A soft partitioning method and apparatus for multi-dimensional IMSA power amplifier model

Abstract

The application discloses a soft partition method and device for a multi-dimensional IMSA power amplifier model. The MD-IMSA model is a low-complexity and high-performance multi-band power amplifier model, and a core operation thereof is to divide an amplitude space of a multi-dimensional input signal into a plurality of regions, and a nonlinearity of a power amplifier is fitted in each partition. However, how to optimally partition has been a difficult problem. The application provides a soft partition method for optimization based on a hybrid expert (MoE), and performance is significantly improved compared with traditional uniform hard partition.

Description

A soft partitioning method and apparatus for multidimensional IMSA power amplifier models Technical Field

[0001] This invention relates to the field of digital predistortion for radio frequency power amplifiers, and more particularly to a soft partitioning method and apparatus for a multidimensional IMSA power amplifier model. Background Technology

[0002] With the increasing demand for high-efficiency and high-linearity amplifiers in modern communication systems, power amplifiers (PAs) play a particularly important role in radio frequency (RF) systems. Multi-band power amplifiers (MPAs) are key components capable of operating simultaneously on multiple frequency bands and providing high power output, widely used in 5G, 6G, and other advanced communication systems. They efficiently support multiple communication standards, reducing system complexity and improving spectrum utilization. Compared to traditional single-band power amplifiers, MPAs offer greater flexibility and adaptability, amplifying power across multiple frequency bands to meet diverse spectrum requirements. However, due to the nonlinear characteristics of power amplifiers, especially at high power outputs, severe signal distortion often occurs, affecting system performance, such as bandwidth limitations, signal distortion, and reduced power efficiency. Intermodulation between signals from multiple frequency bands in MPAs leads to even more severe distortion; effectively suppressing these nonlinear distortions has become a pressing technical challenge in RF power amplifier design.

[0003] Digital predistortion (DPD) technology has been widely used for power amplifier linearization. For multi-band power amplifiers, various multi-band power amplifier behavior models have been proposed and widely applied. Among them, the multidimensional instantaneous sample index amplitude selection affine function (MD-IMSA) power amplifier model proposed by Wang Kai et al. has the advantages of low complexity and high performance. The core operation of this model is to divide the amplitude space of the multidimensional input signal into several regions and fit the nonlinearity of the power amplifier in each region. However, how to optimize the partitioning has always been a difficult problem. The original MD-IMSA model is a uniform hard partition in amplitude space, but the actual communication signal is not uniformly distributed in amplitude, so this is not an optimized partitioning. Summary of the Invention

[0004] The purpose of this invention is to provide an optimized soft partitioning method and apparatus suitable for MD-IMSA power amplifier models, so as to avoid the overfitting problem caused by insufficient data samples in some partitions due to uniform hard partitioning, and improve the performance of MD-IMSA models with the same number of partitions.

[0005] To solve the above-mentioned technical problems, the specific technical solution of the present invention is as follows:

[0006] 1. The apparatus proposed in this invention comprises the following modules:

[0007] A digital predistorter based on an optimized soft-segmentation MD-IMSA model: For D frequency bands of input digital baseband signals x1(n), x2(n), ..., x D (n) Perform predistortion processing to generate digital predistortion signals in each frequency band that have characteristics opposite to those of the RF power amplifier in the corresponding frequency band.

[0008] D digital-to-analog converters: convert D-band digital predistortion signals into analog baseband signals;

[0009] D modulators: Upconvert the analog baseband signals of D frequency bands to their corresponding radio frequency;

[0010] Power combiner: combines radio frequency signals from D frequency bands into one signal;

[0011] Multiband power amplifier: Amplifies and outputs the multiband radio frequency signal synthesized by the power combiner;

[0012] Coupler: Couples a small portion of the power amplifier's output signal for feedback;

[0013] Power divider: splits the feedback signal into D paths for processing in each frequency band;

[0014] D demodulators: down-convert the RF signal of each frequency band to obtain the analog baseband signal;

[0015] D analog-to-digital converters: converting analog baseband signals across D frequency bands into corresponding digital baseband signals y1(n), y2(n), ..., y D (n);

[0016] Model soft partitioning and parameter training algorithm: Based on the D frequency band input and output signals x1(n), x2(n), ..., x D (n) and y1(n),y2(n),…,y D (n) Calculate the soft partition and model parameters for the MD-IMSA model optimization.

