Digital pre-distortion method and pre-distortion system for high-order QAM modulated signals

Abstract

The present invention relates to a digital pre-distortion method and pre-distortion system suitable for high-order QAM modulated signals. Taking a constellation diagram of mQAM modulated signals as an example, the constellation diagram of the mQAM modulated signals is divided according to different signal amplitudes, wherein signals having the same signal amplitude form one circle; the signals are then clustered by means of a fuzzy C-means clustering algorithm; the clustered signals are grouped into several categories, and each category of data corresponds to similar power amplifier behaviors; and after a sub-model is established for each category of signals, different digital pre-distortion sub-models are used for processing. A digital pre-distorter taking a memory polynomial as a model is employed to implement a pre-distortion function for the high-order modulated signals, so as to compensate for nonlinear distortion of the signals caused by a radio frequency power amplifier, wherein the digital pre-distorter uses a memory polynomial model. A pre-distortion parameter extraction algorithm uses a recursive least squares method. Compared with existing traditional digital pre-distortion methods, the whole design scheme has the advantage of lower error vector amplitude.

Description

A digital predistortion method and predistortion system for QAM high-order modulation signals Technical Field

[0001] The present invention relates to the field of power amplifier digital predistortion, and more particularly to a digital predistortion method and a predistortion system for a QAM high-order modulation signal. Background Art

[0002] With the advancement of wireless communication technology, the first generation (1G) of mobile communication systems has evolved into the fifth generation (5G). The commercial launch of the fifth generation (5G) marks the official entry into commercial operation of a new, more advanced wireless communication technology. Compared to 4G, 5G offers the following advantages: First, faster download speeds. Compared to 4G LTE (Long-Term Evolution)'s 5Mbps to 100Mbps, 5G can reach between 1Gbps and 10Gbps, reaching up to a gigabit per second. Second, lower latency. 4G LTE's latency is typically between 20ms and 50ms, while 5G's latency is less than 1ms, which translates to a wider range of applications for 5G. Third, support for emerging technologies. Due to the convenience brought by 5G's low latency and high download speeds, 5G is widely used in areas such as remote healthcare, autonomous vehicles, and virtual reality. Fourth, greater network capacity. 5G can simultaneously support millions of connected devices per square kilometer without compromising network performance.

[0003] However, to meet these requirements, 5G communications employ more complex, higher-order modulation systems, such as 64QAM and 256QAM. Meanwhile, with the development of WiFi 6E and WiFi 7 technologies, they also employ even more complex modulation systems, typically reaching 4096QAM. However, as WiFi technology continues to advance, modulation schemes may become even higher, potentially reaching 8192QAM. However, these systems are still primarily suitable for indoor, short-range wireless communications, resulting in weak signal penetration. To enhance signal transmission, they must be combined with a power amplifier, which inevitably leads to signal distortion during transmission. This distortion is primarily caused by the power amplifier. However, it is noteworthy that relatively little research has focused on addressing the nonlinear distortion caused by power amplifiers during signal transmission, starting with signal modulation methods. This suggests that this issue has been underrepresented in previous research.

[0004] The present invention provides a digital predistortion method and predistortion system for QAM high-order modulation signals. Aiming at the ever-increasing modulation modes in the future, such as 1024QAM and 4096QAM, the present invention can better and more effectively reduce the nonlinear distortion behavior of the signal.

[0005] Summary of the Invention

[0006] The purpose of the present invention is to provide a digital predistortion method and predistortion system for QAM high-order modulation signals, which solve the above problems.

[0007] To solve the above problems, the technical solutions of the present invention are as follows:

[0008] The present invention proposes a digital predistortion method and predistortion system for QAM high-order modulation signals, comprising the following components:

[0009] This method uses a memory polynomial model as the power amplifier model, which is:

[0010] The predistortion method based on this model is:

[0011] The signal generator generates an mQAM (m=16, 64, 256) modulated signal, and decomposes the mQAM modulated signal into a constellation according to the constellation diagram, with the same amplitude of the constellation points as the radius r n The constellation diagram of the mQAM modulated signal can be divided into n circles. Counting from the outside to the inside, the signal amplitude of the outermost circle is the strongest, the second circle is the second, and the innermost circle is the weakest.

