Training method of digital pre-distortion model, signal transmitting method, device and equipment

Abstract

The present disclosure relates to a method, device and equipment for training a digital pre-distortion model, signal transmitting method, device and equipment, belonging to the technical field of wireless communication. The method comprises: training a power amplifier model based on sample input signals, sample output signals and sample scene information of a power amplifier hardware; performing digital pre-distortion processing on the sample input signals based on the sample scene information through the digital pre-distortion model to obtain a predicted pre-distortion signal; performing power amplification processing on the predicted pre-distortion signal associated with the sample scene information through the trained power amplifier model to obtain a first predicted output signal; and training the digital pre-distortion model based on first difference information between the sample input signals and the first predicted output signal to obtain a trained digital pre-distortion model. Thus, a general modeling mechanism of "one model for multiple scenes" is proposed, which can significantly improve the model training efficiency and reduce the model storage overhead.

Description

Technical Field

[0001] This disclosure relates to the field of wireless communication technology, and in particular to a training method for a digital predistortion model, a signal transmission method, an apparatus, an electronic device, a chip, a storage medium, and a computer program product. Background Technology

[0002] Currently, to improve the nonlinear distortion problem of transmitted signals, most methods utilize digital predistortion models to perform digital predistortion processing on the transmitted signals. The nonlinear characteristics of power amplifiers vary significantly across different scenarios. Related technologies require training and storing digital predistortion models separately for different scenarios. This "one scenario, one model" strategy suffers from low training efficiency and high storage overhead for digital predistortion models. Summary of the Invention

[0003] This disclosure provides a training method, signal transmission method, apparatus, electronic device, chip, storage medium, and computer program product for a digital predistortion model, to at least solve the problems of low training efficiency and high storage overhead in related technologies. The technical solution of this disclosure is as follows:

[0004] According to a first aspect of the present disclosure, a method for training a digital predistortion model is provided, comprising: training a power amplifier model based on sample input signals, sample output signals, and sample scene information of power amplifier hardware to obtain a trained power amplifier model; wherein the power amplifier model is a mathematical model of the power amplifier hardware, and the nonlinear characteristics of the power amplifier hardware are associated with the sample scene information; performing digital predistortion processing on the sample input signals based on the sample scene information using the digital predistortion model to obtain a predicted predistortion signal; performing power amplification processing on the predicted predistortion signal associated with the sample scene information using the trained power amplifier model to obtain a first predicted output signal; and training the digital predistortion model based on first difference information between the sample input signals and the first predicted output signal to obtain the trained digital predistortion model.

[0005] According to a second aspect of the present disclosure, a signal transmission method is provided, comprising: performing digital predistortion processing on an original transmission signal based on actual scene information using a trained digital predistortion model to obtain a predistorted signal; performing power amplification processing on the predistorted signal using power amplifier hardware to obtain a target transmission signal; wherein the nonlinear characteristics of the power amplifier hardware are associated with the actual scene information; and transmitting the target transmission signal to a communication receiving end.

[0006] According to a third aspect of the present disclosure, a training apparatus for a digital predistortion model is provided, comprising: a first training module configured to train a power amplifier model based on sample input signals, sample output signals, and sample scene information of power amplifier hardware to obtain a trained power amplifier model; wherein the power amplifier model is a mathematical model of the power amplifier hardware, and the nonlinear characteristics of the power amplifier hardware are associated with the sample scene information; a first processing module configured to perform digital predistortion processing on the sample input signals based on the sample scene information using the digital predistortion model to obtain a predicted predistortion signal; a second processing module configured to perform power amplification processing on the predicted predistortion signal associated with the sample scene information using the trained power amplifier model to obtain a first predicted output signal; and a second training module configured to train the digital predistortion model based on first difference information between the sample input signals and the first predicted output signal to obtain the trained digital predistortion model.

[0007] According to a fourth aspect of the present disclosure, a signal transmitting apparatus is provided, comprising: a first processing module configured to perform digital predistortion processing on an original transmitted signal based on actual scene information using a trained digital predistortion model to obtain a predistorted signal; a second processing module configured to perform power amplification processing on the predistorted signal using power amplifier hardware to obtain a target transmitted signal; wherein the nonlinear characteristics of the power amplifier hardware are associated with the actual scene information; and a transmitting module configured to transmit the target transmitted signal to a communication receiving end.

[0008] According to a fifth aspect of the present disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the training method for the digital predistortion model described in the first aspect of the present disclosure, and / or implements the steps of the signal transmission method described in the second aspect of the present disclosure.

[0009] According to a sixth aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, on which computer program instructions are stored, which, when executed by a processor, implement the steps of the training method for the digital predistortion model described in the first aspect of the present disclosure, and / or implement the steps of the signal transmission method described in the second aspect of the present disclosure.

[0010] According to a seventh aspect of the present disclosure, a chip is provided, the chip including an interface circuit and a processing circuit coupled to each other, the interface circuit being used to input or output signals, and the processing circuit being configured to implement the steps of the training method of the digital predistortion model of the first aspect of the present disclosure, and / or implement the steps of the signal transmission method of the second aspect of the present disclosure.

[0011] According to an eighth aspect of the present disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the steps of the training method for the digital predistortion model described in the first aspect of the present disclosure, and / or implements the steps of the signal transmission method described in the second aspect of the present disclosure.

[0012] The technical solutions provided by the embodiments of this disclosure bring at least the following beneficial effects: a general modeling mechanism of "multi-scenario one model" is proposed. Considering the sample input signal, sample output signal and sample scene information of the power amplifier hardware, the power amplifier model is trained to obtain the trained power amplifier model. During the training process, the power amplifier model can learn the ability to perform power amplification processing on the sample input signal in association with the sample scene information to obtain the sample output signal, and can learn the nonlinear characteristics of the power amplifier hardware under different scenarios. That is, the trained power amplifier model in this solution has scene perception and adaptive reasoning capabilities. Only one general power amplifier model needs to be trained and stored, which can greatly improve the training efficiency of the power amplifier model and significantly reduce the storage overhead of the power amplifier model.

[0013] Furthermore, by using a digital predistortion model to perform digital predistortion processing on the sample input signal based on sample scene information, a predicted predistortion signal can be obtained. During the training process, the digital predistortion model can learn the ability to perform digital predistortion processing on the sample input signal based on sample scene information. That is, the digital predistortion model trained in this scheme has scene awareness and adaptive reasoning capabilities, which can meet the nonlinear compensation requirements of power amplifier hardware in different scenarios. Only one general digital predistortion model needs to be trained and stored, which can significantly improve the training efficiency of the digital predistortion model and significantly reduce the storage overhead of the digital predistortion model.

[0014] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0016] Figure 1 This is a flowchart illustrating a training method for a digital predistortion model according to an exemplary embodiment.

[0017] Figure 2 This is a flowchart illustrating a training method for a digital predistortion model according to another exemplary embodiment.

[0018] Figure 3 This is a flowchart illustrating a training method for a digital predistortion model according to another exemplary embodiment.

[0019] Figure 4 This is a flowchart illustrating a training method for a digital predistortion model according to another exemplary embodiment.

[0020] Figure 5 This is a schematic flowchart illustrating a signal transmission method according to another exemplary embodiment.

[0021] Figure 6 This is a schematic diagram illustrating a power amplifier model according to an exemplary embodiment.

[0022] Figure 7 This is a schematic diagram of a power amplifier model according to another exemplary embodiment.

[0023] Figure 8 This is a schematic diagram illustrating the structure of a training device for a digital predistortion model according to an exemplary embodiment.

[0024] Figure 9 This is a schematic diagram of the structure of a signal transmitting device according to an exemplary embodiment.

[0025] Figure 10 This is a schematic diagram of the structure of an electronic device according to an exemplary embodiment.

[0026] Figure 11 This is a schematic diagram of the structure of a chip according to an exemplary embodiment. Detailed Implementation

[0027] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0028] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The implementations described in the following exemplary embodiments do not represent all implementations consistent with this disclosure.

[0029] The following description, with reference to the accompanying drawings, describes a training method, signal transmission method, apparatus, electronic device, chip, storage medium, and computer program product for a DPD (Digital Predistortion) model according to embodiments of the present disclosure.

[0030] Figure 1 This is a flowchart illustrating a training method for a digital predistortion model according to an exemplary embodiment, such as... Figure 1 As shown, the training method for the digital predistortion model in this embodiment includes the following steps.

[0031] S101, Based on the sample input signal, sample output signal and sample scene information of the power amplifier hardware, the power amplifier model is trained to obtain the trained power amplifier model; wherein, the power amplifier model is the mathematical model of the power amplifier hardware, and the nonlinear characteristics of the power amplifier hardware are related to the sample scene information.

[0032] It should be noted that the execution subject of the digital predistortion model training method in this disclosure is an electronic device, such as a base station, terminal device, vehicle, server, chip, etc. Terminal devices include mobile phones, wearable devices (such as smartwatches, smart glasses), laptops, etc., and vehicles include in-vehicle terminals, in-vehicle controllers, etc. The digital predistortion model training method in this disclosure can be executed by the digital predistortion model training device of this disclosure. This digital predistortion model training device can be configured in any electronic device to execute the digital predistortion model training method of this disclosure.

[0033] Currently, power amplifier (PA) hardware is a critical component in wireless communication transmission links. The nonlinear characteristics of power amplifier hardware can cause nonlinear distortion in the transmitted signal after passing through it, resulting in waveform distortion, reduced modulation accuracy, increased bit error rate, and severely impacting wireless communication quality. Furthermore, the nonlinear characteristics of power amplifier hardware can even trigger spectral regeneration, causing energy leakage from the transmitted signal into adjacent channels and interfering with their communication.

[0034] The nonlinear characteristics of power amplifiers vary significantly in different scenarios. Related technologies require training and storing power amplifier models separately for different scenarios. This "one scenario, one model" strategy results in low training efficiency and high storage overhead for power amplifier models.

