A digital predistortion system and method based on random forest algorithm

Abstract

The present invention discloses a digital predistortion system and method based on a random forest algorithm, the system comprising a digital predistortion model based on the random forest algorithm, a digital-to-analog converter, a transmitter link and a power amplifier; the method comprises establishing a digital predistortion model using a training set and a validation set, building a digital predistortion system, and system operation; the present invention performs predistortion processing on an input signal through a digital predistortion model, so that the input signal has nonlinear characteristics opposite to the input-output characteristics of the power amplifier, and then passes the preprocessed signal through a digital-to-analog converter, a transmitter link and a power amplifier to obtain an output signal having a linear amplification relationship with the original input signal, thereby realizing linear amplification of the signal. The present invention can construct a digital predistortion model with a simple principle, easy implementation and low computational complexity to realize linearization of the power amplifier.

Description

Technical Field

[0001] The present invention belongs to the field of digital predistortion technology, and in particular relates to a digital predistortion system and method based on a random forest algorithm. Background Art

[0002] With the development of wireless mobile communication systems, more users, higher data transmission rates and scarcity of spectrum resources have put forward higher requirements for spectrum resource utilization. Therefore, various multiple access technologies have been applied to communication systems to improve spectrum efficiency. However, at the same time, the resulting non-constant envelope inevitably causes the nonlinearity of the power amplifier. In order to balance the spectrum efficiency and the linearity of the power amplifier, digital pre-distortion technology is generally used to compensate for the nonlinearity of the power amplifier. The principle is to insert a pre-distortion module in front of the power amplifier. This module pre-distorts the input signal by fitting the inverse characteristics of the power amplifier.

[0003] In recent years, digital pre-distortion technology has made great progress due to its advantages of low cost and high precision, and has been widely used in the linearization of communication systems. However, most of the current pre-distortion algorithms are complex in principle, difficult to implement, and have certain limitations. The random forest algorithm is an emerging and highly flexible machine learning algorithm with the advantages of simple principle, high accuracy, ability to process high-dimensional data, and no need for feature selection. These advantages enable the algorithm to quickly build a simple and easy-to-understand behavioral model, providing a new idea for the construction of behavioral models in digital pre-distortion technology. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a digital pre-distortion system and method based on a random forest algorithm, aiming to realize digital pre-distortion with an algorithm that is simple in principle, highly interpretable and easy to implement.

[0005] Technical solution: In order to achieve the above-mentioned invention object, the present invention adopts the following technical solution:

[0006] A digital predistortion system based on a random forest algorithm comprises a digital predistortion model based on the random forest algorithm, a digital-to-analog converter, a transmitter link and a power amplifier; a digital baseband input signal generates a predistortion signal through the digital predistortion model based on the random forest algorithm, and then passes through the digital-to-analog converter, the transmitter link and the power amplifier to finally generate a linearly amplified radio frequency output signal; the signal is then received through a feedback loop and converted into an output digital signal through an analog-to-digital converter, which is basically consistent with the digital baseband input signal; wherein the digital predistortion model based on the random forest algorithm comprises a real digital predistortion model and an imaginary digital predistortion model.

[0007] Furthermore, the predistortion signal is synthesized from two results of the digital baseband input signal passing through a real part digital predistortion model and an imaginary part digital predistortion model.

[0008] Furthermore, the extraction process of the digital pre-distortion model based on the random forest algorithm is: first, the input signal is input into the power amplifier and the output signal is captured; then the input signal and the output signal of the power amplifier are used as the output signal and the input signal of the digital pre-distortion model respectively, and the random forest algorithm is used to construct the behavioral model of the digital pre-distortion model.

[0009] Furthermore, the behavioral model of the digital predistortion model includes a real digital predistortion model and an imaginary digital predistortion model, which are obtained according to equations (1) and (2) respectively:

[0010]

[0011] Where M represents the memory depth, To model the nth sample of the output signal, the value range of n is determined by the specific sampling time; To model the nth sample of the input signal, and are the first-order memory term and the second-order memory term respectively; Re{} represents the real part of the signal, Im{} represents the imaginary part of the signal; f R is the mapping relationship between the real part of the output signal and the input signal, f I It is the mapping relationship between the imaginary part of the output signal and the input signal.

[0012] Furthermore, the random forest algorithm is a parallel ensemble learning method, which samples k training samples with replacement from a given training set data, repeats the sampling T times, and obtains a sampling set of T groups of k samples, then trains a base learner based on each sample set, and then combines the prediction outputs of the T base learners by simple averaging to obtain the final prediction output.

