Pre-distortion signal generation device and method thereof, and target signal generation circuit and method thereof
By optimizing the predistortion signal generation device through neural networks, the complexity and accuracy problems of power amplifier nonlinear modeling under large bandwidth in wireless communication systems are solved, achieving efficient predistortion signal generation. It is suitable for power amplifier applications in RF links and improves the linearity of the signal and the adjacent channel power ratio performance.
Patent Information
- Application Number
- CN202410895611.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-04
- Publication Date
- 2026-01-06
AI Technical Summary
In wireless communication systems, as bandwidth increases, the difficulty of nonlinear modeling and predistortion compensation processing of power amplifiers increases. Existing technologies struggle to guarantee model accuracy and reduce computational complexity in high-bandwidth scenarios, and the real-time performance and engineering deployment feasibility of neural network predistortion processing are limited.
A neural network-based predistortion signal generation device is adopted. The parameters are optimized by using the feedback signal of the target signal generation circuit and the predistortion signal parameter generation model to generate nonlinear basis and weights, thereby reducing network complexity and improving the nonlinear performance of the predistortion signal.
While ensuring the accuracy of predistorted signal parameters, it reduces the complexity of network implementation, is suitable for high bandwidth scenarios, improves the linearity of the signal and the adjacent channel power ratio performance, and has versatility and generalization.
Smart Images
Figure CN121283373A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of communications, and in particular to a predistortion signal generation device and method, and a target signal generation circuit and method. Background Technology
[0002] For the radio frequency (RF) link of a wireless communication system, the nonlinear characteristics of RF power devices limit the behavioral modeling of the power amplifier and the corresponding equalization compensation algorithm, which are crucial for the coverage of the communication cell, inter-user interference within the cell, communication quality assurance, and equipment robustness. With the significant increase in bandwidth of current wireless communication systems in standards and frequency bands such as 5G (fifth-generation mobile communication technology), 6G (sixth-generation mobile communication technology), and millimeter wave, OFDMA (Orthogonal Frequency Division Multiple Access) is often used for multi-user downlink coverage. This makes the peak-to-average power ratio (PAPR) problem of the output signal more pronounced, leading to increased complexity in power amplifier nonlinear modeling and greater difficulty in front-end predistortion compensation processing. Summary of the Invention
[0003] This disclosure provides a predistortion signal generation apparatus and method, and a target signal generation circuit and method.
[0004] In a first aspect, embodiments of this disclosure provide a predistortion signal generation apparatus, which includes: a predistortion signal generation module and a predistortion signal parameter module; the predistortion signal parameter module includes a predistortion signal parameter generation model;
[0005] The predistortion signal parameter module is used to obtain the feedback signal of the target signal generation circuit, and optimize the parameters of the predistortion signal parameter generation model based on the feedback signal to obtain the predistortion signal parameters.
[0006] The predistortion signal generation module is used to generate a predistortion signal corresponding to the target input signal based on the predistortion signal parameters and the target input signal.
[0007] Secondly, embodiments of this disclosure also provide a target signal generation circuit, including the aforementioned predistortion signal generation device.
[0008] Thirdly, embodiments of this disclosure also provide a method for generating a predistortion signal, comprising:
[0009] Obtain the feedback signal from the target signal generation circuit, and optimize the parameters of the preset predistortion signal parameter generation model based on the feedback signal to obtain the predistortion signal parameters;
[0010] Based on the predistortion signal generation parameters and the target input signal, a predistortion signal corresponding to the target input signal is generated.
[0011] Fourthly, embodiments of this disclosure also provide a method for generating a target signal, comprising:
[0012] The target signal output signal is generated based on the predistortion signal generated by the aforementioned predistortion signal generation method.
[0013] The scheme of this disclosure optimizes parameters through the feedback signal of the target signal generation circuit and the predistortion signal parameter generation model, and finally obtains the predistortion signal parameters, ensuring the accuracy of the predistortion signal parameters. It can generate predistortion signal parameters specifically based on the target signal feedback data while ensuring accuracy, such as a nonlinear basis, and generate a predistortion signal based on the predistortion signal parameters. This ensures the nonlinear performance of the predistortion signal, reduces the implementation complexity of the network, and does not limit the type of target signal, thus having the characteristics of versatility and generalization. Attached Figure Description
[0014] In the accompanying drawings of the embodiments disclosed herein:
[0015] Figure 1 A block diagram of the predistortion signal generation apparatus provided in the embodiments of this disclosure;
[0016] Figure 2 This is a schematic diagram of the predistortion signal generation device provided in an embodiment of the present disclosure;
[0017] Figure 3 This is a block diagram of the target signal generation circuit provided in the embodiments of this disclosure;
[0018] Figure 4 A flowchart of a predistortion signal generation method provided in an embodiment of this disclosure. Detailed Implementation
[0019] To enable those skilled in the art to better understand the technical solutions of this disclosure, the communication-sensing data processing method and computer-readable storage medium provided in the embodiments of this disclosure will be described in detail below with reference to the accompanying drawings.
[0020] The present disclosure will be described more fully below with reference to the accompanying drawings; however, the embodiments shown may be embodied in different forms, and the present disclosure should not be construed as limited to the embodiments set forth below. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will enable those skilled in the art to fully understand the scope of the disclosure.