[0035] To address the aforementioned issues, this disclosure proposes a universal modeling mechanism of "one model for multiple scenarios." Taking into account the sample input signals, sample output signals, and sample scenario information of the power amplifier hardware, the power amplifier model is trained to obtain a trained power amplifier model. During training, the power amplifier model can learn to perform power amplification processing on the sample input signals in relation to the sample scenario information to obtain the sample output signals. It can also learn the nonlinear characteristics of the power amplifier hardware under different scenarios. In other words, the trained power amplifier model in this scheme possesses scenario awareness and adaptive reasoning capabilities. Only one universal power amplifier model needs to be trained and stored, which can significantly improve the training efficiency of the power amplifier model and significantly reduce its storage overhead.

[0036] The sample input signal refers to the sample signal input to the power amplifier hardware, and the sample output signal refers to the signal after the sample input signal has been amplified by the power amplifier hardware, which is also the sample signal output by the power amplifier hardware.

[0037] The nonlinear characteristics of the power amplifier hardware are related to the sample scene information; that is, the sample scene information is an influencing factor on the nonlinear characteristics of the power amplifier hardware, and the nonlinear characteristics of the power amplifier change with the sample scene information. No excessive restrictions are placed on the sample scene information.

[0038] In one possible implementation, the sample scenario information includes at least one of the following: The operating frequency band of the power amplifier hardware; Operating frequency of power amplifier hardware; The bandwidth of the sample input signal; Temperature of the power amplifier hardware.

[0039] It should be noted that the power amplification processing associated with different sample scene information may be different.

[0040] A power amplifier model is a mathematical model of the power amplifier hardware. There are no strict limitations on the power amplifier model; it can include polynomial models, deep learning models, etc. Polynomial models include MP (Memory Polynomial) models and GMP (Generalized Memory Polynomial) models, while deep learning models include neural network models, such as fully connected neural network models, convolutional neural network models, and recurrent neural network models.

[0041] It should be noted that the power amplifier model can be trained using any model training method in the relevant technologies based on the sample input signal, sample output signal and sample scene information of the power amplifier hardware. No further restrictions are imposed here.

[0042] In one possible implementation, a power amplifier model is trained based on sample input signals, sample output signals, and sample scene information from the power amplifier hardware to obtain a trained power amplifier model. This includes using the power amplifier model to perform power amplification processing on the sample input signals in association with the sample scene information to obtain a second predicted output signal; and training the power amplifier model based on second difference information between the second predicted output signal and the sample output signal to obtain the trained power amplifier model. Thus, the power amplifier model can perform power amplification processing on the sample input signals in association with the sample scene information to obtain a second predicted output signal, and train the power amplifier model considering the second difference information between the second predicted output signal and the sample output signal to obtain the trained power amplifier model.

[0043] In one possible implementation, a power amplifier model is used to amplify the sample input signal in relation to the sample scene information to obtain a second predicted output signal. This includes amplifying the sample input signal using a power amplifier model based on a power amplification method associated with the sample scene information to obtain the second predicted output signal. It is understood that the power amplification method associated with different sample scene information may differ, and the power amplification method can be characterized by the values ​​of model parameters, the network nodes performing the power amplification, etc.

[0044] In one possible implementation, the power amplifier model is trained based on second difference information between the second predicted output signal and the sample output signal to obtain a trained power amplifier model. This includes determining a loss function for the power amplifier model based on the second difference information, and training the power amplifier model based on the loss function to obtain the trained power amplifier model.

[0045] It should be noted that there are no strict restrictions on the loss function, such as cross-entropy loss function, mean squared error loss function, mean absolute error loss function, KL (Kullback-Leibler) divergence loss function, etc.

[0046] In one possible implementation, the method further includes sampling the actual input signal of the power amplifier hardware to obtain a sample input signal, sampling the actual output signal of the power amplifier hardware to obtain a sample output signal, and sampling the actual scene information of the power amplifier hardware to obtain sample scene information.

[0047] In one possible implementation, the method further includes normalizing the sample scene information under any category, using the normalized sample scene information as the final sample scene information. It should be noted that the normalization of the sample scene information can be achieved using any data normalization method in related technologies, and no further limitations are imposed here.

[0048] For example, sample scene information under any category The maximum and minimum values ​​are normalized to 1 and -1, respectively. , The sample size can be determined using the following formula:

[0049] in, This refers to the normalized sample scene information. For sample scene information The maximum value, For sample scene information The minimum value.

[0050] For example, sample scene information under any category The mean and standard deviation are both normalized to 1, which can be achieved using the following formula:

[0051] in, For sample scene information The mean, For sample scene information The standard deviation.

[0052] S102, based on the sample scene information, the sample input signal is digitally predistorted using a digital predistortion model to obtain a predicted predistortion signal.

[0053] Currently, to improve the nonlinear distortion problem of transmitted signals, most methods utilize digital predistortion models to perform digital predistortion processing on the transmitted signals. The nonlinear characteristics of power amplifiers vary significantly across different scenarios. Related technologies require training and storing digital predistortion models separately for different scenarios. This "one scenario, one model" strategy suffers from low training efficiency and high storage overhead for digital predistortion models.

[0054] To address the aforementioned issues, this disclosure proposes a universal modeling mechanism of "one model for multiple scenarios." Based on sample scenario information, a digital predistortion model performs digital predistortion processing on the sample input signal to obtain a predicted predistortion signal. During training, the digital predistortion model learns the ability to perform digital predistortion processing on the sample input signal based on sample scenario information. In other words, the trained digital predistortion model in this scheme possesses scene awareness and adaptive reasoning capabilities, which can meet the nonlinear compensation requirements of power amplifier hardware in different scenarios. Only one universal digital predistortion model needs to be trained and stored, significantly improving the training efficiency of the digital predistortion model and significantly reducing its storage overhead.

[0055] There are no strict limitations on digital predistortion models, such as including multinomial models and deep learning models.

[0056] In this embodiment, a digital predistortion model is used to perform digital predistortion processing on the sample input signal based on the sample scene information to obtain a predicted predistortion signal, including at least one of the following: Method 1: The sample scene information and sample input signal are spliced ​​together to obtain the first spliced ​​signal. The first spliced ​​signal is then processed by a digital predistortion model to obtain the predicted predistortion signal.

[0057] Therefore, the sample scene information and sample input signal are spliced ​​together to obtain the first spliced ​​signal, which is then used as the input data for the digital predistortion model. The digital predistortion model processes the first spliced ​​signal to obtain the predicted predistortion signal. The splicing operation is a lightweight fusion operation with low computational complexity, making the digital predistortion model more lightweight. This significantly reduces the amount of computation required for digital preprocessing, helps improve signal transmission efficiency, and can reduce the storage overhead of the digital predistortion model.

[0058] In one possible implementation, the digital predistortion model includes network layer A, network layer B, and network layer C; the digital predistortion model processes the first spliced ​​signal to obtain a predicted predistortion signal, including extracting features from sample scene information in the first spliced ​​signal through network layer A to obtain second scene features, extracting features from sample input signals in the first spliced ​​signal through network layer B to obtain second signal features, and processing the second scene features and the second signal features through network layer C to obtain the predicted predistortion signal.

[0059] For example, the second scene features and the second signal features are processed by network layer C to obtain the predicted predistortion signal. This includes fusing the second scene features and the second signal features by network layer C to obtain the second fused features, and then determining the predicted predistortion signal by network layer C based on the second fused features.

[0060] Method 2: Based on the sample scene information, determine the third value of the parameters of the digital predistortion model, configure the parameters of the digital predistortion model to the third value to update the digital predistortion model, and perform digital predistortion processing on the sample input signal through the updated digital predistortion model to obtain the predicted predistortion signal.

[0061] Therefore, the parameters of the digital predistortion model can be adaptively adjusted by taking into account the sample scene information. That is, the parameters of the digital predistortion model change with the sample scene information, which can accurately compensate for the nonlinear characteristics of the power amplifier hardware in a specific scene.

[0062] In one possible implementation, the digital predistortion model includes a supernetwork and a main network. The supernetwork determines a third value for the third parameter of the main network based on sample scene information and configures the third parameter to this third value to update the main network. The third parameter is associated with the sample scene information. The updated main network performs digital predistortion processing on the sample input signal to obtain the predicted predistortion signal. Therefore, the supernetwork can adaptively adjust the value of the third parameter of the main network based on the sample scene information; that is, the third parameter changes with the sample scene information, which can accurately compensate for the nonlinear characteristics of the power amplifier hardware in specific scenarios.

[0063] In one possible implementation, the parameters of the main network also include a fourth parameter, which is unrelated to the sample scene information.

[0064] Method 3: By using network nodes in the digital predistortion model that match the sample scene information, digital predistortion processing is performed on the sample input signal to obtain the predicted predistortion signal.

[0065] Therefore, the digital predistortion model includes network nodes that match the sample scene information. It can take the sample scene information into account and adaptively select network nodes for digital predistortion processing, which can accurately compensate for the nonlinear characteristics of power amplifier hardware in specific scenarios.

[0066] In this embodiment, the digital predistortion model includes network nodes that match the sample scene information, and the network nodes that match different sample scene information may be different.

[0067] In one possible implementation, the method further includes determining the network nodes corresponding to the sample scene information in the digital predistortion model, as the network nodes in the digital predistortion model that match the sample scene information. It is understood that the correspondence between scene information and network nodes can be pre-established, and the network nodes corresponding to different scene information may be different.

[0068] In one possible implementation, the method further includes determining the network nodes corresponding to the value ranges of the sample scene information in the digital predistortion model, as the network nodes in the digital predistortion model that match the sample scene information. It is understood that the correspondence between the value ranges of scene information and network nodes can be established in advance, and different value ranges of scene information may correspond to different network nodes.

[0069] In one possible implementation, the digital predistortion model employs a network tree architecture, comprising multiple sub-models. Each network layer of the digital predistortion model includes network nodes that match the sample scene information, and at least two sub-models share network nodes. This shared network node structure makes the digital predistortion model more lightweight, improves its training efficiency, and significantly reduces its storage overhead.

[0070] It should be noted that the network tree architecture is a model architecture that combines multiple network nodes based on a tree structure.

