An adaptive power amplifier nonlinear correction system based on independent architecture

Abstract

The present application provides an adaptive power amplifier nonlinearity correction system based on an independent architecture, which is composed of a digital processing unit, an analog-to-digital converter, and a first digital-to-analog converter, wherein the digital processing unit is configured with a gated dynamic neural network, which is composed of a gated network and a backbone network; the baseband signal in the communication device forms an initial analog signal after passing through the second digital-to-analog converter of the communication device itself, and the initial analog signal is converted into a digital signal by the analog-to-digital converter and transmitted to the digital processing unit, and data processing is performed through the gated network and the backbone network in the digital processing unit to form a corrected signal, which is then converted into a target analog signal by the first digital-to-analog converter, and the target analog signal is sent to the power amplifier of the communication device. The present application, combined with the gated dynamic neural network, can achieve efficient, flexible and adaptable dynamic configuration, and can meet the requirements of intelligent communication systems for power amplifier linearization performance.

Description

Technical Field

[0001] The present application belongs to the field of communication technology, and in particular relates to an adaptive power amplifier nonlinear correction system based on an independent architecture. Background Art

[0002] Power amplifier linearization technology is a fundamental approach to addressing nonlinearity in power amplifiers. Traditional power amplifier linearization methods, such as predistortion and power feeding, are primarily designed and optimized for fixed-frequency communication systems in the civilian sector. These traditional technologies are difficult to implement in certain application scenarios. For example, the algorithms must be integrated into the baseband unit of existing equipment, limiting system flexibility and optimization potential. Furthermore, for intelligent communication systems with dynamic transmission configurations (such as power, frequency, and modulation), traditional methods struggle to track these changes in real time and adjust linearization strategies in a timely manner, resulting in poor results and failing to meet the stringent requirements for power amplifier linearization performance in intelligent communication systems. Summary of the Invention

[0003] In view of this, the present application aims to propose an adaptive power amplifier nonlinearity correction system based on an independent architecture to solve at least one of the above problems.

[0004] To achieve the above objectives, the technical solution of this application is implemented as follows:

[0005] The present application provides an adaptive power amplifier nonlinearity correction system based on an independent architecture, characterized by:

[0006] The system is composed of a digital processing unit, an analog-to-digital converter, and a first digital-to-analog converter, wherein the digital processing unit is configured with a gated dynamic neural network, and the gated dynamic neural network is composed of a gate network and a backbone network;

[0007] The baseband signal in the communication device forms an initial analog signal after passing through the second digital-to-analog converter of the communication device itself. The initial analog signal is converted into a digital signal by the analog-to-digital converter and transmitted to the digital processing unit. The data is processed by the gating network and the backbone network in the digital processing unit to form a corrected signal, which is then converted into a target analog signal by the first digital-to-analog converter and sent to the power amplifier of the communication device.

[0008] Furthermore, the backbone network is configured to characterize and correct the nonlinear behavior of the power amplifier, and is composed of an input data layer, a filtering layer, a vector decomposition layer, a nonlinear layer, a phase recovery layer, and a synthesis layer.

[0009] Furthermore, the filtering layer is composed of n parallel FIR networks, wherein the FIR network formula is as follows:

[0010] ;

[0011] ;

[0012] Where, represents the weight parameters of each network, represents the bias parameters of each network, represents the Hadamard product, represents the filter layer gating signal output by the gating network, Represents the output data of the input data layer, 、 Represents the output data of the filter layer of Quantity.

[0013] Furthermore, the vector decomposition layer performs vector decomposition on the output of the filtering layer to obtain amplitude and phase components. The specific formula is as follows:

[0014] ;

[0015] ;

[0016] Where, 、 Represents the output data of the filter layer of Quantity, represents the phase component, Indicates amplitude.

[0017] Furthermore, the nonlinear layer is configured to correct the amplitude obtained after decomposition by the vector decomposition layer. The specific formula is as follows:

[0018] ;

[0019] Where, represents the weight parameters of the network, represents the bias parameter of the network, represents the activation function, represents the Hadamard product, represents the nonlinear layer gating signal output by the gating network, Indicates the amplitude after correction.

