A Digital Predistortion Method and System Based on Dimension-Weighted K-Nearest Neighbor Algorithm

Abstract

The present invention relates to the field of wireless communication technologies, and particularly relates to a digital predistortion method and system based on a dimensionality weighted K-nearest neighbor algorithm. The present invention classifies the current original signal U', groups signal elements with similar characteristics into one category, and altogether forms N paths. For each path of signal, a trained predistortion improvement model specific to that path of signal is equipped. N paths of current predistorted signals are processed and inversely synthesized into a total predistorted signal X, and then the current amplified signal Y is obtained through processing by a target power amplifier. The present invention solves the problems of limitations and poor flexibility existing in the existing methods. The amplified signal output by the present invention has higher anti-interference ability, stronger signal quality, and higher prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technologies, and particularly to a digital predistortion method and system based on a dimensionality-weighted K-nearest neighbor algorithm. Background Art

[0002] The development of wireless communication technologies is extremely important for social production and human life. With the expansion of the social group range and the exponential increase in the number of communication users, people have higher requirements for the rate of wireless communication. However, in high-rate communication, the higher the frequency of the signal, the faster its change speed, resulting in the inability of the power amplifier to linearly amplify the signal under the requirement of instantaneous high-power output. For this reason, an efficient power amplifier (PA) linearization technology is required in practical high-speed transmission systems.

[0003] Digital Predistortion (DPD) technology is a method that can effectively compensate for the nonlinearity of power amplifiers. It preprocesses the baseband signal to reduce the nonlinear distortion of the power amplifier output signal.

[0004] In currently commonly used predistortion systems, there is a contradictory relationship between the complexity of the model and the modeling accuracy. Improving the modeling accuracy requires using a model with higher complexity. Since power amplifiers exhibit different nonlinearities and memories at different power levels, the industry has proposed a method for power-segmented modeling of power amplifiers and a model selection method based on an enhanced model tree. However, this power-segmented modeling method does not consider the memory effect of the power amplifier and is a memoryless model. Since the memoryless model only considers the input and output at the current moment and cannot consider the influence of historical data, its performance may be inferior to that of the memory polynomial model for data with obvious time-series dependence, and there are limitations. In addition, the memoryless polynomial model has poor flexibility and cannot adapt to complex time-series data. Summary of the Invention

[0005] Based on this, in view of the limitations and poor flexibility of existing methods, a digital predistortion method and system based on a dimensionality-weighted K-nearest neighbor algorithm are provided.

[0006] The present invention is implemented by adopting the following technical solutions:

[0007] In a first aspect, the present invention discloses a digital predistortion method based on a dimensionality-weighted K-nearest neighbor algorithm, including the following steps:

[0008] Step 1, splitting the current original signal U' into N current signal sequences;

[0009] Wherein, the current original signal U' = [U1, U2,..., Un ;

[0010] U g is the g-th dimensional signal element in U', where g ∈ [1, n];

[0011] The method for splitting the current original signal U' includes:

[0012] S101. Obtain the maximum value U n and the minimum value U max in U min , and calculate the interval division length

[0013] S102. According to ΔU, evenly divide the total interval [U min , U max into N segments of current splitting intervals;

[0014] Among them, the l-th segment of the current splitting interval is [U min + (l - 1)ΔU, U min + lΔU]; the midpoint of the l-th segment of the current splitting interval is where l ∈ [1, N];

[0015] S103. Calculate the Euclidean distance D(U g , U l ); g , U l )

[0016] Among them, for U g , obtain the minimum value D g in D(U g , U1) to D(U N ); if D min is D(U min , U g , U L ), then classify U g into the L-th segment of the current splitting interval; L ∈ [1, N];

[0017] S104. Traverse U1 to U n , and classify U1 to U n into the N segments of the current splitting intervals respectively to form an N-channel current signal sequence;

[0018] Among them, the L-th segment of the current splitting interval is sorted in the order of the elements included to form the L-th channel current signal sequence, with a total of S elements;

