A Simplified Method of Digital Predistortion Model Based on Greville Method

The coefficients of VS model are simplified through the Greville method, and the problems of instability and poor applicability of VS model coefficients are solved, and the efficiency of model simplification and hardware implementation is realized. It is suitable for different modulation signals and application scenarios, reducing the difficulty of hardware implementation .

CN115396274BActive Publication Date: 2024-11-26THE 54TH RESEARCH INSTITUTE OF CHINA ELECTRONICS TECHNOLOGY GROUP CORPORATION
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211087523.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-09-07
Publication Date
2024-11-26
Estimated Expiration
2042-09-07

Smart Images

  • Figure CN115396274B_ABST
    Figure CN115396274B_ABST
Patent Text Reader

Abstract

The present invention discloses a digital pre-distortion model simplification method based on the Greville method, and belongs to the technical field of digital pre-distortion. The present invention comprises the following steps: collecting the input and output baseband signals of a power amplifier; converting the model coefficient simplification problem into a signal reconstruction problem; constructing an observation matrix; and completing the model coefficient simplification process. The method of the present invention can simplify the coefficients of a Volterra series model, and is suitable for different modulation signals and application scenarios. It can greatly reduce the model coefficients on the basis of ensuring the modeling accuracy and pre-distortion effect, enhance the system stability and the flexibility of the model simplification, and provide a new solution to the problem of simplifying the digital pre-distortion power amplifier model. At the same time, the method of the present invention does not include a matrix inversion operation, which reduces the difficulty of hardware implementation and has engineering significance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of digital predistortion, and in particular to a method for simplifying a digital predistortion model based on a Greville method. Background Art

[0002] Wireless communication technology is evolving towards higher-order modulation and broadband. Issues such as high peak-to-average ratio (PAPR) place higher demands on the linearity of power amplifiers (PAs). Consequently, PA linearization has become a key technology driving the development of wireless communication systems. Digital pre-distortion (DPD) has become a research hotspot due to its robust stability and excellent linearization. Accurately modeling PA models is crucial for DPD. VS models, which effectively account for PA nonlinearities and memory effects, have been widely used and are the most common model type in DPD. However, VS models suffer from rapid growth in model size with increasing order and memory depth, as well as redundancy between basis function vectors. This can lead to unstable coefficient solutions and poor model applicability. Therefore, effectively simplifying VS models is a key research focus in DPD.

[0003] Using a compressed sensing (CS) algorithm to simplify VS models is a common method. It can flexibly prune VS models, effectively reducing the number of coefficients while ensuring modeling accuracy and pre-distortion effects, thereby achieving power amplifier linearization. However, the application of compressed sensing algorithms in the field of digital pre-distortion model simplification has been relatively recent, and the effectiveness of simplified models has been mainly verified through computer simulation or instrument platform verification. Currently, there is no hardware implementation of a simplified digital pre-distortion model based on a compressed sensing algorithm. There are two main reasons for this: first, some algorithms use regularization, backtracking, and other methods to improve reconstruction accuracy, but this also increases the complexity of the algorithm; second, compressed sensing algorithms often involve matrix inversion operations, which makes hardware implementation difficult. Summary of the Invention

[0004] The purpose of the present invention is to simplify the digital pre-distortion model coefficients by using the compressed sensing concept, solve the two major problems faced by the VS type power amplifier model, and simplify the algorithm process and reduce the algorithm complexity. To this end, a digital pre-distortion model simplification method based on the Greville method is proposed. This method can greatly simplify the power amplifier model coefficients on the basis of ensuring the model accuracy and pre-distortion effect, enhance the system stability, and the method is applicable to the VS type model, and adaptively prunes the model coefficients according to different modulation signals and application scenarios, with high flexibility. In addition, the method has a simple process and does not require matrix inversion, which reduces the difficulty of hardware implementation of the algorithm and has high engineering implementation value.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for simplifying a digital predistortion model based on the Greville method comprises the following steps:

[0007] (1) Collecting the input baseband signal and output baseband signal of the power amplifier;

[0008] (2) Convert the model coefficient simplification problem into a signal reconstruction problem;

[0009] (3) Construct the observation matrix;

[0010] (4) Complete the model coefficient simplification process.

