Digital pre-distortion parameter dynamic scheduling method based on hierarchical working condition modeling

Abstract

The application belongs to the technical field of radio frequency power amplifiers, and provides a digital pre-distortion parameter dynamic scheduling method based on hierarchical working condition modeling, comprising: S1. constructing a dominant working condition training set, collecting power amplifier input and output data; S2. training basic DPD parameters to obtain a two-dimensional dominant working condition DPD parameter set; S3. establishing a disturbance working condition training set, training disturbance working condition DPD parameters; S4. establishing a disturbance drift model; S5. detecting a current working condition to generate final DPD parameters under the current working condition; and S6. loading the final DPD parameters to a memory polynomial pre-distorter, obtaining an output signal of the pre-distorter for an original input signal, and sending the output signal of the pre-distorter into a power amplifier to compensate the power amplifier. The application classifies five-dimensional working conditions into two-dimensional dominant working conditions and three-dimensional disturbance working conditions, takes input power and power supply voltage as the two-dimensional dominant working conditions, and only needs to perform full combination training on the dominant working conditions, thereby greatly reducing the time and cost of offline training.

Description

Technical Field

[0001] This invention relates to the field of radio frequency power amplifier technology, and more specifically to a dynamic scheduling method for digital predistortion parameters based on hierarchical operating condition modeling. Background Technology

[0002] Digital predistortion (DPD) is a core technology for improving the linearity of RF power amplifiers, such as... Figure 1 As shown, its basic principle is to introduce a predistortion signal that is opposite to the nonlinear characteristics of the power amplifier at the input end, thereby canceling the nonlinear distortion at the output end.

[0003] Existing DPD methods typically employ models such as Memory Polynomial (MP) and Generalized Memory Polynomial (GMP), obtaining DPD parameters through offline training or online adaptive algorithms. These methods have achieved good results in continuous wave or communication signal scenarios.

[0004] However, the pulse RF power amplifier system has the following characteristics: (1) The power amplifier operates in pulse mode and has obvious heat accumulation effect; (2) The system needs to adapt to a wide range of input power, supply voltage, duty cycle, pulse width and temperature changes; (3) The nonlinear characteristics of the power amplifier differ significantly under different operating conditions.

[0005] The existing technology has the following shortcomings: (1) The model has a single dimension and cannot adapt to changes in multiple working conditions; (2) The environmental adaptability is insufficient and lacks thermal effect compensation; (3) The training complexity is high and it is not conducive to rapid deployment; (4) The pulse adaptability is poor and the waveform distortion suppression is insufficient.

[0006] Therefore, there is an urgent need for a DPD parameter modeling and dynamic scheduling method that has low training complexity, strong adaptability to multiple working conditions, and is easy to implement in hardware. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a dynamic scheduling method for digital predistortion parameters based on hierarchical working condition modeling, which solves the problem of exponential growth in training complexity caused by traditional five-dimensional full combination training methods, while also being able to adapt to dynamic nonlinear drift in pulse radio frequency systems.

[0008] The present invention provides a dynamic scheduling method for digital predistortion parameters based on hierarchical working condition modeling, comprising: S1. Construct a training set based on the dominant operating conditions and collect power amplifier input and output data; a two-dimensional training network for input power and supply voltage; S2. Train the basic DPD parameters to obtain the two-dimensional dominant working condition DPD parameter set. ; S3. Establish a training set for disturbance conditions and train the DPD parameters for disturbance conditions; S4. Establish a disturbance drift model; the disturbance drift model is used to characterize the relationship between the drift amount and the disturbance eigenvector, which includes the duty cycle, pulse width and operating temperature; S5. Detect the current operating condition and generate the final DPD parameters under the current operating condition; S6. Load the final DPD parameters under the current operating condition into the memory polynomial predistorter, obtain the output signal of the predistorter for the original input signal, and send the output signal of the predistorter into the power amplifier to achieve compensation for the nonlinearity of the power amplifier.