[0071] For example, sample scene information includes the operating frequency band of the power amplifier hardware and the bandwidth of the sample input signal. The digital predistortion model includes network layer A, network layer B, and network layer C. Network layer A includes network node A1, network layer B includes network nodes B1 and B2, and network layer C includes network nodes C1, C2, C3, and C4. Network node A1 is matched with each sample scene information, network node B1 is matched with the 2.6 GHz (gigahertz) frequency band, network node B2 is matched with the 4.9 GHz frequency band, network node C1 is matched with the 2.6 GHz frequency band and 20 Mbps (megabits per second) bandwidth, network node C2 is matched with the 2.6 GHz frequency band and 50 Mbps bandwidth, network node C3 is matched with the 4.9 GHz frequency band and 20 Mbps bandwidth, and network node C4 is matched with the 4.9 GHz frequency band and 50 Mbps bandwidth.

[0072] Network node A1 is the parent node of network nodes B1 and B2, network node B1 is the parent node of network nodes C1 and C2, and network node B2 is the parent node of network nodes C3 and C4.

[0073] The digital predistortion model includes sub-models 1 to 4. Sub-model 1 is obtained by combining network nodes A1, B1, and C1; sub-model 2 is obtained by combining network nodes A1, B1, and C2; sub-model 3 is obtained by combining network nodes A1, B2, and C3; and sub-model 4 is obtained by combining network nodes A1, B2, and C4.

[0074] Sub-model 1 shares network nodes A1 and B1 with sub-model 2; sub-model 1 shares network node A1 with sub-model 3; sub-model 1 shares network node A1 with sub-model 4; sub-model 2 shares network node A1 with sub-model 3; sub-model 2 shares network node A1 with sub-model 4; sub-model 3 shares network nodes A1 and B2 with sub-model 4.

[0075] S103, through the trained power amplifier model, performs power amplification processing on the predicted predistortion signal in association with the sample scene information to obtain the first predicted output signal.

[0076] In one possible implementation, the predicted predistortion signal is amplified by a power amplifier model associated with the sample scene information to obtain a first predicted output signal. This includes amplifying the predicted predistortion signal by a power amplifier model based on a power amplification method associated with the sample scene information to obtain the first predicted output signal.

[0077] S104, Based on the first difference information between the sample input signal and the first predicted output signal, the digital predistortion model is trained to obtain the trained digital predistortion model.

[0078] It should be noted that the first difference information is used to characterize the nonlinear distortion between the sample input signal and the first predicted output signal, and may include, for example, error vector magnitude, total harmonic distortion, normalized mean square error, etc.

[0079] The digital predistortion model is trained based on the first difference information between the sample input signal and the first predicted output signal. Any model training method in related technologies can be used to achieve this, and no further restrictions are imposed here.

[0080] In one possible implementation, a digital predistortion model is trained based on first difference information between the sample input signal and the first predicted output signal to obtain a trained digital predistortion model. This includes determining a loss function for the digital predistortion model based on the first difference information, and training the digital predistortion model based on the loss function to obtain the trained digital predistortion model.

[0081] In one possible implementation, the digital predistortion model includes a supernetwork and a main network, where the third parameter of the main network is associated with sample scene information. The digital predistortion model is trained based on a first difference information between the sample input signal and the first predicted output signal to obtain a trained digital predistortion model. This includes determining the gradient of the third parameter based on the first difference information, and updating the parameters of the supernetwork based on the gradient of the third parameter to obtain the trained supernetwork. Thus, during training, the supernetwork can learn the ability to generate the third parameter of the main network based on sample scene information.

[0082] In one possible implementation, the parameters of the supernetwork are updated based on the gradient of the third parameter to obtain the trained supernetwork, including backpropagation of the gradient of the third parameter to determine the gradient of the supernetwork parameters, and updating the supernetwork parameters based on the gradient of the supernetwork parameters to obtain the trained supernetwork.

[0083] In one possible implementation, the parameters of the main network further include a fourth parameter, which is independent of the sample scene information. The digital predistortion model is trained based on the first difference information between the sample input signal and the first predicted output signal to obtain the trained digital predistortion model. This includes determining the gradient of the fourth parameter based on the first difference information, and updating the fourth parameter based on the gradient of the fourth parameter to update the main network.

[0084] In one possible implementation, the digital predistortion model includes network nodes matched with sample scene information. The digital predistortion model is trained based on first difference information between the sample input signal and the first predicted output signal to obtain a trained digital predistortion model. This includes updating the parameters of a sub-model obtained by combining network nodes that match the sample scene information in the digital predistortion model based on the first difference information to obtain the corresponding trained sub-model. Therefore, by taking into account the first difference information, the parameters of the sub-model obtained by combining specific network nodes in the digital predistortion model can be locally optimized, which helps to improve the training efficiency of the digital predistortion model.

[0085] For example, let's continue with the example of a digital predistortion model that includes network layer A, network layer B, and network layer C.

[0086] In the first scenario, if the sample scene information includes the power amplifier hardware operating frequency band of 2.6GHz and the sample input signal bandwidth of 20Mbps, then the sub-model 1 obtained by combining network nodes A1, B1, and C1 performs digital predistortion processing on the sample input signal to obtain a predicted predistortion signal. Using the trained power amplifier model, the predicted predistortion signal undergoes power amplification processing related to the sample scene information to obtain a first predicted output signal. Based on the first difference information between the sample input signal and the first predicted output signal, the parameters of sub-model 1 are updated to obtain the trained sub-model 1. For example, the parameters of network nodes A1, B1, and C1 are updated to obtain the trained network nodes A1, B1, and C1, respectively.

[0087] In the second scenario, if the sample scene information includes the power amplifier hardware operating frequency band of 2.6GHz and the sample input signal bandwidth of 50Mbps, then the sub-model 2 obtained by combining network nodes A1, B1, and C2 performs digital predistortion processing on the sample input signal to obtain a predicted predistortion signal. Using the trained power amplifier model, the predicted predistortion signal undergoes power amplification processing related to the sample scene information to obtain a first predicted output signal. Based on the first difference information between the sample input signal and the first predicted output signal, the parameters of sub-model 2 are updated to obtain the trained sub-model 2. For example, the parameters of network nodes A1, B1, and C2 are updated to obtain the trained network nodes A1, B1, and C2, respectively.

[0088] In the third scenario, if the sample scene information includes the power amplifier hardware operating frequency band of 4.9GHz and the sample input signal bandwidth of 20Mbps, then the sub-model 3 obtained by combining network nodes A1, B2, and C3 performs digital predistortion processing on the sample input signal to obtain a predicted predistortion signal. Using the trained power amplifier model, the predicted predistortion signal undergoes power amplification processing related to the sample scene information to obtain a first predicted output signal. Based on the first difference information between the sample input signal and the first predicted output signal, the parameters of sub-model 3 are updated to obtain the trained sub-model 3. For example, the parameters of network nodes A1, B2, and C3 are updated to obtain the trained network nodes A1, B2, and C3, respectively.

[0089] In the fourth scenario, if the sample scene information includes the power amplifier hardware operating frequency band of 4.9GHz and the sample input signal bandwidth of 50Mbps, then the sub-model 4 obtained by combining network nodes A1, B2, and C4 performs digital predistortion processing on the sample input signal to obtain a predicted predistortion signal. Using the trained power amplifier model, the predicted predistortion signal undergoes power amplification processing associated with the sample scene information to obtain a first predicted output signal. Based on the first difference information between the sample input signal and the first predicted output signal, the parameters of sub-model 4 are updated to obtain the trained sub-model 4. For example, the parameters of network nodes A1, B2, and C4 are updated to obtain the trained network nodes A1, B2, and C4, respectively.

[0090] In one possible implementation, the parameters of the trained power amplifier model are fixed during the training of the digital predistortion model. This decouples the training of the power amplifier model from that of the digital predistortion model. The power amplifier model is trained solely on sample input signals, sample output signals, and sample scene information from the power amplifier hardware. This ensures that the power amplifier model is not affected by the training of the digital predistortion model, allowing it to more closely resemble the power amplifier hardware and thus improving the training accuracy of the digital predistortion model.

[0091] In addition, joint training of the power amplifier model and the digital predistortion model can easily lead to problems such as gradient oscillation and difficulty in model convergence. In this scheme, the power amplifier model and the digital predistortion model are trained in a decoupled manner, which helps to accelerate the convergence speed of the power amplifier model and the digital predistortion model respectively, thereby improving the training stability and training efficiency of the power amplifier model and the digital predistortion model respectively.

[0092] It should be noted that, Figure 1 In the embodiments, the supernetwork, main network, sub-model, and network node are all network structures in the digital predistortion model.

[0093] This disclosure does not impose any restrictions on the execution sequence of steps S101-S104. Figure 1 The example is only executed in the order of steps S101-S104.

[0094] The digital predistortion model training method provided in this disclosure proposes a general modeling mechanism of "one model for multiple scenarios". Taking into account the sample input signal, sample output signal and sample scene information of the power amplifier hardware, the power amplifier model is trained to obtain the trained power amplifier model. During the training process, the power amplifier model can learn to perform power amplification processing on the sample input signal in association with the sample scene information to obtain the sample output signal, and can learn the nonlinear characteristics of the power amplifier hardware under different scenarios. That is, the trained power amplifier model in this scheme has scene perception and adaptive reasoning capabilities. Only one general power amplifier model needs to be trained and stored, which can significantly improve the training efficiency of the power amplifier model and significantly reduce the storage overhead of the power amplifier model.

[0095] Furthermore, by using a digital predistortion model to perform digital predistortion processing on the sample input signal based on sample scene information, a predicted predistortion signal can be obtained. During the training process, the digital predistortion model can learn the ability to perform digital predistortion processing on the sample input signal based on sample scene information. That is, the digital predistortion model trained in this scheme has scene awareness and adaptive reasoning capabilities, which can meet the nonlinear compensation requirements of power amplifier hardware in different scenarios. Only one general digital predistortion model needs to be trained and stored, which can significantly improve the training efficiency of the digital predistortion model and significantly reduce the storage overhead of the digital predistortion model.

[0096] Figure 2 This is a flowchart illustrating a training method for a digital predistortion model according to another exemplary embodiment, such as... Figure 2 As shown, the training method for the digital predistortion model in this embodiment includes the following steps.

[0097] S201, the sample scene information and sample input signal are spliced ​​together to obtain the first spliced ​​signal.

