A nonlinear correction method for millimeter wave integrated circuit power amplifier

Abstract

The present invention provides a nonlinear correction method for a millimeter wave integrated circuit power amplifier, which belongs to the technical field of millimeter wave communication and signal processing, and specifically comprises: constructing a mathematical mechanism model based on a Wiener model, and constructing a mathematical mechanism digital predistorter based on the mathematical mechanism model; constructing an artificial intelligence model based on a WOA-LSTM algorithm, wherein the WOA algorithm is used to optimize the initial value of the LSTM algorithm, and constructing an artificial intelligence digital predistorter based on the artificial intelligence model; constructing a digital predistorter based on the mathematical mechanism digital predistorter and the artificial intelligence digital predistorter, and performing nonlinear correction on the power amplifier, thereby combining the two, while improving the overall accuracy, further improving the overall efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of millimeter wave communication and signal processing, and in particular relates to a nonlinear correction method for a millimeter wave integrated circuit power amplifier. Background Art

[0002] At present, the Ministry of Industry and Information Technology has also opened the millimeter wave frequency band for experimental testing. Its spectrum resources are more abundant and can meet the requirements of ultra-large capacity communication bandwidth. In the ultra-high frequency band of millimeter waves, although there are many available spectrum resources, the bandwidth of multiple frequency bands will use more than 500MHz. The use of this ultra-wideband will lead to a sharp increase in the difficulty of hardware equipment design. As one of the important components in the RF transceiver system, the power amplifier has the following indicators: bandwidth, efficiency, linearity, etc. At present, due to the very fast transmission rate of 5G, the bandwidth requirement is also very high. In addition to ensuring a large bandwidth, the broadband power amplifier must also ensure that the output power is large enough, the linearity of the power amplifier is good enough, and the efficiency of the power amplifier is high, which is very difficult for the power amplifier. Among them, the output power of the power amplifier represents the coverage area of ​​the output signal. The operational amplifier working in the compression zone will bring linearity problems, that is, nonlinear distortion. The main effects of distortion can be divided into in-band and out-of-band distortion. In-band distortion will cause EVM to deteriorate, so that the bit error rate will increase during the signal demodulation process. Out-of-band distortion can cause ACPR to deteriorate, mainly manifested as spectrum expansion, affecting communications in out-of-band frequency bands.

[0003] At present, the linear correction method for operational amplifiers is generally achieved through negative feedback and digital pre-simulation technology. In the master's thesis "Research and Implementation of Power Amplifier Pre-distortion Technology Based on BP Neural Network", the author Feng Ajing analyzed the nonlinear characteristics of power amplifiers, discussed the modeling accuracy and complexity of several existing power amplifier models, and pointed out the limitations of each model. At the same time, a BP neural network model was proposed, which has great advantages in modeling accuracy. The model was used to model a group of power amplifier signals with strong nonlinearity. The simulation in MATLAB found that its NMSE was reduced by more than 10dB compared with the memory polynomial model. However, once the initial value is not selected accurately, its processing speed is slow and cannot meet the needs of nonlinear correction. The accuracy of the mathematical model is low. How to combine the two to improve the overall accuracy and further improve the overall efficiency.

[0004] In view of the above technical problems, the present invention provides a nonlinear correction method for a millimeter wave integrated circuit power amplifier. Summary of the invention

[0005] To achieve the purpose of the present invention, the present invention adopts the following technical solutions:

[0006] According to one aspect of the present invention, a nonlinear correction method for a millimeter wave integrated circuit power amplifier is provided.

[0007] A nonlinear correction method for a millimeter wave integrated circuit power amplifier, characterized by comprising:

[0008] S1 constructs a mathematical mechanism model based on the Wiener model, and constructs a mathematical mechanism digital predistorter based on the mathematical mechanism model;

[0009] S2 constructs an artificial intelligence model based on the WOA-LSTM algorithm, wherein the WOA algorithm is used to optimize the initial value of the LSTM algorithm, and an artificial intelligence digital predistorter is constructed based on the artificial intelligence model;

[0010] S3 constructs a digital predistorter based on the mathematical mechanism digital predistorter and the artificial intelligence digital predistorter to perform nonlinear correction on the power amplifier.