[0009] Optionally, S1 includes: A two-dimensional training network is established based on the input power P and the supply voltage V; whereby... , The length of the power signal. The length of the voltage signal; within the input power range Power supply voltage range The internal steps are respectively set according to the preset steps. , Obtained by discretization; At each dominant operating point Next, construct training signals. And simultaneously acquire output signals ;include: By adjusting The amplitude is adjusted so that the average output power of the power amplifier under the excitation of the training signal reaches the target value. ; Maintain the amplitude, and make The instantaneous amplitude fluctuates randomly around the target value, covering the amplitude range of the power amplifier from the linear region to the compression region; Adjust the power supply voltage of the power amplifier to ; The constructed training signal Input power amplifier, synchronously acquire output signal .

[0010] Optionally, S2 includes: Normalized power amplifier output As model input, with power amplifier input As the model output, construct the inverse model of the power amplifier; Define the basis functions and transform the inverse model of the power amplifier into a matrix relationship of the input signal with respect to the basis functions; Each dominant operating condition point is obtained using the least squares method. The DPD parameters are used to obtain the two-dimensional dominant working condition DPD parameter set. .

[0011] Optionally, S3 includes: By fixing the dominant operating point, obtaining different combinations of disturbance operating conditions, and training to obtain the DPD parameter vector; Obtain the drift of the disturbed operating point relative to the basic DPD parameters, and obtain the drift sample. .

[0012] Optionally, the fixed dominant operating point is used to obtain different combinations of disturbance operating conditions, and the DPD parameter vector is trained, including: At the dominant operating point Under these conditions, the duty cycle is changed sequentially. Pulse width and operating temperature To obtain different combinations of disturbance conditions ;in, , ; For each disturbance condition, the DPD parameters are obtained by training using the least squares method.

[0013] Optionally, S4 includes: Constructing perturbation eigenvectors And combine them into a feature matrix ; Perturbation eigenvector , The duty cycle, pulse width, and operating temperature are for the m-th disturbance condition point; The drift sample Combined into a drift matrix Among them, the drift matrix Characteristic matrix and weight matrix The product; The weight matrix is ​​solved using the least squares method. The perturbation drift model is the weight matrix. .

[0014] Optionally, S5 includes: Based on the current power and current voltage Calculate the basic DPD parameters under the current dominant operating condition. ; Based on the perturbation drift model, using the current duty cycle Pulse width Operating temperature Calculate the current disturbance drift amount ; The base DPD parameters are superimposed with the current disturbance drift to obtain the final DPD parameters under the current operating condition.

[0015] Optionally, the statement based on the current power and current voltage Calculate the basic DPD parameters under the current dominant operating condition. ,include: Based on the current power and current voltage In the two-dimensional dominant working condition parameter set Bilinear interpolation is performed to determine the current power. and current voltage Adjacent discrete points at both ends; Calculate the position coefficients of the current power and current voltage relative to adjacent discrete points; Determine the basic DPD parameters under the current dominant operating condition based on the position coefficient. .

[0016] Optionally, the location coefficient is as follows: , ; The basic DPD parameters are: , correspond Basic DPD parameters, correspond Basic DPD parameters, correspond Basic DPD parameters, correspond Basic DPD parameters; , , , From the two-dimensional dominant operating condition parameter set Read from the middle.

