiers.

[0086] After that, when the input signal of the multi-antenna flows to the predistortion model, the input signal at each moment in the current processing period to be processed is stored until the memory length is full, and it is shaped into the format of concatenating the I / Q components and signal amplitudes at each moment in the processing period corresponding to the memory length, that is, the concatenated vector, to be applicable to the input of the predistortion model; after the concatenated vector is transmitted to the predistortion model for processing, the target I / Q components of each antenna can be obtained. Specifically, it can be: using the first number of expert networks in the predistortion model to process the concatenated vector respectively to obtain the common feature vectors corresponding to each expert network; and using the second number of gating networks in the predistortion model to process the concatenated vector to obtain the weights of each expert network; using the second number of private networks in the predistortion model to process the weights of each expert network and the common feature vectors corresponding to each expert network to obtain the target I / Q components of each antenna.

[0087] In this method, the predistortion model performs pre-compensation on the received input signal, outputs the target I / Q components corresponding to the input signal of the multi-antenna power amplifier, and converts the target I / Q components of each antenna into the form of complex signals, and transmits them to the power amplifiers of each antenna for corresponding processing.

[0088] Regarding the linearization compensation of multiple power amplifiers as a multi-task problem, a predistortion model is constructed using a multi-gated mixture of experts network, and only one DPD module is used to perform linearization processing on multiple power amplifiers simultaneously, so as to ensure the linearity of the omnidirectional beam while restricting a large increase in hardware complexity. Through the efficient sharing and independent learning between multiple tasks, the linearization performance of multiple power amplifiers can be effectively guaranteed, and the non-linear crosstalk problem between multiple antennas can be alleviated.

[0089] A digital predistortion method for a multi-antenna satellite system provided by the present application is based on a multi-feedback loop architecture, and uses a multi-gated mixture of experts network to model the predistortion model of the multi-antenna system. Among them, the multi-gated mixture of experts network uses expert networks and private networks to learn common features and private features, and uses gating networks to adaptively select features for different tasks. By automatically adjusting the parameterization between shared information and specific task information, the predistortion model can effectively learn complex data distributions, improve linearization performance, and reduce non-linear crosstalk across channels. Compared with the existing neural network-based predistortion models, the predistortion model of the present application can obtain a more accurate power amplifier inverse model under the same number of training iterations, so as to effectively compensate for the distortion and crosstalk of multiple power amplifiers. Compared with the existing polynomial machine learning methods, the predistortion model has more excellent performance in both non-linear distortion compensation and crosstalk suppression.

[0090] Corresponding to the above method, an embodiment of the present application also provides a digital predistortion device for a multi-antenna satellite system, asFigure 5 As shown in the figure, the device includes:

[0091] An acquisition unit 510, configured to acquire input data of multiple antennas to be processed; the input data includes I / Q components and signal amplitudes of input signals;

[0092] A splicing unit 520, configured to splice the I / Q components and signal amplitudes of multiple antennas to obtain a spliced vector;

[0093] A processing unit 530, configured to process the spliced vector by using a trained predistortion model to obtain target I / Q components of each antenna, and convert the target I / Q components into predistortion signals, so that power amplifiers corresponding to the antennas process the corresponding predistortion signals and then transmit them; wherein, the predistortion model includes a first number of expert networks, a second number of gating networks and a second number of private networks.

[0094] The functions of the functional units of a digital predistortion device for a multi-antenna satellite system provided in the above embodiments of the present application can be implemented by the above method steps. Therefore, the specific working processes and beneficial effects of each unit in the digital predistortion device for a multi-antenna satellite system provided in the embodiments of the present application will not be elaborated herein.

[0095] The embodiments of the present application further provide an electronic device, as Figure 6 shown, including a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640.

[0096] The memory 630 is used for storing a computer program;

[0097] When the processor 610 is configured to execute the program stored in the memory 630, the following steps are implemented:

[0098] Acquire input data of multiple antennas to be processed; the input data includes I / Q components and signal amplitudes of input signals;

[0099] Splice the I / Q components and signal amplitudes of multiple antennas to obtain a spliced vector;

[0100] Process the spliced vector by using a trained predistortion model to obtain target I / Q components of each antenna, and convert the target I / Q components into predistortion signals, so that power amplifiers corresponding to the antennas process the corresponding predistortion signals and then transmit them; wherein, the predistortion model includes a first number of expert networks, a second number of gating networks and a second number of private networks.

[0101] The communication bus mentioned above may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, only a thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0102] The communication interface is used for communication between the above-mentioned electronic device and other devices.

[0103] The memory may include a Random Access Memory (RAM), or may also include a Non-Volatile Memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located far from the aforementioned processor.

[0104] The above-mentioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processing (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0105] Since the implementation manners and beneficial effects of the various devices of the electronic device in the above embodiments can be realized by referring to the steps in the embodiments shown in Figure 2 Therefore, the specific working process and beneficial effects of the electronic device provided in the embodiments of the present application will not be repeated here.