[0013] Furthermore, the base learner is a decision tree model, which recursively calls the process of dividing the feature space, and divides each subspace into two subspaces until the stopping condition is met; the feature space of the input signal is divided into several subspaces by continuously splitting the subtrees, and a specific output is given to each subspace.

[0014] Furthermore, the process of dividing the feature space is: using the least squares method to select the segmentation variable j and the segmentation point s, and by solving equations (3-a), (3-b), and (3-c), we can obtain:

[0015]

[0016] R1(j,s)={x i |x i (j) ≤s}, R2(j,s)={x i |x i (j) >s} (3-b)

[0017]

[0018] Among them, j is the input attribute of the signal, that is, the real part and imaginary part of the current input signal or the previous M input signals, which is used as the segmentation variable when dividing the feature space; x i is the i-th input vector, y i is the same as the input vector x i The corresponding real or imaginary part of the output; R m is the mth partition space, m is the subspace number, each space can only be divided into two subspaces, so m can only be 1 or 2; is the predicted output of subspace m, N m is the number of all input vectors divided into subspace m;

[0019] Then traverse the segmentation variable j, scan the segmentation point s for the fixed segmentation variable j, find the pair (j, s) that minimizes equation (3-a), use the selected pair (j, s) to divide the space, and determine the predicted output value.

[0020] Furthermore, the stopping condition is: when the division is performed again, the minimum value of formula (3-a) is greater than that before the division.

[0021] A digital predistortion method for a digital predistortion system based on a random forest algorithm comprises the following steps:

[0022] S1: The input signal is passed through a power amplifier to generate an output signal, thereby obtaining the actual input signal and output signal of the power amplifier to be linearized, and the input signal data and the output signal data are divided into a training set and a verification set;

[0023] S2: Set the number of base learners in the random forest algorithm and use the training set data to extract the digital pre-distortion model;

[0024] S3: Save the trained digital pre-distortion model, use the validation set data to verify the modeling accuracy of the digital pre-distortion model, and complete the establishment of the digital pre-distortion model based on the random forest algorithm;

[0025] S4: Build a digital pre-distortion system based on random forest algorithm;

[0026] S5: Input the digital baseband input signal into a digital predistortion system based on a random forest algorithm to obtain an output signal that is linearly amplified with the original input signal.

[0027] Beneficial effects: Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1) The digital pre-distortion model based on the random forest algorithm is a white box model, which is easier to understand than neural networks and other machine learning algorithms;

[0029] 2) Compared with other machine learning algorithms, the random forest algorithm is easier to implement and has low computational overhead. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a block diagram of a digital predistortion system in a specific embodiment of the present invention;

[0031] Figure 2 In a specific embodiment of the present invention, an input signal with a center frequency of 2.4 GHz and a bandwidth of 20 MHz is shown, and the amplitude characteristic curve of the original input signal and the linearized signal after passing through a Doherty power amplifier;

[0032] Figure 3 In a specific embodiment of the present invention, an input signal with a center frequency of 2.4 GHz and a bandwidth of 20 MHz is shown, and the phase characteristic curve of the original input signal and the linearized signal after passing through the Doherty power amplifier;

[0033] Figure 4 This is a power spectrum diagram of the original input signal and the linearized signal of an input signal with a center frequency of 2.4 GHz and a bandwidth of 20 MHz in a specific embodiment of the present invention, after passing through a Doherty power amplifier. DETAILED DESCRIPTION

[0034] The technical solution of the present invention is further introduced below in conjunction with specific implementation modes and drawings.

[0035] This specific embodiment discloses a digital predistortion system based on a random forest algorithm, such as Figure 1As shown, the system includes a digital pre-distortion model based on the random forest algorithm, a digital-to-analog converter, a transmitter link and a power amplifier; the digital baseband input signal generates a pre-distortion signal through the digital pre-distortion model based on the random forest algorithm, and then passes through the digital-to-analog converter, the transmitter link and the power amplifier to finally generate a linearly amplified RF output signal; it is then received through a feedback loop and converted into an output digital signal through an analog-to-digital converter, which is basically consistent with the digital baseband input signal; wherein the digital pre-distortion model based on the random forest algorithm includes a real digital pre-distortion model and an imaginary digital pre-distortion model, and after the baseband input signal passes through the real digital pre-distortion model and the imaginary digital pre-distortion model, the real part and the imaginary part pre-distortion results are combined to obtain the final pre-distortion signal.