[0021] The accompanying drawings of the embodiments disclosed herein are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the detailed embodiments to explain this disclosure and do not constitute a limitation thereof. The above and other features and advantages will become more apparent to those skilled in the art from the description of the detailed embodiments with reference to the accompanying drawings.
[0022] This disclosure may be described with reference to plan and / or cross-sectional views using the ideal schematic diagrams of this disclosure. Therefore, the example illustrations may be modified according to manufacturing techniques and / or tolerances.
[0023] Where there is no conflict, the various embodiments of this disclosure and the features thereof in the embodiments may be combined with each other.
[0024] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the disclosure. The term "and / or" as used in this disclosure includes any and all combinations of one or more of the associated enumerated entries. The singular forms "a" and "the" as used in this disclosure are also intended to include the plural forms, unless the context clearly indicates otherwise. The terms "comprising," "made of," etc., as used in this disclosure specify the presence of the stated feature, integral, step, operation, element, and / or component, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof.
[0025] Unless otherwise specified, all terms used in this disclosure (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and will not be interpreted as having an idealized or overly formal meaning, unless expressly so defined in this disclosure.
[0026] This disclosure is not limited to the embodiments shown in the accompanying drawings, but includes modifications to the configuration based on the manufacturing process. Therefore, the areas illustrated in the drawings are schematic, and the shapes of the areas shown illustrate specific shapes of the areas of an element, but are not intended to be limiting.
[0027] In practical applications, the RF link between the antenna and the power amplifier is often far apart. Linear mismatch in this RF link leads to significant standing waves (VSWR), posing a substantial threat to the robustness of the power amplifier. The reflected signals generated by the mismatch between the power amplifier output port and the antenna port not only risk overvoltage breakdown but also severely impact the accuracy of behavioral modeling and predistortion compensation. From the perspective of the power amplifier's nonlinear behavior, both the compression distortion under high bandwidth peak-to-average power ratio (PAPR) and the alteration of the power amplifier's output amplitude by reflected signals under high VSWR can be attributed to similar power amplifier compression distortion behavior problems.
[0028] Based on specific power amplifier application scenarios, it can be seen that high-precision modeling and pre-distortion compensation of the power amplifier's behavior can greatly improve the environmental adaptability and operational robustness of the transmitter in wireless broadband communication.
[0029] Existing typical predistortion processing schemes are mostly based on Volterra series nonlinear modeling using polynomial expansions, such as MP (Memory Polynomial, a digital predistortion polynomial expansion model describing time memory models), GMP (Generalized Memory Polynomial, a digital predistortion polynomial expansion model generalized from the MP model), and DDR (Dynamic Derivation Reduction, a truncated polynomial expansion model for modeling large-bandwidth nonlinear compression). These schemes mainly utilize higher-order terms of the signal's magnitude and time delay cross-terms to form a basis, simulating the nonlinear behavior of the power amplifier or constituting predistortion compensation in the form of polynomial expansion. In high-bandwidth scenarios, traditional polynomial basis requires more higher-order terms and time delay taps to ensure the accuracy of depicting power amplifier behavior or predistortion. However, the extremely high correlation between higher-order terms not only makes the selection and identification of effective basis very difficult, but the resulting ill-conditioned computational matrix is also highly sensitive to noise during training, making it difficult to guarantee the modeling accuracy of power amplifier behavior or predistortion networks in high-bandwidth scenarios.
[0030] In recent years, neural networks have demonstrated significant technological advantages in data feature extraction and nonlinear function fitting. Through various data combination relationships and the design of nonlinear activation functions, processing devices composed of neural networks can more accurately identify the nonlinear characteristic relationships inherent in the data than traditional polynomial expansion methods, and construct higher-performance power amplifier predistortion compensation networks. However, existing neural networks for power amplifier behavior and predistortion compensation are typically end-to-end structures, meaning the input data is directly fitted by the neural network, placing all the iterative computational burden on the neural network itself. This means that as the bandwidth of the predistortion signal increases, the predistortion network needs to stack more neurons to achieve good output linearity, and a large amount of data is required during training to ensure the network's general applicability. Furthermore, existing neural network-based predistortion processing, due to its complex implementation of nonlinear activation functions, cannot operate independently of deep learning platforms, severely limiting the real-time performance and engineering deployment feasibility of such predistortion processing.
[0031] In the field of wireless communication, where high-bandwidth signals are nonlinearly compressed by power amplifiers, the mainstream technique is based on nonlinear modeling expansion using polynomial models such as Volterra series (a type of functional series), and completes the compression by constructing a nonlinear basis that induces the nonlinearity. However, while this approach approximates the nonlinear behavior and predistortion model of the power amplifier with higher-order expansions of the signal, in broadband signal scenarios, it typically requires not only higher-order and more time-delay taps to form the basis, but also excessively high correlation between higher-order terms. This results in a bottleneck in the predistortion accuracy of such polynomial expansion broadband models, and also leads to excessively high computational complexity.
[0032] In view of the many problems existing in the current related technologies, this disclosure proposes a predistortion signal generation device based on neural network parameter optimization.
[0033] The scheme of this disclosure optimizes parameters through the feedback signal of the target signal generation circuit and the predistortion signal parameter generation model, and finally obtains the predistortion signal parameters, ensuring the accuracy of the predistortion signal parameters. It can generate predistortion signal parameters specifically based on the target signal feedback data while ensuring accuracy, such as a nonlinear basis, and generate a predistortion signal based on the predistortion signal parameters, thus ensuring the nonlinear performance of the predistortion signal and reducing the implementation complexity of the network. Moreover, it does not limit the type of target signal, for example, it can include but is not limited to power amplifier signals, and has the characteristics of versatility and generalization.