[0012] Based on the principle that the signal amplitudes in the constellation diagram are similar, a clustering algorithm is used to cluster the signals. The clustered signals are divided into several categories.

[0013] Similar signal amplitudes form a class, corresponding to similar power amplifier behaviors;

[0014] Model the power amplifier behavior of each classified signal, and then process it separately through the corresponding digital predistortion sub-model;

[0015] Each DPD sub-model has independent coefficients A K,M , all DPD sub-models are combined into an overall DPD model to perform digital pre-distortion on the power amplifier.

[0016] Furthermore, the mQAM (m=16, 64, 256) modulated signal is formed into a circle with the same amplitude as the constellation points in the constellation diagram. To facilitate subsequent signal clustering, the number of circles that the constellation diagram of the mQAM modulated signal can be divided into is first calculated. Assuming that it can be divided into n circles, counting from the outside to the inside, the signal amplitude of the outermost circle is the strongest, the second circle is the second strongest, and the innermost circle is the weakest. Since the constellation decomposition is based on the principle of the same amplitude and does not focus on the phase issue, when performing constellation decomposition, only half of the constellation points in a quadrant of the constellation diagram need to be calculated. The half constellation points in this quadrant are half of the constellation points that are symmetrical with y=x as the axis.

[0017] Then, the mQAM (m=16, 64, 256) modulated signal is modulated according to the constellation point with the same amplitude as the radius r n Forming a circle, the constellation diagram of the mQAM modulated signal can be divided into n circles;

[0018] The radius of each circle is r n ;

[0019] Where (z,q) is the coordinate of a half-constellation point in a quadrant of the constellation diagram, and ω is the unit signal amplitude;

[0020] Where m is the modulation base number, n is the number of turns, m can be expressed as m=a*a, then the calculation formula for the number of turns n is: n=h*a1+h(h-1)d / 2

[0021] Wherein, h=a / 2, which is half of the side length of the constellation diagram, a represents the side length of the constellation diagram, a1 represents the first term of the sum of the arithmetic progression and a1=1, and d represents the common difference and d=1.

[0022] Furthermore, clustering is performed based on the principle that the points in the constellation diagram of the mQAM modulation signal are close in distance. The specific formula of the K-means clustering algorithm is as follows: (j) =argmin j ∣∣x (i) -μ j ∣∣ 2 ,j∈1,2,…,g

[0023] Where x (i) Represents an unlabeled data set, that is, each constellation point in the constellation diagram, μ j represents g randomly selected cluster centroids, c (j) Represents each class. For each classified class, the centroid value of the class needs to be recalculated until the algorithm converges.

[0024] Furthermore, the fuzzy C-means clustering algorithm is used to cluster the mQAM modulated signal. Clustering is performed based on the principle that the points in the constellation diagram of the mQAM modulated signal are close to each other. The specific formula of the fuzzy C-means clustering algorithm is as follows:

[0025] In the formula, j represents the jth class, i represents the i-th sample data, p represents the p-th feature of the sample data, and v i,j Indicates the membership of the i-th sample data to the j-th category, the membership degree is between (0,1), c jrepresents the cluster center of the jth class; P(V,C) is the cost function. When its value is minimum, the iteration is terminated by the following conditions:

[0026] Where s is the number of iteration steps and ε is the error threshold. When the error is lower than the threshold, the iteration stops immediately.

[0027] The specific steps are:

[0028] ① Initialization: Initialize V = [v i,j ], initialize the matrix U determined by the membership function 0 ;

[0029] ②Calculate the center value c of the cluster j ;

[0030] ③Calculate the new membership matrix

[0031] ④Comparison and If the change between the two is less than the threshold ε, the algorithm stops, otherwise go to step ② and repeat steps ② to ④ until the change between the two is less than the threshold ε.