[0020] Furthermore, the phase recovery layer is configured to recombine the corrected amplitude and the original phase into a complex signal, and the specific formula is as follows:

[0021] ;

[0022] in, , .

[0023] Furthermore, the synthesis layer is configured to weightedly fuse the correction results of each parallel branch to output m Signal data, the specific formula is as follows:

[0024] ;

[0025] Where, represents the weight parameters of the network, represents the bias parameter of the network, Represents the output data of the phase recovery layer.

[0026] Furthermore, the gating network is configured to learn patterns in how the power amplifier behavior changes with transmission configuration and dynamically adjust weights of the backbone network during power amplifier linearization correction processing, wherein the gating network consists of an input data layer, a feature extraction layer, and a gating weight generation layer.

[0027] Furthermore, the feature extraction layer is configured to extract data features of the input data layer. The specific formula is as follows:

[0028] ;

[0029] Among them, the input data , , ;

[0030] Where, Indicates the maximum value of power, Indicates the currently configured power value. Indicates the maximum value of frequency, Indicates the currently configured frequency, 、 represents the weight parameters of the network, 、 represents the bias parameter of the network, represents the activation function, Represents the data features extracted by the feature extraction layer.

[0031] Furthermore, the gating weight generation layer is configured to map the extracted data features into gating weights, and the specific formula is as follows:

[0032] ;

[0033] ;

[0034] Where, 、 represents the weight parameters of the network, 、 represents the bias parameter of the network, represents the Sigmod activation function, 、 represents the gating weight, Represents the data features extracted by the feature extraction layer.

[0035] Compared with the prior art, the adaptive power amplifier nonlinear correction system based on an independent architecture described in this application has the following beneficial effects:

[0036] The adaptive power amplifier nonlinearity correction system based on an independent architecture described in this application avoids the dependence of the traditional baseband-side power amplifier linearization correction algorithm on the baseband unit, and is suitable for scenarios where the baseband is difficult to customize and develop. The system is combined with a gated dynamic neural network to achieve efficient, flexible and adaptable dynamic configuration, and can meet the strict requirements of intelligent communication systems for power amplifier linearization performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0038] Figure 1 This is a schematic diagram of the radio frequency architecture of an adaptive power amplifier nonlinear correction system based on an independent architecture according to an embodiment of the present application;

[0039] Figure 2 This is a schematic diagram of the network structure of an adaptive power amplifier nonlinear correction system based on an independent architecture described in an embodiment of the present application. DETAILED DESCRIPTION

[0040] In order to make the objectives, technical solutions and advantages of this application more clear, this application is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0041] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should have the usual meanings understood by people with ordinary skills in the field to which this application belongs. The "first", "second" and similar words used in the embodiments of the present application do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0042] See also Figure 1 As shown, this embodiment provides an adaptive power amplifier nonlinear correction system based on an independent architecture. The system architecture is independent of the original communication equipment. Without changing the original communication system, it is directly connected to the original communication system through a radio frequency cable. By using analog signal input and analog signal output This approach avoids the reliance of traditional baseband-side power amplifier linearization correction algorithms on baseband units, and is suitable for scenarios where customized baseband development is difficult.

[0043] The system consists of three parts: a digital processing unit, an analog-to-digital converter (ADC), and a first digital-to-analog converter (DAC).

[0044] The baseband signal in the communication device will form the initial analog signal after passing through the second digital-to-analog converter (DAC) of the communication device itself. , the initial analog signal The analog signal is converted into a digital signal through the digital-to-analog converter of this patented system Connect to the digital unit of this system;

[0045] digital signal After being processed by the gated dynamic neural network (i.e., the pre-distortion network, which consists of a gated network and a backbone network) in the digital processing unit, the corrected signal is formed. , converted into an analog signal by the first digital-to-analog converter , and The signal is sent to the power amplifier (PA) of the communication device.