[0011] A mathematical mechanism model is constructed based on the Wiener model to establish a mathematical mechanism digital pre-distorter, and an artificial intelligence model is constructed based on the WOA-LSTM algorithm, wherein the WOA algorithm is used to optimize the initial value of the LSTM algorithm, and an artificial intelligence digital pre-distorter is constructed based on the artificial intelligence model. By establishing the digital pre-distorter, compensation for the power amplifier is achieved, and correction of the power amplifier output is achieved, thereby solving the problem that once the initial value selection is not accurate enough, the processing speed is slow and cannot meet the needs of nonlinear correction. The mathematical model has low accuracy. Combining the two can further improve the overall efficiency while improving the overall accuracy.

[0012] By constructing an artificial intelligence model based on the WOA-LSTM algorithm, wherein the WOA algorithm is used to optimize the initial value of the LSTM algorithm, and an artificial intelligence digital predistorter is constructed based on the artificial intelligence model, the optimization of the initial value is realized, and the problem that once the initial value selection is not accurate enough, the processing speed is slow and cannot meet the demand for nonlinear correction is solved, and the efficiency of the artificial intelligence digital predistorter is further improved. By adopting the LSTM algorithm, time series data can be better processed, and the problem of inaccurate prediction accuracy of the artificial intelligence model due to memory effect can be better solved, and the demand for nonlinear correction is met. A digital predistorter is constructed based on the mathematical mechanism digital predistorter and the artificial intelligence digital predistorter, and nonlinear correction is performed on the power amplifier, thereby realizing the combination of the mathematical model method and the artificial intelligence method, further improving the overall accuracy, and solving the problem of low accuracy of a single method, so that the final prediction efficiency and accuracy are improved to a certain extent.

[0013] A further technical solution is that the Wiener model includes a linear time-invariant system and a memoryless nonlinear system.

[0014] A further technical solution is that the transfer function of the linear time-invariant system is:

[0015]

[0016] Where d(n) is the nth output signal of the linear time-invariant system, a m is the constant of the input of the mth memory depth, x(nm) is the input of the mth memory depth of the linear time-invariant system, and M is the total memory depth of the linear time-invariant system.

[0017] A further technical solution is that the transfer function of the memoryless nonlinear system is:

[0018]

[0019] Where y(n) is the nth output signal of the memoryless nonlinear system, b k is the constant of the output of the linear time-invariant system, t is the compensation constant of the output of the linear time-invariant system, and k is the order of the memoryless nonlinear system.

[0020] A further technical solution is that the specific steps of constructing the artificial intelligence digital predistorter are:

[0021] S21 extracts a real-time input signal and transmits the real-time input signal to an artificial intelligence digital predistorter based on a WOA-LSTM algorithm;

[0022] S22 obtains an output signal based on the artificial intelligence digital predistorter.

[0023] A further technical solution is that the WOA-LSTM algorithm uses the WOA algorithm to optimize the number of hidden layers of the LSTM algorithm.

[0024] By using the WOA algorithm to optimize the number of hidden layers of the LSTM algorithm, the overall prediction speed is further improved. At the same time, the problem that the original prediction algorithm cannot handle the timing data of the operational amplifier well is solved, thereby further improving the overall efficiency and accuracy to a certain extent.

[0025] A further technical solution is that the mathematical mechanism digital pre-distorter, artificial intelligence digital pre-distorter and digital pre-distorter all adopt a direct learning structure or an indirect learning structure.

[0026] A further technical solution is that the output signal of the digital predistorter is:

[0027] y 2n =k1y 1n +k2y n

[0028] where y 2n ,y 1n ,y n They are respectively the output signal of the digital pre-distorter, the output signal of the artificial intelligence digital pre-distorter, and the output signal of the mathematical mechanism digital pre-distorter. k1 and k2 are respectively the constants of the output signal of the artificial intelligence digital pre-distorter and the output signal of the mathematical mechanism digital pre-distorter, wherein k1+k2=1.