[0034] The solutions disclosed herein can be applied to all nonlinear scenarios. They can be implemented in any scenario requiring a pre-distorted signal, including, but not limited to, power amplifier applications requiring pre-distorted signals. For example, they can be applied to RF links with power amplifiers to solve the nonlinear distortion problem that occurs after the signal passes through the power amplifier. Specifically, the solutions disclosed herein can be applied to the power amplifier circuit of a base station. In RF links, they can be mainly applied at the front end of the RF power amplifier. Before being injected into the RF power amplifier, the digital signal of the target signal is sent to the pre-distortion signal generation device of this disclosure to generate a pre-distorted signal. Then, the pre-distorted signal is injected into the power amplifier through digital-to-analog conversion.
[0035] Furthermore, the nonlinear scenarios covered by the embodiments of this disclosure may include, but are not limited to, narrowband, high-bandwidth power amplifier scenarios, and high VSWR scenarios. In this disclosure, the solution will be illustrated using the scenario in the field of wireless communication where a high-bandwidth signal is nonlinearly compressed by a power amplifier as an example.
[0036] The embodiments of this disclosure will be described in detail below.
[0037] This disclosure provides a predistortion signal generation device 1, which includes: a predistortion signal generation module 11 and a predistortion signal parameter module 12; the predistortion signal parameter module 12 includes a predistortion signal parameter generation model 121;
[0038] The predistortion signal parameter module 12 is used to obtain the feedback signal of the target signal generation circuit, and optimize the parameters of the predistortion signal parameter generation model 121 based on the feedback signal to obtain the predistortion signal parameters.
[0039] The predistortion signal generation module 11 is used to generate a predistortion signal corresponding to the target input signal based on the predistortion signal parameters and the target input signal.
[0040] In this embodiment of the disclosure, the target signal generation circuit is used to generate a target signal based on the input target signal. The target signal can be any signal that needs to be generated based on a predistorted signal, such as, but not limited to, a power amplifier signal. The target signal generation circuit can be, but is not limited to, a power amplifier generation circuit, and the target input signal can be, but is not limited to, a power amplifier input signal.
[0041] In this embodiment of the disclosure, the feedback signal of the target signal generation circuit may include, but is not limited to, the target signal, or it may be an intermediate signal generated by the target signal generation circuit, such as a pre-distortion signal.
[0042] In the embodiments disclosed herein, such as Figure 2 As shown, the predistortion signal generation module 11 can be applied to the target signal generation circuit (which can be called the downlink of the signal), and the predistortion signal parameter module 12 can be applied to the feedback link of the target signal generation circuit.
[0043] In this embodiment of the disclosure, the predistortion signal generation module 11 is a part of the main circuit of the target signal generation circuit. The predistortion signal generation module 11 can generate a predistortion signal according to the target input signal, nonlinear basis parameters and weights. Other parts of the target signal generation circuit (such as the digital-to-analog conversion circuit and the target signal generation device) can generate the target signal according to the predistortion signal.
[0044] In this embodiment, the predistortion signal parameter module 12 is not part of the main circuit of the target signal generation circuit, but can be regarded as the feedback circuit of the target signal generation circuit. It is mainly used to optimize the nonlinear base parameters and weights required by the predistortion signal generation module 11 so that the predistortion signal generation module 11 can generate a more accurate predistortion signal.
[0045] In this embodiment of the disclosure, the predistortion signal parameter module 12 may include an analog-to-digital converter (ADC) module, which converts the analog signal of the feedback signal into a digital signal.
[0046] In this embodiment of the disclosure, the predistortion signal parameters may include, but are not limited to: nonlinear basis parameters and weights required to calculate the predistortion signal; the predistortion signal parameter generation model 121 may include, but is not limited to: nonlinear basis generation network 1211 and weight generation network 1212;
[0047] The nonlinear basis generation network 1211 is used to perform dimensionality reduction and screening of the basis function set contained in the nonlinear basis generation network itself based on the feedback signal, thereby optimizing the nonlinear basis generation network parameters corresponding to the nonlinear basis generation network and extracting the optimized nonlinear basis generation network parameters as nonlinear basis parameters.
[0048] Weight generation network 1212 is used to generate weights.
[0049] In this embodiment of the disclosure, the weights may include, but are not limited to, a set of values (which may be one or more). For each set of nonlinear basis parameters (which may be one or more) generated by the nonlinear basis generation network 1211, the weight generation network 1212 will generate a set of weights (which may be one or more). Therefore, the nonlinear basis parameters generated each time correspond to the weights generated.
[0050] In this embodiment of the disclosure, neural networks, due to their excellent performance in nonlinear feature extraction and behavior fitting, can be applied to signal analysis in the scheme of this embodiment. For example, the predistortion signal parameter generation model 121 may include, but is not limited to, a nonlinear basis generation network 1211 and a weight generation network 1212. The neural network that generates the nonlinear basis parameters (nonlinear basis generation network 1211) and the neural network that generates the corresponding weights (weight generation network 1212) both need to perform nonlinear feature learning of the target signal based on a deep learning platform (e.g., power amplifier nonlinear feature learning). After the learning is completed, the network parameters of the nonlinear basis generation network 1211 and the corresponding weights generated by the weight generation network 1212 are extracted and applied to a digital signal processing platform. First, a nonlinear basis is generated, and then the predistortion signal is generated through weight processing.