[0019] Step 2. Input the L-th channel current signal sequence into the trained L-th pre-distortion improvement model to obtain the L-th channel current pre-distorted signal F L ;

[0020] Among them, F L = [f1, f2, …, f S ; f s is the sth-dimensional signal element in F L , and is obtained by processing U g through the trained Lth predistortion improvement model; s ∈ [1, S];

[0021] The Lth predistortion improvement model adopts a memory polynomial cross-term predistortion model;

[0022] The expression of f s is:

[0023] In the formula, is the predistortion processing coefficient used by the trained Lth predistortion improvement model;

[0024] Step 3: Synthesize N current predistortion signals into a total predistortion signal X;

[0025] Step 4: Process the total predistortion signal X through the target power amplifier to obtain the current amplified signal Y.

[0026] The implementation of this digital predistortion method based on the dimension-weighted K-nearest neighbor algorithm is based on the method or process of the embodiments disclosed in the present invention.

[0027] In the second aspect, the present invention discloses a digital predistortion system based on the dimension-weighted K-nearest neighbor algorithm, which uses the digital predistortion method based on the dimension-weighted K-nearest neighbor algorithm disclosed in the first aspect.

[0028] The digital predistortion system based on the dimension-weighted K-nearest neighbor algorithm includes: a signal splitting module, a predistortion processing module, a signal synthesis module, and an amplifier module.

[0029] The signal splitting module is used to split the current original signal U' into N current signal sequences. The predistortion processing module is used to input the Lth current signal sequence into the trained Lth predistortion improvement model to obtain the Lth current predistortion signal F L . The signal synthesis module is used to synthesize N current predistortion signals into a total predistortion signal X. The amplifier module is used to process the total predistortion signal X through the target power amplifier to obtain the current amplified signal Y.

[0030] The implementation of this digital predistortion system based on the dimension-weighted K-nearest neighbor algorithm is based on the method or process of the embodiments disclosed in the present invention.

[0031] In a third aspect, the present invention discloses a readable storage medium. Computer program instructions are stored in the readable storage medium. When the computer program instructions are read and run by a processor, the steps of the digital predistortion method based on the dimensionality weighted K-nearest neighbor algorithm disclosed in the first aspect are executed.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1. The present invention classifies the current original signal U', groups signal elements with similar characteristics into one category, and altogether forms N paths. For each path of signal, a trained predistortion improvement model specific to that path of signal is equipped. N paths of current predistorted signals are processed, and then inversely synthesized into the total predistorted signal X, and then the current amplified signal Y is obtained through processing by the target power amplifier. Among them, the current original signal U' is split for processing, and a targeted predistortion improvement model is used for processing, reducing the processing error caused by inconsistent features. The adopted predistortion improvement model has a memory polynomial cross term and has a better goodness of fit, can capture and compensate the nonlinear distortion of the power amplifier more comprehensively, make the amplified signal approach the ideal effect, and can reduce the loss caused by the nonlinear distortion.

[0034] 2. Through simulation verification, compared with the traditional method, the amplified signal output by the present invention has higher anti-interference ability, stronger signal quality, and higher prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0036] Figure 1 It is a data flow diagram of the digital predistortion method based on the dimensionality weighted K-nearest neighbor algorithm proposed in Embodiment 1 of the present invention;

[0037] Figure 2 For Figure 1 It is a data flow diagram for training the digital predistortion improvement model in

[0038] Figure 3 It is a comparison diagram of the simulation results of three methods in Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0040] It should be noted that when a component is referred to as being "installed on" another component, it can be directly on the other component or there can be an intermediate component. When a component is considered to be "set on" another component, it can be directly set on the other component or there can be an intermediate component at the same time. When a component is considered to be "fixed to" another component, it can be directly fixed to the other component or there can be an intermediate component at the same time.

[0041] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "or / and" used in the present invention includes any and all combinations of one or more of the related listed items.

[0042] Embodiment 1

[0043] Embodiment 1 of the present invention proposes a digital predistortion method based on a dimensionality-weighted K-nearest neighbor algorithm. Refer to Figure 1 , which is the data flow diagram of this digital predistortion method.