[0029] A further technical solution is that when the change in the input signal of the digital predistorter is greater than a first signal threshold and the duration is greater than a first time threshold, the output signal of the artificial intelligence digital predistorter is updated; when the change in the input signal of the digital predistorter is greater than a third signal threshold, the output signal of the artificial intelligence digital predistorter is updated; when the duration of the input signal of the digital predistorter is greater than a second time threshold, the output signal of the artificial intelligence digital predistorter is updated; and the mathematical mechanism digital predistorter is updated in real time.

[0030] When the change of the input signal of the digital predistorter is greater than the first signal threshold and the duration is greater than the first time threshold, the input signal at this time has undergone a mutation and has reached the specified update time, and the output signal of the artificial intelligence digital predistorter is updated, thereby further improving the overall efficiency while ensuring better output. When the change of the input signal of the digital predistorter is greater than the third signal threshold, the signal mutation at this time is large, and the output signal of the artificial intelligence digital predistorter is updated, further improving the overall accuracy. When the duration of the input signal of the digital predistorter is greater than the second time threshold, the output signal of the artificial intelligence digital predistorter is updated. At this time, the signal input has lasted for a period of time and needs to be updated so that its output can maintain a high accuracy.

[0031] A further technical solution is that the first signal threshold is smaller than the third signal threshold, and the first time threshold is smaller than the second time threshold. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The above and other features and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.

[0033] Figure 1 This is a flow chart of a nonlinear correction method for a millimeter wave integrated circuit power amplifier according to Example 1.

[0034] Figure 2 It is a flowchart of specific steps of constructing an artificial intelligence digital predistorter according to Example 1. DETAILED DESCRIPTION

[0035] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that the present invention will be comprehensive and complete and fully convey the concepts of the example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar structures, and thus their detailed description will be omitted.

[0036] The terms "a", "an", "the", and "said" are used to indicate the presence of one or more elements / components / etc.; the terms "comprising" and "having" are used to express an open-ended inclusive meaning and mean that additional elements / components / etc. may be present in addition to the listed elements / components / etc.

[0037] At present, the Ministry of Industry and Information Technology has also opened the millimeter wave frequency band for experimental testing. Its spectrum resources are more abundant and can meet the requirements of ultra-large capacity communication bandwidth. In the ultra-high frequency band of millimeter waves, although there are many available spectrum resources, the bandwidth of multiple frequency bands will use more than 500MHz. The use of this ultra-wideband will lead to a sharp increase in the difficulty of hardware equipment design. As one of the important components in the RF transceiver system, the power amplifier has the following indicators: bandwidth, efficiency, linearity, etc. At present, due to the very fast transmission rate of 5G, the bandwidth requirement is also very high. In addition to ensuring a large bandwidth, the broadband power amplifier must also ensure that the output power is large enough, the linearity of the power amplifier is good enough, and the efficiency of the power amplifier is high, which is very difficult for the power amplifier. Among them, the output power of the power amplifier represents the coverage area of ​​the output signal. The operational amplifier working in the compression zone will bring linearity problems, that is, nonlinear distortion. The main effects of distortion can be divided into in-band and out-of-band distortion. In-band distortion will cause EVM to deteriorate, so that the bit error rate will increase during the signal demodulation process. Out-of-band distortion can cause ACPR to deteriorate, mainly manifested as spectrum expansion, affecting communications in out-of-band frequency bands.

