A pre-distortion method, system, device and storage medium

Abstract

The application provides a pre-distortion method, system, device and storage medium. The method is applied to a pre-distortion system, the pre-distortion system comprises a pre-distortion multiplier, a complex neural network and a radio frequency power amplifier output feedback loop, and the method comprises the following steps: inputting a training complex signal into the pre-distortion system and outputting a corresponding complex scalar; training the pre-distortion system based on the training complex signal and the complex scalar until the generalization error vector amplitude and the generalization adjacent channel leakage ratio of the pre-distortion system reach a set requirement; and inputting a service complex signal into the trained pre-distortion system to obtain a pre-distortion corrected complex scalar. The above technical scheme effectively solves the problem of insufficient nonlinear representation capability of traditional pre-distortion models and power amplifier models, and solves the problem of poor pre-distortion correction effect caused by the lag of the dynamic change reaction of the power amplifier system due to the separate processing when the neural network is applied.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication, for example, to a predistortion method, system, device and storage medium. BACKGROUND

[0002] The power amplifier is a key component related to energy consumption and signal quality. The power amplifier model and the predistortion model have become a key technology for efficient and high data rate communication.

[0003] The ideal predistortion model and the nonlinear characteristics of the power amplifier model are strictly reciprocal in mathematics, and the predistortion model and the power amplifier model also belong to typical nonlinear fitting problems. However, the nonlinear representation ability of the traditional predistortion model and the power amplifier model is inherently insufficient. SUMMARY

[0004] The present application provides a predistortion method, system, device and storage medium, which solves the problem of insufficient nonlinear representation ability of the traditional predistortion model and the power amplifier model, and also solves the problem of poor predistortion correction effect caused by the lag of the dynamic change reaction of the power amplifier system due to the separate processing when applying the neural network.

[0005] In a first aspect, the embodiments of the present application provide a predistortion method applied to a predistortion system, the predistortion system comprising a predistortion multiplier, a complex neural network and a radio frequency power amplifier output feedback loop; the method comprising:

[0006] inputting a training complex signal into the predistortion system, and outputting a corresponding complex scalar;

[0007] training the predistortion system based on the training complex signal and the complex scalar until the generalization error vector amplitude and the generalization adjacent channel leakage ratio of the predistortion system reach a set requirement;

[0008] inputting a service complex signal into the trained predistortion system to obtain a predistortion corrected complex scalar.

[0009] In a second aspect, the embodiments of the present application provide a predistortion system for executing the predistortion method provided by the embodiments of the present application, the predistortion system comprising a predistortion multiplier, a complex neural network and a radio frequency power amplifier output feedback loop;

[0010] The first input end of the pre-distortion multiplier is the input end of the pre-distortion system, and is connected with the first input end and the second input end of the complex neural network; the output end of the pre-distortion multiplier is connected with the input end of the radio frequency power amplifier output feedback loop; the output end of the radio frequency power amplifier output feedback loop is the output end of the pre-distortion system, and is connected with the second input end of the complex neural network; and the output end of the complex neural network is connected with the second input end of the pre-distortion multiplier.

[0011] In a third aspect, an embodiment of the present application provides a pre-distortion system for executing the pre-distortion method provided in the embodiments of the present application, and the pre-distortion system comprises a pre-distortion multiplier, a complex neural network, a radio frequency power amplifier output feedback loop, a first real-time power normalization unit and a second real-time power normalization unit.

[0012] The input end of the first real-time power normalization unit is the input end of the pre-distortion system; the output ends of the first real-time power normalization unit are respectively connected with the first input end of the pre-distortion multiplier, the first input end of the complex neural network and the second input end of the complex neural network; the output end of the pre-distortion multiplier is connected with the input end of the radio frequency power amplifier output feedback loop; the output end of the radio frequency power amplifier output feedback loop is the output end of the pre-distortion system, and is connected with the input end of the second real-time power normalization unit; the output end of the second real-time power normalization unit is connected with the second input end of the complex neural network; and the output end of the complex neural network is connected with the second input end of the pre-distortion multiplier.

[0013] In a fourth aspect, an embodiment of the present application provides a device, which comprises one or more processors, a storage apparatus configured to store one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method in the first aspect.

[0014] In a fifth aspect, an embodiment of the present application provides a storage medium, which stores a computer program, and when the computer program is executed by a processor, the computer program implements any of the methods in the embodiments of the present application.

[0015] More details about the above embodiments and other aspects of the present application and implementation manners thereof are provided in the description of drawings, specific embodiments and claims. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flowchart of a pre-distortion method provided in an embodiment of the present application;

[0017] Figure 2A schematic diagram of a traditional memory polynomial-based power amplifier model;

[0018] Figure 3a A schematic diagram of a pre-distortion system according to an embodiment of the present application;

[0019] Figure 3b A schematic diagram of another pre-distortion system according to an embodiment of the present application;

[0020] Figure 4 Performance effect of generalized ACLR according to an embodiment of the present application;

[0021] Figure 5 Improvement effect of generalized ACLR according to an embodiment of the present application;

[0022] Figure 6 A schematic diagram of a device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to make the objectives, technical solutions and advantages of the present application clearer, the embodiments of the present application will be described in detail below with reference to the accompanying drawings. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other as long as there is no conflict.

[0024] The steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Moreover, although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown.

[0025] In one exemplary embodiment, Figure 1 A flowchart of a pre-distortion method according to an embodiment of the present application, which can be applied to a pre-distortion processing case, can be executed by a pre-distortion system, which can be realized by software and / or hardware and integrated on a terminal device.

[0026] A power amplifier (PA) generally has three characteristics: 1) static nonlinearity of the device itself; 2) linear memory effect resulting from a matching network and device component delay; and 3) nonlinear memory effect, which mainly results from a transistor trapping effect and non-ideal characteristics of a bias network, as well as dependence of an input level on temperature change, etc.

[0027] Predistorter (PD) is located before the RF power amplifier, to pre-process the inverse of the nonlinear distortion that will be generated when the signal passes through the power amplifier; the predistortion implemented in the digital domain is called Digital Predistortion (DPD).

[0028] Therefore, the ideal predistortion and the nonlinear characteristics of the power amplifier are strictly inverse in mathematics, and both belong to typical nonlinear fitting problems. However, the series or polynomial model used in the traditional power amplifier nonlinear characteristic fitting has a congenital problem of insufficient representation or fitting ability for complex nonlinearities. Taking the memory polynomial power amplifier model as an example, the mathematical expression of the model is as follows:

[0029]

[0030] Figure 2 The traditional predistortion model based on the memory polynomial is shown in the figure. The power amplifier model is equivalent to a zero-hidden-layer network with only an input layer and an output layer, and it is not even a Multi-Layer Perceptron (MLP). However, it is well known in the field that the expression ability of the network is positively correlated with its structural complexity; this is the fundamental reason for the congenital deficiency of the nonlinear expression ability of the traditional power amplifier model. However, the predistortion and the power amplifier nonlinear are inverse functions of each other, so it can be predicted that when using the traditional series or polynomial model to model DPD, the same bottleneck will inevitably exist.

[0031] Further, the traditional power amplifier model is usually truncated or simplified in various ways, making it even more difficult to accurately characterize the nonlinear characteristics of the power amplifier; it also leads to a large deviation between the output obtained by the parameters solved by the model and the expected output when encountering new data or inputs, that is, the model does not have the universality and generalization ability for new data, and the generalization ability is the most important indicator of neural network performance testing. At the same time, neural networks have much richer structural forms, and their nonlinear expression or fitting ability has been widely verified by academia and industry.

[0032] However, existing methods and technologies that use neural networks for predistortion all use a method that separates the two steps of learning the nonlinear characteristics of the power amplifier and performing predistortion processing, that is: first, the neural network does not access the main link of signal processing, that is, it does not perform predistortion processing on the signal, but only accepts the feedback output of the power amplifier to train or learn the nonlinear characteristics or inverse characteristics of the power amplifier; after the nonlinear characteristic learning is completed, the neural network is connected to the main link or the weights obtained by training are transmitted to a copy of the neural network in the main link, and then the predistortion processing begins.

[0033] The separated processing inevitably causes a problem that the neural network lags behind in response to dynamic changes of the power amplifier system or the learning does not match, that is, the weight values of the neural network trained before the access system reflect only the dynamic characteristics of the system before the current time, and the adaptability to the system dynamics at the current time is uncertain or unguaranteed, so that the pre-distortion correction effect is poor.

[0034] To solve the above technical problems, the present application provides a pre-distortion method, which is applied to a pre-distortion system, the pre-distortion system comprising a pre-distortion multiplier, a complex neural network and a radio frequency power amplifier output feedback loop, as shown in the following figure, the method comprising the following steps: Figure 1

[0035] S110, inputting a training complex signal into the pre-distortion system and outputting a corresponding complex scalar.

[0036] The present application performs pre-distortion correction based on a pre-distortion system, which can be considered as a system capable of realizing pre-distortion correction. The pre-distortion system comprises a power amplifier, and the pre-distortion system can integrate the nonlinear characteristic learning and pre-distortion correction of the power amplifier, that is, the pre-distortion system is directly connected to the power amplifier during training, and the nonlinear characteristic learning and pre-distortion correction of the power amplifier are performed simultaneously, thereby improving the efficiency and accuracy. The pre-distortion system further comprises a pre-distortion multiplier and a complex neural network. The complex neural network can provide the complex coefficient for the pre-distortion multiplier. The pre-distortion multiplier can be considered as a multiplier for realizing pre-distortion function.