[0032] Furthermore, the predistortion parameters of each DPD sub-model of the high-order modulated signal clustered by the fuzzy C-means clustering algorithm are obtained. Assuming that the mQAM high-order modulated signal is divided into g categories, the recursive least squares method or the least squares method is used to obtain the predistortion parameters of the DPD sub-model of the corresponding category for each category of signal.

[0033] The predistortion parameter of the DPD model with weak memory effect and weak nonlinear signal can be obtained as A K,M =[a 1,1 ,a 0,1 ], the signal amplitude after clustering is B1; the predistortion parameter of the DPD model with weaker memory effect and weaker nonlinear signal is A K,M =[a 2,2 ,a 1,2 ,a 0,2 ], the signal amplitude after clustering is B2; the predistortion parameter of the DPD model with strong memory effect and strong nonlinear signal is A K,M =[a 3,3 ,a 2,3 ,a 1,3 ,a 0,3 ], the signal amplitude after clustering is B3; the predistortion parameter of the DPD model with stronger memory effect and stronger nonlinear signal is A K,M =[a g-2,g-2 a g-3,g-2 ,…,a 1,g-2 ,a 0,g-2], the signal amplitude after clustering is B g-2 The predistortion parameter of the DPD model for a strong memory effect and a strong nonlinear signal is A K,M =[a g-1,g-1 ,a g-2,g-1 ,…,a 1,g-1 ,a 0,g-1 ], the signal amplitude after clustering is B g-1 .

[0034] Furthermore, the predistortion system consists of: a QAM signal generation module, an up-conversion module, a power amplifier, a coupling module, a low noise amplifier, a down-conversion module, a predistortion parameter extraction module and a digital predistorter.

[0035] Furthermore, the digital predistorter is used to extract the nonlinearity and memory effects of the broadband RF power amplifier. It is composed of multiple sub-predistorters connected in parallel and cascaded to the front end of the power amplifier for predistortion processing.

[0036] Furthermore, the PA model adopts a memory polynomial model, and the behavioral model of the power amplifier is composed of multiple sub-models in parallel, each sub-model corresponds to a DPD sub-model, and the memory polynomial formula is as follows:

[0037] Where K is the nonlinear order, M is the memory depth, and a k,m is the model coefficient, and each power amplifier model corresponds to an independent set of model coefficients.

[0038] Furthermore, the digital predistortion parameter extraction module uses the RLS or LS algorithm to extract the predistortion parameters, and has independent predistortion parameters for each type of classified signal.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention provides a digital predistortion method and predistortion system for QAM high-order modulation signals, which uses QAM high-order modulation as a starting point to solve the nonlinear distortion problem caused by the high-order modulation method. The method is not only applicable to 5G, but also to WiFi7 and later WiFi technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings, which constitute part of the present invention, are provided to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are provided to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0042] FIG1 is a schematic diagram of an overall digital predistortion system of the present invention;

[0043] FIG2 is a structural block diagram of the predistortion system of the present invention. DETAILED DESCRIPTION

[0044] The following is a detailed description of the technical solutions in the embodiments of the present invention, with reference to the accompanying drawings. It is apparent that the described embodiments are only a portion of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0045] As shown in Figure 1, it is a schematic diagram of the overall digital predistortion system provided by the present invention. The mQAM high-order modulation signal is generated by the signal generator module. For the 5G communication system, a 256QAM modulation signal is generated. Then, the fuzzy C-means clustering algorithm is used to classify the signal into 6 categories. The specific steps are as follows:

[0046] ① Initialization: Initialize V = [v i,j ], initialize the matrix U determined by the membership function 0 ;

[0047] ②Calculate the center value c of the cluster j ;

[0048] ③Calculate the new membership matrix

[0049] ④Comparison and If the change between the two is less than the threshold ε, the algorithm stops, otherwise go to step ② and repeat steps ② to ④ until the change between the two is less than the threshold ε.

[0050] The classified signals are: linear signals with no memory effect, weak nonlinear signals with weak memory effect, relatively weak nonlinear signals with relatively weak memory effect, strong nonlinear signals with strong memory effect, relatively strong nonlinear signals with relatively strong memory effect, and very strong nonlinear signals. Since the first type of signal is linear and has no memory effect, there is no need to use digital predistortion for it. For the last five types of signals, the recursive least squares method or the least squares method is used to obtain the predistortion parameters of the DPD sub-model of the corresponding category.