[0017] By adopting the above technical solution, this application has the following beneficial effects: This invention classifies the five-dimensional operating conditions into two-dimensional dominant operating conditions and three-dimensional disturbance operating conditions. Input power and supply voltage are used as the two-dimensional dominant operating conditions. Only the active operating conditions need to be fully combined for training, while the disturbance operating conditions are fitted by a low-dimensional drift model. Compared with the traditional five-dimensional full combination training method, the amount of training data can be reduced by more than 90%, which greatly reduces the time and cost of offline training. This invention establishes a basic DPD parameter table covering a wide power range and power supply fluctuations based on two-dimensional dominant operating conditions; it compensates for the influence of pulse structure and thermal effects on power amplifier linearity through a disturbance drift model of duty cycle, pulse width and operating temperature; during online operation, it quickly generates DPD parameters through two-dimensional difference and disturbance calculation, effectively adapting to any combination of five-dimensional operating conditions. To address the typical distortion problem in pulsed radio frequency systems, this invention uses layered modeling to accurately match the nonlinear characteristics of the power amplifier under different operating conditions, which can effectively suppress waveform distortion such as pulse platform collapse, overshoot, and tailing, and significantly improve the consistency of the output waveform. The online phase only requires two-dimensional bilinear interpolation, six-term weighted summation, and parameter superposition calculation, without the need for complex matrix operations or iterative updates; the computational load fully meets the real-time implementation requirements of FPGA, DSP, and embedded systems, and parameter switching can be completed within the pulse interval without affecting the pulse waveform quality; The dominant operating condition model and the disturbance drift model are independent of each other and can be updated and optimized separately. When the system operating condition range expands or the power amplifier characteristics change, only the corresponding model needs to be updated locally, without the need for retraining across all operating conditions. In addition, this method does not rely on a high-speed feedback link, reducing system hardware costs and demonstrating good engineering practicality. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0019] Figure 1 A schematic diagram of a digital predistorter is shown; Figure 2 One of the flowcharts of the dynamic scheduling method for digital predistortion parameters based on hierarchical working condition modeling provided by an embodiment of the present invention is shown; Figure 3 The flowcharts for S2-S4 provided in the embodiments of the present invention are shown; Figure 4 The second flowchart of the dynamic scheduling method for digital predistortion parameters based on hierarchical working condition modeling provided by the present invention is shown. Figure 5 A waveform diagram of a device provided in the fourth embodiment of the present invention is shown. Detailed Implementation

[0020] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of the present invention and are therefore merely examples, and should not be construed as limiting the scope of protection of the present invention.

[0021] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0022] In one embodiment, such as Figure 2 As shown, a dynamic scheduling method for digital predistortion parameters based on hierarchical working condition modeling is provided, including: S1. Construct a training set based on the dominant operating conditions, and collect power amplifier input and output data; a two-dimensional training network for input power and supply voltage.

[0023] S1 includes: S11. Establish a two-dimensional training network for input power P and supply voltage V; whereby... , in, The length of the power signal. The length of the voltage signal; within the input power range Power supply voltage range The internal steps are respectively set according to the preset steps. , The input power and supply voltage are obtained through discretization. These are used as the dominant operating parameters to determine the fundamental nonlinear state of the power amplifier.

[0024] S12. At each dominant operating point Next, construct training signals. And simultaneously acquire output signals .include: S121. By adjusting The amplitude is adjusted so that the average output power of the power amplifier under the excitation of the training signal (referring to the average power during the pulse plateau period) reaches the target value. In practice, a mapping relationship between the root mean square value of the digital signal and the output power can be established in advance through offline calibration, and the required digital amplitude can be calculated accordingly.

[0025] S122. Maintain amplitude, and make The instantaneous amplitude fluctuates randomly around the target value, covering the amplitude range of the power amplifier from the linear region to the compression region, so as to fully stimulate its dynamic nonlinear characteristics.

[0026] S123. Adjust the power supply voltage of the power amplifier to .

[0027] S124. Transfer the constructed training signal Input power amplifier, synchronously acquire output signal .

[0028] Training signal A pulse waveform is used, and its timing parameters are set as follows: Pulse width τ Set to the typical value for this dominant operating point (e.g., τ =500μs); Duty cycle D: Set to the typical value for this dominant operating condition (e.g., D=50%). Signal length N: should contain at least 10 complete pulse cycles, and the total number of sampling points during the plateau period should not be less than 1024, in order to ensure the numerical stability of the least squares estimation.

[0029] S2. Train the basic DPD parameters to obtain the two-dimensional dominant working condition DPD parameter set. .