[0011] Furthermore, the observation matrix constructed in step (3) is a model basis function matrix composed of the input signal.

[0012] Furthermore, step (4) specifically includes the following steps:

[0013] (401) Input the observation matrix X, observation vector y, and termination condition, and set the output model coefficient estimate to Index set index;

[0014] (402) Initialize the residual r0 = y, support set Index Set Iteration number k = 1;

[0015] (403) Calculate the inner product u of the observation matrix and the residual = |X H ·r0|, take the maximum value in u, the vectors at the corresponding position in X form the set S1, and the vectors at the corresponding positions form the set J1; the superscript H indicates conjugate transpose;

[0016] (404) Update the support set F1 = F0∪S1 and the index set index1 = index0∪J1;

[0017] (405) Update the generalized inverse matrix T1 = (F1 H Formula 1 -1 Formula One H ,coefficient residual The number of iterations k = 2; represents the least squares solution at the first iteration, the superscript -1 represents matrix inversion, and the superscript ^ represents the estimated value;

[0018] (406) Determine whether the termination condition is met, if so, exit, if not, enter the iterative process;

[0019] (407) Start iteration: Calculate the inner product u of the observation matrix and the residual H ·r k-1|, take out the maximum value in u, and the vectors at the corresponding position in X form a set S k , the corresponding positions constitute the set J k ;

[0020] (408) Update support set F k =F k-1 ∪S k , index set index k =index k-1 ∪J k ;

[0021] (409) Set the intermediate variable d k =T k-1 ·S k , c k =S k -F k-1 ·d k ;

[0022] (410) If c k =0, then update the intermediate variable b k H =(1+d k H ·d k ) -1 ·d k H ·T k-1 , if c k ≠0, then update the intermediate variable b k H =(c k H c k ) -1 c k H ;

[0023] (411) Update the generalized inverse matrix Find coefficient estimates Update residual represents the least squares solution at the kth iteration;

[0024] (412) When the termination condition is met or the number of coefficient estimates is greater than the maximum coefficient of the model, the iteration is exited and the output is Index set index, otherwise update k = k + 1 and continue iterating.

[0025] The beneficial effects of the above technical solution of the present invention are:

[0026] 1. The present invention uses compressed sensing technology to simplify the number of coefficients of the VS model, and uses the sparsity of the model itself to complete the sparse process of the model coefficients, thereby reducing the number of coefficients.

[0027] 2. The model simplification method of the present invention can be applied to all VS-type polynomial models and has strong applicability. At the same time, it can significantly reduce the model coefficients while ensuring the modeling accuracy and pre-distortion effect, and improve the stability of the coefficient solution.

[0028] 3. The model simplification method of the present invention has a simple process, does not require matrix inversion, has low hardware implementation difficulty, and is of great value to technical engineering. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Flowchart of the method according to the embodiment of the present invention (hereinafter referred to as GOMP algorithm, namely Greville Orthogonal Matching Pursuit, orthogonal matching pursuit based on Greville method).

[0030] Figure 2 This is a table of predistortion effects of the MP-GOMP model in an embodiment of the present invention.

[0031] Figure 3 3 is a comparison diagram of the AMAM curves before and after pre-distortion of the simplified model in an embodiment of the present invention.

[0032] Figure 4 3 is a comparison diagram of AMPM curves before and after pre-distortion of the simplified model in an embodiment of the present invention.