[0051] The predistortion parameter of the DPD model with weak memory effect and weak nonlinear signal can be obtained as A K,M =[a 1,1 ,a 0,1 ], the signal amplitude after clustering is B1; the predistortion parameter of the DPD model with weaker memory effect and weaker nonlinear signal is A K,M =[a 2,2 ,a 1,2 ,a 0,2], the signal amplitude after clustering is B2; the predistortion parameter of the DPD model with strong memory effect and strong nonlinear signal is A K,M =[a 3,3 ,a 2,3 ,a 1,3 ,a 0,3 ], the signal amplitude after clustering is B3; the predistortion parameter of the DPD model with stronger memory effect and stronger nonlinear signal is A K,M =[a 4,4 a 3,4 ,a 2,4 ,a 1,4 ,a 0,4 ], the signal amplitude after clustering is B4; the predistortion parameter of the DPD model with strong memory effect and strong nonlinear signal is A K,M =[a 5,5 ,a 4,5 a 3,5 ,a 2,5 ,a 1,5 ,a 0,5 ], and the signal amplitude after clustering is B5.

[0052] The classified signal is amplified by the up-conversion module and the power amplifier module, and then input into the digital pre-distortion parameter extraction module after passing through the low-noise amplifier and the down-conversion module. It is used to train and extract the parameters of the digital pre-distorter model. Finally, the trained parameters are input into the digital pre-distorter to compensate for the nonlinearity of the power amplifier.

[0053] The digital predistorter is used to extract the nonlinear and memory effects of a broadband RF power amplifier. It is composed of multiple sub-predistorters connected in parallel and cascaded to the front end of a power amplifier for predistortion processing. The predistorter model adopts a memory polynomial model. The up-conversion module is used to modulate the signal so that its center frequency is the same as the operating frequency of the power amplifier. The power amplifier module is used to amplify the modulated signal so that it can be transmitted farther, but nonlinear distortion will inevitably be generated in this process. The coupler module is used to divide the signal amplified by the amplifier into two paths, one for spectrum observation and the other for subsequent predistortion parameter extraction. The low-noise amplifier is used to reduce noise and enhance the signal output by the coupler. The down-conversion module is used to demodulate the signal output by the low-noise amplifier. The predistortion parameter extraction module is used to extract the parameters of the digital predistorter model. The parameter extraction algorithm adopted is the RLS or LS algorithm, and the extracted parameters are passed to the digital predistorter.

[0054] More specifically, as shown in FIG2 , which is a block diagram of the predistortion system structure of the present invention, the signal generated by the signal generator module is classified by a clustering algorithm. The specific classification method is:

[0055] The K-means clustering algorithm is used to cluster the mQAM modulated signal. Clustering is performed based on the principle that the points in the constellation diagram of the mQAM modulated signal are close to each other. The specific formula of the K-means clustering algorithm is as follows: (j) =argmin j ∣∣x (i) -μ j ∣∣ 2 ,j∈1,2,…,g

[0056] Where x (i) represents the unlabeled data set, μ j represents k randomly selected cluster centroids, c (j) Represents each class. For each classified class, the centroid value of the class needs to be recalculated until the algorithm converges.

[0057] More specifically, the fuzzy C-means clustering algorithm is used to cluster the mQAM modulated signal. Clustering is performed based on the principle that the points in the constellation diagram of the mQAM modulated signal are close to each other. The specific formula of the fuzzy C-means clustering algorithm is as follows:

[0058] In the formula, j represents the jth class, i represents the i-th sample data, p represents the p-th feature of the sample data, and v i,j Indicates the membership of the i-th sample data to the j-th category, the membership degree is between (0,1), c j represents the cluster center of the jth class; P(V,C) is the cost function. When its value is minimum, the iteration is terminated by the following conditions:

[0059] Where k is the number of iteration steps and ε is the error threshold. When the error is lower than the threshold, the iteration stops immediately.