[0017] 2. Building upon step 1, the derivation is performed using the power amplifier forward model corresponding to the signal in frequency band 1 as an example. The derivation for other frequency bands follows the same pattern. The corresponding MD-IMSA model expression is then:

[0018]

[0019] In formula (1), n∈[1,N] represents the index position of the digital baseband signal sampling point, and the total length of the baseband signal is N sampling points; m is the memory depth and m∈[1,M]. M is the maximum memory depth. p∈[1,P] is the index position of the partition, and P is the total number of partitions. d∈[1,D] is the index number of the frequency band, and D is the total number of frequency bands of the power amplifier. x1(n),x2(n),…,x D (n) represents the input complex signal for each frequency band, and || is defined as the modulo operation on the complex number. The output complex signal of the model, For the output of the sub-model of the p-th partition, a md,p and b m,p Let x be the parameters of the sub-model for the p-th partition. Define a D-dimensional magnitude vector A(n) = [|x1(n)|,|x2(n)|,…,|x...]. D (n)|]. For the MD-IMSA model, the partitioning method is to uniformly divide the amplitude space corresponding to A(n) into P partitions, which are represented as g p (n) is the gate function that selects each partition, defined as

[0020]

[0021] As can be seen, MD-IMSA can be viewed as being transmitted via g. p (n) selects the output of the sub-model of the corresponding p-th partition, thus it is a uniform hard partition. Since the actual communication signal is not uniformly distributed in amplitude, and only one partition sub-model works at any given time, if the partition can be optimized and multiple sub-models work at the same time, the model performance can be further improved. The model error and gate function are modeled using a probability model, as shown in (3).

[0022]

[0023] where x(n)=[x1(n),x2(n),…,x D [y1(n)], where y1(n) is the actual output of the power amplifier in band 1. The probability of error in the sub-model of the p-th partition is... It is modeled using a complex Gaussian process, as shown in (4).

[0024]

[0025] in It is the variance of the Gaussian process, θ p It is the set of parameters for this probability model, defined as follows: Where c p=[a 00,p ,b 00,p …,a MD,p ,b MD,p ] is the parameter vector of the sub-model of the p-th partition. Regarding g p The probability model (n) defines a variable z. p (n), if A(n) belongs to partition p, this value is 1; otherwise, it is 0. p Defined as the prior probability that any data belongs to the p-th partition. It is a multidimensional Gaussian distribution, which represents the probability that A(n) appears at the corresponding position in the p-th partition, as shown in expression (5).

[0026]

[0027] In the formula α p ={μ p ,∑ p} is the set of parameters of the Gaussian distribution, μ p and ∑ p These are the mean vector and covariance matrix of the Gaussian process, respectively, where T represents the transpose of the vector or matrix. -1 This represents the matrix inversion operation. We can see the gate function g for each partition at this point. p (n) are all not equal to 0, thus achieving soft partitioning. This method of weighting multiple sub-models using a probability model is called the Hybrid Expert (MoE) method. If the gate function g defined in (3) is... p Substituting (n) into (1) yields the MD-IMSA model of soft partitions optimized by the hybrid expert method (MoE-MD-IMSA).

[0028] 3. Based on the MoE-MD-IMSA model for multi-band soft partitioning described in section 2, the parameter solution method for the model is introduced below. This model is solved using the Expectation-Maximization (EM) method. First, an auxiliary variable h is defined. p (n) is

[0029]

[0030] The model parameters in equation (3) can be solved sequentially according to equation (7).