[0037] In one embodiment, the pre-distortion system can further comprise a first real-time power normalization unit and a second real-time power normalization unit. Wherein, "first" and "second" are only used to distinguish the real-time power normalization units. The real-time power normalization unit can perform normalization processing. For example, the first real-time power normalization unit can normalize the training complex signal before inputting it into the pre-distortion system. The second power normalization unit can normalize the corresponding complex scalar of the training complex signal before returning it to the pre-distortion system.

[0038] The training complex signal can be considered as a complex signal used for training the pre-distortion system. This step can input the training complex signal into the pre-distortion system to output a corresponding complex scalar, which can be output by the radio frequency power amplifier output feedback loop in the pre-distortion system. The complex scalar and the training complex signal can be used as samples for training the pre-distortion system, and the pre-distortion system can be trained based on the complex scalar and learn the nonlinear characteristics of the power amplifier during the training process.

[0039] ​S120, training the pre-distortion system based on the training complex signal and the complex scalar until the generalization error vector magnitude and the generalization adjacent channel leakage ratio corresponding to the pre-distortion system reach a set requirement.

[0040] The application can input the training complex signal and the complex scalar to the pre-distortion system to train the pre-distortion system. The training can be supervised training. The weight parameters and bias parameters of each layer can be updated through the loss function of the complex neural network.

[0041] The condition for ending the training of the pre-distortion system can be determined based on the generalization error vector magnitude and the generalization adjacent channel leakage ratio. In the case where the generalization error vector magnitude and the generalization adjacent channel leakage ratio reach a set requirement, it can be considered that the training of the pre-distortion system is completed. The set requirement can be set according to actual needs, which is not limited here, such as determining the set requirement based on the pre-distortion index.

[0042] In one embodiment, in the case where the values corresponding to the generalization error vector magnitude and the generalization adjacent channel leakage ratio are greater than or equal to the corresponding set threshold, the training of the pre-distortion system is completed, and in the case where the values corresponding to the generalization error vector magnitude and the generalization adjacent channel leakage ratio are less than the corresponding set threshold, the pre-distortion system is continuously trained.

[0043] The set threshold can be set according to actual needs, which is not limited here. In the case where the generalization error vector magnitude and the generalization adjacent channel leakage ratio do not reach the set requirement, the pre-distortion system can be continuously trained using the training complex signal.

[0044] The application uses the generalization error vector magnitude and the generalization adjacent channel leakage ratio as the condition for ending the training of the pre-distortion system, which can measure and reflect the universality and generalization ability of the pre-distortion system when performing pre-distortion correction. The generalization ability refers to repeatedly inputting the complex signal in the training set into the pre-distortion system to train the system to learn known data in essence, and then inputting new data in the test set unknown to the system to statistically and investigate the adaptability or pre-distortion ability of the system to new input or new data.

[0045] S130, inputting a service complex signal into the trained pre-distortion system to obtain a pre-distortion corrected complex scalar.

[0046] After the training of the pre-distortion system is completed, the service complex signal can be input into the trained pre-distortion system to obtain a pre-distortion corrected complex scalar. The service complex signal can be a service signal in the process of applying the pre-distortion system, which is a complex signal.

[0047] The application provides a pre-distortion method applied to a pre-distortion system, the pre-distortion system comprising a pre-distortion multiplier, a complex neural network and a radio frequency power amplifier output feedback loop, the method comprising the following steps: firstly, inputting a training complex signal into the pre-distortion system to output a corresponding complex scalar; then, training the pre-distortion system based on the training complex signal and the complex scalar until the generalization error vector magnitude and the generalization adjacent channel leakage ratio of the pre-distortion system reach a set requirement; finally, inputting a service complex signal into the trained pre-distortion system to obtain a pre-distortion corrected complex scalar. The pre-distortion system is used to replace the pre-distortion model in the related art, the complex neural network in the pre-distortion system has a rich structure form and strong non-linear expression or fitting capability, and can effectively solve the problem of the innate deficiency of the non-linear expression capability of the traditional pre-distortion model and the power amplifier.

[0048] On the basis of the above-mentioned embodiments, variant embodiments of the above-mentioned embodiments are provided, and it should be noted that, in order to make the description brief, only the differences from the above-mentioned embodiments are described in the variant embodiments.

[0049] In one embodiment, the pre-distortion system is trained using a complex error back propagation algorithm.

[0050] The complex error back propagation algorithm can be regarded as a complex error back propagation algorithm. The error back propagation algorithm learning process comprises two processes of signal forward propagation and error back propagation.

[0051] In one embodiment, the pre-distortion system is trained by:

[0052] initializing system parameters of the pre-distortion system;

[0053] training the pre-distortion system based on a training set and the complex scalar;

[0054] testing the trained pre-distortion system based on a test set to obtain the generalization error vector magnitude and the generalization adjacent channel leakage ratio corresponding to the pre-distortion system;

[0055] in the case that the values corresponding to the generalization error vector magnitude and the generalization adjacent channel leakage ratio are greater than or equal to the corresponding set threshold, the training of the pre-distortion system is completed, and in the case that the values corresponding to the generalization error vector magnitude and the generalization adjacent channel leakage ratio are less than the corresponding set threshold, the pre-distortion system is continuously trained based on the training set;

[0056] The training set and the test set are obtained based on normalized complex vectors, or are obtained based on the training complex signals.

[0057] The system parameters can be considered as parameters required when the pre-distortion system is initialized, and the specific content is not limited here, and can be set according to actual conditions. Exemplarily, the system parameters include but are not limited to: a non-linear order parameter, a power amplifier memory effect parameter, and an initial output of a complex neural network included in the pre-distortion system.

[0058] The initialized system parameters can be considered as setting the system parameters, and the specific values set can be an empirical value or a system parameter determined based on history.

[0059] The training set can be considered as a set of complex signals for training the pre-distortion system. The test set can be considered as a set of complex signals for verifying the generalization performance of the pre-distortion system. The complex signals included in the training set and the test set are different, and the combination of the two is the training complex signals.

[0060] The complex vector input into the pre-distortion system from each part of the training set, and the complex scalar output by the pre-distortion system corresponding to the complex vector, are one-to-one corresponding; when training the pre-distortion system, it is based on the element corresponding to the current value in the complex vector and the complex scalar to calculate the loss function of the complex neural network in the pre-distortion system, and then update the weight parameters and bias parameters of each layer of the complex neural network based on the loss function.

[0061] The complex vector input into the pre-distortion system from each part of the test set, and the complex scalar output by the pre-distortion system corresponding to the complex vector, are also one-to-one corresponding; when verifying the generalization performance of the pre-distortion system, it is based on the output set composed of the entire test set and all the complex scalars to statistically obtain the Generalization Error Vector Magnitude (GEVM) and the Generalization Adjacent Channel Power Ratio (GACLR) performance, and to determine whether to end the training of the pre-distortion system.

[0062] In one embodiment, the elements in the normalized complex vector are determined by the following calculation expression:

[0063]

[0064]

[0065] wherein x1(n) represents the nth element in the normalized complex vector, x 1,raw (n) represents the nth element in the complex signal for training, x 1,raw (d) represents the dth element in the complex signal for training.

[0066] In one embodiment, the system parameters of the predistortion system are initialized, including:

[0067] According to the layer type corresponding to each layer in the complex neural network, the initialization of the corresponding layer is completed.

[0068] The initialization of the system parameters of the predistortion system further includes setting the nonlinear order parameter, the power amplifier memory effect parameter, and the initial output of the complex neural network included in the predistortion system. It should be noted that the power amplifier memory effect parameter set is not limited to the power amplifier. The nonlinear order parameter and the power amplifier memory effect parameter are only used for the training of the predistortion system.

[0069] The specific values of the nonlinear order parameter, the power amplifier memory effect parameter, and the initial output of the complex neural network included in the predistortion system are not limited here. The execution order of the operations of setting the nonlinear order parameter, the power amplifier memory effect parameter, and the initial output of the complex neural network included in the predistortion system and the operations of initializing each layer of the complex neural network is not limited.

[0070] The initialization of each layer of the complex neural network can be completed based on the corresponding initialization setting parameters, which are not limited here, such as distribution type and distribution parameter.

[0071] In one embodiment, the initialization of each layer of the complex neural network based on the corresponding layer type includes:

[0072] According to the distribution type and the distribution parameter of each layer of the complex neural network, the weight parameter and the bias parameter of the corresponding layer are initialized.

[0073] For example, if the current (to be initialized) layer is a fully connected layer, the weight parameter and the bias parameter of the layer are initialized according to the distribution type (such as Gaussian distribution, uniform distribution, etc.) and the distribution parameter (mean, variance, standard deviation, etc.) of the random initialization set for the layer.

[0074] In one embodiment, the training of the predistortion system based on the training set and the complex scalar includes:

[0075] The complex signals in the training set are input into the predistortion system according to the signal index. The length of each input is determined based on the power amplifier memory effect parameter. The complex vector input into the predistortion system each time is obtained by combining the historical value and the current value in the training set.

[0076] The output of the radio frequency power amplifier output feedback loop in the pre-distortion system is input into the complex neural network after passing through a second real-time power normalization unit included in the pre-distortion system, or is directly input into the complex neural network;

[0077] The partial derivatives or sensitivities of the loss function of the complex neural network in the pre-distortion system with respect to the weight parameters and bias parameters of each layer in the complex neural network are determined according to the partial derivatives or sensitivities of the loss function of the complex neural network with respect to each element of the correction vector output by the complex neural network;

[0078] The weight parameters and bias parameters of the corresponding layer are updated according to the determined partial derivatives or sensitivities of each layer;

[0079] wherein, the expression of the complex vector input into the pre-distortion system each time is as follows:

[0080]

[0081] wherein, is the complex vector input into the pre-distortion system each time, M1 is a power amplifier memory effect parameter, 0≤M1≤TrainSize, TrainSize is the length of the training set, is the nth element in the training set, is the current value, m is an integer greater than 1 and less than M1, and are the (n-1)th, (n-m)th and (n-M1)th elements in the training set, and is the historical value.

