Training method of digital pre-distortion model and pre-distortion compensation method of signal

Abstract

The embodiment of the invention provides a training method of a digital pre-distortion model and a pre-distortion compensation method of a signal. The method comprises the steps that training data of a power amplifier under different working conditions are obtained, the training data are collected multi-frame signals, each frame signal comprises a plurality of initial signals, for any working condition, the initial signals included in each frame are preprocessed, and first signals included in each frame are obtained; and performing power back-off processing on the first signal included in each frame to obtain a second signal included in each frame, training the initial neural network model according to the second signal included in each frame until a convergence condition is reached, and obtaining a target digital pre-distortion model trained under a working condition. According to the method, the limitation of a fixed threshold value is broken through, accurate adaptation can be achieved for different working conditions, the training data covers multiple working conditions, the trained model better fits the actual working characteristics of the power amplifier, the nonlinear influence of the power amplifier can be offset more accurately, and therefore the compensation accuracy is greatly improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a training method for a digital predistortion model and a signal predistortion compensation method. Background Technology

[0002] In satellite communication systems, the power amplifier (PA) is a core component of the transmitter, its main function being to amplify the baseband signal to a sufficiently high power to meet the requirements of long-distance transmission. However, the nonlinear characteristics of the PA are one of the main problems affecting signal quality.

[0003] In related technologies, when compensating for the signal output of a power amplifier, a fixed power compensation threshold is usually set based on the distortion under rated power. When the power of the signal output by the PA is within the range of this threshold, the hardware circuit performs pre-compensation according to preset parameters, thereby reducing the nonlinear effect of the PA.

[0004] However, the above method has poor compensation accuracy for the power amplifier output signal. Summary of the Invention

[0005] This application provides a training method for a digital predistortion model and a signal predistortion compensation method to achieve the technical effect of improving the compensation accuracy of the power amplifier output signal.

[0006] In a first aspect, embodiments of this application provide a training method for a digital predistortion model, comprising:

[0007] Acquire training data of the power amplifier under different operating conditions. The training data consists of multiple frames of collected signals, each frame of which includes multiple initial signals.

[0008] For any given operating condition, perform the following operations to obtain the corresponding target digital predistortion model. The following operations include:

[0009] The initial signal included in each frame is preprocessed to obtain the first signal included in each frame. The first signal included in each frame is then subjected to power back-off processing to obtain the second signal included in each frame. Based on the second signal included in each frame, the initial neural network model is trained until the convergence condition is reached, thereby obtaining the target digital predistortion model trained under the stated operating condition.

[0010] In one possible implementation, the first signal includes a pilot signal;

[0011] The power back-off processing of the preprocessed first signal of each frame yields the second signal included in each frame, including:

[0012] For any frame, obtain the total number of signals included in the first signal, and the target position information of the pilot signal in the first signal;

[0013] The target location information, the preset power backoff value, and the total number of signals are input into the preset power backoff algorithm for power backoff processing to determine the second signal included in the frame.

[0014] In one possible implementation, training the initial neural network model based on the second signal included in each frame until a convergence condition is met, to obtain the trained target digital predistortion model under the specified working condition, includes:

[0015] Obtain the third signal preset under the initial operating conditions of the power amplifier;

[0016] Based on the third signal, the initial neural network model is initially trained offline until the convergence condition is met, and the first digital predistortion model after training is obtained.

[0017] Based on the second signal included in each frame, the first digital predistortion model is retrained in the online state until the convergence condition is met, thereby obtaining the target digital predistortion model trained under the stated working condition.

[0018] In one possible implementation, the first digital predistortion model includes an adaptation layer and a migration layer;

[0019] The step of retraining the first digital predistortion model in online mode based on the second signal included in each frame until the convergence condition is met, to obtain the target digital predistortion model trained under the specified working condition, includes:

[0020] Extract the feature information corresponding to the second signal included in each frame;

[0021] Based on the feature information corresponding to each frame, the target operating condition of the power amplifier corresponding to the second signal included in each frame is determined;

[0022] If the target working condition is different from the initial working condition, the hyperparameters of the adaptation layer in the first digital predistortion model are randomly initialized, while the hyperparameters of the transfer layer in the first digital predistortion model are kept, to obtain the adjusted second digital predistortion model.

[0023] The second signal included in each frame is divided into two paths, and the second digital predistortion model is retrained until the convergence condition is met, so as to obtain the target digital predistortion model trained under the said working condition.

[0024] In one possible implementation, the step of splitting the second signal included in each frame into two paths and retraining the second digital predistortion model until the convergence condition is met, to obtain the target digital predistortion model trained under the stated operating condition, includes:

[0025] The second signal included in each frame is divided into two paths, and the second digital predistortion model is iteratively calculated according to the preset loss function and the preset update rule. When the preset loss function meets the preset condition or the number of iterations reaches the preset number of iterations threshold, the target digital predistortion model trained under the working condition is obtained.

[0026] The preset loss function is determined by a preset metric parameter that affects the nonlinearity of the power amplifier.

[0027] In one possible implementation, it also includes:

[0028] For any given working condition, determine the current hyperparameters of the adaptation layer and the migration layer in the target digital predistortion model based on the target digital predistortion model corresponding to the working condition;

[0029] The signal power is determined based on the second signal included in each frame corresponding to the operating condition;

[0030] Based on the signal power and the current hyperparameters, establish a mapping relationship between the signal power and the current hyperparameters;

[0031] The mapping relationship is stored in a preset lookup table, which stores the mapping relationship between different signal powers and hyperparameters.

[0032] Secondly, embodiments of this application provide a signal pre-distortion compensation method, comprising:

[0033] Obtain the fourth signal included in each frame after power back-off processing;

[0034] The target signal power of the fourth signal is determined based on the fourth signal included in each frame;

[0035] Based on the target signal power, the target hyperparameters that map to the target signal power are determined from the lookup table;

[0036] The current hyperparameters in the target digital predistortion model are replaced with the target hyperparameters to obtain the adjusted digital predistortion model, wherein the target digital predistortion model is the target digital predistortion model trained by the method described in any of the first aspects;

[0037] The fourth signal included in each frame is input into the adjusted digital predistortion model to obtain the compensation signal of the fourth signal included in each frame;

[0038] The fourth signal included in each frame is pre-compensated according to the compensation signal to obtain the fifth signal included in each frame after pre-compensation processing.