[0044] The digital predistortion method based on the dimensionality-weighted K-nearest neighbor algorithm includes steps one to four:

[0045] Step one, split the current original signal U' into N current signal sequences. Among them, U' = [U1, U2,..., U n . U g is the g-th dimensional signal element in U', and g ∈ [1, n].

[0046] In this step, the original data stream (i.e., the current original signal U') is used as a sample, and the amplitude of each path of data is used as a feature to classify the sample. In this way, each path of data after classification has the characteristic of being close in amplitude.

[0047] Specifically, the method of splitting the current original signal U' is a specific application of the dimensionality-weighted K-nearest neighbor algorithm, including S101 to S104:

[0048] S101, obtain the maximum value U n in U1 to U max and the minimum value Umin , and calculate the interval division length

[0049] S102. According to ΔU, divide the total interval [U min , U max evenly into N segments of current shunt intervals, namely: [U min , U min + ΔU], [U min + ΔU, U min + 2ΔU], …, [U min + (N - 1)ΔU, U max .

[0050] Among them, the l-th segment of the current shunt interval is [U min + (l - 1)ΔU, U min + lΔU].

[0051] The midpoint of the l-th segment of the current shunt interval is l ∈ [1, N].

[0052] S103. Calculate the Euclidean distance D(U g , U l ) between U g , U l .

[0053] The expression of D(U g , U l ) is:

[0054]

[0055] Among them, ω g represents the weighting value of the g-th dimension.

[0056] For U g , obtain the minimum value D g in D(U g , U1) to D(U N ); if D min is D(U min , U g , U L ), then classify U g into the L-th segment of the current shunt interval; L ∈ [1, N].[[]END]

[0057] S104. Traverse U1 to U n , and classify U1 to U n into the N segments of the current shunt intervals respectively to form an N-channel current signal sequence.

[0058] Of course, the elements in the current signal sequence are not in disorder, but are arranged in the order in which the elements are classified. That is to say, the L-th current branch interval is sorted in the order in which the elements are classified to form the L-th current signal sequence, with a total of S elements.

[0059] Step 2: Input the L-th current signal sequence into the trained L-th pre-distortion improvement model to obtain the L-th current pre-distorted signal F L 。

[0060] Since there are N current signal sequences, there are also N trained pre-distortion improvement models.

[0061] It should be noted that the pre-distortion improvement model uses a pre-distortion model with memory polynomial cross terms. In this way, in addition to the historical data of the input and output, the cross influence between the input and output historical data is also considered, which can improve the prediction accuracy of the model.

[0062] F L =[f1,f2,…,f S 。Where f s is the s-th dimensional signal element in F L and is obtained by processing U g through the trained L-th pre-distortion improvement model; s ∈ [1, S].

[0063] f s The expression of is:

[0064]

[0065] In the formula, is the pre-distortion processing coefficient used by the trained L-th pre-distortion improvement model;

[0066] Among them, belongs to the first type of pre-distortion processing coefficient a mq and is a basic parameter affecting the pre-distortion processing process. belongs to the second type of pre-distortion processing coefficient b q and it reflects the cross influence between the input and output historical data. belongs to the third type of pre-distortion processing coefficient c, which reflects the influence intensity of the memory polynomial cross terms.

[0067] Of course, for the N trained pre-distortion improvement models, N original pre-distortion improvement models need to be used for training. Specifically, see Figure 2 , the training method includes S201~S202:

[0068] S201, divide the original sample signal W’ into N test signal sequences. Among them, the original sample signal W’ = [W1, W2, …, W n ; W g is the g-th dimensional signal element in W’.