[0082] The output of the radio frequency power amplifier output feedback loop in the pre-distortion system is a complex scalar. The complex signals in the training set are input into the pre-distortion system according to the signal index, that is, into the pre-distortion multiplier and the complex neural network in the pre-distortion system. The length input into the pre-distortion system each time can be determined by the power amplifier memory effect parameter, such as inputting one current value in the training set each time, and then determining the number of historical values input into the pre-distortion system by the power amplifier memory effect parameter. The specific determination means is not limited here, such as the number of historical values is not greater than the power amplifier memory effect parameter.

[0083] After the complex signals in the training set are input into the pre-distortion system and the output of the radio frequency power amplifier output feedback loop is fed back to the pre-distortion system, the loss function of the complex neural network in the pre-distortion system can be determined, and then the partial derivatives or sensitivities of the loss function with respect to the weight parameters and bias parameters of each layer in the complex neural network are determined based on the partial derivatives or sensitivities of the loss function with respect to each element of the correction vector output by the complex neural network, so as to update the weight parameters and bias parameters of each layer of the complex neural network, so as to realize the training of the pre-distortion system.

[0084] In one embodiment, the relationship between the complex vector input to the pre-distortion system and the correction vector output by the complex neural network is shown as follows:

[0085]

[0086] wherein, is the correction vector, and are the n, n-1, n-m and n-M1 elements in the correction vector, respectively, and ComplexNN represents the composite function of the internal layer-by-layer operation function of the complex neural network.

[0087] The relationship between the complex vector input to the pre-distortion multiplier in the pre-distortion system, the correction vector, and the output of the pre-distortion multiplier is shown as follows:

[0088]

[0089] wherein, is the complex vector output by the pre-distortion multiplier, and are the n, n-1, n-M+1 and n-M1 elements in the complex vector output by the pre-distortion multiplier, is the complex vector input to the pre-distortion multiplier, denotes the dot product.

[0090] The relationship between the output of the pre-distortion multiplier and the output of the radio frequency power amplifier output feedback loop is shown as follows:

[0091]

[0092] wherein, is the output of the radio frequency power amplifier output feedback loop, and PA denotes the processing function of the radio frequency power amplifier output feedback loop on the input signal.

[0093] In the case where the pre-distortion system includes a second real-time power normalization unit, the relationship between the input and output of the second real-time power normalization unit is shown as follows:

[0094]

[0095]

[0096] wherein, is the n output of the second real-time power normalization unit, For the n-th input of the second real-time power normalization unit, n is a positive integer, d is a positive integer greater than or equal to 1 and less than or equal to n, For the d-th output of the radio frequency power amplifier output feedback loop.

[0097] In one embodiment, the partial derivative or sensitivity of the loss function of the complex neural network with respect to each element of the correction vector of the output of the complex neural network is determined according to the partial derivative or sensitivity of the loss function of the complex neural network with respect to each element of the correction vector of the output of the complex neural network, including:

[0098] The loss function of the complex neural network is determined based on the feedback quantity of the radio frequency power amplifier output feedback loop and the complex vector input to the complex neural network.

[0099] The partial derivative or sensitivity of the loss function with respect to each element of the correction vector of the output of the complex neural network is determined.

[0100] The partial derivative or sensitivity of the loss function with respect to the weight parameters and bias parameters of each layer inside the complex neural network is determined according to the partial derivative or sensitivity of the correction vector.

[0101] The weight parameters and bias parameters of the corresponding layer are updated based on the partial derivative or sensitivity of the weight parameters and bias parameters of each layer.

[0102] In one embodiment, the loss function is determined by the following calculation expression:

[0103]

[0104] The loss function is.

[0105] In the case where the pre-distortion system includes a second real-time power normalization unit, the feedback quantity of the radio frequency power amplifier output feedback loop is the output of the radio frequency power amplifier output feedback loop input to the output of the second real-time power normalization unit; in the case where the pre-distortion system does not include a second real-time power normalization unit, the feedback quantity of the radio frequency power amplifier output feedback loop is the output of the radio frequency power amplifier output feedback loop.

[0106] The partial derivative or sensitivity of the loss function with respect to each element of the correction vector of the output of the complex neural network can be determined based on the partial derivative or sensitivity of the loss function with respect to the output of the pre-distortion multiplier, the radio frequency power amplifier output feedback loop and / or the second real-time power normalization unit in the pre-distortion system.

[0107] After the partial derivative or sensitivity of the correction vector is determined, the weight parameters and bias parameters of each layer of the complex neural network can be updated based on the partial derivative or sensitivity of the correction vector.

[0108] In one embodiment, the partial derivative or sensitivity of each element of the correction vector is calculated according to the following expression:

[0109]

[0110] wherein, is the partial derivative or sensitivity of each element of the initial correction vector, is the partial derivative or sensitivity of each element of the complex vector of the pre-distortion multiplier output, conj(·) represents taking the conjugate of a complex number;

[0111] is determined according to the following expression:

[0112]

[0113] wherein, “·” represents scalar multiplication, L1 is a non-linear order parameter, is an element in the complex vector of the pre-distortion multiplier output, real(·) represents taking the real part, imag(·) represents taking the imaginary part, δ l,m represents an intermediate partial derivative or sensitivity.

[0114] In one embodiment, in the case where the pre-distortion system does not include a second real-time power normalization unit, δ l,m is determined according to the following expression:

[0115]

[0116] In the case where the pre-distortion system includes a second real-time power normalization unit, δ l,m is determined according to the following expression:

[0117]

[0118] wherein, c l,m is a complex coefficient obtained according to a memory polynomial model, based on the complex signal in the training set and the output obtained by inputting the output of the radio frequency power amplifier output feedback loop into the radio frequency power amplifier output feedback loop, using a least squares algorithm; is the partial derivative or sensitivity of the loss function with respect to the complex scalar of the radio frequency power amplifier output feedback loop output;

[0119] is determined according to the following expression:

[0120]

[0121] wherein, is the partial derivative or sensitivity of the loss function with respect to the feedback quantity of the radio frequency power amplifier output feedback loop;

[0122] The determination is calculated by the following calculation expression:

[0123]

[0124] In one embodiment, the partial derivatives or sensitivities of the weight parameters and bias parameters of the fully connected layer inside the complex neural network are respectively:

[0125]

[0126]

[0127] wherein, represents the complex weight parameter connected from the u-th neuron of the j-th layer to the v-th neuron of the k-th layer of the complex neural network, represents the bias parameter of the v-th neuron of the k-th layer of the complex neural network, and represent the output complex signals of the u-th neuron of the j-th layer and the v-th neuron of the k-th layer of the complex neural network respectively; f'(·) represents the derivative of the neuron activation function to the input signal, is the partial derivative or sensitivity of the weight parameter and bias parameter of the fully connected layer, is the partial derivative or sensitivity of the bias parameter of the fully connected layer, is the partial derivative or sensitivity of the loss function to .

[0128] When the current layer is the last layer of the complex neural network, the partial derivative or sensitivity of the loss function to is equal to the partial derivative or sensitivity of the correction vector.

[0129] In one embodiment, the partial derivatives or sensitivities of the weight parameters and bias parameters of the q-th convolution kernel of the convolution layer inside the complex neural network are respectively:

[0130]

[0131]

[0132] wherein, is the partial derivative or sensitivity of the weight parameter and bias parameter of the q-th convolution kernel, is the partial derivative or sensitivity of the bias parameter of the q-th convolution kernel, represents the p-th complex vector output by the j-th layer of the previous layer of the convolution layer, represents the q-th complex vector of the k-th layer output, q = 1, 2, …, Q, p = 1, 2, …, P, where Q and P are the number of output feature vectors of the k-th layer and the j-th layer, respectively, represents the partial derivative or sensitivity of the loss function with respect to , Fliplr(·) represents the position inversion of the input vector. Each feature map can be a complex vector form.

[0133] In one embodiment, the weight parameters and bias parameters of the fully connected layer inside the complex neural network are updated by the following expressions:

[0134]

[0135]

[0136] wherein, and represent the weight parameters of the fully connected layer at the current moment and the previous moment, respectively; and represent the bias parameters of the fully connected layer at the current moment and the previous moment, respectively; represents the update step size of the weight parameters at the current moment; represents the update step size of the bias parameters at the current moment.

[0137] In one embodiment, the trained pre-distortion system is tested based on the test set to obtain the error vector magnitude and the adjacent channel leakage ratio corresponding to the pre-distortion system, comprising:

[0138] The complex vectors in the test set are input into the pre-distortion system according to the signal index, the length of each input is determined based on the power amplifier memory effect parameter, and the complex vectors input into the pre-distortion system each time are obtained by combining the historical values and the current values in the training set;

[0139] The output of the radio frequency power amplifier output feedback loop in the pre-distortion system is input into the pre-distortion system to determine the generalization error vector magnitude and the generalization adjacent channel leakage ratio corresponding to the pre-distortion system.

[0140] The method used for training a predistortion system based on a test set is similar to that used for training a predistortion system based on a training set. The difference lies in that, based on the test set, there's no need to update the weight and bias parameters of each layer of the complex neural network. Instead, the generalization error vector magnitude and the generalized adjacent channel leakage ratio are directly calculated and determined to confirm whether they meet the set requirements. The generalization error vector magnitude and the generalized adjacent channel leakage ratio can be determined based on the feedback quantity of the RF power amplifier output feedback loop and the corresponding complex vectors in the test set.