[0039] In one possible implementation, acquiring the fourth signal included in each frame after power back-off processing includes:

[0040] Acquire the currently acquired multi-frame signal;

[0041] The multi-frame signals are preprocessed to obtain the sixth signal included in each frame;

[0042] The sixth signal included in each frame is processed by power back-off to obtain the fourth signal included in each frame.

[0043] Thirdly, embodiments of this application provide a training apparatus for a digital predistortion model, comprising:

[0044] The first acquisition module is used to acquire training data of the power amplifier under different operating conditions. The training data is a multi-frame signal, and each frame signal includes multiple initial signals.

[0045] The first processing module is used to perform the following operations for any given working condition to obtain the corresponding target digital predistortion model. The following operations include:

[0046] The initial signal included in each frame is preprocessed to obtain the first signal included in each frame. The first signal included in each frame is then subjected to power back-off processing to obtain the second signal included in each frame. Based on the second signal included in each frame, the initial neural network model is trained until the convergence condition is reached, thereby obtaining the target digital predistortion model trained under the stated operating condition.

[0047] Fourthly, embodiments of this application provide a signal pre-distortion compensation device, comprising:

[0048] The second acquisition module is used to acquire the fourth signal included in each frame after power back-off processing;

[0049] The determining module is used to determine the target signal power of the fourth signal based on the fourth signal included in each frame;

[0050] The determining module is further configured to determine, from a lookup table, a target hyperparameter that maps to the target signal power based on the target signal power;

[0051] The second processing module is used to replace the current hyperparameters in the target digital predistortion model with the target hyperparameters to obtain the adjusted digital predistortion model, wherein the target digital predistortion model is the target digital predistortion model trained by the method described in any of the first aspects.

[0052] The second processing module is further configured to input the fourth signal included in each frame into the adjusted digital predistortion model to obtain the compensation signal of the fourth signal included in each frame;

[0053] The second processing module is further configured to perform pre-compensation processing on the fourth signal included in each frame according to the compensation signal, so as to obtain the fifth signal included in each frame after pre-compensation processing.

[0054] Fifthly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0055] The memory stores computer-executed instructions;

[0056] The processor executes computer execution instructions stored in the memory, causing the processor to perform various possible implementations of the first and / or second aspects described above.

[0057] In a sixth aspect, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement various possible implementations of the first and / or second aspects described above.

[0058] In a seventh aspect, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements various possible implementations of the first and / or second aspects described above.

[0059] The digital predistortion model training method and signal predistortion compensation method provided in this application acquire training data of a power amplifier under different operating conditions. The training data consists of multiple frames of acquired signals, each frame containing multiple initial signals. For any given operating condition, the initial signals in each frame are preprocessed to obtain a first signal. Power back-off processing is then performed on the first signals to obtain a second signal. Based on the second signals, the initial neural network model is trained until convergence is achieved, resulting in the target digital predistortion model trained for that operating condition. This method acquires multiple frames of signals from the power amplifier under different operating conditions as training data. The initial signals in each frame under each operating condition are preprocessed to obtain a first signal, and then power back-off processing is performed to generate a second signal. This process is used to train the initial neural network model until convergence, resulting in the target digital predistortion model for the corresponding operating condition. This approach overcomes the limitations of fixed thresholds and can accurately adapt to different operating conditions. Furthermore, the training data covers multiple operating conditions, making the model more closely reflect the actual operating characteristics of the power amplifier and more accurately offsetting its nonlinear effects, thereby significantly improving compensation accuracy. Attached Figure Description

[0060] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0061] Figure 1 This is a schematic diagram of an application scenario provided by an embodiment of this application;

[0062] Figure 2 A flowchart illustrating a training method for a digital predistortion model provided in an embodiment of this application;

[0063] Figure 3 A schematic diagram of a model parameter training architecture provided for an embodiment of this application;

[0064] Figure 4 This is a schematic diagram illustrating the composition of a first signal structure included in a frame of signal, as provided in an embodiment of this application.

[0065] Figure 5 A flowchart illustrating a method for power back-off processing of a first signal provided in an embodiment of this application;

[0066] Figure 6 This is a schematic diagram of the structure of an initial neural network model provided in an embodiment of this application;

[0067] Figure 7 A flowchart illustrating a method for obtaining a target digital predistortion model, provided in an embodiment of this application;

[0068] Figure 8A A schematic flowchart illustrating a signal pre-distortion compensation method provided in an embodiment of this application;

[0069] Figure 8B A schematic diagram illustrating a pre-distortion compensation method provided in an embodiment of this application;

[0070] Figure 9 A schematic diagram of the structure of a training device for a digital predistortion model provided in an embodiment of this application;

[0071] Figure 10 A schematic diagram of the structure of a signal pre-distortion compensation device provided in an embodiment of this application;

[0072] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0073] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0074] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0075] In wireless communication and radio frequency transmission systems, power amplifiers are the core components that determine the quality of signal transmission. However, their nonlinear characteristics can lead to problems such as harmonic distortion and adjacent channel power leakage in the output signal, which seriously affect communication performance.

[0076] Achieving high-quality, high-efficiency broadband waveform transmission on satellite is an inevitable trend in satellite communication systems. The limitations on the waveform quality at the transmitter output of a satellite largely stem from the influence of the nonlinear characteristics of the power amplifier. As a typical nonlinear device, the power amplifier inherently presents a trade-off between efficiency and linearity. Operating in the high-efficiency region results in significant linear distortion, while maintaining high linearity means a decrease in power amplifier efficiency and output power. To improve linearity, power amplifier efficiency, and output power, pre-compensation of the power amplifier's output signal is necessary.

[0077] In related technologies, pre-compensation of the power amplifier output signal is based on the nonlinear distortion at the PA's rated power. First, the PA is adjusted to its rated power condition, and distortion data such as harmonic distortion are collected using a signal analyzer to determine the degree of distortion under this condition. Then, a fixed power compensation threshold is preset based on this, typically taking a certain percentage fluctuation range centered on the rated power. Simultaneously, compensation parameters are preset in the hardware circuit based on the distortion data at the rated power. During actual PA operation, a hardware power detection module, such as a power sensor or detector, collects its output signal power in real time and compares it with the power compensation threshold. If the output signal power is within the power compensation threshold, the hardware circuit automatically triggers the setting of compensation parameters, performing amplitude / phase pre-compensation on the signal input to the PA before inputting the adjusted signal back into the PA. This pre-compensation reduces the PA's nonlinear effects through the reverse distortion of the distortion.