[0030] S2 includes: S21. An inverse model of the power amplifier (i.e., a DPD (Digital Pre-Distortion) model) is established using a memory polynomial model. The normalized power amplifier output... As model input, with power amplifier input As the model output, an inverse model of the power amplifier is constructed, and the input and output satisfy the following relationship; , Where G is the expected linear gain of the power amplifier, K is the nonlinear order, odd indicates that only odd numbers are used, and Q is the memory depth. The DPD coefficients to be estimated are denoted as .

[0031] S22. Define the basis functions and convert the inverse model of the power amplifier into a matrix relationship of the input signal with respect to the basis functions; Basis function sequence: ; The above relationships can be expressed in matrix form: , In the formula: , , , .

[0032] According to the least squares method, we can obtain: In the formula This represents the conjugate transpose of a matrix. This represents the DPD parameter vector obtained during training under a certain working condition.

[0033] S23. Obtain each dominant operating point based on the least squares method. The DPD parameters are used to obtain the two-dimensional dominant working condition DPD parameter set. .

[0034] For each dominant operating point Collect input and output data and construct the basis function matrix. and input signal vector The DPD parameters at this operating point are estimated using the least squares method: , Will The basic DPD parameter for this dominant operating condition is denoted as: , Indicates the dominant operating point The DPD parameter vector below is A special case. For all The two-dimensional dominant working condition DPD parameter set is obtained through sequential training. , This represents the set of all dominant operating point parameters, forming a two-dimensional lookup table.

[0035] S3. Establish a training set for disturbance conditions and train the DPD parameters for disturbance conditions.

[0036] Specifically, S3 includes: S31. After completing the basic DPD parameter training, fix the dominant operating point. Different combinations of disturbance conditions are obtained, and the DPD parameter vector is trained.

[0037] S31 includes: S311. Adjust the signal amplitude to make the average output power of the power amplifier reach the target value during the pulse plateau period. At the same time, maintain the power supply voltage at .

[0038] S312. At the dominant operating point Under these conditions, the duty cycle is changed sequentially. (e.g., 10%, 30%, 50%, 70%, 90%), pulse width (e.g., 10μs, 100μs, 500μs, 1000μs) and operating temperature (e.g., 25℃, 40℃, 55℃, 70℃, 85℃), where the temperature is stabilized to the target value by a temperature control device.

[0039] S313. Obtain different combinations of disturbance conditions And construct the corresponding training signal; where, , .

[0040] The signal amplitude setting is the same as in step S312 (making the average output power equal to...). ), but change the pulse width to The duty cycle was changed to The signal length remains consistent with step S312, and the instantaneous amplitude fluctuates randomly around the target value. After the temperature stabilizes, the training signal is input and the output signal is acquired synchronously.

[0041] S32. For each disturbance condition, the DPD parameters are obtained by training using the least squares method.

[0042] Under fixed dominant operating conditions For each combination of disturbance conditions The DPD model was trained using the least squares method to obtain the DPD parameter vector under the disturbance condition, denoted as . .

[0043] S32. Obtain the drift of the disturbance operating point relative to the basic DPD parameters, and obtain the drift sample. Calculate the drift of the disturbed operating point relative to the base DPD parameters: ; In the formula: : No. DPD parameter vectors obtained by training under each disturbance condition; Basic DPD parameters under the same dominant operating condition; : No. The parameter drift vector corresponding to each disturbance condition point.

[0044] For all Perform the above calculations at each disturbance operating point to obtain... Group drift sample .

[0045] S4. Establish a disturbance drift model; the disturbance drift model is used to characterize the relationship between the drift amount and the disturbance eigenvector, which includes the duty cycle, pulse width and operating temperature.

[0046] Optionally, S4 includes: Constructing perturbation eigenvectors And combine them into a feature matrix .

[0047] For the m-th sample, the perturbation feature vector , Let be the duty cycle, pulse width, and operating temperature of the m-th disturbance condition.

[0048] Suppose there are M training samples under disturbance conditions. Combine the feature vectors of all samples into a feature matrix: .