[0031]

[0032] Where Y1 = [y1(1), ..., y1(N)] T Ψ is the vector composed of the output signals, H is the model matrix corresponding to (1), and H represents the conjugate transpose of the complex matrix. pThe diagonal elements are {h p (1),…,h p The diagonal weighted matrix of {N}. To achieve optimal performance, the solution process requires multiple iterations, as detailed below:

[0033] Step 3.1: Initialize the model parameters in formula (3).

[0034] Step 3.2: Substitute the current model parameters and power amplifier input / output data into equations (3) and (4) to calculate the model error probability corresponding to each partition. The probability g of the gate function p (n).

[0035] Step 3.3: Substitute the current model parameters and power amplifier input / output data into equation (6) to calculate the auxiliary variable h corresponding to each partition. p (n).

[0036] Step 3.4: Update the model parameters according to equation (7).

[0037] Step 3.5: Substitute the updated model parameters into equation (1) to obtain the model output under the current parameters, and determine whether the modeling error meets the requirements. If it does, the parameter solution process is complete. Otherwise, proceed to step 3.2 to start the next iteration.

[0038] Beneficial Effects: This invention provides a soft partitioning method and apparatus for optimizing MD-IMSA power amplifier models. It has the following advantages: Novelly, by using a multidimensional Gaussian distribution probability to model the contribution of each partition sub-model, soft partitioning is achieved, enabling multiple partition-corresponding sub-models to work collaboratively, thus improving the final model accuracy. Simultaneously, it effectively avoids overfitting problems caused by insufficient data samples in some partitions due to uniform hard partitioning. Attached Figure Description

[0039] Figure 1 is a schematic diagram of the power amplifier digital predistortion device of the present invention;

[0040] Figure 2 shows the power spectrum of the power amplifier output signal before and after predistortion for two dual-band signals with a bandwidth of 30MHz, a sampling rate of 168.84MSPS, and a peak-to-average ratio of 7dB. Detailed Implementation

[0041] To better understand the purpose, structure, and function of this invention, the following detailed description, in conjunction with the accompanying drawings, provides an example of a soft partitioning device and method for a multidimensional IMSA power amplifier model.

[0042] As shown in Figure 1, considering the scenario of a D-band power amplifier, the power amplifier digital predistortion device of the present invention comprises a predistorter based on a soft-segment MD-IMSA model, an analog-to-digital converter from band 1 to band D, a modulator from band 1 to band D, a power combiner, a multi-band power amplifier, a coupler, a power divider, a demodulator from band 1 to band D, an analog-to-digital converter from band 1 to band D, and modules such as model soft partitioning and parameter training algorithm.

[0043] The overall process of this invention is as follows: First, multi-band baseband signals x1(n), x2(n), ..., x that have not undergone pre-distortion are used. D (n) The analog RF signals of their respective frequency bands are obtained by sequentially passing through digital-to-analog converters and quadrature modulators in their respective frequency bands. After being combined by a power combiner, they drive a multi-band RF power amplifier. The RF signal output from the power amplifier, which contains distortion, is coupled by a coupler for a portion of the signal for feedback. The coupled signal is then output to the quadrature demodulator and analog-to-digital converter corresponding to each frequency band by a power divider, resulting in the distorted digital baseband signals y1(n), y2(n), ..., y1(n) output from each frequency band of the power amplifier. D (n). Then, the model configuration parameters to be used are set, such as the number of partitions and memory depth. Finally, the soft partitioning method proposed in this invention is used to establish the power amplifier model for the output signal of each frequency band. This process is described in detail using frequency band 1 as an example.