[0098] S202, the first spliced ​​signal is processed by a power amplifier model to obtain the second predicted output signal.

[0099] In this embodiment, the sample scene information and sample input signal are concatenated to obtain a first concatenated signal. The first concatenated signal is then used as the input data for the power amplifier model. The power amplifier model processes the first concatenated signal to obtain a second predicted output signal. The concatenation operation is a lightweight fusion operation with low computational complexity, making the power amplifier model more lightweight and significantly improving its inference speed. This, in turn, helps to improve the training efficiency of the power amplifier model and reduces its storage overhead.

[0100] It should be noted that the first spliced ​​signal can be obtained by splicing the sample scene information and the sample input signal using any data splicing method in the relevant technology, and no further restrictions are imposed here.

[0101] In one possible implementation, the power amplifier model includes a first network layer, a second network layer, and a third network layer; the first spliced ​​signal is processed by the power amplifier model to obtain a second predicted output signal, including extracting features from sample scene information in the first spliced ​​signal through the first network layer to obtain first scene features, extracting features from sample input signals in the first spliced ​​signal through the second network layer to obtain first signal features, and processing the first scene features and the first signal features through the third network layer to obtain the second predicted output signal.

[0102] For example, the first scene features and the first signal features are processed by the third network layer to obtain the second predicted output signal. This includes fusing the first scene features and the first signal features by the third network layer to obtain the first fused features, and then determining the second predicted output signal by the third network layer based on the first fused features.

[0103] It should be noted that the first scene feature and the first signal feature are fused through the third network layer to obtain the first fused feature. This can be achieved using any feature fusion method in related technologies, without further limitations. For example, feature fusion methods include attention mechanisms, gating mechanisms, weighted summation, etc.

[0104] S203, Based on the second difference information between the second predicted output signal and the sample output signal, the power amplifier model is trained to obtain the trained power amplifier model.

[0105] S204 uses a digital predistortion model to perform digital predistortion processing on the sample input signal based on the sample scene information in order to obtain the predicted predistortion signal.

[0106] The details of steps S203-S204 can be found in the above embodiments and will not be repeated here.

[0107] S205, the sample scene information and the predicted predistortion signal are spliced ​​together to obtain the second spliced ​​signal.

[0108] S206, the second spliced ​​signal is processed by the trained power amplifier model to obtain the first predicted output signal.

[0109] In this embodiment, the sample scene information and the predicted predistortion signal are spliced ​​together to obtain a second spliced ​​signal. The second spliced ​​signal is then used as input data for the power amplifier model. The trained power amplifier model processes the second spliced ​​signal to obtain a first predicted output signal. The splicing operation is a lightweight fusion operation with low computational complexity, making the power amplifier model more lightweight and significantly improving the inference speed of the power amplifier model. This, in turn, helps to improve the training efficiency of the digital predistortion model.

[0110] In one possible implementation, the power amplifier model includes a first network layer, a second network layer, and a third network layer. The trained power amplifier model processes the second spliced ​​signal to obtain a first predicted output signal. This process includes extracting features from sample scene information in the second spliced ​​signal using the trained first network layer to obtain third scene features, extracting features from the predicted predistortion signal in the second spliced ​​signal using the trained second network layer to obtain third signal features, and processing the third scene features and the third signal features using the trained third network layer to obtain the first predicted output signal.

[0111] For example, the third scene features and the third signal features are processed by the trained third network layer to obtain the first predicted output signal. This includes fusing the third scene features and the third signal features by the trained third network layer to obtain the third fused features, and determining the first predicted output signal based on the third fused features by the trained third network layer.

[0112] It should be noted that the relevant content of step S206 can be referred to the relevant content of step S202, and will not be repeated here.

[0113] S207, Based on the first difference information between the sample input signal and the first predicted output signal, the digital predistortion model is trained to obtain the trained digital predistortion model.

[0114] The details of step S207 can be found in the above embodiments and will not be repeated here.

[0115] It should be noted that this disclosure does not impose any restrictions on the execution sequence of steps S201-S207. Figure 2 The example only demonstrates the execution of steps S201-S207 in sequence.

[0116] The digital predistortion model training method provided in the embodiments of this disclosure concatenates sample scene information and sample input signals to obtain a first concatenated signal, and uses the first concatenated signal as input data for a power amplifier model. The power amplifier model processes the first concatenated signal to obtain a second predicted output signal. The concatenation operation is a lightweight fusion operation with low computational complexity, thereby making the power amplifier model more lightweight, significantly improving the inference speed of the power amplifier model, which in turn helps to improve the training efficiency of the power amplifier model and can reduce the storage overhead of the power amplifier model.

[0117] In addition, the sample scene information and the predicted predistortion signal are concatenated to obtain a second concatenated signal, which is then used as input data for the power amplifier model. The trained power amplifier model processes the second concatenated signal to obtain the first predicted output signal. The concatenation operation is a lightweight fusion operation with low computational complexity, making the power amplifier model more lightweight and significantly improving its inference speed, thereby helping to improve the training efficiency of the digital predistortion model.

[0118] Figure 3 This is a flowchart illustrating a training method for a digital predistortion model according to another exemplary embodiment, such as... Figure 3 As shown, the training method for the digital predistortion model in this embodiment includes the following steps.

[0119] S301, the supernetwork determines the first value of the first parameter of the main network based on the sample scene information, and configures the first parameter to the first value to update the main network; wherein, the first parameter is associated with the sample scene information.

[0120] S302, the sample input signal is amplified by the updated main network to obtain the second predicted output signal.

[0121] In this embodiment, as Figure 6 As shown, the power amplifier model includes a supernetwork and a main network.

[0122] Based on sample scene information, a hypernetwork determines the first value of the first parameter of the main network and configures it to this first value to update the main network; the first parameter is associated with the sample scene information. The updated main network then amplifies the sample input signal to obtain the second predicted output signal. Therefore, by adaptively adjusting the value of the first parameter of the main network based on sample scene information, the hypernetwork can accurately characterize the nonlinear characteristics of the power amplifier hardware under specific scenarios.

[0123] S303, based on the second difference information, determine the gradient of the first parameter.

[0124] S304 updates the parameters of the supernetwork based on the gradient of the first parameter to obtain the trained supernetwork.

[0125] In this embodiment, the gradient of the first parameter can be determined by taking into account the second difference information in order to update the parameters of the super network and obtain the trained super network. During the training process, the super network can learn the ability to generate the first parameters of the main network based on the sample scene information.

[0126] It should be noted that the parameters of the supernetwork are updated based on the gradient of the first parameter to obtain the trained supernetwork. This can be achieved using any supernetwork training method in related technologies, without further restrictions here.

[0127] In one possible implementation, the parameters of the supernetwork are updated based on the gradient of the first parameter to obtain the trained supernetwork, including backpropagation of the gradient of the first parameter to determine the gradient of the supernetwork parameters, and updating the supernetwork parameters based on the gradient of the supernetwork parameters to obtain the trained supernetwork.

[0128] S305, Based on the second difference information, determine the gradient of the second parameter.

[0129] S306 updates the second parameter based on its gradient to update the main network.

[0130] In this embodiment, the parameters of the main network also include a second parameter, which is unrelated to the sample scene information. The gradient of the second parameter can be determined by taking into account the second difference information, thereby updating the second parameter and thus the main network.

[0131] It should be noted that updating the second parameter based on its gradient can be achieved using any model training method in the relevant technologies, and no further restrictions are imposed here.

[0132] S307 uses a digital predistortion model to perform digital predistortion processing on the sample input signal based on the sample scene information in order to obtain the predicted predistortion signal.

[0133] The details of step S307 can be found in the above embodiments and will not be repeated here.

[0134] S308: Based on the sample scene information, the trained supernetwork determines the second value of the first parameter and configures the first parameter to the second value to update the main network.

[0135] S309, the predicted predistortion signal is amplified by the updated main network to obtain the first predicted output signal.

[0136] In this embodiment, the trained supernetwork determines a second value for the first parameter of the main network based on sample scene information, and configures the first parameter to the second value to update the main network. The updated main network then performs power amplification processing on the predicted predistortion signal to obtain the first predicted output signal. Therefore, the trained supernetwork can adaptively adjust the value of the first parameter of the main network based on sample scene information; that is, the first parameter changes with the sample scene information. This accurately characterizes the nonlinear characteristics of the power amplifier hardware in specific scenarios, thereby helping to improve the training accuracy of the digital predistortion model.

[0137] S310, the digital predistortion model is trained based on the first difference information between the sample input signal and the first predicted output signal to obtain the trained digital predistortion model.

[0138] The details of step S310 can be found in the above embodiments and will not be repeated here.

[0139] It should be noted that, Figure 3 In the embodiments, both the supernetwork and the main network are network structures in the power amplifier model.

[0140] This disclosure does not limit the execution order of steps S301-S310. For example, steps S301-S304 and S307-S310 can be implemented as independent embodiments, and steps S303-S304 and S305-S306 can be executed in parallel.

[0141] The digital predistortion model training method provided in this disclosure involves a hypernetwork determining a first value for a first parameter of the main network based on sample scene information, and configuring the first parameter to this first value to update the main network; wherein the first parameter is associated with the sample scene information. The updated main network then performs power amplification processing on the sample input signal to obtain a second predicted output signal. Therefore, by adaptively adjusting the value of the first parameter of the main network based on sample scene information, the hypernetwork can accurately characterize the nonlinear characteristics of the power amplifier hardware in a specific scenario.

[0142] In addition, the gradient of the first parameter can be determined by taking into account the second difference information, so as to update the parameters of the super network and obtain the trained super network. During the training process, the super network can learn the ability to generate the first parameters of the main network based on the sample scene information.

[0143] Furthermore, based on sample scene information, the trained supernetwork determines a second value for the first parameter of the main network and configures the first parameter to the second value to update the main network. The updated main network then performs power amplification on the predicted predistortion signal to obtain the first predicted output signal. Thus, by adaptively adjusting the value of the first parameter of the main network based on sample scene information—meaning the first parameter changes with the sample scene information—the nonlinear characteristics of the power amplifier hardware under specific scenarios can be accurately characterized, thereby improving the training accuracy of the digital predistortion model.