[0060] The classified signals are input into the algorithm training module, and the coefficients of the DPD model for each type of signal are extracted. Multiple DPD sub-models are then connected in parallel to form an overall DPD model that is cascaded to the front end of the power amplifier to perform digital pre-distortion on the power amplifier.

[0061] More specifically, the PA model adopts a memory polynomial model, and the behavioral model of the power amplifier is composed of multiple sub-models in parallel, each sub-model corresponds to a DPD sub-model. The memory polynomial formula is as follows:

[0062] Where K is the nonlinear order, M is the memory depth, and a k,m is the model coefficient, and each power amplifier model corresponds to an independent set of model coefficients.

[0063] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A digital predistortion method for a QAM high-order modulation signal, characterized in that: include: This method uses a memory polynomial model as the power amplifier model, which is: The predistortion method based on this model is: The signal generator generates an mQAM (m=16, 64, 256) modulated signal, and the mQAM modulated signal is decomposed according to the constellation diagram, with the same amplitude of the constellation point as the radius r n The constellation diagram of the mQAM modulation signal can be divided into n circles. Counting from the outside to the inside, the signal amplitude of the outermost circle is the strongest, followed by the second circle, and the innermost circle is the weakest. Based on the principle that the signal amplitudes in the constellation diagram are similar, a clustering algorithm is used to cluster the signals, and the clustered signals are divided into several categories; Similar signal amplitudes form a class, corresponding to similar power amplifier behaviors; Model the power amplifier behavior for each type of classified signal, and then process it separately through the corresponding digital pre-distortion sub-model; Each DPD sub-model has its own independent coefficient A K,M , all DPD sub-models are combined into an overall DPD model to perform digital pre-distortion on the power amplifier.

2. The digital predistortion method for QAM high-order modulation signals according to claim 1, characterized in that: The mQAM (m=16, 64, 256) modulated signal is formed into a circle with the same amplitude of the constellation points in the constellation diagram as the radius. To facilitate the subsequent clustering of the signal, the constellation diagram of the mQAM modulated signal is first calculated to be divided into several circles. Assuming that it can be divided into n circles, counting from the outside to the inside, the signal amplitude of the outermost circle is the strongest, the second circle is the second, and the innermost circle is the weakest, including: First, since the constellation decomposition is based on the principle of the same amplitude and does not focus on the phase problem, when performing constellation decomposition, it is only necessary to calculate half of the constellation points in a certain quadrant of the constellation diagram. The half constellation points in the quadrant are half of the constellation points obtained by symmetry with y=x as the axis; Then, the mQAM (m = 16, 64, 256) modulated signal is modulated according to the constellation point in the constellation diagram with the same amplitude as the radius r n To form a circle, the constellation diagram of the mQAM modulated signal can be divided into n circles; The radius of each circle is r n ; Among them, (z, q) is the point coordinates of a half constellation point in a quadrant of the constellation diagram, and ω is the unit signal amplitude; Where m is the modulation base number, n is the number of turns, m can be expressed as m=a*a, then the calculation formula for the number of turns n is: n=h*a1+h(h-1)d / 2 Wherein, h=a / 2, which is half of the side length of the constellation diagram, a represents the side length of the constellation diagram, a1 represents the first term of the sum of the arithmetic progression and a1=1, and d represents the common difference and d=1.

3. The digital predistortion method for QAM high-order modulation signals according to claim 1, characterized in that: The method of clustering the mQAM modulated signal using a K-means clustering algorithm includes: Clustering is performed based on the principle that the points in the constellation diagram of the mQAM modulation signal are close in distance. The specific formula of the K-means clustering algorithm is as follows: c (j) =argmin j ∣∣x (i) -m j ∣∣ 2 ,j∈1,2,…,g In the formula, x (i) represents an unlabeled data set, that is, each constellation point in the constellation diagram, μ j represents the g initial cluster centroid points randomly selected, c (j) Represents each class. For each classified class, the centroid value of the class needs to be recalculated until the algorithm converges.