[0044] Step 1. The corresponding MD-IMSA model expression for frequency band 1 at this time is:

[0045]

[0046] In formula (1), n∈[1,N] represents the index position of the digital baseband signal sampling point, and the total length of the baseband signal is N sampling points; m is the memory depth and m∈[1,M]. M is the maximum memory depth. p∈[1,P] is the index position of the partition, and P is the total number of partitions. d∈[1,D] is the index number of the frequency band, and D is the total number of frequency bands of the power amplifier. x1(n),x2(n),…,x D (n) represents the input complex signal for each frequency band, and || is defined as the modulo operation on the complex number. The output complex signal of the model, For the output of the sub-model of the p-th partition, a md,p and b m,p Let x be the parameters of the sub-model for the p-th partition. Define a D-dimensional magnitude vector A(n) = [|x1(n)|,|x2(n)|,…,|x...]. D (n)|]. For the MD-IMSA model, the partitioning method is to uniformly divide the amplitude space corresponding to A(n) into P partitions, which are represented as gp (n) is the gate function that selects each partition, defined as

[0047]

[0048] As can be seen, MD-IMSA can be viewed as being achieved through gp ( n) selects the output of the corresponding p-th partition sub-model, thus it is a uniform hard partition. Since the actual communication signal is not uniformly distributed in amplitude, and only one partition sub-model works at a time, if the partition can be optimized and multiple sub-models work at the same time, the model performance can be further improved. The model error and gate function are modeled using a probability model, as shown in (3).

[0049]

[0050] where x(n)=[x1(n),x2(n),…,x D [y1(n)], where y1(n) is the actual output of the power amplifier in band 1. The probability of error in the sub-model of the p-th partition is... It is modeled using a complex Gaussian process, as shown in (4).

[0051]

[0052] in It is the variance of the Gaussian process, θ p It is the set of parameters for this probability model, defined as follows: Where c p =[a 00,p ,b 00,p …,a MD,p ,b MD,p ] is the parameter vector of the sub-model of the p-th partition. Regarding g p The probability model (n) defines a variable z. p (n), if A(n) belongs to partition p, this value is 1; otherwise, it is 0. p Defined as the prior probability that any data belongs to the p-th partition. It is a multidimensional Gaussian distribution, which represents the probability that A(n) appears at the corresponding position in the p-th partition, as shown in expression (5).

[0053]

[0054] In the formula α p ={μ p ,∑ p} is the set of parameters of the Gaussian distribution, μ pand ∑ p These are the mean vector and covariance matrix of the Gaussian process, respectively. T represents the transpose of the vector or matrix, and -1 represents the inversion of the matrix. We can see that the gate function g for each partition is... p (n) are all not equal to 0, thus achieving soft partitioning. This method of weighting multiple sub-models using a probability model is called the Hybrid Expert (MoE) method. If the gate function g defined in (3) is... p Substituting (n) into (1) yields the MD-IMSA model of soft partitions optimized by the hybrid expert method (MoE-MD-IMSA).

[0055] Step 2. Based on the multi-band soft partitioning MoE-MD-IMSA model described in Step 1, the parameter solution method for the model is introduced below. This model is solved using the Expectation Maximum (EM) method. First, an auxiliary variable h is defined. p (n) is

[0056]

[0057] The model parameters in equation (3) can be solved sequentially according to equation (7).

[0058]

[0059] Where Y1 = [y1(1), ..., y1(N)] T W is a vector composed of output signals, ψ is the corresponding model matrix in (1), and H represents the conjugate transpose of the complex matrix. p The diagonal elements are {h p (1),…,h p The diagonal weighted matrix of (N)}.

[0060] Step 3. To achieve optimal performance in the parameter solving process of Step 2, multiple iterations are required. The specific process is as follows:

[0061] Step 3.1: Initialize the model parameters in formula (3).

[0062] Step 3.2: Substitute the current model parameters and power amplifier input / output data into equations (3) and (4) to calculate the model error probability corresponding to each partition. The probability g of the gate function p (n).

[0063] Step 3.3: Substitute the current model parameters and power amplifier input / output data into equation (6) to calculate the auxiliary variable h corresponding to each partition. p (n).