[0069] The method for dividing the original sample signal W’ in this step is the same in principle as the method for dividing U’ in step one, including S301 to S304:

[0070] S301, obtain the maximum value W n and the minimum value W max in W1 to W min , and calculate the interval division length

[0071] S302, according to ΔW, evenly divide the total interval [W min , W max into N test division intervals;

[0072] Among them, the l-th test division interval is [W min + (l - 1)ΔW, W min + lΔW]; the midpoint of the l-th test division interval is l ∈ [1, N];

[0073] S303, calculate the Euclidean distance D(W g , W l ); g , W l )

[0074] Among them, for W g , obtain the minimum value W g in D(W g , W1) to D(W N ); if W min is D(W min , W g ), then classify W L into the L-th division interval; L ∈ [1, N]; g

[0075] S304, traverse W1 to W n , and classify W1 to W n into the N test division intervals respectively to form N test signal sequences;

[0076] Among them, the L-th test division interval is sorted according to the order of the elements included to form the L-th test signal sequence.

[0077] Meanwhile, multiply the original sample signal W’ by the amplification gain G of the target power amplifier to obtain the desired amplified signal d’. Among them, the desired amplified signal d’ = [d1, d2, …, d n ; d g is the g-th dimensional signal element in d’, and d g = W g *G.

[0078] S202. Iterate the L-th original predistortion improvement model a number of times according to the L-th test signal sequence.

[0079] Among them, in the i-th iteration, input the L-th test signal sequence into the L-th original predistortion improvement model to obtain the L-th training signal; the L-th training signal is processed by the target power amplifier to obtain the L-th test amplified signal Then calculate the signal difference of the L-th

[0080] If is lower than the preset threshold, then retain the predistortion processing coefficient of the L-th original predistortion improvement model in the i-th iteration as c i , and let

[0081] If is greater than or equal to the preset threshold, update the predistortion processing coefficient of the L-th original predistortion improvement model according to to c i+1 , and perform the (i + 1)-th iteration.

[0082] Among them, the specific method for updating the predistortion processing coefficient includes:

[0083] S301. Construct a cost function Among them, () * represents complex conjugate;

[0084] S302. Let to obtain s mq (L);

[0085] Among them, s mq (L) represents the correlation formula of a mq , and a mq represents the first type of predistortion processing coefficient;

[0086] Take the a mq corresponding to the minimum value of s mq as

[0087] Let to obtain tq (L);

[0088] where t q (L) represents the correlation formula of b q and b q represents the second - type predistortion processing coefficient;

[0089] Take the b corresponding to the minimum value of t q (L) as q as

[0090] Take the partial derivative of J with respect to c and transform to get:

[0091] In the formula, μ represents the iteration step value, and h0 represents the gain coefficient; represents taking the conjugate and then transposing each element of F L Each element is conjugated and then transposed.

[0092] Generally speaking, the purpose of training the original predistortion improvement model is to iteratively obtain the predistortion processing coefficient that meets the preset threshold requirements, realize compensating for signal distortion, and improve the transmission quality of the model.

[0093] Step 3: Synthesize N current predistorted signals into a total predistorted signal X. Where X = [x1, x2,..., x n ; x g is the g - th dimensional signal element in X, and x g = f s , r ∈ [1, n].

[0094] This step is to perform the combination by reverse - processing according to the timing sequence of each data stream. Specifically, Step 3 includes:

[0095] Obtain the position of U g in U' and the position of f s in the N current predistorted signals, and construct a mapping relationship of the position change. Since f s is obtained by predistortion processing of U g , then the position of f s in the N current predistorted signals can be associated with the position of U g in U'.

[0096] Reverse - adjust the signal positions of the N current predistorted signals according to the mapping relationship to obtain the total predistorted signal X. That is to say, adjust f s back to the position of U g in U', traverse all elements of the N current predistorted signals, and generate the total predistorted signal X. Then, like U', X has n signal elements, where x g = fs 。

[0097] Step 4: Process the total predistortion signal X through the target power amplifier to obtain the current amplified signal Y.

[0098] The current amplified signal Y obtained in this way can effectively reduce nonlinear distortion.

[0099] In addition, it should be noted that theoretically, the larger the value of N, the better the effect of the entire method. However, both the running load and the model training load will be too large. Therefore, the value of N should consider the actual computing power. Generally, N takes an integer value between 8 and 13, such as 10.