[0046] The gated dynamic neural network consists of two parts: the gating network and the backbone network. The gated dynamic neural network runs in the digital processing unit of this system and is used to correct the nonlinearity of the communication equipment PA; the backbone network can be any neural network-based power amplifier nonlinearity correction model designed for a fixed configuration. The backbone network is mainly used to characterize and correct the nonlinear behavior of the power amplifier.

[0047] The backbone network used in this system consists of six parts: input data layer, filtering layer, vector decomposition layer, nonlinear layer, phase recovery layer, and synthesis layer. The details are as follows:

[0048] 1) Input data layer: input into this system The signal needs to be organized into two-dimensional data, which consists of the I / Q components of the input data, the cubic terms of the I / Q data, the quintic terms of the I / Q data, and the septic terms of the I / Q data. The details are as follows:

[0049]

[0050] Where, 、 The data is represented by I / Q components, 、 、 is the cubic term of I / Q data, the quintic term of I / Q data, and the septic term of I / Q data. The specific calculation formulas are as follows:

[0051]

[0052] Where n is the memory depth, which indicates how many previous input data the amplifier output data is related to. It is usually selected based on the actual amplifier situation, and m=3, 5, or 7.

[0053] 2) Filter layer: The filter layer consists of n parallel FIR networks. The number of FIR networks n is consistent with the memory depth of the input data. It is used to model the memory effect of the power amplifier. The output of the filter layer is .

[0054]

[0055] Where, 、 Represents the output data of the filter layer of Quantity, represents the phase component, Indicates amplitude.

[0056] The calculation formula of the FIR network is as follows:

[0057]

[0058] Where, represents the weight parameters of each network, represents the bias parameters of each network, represents the Hadamard product, Represents the gating signal of the filter layer output by the gating network.

[0059] 3) Vector decomposition layer: Decompose the output of the FIR network into vectors to obtain the amplitude and phase The specific calculation formula is as follows:

[0060]

[0061]

[0062] 4) Nonlinear layer: amplitude after decomposition It is fed into the nonlinear layer to model and compensate for the nonlinear behavior of the power amplifier. The output of the nonlinear layer is The calculation formula of the nonlinear layer is as follows:

[0063]

[0064] Where, represents the weight parameters of the network, represents the bias parameter of the network, represents the activation function, represents the Hadamard product, The gating signal of the nonlinear layer representing the output of the gating network.

[0065] 5) Phase recovery layer: convert the corrected amplitude With the original phase Recombined into complex signal .

[0066]

[0067] Where, 、 The specific calculation formula is as follows:

[0068]

[0069]

[0070] 6) Synthesis layer: The synthesis layer weights and fuses the correction results of each parallel branch to enhance the nonlinear compensation capability and adapt to complex PA characteristics. The synthesis layer outputs m I / Q signal data. The specific calculation formula is as follows:

[0071]

[0072] Where, represents the weight parameters of the network, represents the bias parameter of the network, Indicates the output of m I / Q signal data.

[0073] The gating network used in this system is used to learn the pattern of PA behavior as the transmission configuration (power, frequency, etc.) changes, and dynamically adjusts the weights of the backbone network during the PA linearization correction process, allowing the system to quickly adapt to different transmission configurations.

[0074] The gated network consists of three parts: the input data layer, the feature extraction layer, and the gated weight generation layer. The details are as follows:

[0075] 1) Input data layer: input data Including normalized power ( ), frequency ( ) and other information related to transmission configuration.

[0076]

[0077] in, , ;

[0078] Indicates the maximum value of power, Indicates the currently configured power value; Indicates the maximum value of frequency, Indicates the currently configured frequency.

[0079] 2) Feature extraction layer: The feature extraction layer is used to extract the features of the input layer data. The specific calculation formula is as follows:

[0080]

[0081] in: 、 represents the weight parameters of the network, 、 represents the bias parameter of the network, Represents the activation function.

[0082] 3) Gating weight generation layer: The gating weight generation layer maps the extracted features into gating weights and The specific calculation formula is as follows:

[0083]

[0084]

[0085] Where, 、 represents the weight parameters of the network, 、 represents the bias parameter of the network, Represents the Sigmod activation function.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

[0087] The embodiments of the present application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of this application.