[0141] The input to the predistortion information system includes passing through the second real-time power normalization unit included in the predistortion system before being input into the complex neural network; or, the input can be directly input into the complex neural network.

[0142] In one embodiment, the generalization error vector magnitude and the generalization adjacent channel leakage ratio are determined by the following calculation expressions:

[0143]

[0144]

[0145]

[0146]

[0147] Where GEVM1 represents the magnitude of the error vector. express The adjacent channel leakage ratio, This indicates the feedback quantity of the output feedback loop of the RF power amplifier. This represents the feedback quantity of the nth element in the output feedback loop of the RF power amplifier. This represents the nth element of the complex vector corresponding to the test set; TestSize represents the length of the training set; the sum of TestSize and TrainSize is N1, where N1 represents the length of the complex signal used for training; HBW represents half of the effective signal bandwidth; GBW represents half of the guard bandwidth; and NFFT represents the number of points in the discrete Fourier transform. NFFT It is the coefficient vector of a window function with an NFFT window length. Represents the output complex signal randomly selected from the test set. Extract the NFFT length of the signal; It is a uniformly distributed random positive integer in the range [1, TestSize]; K is the number of random cuts.

[0148] In the case that the pre-distortion system comprises the second real-time power normalization unit, the feedback quantity of the radio frequency power amplifier output feedback loop can be considered as the output of the second real-time power normalization unit; in the case that the pre-distortion system does not comprise the second real-time power normalization unit, the feedback quantity of the radio frequency power amplifier output feedback loop can be the output of the radio frequency power amplifier output feedback loop.

[0149] The pre-distortion method provided by the application solves the problems of the innate insufficient nonlinear expression ability and the missing generalization ability of the traditional power amplifier model and the pre-distortion model, and better error vector magnitude (EVM) performance and adjacent channel power ratio (ACLR) performance are obtained.

[0150] The application adopts a neural network with more abundant structure to model the system, solves the innate defect of the insufficient nonlinear expression ability of the traditional power amplifier model and the DPD model, completely discards the separate processing mode of the existing pre-distortion method and technology using the neural network, and integrates and processes the characteristic learning and the pre-distortion correction of the power amplifier as a whole throughout, and first proposes an AI-DPD integrated scheme, i.e., an integrated scheme of the artificial intelligence replacing the DPD.

[0151] In one embodiment, a known training complex signal is sent before the sending of a service complex signal, and the training complex signal is used for the training of the AI-DPD integrated scheme system (i.e., the pre-distortion system) together with the service complex signal through the pre-distortion multiplier, the complex neural network and the radio frequency power amplifier output feedback loop; the training is stopped when the required GEVM and GACLR are reached; and meanwhile, the service complex signal is sent through the pre-distortion system, and a pre-distortion corrected complex scalar is obtained.

[0152] Figure 3a A structure schematic diagram of a pre-distortion system provided by the embodiment of the application is shown in Figure 3a The pre-distortion system, i.e., the AI-DPD integrated scheme system mainly comprises a first real-time power normalization (i.e., PNorm) unit, a second real-time power normalization unit, a pre-distortion multiplier (i.e., DPD Multiplier) unit, a complex neural network (i.e., Complex Neural Network) unit and a radio frequency power amplifier output feedback loop. The pre-distortion multiplier unit, i.e., the pre-distortion multiplier. The complex neural network unit can comprise a complex neural network, a selector (i.e., MUX) and an adder. The radio frequency power amplifier output feedback loop comprises a digital-to-analog converter unit (i.e., D / A), a radio frequency modulation unit, a power amplifier unit, an analog-to-digital converter unit (i.e., A / D) and a radio frequency demodulation unit.

[0153] wherein, is the original input complex vector composed of historical values (the length of historical values is set by the power amplifier memory effect parameter M1) and current values; is the complex vector after real-time power normalization of , is also the input of the complex neural network; is the complex correction vector after pre-distortion of the input vector , is also the output of the complex neural network; is the complex vector after pre-distortion correction of ; is the final original output of the AI-DPD integration scheme, y 1，raw is a complex scalar; y1 is the complex scalar after real-time power normalization of y 1，raw ; 1，raw

[0154] The complex neural network is a neural network: the complex number includes a real part (Real Part, hereinafter abbreviated as I path) and an imaginary part (Imaginary Part, hereinafter abbreviated as Q path), and usually refers to the case where the imaginary part is not 0, in order to be distinguished from real numbers. The input and output of the neural network can be directly a complex variable, vector, matrix or tensor, or can be in the form of combination of I path and Q path; the input and output of the neuron activation function and all other processing functions of each layer (i.e. Layer) of the neural network can be directly a complex variable, vector, matrix or tensor, or can be in the form of combination of I path and Q path; the neural network can be trained using a complex error back propagation algorithm (Complex BP) or a real error back propagation algorithm (BP).

[0155] The complex error back propagation algorithm includes but is not limited to the following steps:

[0156] Step 1: initialization of AI-DPD integration scheme system parameters.

[0157] Set the non-linear order parameter L1 and the power amplifier memory effect parameter M1, for example: L1 = 5, M1 = 4; set the initial output of the complex neural network, i.e. each element of the initial pre-distortion correction vector is 1; and according to the index of forward running each layer, layer by layer, or all layers simultaneously in parallel, complete the corresponding initialization steps according to the layer type. The initialization of different types of layers mainly includes but is not limited to the following steps:

[0158] ​Step 1-1: If the current (to be initialized) layer is a fully connected layer, initialize the weight parameters and bias parameters of the layer according to the distribution type (e.g. Gaussian distribution, uniform distribution, etc.) and distribution parameters (mean, variance, standard deviation, etc.) of the random initialization set by the layer;

[0159] Step 1-2: If the current (to be initialized) layer is a convolutional layer, initialize the weight parameters and bias parameters of each convolutional kernel of the layer according to the distribution type and distribution parameters of the random initialization set by the layer;

[0160] Step 1-3: If the current (to be initialized) layer is of other types, such as Fractionally Strided Convolutions layer, etc., initialize the weight and bias parameters of the layer according to the initialization setting parameters (such as distribution type and distribution parameters of random initialization) of the type of layer.

[0161] Step 2: AI-DPD integrated scheme system training based on training set. First, the known complex signal sequence of length N is subjected to real-time power normalization (PNorm). Since there are various forms of real-time power normalization, only one example is given below:

[0162]

[0163]

[0164] Then, the real-time power normalized sequence is divided into training set and test set by a certain proportion (e.g. 0.5:0.5), with the lengths of the two sets being TrainSize and TestSize respectively; wherein the training set is used for overall learning or training of the AI-DPD integrated scheme system; the test set is used to test the generalization and adaptation performance of the scheme system to new data in the latter half of the training; this performance includes the generalization EVM performance and generalization ACPR performance, which are defined as follows:

[0165]

[0166]

[0167] wherein HBW refers to half of the effective signal bandwidth; GBW refers to half of the guard bandwidth; The calculation is as follows:

[0168]

[0169]

[0170] where NFFT is the number of points of Discrete Fourier Transform (FFT) ; Win NFFT is the coefficient vector of window function with length NFFT, for example, the window function can be Blackman window; represents the output complex signal of NFFT length randomly intercepted from the test set; represents the output complex signal of NFFT length randomly intercepted from the test set; is a uniformly distributed random positive integer in the range [1, TestSize] ; K is the number of random intercepts; represents point multiplication.

[0171] The training of the AI-DPD integrated scheme system includes but is not limited to the following sub-steps:

[0172] Step 2-1: the training set signal is organized into the form of complex vector of history value and current value combination as follows: each time according to signal index n (n = 1, 2, …, TrainSize) :

[0173]

[0174] where M1 is the power amplifier memory effect setting, 0≤M1≤TrainSize.

[0175] Step 2-2: the complex vector is obtained through the complex neural network (Complex Neural Network, hereinafter referred to as ComplexNN) unit, and the pre-distorted complex correction vector The input and output relationship of this unit is shown in the following formula:

[0176]

[0177]

[0178] Step 2-3: the complex vector is obtained through the pre-distortion multiplier (DPD Multiplier) unit, and the pre-distorted complex correction vector The input and output relationship of this unit is shown in the following formula:

[0179]

[0180]

[0181] Above, represents point multiplication.

[0182] Step 2-4: the pre-distortion corrected complex vector Step 2-5: the AI-DPD integrated scheme final original output The relationship between the output of the pre-distortion multiplier and the output of the RF PA output feedback loop is shown in the following formula:

[0183]

[0184] It should be particularly noted that the power amplifier (PA) unit in the RF PA output feedback loop can be any actual power amplifier product; in other words, the present embodiment has no limitation on the non-linear characteristics of the power amplifier model.

[0185] Step 2-4: the pre-distortion corrected complex vector After RF demodulation and analog-to-digital conversion (A / D), the real-time power normalization (PNorm) unit is inputted, and the PA output feedback quantity is obtained The input and output relationship of this unit is shown in the following two formulas:

[0186]

[0187]

[0188] Step 2-6: according to the PA output feedback quantity and the complex vector The loss function of the complex neural network is calculated. Because there are various forms of additional regularization terms, the following loss function expression is only a simple example:

[0189]

[0190] Step 2-7: the loss function is calculated according to the following formula (2-16) The partial derivative or sensitivity of the PA output feedback quantity

[0191]

[0192] Step 2-8: according to the above sensitivity The loss function is calculated according to the following formula (2-17) The partial derivative or sensitivity of the AI-DPD integrated scheme final original output

[0193] ​​

[0194] Above, is calculated from the above formula (2-13).