[0049] The drift sample Combined into a drift matrix ; ; Among them, the drift matrix Characteristic matrix and weight matrix The product of.

[0050] The perturbation drift model assumes a linear relationship between the drift amount and the perturbation eigenvector: , In the formula, Let be the perturbation weight coefficients to be determined. The above relationship can be written in matrix form: , In the formula, This is a weight matrix with a dimension of 6×L (L is the length of the DPD parameter vector), and each column corresponds to 6 weight coefficients at a DPD coefficient position.

[0051] The weight matrix is ​​solved using the least squares method. The perturbation drift model is the weight matrix. .

[0052] , In the formula: This is the transpose of the characteristic matrix; Invert a matrix.

[0053] In other implementations, the perturbation drift model may also employ nonlinear models such as polynomial regression, radial basis functions, or neural networks.

[0054] In summary, the process from S2 to S4 is as follows: Figure 3 As shown.

[0055] S5. Detect the current operating condition and generate the final DPD parameters under the current operating condition.

[0056] Specifically, S5 includes: S51. Based on the current power and current voltage Calculate the basic DPD parameters under the current dominant operating condition. S51 includes: S511. Based on the current power and current voltage In the two-dimensional dominant working condition parameter set Bilinear interpolation is performed to determine the current power. and current voltage Adjacent discrete points at both ends; S512. Calculate the position coefficients of the current power and current voltage relative to adjacent discrete points. The position coefficients are specifically: , .

[0057] S513. Determine the basic DPD parameters under the current dominant operating condition based on the position coefficient. The basic DPD parameters are: , correspond Basic DPD parameters, correspond Basic DPD parameters, correspond Basic DPD parameters, correspond The basic DPD parameters. All four parameters are derived from the two-dimensional dominant operating condition parameter set established in step S2. Read from the middle.

[0058] S52. Based on the perturbation drift model, using the current duty cycle Pulse width Operating temperature Calculate the current disturbance drift amount .

[0059] , In the formula These are the perturbation weight coefficients obtained through least squares fitting in step S4.

[0060] S53. Superimpose the basic DPD parameters with the current disturbance drift to obtain the final DPD parameters under the current operating condition.

[0061] In the formula This is the DPD parameter vector that is ultimately used in the predistorter.

[0062] S6. Load the final DPD parameters under the current operating condition into the memory polynomial predistorter, obtain the output signal of the predistorter for the original input signal, and send the output signal of the predistorter into the power amplifier to achieve compensation for the nonlinearity of the power amplifier.

[0063] The final DPD parameters generated in step S5 Loaded to the memory polynomial predistorter. For the original input signal... The output of the predistorter is: , In the formula: The original input signal when used online; The predistorted output signal is sent to the power amplifier. Final DPD parameters The corresponding coefficients in; K: Nonlinear order; Q: Memory depth.

[0064] like Figure 4 As shown, the output of the predistorter By feeding the signal into a power amplifier, nonlinearity can be compensated for, resulting in a linearized output. Furthermore, as... Figure 5 As shown, by comparing the output waveform of the method provided by the present invention with the waveform without DPD, it can be seen that the method of the present invention can effectively suppress waveform distortion such as pulse platform collapse, overshoot and tailing, and significantly improve the consistency of the output waveform.

[0065] The above embodiments are only used to provide a detailed description of the technical solutions of this application. However, the descriptions of the above embodiments are only for helping to understand the methods of the embodiments of the present invention and should not be construed as limiting the embodiments of the present invention. Variations or substitutions that can be easily conceived by those skilled in the art should be covered within the protection scope of the embodiments of the present invention.