[0078] However, the above methods only consider a single rated power operating condition and set fixed power compensation thresholds and parameters. The nonlinear distortion characteristics of a power amplifier (PA) change dynamically with varying operating conditions, and the fixed power compensation thresholds and parameters in related technologies cannot respond to these changes. When the actual operating condition of the PA deviates from its rated value, the preset reverse distortion deviates significantly from the actual distortion, leading to a decrease in compensation accuracy.

[0079] Therefore, addressing the aforementioned problems in related technologies, the research process revealed that by acquiring training data of a power amplifier under different operating conditions—specifically, multi-frame signals collected, each frame containing multiple initial signals—the nonlinear characteristics of the PA under different operating conditions can be covered. Preprocessing the initial signal for any operating condition removes interference signals to ensure signal quality. By introducing power back-off processing, it can adapt to distortion differences in different power ranges. Then, using the processed second signal specific to each operating condition, an initial neural network model is trained to obtain a target predistortion model specific to each operating condition, thereby improving compensation accuracy and adaptation flexibility. Based on this, this application proposes a training method for a digital predistortion model and a signal predistortion compensation method.

[0080] To facilitate understanding of the methods in this application, the following exemplary application scenarios are provided. Please refer to [link / reference]. Figure 1 , Figure 1 This is a schematic diagram of an application scenario provided by an embodiment of the present application, including a satellite 01 and a ground terminal 02.

[0081] Satellite 01 is equipped with an onboard phased array transmitter, which includes a power amplifier to amplify the weak signals emitted by the satellite to meet the required signal strength.

[0082] Since the baseband signal generated by Satellite 01 is transmitted only within the internal circuitry of Satellite 01 and does not enter the space-to-ground communication link, and Satellite 01 and Ground Terminal 02 are independent signal systems, Ground Terminal 02 lacks the physical path and access permission to obtain this internal signal. Therefore, Ground Terminal 02 cannot directly acquire it. Thus, Ground Terminal 02 simulates and generates the baseband signal of the satellite power amplifier under different operating conditions, and uses this as training data to train the digital predistortion model.

[0083] In this application scenario, it could be a scenario where nonlinear pre-compensation is performed on the output signal of the power amplifier in a self-loop-free satellite-borne phased array transmitter.

[0084] It is understood that this application does not limit the number, type, or function of satellite 01 and ground terminal 02. The above application scenarios are only for illustrative purposes and can be determined according to the actual application situation.

[0085] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0086] Please see Figure 2 , Figure 2 This is a flowchart illustrating a training method for a digital predistortion model provided in this application embodiment. The execution entity of this method can be a training device for the digital predistortion model, which can be implemented through a computer program; it can also be implemented through a medium storing the relevant computer program, such as a USB flash drive and / or optical disc; or it can be implemented through a physical device integrating or installing the relevant computer program, such as a chip or electronic device. The electronic device can be a server, server cluster, smart terminal, etc. The method may include the following steps:

[0087] S201. Acquire training data of the power amplifier under different operating conditions. The training data consists of multiple frames of collected signals, each frame of which includes multiple initial signals.

[0088] Optionally, the execution entity can be a ground terminal, taking a server as an example, which utilizes the server's powerful computing capabilities to train the model.

[0089] In this embodiment, model parameters are trained using the entire satellite-to-ground communication link. To facilitate understanding of this application, the communication between the satellite and the ground terminal is first described, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of an architecture for training model parameters provided in an embodiment of this application.

[0090] After the satellite generates a signal, it passes through a preprocessing module and then reaches the phased array transmission module. The power amplifier in the phased array transmission module amplifies the signal and then transmits the amplified signal to the ground terminal through a low-orbit satellite channel, so as to utilize the computing resources of the ground terminal for model training.

[0091] For the signal input to the power amplifier in the phased array transmitter module Since this signal is a baseband signal generated at the satellite end, it cannot be directly obtained on the ground. Therefore, it is chosen to simulate the same signal at the ground terminal, that is, to generate a set of identical reference signals.

[0092] Optionally, such as Figure 3 As shown, after the ground terminal simulates the signal generated by the satellite, it undergoes the same steps as the satellite signal processing, namely, the same encoding, modulation, and preprocessing processes, to obtain the signal. .

[0093] Therefore, in this embodiment, the signal generated by the satellite is simulated by the signal generated by the ground terminal.

[0094] The ground terminal acquires training data of the power amplifier under different operating conditions by simulating the power amplifier's operation. The training data consists of multiple frames of acquired signals, each frame containing multiple initial signals. .

[0095] Alternatively, different operating conditions could include different operating states of the power amplifier, fluctuations in ambient temperature, and differences in signal type.

[0096] S202. For any operating condition, preprocess the initial signal included in each frame to obtain the first signal included in each frame.

[0097] For any one of the multiple operating conditions, based on any frame of that operating condition, the initial signal included in that frame is preprocessed to obtain the first signal corresponding to that frame after preprocessing.

[0098] Optionally, preprocessing includes, but is not limited to, noise filtering, signal synchronization, and outlier removal.

[0099] Preprocessing can reduce noise, electromagnetic interference, and abnormal distortion in the signal, calibrate timing and phase deviations, reduce the accuracy of subsequent power back-off processing, provide a unified timing and phase reference for signal processing under different operating conditions, improve the stability of the model's learning of PA nonlinear features, and thus ensure the compensation accuracy of the target DPD model.

[0100] S203. Perform power back-off processing on the first signal included in each frame to obtain the second signal included in each frame.

[0101] Because a low-Earth orbit satellite channel (LEO) is used as the feedback loop, the signal processed by the power amplifier in the satellite needs to pass through the LEO link before reaching the ground terminal. However, the non-ideal factors of the LEO link and receiver noise can severely affect the accuracy of the parameter extraction results. Therefore, it is necessary to process the signal transmitted through the LEO link to recover the signal processed by the power amplifier as much as possible. .

[0102] In this embodiment, since the satellite-generated signal is simulated by a ground terminal, it is necessary to perform power back-off processing on the first signal included in each frame under different operating conditions after preprocessing, so as to obtain the second signal included in each frame under different operating conditions.

[0103] S204. Based on the second signal included in each frame, train the initial neural network model until the convergence condition is met, and obtain the target digital predistortion model trained under this working condition.