[0033] Figure 5 3 is a spectrum comparison diagram before and after predistortion of the simplified model in an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The following is a clear and complete description of the technical solutions in the embodiments of the present invention with reference to the accompanying drawings and examples. However, the embodiments described below are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0035] A digital pre-distortion model simplification method based on the Greville method uses the baseband input signal of the power amplifier to construct a basis function matrix according to the power amplifier model, and uses a model simplification algorithm to simplify the model coefficients to obtain a simplified model. Finally, instrument platform testing verifies that the simplified model can significantly reduce the number of model coefficients while ensuring modeling accuracy and pre-distortion effect.

[0036] The specific steps of this method are as follows:

[0037] (1) Collecting the input baseband signal and output baseband signal of the power amplifier;

[0038] (2) Convert the model coefficient simplification problem into a signal reconstruction problem;

[0039] (3) Construct the observation matrix;

[0040] (4) Complete the model coefficient simplification process.

[0041] Wherein, step (4) specifically includes the following steps:

[0042] (401) Input observation matrix X, observation vector y, termination condition, and set the output model coefficient estimate to be Index set index;

[0043] (402) Initialize the residual r0 = y, support set Index Set Iteration number k = 1;

[0044] (403) Calculate the inner product u of the observation matrix and the residual = |X H ·r0|, take out the maximum value in u, the vectors at the corresponding position in X form set S1, and the corresponding positions form set J1;

[0045] (404) Update the support set F1 = F0∪S1 and the index set index1 = index0∪J1;

[0046] (405) Update the generalized inverse T1 = (F1 H Formula 1 -1 Formula One H ,coefficient residual The number of iterations k = 2;

[0047] (406) Determine whether the termination condition is met, exit if it is met, and enter the iterative process if it is not met;

[0048] (407) Start iteration: Calculate the inner product u of the observation matrix and the residual H ·r k-1 |, take out the maximum value in u, and the vectors at the corresponding position in X form a set S k , the corresponding positions constitute the set J k ;

[0049] (408) Update support set F k =F k-1 ∪S k , index set index k =index k-1 ∪Jk ;

[0050] (409)d k =T k-1 ·S k , c k =S k -F k-1 ·d k ;

[0051] (410) If c k =0,b k H =(1+d k H ·d k ) -1 ·d k H ·T k-1 , if c k ≠0, b k H =(c k H c k ) -1 c k H ;

[0052] (411) Updated generalized inverse Find coefficient estimates Update residual

[0053] (412) When the termination condition is met or the number of coefficient estimates is greater than the maximum coefficient of the model, the iteration is exited and the output is If the index set index is not satisfied, update k=k+1 and continue iterating.

[0054] Here's a more specific example:

[0055] A method for simplifying a digital predistortion model based on the Greville method mainly includes the following steps:

[0056] (1) Collecting the input and output baseband signals of the power amplifier:

[0057] (2) Convert the model coefficient simplification problem into a signal reconstruction problem:

[0058] The expression of the VS class model can be expressed as follows

[0059] y=X·a

[0060] Here, y, X, and a represent the model basis function matrix and model coefficients, respectively, consisting of the output and input signals of the power amplifier. Because the coefficients of VS models often vary significantly, this indicates that the nonlinear basis functions corresponding to different coefficients contribute differently to model building. If the basis functions with high contributions are retained and those with low contributions are removed, that is, if the values ​​of the model coefficients approaching zero are set to zero, the loss in modeling accuracy is minimal. Based on this, the coefficient vector can be considered to contain zeros, meaning it is sparse. Therefore, the simplification of the model coefficient a can be viewed as a signal reconstruction problem in compressed sensing. Using reconstruction algorithms, the number of model coefficients can be reduced.

[0061] (3) Construct the observation matrix:

[0062] The coefficient a can be considered to be sparse. It can be obtained from the VS model expression that the output signal y is the observation vector, and the model basis function matrix X composed of the input signal is the observation matrix. The construction of X depends on the selected VS simplified model.