[0106] In another embodiment provided by the present application, a computer-readable storage medium is also provided. Instructions are stored in the computer-readable storage medium. When it runs on a computer, it causes the computer to execute the digital predistortion method of a multi-antenna satellite system described in any one of the above embodiments.

[0107] In another embodiment provided by the present application, there is also provided a computer program product containing instructions. When it runs on a computer, it causes the computer to execute the digital pre-distortion method of a multi-antenna satellite system described in any one of the above embodiments.

[0108] Those skilled in the art should understand that the embodiments in the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the embodiments in the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments in the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0109] The embodiments in the present application are described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments in the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and 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 the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0110] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0112] Unless otherwise defined, the technical terms or scientific terms used in this application shall have the ordinary meanings as understood by those of ordinary skill in the art to which this invention pertains. The terms "first", "second" and similar terms used in this application do not denote any order, quantity or importance, but are only used to distinguish different components. Words such as "comprising" or "including" mean that the elements or items appearing before this word cover the elements or items listed after this word and their equivalents, without excluding other elements or items. Words such as "connected", "coupled" or "linked" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Upper", "lower", "left", "right", etc. are only used to indicate relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0113] Although the preferred embodiments in the embodiments of this application have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concepts. Therefore, the embodiments of this application are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of this application.

[0114] Obviously, those skilled in the art can make various changes and modifications to the embodiments in the embodiments of this application without departing from the spirit and scope of the embodiments in the embodiments of this application. Thus, if these modifications and variations of the embodiments in the embodiments of this application fall within the scope of the embodiments of this application and their equivalent technologies, the embodiments of this application also intend to include these changes and modifications therein.

Claims

1. A digital predistortion method for a multi-antenna satellite system, characterized in that, The method includes: Obtaining input data of multiple antennas to be processed; the input data includes I / Q components and signal amplitudes of input signals; Concatenating the I / Q components and signal amplitudes of the multiple antennas to obtain a concatenated vector; Using a trained predistortion model to process the concatenated vector to obtain target I / Q components of each antenna, and converting the target I / Q components into predistortion signals, so that power amplifiers corresponding to the antennas process the corresponding predistortion signals and then transmit them; wherein, the predistortion model includes a first number of expert networks, a second number of gating networks, and a second number of private networks; Wherein, using a trained predistortion model to process the concatenated vector to obtain target I / Q components of each antenna includes: Using the first number of expert networks in the predistortion model to process the concatenated vector respectively to obtain common feature vectors corresponding to the expert networks; And, using the second number of gating networks in the predistortion model to process the concatenated vector to obtain weights of the expert networks; Using the second number of private networks in the predistortion model to process the weights of the expert networks and the common feature vectors corresponding to the expert networks to obtain target I / Q components of each antenna.

2. The method according to claim 1, characterized in that, For any one of the expert networks, the expert network has a fully connected network structure, including an expert input layer, an expert hidden layer, and an expert output layer; The activation functions of the expert hidden layer and the expert output layer are both Tanh functions.

3. The method according to claim 1, characterized in that, For any one of the gating networks, the gating network only includes a network layer; The activation function of the network layer is a Softmax function.

4. The method according to claim 1, wherein The private network has a fully connected network structure, including a private input layer, a private hidden layer, and a private output layer; The activation functions of the private hidden layer and the private output layer are both Leaky ReLU functions.

5. A digital predistortion device for a multi-antenna satellite system, characterized in that, The apparatus includes: An acquisition unit, configured to acquire input data of multiple antennas to be processed; the input data includes I / Q components and signal amplitudes of input signals; A concatenation unit, configured to concatenate the I / Q components and signal amplitudes of the multiple antennas to obtain a concatenated vector; A processing unit, configured to use a trained predistortion model to process the concatenated vector to obtain target I / Q components of each antenna, and convert the target I / Q components into predistortion signals, so that power amplifiers corresponding to the antennas process the corresponding predistortion signals and then transmit them; wherein, the predistortion model includes a first number of expert networks, a second number of gating networks, and a second number of private networks; Wherein, using a trained predistortion model to process the concatenated vector to obtain target I / Q components of each antenna includes: Using the first number of expert networks in the predistortion model to process the concatenated vector respectively to obtain common feature vectors corresponding to the expert networks; And, using the second number of gating networks in the predistortion model to process the concatenated vector to obtain weights of the expert networks; Process the weights of each expert network and the corresponding common feature vectors of each expert network by using the second quantity of private networks in the pre-distortion model to obtain the target I / Q components of each antenna.

6. The device according to claim 5, characterized in that For any expert network, the expert network has a fully connected network structure, including an expert input layer, an expert hidden layer, and an expert output layer. The activation functions of the expert hidden layer and the expert output layer are both Tanh functions.

7. The device according to claim 5, characterized in that, For any gating network, the gating network only includes a network layer. The activation function of the network layer is a Softmax function.

8. An electronic device, characterized in that, The electronic device includes a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus. The memory is used to store a computer program. When the processor is used to execute the program stored on the memory, it implements the method steps described in any one of claims 1-4.

Images