[0104] After obtaining the second signal included in each frame under different working conditions through the above steps, the initial neural network model is trained until the convergence condition is met, thereby obtaining the target digital predistortion model trained under different working conditions.

[0105] Among them, the target digital predistortion model trained under different working conditions has different model parameters, which can be multiple hyperparameters of the model during the training process.

[0106] In the above embodiments of this application, training data of the power amplifier under different operating conditions is acquired. The training data consists of multiple frames of collected signals, each frame including multiple initial signals. For any operating condition, the initial signals in each frame are preprocessed to obtain a first signal, and power back-off processing is performed on the first signals to obtain a second signal. Then, based on the second signals, the initial neural network model is trained until convergence is achieved, resulting in the target digital predistortion model trained for the corresponding operating condition. This method acquires multiple frames of signals from the power amplifier under different operating conditions as training data, preprocesses the initial signals of each frame under each operating condition to obtain a first signal, and then performs power back-off processing to generate a second signal. This is used to train the initial neural network model until convergence, resulting in the target digital predistortion model for the corresponding operating condition. This approach overcomes the limitations of fixed thresholds, can accurately adapt to different operating conditions, and the training data covers multiple operating conditions. The trained model better matches the actual operating characteristics of the power amplifier, more accurately offsetting its nonlinear effects, thereby significantly improving compensation accuracy.

[0107] Furthermore, based on the above embodiments, the following embodiments illustrate the process of performing power back-off processing on the preprocessed first signal of each frame to obtain the second signal included in each frame.

[0108] In this embodiment, a partial power back-off method is used to perform power back-off processing on the first signal. The following will combine... Figure 4 This embodiment will be described below. Figure 4 This is a schematic diagram illustrating the composition of a first signal structure included in a frame of signal, as provided in an embodiment of this application. Figure 4 The Z-axis represents power, the Y-axis represents carrier frequency, and the X-axis represents a frame of signal at a corresponding time. A frame of signal includes a first signal, which in turn includes pilot signals and digital signals. The pilot signal is as follows: Figure 4 As shown by the black bar in the image, the digital signal is as follows: Figure 4 As shown in the white bar.

[0109] To address the impact of non-ideal factors in low-Earth orbit satellite channels, channel estimation and compensation are performed by adding pilot signals to the signal. However, if the first signal is amplified in the nonlinear region, the pilot sequence itself will also be affected by the power amplifier's nonlinearity. This not only reduces the accuracy of channel estimation but may also introduce additional distortion during channel estimation and signal detection, leading to a complete deterioration of the model parameter training results.

[0110] Therefore, in this embodiment, a power back-off method is used to amplify the pilot signal in the linear region. The first signal is then compensated at the ground terminal using the channel matrix estimated based on the pilot signal, thereby recovering the signal processed by the power amplifier as much as possible.

[0111] Specifically, based on the needs of actual engineering applications, the pilot signal within the first signal can be backed up in terms of power. At the same time, considering the influence of power amplifier memory effect, in order to reduce the influence of adjacent high-power signals on the pilot signal, the scheme will back up the power of the two digital signals adjacent to the pilot signal to a certain extent.

[0112] Please see Figure 5 , Figure 5 This application provides a flowchart illustrating a method for power back-off processing of a first signal, which may include the following steps:

[0113] S501. For any frame, obtain the total number of signals included in the first signal, and the target position information of the pilot signal in the first signal.

[0114] For any given frame, the total number of signals included in the first signal corresponding to that frame is obtained, and the target position information of the pilot signal in the first signal is determined. It also includes obtaining a preset power back-off value and a preset memory depth factor.

[0115] S502. Input the target location information, the preset power backoff value, and the total number of signals into the preset power backoff algorithm to perform power backoff processing and determine the second signal included in the frame.

[0116] The second signal included in the frame after power back-off is determined by the following formula (1):

[0117] (1)

[0118] in, This indicates the second signal after power back-off processing; This indicates the preset power backoff value; Indicates the total number of signals; Indicates the depth of memory factor; This indicates the target position information of the pilot signal in the first signal.

[0119] like Figure 4 As shown, in Figure 4 middle , , The target position information of the pilot signal in the first signal is the fourth one, as shown by the black bar.

[0120] In this application, considering the impact of noise on the accuracy of subsequent model training, a method of signal accumulation by sending repeated sequences can be adopted to address the deterioration of the signal-to-noise ratio. Specifically, when generating the signal, the sequence is repeated periodically with one or several symbols within a frame, and the number of repetitions is determined based on the actual signal-to-noise ratio conditions achievable by the system. At the ground terminal, the accumulated repeated signals are averaged to reduce the impact of noise on the training results to some extent. It should be noted that as the number of accumulations increases, the noise resistance of the recovered signal becomes stronger, but the time resources required also increase. To ensure efficiency, the repetition period needs to be reduced, meaning the length of the signal sequence actually used for new training of model parameters will decrease. The reduction in signal sequence length will decrease the accuracy of the training results to some extent, but as long as the amplitude range of the transmitted signal is within the preset amplitude range in practical applications, the decrease in accuracy caused by the reduction in signal sequence length is almost negligible.

[0121] In the above embodiments of this application, by combining the total number of first signals, the target position information of pilot signals, and a preset power back-off value, and performing power back-off processing through a preset algorithm, the pilot signals can be accurately located. Furthermore, it can specifically cover different power ranges, allowing the subsequently trained digital predistortion model to better fit the nonlinear characteristics of each power range of the power amplifier, improving the model's adaptability to multiple operating conditions, and thus enhancing the compensation accuracy for the nonlinear distortion of the power amplifier.

[0122] Furthermore, based on any of the above embodiments, the following embodiments illustrate the process of training an initial neural network model according to the second signal included in each frame until the convergence condition is met, thereby obtaining the target digital predistortion model trained under the working condition.

[0123] In this embodiment, the structure of the initial neural network model will first be described, as follows: Figure 6 As shown, Figure 6 This is a schematic diagram of the structure of an initial neural network model provided in an embodiment of this application, including an input layer, a hidden layer, and an output layer.

[0124] Since the second signal is generally a complex signal, it can be divided into two paths, I and Q, when entering the input layer, with the real part data as one path. Imaginary data as a path By splitting the input complex signal into I and Q channels for processing, the computational complexity of the neural network can be reduced, as well as the training time and computational resource requirements. Furthermore, the hardware requirements are relatively low when subsequently deployed to a real-world system.