[0051] In this embodiment of the disclosure, the nonlinear basis generation network 1211 may include, but is not limited to, a fully connected layer network. Accordingly, the network parameters of the nonlinear basis generation network 1211 include fully connected parameters.
[0052] In this embodiment of the disclosure, the nonlinear basis generation network 1211 can be a multi-layer fully connected structure. The nonlinear basis parameters output by the nonlinear basis generation network 1211 can be realized by projecting nonlinear activation onto a higher-order nonlinear space. Each fully connected layer combined with nonlinear activation can be represented in the following form:
[0053]
[0054] In this embodiment of the disclosure, g is a nonlinear basis parameter of a fully connected layer, and f Nonlinear () represents the nonlinear activation function, X represents the nonlinear basis parameters, and W and b represent the weights and biases of the fully connected layer.
[0055] In this embodiment, the nonlinear activation function can be implemented using an existing nonlinear activation function that has been verified for functionality and convergence.
[0056] In this embodiment of the disclosure, the nonlinear basis generation network 1211 has a multi-layer fully connected structure, which can be expressed in the following form:
[0057] g = f n …f3(f2(f1(X)));
[0058] In this embodiment, g represents the final generated nonlinear basis parameters, ranging from n (where n is a positive integer) to 1. The weight scale in the fully connected architecture decreases step by step, and the implicit screening of the nonlinear space is actually completed by combining the nonlinear activation function.
[0059] In this embodiment of the disclosure, since there is no input, the weight generation network 1212 can be regarded as a "black box", and the weight optimization is completed together by the neural network optimizer and the nonlinear basis generation network 1211.
[0060] In this embodiment, since the generated nonlinear basis parameters and corresponding weights are intermediate results, they cannot be iteratively solved using optimization methods such as the LMS (Least Mean Square) algorithm. However, other neural networks, including but not limited to recurrent neural network structures, can be used as substitutes. The nonlinear basis generation network 1211 can complete the substitution as long as it can ensure that the substitution network can be converted into parameter operations, while the weight generation has no limitations and any neural network can be used as a substitute.
[0061] In this embodiment, the predistortion signal generation device comprehensively considers the strong nonlinear behavior extraction capability of the neural network model and the effectiveness of traditional basic elements. This allows the predistortion signal generation device to utilize multi-layer fully connected layers and nonlinear activation functions for approximate representation, selectively generating predistortion nonlinear bases. Through successive dimensionality reduction, it can implicitly filter redundant nonlinear features while acquiring nonlinear behavior during training, achieving implicit filtering of high-dimensional features. This reduces the number of nonlinear bases during network deployment and significantly improves the linearity of the predistortion device while ensuring that complexity does not increase with bandwidth. It also reduces the design complexity of the internal neural network while maintaining the linearity of the predistortion device's output signal.
[0062] In this embodiment of the disclosure, by projecting a nonlinear activation function onto a higher-order nonlinear space, more higher-order terms can be excited under the constraints of order and time delay terms, and higher performance can be obtained than other similar technologies.
[0063] In this embodiment of the disclosure, the nonlinear basis generation network 1211 and the weight generation network 1212 can be used offline after one training optimization, that is, no further optimization is required. The nonlinear basis generation network 1211 can directly generate nonlinear basis parameters based on the feedback signal, and the weight generation network 1212 can directly generate weights corresponding to the nonlinear basis parameters. The nonlinear basis parameters and weights are synchronously input into the predistortion signal generation module to generate a predistortion signal, and then the final target signal (e.g., the power amplifier signal obtained after power amplification) is generated based on the predistortion signal.
[0064] In this embodiment of the disclosure, when the nonlinear basis generation network 1211 and the weight generation network 1212 are trained and optimized only once, a large number of target signals and target input signals can be collected in advance. Based on these target signals and target input signals, a training set and a validation set are constructed. The predistortion signal parameter module 12 of this embodiment of the disclosure is used to train the nonlinear basis generation network 1211 and the weight generation network 1212.
[0065] In this embodiment of the disclosure, for example, for a high-bandwidth power amplifier scenario, the preparation of the training set can mainly consider the configuration of odd harmonics under the high-bandwidth scenario, and the mathematical expression of the training set can be expressed as follows:
[0066]
[0067] Where x(n) is the power amplifier output signal, M corresponds to the number of time delay taps in the instance, p is the signal order, q describes the order of the signal magnitude, T represents the transpose operation, and the validation set corresponding to the training set is the target input signal corresponding to the target signal.
[0068] In this embodiment, the nonlinear basis generation network 1211 and the weight generation network 1212 can also be used online. That is, the training set for each training of the nonlinear basis generation network 1211 and the weight generation network 1212 is only the target signal output in that iteration, and the corresponding validation set is only the target input signal input in that iteration. The nonlinear basis generation network 1211 can generate nonlinear basis parameters based on the feedback signal, and the weight generation network 1212 generates weights corresponding to the nonlinear basis parameters. Based on the generated nonlinear basis parameters and weights, the network parameters of the nonlinear basis generation network 1211 and the weight generation network 1212 are iteratively optimized until the optimization is completed. Only then does the nonlinear basis generation network 1211 output the final nonlinear basis parameters, and the weight generation network 1212 generates the corresponding weights. Simultaneously, a predistortion signal generation module is input to generate a predistortion signal, and the final target signal is then generated based on the predistortion signal.