[0144] Figure 4 This is a flowchart illustrating a training method for a digital predistortion model according to another exemplary embodiment, such as... Figure 4 As shown, the training method for the digital predistortion model in this embodiment includes the following steps.

[0145] S401 uses a sub-model obtained by combining network nodes in each network layer that match the sample scene information to amplify the power of the sample input signal in order to obtain the second predicted output signal.

[0146] In this embodiment, the power amplifier model adopts a network tree architecture, comprising multiple sub-models. Each network layer of the power amplifier model includes network nodes that match the sample scene information, and at least two sub-models share network nodes. Therefore, the shared network nodes between at least two sub-models make the power amplifier model more lightweight, which helps improve the training efficiency of the power amplifier model and significantly reduces its storage overhead.

[0147] In addition, considering the sample scenario information, network nodes for power amplification processing can be adaptively selected, and the selected network nodes can be combined to obtain a sub-model. This sub-model can accurately characterize the nonlinear characteristics of the power amplifier hardware in a specific scenario.

[0148] For example, sample scene information includes the operating frequency band of the power amplifier hardware and the bandwidth of the sample input signal.

[0149] like Figure 7 As shown, the power amplifier model includes network layer 1, network layer 2 and network layer 3. Network layer 1 includes network node 1, network layer 2 includes network nodes 2 and 3, and network layer 3 includes network nodes 4, 5, 6 and 7.

[0150] Network node 1 is matched with the information of each sample scenario; network node 2 is matched with the 2.6GHz frequency band; network node 3 is matched with the 4.9GHz frequency band; network node 4 is matched with the 2.6GHz frequency band and 20Mbps bandwidth; network node 5 is matched with the 2.6GHz frequency band and 50Mbps bandwidth; network node 6 is matched with the 4.9GHz frequency band and 20Mbps bandwidth; and network node 7 is matched with the 4.9GHz frequency band and 50Mbps bandwidth.

[0151] Network node 1 is the parent node of network nodes 2 and 3, network node 2 is the parent node of network nodes 4 and 5, and network node 3 is the parent node of network nodes 6 and 7.

[0152] The power amplifier model includes sub-models 1 to 4. Sub-model A is obtained by combining network nodes 1, 2, and 4; sub-model B is obtained by combining network nodes 1, 2, and 5; sub-model C is obtained by combining network nodes 1, 3, and 6; and sub-model D is obtained by combining network nodes 1, 3, and 7.

[0153] Sub-model A shares network nodes 1 and 2 with sub-model B; sub-model A shares network node 1 with sub-model C; sub-model A shares network node 1 with sub-model D. Sub-model B shares network node 1 with sub-model C; sub-model B shares network node 1 with sub-model D. Sub-model C shares network nodes 1 and 3 with sub-model D.

[0154] Understandably, the selection of parent nodes to child nodes in the network tree is based on the sample scene information. During digital predistortion model inference, the network nodes to be processed for digital predistortion are determined according to the combination of sample scene information (i.e., the network nodes in each network layer that match the sample scene information).

[0155] S402, based on the second difference information, update the parameters of the sub-models obtained by combining network nodes in each network layer that match the sample scene information, so as to obtain the corresponding trained sub-models.

[0156] In this embodiment, the second difference information can be taken into account to locally optimize the parameters of the sub-model obtained by combining specific network nodes in the power amplifier model, which helps to improve the training efficiency of the power amplifier model.

[0157] For example, continue with Figure 7 Let's take an example to illustrate the training process of a power amplifier model.

[0158] In the first scenario, if the sample scene information includes the power amplifier hardware operating frequency band of 2.6GHz and the sample input signal bandwidth of 20Mbps, then the sub-model A obtained by combining network nodes 1, 2, and 4 performs power amplification on the sample input signal to obtain the second predicted output signal. Based on the second difference information, the parameters of sub-model A are updated to obtain the trained sub-model A, for example, the parameters of network node 1 are updated. Parameters of network node 2 Parameters of network node 4 Update the network to obtain trained network nodes 1, 2, and 4 respectively.

[0159] In the second scenario, if the sample scene information includes the power amplifier hardware operating frequency band of 2.6GHz and the sample input signal bandwidth of 50Mbps, then the sub-model B obtained by combining network nodes 1, 2, and 5 performs power amplification on the sample input signal to obtain the second predicted output signal. Based on the second difference information, the parameters of sub-model B are updated to obtain the trained sub-model B, for example, the parameters of network node 1 are updated. Parameters of network node 2 Parameters of network node 5 Update the network to obtain trained network nodes 1, 2, and 5 respectively.

[0160] In the third scenario, if the sample scene information includes the power amplifier hardware operating frequency band of 4.9GHz and the sample input signal bandwidth of 20Mbps, then the sub-model C obtained by combining network nodes 1, 3, and 6 performs power amplification on the sample input signal to obtain the second predicted output signal. Based on the second difference information, the parameters of sub-model C are updated to obtain the trained sub-model C, for example, the parameters of network node 1 are updated. Parameters of network node 3 Parameters of network node 6 Update the network to obtain trained network nodes 1, 3, and 6 respectively.

[0161] In the fourth scenario, if the sample scene information includes the power amplifier hardware operating frequency band of 4.9GHz and the sample input signal bandwidth of 50Mbps, then the sub-model D obtained by combining network nodes 1, 3, and 7 performs power amplification on the sample input signal to obtain the second predicted output signal. Based on the second difference information, the parameters of sub-model D are updated to obtain the trained sub-model D, for example, the parameters of network node 1 are updated. Parameters of network node 3 Parameters of network node 7 Update the network to obtain trained network nodes 1, 3, and 7 respectively.

[0162] To facilitate understanding of the training process of a network tree, an exemplary embodiment is provided: Step a. For network layer 1 corresponding to the root node, collect the training dataset with the scene information of the sample corresponding to the root node, or select a training dataset that meets the scene information of the corresponding sample from the existing training dataset. Concatenate network layer 1 with the output layer, and train the neural network based on the training dataset.

[0163] Step b. Perform the following steps one by one, from the first child node of the root node to the last leaf node.

[0164] For any node's corresponding network layer Fix all nodes from the parent node of the current node up to the root node. Layer parameters of the network. Collect the training dataset containing all sample scene information corresponding to the node and its parent nodes up to the root node, or select a training dataset from the existing training dataset that satisfies all the above sample scene information. The network layers are cascaded with the output layer, and the neural network is trained based on the training dataset.

[0165] If the network layer parameters corresponding to the other child nodes of the parent node have been trained, the parameters of this node can be initialized to the network layer parameters corresponding to the other child nodes of its parent node.

[0166] To facilitate understanding of the growth process of a network tree, an exemplary embodiment is provided: Step a. Preset the sample scene information corresponding to each network layer, such as network layer 1 covering the entire scene, network layer 2 for frequency points, and network layer 3 for bandwidth.

[0167] Step b. Grow the tree starting from the root node, which is also a leaf node initially. When the performance of a leaf node (after model training) falls below a preset threshold in the corresponding sample scenario information... If the leaf node is the parent node, a new leaf node is generated based on the sample scene information corresponding to the next layer of the network. The leaf node corresponds to the possible values ​​of the sample scene information (if it is a continuous value, a preset segment interval is used, and each interval corresponds to a leaf node). If the leaf node is already located in the last layer of the network, no more leaf nodes will be generated.

[0168] S403 uses a digital predistortion model to perform digital predistortion processing on the sample input signal based on the sample scene information to obtain the predicted predistortion signal.

[0169] The details of step S403 can be found in the above embodiments and will not be repeated here.

[0170] S404 uses a sub-model obtained by combining network nodes in each network layer after training that match the sample scene information to perform power amplification processing on the predicted predistortion signal in order to obtain the first predicted output signal.

[0171] In this embodiment, a sub-model is obtained by combining network nodes in each trained network layer that match the sample scene information. This sub-model is then used to amplify the predicted predistortion signal to obtain the first predicted output signal. Thus, considering the sample scene information, network nodes for power amplification are adaptively selected, and these selected network nodes are combined to form a sub-model. This sub-model can accurately characterize the nonlinear characteristics of the power amplifier hardware in a specific scenario, thereby helping to improve the training accuracy of the digital predistortion model.

[0172] For example, continue with Figure 7 For example.

[0173] In the first case, if the sample scene information includes the power amplifier hardware operating frequency band of 2.6GHz and the sample input signal bandwidth of 20Mbps, then the sub-model A obtained by combining the trained network nodes 1, 2, and 4 will perform power amplification processing on the predicted predistortion signal to obtain the first predicted output signal.

[0174] In the second scenario, if the sample scene information includes the power amplifier hardware operating frequency band of 2.6GHz and the sample input signal bandwidth of 50Mbps, then the sub-model B obtained by combining the trained network nodes 1, 2, and 5 will perform power amplification processing on the predicted predistortion signal to obtain the first predicted output signal.

[0175] In the third case, if the sample scene information includes the power amplifier hardware operating frequency band of 4.9GHz and the sample input signal bandwidth of 20Mbps, then the sub-model C obtained by combining the trained network nodes 1, 3, and 6 will perform power amplification processing on the predicted predistortion signal to obtain the first predicted output signal.

[0176] In the fourth case, if the sample scene information includes the power amplifier hardware operating frequency band of 4.9GHz and the sample input signal bandwidth of 50Mbps, then the sub-model D obtained by combining the trained network nodes 1, 3, and 7 will perform power amplification processing on the predicted predistortion signal to obtain the first predicted output signal.

[0177] S405, Based on the first difference information between the sample input signal and the first predicted output signal, the digital predistortion model is trained to obtain the trained digital predistortion model.

[0178] The details of step S405 can be found in the above embodiments and will not be repeated here.

[0179] It should be noted that, Figure 4 In the embodiments, the sub-models and network nodes are all network structures in the power amplifier model.

[0180] This disclosure does not impose any restrictions on the execution sequence of steps S401-S405. Figure 4 The example only demonstrates the execution of steps S401-S405 in sequence.