[0125] Before entering the input layer, the I and Q signals each pass through a set of M-order delay circuits. This delay processing generates... , … Equal delay sequences are used to construct temporal or multi-dimensional features of signals. Through this operation, the signal input at each time step contains the signal features of that time step and several previous time steps, and the entire neural network also gains the ability to describe and compensate for nonlinear distortion caused by memory effects.

[0126] After the input layer, the data enters multiple hidden layers, such as 3-5 hidden layers. Each hidden layer undergoes batch normalization and a Rectified Linear Unit (ReLU) activation function to introduce non-linearity, enhancing the model's ability to fit complex relationships. The final output consists of two processed paths. and .

[0127] In this embodiment, as Figure 6 As shown, the input layer and some hidden layers can also be designated as adaptation layers, while the remaining hidden layers and the output layer can be designated as transfer layers. The specific number of hidden layers in the adaptation layers and the number of hidden layers in the transfer layers can be preset according to actual application conditions; this embodiment does not impose a limitation. This layering operation is related to the transfer learning assistance process during model parameter training. In this embodiment, several layers closer to the input layer are designed as adaptation layers, and several layers closer to the output layer are designed as transfer layers because, during online transfer learning, the data in the input layer includes output signals from the power amplifier under different operating conditions, and these signals change with the power amplifier's operating conditions. The data in the output layer consists of signals input to the power amplifier, which are relatively stable. Therefore, it is more reasonable to designate several layers closer to the input layer as adaptation layers, and designing several layers closer to the output layer as transfer layers to transfer the initial hyperparameter values ​​under new operating conditions will be more efficient.

[0128] Furthermore, based on the structure of the initial neural network model described above, the process of obtaining the target digital predistortion model is explained. Please refer to [link to relevant documentation]. Figure 7 , Figure 7 A flowchart illustrating a method for obtaining a target digital predistortion model, provided in this application embodiment, may include the following steps:

[0129] S701, Obtain the preset third signal when the power amplifier is in its initial operating condition.

[0130] In this embodiment, considering that the nonlinear characteristics of the power amplifier change slowly, the nonlinear characteristics of signals at two adjacent time points are quite similar. Therefore, this embodiment divides the training process of the model-driven neural network model into two stages: offline learning and transfer learning. This reduces the need to update all hyperparameters corresponding to each layer, thereby improving the iteration speed and reducing the amount of data required for training.

[0131] S702. Based on the third signal, the initial neural network model is initially trained offline until the convergence condition is met, and the first digital predistortion model after training is obtained.

[0132] During the offline learning phase, a preset third signal is acquired under the initial operating conditions of the power amplifier. These initial conditions are the pre-defined operating conditions of the power amplifier. By initializing the neurons corresponding to each layer, the basic hyperparameters (baseline hyperparameters) of each layer are trained from scratch. This step may require a relatively large amount of training data and a large number of third signals; however, for a phased array transmitter system, only one offline learning session is needed. Furthermore, this step is performed on a ground terminal, which utilizes its relatively abundant computing resources to improve training speed.

[0133] S703. Based on the second signal included in each frame, the first digital predistortion model is retrained in the online state until the convergence condition is met, and the target digital predistortion model trained under the working condition is obtained.

[0134] One possible implementation is to extract the feature information corresponding to the second signal included in each frame, and determine the target operating condition of the power amplifier corresponding to the second signal included in each frame based on the feature information. If the target operating condition is different from the initial operating condition, the hyperparameters of the adaptation layer in the first digital predistortion model are randomly initialized, while the hyperparameters of the transfer layer in the first digital predistortion model are maintained, resulting in an adjusted second digital predistortion model. Then, the second signal included in each frame is divided into two paths, and the second digital predistortion model is retrained until the convergence condition is met, resulting in the target digital predistortion model trained under the operating condition.

[0135] During the transfer learning phase, the operation is performed when the power amplifier's operating conditions differ from the initial operating conditions. During training in the transfer learning phase, only the hyperparameters of the adaptation layer are randomly initialized, while the hyperparameters of the transfer layer can be initialized using hyperparameters obtained from offline learning. Finally, the entire neural network is trained to obtain the target digital predistortion model trained for the corresponding operating conditions.

[0136] Optionally, the process of retraining the second digital predistortion model until convergence is achieved to obtain the target digital predistortion model trained under the working condition can be as follows: the second signal included in each frame is divided into two paths, and the second digital predistortion model is iteratively calculated according to the preset loss function and the preset update rule. When the preset loss function meets the preset condition or the number of iterations reaches the preset number of iterations threshold, the target digital predistortion model trained under the working condition is obtained.

[0137] The preset loss function is determined by preset metric parameters that affect the nonlinearity of the power amplifier. These preset metric parameters can be the error vector magnitude (EVM) and the adjacent channel power ratio (ACPR).

[0138] Alternatively, the preset loss function can be as shown in the following formula (2):

[0139] (2)

[0140] in, Represents the loss function; This represents the second signal simulated by the ground terminal; This indicates the second signal generated by the satellite; Indicates the imaginary part; Indicates the real part; Indicates the first weight; Indicates the second weight; Indicates the magnitude of the error vector; This represents the power ratio difference between adjacent channels of two signals.

[0141] The design of the loss function comprehensively considers the impact of the power amplifier's nonlinear characteristics on the signal. Among other things, This is used to control the transition of the loss function from a quadratic function to a linear function. When the absolute value of the prediction error is less than or equal to... When the error is large, the loss function is expressed as mean squared error (MSE), which is sensitive to small errors. When the error is large... When the loss function is expressed as Mean Absolute Error (MAE), its sensitivity to large errors is reduced, thus minimizing the impact of outliers on the model. When using it, for... The selection of [the signal-to-noise ratio] is related to prior knowledge of the system's signal-to-noise ratio. As a baseline, when the signal-to-noise ratio is low, reduce To enhance robustness, when the signal-to-noise ratio is high, increase To improve the convergence speed.

[0142] Optionally, the preset update rules can be as shown in formulas (3) to (5) below:

[0143] (3)

[0144] (4)

[0145] (5)

[0146] in, This represents the exponentially weighted moving average of the squared gradient of the parameter at time t; This represents the decay rate, used to calculate the exponentially weighted moving average of the squared gradient; Let represent the gradient at time t; Indicates the initial learning rate; This represents a preset constant used to prevent division by zero; This represents the amount of parameter update at time t; This represents the parameter value at time t.