[0064] Step 3.4: Update the model parameters according to equation (7).

[0065] Step 3.5: Substitute the updated model parameters into equation (1) to obtain the model output under the current parameters, and determine whether the modeling error meets the requirements. If it does, the parameter solution process is complete. Otherwise, proceed to step 3.2 to start the next iteration.

[0066] After the above soft partitioning optimization and parameter training process, a forward model for power amplifier band 1 can be obtained. By swapping the output and input positions of the power amplifier, the basis function optimization and coefficient estimation of the inverse model of the power amplifier, i.e., the predistorter, can be completed. The final selected basis functions and model coefficients are used to update the predistorter, and then the input signal is injected into the predistorter to obtain the pre-distorted signal. The entire pre-distortion process is completed, and other frequency bands follow the same procedure.

[0067] The baseband signal is a complex signal containing real and imaginary parts. The power amplifier being tested is a dual-band power amplifier with center frequencies of 3.42GHz and 3.58GHz for the two bands, respectively. The two baseband signals with a bandwidth of 30MHz, a sampling rate of 168.84MSPS, and a peak-to-average power ratio of 7dB are modulated to the corresponding radio frequency.

[0068] In formula (1), the model is configured with dimension D = 2, memory depth M = 5, and number of partitions P = 25. Then, the model is soft-partitioned for optimization and parameter training according to the steps proposed in this invention, and corresponding predistortion signals are generated to linearize the power amplifier. Table 1 shows the normalized mean square error (NMSE) and adjacent channel leakage suppression ratio (ACLR) before and after predistortion for the two frequency bands. Figure 2 shows the power spectrum of the power amplifier output before and after predistortion for the two frequency bands. Combining Table 1 and Figure 2, it can be seen that after predistortion using the model proposed in this invention, the signals in both frequency bands can achieve an NMSE performance improvement of approximately 20 dB, and the ACLR of the corresponding sidebands in both frequency bands is reduced by approximately 19-22 dB, indicating a significant improvement in the linearity of the power amplifier.

[0069] Table 1. NMSE and ACPR of the two frequency bands

[0070]

[0071] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A soft partitioning method for a multidimensional IMSA power amplifier model, characterized in that, The process includes the following steps: Taking the forward model of the power amplifier corresponding to the signal in frequency band 1 as an example, the derivation is performed, and so on for the other frequency bands. The corresponding MD-IMSA power amplifier model expression is as follows: In formula (1), n∈[1,N] represents the index position of the sampling point of the digital baseband signal, and the total length of the baseband signal is N sampling points; m is the memory depth and m∈[1,M], where M is the maximum memory depth; p∈[1,P] is the index position of the partition, where P is the total number of partitions; Index numbers of the frequency band d∈[1,D]; x1(n), x2(n), ..., x D (n) represents the input complex signal for each frequency band; || is defined as the modulo operation on the complex number; The output complex signal of the model, For the output of the sub-model of the p-th partition, a md，p and b m，p These are the parameters of the sub-model for the p-th partition; Define a D-dimensional magnitude vector A(n) = [|x1(n)|, |x2(n)|, ..., |x...]. D For the MD-IMSA power amplifier model, the partitioning method is to uniformly divide the amplitude space corresponding to A(n) into P partitions, which are represented as... g p (n) is the gate function that selects each partition, defined as The model error and gate function of the MD-IMSA power amplifier are modeled using a probabilistic model, as shown in expression (3). where x(n)=[x1(n),x2(n),…,x D [y1(n)], where y1(n) is the actual output of the power amplifier in band 1; the probability of error in the sub-model of the p-th partition is... It is modeled using a complex Gaussian process, as shown in (4). in It is the variance of the Gaussian process, θ p It is the set of parameters for a probabilistic model, defined as follows: Where c p =[a 00,p b 00,p …, a MD,p b MD，p ] is the parameter vector of the sub-model of the p-th partition; regarding g p A probability model of (n) is defined, with a variable z. p (n), if A(n) belongs to partition p, then the value of this variable is 1, otherwise it is 0; a p Defined as the prior probability that any data belongs to the p-th partition. It is a multidimensional Gaussian distribution, which represents the probability that A(n) appears at the corresponding position in the p-th partition, as shown in expression (5). In the formula α p ={μ p , ∑ p } is the set of parameters of a Gaussian distribution, μ p and ∑ p These are the mean vector and covariance matrix of the Gaussian process, respectively. T This represents the transpose operation of a vector or matrix. -1 This represents the matrix inversion operation; the gate function g defined in (3) is used to... p Substituting (n) into (1), we obtain the MD-IMSA power amplifier model based on optimized soft partitioning.