[0100] Embodiment 2

[0101] This Embodiment 2 discloses a first digital predistortion system based on the dimension-weighted K-nearest neighbor algorithm, which uses the digital predistortion method based on the dimension-weighted K-nearest neighbor algorithm of Embodiment 1.

[0102] The first digital predistortion system based on the dimension-weighted K-nearest neighbor algorithm includes: a signal splitting module, a predistortion processing module, a signal synthesis module, and an amplifier module.

[0103] The signal splitting module is used to split the current original signal U' into N current signal sequences. The predistortion processing module is used to input the L-th current signal sequence into the trained L-th predistortion improvement model to obtain the L-th current predistortion signal F L 。 The signal synthesis module is used to synthesize the N current predistortion signals into a total predistortion signal X. The amplifier module is used to process the total predistortion signal X through the target power amplifier to obtain the current amplified signal Y.

[0104] The trained L-th predistortion improvement model in the first digital predistortion system based on the dimension-weighted K-nearest neighbor algorithm can be pre-trained on the original predistortion improvement model using other training devices, and then the trained L-th predistortion improvement model is placed into the predistortion processing module.

[0105] Of course, the function of model training can also be integrated into the digital predistortion system based on the dimension-weighted K-nearest neighbor algorithm, which is the second digital predistortion system based on the dimension-weighted K-nearest neighbor algorithm. The structure of the second digital predistortion system based on the dimension-weighted K-nearest neighbor algorithm is similar to the first one, the difference is that:

[0106] A model training module is added to train the original predistortion improvement model.

[0107] The signal splitting module is also used to split the original sample signal W’ into N test signal sequences during model training. The amplifier module is also used to process the L-th training signal through the target power amplifier at the i-th iteration of model training to obtain the L-th test amplified signal. The predistortion processing module is also used to input the L-th test signal sequence into the L-th predistortion improvement model at the i-th iteration of model training to obtain the L-th training signal.

[0108] The model training module includes a signal difference calculation module and a model coefficient update module. The signal difference calculation module is used to calculate the signal difference of the L-th path at the i-th iteration of model training. The model coefficient update module is used to when it is greater than or equal to the preset threshold, according to update the predistortion processing coefficient of the L-th predistortion improvement model to c i+1 .

[0109] Embodiment 3

[0110] To illustrate the usage effect of the digital predistortion method based on the dimensionality-weighted K-nearest neighbor algorithm in Embodiment 1 (the proposed digital predistortion method), this Embodiment 3 additionally introduces two cases - the no-predistortion method and the traditional digital predistortion method, and conducts simulation comparisons on the three methods.

[0111] The current original signal is selected to use the same OFDM wave. The target power amplifier is constructed into an amplifier model in a digital simulation manner for use.

[0112] For the first case (the proposed digital predistortion method), the OFDM wave is processed according to the method of Embodiment 1 to obtain the first current amplified signal; for the second case (the no-predistortion method), the OFDM wave is directly processed through the target power amplifier to obtain the second current amplified signal; for the third case (the traditional digital predistortion method), the OFDM wave is processed according to the traditional digital predistortion method to obtain the third current amplified signal.

[0113] First, refer to Figure 3 , which is a comparison chart of the simulation results of the three methods, showing the power spectrum distribution of the three current amplified signals. From Figure 3 it can be seen that the second current amplified signal has the most serious adjacent-channel interference; although the third current amplified signal reduces the adjacent-channel interference to some extent, it is still not as good as the first current amplified signal. It shows that the signal power ratio of the first current amplified signal falling into the adjacent sub-channel frequency band is small and the anti-interference ability is strong.

[0114] Referring again to Table 1, which shows the index comparison of three methods. Among them, the Adjacent Channel Power Ratio (ACPR), Error Vector Magnitude (EVM), and Normalized Mean Square Error (NMSE) are used to measure the advantages and disadvantages of the three methods.

[0115] Table 1 Index Comparison of Three Methods

[0116]

[0117] Among them, the larger the absolute value of ACPR, the better; the smaller the value of EVM, the better; and the smaller the value of NMSE, the better.