[0147] Based on the above, iterative calculations are performed. After each iteration, the model parameters obtained from the iterative calculations are synchronized to all data predistortion modules, and then the signal is regenerated to start the next round of iterations, until the preset loss function meets the preset conditions or the number of iterations reaches the preset iteration threshold.

[0148] In the above embodiments of this application, an initial neural network model is first trained offline based on a third signal preset for the initial operating condition. This allows for the construction of the basic model without real-time operating condition pressure, avoiding model instability caused by insufficient data in the early stages of online training. Then, the first digital predistortion model is trained online using the second signal of each frame. This allows for precise adaptation to the signal characteristics of the current operating condition, enabling the model to quickly conform to the nonlinear characteristics of the power amplifier under that condition. This phased training method ensures both the efficiency and initial stability of model training, while also improving the model's adaptation accuracy to the target operating condition.

[0149] Furthermore, based on any of the above embodiments, after training the target digital predistortion model corresponding to each working condition, for any working condition, the current hyperparameters of the adaptation layer and the transfer layer in the target digital predistortion model are determined according to the target digital predistortion model corresponding to that working condition, and the signal power is determined according to the second signal included in each frame corresponding to that working condition. Then, based on the signal power and the current hyperparameters, a mapping relationship between the signal power and the current hyperparameters is established, and this mapping relationship is stored in a preset lookup table. The lookup table stores the mapping relationships between different signal powers and hyperparameters.

[0150] The lookup table can be shown in Table 1:

[0151] Table 1

[0152]

[0153] A lookup table is used to store the model parameters (hyperparameters) of the power amplifier under different operating conditions, which can be looked up based on the signal power when needed. Simultaneously, a ground-based monitoring mechanism can be established; when signal quality deterioration is detected, the model is retrained and the lookup table is updated.

[0154] In this embodiment, the lookup table after training can be stored on the ground terminal, or the power and hyperparameters of the predistortion model under different operating conditions can be uploaded to the satellite's lookup table through the power supply circuit for satellite use.

[0155] In the above embodiments of this application, by matching a target digital predistortion model to any operating condition, the current hyperparameters of the adaptation and migration layers are determined. Then, a mapping relationship between the two is established and stored in a lookup table, based on the signal power of the second signal under the operating condition. This mapping relationship directly links the signal power and the model's hyperparameters, reducing redundant steps of repeatedly calculating hyperparameters during subsequent calls and significantly improving the efficiency of model parameter retrieval. Furthermore, the lookup table stores hyperparameters corresponding to different signal powers. When the power amplifier operating condition changes with the signal power, the appropriate hyperparameters can be quickly matched from the table without retraining the model. This ensures both rapid adaptation of the model to different power scenarios and stable maintenance of the digital predistortion model's compensation accuracy.

[0156] Please see Figure 8A , Figure 8A A flowchart illustrating a signal predistortion compensation method provided in this application embodiment is shown. The method may include the following steps:

[0157] S801. Obtain the fourth signal included in each frame after power back-off processing.

[0158] The following is combined Figure 8B The method of this embodiment will be described below. Figure 8B This is a schematic diagram of a predistortion compensation method provided in an embodiment of this application. In this embodiment, the executing entity can be a satellite or a ground terminal, and a digital predistortion model is deployed in both the satellite and the ground terminal.

[0159] The satellite acquires multiple frames of signals currently being collected, preprocesses these frames to obtain the sixth signal included in each frame, and performs power back-off processing on the sixth signal included in each frame to obtain the fourth signal included in each frame.

[0160] S802. Determine the target signal power of the fourth signal based on the fourth signal included in each frame.

[0161] Based on a preset power calculation algorithm, the corresponding target signal power is determined according to the fourth signal included in each frame.

[0162] S803. Based on the target signal power, determine the target hyperparameter that maps to the target signal power from the lookup table.

[0163] The target signal power is determined from a pre-stored lookup table, and the hyperparameters mapped to the target signal power are then determined as the target hyperparameters.

[0164] S804. Replace the current hyperparameters in the target digital predistortion model with the target hyperparameters to obtain the adjusted digital predistortion model.

[0165] The target digital predistortion model is the target digital predistortion model trained in any of the above embodiments.

[0166] S805. Input the fourth signal included in each frame into the adjusted digital predistortion model to obtain the compensation signal of the fourth signal included in each frame.

[0167] By using the adjusted digital predistortion model, the fourth signal included in each frame is input into the model, resulting in a reduced output compensation signal.

[0168] S806. Perform pre-compensation processing on the fourth signal included in each frame according to the compensation signal to obtain the fifth signal included in each frame after pre-compensation processing.

[0169] Then, based on the compensation signal, the fourth signal included in each frame is pre-compensated, and finally the fifth signal included in each frame after pre-compensation processing is obtained.

[0170] In the above embodiments of this application, by quickly matching the corresponding target hyperparameters from a lookup table based on the target signal power of the fourth signal, recalculation or retraining is unnecessary, significantly improving the efficiency of hyperparameter adjustment. The current hyperparameters of the target digital predistortion model are then replaced, enabling the model to accurately adapt to the operating conditions corresponding to the current signal power. The adjusted model generates a compensation signal and completes pre-compensation, effectively offsetting the nonlinear distortion of the power amplifier at the current power. The entire process relies on a lookup table for rapid hyperparameter retrieval, shortening the compensation response time while ensuring pre-compensation accuracy under different signal power scenarios.

[0171] The method of this application also has, but is not limited to, the following technical effects.

[0172] The method presented in this application does not require additional hardware to the satellite transmitter and can achieve power amplifier nonlinearity compensation without altering the onboard payload. This approach is suitable for upgrading existing satellite communication systems and also facilitates the design and deployment of onboard payloads for future satellite communication systems. It reduces the need for additional hardware in traditional digital predistortion techniques that rely on self-loop feedback structures, thus addressing the limitations of valuable onboard payloads and the difficulty in modifying hardware structures.

[0173] The method of this application can quickly update the model parameters as the characteristics of the power amplifier change, overcoming the shortcomings of traditional lookup table-based digital predistortion technology that makes it difficult to update parameters when the power amplifier characteristics change. It can effectively cope with the situation where the characteristics of the spaceborne power amplifier change due to the working environment and lifespan, and ensure that the digital predistortion system continuously adapts to the power amplifier status.