[0181] The digital predistortion model training method provided in this disclosure uses a sub-model obtained by combining network nodes in each network layer that match the sample scene information to perform power amplification processing on the sample input signal to obtain a second predicted output signal. Thus, considering the sample scene information, the network nodes used for power amplification processing can be adaptively selected, and the selected network nodes can be combined to obtain a sub-model. This sub-model can accurately characterize the nonlinear characteristics of the power amplifier hardware in a specific scene.

[0182] In addition, taking into account the second difference information, local optimization of the parameters of the sub-model obtained by combining specific network nodes in the power amplifier model can help improve the training efficiency of the power amplifier model.

[0183] Furthermore, a sub-model is obtained by combining network nodes in each trained network layer that match the sample scene information. This sub-model then performs power amplification on the predicted predistortion signal to obtain the first predicted output signal. Thus, considering the sample scene information, network nodes for power amplification can be adaptively selected, and these selected nodes can be combined to obtain a sub-model. This sub-model can accurately characterize the nonlinear characteristics of the power amplifier hardware in a specific scenario, thereby helping to improve the training accuracy of the digital predistortion model.

[0184] Figure 5 This is a flowchart illustrating a signal transmission method according to an exemplary embodiment, such as... Figure 5 As shown, the signal transmission method of this disclosure includes the following steps.

[0185] S501 uses a trained digital predistortion model to perform digital predistortion processing on the original transmitted signal based on real-world scene information to obtain a predistorted signal.

[0186] S502 amplifies the predistorted signal using power amplifier hardware to obtain the target transmission signal; the nonlinear characteristics of the power amplifier hardware are related to the actual scene information.

[0187] S503 transmits the target transmission signal to the communication receiver.

[0188] It should be noted that the execution subject of the signal transmission method in this embodiment is an electronic device, such as a communication transmitter. The communication transmitter can be a base station, terminal device, vehicle, robot, server, chip, etc. Terminal devices include mobile phones, wearable devices (such as smartwatches, smart glasses), laptops, etc., and vehicles include vehicle terminals, vehicle controllers, etc. The signal transmission method in this embodiment can be executed by the signal transmission device in this embodiment. The signal transmission device in this embodiment can be configured in any electronic device to execute the signal transmission method in this embodiment.

[0189] This disclosure proposes a "multi-scenario, one-model" digital predistortion processing mechanism. Based on actual scenario information, the trained digital predistortion model performs digital predistortion processing on the original transmitted signal to obtain a predistorted signal. The trained digital predistortion model in this scheme has scene awareness and adaptive reasoning capabilities, which can meet the nonlinear compensation requirements of power amplifier hardware in different scenarios. Only one general digital predistortion model needs to be trained and stored, which can significantly improve the training efficiency of the digital predistortion model and significantly reduce the storage overhead of the digital predistortion model.

[0190] In this embodiment, the trained digital predistortion model performs digital predistortion processing on the original transmitted signal based on actual scene information to obtain a predistorted signal, including at least one of the following: Method 1: The actual scene information and the original transmitted signal are spliced ​​together to obtain a third spliced ​​signal. The third spliced ​​signal is then processed by a trained digital predistortion model to obtain a predistortion signal.

[0191] Therefore, the actual scene information and the original transmitted signal are spliced ​​together to obtain a third spliced ​​signal. The third spliced ​​signal is then processed by the trained digital predistortion model to obtain a predistorted signal. The splicing operation is a lightweight fusion operation with low computational complexity, which makes the digital predistortion model more lightweight, significantly reduces the amount of computation required for digital preprocessing, helps to improve signal transmission efficiency, and can reduce the storage overhead of the digital predistortion model.

[0192] In one possible implementation, the digital predistortion model includes network layer A, network layer B, and network layer C; the trained digital predistortion model is used to process the third spliced ​​signal to obtain a predistorted signal, including extracting features of the actual scene information in the third spliced ​​signal through the trained network layer A to obtain a fourth scene feature, extracting features of the original transmitted signal in the third spliced ​​signal through the trained network layer B to obtain a fourth signal feature, and processing the fourth scene feature and the fourth signal feature through the trained network layer C to obtain the predistorted signal.

[0193] For example, the fourth scene features and the fourth signal features are processed by the trained network layer C to obtain the predistorted signal. This includes fusing the fourth scene features and the fourth signal features by the trained network layer C to obtain the fourth fused feature, and determining the predistorted signal based on the fourth fused feature by the trained network layer C.

[0194] Method 2: Based on the actual scene information, determine the fourth value of the parameters of the trained digital predistortion model, configure the parameters of the trained digital predistortion model to the fourth value, update the trained digital predistortion model, and perform digital predistortion processing on the original transmitted signal through the updated digital predistortion model to obtain the predistorted signal.

[0195] Therefore, the parameters of the digital predistortion model can be adaptively adjusted by taking into account the actual scene information. That is, the parameters of the digital predistortion model change with the actual scene information, which can accurately compensate for the nonlinear characteristics of the power amplifier hardware in a specific scene.

[0196] In one possible implementation, the digital predistortion model includes a supernetwork and a main network. The trained supernetwork determines a fourth value for the third parameter of the main network based on real-world scene information, and configures the third parameter to this fourth value to update the main network. The third parameter is associated with the real-world scene information. The updated main network then performs digital predistortion processing on the original transmitted signal to obtain a predistorted signal. Therefore, the trained supernetwork can adaptively adjust the value of the third parameter of the main network based on the real-world scene information; that is, the third parameter changes with the real-world scene information, which can accurately compensate for the nonlinear characteristics of the power amplifier hardware in specific scenarios.

[0197] In one possible implementation, the parameters of the main network also include a fourth parameter, which is unrelated to the actual scene information.

[0198] Method 3: The original transmitted signal is digitally predistorted by using network nodes in the trained digital predistortion model that match the information of the actual scene to obtain the predistorted signal.

[0199] Therefore, the digital predistortion model includes network nodes that match the actual scene information. It can take into account the actual scene information and adaptively select network nodes for digital predistortion processing, which can accurately compensate for the nonlinear characteristics of power amplifier hardware in specific scenarios.

[0200] In this embodiment, the digital predistortion model includes network nodes that match the actual scene information, and the network nodes that match different actual scene information may be different.

[0201] In one possible implementation, the method further includes determining the network nodes corresponding to the actual scene information in the trained digital predistortion model, as the network nodes in the trained digital predistortion model that match the actual scene information. It is understood that the correspondence between scene information and network nodes can be established in advance, and the network nodes corresponding to different scene information may be different.

[0202] In one possible implementation, the method further includes determining the network nodes corresponding to the value ranges of the actual scene information in the trained digital predistortion model, and using these as the network nodes in the trained digital predistortion model that match the actual scene information. It is understood that the correspondence between the value ranges of scene information and network nodes can be established in advance, and different value ranges of scene information may correspond to different network nodes.

[0203] In one possible implementation, the trained digital predistortion model is obtained by training using the digital predistortion model training method provided in this disclosure.

[0204] In one possible implementation, the actual scenario information includes at least one of the following: The operating frequency band of the power amplifier hardware; Operating frequency of power amplifier hardware; The bandwidth of the original transmitted signal; Temperature of the power amplifier hardware.

[0205] It should be noted that the relevant content of step S501 can be referred to the relevant content of step S102, and will not be repeated here.

[0206] The predistorted signal is amplified by power amplifier hardware to obtain the target transmission signal. This can be achieved using any power amplifier hardware in related technologies, and no further restrictions are imposed here.

[0207] Transmitting a target signal to a communication receiver can be achieved using any signal transmission method from relevant technologies; no specific limitations are imposed here. For example, a target signal can be transmitted to the communication receiver via an antenna.

[0208] This disclosure does not impose any restrictions on the execution sequence of steps S501-S503. Figure 5 The example only demonstrates the execution of steps S501-S503 in sequence.

[0209] The embodiments of this disclosure provide a training method for a digital predistortion model, proposing a "multi-scenario, one-model" digital predistortion processing mechanism. Based on actual scenario information, the trained digital predistortion model performs digital predistortion processing on the original transmitted signal to obtain a predistorted signal. The trained digital predistortion model in this scheme has scene awareness and adaptive reasoning capabilities, which can meet the nonlinear compensation requirements of power amplifier hardware in different scenarios. Only one general digital predistortion model needs to be trained and stored, which can significantly improve the training efficiency of the digital predistortion model and significantly reduce the storage overhead of the digital predistortion model.

[0210] Figure 8 This is a schematic diagram illustrating the structure of a training device for a digital predistortion model according to an exemplary embodiment.

[0211] Reference Figure 8 The training device 800 for the digital predistortion model in this embodiment includes: a first training module 801, a first processing module 802, a second processing module 803, and a second training module 804.

[0212] The first training module 801 is configured to train the power amplifier model based on the sample input signal, sample output signal, and sample scene information of the power amplifier hardware to obtain the trained power amplifier model; wherein, the power amplifier model is a mathematical model of the power amplifier hardware, and the nonlinear characteristics of the power amplifier hardware are related to the sample scene information. The first processing module 802 is configured to perform digital predistortion processing on the sample input signal based on the sample scene information using a digital predistortion model to obtain a predicted predistortion signal. The second processing module 803 is configured to perform power amplification processing on the predicted predistortion signal in association with the sample scene information through the trained power amplifier model to obtain a first predicted output signal. The second training module 804 is configured to train the digital predistortion model based on the first difference information between the sample input signal and the first predicted output signal to obtain the trained digital predistortion model.

[0213] In some possible implementations, the first training module 801 is further configured to: perform power amplification processing on the sample input signal associated with the sample scene information through the power amplifier model to obtain a second predicted output signal; and train the power amplifier model based on the second difference information between the second predicted output signal and the sample output signal to obtain the trained power amplifier model.

[0214] In some possible implementations, the first training module 801 is further configured to: concatenate the sample scene information and the sample input signal to obtain a first concatenated signal; and process the first concatenated signal through the power amplifier model to obtain a second predicted output signal.

[0215] In some possible implementations, the power amplifier model includes a first network layer, a second network layer, and a third network layer; the first training module 801 is further configured to: extract features from the sample scene information in the first spliced ​​signal through the first network layer to obtain a first scene feature; extract features from the sample input signal in the first spliced ​​signal through the second network layer to obtain a first signal feature; and process the first scene feature and the first signal feature through the third network layer to obtain a second predicted output signal.