Claims

1. An adaptive power amplifier nonlinearity correction system based on an independent architecture, characterized by: The system is composed of a digital processing unit, an analog-to-digital converter, and a first digital-to-analog converter, wherein the digital processing unit is configured with a gated dynamic neural network, and the gated dynamic neural network is composed of a gate network and a backbone network; The baseband signal in the communication device is converted into an initial analog signal after passing through the second digital-to-analog converter of the communication device itself. The initial analog signal is converted into a digital signal by the analog-to-digital converter and transmitted to the digital processing unit. The digital processing unit performs data processing on the gating network and the backbone network to form a corrected signal. The signal is then converted into a target analog signal by the first digital-to-analog converter, and the target analog signal is sent to the power amplifier of the communication device. The backbone network is configured to characterize and correct the nonlinear behavior of the power amplifier, and is composed of an input data layer, a filtering layer, a vector decomposition layer, a nonlinear layer, a phase recovery layer, and a synthesis layer; The filtering layer is composed of n parallel FIR networks. The vector decomposition layer performs vector decomposition on the output of the filtering layer to obtain amplitude and phase components. The nonlinear layer is configured to correct the amplitude obtained after decomposition by the vector decomposition layer. The phase recovery layer is configured to recombine the corrected amplitude and the original phase into a complex signal. The synthesis layer is configured to weightedly fuse the correction results of each parallel branch to output m I / Q signal data. The gating network is configured to learn the pattern of power amplifier behavior as the transmission configuration changes and dynamically adjust the weights of the backbone network during the power amplifier linearization correction process, wherein the gating network consists of an input data layer, a feature extraction layer and a gating weight generation layer.

2. The system according to claim 1, wherein: The FIR network formula is as follows: u (n) =(w Fir x (n) +b Fir )⊙g f ; Where w Fir represents the weight parameters of each network, b Fir represents the bias parameter of each network, ⊙ represents the Hadamard product, g f Represents the filter layer gating signal output by the gating network, x (n) Represents the output data of the input data layer, u nI 、u nQ Represents the filter layer output data u (n) I / Q components.

3. The system according to claim 1, wherein: The vector decomposition formula is as follows: Where u nI 、u nQ Represents the filter layer output data u (n) I / Q components, represents the phase component, m (n) Indicates amplitude.

4. The system according to claim 3, characterized in that The amplitude correction formula is as follows: nl (n) =tanh(w nl m (n) +b nl )⊙g n ; Where w nl represents the weight parameter of the network, b nl represents the bias parameter of the network, tanh represents the activation function, ⊙ represents the Hadamard product, g n Represents the nonlinear layer gating signal output by the gating network, nl (n) Indicates the amplitude after correction.

5. The system according to claim 4, characterized in that The complex signal calculation formula is as follows: in, 6. The system according to claim 1, wherein: The weighted fusion formula is as follows: y (m) =w out Z (n) +b out ; Where w out represents the weight parameter of the network, b out represents the bias parameter of the network, Z (n) Represents the output data of the phase recovery layer.

7. The system according to claim 1, wherein: The feature extraction layer is configured to extract data features of the input data layer. The specific formula is as follows: S=ReLU(w1ReLU(w2C k +b2)+b1); Among them, the input data Where p max Indicates the maximum power value, p indicates the current configured power value, and f max Represents the maximum frequency, f represents the currently configured frequency, w1 and w2 represent the weight parameters of the network, b1 and b2 represent the bias parameters of the network, ReLU represents the activation function, and S represents the data features extracted by the feature extraction layer.

8. The system according to claim 1, wherein: The gating weight generation layer is configured to map the extracted data features into gating weights, and the specific formula is as follows: g f =σ(w f S+b f ); g n =σ(w n S+b n ); Where w f 、w n represents the weight parameter of the network, b f 、b n represents the bias parameter of the network, σ represents the Sigmod activation function, g f 、g n Represents the gate weight, and S represents the data features extracted by the feature extraction layer.

Images