[0036] The model extraction process of the digital pre-distortion model based on the random forest algorithm is as follows: input the input signal into the power amplifier and capture the output signal. The input signal and the output signal are used as the output signal and the input signal of the digital pre-distortion model respectively, and the inverse model of the power amplifier, that is, the digital pre-distortion model, is constructed using the random forest algorithm.

[0037] In the model extraction of the digital pre-distortion model based on the random forest algorithm, the behavior model of the digital pre-distortion model is divided into two parts: a real digital pre-distortion model and an imaginary digital pre-distortion model. According to equations (1) and (2), it is obtained:

[0038]

[0039] Where M represents the memory depth, To model the nth sample of the output signal, the value range of n is determined by the specific sampling time. To model the nth sample of the input signal, and are the first-order memory term and the second-order memory term, Re{} represents the real part of the signal, and Im{} represents the imaginary part of the signal; f R is the mapping relationship between the real part of the output signal and the input signal, f I It is the mapping relationship between the imaginary part of the output signal and the input signal.

[0040] In the model extraction of the digital pre-distortion model based on the random forest algorithm, the random forest algorithm is a parallel ensemble learning method. The algorithm samples k training samples with replacement from a given training set data, repeats the sampling T times, obtains a sampling set of T groups of k samples, and then trains a base learner based on each sample set, and then combines the prediction outputs of the T base learners with a simple average method to obtain the final prediction output.

[0041] In the model extraction of the digital pre-distortion model based on the random forest algorithm, the base learner of the random forest algorithm is a decision tree. The decision tree model divides the feature space of the input signal into several units by continuously splitting subtrees, and gives each unit a specific output. Among them, the selection of the segmentation variable j and the segmentation point s when dividing the feature space adopts the least squares method, which is obtained by solving equations (3-a), (3-b), and (3-c);

[0042]

[0043] R1(j,s)={x i |x i (j) ≤s}, R2(j,s)={x i |x i (j) >s} (3-b)

[0044]

[0045] Among them, j is the input attribute of the signal, that is, the real part and imaginary part of the current input signal or the previous M input signals, which is used as the segmentation variable when dividing the feature space; x i is the i-th input vector, y i is the real or imaginary part of the output corresponding to the input vector; R m is the mth partition space, m is the subspace number, each unit can only be divided into two subunits, so m can only be 1 or 2. is the predicted output of subspace unit m. m is the number of all input vectors divided into subspace m.

[0046] Traverse the segmentation variable j, scan the segmentation point s for the fixed segmentation attribute j, find the pair (j, s) that minimizes equation (3-a), and use the selected pair (j, s) to divide the space R m And determine the predicted output value.

[0047] In the model extraction of the digital pre-distortion model based on the random forest algorithm, the decision tree algorithm recursively calls the above-mentioned step of dividing the feature space on the input space where the training data set is located to divide each subspace into two subspaces until the stopping condition is met. The stopping condition is: when the division is performed again, the minimum value of formula (3-a) is greater than that before the division.

[0048] A digital predistortion method of a digital predistortion system based on a random forest algorithm of the present invention comprises the following steps:

[0049] S1: The input signal is passed through a power amplifier to generate an output signal, thereby obtaining the actual input signal and output signal of the power amplifier to be linearized, and the input signal data and the output signal data are divided into a training set and a verification set;

[0050] S2: Set the number of base learners in the random forest algorithm and use the training set data to extract the digital pre-distortion model;

[0051] S3: Save the trained random forest model, use the validation set data to verify the modeling accuracy of the digital pre-distortion model, and complete the establishment of the digital pre-distortion model based on the random forest algorithm;

[0052] S4: Construct a digital pre-distortion system based on random forest algorithm;

[0053] S5: Input the digital baseband input signal into a digital predistortion system based on a random forest algorithm to obtain an output signal that is linearly amplified with the original input signal.

[0054] Take the input signal with a center frequency of 2.4GHz and a bandwidth of 20MHz as an example, and pass through the Doherty power amplifier. The linearization of the power amplifier using the digital predistortion technology based on the random forest algorithm proposed in the present invention can achieve better results. The amplitude characteristic curve, phase characteristic curve, and power spectrum density diagram of the original input signal and the output signal after linearization are shown in Figure 1. Figure 2-4 As shown in the figure, the linearization of the power amplifier is achieved by using the Random Forest algorithm, a simple and highly flexible machine learning algorithm, which makes the digital pre-distortion process faster, simpler and easier to understand.