[0038] At present, the linear correction method for operational amplifiers is generally achieved through negative feedback and digital pre-simulation technology. In the master's thesis "Research and Implementation of Power Amplifier Pre-distortion Technology Based on BP Neural Network", the author Feng Ajing analyzed the nonlinear characteristics of power amplifiers, discussed the modeling accuracy and complexity of several existing power amplifier models, and pointed out the limitations of each model. At the same time, a BP neural network model was proposed, which has great advantages in modeling accuracy. The model was used to model a group of power amplifier signals with strong nonlinearity. The simulation in MATLAB found that its NMSE was reduced by more than 10dB compared with the memory polynomial model. However, once the initial value is not selected accurately, its processing speed is slow and cannot meet the needs of nonlinear correction. The accuracy of the mathematical model is low. How to combine the two to improve the overall accuracy and further improve the overall efficiency.

[0039] Example 1

[0040] To solve the above problems, according to one aspect of the present invention, Figure 1 As shown, a nonlinear correction method for a millimeter wave integrated circuit power amplifier is provided.

[0041] A nonlinear correction method for a millimeter wave integrated circuit power amplifier, characterized by comprising:

[0042] S1 constructs a mathematical mechanism model based on the Wiener model, and constructs a mathematical mechanism digital predistorter based on the mathematical mechanism model;

[0043] S2 constructs an artificial intelligence model based on the WOA-LSTM algorithm, wherein the WOA algorithm is used to optimize the initial value of the LSTM algorithm, and an artificial intelligence digital predistorter is constructed based on the artificial intelligence model;

[0044] S3 constructs a digital predistorter based on the mathematical mechanism digital predistorter and the artificial intelligence digital predistorter to perform nonlinear correction on the power amplifier.

[0045] For example, if the input signal of the mathematical mechanism digital pre-simulator and the artificial intelligence digital pre-distorter is x, then the output signal of the mathematical mechanism digital pre-simulator is yn, and the output signal of the artificial intelligence digital pre-distorter is yn+1. At this time, the digital pre-distorter can be written as y2n=0.5yn+0.5yn+1.

[0046] A mathematical mechanism model is constructed based on the Wiener model to establish a mathematical mechanism digital pre-distorter, and an artificial intelligence model is constructed based on the WOA-LSTM algorithm, wherein the WOA algorithm is used to optimize the initial value of the LSTM algorithm, and an artificial intelligence digital pre-distorter is constructed based on the artificial intelligence model. By establishing the digital pre-distorter, compensation for the power amplifier is achieved, and correction of the power amplifier output is achieved, thereby solving the problem that once the initial value selection is not accurate enough, the processing speed is slow and cannot meet the needs of nonlinear correction. The mathematical model has low accuracy. Combining the two can further improve the overall efficiency while improving the overall accuracy.

[0047] By constructing an artificial intelligence model based on the WOA-LSTM algorithm, wherein the WOA algorithm is used to optimize the initial value of the LSTM algorithm, and an artificial intelligence digital predistorter is constructed based on the artificial intelligence model, the optimization of the initial value is realized, and the problem that once the initial value selection is not accurate enough, the processing speed is slow and cannot meet the demand for nonlinear correction is solved, and the efficiency of the artificial intelligence digital predistorter is further improved. By adopting the LSTM algorithm, time series data can be better processed, and the problem of inaccurate prediction accuracy of the artificial intelligence model due to memory effect can be better solved, and the demand for nonlinear correction is met. A digital predistorter is constructed based on the mathematical mechanism digital predistorter and the artificial intelligence digital predistorter, and nonlinear correction is performed on the power amplifier, thereby realizing the combination of the mathematical model method and the artificial intelligence method, further improving the overall accuracy, and solving the problem of low accuracy of a single method, so that the final prediction efficiency and accuracy are improved to a certain extent.

[0048] In another possible embodiment, the Wiener model includes a linear time-invariant system and a memoryless nonlinear system.

[0049] In another possible embodiment, the transfer function of the linear time-invariant system is:

[0050]

[0051] Where d(n) is the nth output signal of the linear time-invariant system, a m is the constant of the input of the mth memory depth, x(nm) is the input of the mth memory depth of the linear time-invariant system, and M is the total memory depth of the linear time-invariant system.