[0174] The method of this application is also applicable to satellite communication systems that use multi-stage power amplifiers and phased array antennas. It reduces the high system complexity and resource waste caused by separate compensation for multi-channel power amplifiers in phased arrays, and also solves the problem that it is difficult to deploy a single power amplifier predistortion scheme under the high integration and packaging of traditional phased array transmitter units, thus adapting to the application requirements of highly integrated phased array antennas.

[0175] The method in this application balances high modeling accuracy with low computational complexity through model training, resolving the contradictions of poor compensation accuracy of traditional lookup table and memory polynomial models, complex parameters and high training costs of neural network models, and adapting to the limited storage and computing resources on the satellite.

[0176] Please see Figure 9 , Figure 9 This application provides a schematic diagram of the structure of a training device for a digital predistortion model, which may include:

[0177] The first acquisition module 901 is used to acquire training data of the power amplifier under different operating conditions. The training data consists of multiple frames of signals, each of which includes multiple initial signals.

[0178] The first processing module 902 is used to perform the following operations for any given working condition to obtain the corresponding target digital predistortion model, including:

[0179] The initial signal included in each frame is preprocessed to obtain the first signal included in each frame. The first signal included in each frame is then subjected to power back-off processing to obtain the second signal included in each frame. Based on the second signal included in each frame, the initial neural network model is trained until the convergence condition is reached, and the target digital predistortion model trained under the working condition is obtained.

[0180] One possible implementation is that each frame includes a pilot signal in its first signal, and the first processing module 902 is specifically used for:

[0181] For any given frame, obtain the total number of signals included in the first signal, and the target position information of the pilot signal within the first signal.

[0182] The target location information, the preset power backoff value, and the total number of signals are input into the preset power backoff algorithm for power backoff processing to determine the second signal included in the frame.

[0183] One possible implementation is that the first processing module 902 is specifically used for:

[0184] Obtain the preset third signal when the power amplifier is in its initial operating condition.

[0185] Based on the third signal, the initial neural network model is initially trained offline until the convergence condition is met, resulting in the first digital predistortion model after training.

[0186] Based on the second signal included in each frame, the first digital predistortion model is retrained in the online state until the convergence condition is met, thus obtaining the target digital predistortion model trained under the working condition.

[0187] One possible implementation is that the first digital predistortion model includes an adaptation layer and a transfer layer, and the first processing module 902 is specifically used for:

[0188] Extract the feature information corresponding to the second signal included in each frame.

[0189] Based on the feature information corresponding to each frame, the target operating condition of the power amplifier corresponding to the second signal included in each frame is determined.

[0190] If the target operating condition is different from the initial operating condition, the hyperparameters of the adaptation layer in the first digital predistortion model are randomly initialized, while the hyperparameters of the transfer layer in the first digital predistortion model are kept, to obtain the adjusted second digital predistortion model.

[0191] The second signal included in each frame is divided into two paths, and the second digital predistortion model is retrained until the convergence condition is met, so as to obtain the target digital predistortion model trained under the working condition.

[0192] One possible implementation is that the first processing module 902 is specifically used for:

[0193] The second signal included in each frame is divided into two paths, and the second digital predistortion model is iteratively calculated according to the preset loss function and the preset update rule. When the preset loss function meets the preset conditions or the number of iterations reaches the preset number of iterations threshold, the target digital predistortion model trained under the working condition is obtained.

[0194] The preset loss function is determined by preset metric parameters that affect the nonlinearity of the power amplifier.

[0195] One possible implementation is that the first processing module 902 is also used for:

[0196] For any given working condition, the current hyperparameters of the adaptation layer and the migration layer in the target digital predistortion model are determined based on the target digital predistortion model corresponding to the working condition.

[0197] The signal power is determined based on the second signal included in each frame corresponding to the operating condition.

[0198] Based on the signal power and the current hyperparameters, establish a mapping relationship between the signal power and the current hyperparameters.

[0199] The mapping relationship is stored in a preset lookup table, which contains the mapping relationship between different signal powers and hyperparameters.

[0200] The training apparatus for the digital predistortion model provided in this embodiment can execute the training method for the digital predistortion model provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0201] Please see Figure 10 , Figure 10 This is a schematic diagram of a signal predistortion compensation device provided in an embodiment of this application. The device may include:

[0202] The second acquisition module 1001 is used to acquire the fourth signal included in each frame after power back-off processing.

[0203] The determining module 1002 is used to determine the target signal power of the fourth signal based on the fourth signal included in each frame.

[0204] The determination module 1002 is also used to determine the target hyperparameters that are mapped to the target signal power from a lookup table based on the target signal power.

[0205] The second processing module 1003 is used to replace the current hyperparameters in the target digital predistortion model with the target hyperparameters to obtain the adjusted digital predistortion model. The target digital predistortion model is the target digital predistortion model trained by the method in any of the above embodiments.

[0206] The second processing module 1003 is also used to input the fourth signal included in each frame into the adjusted digital predistortion model to obtain the compensation signal of the fourth signal included in each frame.

[0207] The second processing module 1003 is further configured to perform pre-compensation processing on the fourth signal included in each frame according to the compensation signal, so as to obtain the fifth signal included in each frame after pre-compensation processing.

[0208] One possible implementation is that the second acquisition module 1001 is specifically used for:

[0209] Acquire the currently acquired multi-frame signal.

[0210] Preprocessing of multiple frames of signals yields the sixth signal included in each frame.

[0211] The sixth signal included in each frame is processed by power back-off to obtain the fourth signal included in each frame.

[0212] The signal predistortion compensation device provided in this embodiment can execute the signal predistortion compensation method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0213] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 11 As shown, the electronic device provided in this embodiment includes at least one processor 1101 and a memory 1102. Optionally, the electronic device further includes a communication component 1103. The processor 1101, the memory 1102, and the communication component 1103 are connected via a bus 1104.

[0214] In a specific implementation, at least one processor 1101 executes computer execution instructions stored in memory 1102, causing at least one processor 1101 to perform the above-described method.