[0118] As can be seen from Table 1, compared with the traditional predistortion method, the digital predistortion method proposed in Embodiment 1 has an increase of approximately 3.2 dB in the ACPR value, an increase of 0.48% in the EVM value, and an increase of 0.049889% in the NMSE value, indicating that the present invention has been improved in anti-interference ability, signal quality, and prediction accuracy.

[0119] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0120] The above-described embodiments only represent several implementation manners of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the invention patent should be subject to the appended claims.

Claims

1. A digital predistortion method based on a dimensionality-weighted K-nearest neighbor algorithm, characterized in that, Including the following steps: Step 1, splitting the current original signal U’ into N current signal sequences; where U’ = [U1, U2, …, U n ; U g is the g-th dimensional signal element in U’, g ∈ [1, n]; The method for splitting the current original signal U’ includes: S101, Obtain the maximum value U n among U1 to U max , the minimum value U min , and calculate the interval division length S102, according to ΔU, evenly divide the total interval [U min , U max into N current shunt intervals; Among them, the current shunt interval of the l-th segment is [U min +(l - 1)ΔU, U min +lΔU]; the midpoint of the current shunt interval of the l-th segment is S103, calculate U g , U l 's Euclidean distance D(U g , U l ); Among them, for U g , obtain the minimum value D among D(U g , U1) to D(U g , U N ); min If D min is D(U g , U L ), then classify U g into the current shunt interval of the L-th segment; L ∈ [1, N]; S104, traverse U1 to U n , and classify U1 to U n into the current sub - interval of N segments correspondingly to form an N - path current signal sequence; Among them, the L-th current splitting interval is sorted according to the order of element inclusion to form the L-th current signal sequence, with a total of S elements; Step 2: Input the current signal sequence of the L-th path into the trained L-th pre-distortion improvement model to obtain the current pre-distorted signal F of the L-th path L ; Among them, the predistortion improvement model adopts a predistortion model with memory polynomial cross terms added; F L = [f1, f2, …, f S ; f s is the sth - dimensional signal element in F L and is obtained by processing U g through the trained Lth pre - distortion improvement model; s ∈ [1, S]; Step 3: Synthesize the N current pre-distorted signals into the total pre-distorted signal X; where X = [x1, x2, …, x n ; x g is the g-th dimensional signal element in X, and x g = f s , r ∈ [1, n]; Step 4, processing the total predistortion signal X through the target power amplifier to obtain the current amplified signal Y; In Step 2, the training method of the predistortion improvement model includes: S201, splitting the sample original signal W’ into N test signal sequences; Multiplying the sample original signal W’ by the amplification gain G of the target power amplifier to obtain the desired amplified signal d’; Among them, the sample original signal W’ = [W1, W2, …, W n ; W g is the g-th dimensional signal element in W’; Desired amplified signal d’ = [d1, d2, …, d n ; d g is the g-th dimensional signal element in d’, d g = W g * G; S202, performing several iterations on the L-th original predistortion improvement model according to the L-th test signal sequence; Among them, in the i-th iteration, the L-th test signal sequence is input into the L-th original pre-distortion improvement model to obtain the L-th training signal; the L-th training signal is processed by the target power amplifier to obtain the L-th test amplified signal Then calculate the signal difference of the L-th path If is lower than a preset threshold, then retain the pre-distortion processing coefficient of the L-th original pre-distortion improvement model at the i-th iteration as c i , and let be the pre-distortion processing coefficient used by the trained L-th pre-distortion improvement model; If is greater than or equal to a preset threshold, then according to update the pre-distortion processing coefficient of the L-th original pre-distortion improvement model to c i+1 , and perform the (i + 1)-th iteration.

2. The digital predistortion method based on the dimension weighted K-nearest neighbor algorithm according to claim 1, wherein In Step 1, the expression of D(U g , U l ) is as follows: Among them, ω g represents the weighted value of the g-th dimension.