[0052] In another possible embodiment, the transfer function of the memoryless nonlinear system is:

[0053]

[0054] Where y(n) is the nth output signal of the memoryless nonlinear system, b k is the constant of the output of the linear time-invariant system, t is the compensation constant of the output of the linear time-invariant system, and k is the order of the memoryless nonlinear system.

[0055] In another possible embodiment, Figure 2 As shown, the specific steps of constructing the artificial intelligence digital predistorter are:

[0056] S21 extracts a real-time input signal and transmits the real-time input signal to an artificial intelligence digital predistorter based on a WOA-LSTM algorithm;

[0057] S22 obtains an output signal based on the artificial intelligence digital predistorter.

[0058] In another possible embodiment, the WOA-LSTM algorithm uses the WOA algorithm to optimize the number of hidden layers of the LSTM algorithm.

[0059] By using the WOA algorithm to optimize the number of hidden layers of the LSTM algorithm, the overall prediction speed is further improved. At the same time, the problem that the original prediction algorithm cannot handle the timing data of the operational amplifier well is solved, thereby further improving the overall efficiency and accuracy to a certain extent.

[0060] In another possible embodiment, the mathematical mechanism digital pre-distorter, the artificial intelligence digital pre-distorter, and the digital pre-distorter all adopt a direct learning structure or an indirect learning structure.

[0061] In another possible embodiment, the output signal of the digital predistorter is:

[0062] y 2n =k1y 1n +k2y n

[0063] where y 2n ,y 1n ,y n They are respectively the output signal of the digital pre-distorter, the output signal of the artificial intelligence digital pre-distorter, and the output signal of the mathematical mechanism digital pre-distorter. k1 and k2 are respectively the constants of the output signal of the artificial intelligence digital pre-distorter and the output signal of the mathematical mechanism digital pre-distorter, wherein k1+k2=1.

[0064] In another possible embodiment, when the change amount of the input signal of the digital predistorter is greater than the first signal threshold and the duration is greater than the first time threshold, the output signal of the artificial intelligence digital predistorter is updated; when the change amount of the input signal of the digital predistorter is greater than the third signal threshold, the output signal of the artificial intelligence digital predistorter is updated; when the duration of the input signal of the digital predistorter is greater than the second time threshold, the output signal of the artificial intelligence digital predistorter is updated; and the mathematical mechanism digital predistorter is updated in real time.

[0065] For example, if the first signal threshold is 0.3 and the first time threshold is 5s, when the signal changes from 1.5 to 1.9, the change is greater than 0.3, and the duration is 6s, which is greater than 5s. At this time, the output signal of the artificial intelligence digital pre-distorter is updated. If the third signal threshold is 1.0, when the signal changes from 1.5 to 2.6, the change is greater than 1.0, and the output signal of the artificial intelligence digital pre-distorter needs to be updated immediately. When the duration of the signal input is greater than 1 minute, the output signal of the artificial intelligence digital pre-distorter still needs to be updated.

[0066] When the change of the input signal of the digital predistorter is greater than the first signal threshold and the time is greater than the first time threshold, the input signal at this time has undergone a mutation and has reached the specified update time, and the output signal of the artificial intelligence digital predistorter is updated, thereby further improving the overall efficiency while ensuring better output. When the change of the input signal of the digital predistorter is greater than the third signal threshold, the signal mutation at this time is large, and the output signal of the artificial intelligence digital predistorter is updated, further improving the overall accuracy. When the duration of the input signal of the digital predistorter is greater than the second time threshold, the output signal of the artificial intelligence digital predistorter is updated. At this time, the signal input has lasted for a period of time and needs to be updated so that its output can maintain a high accuracy.

[0067] In another possible embodiment, the first signal threshold is smaller than the third signal threshold, and the first time threshold is smaller than the second time threshold.

[0068] For example, the first signal threshold and the third signal threshold are 0.3 and 1.0 respectively, and the first time threshold and the second time threshold are 5s and 1 minute respectively.

[0069] In the embodiments of the present invention, the term "plurality" refers to two or more than two, unless otherwise clearly defined. The terms "installation", "connection", "fixation" and the like should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection. For those skilled in the art, the specific meanings of the above terms in the embodiments of the present invention can be understood according to the specific circumstances.