[0216] In some possible implementations, the second processing module 803 is further configured to: splice the sample scene information and the predicted predistortion signal to obtain a second spliced ​​signal; and process the second spliced ​​signal through the trained power amplifier model to obtain the first predicted output signal.

[0217] In some possible implementations, the power amplifier model includes a supernetwork and a main network; the first training module 801 is further configured to: determine a first value of a first parameter of the main network based on the sample scene information through the supernetwork, and configure the first parameter to the first value to update the main network; wherein the first parameter is associated with the sample scene information; and perform power amplification processing on the sample input signal through the updated main network to obtain the second predicted output signal.

[0218] In some possible implementations, the first training module 801 is further configured to: determine the gradient of the first parameter based on the second difference information; and update the parameters of the supernetwork based on the gradient of the first parameter to obtain the trained supernetwork.

[0219] In some possible implementations, the parameters of the main network further include a second parameter, which is unrelated to the sample scene information; the first training module 801 is further configured to: determine the gradient of the second parameter based on the second difference information; and update the second parameter based on the gradient of the second parameter to update the main network.

[0220] In some possible implementations, the second processing module 803 is further configured to: determine a second value of the first parameter based on the sample scene information using the trained supernetwork, and configure the first parameter as the second value to update the main network; and perform power amplification processing on the predicted predistortion signal using the updated main network to obtain the first predicted output signal.

[0221] In some possible implementations, the power amplifier model adopts a network tree architecture, the power amplifier model includes multiple sub-models, each network layer of the power amplifier model includes network nodes that match the sample scene information, and at least two sub-models share network nodes; the first training module 801 is further configured to: perform power amplification processing on the sample input signal through the sub-models obtained by combining the network nodes that match the sample scene information in each network layer, so as to obtain the second predicted output signal.

[0222] In some possible implementations, the first training module 801 is further configured to: update the parameters of the sub-model obtained by combining network nodes that match the sample scene information in each network layer based on the second difference information, so as to obtain the corresponding trained sub-model.

[0223] In some possible implementations, the second processing module 803 is further configured to: perform power amplification processing on the predicted predistortion signal using a sub-model obtained by combining network nodes in each of the trained network layers that match the sample scene information, so as to obtain the first predicted output signal.

[0224] In some possible implementations, the parameters of the trained power amplifier model are fixed during the training of the digital predistortion model.

[0225] In some possible implementations, the sample scenario information includes at least one of the following: The operating frequency band of the power amplifier hardware; The operating frequency of the power amplifier hardware; The bandwidth of the sample input signal; The temperature of the power amplifier hardware.

[0226] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0227] The digital predistortion model training device provided in the embodiments of this disclosure proposes a general modeling mechanism of "one model for multiple scenarios". Taking into account the sample input signal, sample output signal and sample scene information of the power amplifier hardware, the power amplifier model is trained to obtain the trained power amplifier model. During the training process, the power amplifier model can learn to perform power amplification processing on the sample input signal in association with the sample scene information to obtain the sample output signal, and can learn the nonlinear characteristics of the power amplifier hardware under different scenarios. That is, the trained power amplifier model in this scheme has scene perception and adaptive reasoning capabilities. Only one general power amplifier model needs to be trained and stored, which can greatly improve the training efficiency of the power amplifier model and significantly reduce the storage overhead of the power amplifier model.

[0228] Furthermore, by using a digital predistortion model to perform digital predistortion processing on the sample input signal based on sample scene information, a predicted predistortion signal can be obtained. During the training process, the digital predistortion model can learn the ability to perform digital predistortion processing on the sample input signal based on sample scene information. That is, the digital predistortion model trained in this scheme has scene awareness and adaptive reasoning capabilities, which can meet the nonlinear compensation requirements of power amplifier hardware in different scenarios. Only one general digital predistortion model needs to be trained and stored, which can significantly improve the training efficiency of the digital predistortion model and significantly reduce the storage overhead of the digital predistortion model.

[0229] Figure 9 This is a schematic diagram of the structure of a signal transmitting device according to an exemplary embodiment.

[0230] Reference Figure 9 The signal transmitting device 900 of this embodiment includes: a first processing module 901, a second processing module 902 and a transmitting module 903.

[0231] The first processing module 901 is configured to perform digital predistortion processing on the original transmitted signal based on actual scene information using a trained digital predistortion model to obtain a predistorted signal. The second processing module 902 is configured to amplify the predistorted signal using power amplifier hardware to obtain the target transmission signal; wherein the nonlinear characteristics of the power amplifier hardware are associated with the actual scene information. The transmitting module 903 is configured to transmit the target transmission signal to the communication receiving end.

[0232] In some possible implementations, the trained digital predistortion model is obtained by training the digital predistortion model using the training method provided in this disclosure.

[0233] In some possible implementations, the actual scenario information includes at least one of the following: The operating frequency band of the power amplifier hardware; The operating frequency of the power amplifier hardware; The bandwidth of the original transmitted signal; The temperature of the power amplifier hardware.

[0234] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0235] The signal transmitting apparatus provided in the embodiments of this disclosure proposes a "multi-scenario, one-model" digital predistortion processing mechanism. Based on actual scene information, the trained digital predistortion model performs digital predistortion processing on the original transmitted signal to obtain a predistorted signal. The trained digital predistortion model in this scheme has scene awareness and adaptive reasoning capabilities, which can meet the nonlinear compensation requirements of power amplifier hardware in different scenarios. Only one general digital predistortion model needs to be trained and stored, which can significantly improve the training efficiency of the digital predistortion model and significantly reduce the storage overhead of the digital predistortion model.

[0236] To implement the above embodiments, this disclosure also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the training method for the digital predistortion model provided in this disclosure, and / or the steps of the signal transmission method provided in this disclosure.

[0237] Figure 10 This is a schematic diagram illustrating the structure of an electronic device according to an exemplary embodiment. For example, the electronic device 1000 may be a vehicle, mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0238] Reference Figure 10 The electronic device 1000 may include one or more of the following components: a processing component 1002, a memory 1004, a power component 1006, a multimedia component 1008, an audio component 1010, an input / output (I / O) interface 1012, a sensor component 1014, and a communication component 1016.

[0239] Processing component 1002 typically controls the overall operation of electronic device 1000, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. Processing component 1002 may include one or more processors 1020 to execute instructions to complete all or part of the steps of the above-described digital predistortion model training method, and / or complete all or part of the steps of the above-described signal transmission method. Furthermore, processing component 1002 may include one or more modules to facilitate interaction between processing component 1002 and other components. For example, processing component 1002 may include a multimedia module to facilitate interaction between multimedia component 1008 and processing component 1002.

[0240] Memory 1004 is configured to store various types of data to support the operation of electronic device 1000. Examples of this data include instructions for any application or method operating on electronic device 1000, contact data, phonebook data, messages, pictures, videos, etc. Memory 1004 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.

[0241] Power component 1006 provides power to various components of electronic device 1000. Power component 1006 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 1000.

[0242] Multimedia component 1008 includes a screen that provides an output interface between electronic device 1000 and user. In some embodiments, the screen may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a Touch Panel, the screen may be implemented as a touchscreen to receive input signals from the user. The Touch Panel includes one or more touch sensors to sense touches, swipes, and gestures on the Touch Panel. The touch sensors may sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 1008 includes a front-facing camera and / or a rear-facing camera. When electronic device 1000 is in an operating mode, such as a shooting mode or video mode, the front-facing camera and / or rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0243] Audio component 1010 is configured to output and / or input audio signals. For example, audio component 1010 includes a microphone (MIC) configured to receive external audio signals when electronic device 1000 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 1004 or transmitted via communication component 1016. In some embodiments, audio component 1010 also includes a speaker for outputting audio signals.

[0244] I / O interface 1012 provides an interface between processing component 1002 and peripheral interface modules, which may be keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, start buttons, and lock buttons.

[0245] Sensor assembly 1014 includes one or more sensors for providing state assessment of various aspects of electronic device 1000. For example, sensor assembly 1014 may detect the on / off state of electronic device 1000, the relative positioning of components such as the display and keypad of electronic device 1000, changes in position of electronic device 1000 or a component of electronic device 1000, the presence or absence of user contact with electronic device 1000, orientation or acceleration / deceleration of electronic device 1000, and temperature changes of electronic device 1000. Sensor assembly 1014 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 1014 may also include an optical sensor, such as a complementary metal-oxide-semiconductor (CMOS) or charge-coupled device (CCD) image sensor, for use in imaging applications. In some embodiments, sensor assembly 1014 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0246] Communication component 1016 is configured to facilitate wired or wireless communication between electronic device 1000 and other devices. Electronic device 1000 can access wireless networks based on communication standards, such as WiFi, 4G, or 5G, or combinations thereof. In one exemplary embodiment, communication component 1016 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 1016 also includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Ultra-Wideband (UWB), Bluetooth, and other technologies.

[0247] In an exemplary embodiment, the electronic device 1000 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the steps of the training method for the digital predistortion model described above, and / or the steps of the signal transmission method described above.

[0248] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 1004 including instructions, which can be executed by a processor 1020 of an electronic device 1000 to complete the training method of the digital predistortion model and / or the signal transmission method described above. For example, the non-transitory computer-readable storage medium may be a read-only memory (ROM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0249] To implement the above embodiments, this disclosure also proposes a non-transitory computer-readable storage medium storing computer program instructions thereon, which, when executed by a processor, implement the steps of the training method for the digital predistortion model provided in this disclosure, and / or the steps of the signal transmission method provided in this disclosure.

[0250] To implement the above embodiments, this disclosure also proposes a chip including an interface circuit and a processing circuit coupled to each other. The interface circuit is used to input or output signals, and the processing circuit is configured to implement the steps of the training method of the digital predistortion model provided in this disclosure, and / or the steps of the signal transmission method provided in this disclosure.

[0251] Figure 11 This is a schematic diagram illustrating the structure of a chip according to an exemplary embodiment. See also... Figure 11 The diagram shown is a schematic representation of the structure of chip 1100, but it is not limited to this.