[0195] Step 2-9: According to the above sensitivity The intermediate partial derivative or sensitivity δ is calculated as follows according to the following formula (2-18) l,m :

[0196]

[0197] Above, c l,m is the complex coefficient obtained by using the least square algorithm based on the complex signal in the training set and the output obtained by inputting the output feedback loop of the radio frequency power amplifier; conj(·) represents taking the conjugate of the complex number.

[0198] Step 2-10: According to the above sensitivity l,m The loss function is calculated according to the following formula (2-19) For the complex vector after pre-distortion correction The partial derivative or sensitivity of each element

[0199]

[0200]

[0201] In order to avoid the extreme value situation of input 0+0j, in particular: if there is a zero element (i.e. 0+0j) in the pre-distortion corrected complex vector , set B l,m defined in the above formula (2-20) to 0.

[0202] Step 2-11: According to the above sensitivity The loss function is calculated according to the following formula (2-21) For the pre-distorted complex correction vector The partial derivative or sensitivity of each element

[0203]

[0204] Step 2-12: According to the above sensitivity According to the reverse order of the forward operation of the complex neural network unit, the loss function is calculated in turn For the partial derivative or sensitivity of the weight and bias parameters of each layer in it, including but not limited to the following sub-steps:

[0205] Step 2-12-1: If the current layer to be calculated is a full-connected layer, the loss function is calculated according to the following formula (2-22) and formula (2-23) respectively The partial derivative or sensitivity of the complex weight parameter and the complex bias parameter of this layer:

[0206]

[0207]

[0208] Above, represents the complex weight connected from the u-th neuron of the j-th layer of the network to the v-th neuron of the k-th layer; and respectively represent the output complex signal of the u-th neuron of the j-th layer and the v-th neuron of the k-th layer of the network; f'(·) represents the derivative of the neuron activation function with respect to the input signal.

[0209] Step 2-12-2: If the current layer (set as the k-th layer) to be calculated is a convolutional layer, the loss function is calculated according to the following formula (2-24) and formula (2-25) respectively The partial derivative or sensitivity of the complex weight and the complex bias of the q-th (q=1, 2, …, Q) convolution kernel of this layer:

[0210]

[0211]

[0212] Above, represents the p-th (p=1, 2, …, P) complex vector output by the previous layer (j-th layer) of the convolutional layer; represents the q-th complex vector output by the current previous convolutional layer (k-th layer); represents the loss function The partial derivative or sensitivity of the above ; Conv(·) represents convolution operation; Fliplr(·) represents position inversion of the input vector.

[0213] Step 2-13: According to the loss function calculated by each layer The sensitivity of the weight parameter and the bias parameter is updated, because the parameter updating method of different training algorithms is various, the following is only a simple example:

[0214]

[0215]

[0216] Step 3: AI-DPD integrated scheme system based on performance statistics of test set.

[0217] Step 3-1: the test set signal is organized in the form of a complex vector of historical and current value combinations as follows:

[0218]

[0219] Step 3-2: the complex vector is input into the complex neural network (Complex Neural Network) unit to obtain the pre-distorted complex correction vector

[0220]

[0221]

[0222] Step 3-3: the complex vector is input into the pre-distortion multiplier (DPD Multiplier) unit to obtain the pre-distorted corrected complex vector

[0223]

[0224]

[0225] Step 3-4: the pre-distorted corrected complex vector is input into the radio frequency power amplifier output feedback loop to obtain the output

[0226]

[0227] Step 3-5: the is input into the real-time power normalization (PNorm) unit after radio frequency demodulation and analog-to-digital conversion (A / D) to obtain the power amplifier output feedback quantity

[0228]

[0229]

[0230] Step 3-6: according to the above​ and the test set The generalized EVM performance (GEVM) and generalized ACLR performance (GACLR) of the system based on the test set are statistically calculated according to Equations (2-3) and (2-4) above.

[0231] Steps 3-7: If GEVM and GACLR meet the set performance requirements, stop training of the AI-DPD integrated solution system; otherwise, return to all steps in step 2 and start a new round of training.

[0232] In one embodiment, a known training complex signal is transmitted before the service complex signal is transmitted. The training complex signal is then transmitted through a predistortion multiplier, a complex neural network, and an RF power amplifier output feedback loop to obtain a predistortion-corrected complex signal, which is fed back to the transmitter. This signal, along with the training complex signal, is used for training the AI-DPD integrated solution system. Training stops once the required generalized EVM and generalized ACLR performances are achieved. Simultaneously, the service complex signal is transmitted through the predistortion system to output a predistortion-corrected complex scalar.

[0233] Figure 3b A schematic diagram of another predistortion system provided in this application embodiment is shown below. Figure 3b This predistortion system, also known as the AI-DPD integrated solution, mainly includes: a predistortion multiplier unit, a complex neural network unit, and an RF power amplifier output feedback loop. It is a raw input complex vector composed of historical values ​​(the length of which is set by the amplifier memory effect parameter M2) and the current value; For the input vector Perform pre-distortion complex correction vector It is also the output of a complex neural network; y1 is the complex vector after pre-distortion correction; y2 is the final output of the AI-DPD integrated scheme, and y2 is a complex scalar. It should be noted that the subscripts "1" and "2" in the formulas of this application have no practical meaning and are used to refer to... Figure 3a and Figure 3b The corresponding two predistortion systems. For example, M1 and M2 only represent the power amplifier memory effect parameters, i.e., the power amplifier memory effect parameters; the subscript "1" refers to... Figure 3a The predistortion system shown, with the subscript "2" referring to... Figure 3b The predistortion system shown.

[0234] The complex neural network is a neural network: the complex number includes two parts of real part and imaginary part, and usually refers to the case that the imaginary part is not 0, so as to be distinguished from the real number; the input and output of the neural network can be directly a complex variable, vector, matrix or tensor, or can be in the form of combination of I and Q of the complex number; the input and output of the neuron activation function and all other processing functions of each layer of the neural network can be directly a complex variable, vector, matrix or tensor, or can be in the form of combination of I and Q of the complex number; the neural network can be trained by using a complex error back propagation algorithm or an real error back propagation algorithm.

[0235] The complex error back propagation algorithm includes but is not limited to the following steps:

[0236] Step 1: initialization of system parameters of the AI-DPD integrated scheme.

[0237] Set the nonlinear order parameter L2 and the power amplifier memory effect parameter M2, for example: L2 = 5, M2 = 4; set the initial output of the complex neural network, that is, each element of the initial pre-distortion correction vector is 1; and according to the index of forward running each layer, layer by layer, or all layers simultaneously and in parallel, complete the corresponding initialization steps according to the layer type. The initialization of the different types of layers mainly includes but is not limited to the following steps:

[0238] Step 1-1: if the current (to be initialized) layer is a fully connected layer, initialize the weight parameters and bias parameters of the layer according to the distribution type (for example: Gaussian distribution, uniform distribution, etc.) and distribution parameters (mean, variance, standard deviation, etc.) of the random initialization set for the layer;

[0239] Step 1-2: if the current (to be initialized) layer is a convolutional layer, initialize the weight parameters and bias parameters of each convolutional kernel of the layer according to the distribution type and distribution parameters of the random initialization set for the layer;

[0240] Step 1-3: if the current (to be initialized) layer is a layer of other types, complete the initialization of the weight and bias parameters of the layer according to the initialization setting parameters of the layer of the type.

[0241] Step 2: training of the AI-DPD integrated scheme system based on the training set. Divide the known complex signal sequence of length N into a training set and a test set The lengths of the two are recorded as TrainSize and TestSize respectively; wherein the training set is used for the overall learning or training of the AI-DPD integrated scheme system; the test set is used to test the generalization adaptation performance of the scheme system to new data in the latter half of the training; the performance includes the generalization EVM performance (GEVM) and the generalization ACPR performance (GACLR), which are defined as follows respectively:

[0242]

[0243]

[0244] Wherein, HBW refers to half of the effective signal bandwidth; GBW refers to half of the guard bandwidth; The calculation is as follows:

[0245]

[0246]

[0247] Above, NFFT is the number of points of discrete Fourier transform (FFT); Win NFFT is the coefficient vector of the window function with window length NFFT, for example: the window function can use Blackman window; represents randomly intercepting NFFT long signals from the output complex signal of the test set . is a uniformly distributed random positive integer in the range [1, TestSize]; K is the number of random intercepts. represents point multiplication.

[0248] The training of the AI-DPD integrated scheme system includes but is not limited to the following sub-steps:

[0249] Step2-1: Organize the training set signal into the form of complex vector composed of the following historical values and current values by signal index n (n = 1, 2, …, TrainSize) each time and input into the system continuously:

[0250]

[0251] Wherein, M2 is the power amplifier memory effect setting, 0≤M2≤TrainSize.

[0252] Step2-2: The complex vector is input into the complex neural network unit (ComplexNN) to obtain the pre-distorted complex correction vector The input and output relationship of the unit is shown in the following formula:

[0253]

[0254]

[0255] Step2-3: The complex vector The pre-distortion correction complex vector The input and output relationship of this unit is shown in the following formula:

[0256]

[0257]

[0258] The above, Indicates the dot product.

[0259] Step2-4: The pre-distortion correction complex vector Input the radio frequency power amplifier output feedback loop to obtain the output As shown in the following formula:

[0260]

[0261] It should be particularly noted that the power amplifier (PA) unit in the radio frequency power amplifier output feedback loop can be any actual power amplifier product; in other words, the present embodiment has no any limitation condition on the non-linear characteristics of the power amplifier model.