[0215] The specific implementation process of processor 1101 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0216] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0217] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0218] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0219] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0220] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0221] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0222] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0223] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0224] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0225] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0226] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0227] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0228] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for training a digital pre-distortion model, characterized in that, The application relates to a method for training a target digital pre-distortion model of a power amplifier under different working conditions. The method comprises the following steps: acquiring training data of the power amplifier under different working conditions, wherein the training data is a plurality of collected frames of signals, and each frame of signal comprises a plurality of initial signals; for any working condition, the following operations are performed to obtain a corresponding target digital pre-distortion model, the operations comprising:

2. The method of claim 1, wherein, pre-processing the initial signals included in each frame to obtain first signals included in each frame, performing power back-off processing on the first signals included in each frame to obtain second signals included in each frame, and training an initial neural network model according to the second signals included in each frame until a convergence condition is reached to obtain a target digital pre-distortion model trained under the working condition. The first signals comprise pilot signals; the power back-off processing on the first signals pre-processed in each frame to obtain the second signals included in each frame comprises: for any frame, acquiring a total number of signals included in the first signals and target position information of the pilot signals in the first signals; 3. The method of claim 1, wherein, inputting the target position information, a preset power back-off value and the total number of signals into a preset power back-off algorithm for power back-off processing to determine the second signals included in the frame. the training of the initial neural network model according to the second signals included in each frame until the convergence condition is reached to obtain the target digital pre-distortion model trained under the working condition comprises: acquiring a preset third signal when the power amplifier is in an initial working condition; initially training the initial neural network model based on the third signal in an offline state until a convergence condition is reached to obtain a first digital pre-distortion model trained; 4. The method of claim 3, wherein, re-training the first digital pre-distortion model based on the second signals included in each frame in an online state until a convergence condition is reached to obtain the target digital pre-distortion model trained under the working condition. the first digital pre-distortion model comprises an adaptive layer and a migration layer; the re-training of the first digital pre-distortion model based on the second signals included in each frame in an online state until a convergence condition is reached to obtain the target digital pre-distortion model trained under the working condition comprises: extracting feature information corresponding to the second signals included in each frame; determining a target working condition of the power amplifier corresponding to the second signals included in each frame according to the feature information corresponding to the frame; if the target working condition is different from the initial working condition, randomly initializing hyperparameters of the adaptive layer in the first digital pre-distortion model and keeping hyperparameters of the migration layer in the first digital pre-distortion model to obtain a second digital pre-distortion model adjusted; 5. The method of claim 4, wherein, dividing the second signals included in each frame into two paths, re-training the second digital pre-distortion model until a convergence condition is reached to obtain the target digital pre-distortion model trained under the working condition. the re-training of the second digital pre-distortion model until a convergence condition is reached to obtain the target digital pre-distortion model trained under the working condition comprises: The second signals included in each frame are divided into two paths, and the second digital pre-distortion model is iteratively calculated according to a preset loss function and a preset update rule, and when the preset loss function meets a preset condition or the number of iterations reaches a preset iteration threshold, a target digital pre-distortion model trained under the working condition is obtained. The preset loss function is determined by a preset metric parameter affecting the nonlinearity of the power amplifier.

6. The method according to any one of claims 1 to 5, characterized in that, Further comprising: For any working condition, determining current hyperparameters of an adaptive layer and a migration layer in the target digital pre-distortion model corresponding to the working condition according to the target digital pre-distortion model corresponding to the working condition; Determining a signal power according to the second signals included in each frame corresponding to the working condition; Establishing a mapping relationship between the signal power and the current hyperparameters according to the signal power and the current hyperparameters; Storing the mapping relationship in a preset lookup table, and the lookup table stores mapping relationships between different signal powers and hyperparameters.

7. A method of pre-distortion compensation of a signal, characterized by, Comprising: Obtaining fourth signals included in each frame after power backoff processing; Determining a target signal power of the fourth signals according to the fourth signals included in each frame; Determining a target hyperparameter corresponding to the target signal power from the lookup table according to the target signal power; Replacing current hyperparameters in a target digital pre-distortion model with the target hyperparameters to obtain an adjusted digital pre-distortion model, the target digital pre-distortion model being a target digital pre-distortion model trained by the method of any one of claims 1-6; Inputting the fourth signals included in each frame to the adjusted digital pre-distortion model to obtain compensation signals of the fourth signals included in each frame; Performing pre-compensation processing on the fourth signals included in each frame according to the compensation signals to obtain fifth signals included in each frame after pre-compensation processing.

8. The method of claim 7, wherein, The obtaining of the fourth signals included in each frame after power backoff processing comprises: Obtaining a plurality of frames of signals currently collected; Performing preprocessing on the plurality of frames of signals to obtain sixth signals included in each frame; Performing power backoff processing on the sixth signals included in each frame to obtain the fourth signals included in each frame.

9. A training device for a digital predistortion model, characterized in that, Comprising: A first obtaining module is configured to obtain training data of a power amplifier under different working conditions, the training data being a plurality of frames of signals collected, and each frame of signal including a plurality of initial signals; A first processing module is configured to, for any working condition, perform the following operations to obtain a corresponding target digital pre-distortion model, the operations comprising: Performing preprocessing on the initial signals included in each frame to obtain first signals included in each frame, performing power backoff processing on the first signals included in each frame to obtain second signals included in each frame, and training an initial neural network model according to the second signals included in each frame until a convergence condition is met to obtain a target digital pre-distortion model trained under the working condition.

10. A pre-distortion compensation apparatus for a signal, characterized by, Comprising: A second obtaining module is configured to obtain fourth signals included in each frame after power backoff processing; A determining module is configured to determine a target signal power of the fourth signals according to the fourth signals included in each frame; The determining module is further configured to determine, according to the target signal power, a target hyper-parameter mapped to the target signal power from a lookup table. The second processing module is configured to replace a current hyper-parameter in a target digital pre-distortion model with the target hyper-parameter to obtain an adjusted digital pre-distortion model, the target digital pre-distortion model being a target digital pre-distortion model trained by the method according to any one of claims 1-6. The second processing module is further configured to input the fourth signal included in each frame into the adjusted digital pre-distortion model to obtain a compensation signal of the fourth signal included in each frame. The second processing module is further configured to perform pre-compensation processing on the fourth signal included in each frame according to the compensation signal to obtain a fifth signal included in each frame after pre-compensation processing.

11. An electronic device, comprising: comprising: a memory, a processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory, so that the processor executes the method according to any one of claims 1-6 and / or 7-8.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed by the processor to implement the method according to any one of claims 1-6 and / or 7-8.

13. A computer program product, characterised in that, comprising a computer program, which is executed by the processor to implement the method according to any one of claims 1-6 and / or 7-8.

Images