4. The digital predistortion method for QAM high-order modulation signals according to claim 1, characterized in that: The method of clustering the mQAM modulated signal using a fuzzy C-means clustering algorithm includes: Clustering is performed based on the principle that the points in the constellation diagram of the mQAM modulation signal are close in distance. The specific formula of the fuzzy C-means clustering algorithm is as follows: In the formula, j represents the jth class, i represents the i-th sample data, p represents the p-th feature of the sample data, and v i,j Indicates the membership of the i-th sample data to the j-th category, the membership degree is between (0,1), c j represents the cluster center of the jth class; P(V,C) is the cost function. When its value is minimum, the iteration is terminated by the following conditions: Where s is the number of iteration steps, and ε is the error threshold. When the error is lower than the threshold, the iteration is stopped immediately. The specific steps are: ① Initialization: Initialize V = [v i,j ], initialize the matrix U determined by the membership function 0 ; ② Calculate the center value c of the cluster j ; ③Calculate the new membership matrix ④Comparison and If the change of the two is less than the threshold ε, the algorithm stops, otherwise go to step ② and repeat steps ② to ④ until the change of the two is less than the threshold ε.

5. The digital predistortion method for QAM high-order modulation signals according to claim 1, characterized in that: The method of obtaining the predistortion parameters of each type of DPD sub-model of the high-order modulated signal clustered by the fuzzy C-means clustering algorithm includes: Assume that the mQAM high-order modulation signal is divided into g categories, namely: linear signal with no memory effect, signal with weak memory effect and weak nonlinearity, signal with relatively weak memory effect and relatively weak nonlinearity, signal with strong memory effect and strong nonlinearity, signal with relatively strong memory effect and relatively strong nonlinearity, signal with very strong memory effect and very strong nonlinearity, etc.; since the first category signal is linear and has no memory effect, there is no need to use digital pre-distortion for it. For the latter g-1 category signals, the recursive least squares method or the least squares method is used to obtain the pre-distortion parameters of the DPD sub-model of the corresponding category; The predistortion parameter of the DPD model with weak memory effect and weak nonlinear signal can be obtained as A K,M =[a 1,1 ,a 0,1 ], the signal amplitude after clustering is B1; the predistortion parameter of the DPD model with weaker memory effect and weaker nonlinear signal is A K,M =[a 2,2 ,a 1,2 ,a 0,2 ], the signal amplitude after clustering is B2; the predistortion parameter of the DPD model with strong memory effect and strong nonlinear signal is A K,M =[a 3,3 ,a 2,3 ,a 1,3 ,a 0,3 ], the signal amplitude after clustering is B3; the predistortion parameter of the DPD model with stronger memory effect and stronger nonlinear signal is A K,M =[a g-2,g-2 a g-3,g-2 ,…,a 1,g-2 ,a 0,g-2 ], and the signal amplitude after clustering is B g-2 ; The predistortion parameter of the DPD model for a strong memory effect and a strong nonlinear signal is A K,M =[a g-1,g-1 ,a g-2,g-1 ,…,a 1,g-1 ,a 0,g-1 ], and the signal amplitude after clustering is B g-1 .

6. The predistortion system according to claim 1, characterized in that include: QAM signal generation module, up-conversion module, power amplifier, coupling module, low noise amplifier, down-conversion module, pre-distortion parameter extraction module and digital pre-distorter.

7. The system of claim 1, wherein: The digital predistorter is used to extract the nonlinearity and memory effects of the broadband RF power amplifier. It is composed of multiple sub-predistorters connected in parallel and cascaded to the front end of the power amplifier for predistortion processing.

8. The system of claim 1, wherein: The PA model adopts a memory polynomial model, and the behavior model of the power amplifier is composed of multiple sub-models in parallel. Each sub-model corresponds to a DPD sub-model. The memory polynomial formula is as follows: In the formula, K is the nonlinear order, M is the memory depth, and a k,m are model coefficients, and each power amplifier model corresponds to an independent set of model coefficients.

9. The system of claim 1, wherein: The digital pre-distortion parameter extraction module uses the RLS or LS algorithm to extract the pre-distortion parameters, and each classified signal has an independent pre-distortion parameter.

Images