[0069] In this embodiment of the disclosure, the online application scheme of the nonlinear basis generation network 1211 and the weight generation network 1212 will be described in detail below.
[0070] In this embodiment of the disclosure, the predistortion signal parameter module 12 may further include: a projection layer 122;
[0071] Projection layer 122 is used to linearly combine nonlinear basis parameters and weights to obtain the network output signal.
[0072] In this embodiment of the disclosure, the projection layer 122 projecting the nonlinear basis parameters and weights into a higher-order nonlinear space can be understood as performing a preset operation on the nonlinear basis parameters and weights, which may include, but is not limited to, multiplication and addition operations.
[0073] In this embodiment of the disclosure, the projection layer 122 can be a custom projection layer used to generate a linear combination of nonlinear basis parameters and weights, and the mathematical expression can be as follows:
[0074]
[0075] in, H represents the signal estimate of the neural network output (i.e., the network output signal), and H represents the weight.
[0076] In this embodiment of the disclosure, the predistortion signal parameter module 12 may further include: a network optimizer 123;
[0077] The network optimizer 123 is used to perform loss calculation based on the network output signal and the target input signal corresponding to the feedback signal. If the loss calculation result does not meet the preset requirements, it generates parameter optimization instructions and inputs them into the nonlinear basis generation network 1211 and the weight generation network 1212.
[0078] Nonlinear basis generation network 1211 and weight generation network 1212 are used to iteratively optimize their respective network parameters based on parameter optimization instructions until the loss calculation results meet the preset requirements; wherein, the network parameters corresponding to nonlinear basis generation network 1211 are nonlinear basis generation network parameters, and the network parameters corresponding to weight generation network 1212 are weight generation network parameters.
[0079] In this embodiment, the network output signal output by the projection layer 122 can be input into the network optimizer 123 to perform loss calculation with the validation set (e.g., the data corresponding to the target input signal). If the loss calculation result does not meet the preset requirements, a parameter optimization instruction is generated, and feedback optimization is performed on the nonlinear basis generation network 1211 and the weight generation network 1212 based on the parameter optimization instruction. The parameter optimization instruction may include a parameter optimization scheme.
[0080] In this embodiment of the disclosure, the optimization method in the network optimizer 123 may include, but is not limited to, optimization based on the nonlinear Winner model (output nonlinear model), or the nonlinear Hammerstein model (input nonlinear model), or the Winner-Hammerstein model. Since all three have been verified to solve nonlinear problems, they can be used interchangeably.
[0081] In this embodiment of the disclosure, the nonlinear basis parameters may include first nonlinear basis parameters, and the weights may include first weights; the nonlinear basis generation network and the weight generation network iteratively optimize their respective network parameters based on parameter optimization instructions, which may include:
[0082] The nonlinear basis generation network 1211 is used to optimize the parameters of the nonlinear basis generation network based on the parameter optimization instructions, obtain the first nonlinear basis parameters based on the optimized nonlinear basis generation network parameters, and input them into the predistortion signal generation module 11 so that the predistortion signal generation module 11 generates the first nonlinear basis based on the first nonlinear basis parameters and the target input signal.
[0083] The weight generation network 1212 is used to optimize the parameters of the weight generation network based on the parameter optimization instructions, and obtain the first weight based on the optimized weight generation network parameters, so that the predistortion signal generation module 11 generates the first predistortion signal based on the first nonlinear basis and the first weight, and the target signal generation circuit uses the target signal generated based on the first predistortion signal as a new feedback signal for the next optimization of the network parameters of the nonlinear basis generation network 1211 and the weight generation network 1212.
[0084] In this embodiment of the disclosure, during the next iteration of optimization, the nonlinear basis generation network 1211 performs dimensionality reduction and screening on the basis function set contained in the nonlinear basis generation network itself based on the new feedback signal, thereby optimizing the nonlinear basis generation network parameters corresponding to the nonlinear basis generation network again, and extracting the optimized nonlinear basis generation network parameters as new nonlinear basis parameters, which are then input into the projection layer; the weight generation network 1212 also inputs the first weight into the projection layer.
[0085] In this embodiment of the disclosure, after the new nonlinear basis parameters and the first weights are input into the projection layer, a linear combination can be performed again based on the projection layer to obtain a new network output signal. Based on the new network output signal, loss calculation is performed in the network optimizer 123. Based on the loss calculation result, it is determined whether to continue optimizing the network parameters of the nonlinear basis generation network 1211 and the weight generation network 1212, or to end the optimization and output the final nonlinear basis and weights.
[0086] In this embodiment of the disclosure, the nonlinear basis parameters may further include a second nonlinear basis parameter, and the weights may further include a second weight.
[0087] The network optimizer 123 is used to perform loss calculation based on the network output signal and the target input signal corresponding to the feedback signal. When the loss calculation result meets the preset requirements, it generates an optimization completion instruction and inputs it into the nonlinear basis generation network and the weight generation network.