3. The digital predistortion method based on the dimensionality weighted K-nearest neighbor algorithm according to claim 1, wherein In step two, f s has the following expression: In the formula, 4. The digital predistortion method based on the dimension-weighted K-nearest neighbor algorithm according to claim 1, wherein In S201, the method for splitting the sample original signal W’ includes: S301, Obtain the maximum value W n among W1 to W max , the minimum value W min , and calculate the interval division length S302, according to ΔW, uniformly divide the total interval [W min , W max into N segmented test branch intervals; Among them, the l-th test shunt interval is [W min +(l - 1)ΔW, W min + lΔW]; the midpoint of the l-th test shunt interval is S303, calculate W g , W l 's Euclidean distance D(W g , W l ); Among them, for W g , obtain the minimum value W among D(W g , W1) to D(W g , W N ); min If W min is D(W g , W L ), then classify W g into the L-th shunt interval; L ∈ [1, N]; S304, traverse W1 to W n , and classify W1 to W n into N test branch intervals correspondingly to form an N-channel test signal sequence; Among them, the L-th test splitting interval is sorted according to the order of element inclusion to form the L-th test signal sequence.

5. The digital predistortion method based on the dimension-weighted K-nearest neighbor algorithm according to claim 4, characterized in that S202, according to update the pre-distortion processing coefficient of the L-th pre-distortion improvement model to c i+1 The method includes: S301, construct a cost function where, () * denotes complex conjugate; S302, let obtain s mq (L); where s mq (L) represents a mq correlation expression of a mq indicating the first - type predistortion processing coefficient; Take s mq The a corresponding to the minimum value of (L) mq As Let obtain t q (L); where t q (L) represents b q 's relational expression, and b q represents the second-order predistortion processing coefficient; Take t q The b corresponding to when (L) is the minimum value q As Taking the partial derivative of J with respect to c and transforming, we get: Where μ represents the iterative step value and h0 represents the gain coefficient; denotes taking the conjugate and then transposing for each element of F L ​ 6. The digital predistortion method based on the dimensionality-weighted K-nearest neighbor algorithm according to claim 1, wherein Step 3 includes: Obtain U g Position in U', f s Position in the N-channel current predistorted signal, and construct a mapping relationship of position change; Adjusting the signal positions of the N current predistortion signals in reverse according to the mapping relationship to obtain the total predistortion signal X.

7. A digital predistortion system based on a dimensionality-weighted K-nearest neighbor algorithm, characterized in that, Using the digital predistortion method based on the dimension-weighted K-nearest neighbor algorithm described in any one of claims 1-6; The digital predistortion system based on the dimension-weighted K-nearest neighbor algorithm includes: A signal splitting module, which is used to split the current original signal U’ into N current signal sequences; A predistortion processing module, which is used to input the current signal sequence of the L-th path into the trained L-th predistortion improvement model to obtain the current predistorted signal F of the L-th path L ; A signal synthesis module, which is used to synthesize the N current predistortion signals into the total predistortion signal X; and An amplifier module, which is used to process the total predistortion signal X through the target power amplifier to obtain the current amplified signal Y.

8. The digital predistortion system based on the dimension-weighted K-nearest neighbor algorithm according to claim 7, characterized in that, The signal splitting module is also used to split the sample original signal W’ into N test signal sequences during model training; The amplifier module is further configured to, at the i-th iteration of model training, process the L-th training signal through the target power amplifier to obtain the L-th test amplified signal The predistortion processing module is also used to input the L-th test signal sequence into the L-th predistortion improvement model during the i-th iteration of model training to obtain the L-th training signal; The digital predistortion system based on the dimension-weighted K-nearest neighbor algorithm further includes a model training module; The model training module includes: A signal difference calculation module, which is used to calculate the signal difference of the L-th path during the i-th iteration of model training And A model coefficient update module, which is used to when it is greater than or equal to a preset threshold, according to update the pre-distortion processing coefficient of the L-th pre-distortion improvement model to c i+1 .

9. A readable storage medium, characterized in that, The readable storage medium stores computer program instructions, and when the computer program instructions are read and run by a processor, they execute the steps of the digital predistortion method based on the dimension-weighted K-nearest neighbor algorithm described in any one of claims 1-6.

Images