[0262] Step2-5: According to the And the complex vector Calculate the loss function of the complex neural network. Because there are many variations of additional regularization terms, the following loss function expression is only a simple example:

[0263]

[0264] Step2-6: Calculate the loss function according to the following formula (2-47) The partial derivative or sensitivity of the power amplifier output feedback quantity

[0265]

[0266] Step2-7: According to the above sensitivity Calculate the following intermediate partial derivative or sensitivity δ l,m According to the following formula (2-48):

[0267] ​

[0268] Above, c l,m is the complex coefficient obtained by least square algorithm based on the complex signal in the training set and the output obtained by inputting the output of the radio frequency power amplifier output feedback loop; conj(·) represents taking the conjugate of the complex number.

[0269] Step2-8: According to the above sensitivity δ l,m The loss function is calculated according to the following formula (2-49) For the complex vector after pre-distortion correction The partial derivative or sensitivity of each element

[0270]

[0271]

[0272] In order to avoid the extreme value situation of input 0+0j, especially: if there is a zero element (i.e. 0+0j) in the complex vector after pre-distortion correction , set B l,m defined in the above formula (2-50) to 0.

[0273] Step2-9: According to the above sensitivity δ The loss function is calculated according to the following formula (2-51) For the complex correction vector of the pre-distortion The partial derivative or sensitivity of each element

[0274]

[0275] Step2-10: According to the above sensitivity δ The loss function is calculated according to the reverse order of the forward operation of the complex neural network unit, in turn The partial derivative or sensitivity of the weight and bias parameters of each layer inside, including but not limited to the following sub-steps:

[0276] Step2-10-1: If the current layer to be calculated is a fully connected layer, the loss function is calculated according to the following formula (2-52) and formula (2-53) The partial derivative or sensitivity of the complex weight parameters and the complex bias parameters of this layer:

[0277]

[0278]

[0279] Above, represents the complex weight value of the connection from the u-th neuron of the j-th layer to the v-th neuron of the k-th layer of the network; and respectively represent the output complex signal of the u-th neuron of the j-th layer and the v-th neuron of the k-th layer of the network; f'(·) represents the derivative of the neuron activation function with respect to the input signal.

[0280] Step2-10-2: If the current layer to be calculated (set as the k-th layer) is a convolutional layer, the loss function is calculated according to the following formula (2-54) and formula (2-55) respectively The partial derivative or sensitivity of the complex weight value of the q-th (q = 1, 2, …, Q) convolution kernel (Kernel) of this layer and the complex bias :

[0281]

[0282]

[0283] Above, represents the p-th (p = 1, 2, …, P) complex vector output by the previous layer (j-th layer) of the convolutional layer; represents the q-th complex vector output by the current pre-convolutional layer (k-th layer); represents the loss function The partial derivative or sensitivity of the above ; Conv(·) represents convolution operation; Fliplr(·) represents position inversion of the input vector.

[0284] Step2-11: According to the loss function calculated by each layer The sensitivity of the weight parameter and the bias parameter is updated, because the parameter updating method of different training algorithms is various, and the following is only a simple example:

[0285]

[0286]

[0287] Step3: AI-DPD integrated scheme system based on performance statistics of test set.

[0288] Step3-1: The test set signal is organized as a complex vector in the form of the following historical value and current value combination and input into the system continuously according to the signal index n (n = 1, 2, …, TestSize):

[0289]

[0290] Step3-2: the complex vector is input into the complex neural network (Complex NN) unit Through the complex neural network (Complex NN) unit, the pre-distorted complex correction vector is obtained

[0291]

[0292]

[0293] Step3-3: the complex vector is input into the pre-distortion multiplier (DPD Multiplier) unit Through the pre-distortion multiplier (DPD Multiplier) unit, the pre-distorted corrected complex vector is obtained

[0294]

[0295]

[0296] Step3-4: the pre-distorted corrected complex vector is input into the radio frequency power amplifier output feedback loop The output is obtained by inputting the radio frequency power amplifier output feedback loop

[0297]

[0298] Step3-5: according to the above and the test set According to the above formula (2-36) and formula (2-37), the generalization EVM performance (GEVM) and the generalization ACLR performance (GACLR) of the system based on the test set are calculated.

[0299] Step3-6: if the GEVM and GACLR meet the set index requirements, stop the training of the AI-DPD integrated scheme system; otherwise, return to all steps of Step2 and start a new round of training.

[0300] Table 1 is a generalization performance effect table provided by an embodiment of the present application. According to the pre-distortion system provided by the present application, the generalization EVM performance is guaranteed. The complex MLP network includes 3 hidden layers, each layer has 5 neurons; the nonlinear order P=5; the memory effect length M=6; the length of the complex vector input into the pre-distortion system each time =P×(M+1)=35. The training set includes 39320 complex signals; the verification set includes 13107 complex signals; and the test set includes 13105 complex signals.

[0301] Table 1 is a generalization performance effect table provided by an embodiment of the present application. According to the pre-distortion system provided by the present application, the generalization EVM performance is guaranteed. The complex MLP network includes 3 hidden layers, each layer has 5 neurons; the nonlinear order P=5; the memory effect length M=6; the length of the complex vector input into the pre-distortion system each time =P×(M+1)=35. The training set includes 39320 complex signals; the verification set includes 13107 complex signals; and the test set includes 13105 complex signals.

[0302]

[0303] Figure 4 The performance effect of the generalized adjacent channel leakage ratio obtained by the embodiment of the present application is shown in the following table: Figure 4 The integrated method, that is, the deviation between the generalized adjacent channel leakage ratio of the actual output of the pre-distortion system provided by the present application and the generalized adjacent channel leakage ratio of the expected output is small. Figure 5 The improvement effect of the generalized adjacent channel leakage ratio obtained by the embodiment of the present application is shown in the following table: Figure 5 Compared with the signal only passing through the power amplifier, the generalized adjacent channel leakage ratio of the pre-distortion system provided by the present application is obviously improved.

[0304] It should be easily recognized by those skilled in the art that different steps of the above method can be realized by programming a computer. Herein, some embodiments also include a machine-readable or computer-readable program storage device (for example, a digital data storage medium) and a coded machine-executable or computer-executable program instruction, wherein the instruction performs some or all steps of the above method. For example, the program storage device can be a digital memory, a magnetic storage medium (for example, a magnetic disk and a magnetic tape), a hardware or an optically readable digital data storage medium. The embodiments also include a programmed computer that performs the steps of the above method.

[0305] The description and the drawings only show the principles of the present application. Therefore, it should be realized that those skilled in the art can suggest different structures, although these different structures are not explicitly described or shown herein, but embody the principles of the present application and are included in the spirit and scope thereof.

[0306] In an exemplary embodiment, the present application provides a pre-distortion system, a structural schematic diagram of which is shown in the following figure: Figure 3b The pre-distortion system can perform the pre-distortion method provided by the embodiment of the present application, and the pre-distortion system includes a pre-distortion multiplier, a complex neural network and a radio frequency power amplifier output feedback loop.

[0307] The first input end of the pre-distortion multiplier is the input end of the pre-distortion system, and is connected with the first input end and the second input end of the complex neural network. The output end of the pre-distortion multiplier is connected with the input end of the radio frequency power amplifier output feedback loop. The output end of the radio frequency power amplifier output feedback loop is the output end of the pre-distortion system, and is connected with the second input end of the complex neural network. The output end of the complex neural network is connected with the second input end of the pre-distortion multiplier.

[0308] In one embodiment, the radio frequency power amplifier output feedback loop comprises: a digital-to-analog converter unit, a radio frequency modulation unit, a power amplifier unit, an analog-to-digital converter unit, and a radio frequency demodulation unit.

[0309] The output end of the pre-distortion multiplier is connected to the input end of the power amplifier unit of the radio frequency power amplifier output feedback loop through the digital-to-analog converter unit (i.e., D / A) and the radio frequency modulation unit of the radio frequency power amplifier output feedback loop, and the output end of the power amplifier unit is connected to the second input end of the complex neural network through the radio frequency demodulation unit and the digital-to-analog converter unit (i.e., A / D).

[0310] The pre-distortion system provided in this embodiment is used to implement the pre-distortion method provided in this application, and the pre-distortion system provided in this embodiment has similar implementation principles and technical effects to the pre-distortion method provided in this application, which will not be described here.

[0311] In one example embodiment, the present application provides a pre-distortion system, a structural schematic diagram of which is shown in Figure 3a The pre-distortion system can execute the pre-distortion method provided in this application, and the pre-distortion system comprises: a pre-distortion multiplier, a complex neural network, a radio frequency power amplifier output feedback loop, a first real-time power normalization unit, and a second real-time power normalization unit.

[0312] The input end of the first real-time power normalization unit is the input end of the pre-distortion system, the output end of the first real-time power normalization unit is connected to the first input end of the pre-distortion multiplier, the first input end of the complex neural network, and the second input end of the complex neural network, respectively, the output end of the pre-distortion multiplier is connected to the input end of the radio frequency power amplifier output feedback loop, the output end of the radio frequency power amplifier output feedback loop is the output end of the pre-distortion system, the output end of the radio frequency power amplifier output feedback loop is connected to the input end of the second real-time power normalization unit, the output end of the second real-time power normalization unit is connected to the second input end of the complex neural network, and the output end of the complex neural network is connected to the second input end of the pre-distortion multiplier.