[0088] The nonlinear basis generation network 1211 is also used to save the optimized nonlinear basis generation network parameters based on the optimization completion instruction, generate second nonlinear basis parameters based on the optimized nonlinear basis generation network parameters, and send the second nonlinear basis parameters to the predistortion signal generation module.
[0089] The weight generation network 1212 is used to save the optimized weight generation network parameters based on the optimization completion instruction, generate a second weight based on the optimized weight generation network parameters, and send the second weight to the predistortion signal generation module 11.
[0090] In this embodiment of the disclosure, if the loss calculation result meets the preset requirements, it is said that the optimization is complete. At this time, the network parameters of the nonlinear basis generation network 1211 and the weight generation network 1212 no longer need to be optimized, and the nonlinear basis generation network parameters of the nonlinear basis generation network 1211 are directly output as nonlinear basis parameters. The weight generation network 1212 outputs the corresponding weights, thereby calculating the predistortion signal based on the output nonlinear basis parameters and weights.
[0091] In this embodiment of the disclosure, in the online usage scheme of the nonlinear basis generation network 1211 and the weight generation network 1212, the parameters of the nonlinear basis generation network and the weight generation network obtained after each optimization can be saved for use in the next iteration optimization step.
[0092] In this embodiment of the disclosure, the nonlinear basis generation network 1211 and the weight generation network 1212 are further configured to synchronously send the second nonlinear basis parameters and the second weights to the predistortion signal generation module 11.
[0093] In this embodiment of the disclosure, the nonlinear basis parameters (e.g., fully connected parameters) output by the nonlinear basis generation network 1211 and the weights output by the weight generation network 1212 are synchronously exported to the downlink, ensuring that the nonlinear basis parameters and weights are matched.
[0094] In this embodiment of the disclosure, the predistortion signal generation module 11 includes: a nonlinear basis generator 111 and a predistortion signal generator 112;
[0095] A nonlinear basis generation network 1211 is used to input nonlinear basis parameters into a nonlinear basis generator 111;
[0096] The weight generation network 1212 is used to synchronously input weights into the predistortion signal generator 112.
[0097] In this embodiment of the disclosure, the nonlinear basis generator 111 is used to generate a nonlinear basis based on the target input signal and the nonlinear basis parameters;
[0098] A predistortion signal generator 112 is used to generate a predistortion signal based on a nonlinear basis and weights.
[0099] In this embodiment of the present disclosure, the predistortion signal generation device based on the solution of this embodiment of the present disclosure can generate the predistortion signal required for any target signal, thereby effectively reducing the nonlinear distortion problem in the target signal generation circuit.
[0100] This disclosure embodiment includes at least the following advantages:
[0101] 1. The present invention proposes a predistortion signal generation device with an innovative structure, including a nonlinear basis generation network, a weight generation network separated from the nonlinear basis generation network, a special parameter iteration method, and synchronous updates of the two networks, which has the ability to process nonlinear compressed signals in real time.
[0102] 2. By generating nonlinear bases in a targeted manner through a fully connected network structure, the problems of high-order term correlation and base selection in traditional polynomial expansion models, including those with large bandwidth, are solved. Furthermore, implicit selection of nonlinear bases is achieved through a multi-level fully connected layer with dimensionality reduction, which reduces the number of bases and lowers the computational complexity of the model.
[0103] 3. Some other neural network-based nonlinear predistorters can only be applied in scenarios with narrow bandwidth. If applied in scenarios with large bandwidth, their performance will deteriorate rapidly. The solution of this disclosure can achieve better predistortion performance in extended scenarios (e.g., including but not limited to scenarios with large bandwidth) without increasing complexity.
[0104] 4. The scheme of this disclosure adopts a structure in which nonlinear basis parameters and weights are generated separately, so that the generation of weights does not depend on the target input signal, which accelerates the speed of feedback optimization and reduces the complexity of engineering deployment.
[0105] 5. Compared with the traditional polynomial expansion model, the predistortion signal generation device involved in the embodiments of this disclosure has a stronger ability to extract the behavioral features of the target signal (e.g., power amplifier), which can ensure that the target signal after predistortion processing has higher linearity and better ACPR (adjacent channel power ratio) performance.
[0106] This disclosure also provides a target signal generation circuit 2, such as... Figure 2 , Figure 3 As shown, it includes the aforementioned predistortion signal generation device 1.
[0107] In this embodiment of the disclosure, the target signal generation circuit 2 may further include: a signal preprocessing module 21, used to preprocess the input target input signal.
[0108] The input terminal of the signal preprocessing module 21 is used to receive the target input signal, and the output terminal is connected to the input terminal of the nonlinear basis generator 111 in the predistortion signal generation device 1. The output terminal of the nonlinear basis generator 111 is connected to the input terminal of the predistortion signal generator 112.
[0109] In this embodiment of the disclosure, the target signal generation circuit 2 may further include: a digital-to-analog converter (DAC) circuit, the input of which is connected to the output of the predistortion signal generator 112 in the predistortion signal generation device 1, and the digital signal of the predistortion signal is converted into an analog signal through the digital-to-analog converter (DAC) module.
[0110] In this embodiment of the disclosure, the target signal generation circuit 2 may further include a target signal generation device 22, such as a power amplifier PA. The input terminal of the target signal generation device 22 is connected to the output terminal of the digital-to-analog converter circuit DAC, and the output terminal of the target signal generation device 22 is used to output a target signal, such as a power amplifier signal.