[0252] Chip 1100 includes processing circuit 1101, which is configured to perform the steps of any of the above-described digital predistortion model training methods and / or the steps of any of the above-described signal transmission methods.

[0253] In some embodiments, chip 1100 further includes one or more interface circuits 1102. In some possible embodiments, interface circuit 1102 is connected to memory 1103, and interface circuit 1102 can be used to receive signals from memory 1103 or other devices, and interface circuit 1102 can be used to send signals to memory 1103 or other devices. For example, interface circuit 1102 can read instructions stored in memory 1103 and send the instructions to processing circuit 1101.

[0254] In some embodiments, the interface circuit 1102 performs at least one of the communication steps such as sending and / or receiving in the above method, while the processing circuit 1101 performs other steps.

[0255] In some embodiments, the terms interface circuit, interface, transceiver pin, transceiver, etc., can be used interchangeably.

[0256] In some embodiments, chip 1100 further includes one or more memories 1103 for storing instructions. In some possible implementations, all or part of the memories 1103 may be located outside of chip 1100.

[0257] To implement the above embodiments, this disclosure also proposes a computer program product, including a computer program that, when executed by a processor, implements the steps of the training method for the digital predistortion model provided in this disclosure, and / or the steps of the signal transmission method provided in this disclosure.

[0258] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0259] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0260] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of this disclosure includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of this disclosure pertain.

[0261] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and compact disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0262] It should be understood that various parts of this disclosure can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0263] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0264] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0265] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present disclosure have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present disclosure. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present disclosure.

Claims

1. A training method for a digital predistortion model, characterized in that, include: Based on the sample input signal, sample output signal, and sample scene information of the power amplifier hardware, the power amplifier model is trained to obtain the trained power amplifier model; wherein, the power amplifier model is the mathematical model of the power amplifier hardware, and the nonlinear characteristics of the power amplifier hardware are related to the sample scene information; Based on the sample scene information, the sample input signal is digitally predistorted using a digital predistortion model to obtain a predicted predistortion signal. The predicted predistortion signal is amplified by the trained power amplifier model and associated with the sample scene information to obtain the first predicted output signal. Based on the first difference information between the sample input signal and the first predicted output signal, the digital predistortion model is trained to obtain the trained digital predistortion model.

2. The method according to claim 1, characterized in that, The power amplifier model is trained using the sample input signal, sample output signal, and sample scene information based on the power amplifier hardware to obtain the trained power amplifier model, including: The sample input signal is amplified by the power amplifier model and associated with the sample scene information to obtain a second predicted output signal. The power amplifier model is trained based on the second difference information between the second predicted output signal and the sample output signal to obtain the trained power amplifier model.

3. The method according to claim 2, characterized in that, The step of performing power amplification processing on the sample input signal, which is associated with the sample scene information, through the power amplifier model to obtain the second predicted output signal includes: The sample scene information and the sample input signal are spliced ​​together to obtain a first spliced ​​signal; The first spliced ​​signal is processed using the power amplifier model to obtain the second predicted output signal.

4. The method according to claim 3, characterized in that, The power amplifier model includes a first network layer, a second network layer, and a third network layer; The step of processing the first spliced ​​signal through the power amplifier model to obtain the second predicted output signal includes: The first scene features are obtained by extracting features from the sample scene information in the first spliced ​​signal through the first network layer. The second network layer extracts features from the sample input signal in the first spliced ​​signal to obtain the first signal features; The third network layer processes the first scene features and the first signal features to obtain the second predicted output signal.

5. The method according to claim 3, characterized in that, The step of amplifying the predicted predistortion signal using the trained power amplifier model, in association with the sample scene information, to obtain a first predicted output signal includes: The sample scene information and the predicted predistortion signal are spliced ​​together to obtain a second spliced ​​signal; The second spliced ​​signal is processed by the trained power amplifier model to obtain the first predicted output signal.

6. The method according to claim 2, characterized in that, The power amplifier model includes a supernetwork and a main network; the step of performing power amplification processing on the sample input signal, which is associated with the sample scene information, through the power amplifier model to obtain a second predicted output signal includes: The supernetwork determines a first value for a first parameter of the main network based on the sample scene information, and configures the first parameter to the first value to update the main network; wherein the first parameter is associated with the sample scene information. The sample input signal is amplified by the updated main network to obtain the second predicted output signal.

7. The method according to claim 6, characterized in that, The step of training the power amplifier model based on the second difference information between the second predicted output signal and the sample output signal to obtain the trained power amplifier model includes: Based on the second difference information, the gradient of the first parameter is determined; Based on the gradient of the first parameter, the parameters of the supernetwork are updated to obtain the trained supernetwork.

8. The method according to claim 6, characterized in that, The parameters of the main network also include a second parameter, which is unrelated to the sample scene information; The step of training the power amplifier model based on the second difference information between the second predicted output signal and the sample output signal to obtain the trained power amplifier model includes: Based on the second difference information, the gradient of the second parameter is determined; The second parameter is updated based on its gradient to update the main network.

9. The method according to claim 6, characterized in that, The step of amplifying the predicted predistortion signal using the trained power amplifier model, in association with the sample scene information, to obtain a first predicted output signal includes: The trained supernetwork determines a second value for the first parameter based on the sample scene information, and configures the first parameter to the second value to update the main network; The predicted predistortion signal is amplified by the updated main network to obtain the first predicted output signal.

10. The method according to claim 2, characterized in that, The power amplifier model adopts a network tree architecture, and the power amplifier model includes multiple sub-models. Each network layer of the power amplifier model includes network nodes that match the sample scene information, and at least two sub-models share network nodes. The step of performing power amplification processing on the sample input signal, which is associated with the sample scene information, through the power amplifier model to obtain the second predicted output signal includes: The sample input signal is amplified by a sub-model obtained by combining network nodes in each network layer that match the sample scene information, in order to obtain the second predicted output signal.

11. The method according to claim 10, characterized in that, The step of training the power amplifier model based on the second difference information between the second predicted output signal and the sample output signal to obtain the trained power amplifier model includes: Based on the second difference information, the parameters of the sub-models obtained by combining network nodes in each network layer that match the sample scene information are updated to obtain the corresponding trained sub-models.

12. The method according to claim 10, characterized in that, The step of amplifying the predicted predistortion signal using the trained power amplifier model, in association with the sample scene information, to obtain a first predicted output signal includes: The predicted predistortion signal is amplified by a sub-model obtained by combining network nodes that match the sample scene information in each of the trained network layers to obtain the first predicted output signal.

13. The method according to any one of claims 1-12, characterized in that, During the training of the digital predistortion model, the parameters of the trained power amplifier model are fixed.

14. The method according to any one of claims 1-12, characterized in that, The sample scene information includes at least one of the following: The operating frequency band of the power amplifier hardware; The operating frequency of the power amplifier hardware; The bandwidth of the sample input signal; The temperature of the power amplifier hardware.

15. A signal transmission method, characterized in that, include: Based on real-world scenario information, the trained digital predistortion model performs digital predistortion processing on the original transmitted signal to obtain a predistorted signal. The predistorted signal is amplified by power amplifier hardware to obtain the target transmission signal; wherein the nonlinear characteristics of the power amplifier hardware are related to the actual scene information. The target transmission signal is transmitted to the communication receiving end.

16. The method according to claim 15, characterized in that, The trained digital predistortion model is obtained by training using the method described in any one of claims 1-14.

17. The method according to claim 15, characterized in that, The actual scenario information includes at least one of the following: The operating frequency band of the power amplifier hardware; The operating frequency of the power amplifier hardware; The bandwidth of the original transmitted signal; The temperature of the power amplifier hardware.

18. A training device for a digital predistortion model, characterized in that, include: The first training module is configured to train the power amplifier model based on the sample input signal, sample output signal, and sample scene information of the power amplifier hardware, so as to obtain the trained power amplifier model; wherein, the power amplifier model is a mathematical model of the power amplifier hardware, and the nonlinear characteristics of the power amplifier hardware are related to the sample scene information; The first processing module is configured to perform digital predistortion processing on the sample input signal based on the sample scene information using a digital predistortion model to obtain a predicted predistortion signal. The second processing module is configured to perform power amplification processing on the predicted predistortion signal in association with the sample scene information through the trained power amplifier model to obtain the first predicted output signal. The second training module is configured to train the digital predistortion model based on the first difference information between the sample input signal and the first predicted output signal, so as to obtain the trained digital predistortion model.

19. The apparatus according to claim 18, characterized in that, The first training module is also configured as follows: The sample input signal is amplified by the power amplifier model and associated with the sample scene information to obtain a second predicted output signal. The power amplifier model is trained based on the second difference information between the second predicted output signal and the sample output signal to obtain the trained power amplifier model.

20. The apparatus according to claim 19, characterized in that, The first training module is also configured as follows: The sample scene information and the sample input signal are spliced ​​together to obtain a first spliced ​​signal; The first spliced ​​signal is processed using the power amplifier model to obtain the second predicted output signal.

21. A signal transmitting device, characterized in that, include: The first processing module is configured to perform digital predistortion processing on the original transmitted signal based on the actual scene information using a trained digital predistortion model to obtain a predistorted signal. The second processing module is configured to amplify the predistorted signal using power amplifier hardware to obtain the target transmission signal; wherein the nonlinear characteristics of the power amplifier hardware are associated with the actual scene information. The transmitting module is configured to transmit the target transmission signal to the communication receiving end.

22. The apparatus according to claim 21, characterized in that, The trained digital predistortion model is obtained by training using the method described in any one of claims 1-14.

23. The apparatus according to claim 21, characterized in that, The actual scenario information includes at least one of the following: The operating frequency band of the power amplifier hardware; The operating frequency of the power amplifier hardware; The bandwidth of the original transmitted signal; The temperature of the power amplifier hardware.

24. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the steps of the method according to any one of claims 1-17.

25. A non-transitory computer-readable storage medium having computer program instructions stored thereon, characterized in that, When executed by a processor, the program instructions implement the steps of the method described in any one of claims 1-17.

26. A chip, characterized in that, The chip includes an interface circuit and a processing circuit coupled to each other. The interface circuit is used to input or output signals, and the processing circuit is configured to implement the steps of the method according to any one of claims 1-17.

27. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1-17.

Images