[0313] In one embodiment, the radio frequency power amplifier output feedback loop comprises: a digital-to-analog converter unit, a radio frequency modulation unit, a power amplifier unit, an analog-to-digital converter unit, and a radio frequency demodulation unit.

[0314] An input end of the digital-to-analog converter unit is an input end of the radio frequency power amplifier output feedback loop, an output end of the digital-to-analog converter unit is connected with an input end of the radio frequency modulation unit, an output end of the radio frequency modulation unit is connected with an input end of the power amplifier unit, an output end of the power amplifier unit is connected with an input end of the radio frequency demodulation unit, an output end of the radio frequency demodulation unit is connected with an input end of the analog-to-digital converter unit, and an output end of the analog-to-digital converter unit is an output end of the radio frequency power amplifier output feedback loop.

[0315] The pre-distortion system provided in the embodiment is used to implement the pre-distortion method provided in the application, and the pre-distortion system provided in the embodiment has similar implementation principles and technical effects to the pre-distortion method provided in the application. Details are not described herein again.

[0316] In one example embodiment, the application also provides a device, Figure 6 The device provided in the application has a structure as shown in the figure, Figure 6 The device provided in the application includes one or more processors 21 and storage devices 22. The processor 21 in the device can be one or more, Figure 6 The storage device 22 is configured to store one or more programs. The one or more programs are executed by the one or more processors 21, so that the one or more processors 21 implement the method described in the embodiments of the application.

[0317] The device also includes a communication device 23, an input device 24 and an output device 25.

[0318] The processor 21, the storage device 22, the communication device 23, the input device 24 and the output device 25 in the device can be connected through a bus or other means, Figure 6 For example, the connection through the bus.

[0319] The input device 24 can be used to receive input digital or character information, and generate key signal input related to user settings and function control of the device. The output device 25 can include a display device such as a display screen.

[0320] The communication device 23 can include a receiver and a transmitter. The communication device 23 is configured to perform information receiving and transmitting communication according to the control of the processor 21.

[0321] The storage device 22, as a computer readable storage medium, can be configured to store software programs, computer executable programs and modules, such as program instructions corresponding to the method of the embodiments of the present application and the pre-distortion system. The storage device 22 can include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application required by a function; the data storage area can store data created according to the use of the device, etc. In addition, the storage device 22 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state memory device. In some examples, the storage device 22 can further include a memory remotely arranged with respect to the processor 21, and these remote memories can be connected to the device through a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0322] The embodiments of the present application also provide a storage medium, which stores a computer program, and the computer program is executed by a processor to implement any method of the present application. The storage medium stores a computer program, and the computer program is executed by a processor to implement the pre-distortion method provided by the embodiments of the present application, and is applied to a pre-distortion system, the pre-distortion system includes a pre-distortion multiplier, a complex neural network and a radio frequency power amplifier output feedback loop, and the method includes: inputting a training complex signal into the pre-distortion system to output a corresponding complex scalar; training the pre-distortion system based on the training complex signal and the complex scalar until the generalization error vector amplitude and the generalization adjacent channel leakage ratio of the pre-distortion system reach a set requirement; inputting a service complex signal into the trained pre-distortion system to obtain a pre-distortion corrected complex scalar.

[0323] The computer storage medium of the embodiments of the present application can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples (non-exhaustive list) of the computer readable storage medium include: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read only memory (ROM), an erasable programmable read only memory (EPROM), a flash memory, an optical fiber, a portable CD-ROM, an optical storage device, a magnetic storage device, or any suitable combination of the above. The computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus or device.

[0324] The computer readable signal medium can include a data signal propagated in baseband or propagated as a carrier wave in a propagated data signal, in which the computer readable program code is embodied. Such propagated data signal can take a variety of forms, including but not limited to electro-magnetic, optical or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a storage medium and that can communicate, propagate or transport programming for use by or in connection with an instruction execution system, apparatus or device.

[0325] The program code embodied on the computer readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire line, optical fiber cable, Radio Frequency (RF), or any suitable combination thereof.

[0326] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0327] The specific embodiments described hereinabove are illustrative of specific embodiments of the present application and are not meant to be limiting of the scope of the application.

[0328] Those skilled in the art will appreciate that the term device (e.g., terminal device) encompasses any suitable type of wireless user equipment, such as a mobile phone, a portable data processing apparatus, a portable web browser, or a vehicle-mounted mobile station.

[0329] In general, the various embodiments of the application can be implemented in hardware or special purpose circuits, software, logic or any combination thereof. For example, some aspects can be implemented in hardware, while other aspects can be implemented in

[0330] Embodiments of the application can be implemented by computer program instructions executed by a data processing apparatus of a mobile device, for example in a processor entity, or by hardware, or by a combination of software and hardware. The computer program instructions can be in the form of assembly instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or in any combination of one or more programming languages, written in any combination of one or more of a plurality of programming languages.

[0331] The block diagrams of any logical flow of the present application in the accompanying drawings can represent program steps or can represent interconnected logic circuits, modules, and functions, or can represent a combination of program steps and logic circuits, modules, and functions. The computer program can be stored on a memory. The memory can have any type suitable for the local technical environment and can be implemented using any suitable data storage technology, such as, but not limited to, a Read-Only Memory (ROM), a Random Access Memory (RAM), an optical storage device and system, a compact disc (CD) or a digital versatile disc (DVD), and the like. The computer readable medium can include a non-transitory storage medium. The data processor can be of any type suitable for the local technical environment, and can include, but is not limited to, a general purpose computer, a special purpose computer, a microprocessor, a Digital Signal Processing (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FGPA) and processors based on a multi-core processor architecture, as non-limiting examples.

[0332] A detailed description of exemplary embodiments of the present application has been provided above with reference to the accompanying drawings. However, various modifications and changes can be made to the above embodiments by those skilled in the art without departing from the scope of the present application, which is defined by the appended claims. Accordingly, the proper scope of the present application is determined by the appended claims.

Claims

1. A predistortion method, characterized by, The application is applied to a predistortion system, the predistortion system comprises a predistortion multiplier, a complex neural network and a radio frequency power amplifier output feedback loop, an output end of the complex neural network is connected with an input end of the predistortion multiplier, an output end of the predistortion multiplier is connected with an input end of the radio frequency power amplifier output feedback loop, and an output end of the radio frequency power amplifier output feedback loop is an output end of the predistortion system; the method comprises: inputting a training complex signal into the predistortion system to output a corresponding complex scalar; training the predistortion system based on the training complex signal and the complex scalar until a generalization error vector amplitude and a generalization adjacent channel leakage ratio of the predistortion system reach a set requirement; inputting a service complex signal into the trained predistortion system to obtain a predistortion corrected complex scalar.

2. The method of claim 1, wherein, The predistortion system is trained by using a complex error back propagation algorithm.

3. The method of claim 1, wherein, The predistortion system is trained, comprising: initializing system parameters of the predistortion system; training the predistortion system based on a training set and the complex scalar; testing the trained predistortion system based on a test set to obtain a generalization error vector amplitude and a generalization adjacent channel leakage ratio of the predistortion system; in a case that the generalization error vector amplitude and the generalization adjacent channel leakage ratio correspond to values greater than or equal to corresponding set thresholds, completing the training of the predistortion system, and in a case that the generalization error vector amplitude and the generalization adjacent channel leakage ratio correspond to values less than the corresponding set thresholds, continuing to train the predistortion system based on the training set; wherein the training set and the test set are obtained based on a normalized complex vector, the normalized complex vector being an output of a first real-time power normalization unit included in the predistortion system and inputted with the training complex signal; or the training set and the test set are obtained based on the training complex signal.

4. The method of claim 3, wherein, An element in the normalized complex vector is determined by a calculation expression as follows: ； ； in, Represents the first element in the normalized complex vector. One element, In the training complex signal, the first... One element, In the training complex signal, the first... Each element.

5. The method of claim 3, wherein, The initialization of the system parameters of the predistortion system comprises: completing the initialization of a corresponding layer according to a layer type of each layer in the complex neural network.

6. The method of claim 5, wherein, The initialization of the corresponding layer according to the layer type of each layer in the complex neural network comprises: initializing weight parameters and bias parameters of the corresponding layer according to a distribution type and distribution parameters of each layer in the complex neural network.

7. The method of claim 3, wherein, The training of the predistortion system based on the training set and the complex scalar comprises: inputting the complex signal in the training set into the predistortion system according to a signal index, the length of each input being determined based on a power amplifier memory effect parameter, and the complex vector inputted into the predistortion system each time being obtained by combining a historical value and a current value in the training set; inputting the output of the radio frequency power amplifier output feedback loop in the predistortion system into the complex neural network through a second real-time power normalization unit included in the predistortion system, or directly inputting the output into the complex neural network; According to the partial derivative or sensitivity of the loss function of the complex neural network in the pre-distortion system with respect to each element of the correction vector output by the complex neural network, the partial derivative or sensitivity of the loss function with respect to the weight parameter and the bias parameter of each layer in the complex neural network is determined; The weight parameter and the bias parameter of the corresponding layer are updated according to the determined partial derivative or sensitivity of each layer; Wherein, the expression of the complex vector input into the pre-distortion system each time is as follows: ； wherein, is a complex vector input to the predistortion system, is a power amplifier memory effect parameter, , is the length of the training set, is the element of the training set, is the current value, is an integer greater than 1 and less than , , , and are the elements of the training set, , and , , and are the history values.