[0070] In the description of the embodiments of the present invention, it needs to be understood that the directions or positional relationships indicated by the terms "upper" and "lower" are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the embodiments of the present invention and simplifying the description, rather than indicating or implying that the device or unit referred to must have a specific direction, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the embodiments of the present invention.

[0071] In the description of this specification, the description of the terms "one embodiment", "a preferred embodiment", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0072] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for nonlinear correction of a millimeter wave integrated circuit power amplifier, characterized in that: Specifically include: S1 constructing a mathematical mechanism model based on the Wiener model, and constructing a mathematical mechanism digital predistorter based on the mathematical mechanism model; S2 constructing an artificial intelligence model based on the WOA-LSTM algorithm, wherein the WOA algorithm is used to optimize the initial value of the LSTM algorithm, and constructing an artificial intelligence digital predistorter based on the artificial intelligence model; S3 constructing a digital predistorter based on the mathematical mechanism digital predistorter and the artificial intelligence digital predistorter to perform nonlinear correction on the power amplifier; The output signal of the digital predistorter is: ； in y 2n , y 1n , y n are respectively the output signal of the digital predistorter, the output signal of the artificial intelligence digital predistorter, and the output signal of the mathematical mechanism digital predistorter, k 1 , k 2 are respectively the constants of the output signal of the artificial intelligence digital predistorter and the constants of the output signal of the mathematical mechanism digital predistorter, wherein k 1 + k 2 =1; When the change amount of the input signal of the digital predistorter is greater than the first signal threshold and the duration is greater than the first time threshold, the output signal of the artificial intelligence digital predistorter is updated; when the change amount of the input signal of the digital predistorter is greater than the third signal threshold, the output signal of the artificial intelligence digital predistorter is updated; when the duration of the input signal of the digital predistorter is greater than the second time threshold, the output signal of the artificial intelligence digital predistorter is updated; the mathematical mechanism digital predistorter is updated in real time; The first signal threshold is smaller than the third signal threshold, and the first time threshold is smaller than the second time threshold.

2. The nonlinear correction method for a millimeter wave integrated circuit power amplifier according to claim 1, characterized in that: The Wiener model includes a linear time-invariant system and a memoryless nonlinear system.

3. The nonlinear correction method for a millimeter wave integrated circuit power amplifier as claimed in claim 2, characterized in that: The transfer function of the linear time-invariant system is: ； in d(n) is the nth output signal of the linear time-invariant system, a m is the constant of the input quantity of the mth memory depth, x (nm) is the input of the mth memory depth of the linear time-invariant system, and M is the total memory depth of the linear time-invariant system.

4. The nonlinear correction method for a millimeter wave integrated circuit power amplifier as claimed in claim 3, characterized in that: The transfer function of the memoryless nonlinear system is: ； in y(n) is the nth output signal of the memoryless nonlinear system, b k is the constant of the output of the linear time-invariant system, t is the compensation constant of the output of the linear time-invariant system, and k is the order of the memoryless nonlinear system.

5. The nonlinear correction method for a millimeter wave integrated circuit power amplifier according to claim 1, characterized in that: The specific steps of constructing the artificial intelligence digital predistorter are: S21 extracts a real-time input signal, and transmits the real-time input signal to an artificial intelligence digital predistorter based on a WOA-LSTM algorithm; S22 obtains an output signal based on the artificial intelligence digital predistorter.

6. The nonlinear correction method for a millimeter wave integrated circuit power amplifier as claimed in claim 5, characterized in that: The WOA-LSTM algorithm uses the WOA algorithm to optimize the number of hidden layers of the LSTM algorithm.

7. The nonlinear correction method for a millimeter wave integrated circuit power amplifier according to claim 1, characterized in that: The mathematical mechanism digital predistorter, artificial intelligence digital predistorter and digital predistorter all adopt a direct learning structure or an indirect learning structure.

Images