Claims

1. A dynamic scheduling method for digital predistortion parameters based on hierarchical working condition modeling, characterized in that, include: S1. Construct a training set for the dominant operating conditions and collect power amplifier input and output data; S2. Train the basic DPD parameters to obtain the two-dimensional dominant working condition DPD parameter set. ; S3. Establish a training set for disturbance conditions and train the DPD parameters for disturbance conditions; S4. Establish a disturbance drift model; the disturbance drift model is used to characterize the relationship between the drift amount and the disturbance eigenvector, which includes the duty cycle, pulse width and operating temperature; S5. Detect the current operating condition and generate the final DPD parameters under the current operating condition; S6. Load the final DPD parameters under the current operating condition into the memory polynomial predistorter, obtain the output signal of the predistorter for the original input signal, and send the output signal of the predistorter into the power amplifier to achieve compensation for the nonlinearity of the power amplifier.

2. The method according to claim 1, characterized in that, S1 includes: A two-dimensional training network is established based on the input power P and the supply voltage V; whereby... , The length of the power signal. The length of the voltage signal; within the input power range Power supply voltage range The internal steps are respectively set according to the preset steps. , Obtained by discretization; At each dominant operating point Next, construct training signals. And simultaneously acquire output signals ;include: By adjusting The amplitude is adjusted so that the average output power of the power amplifier under the excitation of the training signal reaches the target value. ; Maintain the amplitude, and make The instantaneous amplitude fluctuates randomly around the target value, covering the amplitude range of the power amplifier from the linear region to the compression region; Adjust the power supply voltage of the power amplifier to ; The constructed training signal Input power amplifier, synchronously acquire output signal .

3. The method according to claim 2, characterized in that, S2 includes: Normalized power amplifier output As model input, with power amplifier input As the model output, construct the inverse model of the power amplifier; Define the basis functions and transform the inverse model of the power amplifier into a matrix relationship of the input signal with respect to the basis functions; Each dominant operating condition point is obtained using the least squares method. The DPD parameters are used to obtain the two-dimensional dominant working condition DPD parameter set. .

4. The method according to claim 3, characterized in that, S3 includes: By fixing the dominant operating point, obtaining different combinations of disturbance operating conditions, and training to obtain the DPD parameter vector; Obtain the drift of the disturbed operating point relative to the basic DPD parameters, and obtain the drift sample. .

5. The method according to claim 4, characterized in that, The fixed dominant operating point is used to obtain different combinations of disturbance operating conditions, and the DPD parameter vector is trained, including: At the dominant operating point Under these conditions, the duty cycle is changed sequentially. Pulse width and operating temperature To obtain different combinations of disturbance conditions ;in, , ; For each disturbance condition, the DPD parameters are obtained by training using the least squares method.

6. The method according to claim 4, characterized in that, S4 includes: Constructing perturbation eigenvectors And combine them into a feature matrix ; Perturbation eigenvector , The duty cycle, pulse width, and operating temperature are for the m-th disturbance condition point; The drift sample Combined into a drift matrix Among them, the drift matrix Characteristic matrix and weight matrix The product; The weight matrix is ​​solved using the least squares method. The perturbation drift model is the weight matrix. .

7. The method according to claim 6, characterized in that, S5 includes: Based on the current power and current voltage Calculate the basic DPD parameters under the current dominant operating condition. ; Based on the perturbation drift model, using the current duty cycle Pulse width Operating temperature Calculate the current disturbance drift amount ; The base DPD parameters are superimposed with the current disturbance drift to obtain the final DPD parameters under the current operating condition.

8. The method according to claim 7, characterized in that, According to the current power and current voltage Calculate the basic DPD parameters under the current dominant operating condition. ,include: Based on the current power and current voltage In the two-dimensional dominant working condition parameter set Bilinear interpolation is performed to determine the current power. and current voltage Adjacent discrete points at both ends; Calculate the position coefficients of the current power and current voltage relative to adjacent discrete points; Determine the basic DPD parameters under the current dominant operating condition based on the position coefficient. .

9. The method according to claim 8, characterized in that, The position coefficient is as follows: ， ； The basic DPD parameters are: ， correspond Basic DPD parameters, correspond Basic DPD parameters, correspond Basic DPD parameters, correspond Basic DPD parameters; , , , From the two-dimensional dominant operating condition parameter set Read from the middle.

Images