2. A soft partitioning device for a multidimensional IMSA power amplifier model, characterized in that, include: The digital predistorter of the optimized soft-segmented MD-IMSA power amplifier model obtained based on the method described in claim 1: For D frequency bands of input digital baseband signals x1(n), x2(n), ..., x... D (n) Perform predistortion processing to generate digital predistortion signals in each frequency band that have characteristics opposite to those of the RF power amplifier in the corresponding frequency band. This represents the index position of the sampling point of the digital baseband signal, and the total length of the baseband signal is N sampling points; D digital-to-analog converters: convert the digital predistortion signals of D frequency bands into analog baseband signals; D modulators: upconvert the analog baseband signals of D frequency bands to their corresponding RF frequencies; Power combiner: sums and combines the RF signals of D frequency bands into one channel; Multi-band power amplifier: amplifies the multi-band RF signal synthesized by the power combiner and outputs it; Coupler: couples a portion of the output signal from the power amplifier for feedback; Power divider: divides the feedback signal into D channels for processing each frequency band; D demodulators: downconvert the RF signal of each frequency band to obtain an analog baseband signal; D analog-to-digital converters: convert the analog baseband signals of D frequency bands from analog to digital to the corresponding digital baseband signals y1(n), y2(n), ..., y D (n); Model soft partitioning and parameter training algorithm module: Based on the D frequency band input and output signals x1(n), x2(n), ..., x D y(n) and y1(n), y2(n), ..., y D (n) Perform the soft partitioning and calculation of model parameters for the MD-IMSA power amplifier model optimization.

3. A soft partitioning device for a multidimensional IMSA power amplifier model according to claim 2, characterized in that, The parameters of the optimized soft partition-based MD-IMSA power amplifier model are solved using the expectation-maximization method.

4. A soft partitioning device for a multidimensional IMSA power amplifier model according to claim 3, characterized in that, The specific steps for solving the parameters of the optimized soft partition-based MD-IMSA power amplifier model using the expectation-maximization method include: Step 4.1, initializing the MD-IMSA power amplifier model parameters in formula (3); Step 4.2, substituting the current MD-IMSA power amplifier model parameters and power amplifier input / output data into formulas (3) and (4) to calculate the MD-IMSA power amplifier model error probability corresponding to each partition. The probability g of the gate function p (n); Step 4.3, substitute the current MD-IMSA power amplifier model parameters and power amplifier input and output data into formula (6) to calculate the auxiliary variable h corresponding to each partition. p (n); Step 4.4: Update the MD-IMSA power amplifier model parameters according to formula (7); Where Y1 = [y1(1), ..., y1(N)] T It is a vector composed of output signals, and Ψ is the corresponding MD-IMSA power amplifier model matrix in (1). H W represents the conjugate transpose of a complex matrix. p The diagonal elements are {h p (1), ..., h p (N)} diagonal weighted matrix; Step 4.5, substitute the updated MD-IMSA power amplifier model parameters into formula (1) to obtain the MD-IMSA model output under the current parameters, and determine whether the modeling error meets the requirements. If it does, the parameter solution process is completed; otherwise, proceed to step 4.2 to start the next iteration.

Images