8. The method of claim 7, wherein, The relationship between the complex vector input into the complex neural network in the pre-distortion system and the correction vector output by the complex neural network is as follows: ； wherein, , is a correction vector, , , and are the first , , and elements of the correction vector, denotes a composite function of the complex neural network internal layer-wise operation function; The relationship between the complex vector input into the pre-distortion multiplier in the pre-distortion system, the correction vector and the output of the pre-distortion multiplier is as follows: ； wherein , is a complex vector of pre-distortion multiplier outputs, , , and is the kth element of the complex vector of pre-distortion multiplier outputs, , , and is the kth element of the complex vector of pre-distortion multiplier outputs, is a complex vector of inputs to the pre-distortion multiplier, denotes a dot product; The relationship between the output of the pre-distortion multiplier and the output of the radio frequency power amplifier output feedback loop is as follows: ； wherein, is an output of a radio frequency power amplifier output feedback loop, represents a processing function of the radio frequency power amplifier output feedback loop on an input signal; In the case that the pre-distortion system comprises a second real-time power normalization unit, the relationship between the input and the output of the second real-time power normalization unit is as follows: ； ； wherein, is the first output of the second real-time power normalization unit, is the second output of the second real-time power normalization unit, is the first input of the second real-time power normalization unit, is the second input of the second real-time power normalization unit, is a positive integer, is a positive integer greater than or equal to 1 and less than or equal to is a positive integer greater than or equal to 1 and less than or equal to is the first output of the radio frequency power amplifier output feedback loop, and is the second output of the radio frequency power amplifier output feedback loop.

9. The method of claim 8, wherein, According to the partial derivative or sensitivity of the loss function of the complex neural network in the pre-distortion system with respect to each element of the correction vector output by the complex neural network, the partial derivative or sensitivity of the loss function with respect to the weight parameter and the bias parameter of each layer in the complex neural network is determined, comprising: Based on the feedback quantity of the radio frequency power amplifier output feedback loop and the complex vector input into the complex neural network, the loss function of the complex neural network is determined; The partial derivative or sensitivity of the loss function with respect to each element of the correction vector output by the complex neural network is determined; According to the partial derivative or sensitivity of the correction vector, the partial derivative or sensitivity of the loss function with respect to the weight parameter and the bias parameter of each layer in the complex neural network is determined; The weight parameter and the bias parameter of the corresponding layer are updated based on the partial derivative or sensitivity of the weight parameter and the bias parameter of each layer.

10. The method of claim 9, wherein the loss function is determined by the following calculation expression: ； is the loss function.

11. The method of claim 9, wherein, The calculation expression of the partial derivative or sensitivity of each element of the correction vector is as follows: ； wherein is a loss function, is a partial derivative or sensitivity of an initial correction vector element, is a partial derivative or sensitivity of the loss function with respect to a complex vector element of the pre-distortion multiplier output, denotes taking the complex conjugate of a complex number; The determination is calculated by the following calculation expression: ； wherein , " denotes a scalar multiplication; " denotes a scalar multiplication; is a non-linear order parameter, is an element of a complex vector output by the pre-distortion multiplier, denotes taking the real part, denotes taking the imaginary part, denotes an intermediate partial derivative or sensitivity.

12. The method of claim 11, wherein, in case the pre-distortion system does not comprise a second real-time power normalization unit, is determined by the following calculation expression: ； In case the pre-distortion system comprises a second real-time power normalization unit, is determined by the following calculation expression: ； wherein, is a complex coefficient obtained by a least square algorithm based on the complex signal in the training set and the output of the radio frequency power amplifier output feedback loop with the input of the complex signal; is a partial derivative or sensitivity of the loss function with respect to a complex scalar of the output of the radio frequency power amplifier output feedback loop. The determination is calculated by the following calculation expression: ； wherein, is the derivative or sensitivity of the loss function with respect to the feedback quantity of the radio frequency power amplifier output feedback loop; The determination is calculated by the following calculation expression: 。 13. The method of claim 9, wherein, The partial derivative or sensitivity of the weight parameter and the bias parameter of the fully connected layer in the complex neural network is respectively as follows: ； ； wherein represents a complex weight parameter of a connection from a th neuron of a th layer to a th neuron of a th layer of the complex neural network, represents a bias parameter of a th neuron of a th layer of the complex neural network, and respectively represent output complex signals of a th neuron of a th layer and a th neuron of a th layer of the complex neural network; represents a derivative of a neuron activation function with respect to an input signal, is a partial derivative or sensitivity of a weight parameter of a fully connected layer, is a partial derivative or sensitivity of a bias parameter of a fully connected layer, is a partial derivative or sensitivity of a loss function with respect to , is a loss function.

14. The method of claim 9, wherein, The convolutional layer inside the complex neural network is the first... The partial derivatives or sensitivities of the weight parameters and bias parameters of each convolution kernel are as follows: ； ； wherein, is the derivative or sensitivity of the weight parameter of the th convolution kernel, is the derivative or sensitivity of the bias parameter of the th convolution kernel, denotes the previous layer of the convolution layer, is the th complex vector of the layer output, denotes the th complex vector of the layer output, denotes the th complex vector of the layer output, wherein, and are the number of output feature vectors of the th layer and the th layer, respectively, denotes the derivative or sensitivity of the loss function with respect to , denotes the position inverse of the input vector, is the loss function, is the complex weight of the th convolution kernel, is the complex bias of the th convolution kernel, denotes the convolution operation.

15. The method of claim 9, wherein, The weight parameter and the bias parameter of the fully connected layer in the complex neural network are updated by the following expression: ； ； wherein, and respectively represent the weight parameters of the fully connected layer at the current time and the previous time; and respectively represent the bias parameters of the fully connected layer at the current time and the previous time; represents the update step size of the weight parameters at the current time; represents the update step size of the bias parameters at the current time, is the value of the partial derivative or sensitivity of the weight parameters of the fully connected layer at the current time, is the value of the partial derivative or sensitivity of the bias parameters of the fully connected layer at the current time.

16. The method of claim 3, wherein, Based on the test set, the trained pre-distortion system is tested to obtain the error vector magnitude and the adjacent channel leakage ratio corresponding to the pre-distortion system, comprising: The complex vector in the test set is input into the pre-distortion system according to the signal index, the length of each input is determined based on the power amplifier memory effect parameter, and the complex vector input into the pre-distortion system each time is obtained by combining the historical value and the current value in the training set; The output of the radio frequency power amplifier output feedback loop in the pre-distortion system is input into the pre-distortion system to determine the generalization error vector magnitude and the generalization adjacent channel leakage ratio corresponding to the pre-distortion system.

17. The method of claim 16, the generalization error vector magnitude and the generalization crosstalk leakage ratio are determined by the following computational expressions, respectively: ； ； ； ； wherein denotes error vector magnitude, denotes adjacent channel leakage ratio, denotes the feedback quantity of the radio frequency power amplifier output feedback loop, denotes the first element feedback quantity of the radio frequency power amplifier output feedback loop, denotes the first element of the corresponding complex vector in the test set, denotes the length of the training set, denotes the sum of and , denotes the length of the complex signal for training, denotes half of the effective signal bandwidth, denotes half of the guard bandwidth, denotes the number of points of the discrete Fourier transform; is the coefficient vector of the window function with window length , denotes a signal of length randomly cut from the output complex signal of the test set; is a uniformly distributed random positive integer in the range , is the number of random cuts, denotes the dot product.

18. A predistortion system characterized by, The pre-distortion method as claimed in any one of claims 1-17 is executed, and the pre-distortion system comprises a pre-distortion multiplier, a complex neural network, and a radio frequency power amplifier output feedback loop; a first input terminal of the pre-distortion multiplier is an input terminal of the pre-distortion system, and is connected with a first input terminal and a second input terminal of the complex neural network, an output terminal of the pre-distortion multiplier is connected with an input terminal of the radio frequency power amplifier output feedback loop, an output terminal of the radio frequency power amplifier output feedback loop is an output terminal of the pre-distortion system, and the output terminal of the radio frequency power amplifier output feedback loop is connected with the second input terminal of the complex neural network, and an output terminal of the complex neural network is connected with a second input terminal of the pre-distortion multiplier.

19. A predistortion system according to claim 18, characterized in that The radio frequency power amplifier output feedback loop comprises a digital-to-analog converter unit, a radio frequency modulation unit, a power amplifier unit, an analog-to-digital converter unit, and a radio frequency demodulation unit.

20. A predistortion system characterized by, The pre-distortion method as claimed in any one of claims 1-17 is executed, and the pre-distortion system comprises a pre-distortion multiplier, a complex neural network, a radio frequency power amplifier output feedback loop, a first real-time power normalization unit, and a second real-time power normalization unit; an input terminal of the first real-time power normalization unit is an input terminal of the pre-distortion system, an output terminal of the first real-time power normalization unit is connected with the first input terminal of the pre-distortion multiplier, the first input terminal of the complex neural network, and the second input terminal of the complex neural network, respectively, an output terminal of the pre-distortion multiplier is connected with an input terminal of the radio frequency power amplifier output feedback loop, an output terminal of the radio frequency power amplifier output feedback loop is an output terminal of the pre-distortion system, the output terminal of the radio frequency power amplifier output feedback loop is connected with an input terminal of the second real-time power normalization unit, an output terminal of the second real-time power normalization unit is connected with the second input terminal of the complex neural network, and an output terminal of the complex neural network is connected with a second input terminal of the pre-distortion multiplier.

21. An apparatus, comprising: comprise: one or more processors; a memory device for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in any one of claims 1-17.

22. A storage medium, characterized by The storage medium stores a computer program, and the computer program is executed by a processor to implement the method as claimed in any one of claims 1-17. The storage medium stores a computer program, and the computer program is executed by a processor to implement the method as claimed in any one of claims 1-17.

Images