[0111] This disclosure also provides a method for generating a predistortion signal, such as... Figure 4 As shown, steps S11-S12 are included:
[0112] S11. Obtain the feedback signal of the target signal generation circuit, optimize the parameters of the preset predistortion signal parameter generation model based on the feedback signal, and obtain the predistortion signal parameters.
[0113] S12. Based on the predistortion signal generation parameters and the target input signal, generate a predistortion signal corresponding to the target input signal.
[0114] In this embodiment of the disclosure, the predistortion signal parameters may include: nonlinear basis parameters and weights required to calculate the predistortion signal; the predistortion signal parameter generation model includes: a nonlinear basis generation network and a weight generation network;
[0115] Based on the feedback signal and a pre-defined predistortion signal parameter generation model, parameter optimization is performed to obtain the predistortion signal parameters, including:
[0116] The feedback signal is input into the nonlinear basis generation network to generate nonlinear basis parameters.
[0117] The nonlinear basis parameters and the weights generated by the weight generation network are projected into a higher-order nonlinear space to obtain the network output signal.
[0118] Loss calculation is performed based on the network output signal and the target input signal corresponding to the feedback signal;
[0119] If the loss calculation result meets the preset requirements, the nonlinear basis parameters and weights are output as predistortion signal generation parameters. If the loss calculation result does not meet the preset requirements, the network parameters of the nonlinear basis generation network and the weight generation network are optimized. New nonlinear basis parameters and new weights are generated based on the optimized network parameters. A new feedback signal is generated based on the new nonlinear basis parameters, the new weights, and the target signal generation circuit. The feedback signal is updated using the new feedback signal, and the process of inputting the feedback signal into the nonlinear basis generation network to generate nonlinear basis parameters is returned to achieve iterative optimization of the predistortion signal parameter generation model.
[0120] In this embodiment of the disclosure, the feedback signal includes the target signal output signal of the target signal generation circuit.
[0121] In this embodiment of the disclosure, a predistortion signal corresponding to the target input signal is generated based on predistortion signal generation parameters and the target input signal, including:
[0122] The nonlinear basis parameters and weights are synchronously input into the predistortion signal generation module;
[0123] The predistortion signal generation module generates a nonlinear basis based on the target input signal and nonlinear basis parameters;
[0124] The predistortion signal generation module generates a predistortion signal based on a nonlinear basis and weights.
[0125] This disclosure also provides a method for generating a target signal, including:
[0126] The target signal is generated based on the predistorted signal generated by the aforementioned predistorted signal generation method.
[0127] Those skilled in the art will understand that all or some of the functional modules / units disclosed above can be implemented as software, firmware, hardware, or suitable combinations thereof.
[0128] In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be executed by several physical components working together.
[0129] Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit (CPU), digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technique for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory (FLASH) or other disk storage; read-only optical disc (CD-ROM), digital versatile disc (DVD) or other optical disc storage; magnetic cartridges, magnetic tapes, disk storage or other magnetic storage; and any other media that can be used to store desired information and can be accessed by a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.
[0130] This disclosure has disclosed exemplary embodiments, and although specific terminology has been used, it is for general illustrative purposes only and should not be construed as limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in conjunction with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in conjunction with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of this disclosure as set forth by the appended claims.
Claims
1. A predistortion signal generation apparatus comprising: The pre-distortion signal generation module and the pre-distortion signal parameter module; The pre-distortion signal parameter module comprises a pre-distortion signal parameter generation model; The pre-distortion signal parameter module is configured to obtain a feedback signal of the target signal generation circuit, perform parameter optimization on the pre-distortion signal parameter generation model based on the feedback signal, and obtain pre-distortion signal parameters; The pre-distortion signal generation module is configured to generate a pre-distortion signal corresponding to the target input signal based on the pre-distortion signal parameters and the target input signal.
2. The predistortion signal generation device of claim 1, wherein The pre-distortion signal parameters comprise nonlinear basis parameters and weights required for calculating the pre-distortion signal, and the pre-distortion signal parameter generation model comprises a nonlinear basis generation network and a weight generation network. The nonlinear basis generation network is configured to perform dimension reduction screening on a basis function set contained in the nonlinear basis generation network itself based on the feedback signal, optimize nonlinear basis generation network parameters corresponding to the nonlinear basis generation network, and extract the optimized nonlinear basis generation network parameters as the nonlinear basis parameters. The weight generation network is configured to generate the weights.
3. The predistortion signal generation device of claim 2, wherein The nonlinear basis generation network is a fully connected layer network.
4. The predistortion signal generation device of claim 2, wherein The pre-distortion signal parameter module further comprises a projection layer. The projection layer is configured to linearly combine the nonlinear basis parameters and the weights to obtain a network output signal.
5. The predistortion signal generation device of claim 4, wherein The pre-distortion signal parameter module further comprises a network optimizer. The network optimizer is configured to perform loss calculation based on the network output signal and a target input signal corresponding to the feedback signal, generate a parameter optimization instruction in a case where a loss calculation result does not meet a preset requirement, and input the nonlinear basis generation network and the weight generation network. The nonlinear basis generation network and the weight generation network are configured to iteratively optimize network parameters of the respective networks based on the parameter optimization instruction until the loss calculation result meets the preset requirement; wherein the network parameters corresponding to the nonlinear basis generation network are the nonlinear basis generation network parameters, and the network parameters corresponding to the weight generation network are weight generation network parameters.