[0055] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes in form and details may be made without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A digital predistortion system based on random forest algorithm, characterized in that: The system includes a digital predistortion model based on a random forest algorithm, a digital-to-analog converter, a transmitter link and a power amplifier; a digital baseband input signal generates a predistortion signal through the digital predistortion model based on the random forest algorithm, and then passes through the digital-to-analog converter, the transmitter link and the power amplifier to finally generate a linearly amplified radio frequency output signal; the signal is then received through a feedback loop and converted into an output digital signal through an analog-to-digital converter, which is basically consistent with the digital baseband input signal; wherein the digital predistortion model based on the random forest algorithm includes a real digital predistortion model and an imaginary digital predistortion model; The extraction process of the digital pre-distortion model based on the random forest algorithm is as follows: first, an input signal is input into a power amplifier to capture an output signal; then the input signal and the output signal of the power amplifier are used as the output signal and the input signal of the digital pre-distortion model respectively, and a behavior model of the digital pre-distortion model is constructed by using the random forest algorithm. The behavior model of the digital pre-distortion model includes a real digital pre-distortion model and an imaginary digital pre-distortion model, which are obtained according to equations (1) and (2) respectively: Where M represents the memory depth, To model the nth sample of the output signal, the value range of n is determined by the specific sampling time; To model the nth sample of the input signal, and are the first-order memory term and the second-order memory term respectively; Re{} represents the real part of the signal, Im{} represents the imaginary part of the signal; f R is the mapping relationship between the real part of the output signal and the input signal, f I It is the mapping relationship between the imaginary part of the output signal and the input signal.

2. The digital predistortion system based on random forest algorithm according to claim 1, characterized in that: The predistortion signal is synthesized from two results of a digital baseband input signal passing through a real part digital predistortion model and an imaginary part digital predistortion model.

3. The digital predistortion system based on random forest algorithm according to claim 1, characterized in that: The random forest algorithm is a parallel ensemble learning method, which samples k training samples with replacement from a given training set data, repeats the sampling T times, and obtains a sampling set of T groups of k samples, then trains a base learner based on each sample set, and then combines the prediction outputs of the T base learners with a simple averaging method to obtain the final prediction output.

4. The digital predistortion system based on random forest algorithm according to claim 3, characterized in that: The base learner is a decision tree model, which recursively calls the process of dividing the feature space, and divides each subspace into two subspaces until the stopping condition is met; the feature space of the input signal is divided into several subspaces by continuously splitting the subtree, and a specific output is given to each subspace.

5. The digital predistortion system based on random forest algorithm according to claim 4, characterized in that: The process of dividing the feature space is: using the least squares method to select the segmentation variable j and the segmentation point s, and by solving equations (3-a), (3-b), and (3-c), we get: R1(j,s)={x i |x i (j) ≤s},R2(j,s)={x i |x i (j)>s} (3-b) Among them, j is the input attribute of the signal, that is, the real part and imaginary part of the current input signal or the previous M input signals, which is used as the segmentation variable when dividing the feature space; x i is the i-th input vector, y i is the same as the input vector x i The corresponding real or imaginary part of the output; R m is the mth partition space, m is the subspace number, each space can only be divided into two subspaces, so m can only be 1 or 2; is the predicted output of subspace m, N m is the number of all input vectors divided into subspace m; Then traverse the segmentation variable j, scan the segmentation point s for the fixed segmentation variable j, find the pair (j, s) that minimizes equation (3-a), divide the space with the selected pair (j, s), and determine the predicted output value.

6. A digital predistortion system based on random forest algorithm according to claim 4 or 5, characterized in that: The stopping condition is: when the division is performed again, the minimum value of formula (3-a) is greater than that before the division.

7. A digital predistortion method for a digital predistortion system based on a random forest algorithm as claimed in claim 1, characterized in that: The following steps are involved: S1: The input signal is passed through a power amplifier to generate an output signal, thereby obtaining the actual input signal and output signal of the power amplifier to be linearized, and the input signal data and the output signal data are divided into a training set and a verification set; S2: Set the number of base learners in the random forest algorithm and use the training set data to extract the digital pre-distortion model; S3: Save the trained digital pre-distortion model, use the validation set data to verify the modeling accuracy of the digital pre-distortion model, and complete the establishment of the digital pre-distortion model based on the random forest algorithm; S4: Build a digital pre-distortion system based on random forest algorithm; S5: Input the digital baseband input signal into a digital predistortion system based on a random forest algorithm to obtain an output signal that is linearly amplified with the original input signal.

Images