6. The predistortion signal generation device of claim 5, wherein The nonlinear basis parameters comprise first nonlinear basis parameters, and the weights comprise first weights. The nonlinear basis generation network and the weight generation network iteratively optimize the network parameters of the respective networks based on the parameter optimization instruction, comprising: The nonlinear basis generation network is configured to optimize the nonlinear basis generation network parameters based on the parameter optimization instruction, obtain the first nonlinear basis parameters based on the optimized nonlinear basis generation network parameters, and input the pre-distortion signal generation module to enable the pre-distortion signal generation module to generate first nonlinear basis based on the first nonlinear basis parameters and the target input signal. The weight generation network is configured to optimize the weight generation network parameter based on the parameter optimization instruction, obtain the first weight based on the optimized weight generation network parameter, and enable the pre-distortion signal generation module to generate a first pre-distortion signal based on the first nonlinear basis and the first weight, and enable the target signal generation circuit to generate a target signal based on the first pre-distortion signal as a new feedback signal for next optimization of the network parameter of the nonlinear basis generation network and the weight generation network.
7. The predistortion signal generation device of claim 4, wherein The pre-distortion signal parameter module further comprises a network optimizer, the nonlinear basis parameter comprises a second nonlinear basis parameter, and the weight comprises a second weight. The network optimizer is configured to perform loss calculation based on the network output signal and a target input signal corresponding to the feedback signal, and generate an optimization completion instruction and input the nonlinear basis generation network and the weight generation network when the loss calculation result meets a preset requirement. The nonlinear basis generation network is further configured to save an optimized nonlinear basis generation network parameter based on the optimization completion instruction, generate the second nonlinear basis parameter based on the optimized nonlinear basis generation network parameter, and send the second nonlinear basis parameter to the pre-distortion signal generation module. The weight generation network is configured to save an optimized weight generation network parameter based on the optimization completion instruction, generate the second weight based on the optimized weight generation network parameter, and send the second weight to the pre-distortion signal generation module.
8. The pre-distortion signal generation apparatus according to claim 7, wherein The nonlinear basis generation network and the weight generation network are further configured to synchronously send the second nonlinear basis parameter and the second weight to the pre-distortion signal generation module.
9. The predistortion signal generation device according to claim 6 or 7, wherein The pre-distortion signal generation module comprises a nonlinear basis generator and a pre-distortion signal generator. The nonlinear basis generation network is configured to input the nonlinear basis parameter to the nonlinear basis generator. The weight generation network is configured to synchronously input the weight to the pre-distortion signal generator.
10. The pre-distortion signal generation apparatus of any of claims 6-8, wherein, The pre-distortion signal generation module comprises a nonlinear basis generator and a pre-distortion signal generator. The nonlinear basis generator is configured to generate a nonlinear basis based on the target input signal and the nonlinear basis parameter. The pre-distortion signal generator is configured to generate the pre-distortion signal based on the nonlinear basis and the weight.
11. A target signal generation circuit comprising the pre-distortion signal generation apparatus according to any one of claims 1-10.
12. A pre-distortion signal generation method, comprising: obtaining a feedback signal of a target signal generation circuit, performing parameter optimization on a preset pre-distortion signal parameter generation model based on the feedback signal, and obtaining a pre-distortion signal parameter; generating a pre-distortion signal corresponding to a target input signal based on the pre-distortion signal generation parameter and the target input signal.
13. The pre-distortion signal generation method of claim 12, wherein, The pre-distortion signal parameters include non-linear base parameters and weights required for calculating the pre-distortion signal; and the pre-distortion signal parameter generation model includes a non-linear base generation network and a weight generation network. The parameter optimization based on the feedback signal and the preset pre-distortion signal parameter generation model obtains pre-distortion signal parameters, including: The feedback signal is input into the non-linear base generation network to generate non-linear base parameters; The non-linear base parameters and the weights generated by the weight generation network are linearly combined to obtain a network output signal; Loss calculation is performed based on the network output signal and a target input signal corresponding to the feedback signal; In a case where the loss calculation result meets a preset requirement, the non-linear base parameters and the weights are output as the pre-distortion signal generation parameters; in a case where the loss calculation result does not meet the preset requirement, network parameters of the non-linear base generation network and the weight generation network are optimized, new non-linear base parameters and new weights are generated based on the optimized network parameters, a new feedback signal is generated based on the new non-linear base parameters, the new weights and the target signal generation circuit, the feedback signal is updated with the new feedback signal, and the step of inputting the feedback signal into the non-linear base generation network to generate non-linear base parameters is returned, so as to realize parameter iterative optimization of the pre-distortion signal parameter generation model.
14. The pre-distortion signal generation method of claim 13, wherein, The pre-distortion signal generation parameters and the target input signal are used to generate a pre-distortion signal corresponding to the target input signal, including: The non-linear base parameters and the weights are synchronously input into a pre-distortion signal generation module; The pre-distortion signal generation module generates a non-linear base based on the target input signal and the non-linear base parameters; The pre-distortion signal generation module generates the pre-distortion signal based on the non-linear base and the weights.
15. A target signal generation method comprising: The pre-distortion signal generated by the pre-distortion signal generation method according to any one of claims 12-14 is used to generate a target signal.