[0063] (4) Complete the model coefficient simplification process:

[0064] The GOMP algorithm is used to reconstruct the signal, that is, to complete the process of simplifying the model coefficients. The flow chart is as follows: Figure 1 The GOMP algorithm updates the support set with each iteration using a greedy approach, continuously approximating the original signal. It does not require the coefficient sparsity to be known in advance, nor does it require setting termination conditions. It can adaptively output results that meet the conditions. Furthermore, the Greville method can use a finite number of iterations to find the generalized inverse of the matrix, eliminating the need for matrix inversion and reducing hardware implementation difficulty. The algorithm steps are as follows:

[0065] (401) Input observation matrix X, observation vector y, termination condition, and set the output model coefficient estimate to be Index set index;

[0066] (402) Initialize the residual r0 = y, support set Index Set Iteration number k = 1;

[0067] (403) Calculate the inner product u of the observation matrix and the residual = |X H ·r0|, take out the maximum value in u, the vectors at the corresponding position in X form set S1, and the corresponding positions form set J1;

[0068] (404) Update the support set F1 = F0∪S1 and the index set index1 = index0∪J1;

[0069] (405) Update the generalized inverse T1 = (F1 H Formula 1 -1 Formula OneH ,coefficient residual The number of iterations k = 2;

[0070] (406) Determine whether the termination condition is met, exit if it is met, and enter the iterative process if it is not met;

[0071] (407) Start iteration: Calculate the inner product u of the observation matrix and the residual H ·r k-1 |, take out the maximum value in u, and the vectors at the corresponding position in X form a set S k , the corresponding positions constitute the set J k ;

[0072] (408) Update support set F k =F k-1 ∪S k , index set index k =index k-1 ∪J k ;

[0073] (409)d k =T k-1 ·S k , c k =S k -F k-1 ·d k ;

[0074] (410) If c k =0,b k H =(1+d k H ·d k ) -1 ·d k H ·T k-1 , if c k ≠0, b k H =(c k H c k ) -1 c k H ;

[0075] (411) Updated generalized inverse Find coefficient estimates Update residual

[0076] (412) When the termination condition is met or the number of coefficient estimates is greater than the maximum coefficient of the model, the iteration is exited and the output is If the index set index is not satisfied, update k=k+1 and continue iterating.

[0077] (5) Test verification

[0078] The effectiveness of the GOMP algorithm was verified using an instrument platform, using the MP full-coefficient model as a comparison. A 16QAM signal with a peak-to-average ratio of 7.8dB and a bandwidth of 25MHz was generated by a computer and downloaded to a vector signal generator. The signal was then amplified by a gallium arsenide power amplifier and passed through an attenuator. Data was collected by a real-time signal analyzer. Finally, the data was sent back to the computer for signal delay calibration, model coefficient extraction, and predistortion signal generation.

[0079] The model coefficients, modeling accuracy and predistortion effect of the MP full-coefficient model are as follows: Figure 2 As shown in the figure, after simplification by the GOMP algorithm, the number of coefficients of the SDDR model is 7, which is 84% ​​less than the 45 coefficients of the SDDR full coefficient model. The comparison of the AMAM curve, AMPM curve and adjacent channel power ratio (ACPR) before and after digital predistortion of the simplified model is shown in the figure. Figure 3 、 4 As shown in Figure 5, it can be seen that the MP model can still ensure the same predistortion effect as the full-coefficient model after the coefficient simplification of the GOMP algorithm.

[0080] In summary, the method of the present invention can simplify the coefficients of the Voltera series (VS) model and is applicable to different modulation signals and application scenarios. It can significantly reduce the model coefficients while ensuring the modeling accuracy and predistortion effect, enhance the system stability and the flexibility of the model simplification, and at the same time, it does not require matrix inversion, which reduces the difficulty of hardware implementation. It has strong engineering application value and provides a new solution to the problems of model simplification and hardware implementation of digital predistortion power amplifiers.

[0081] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown above, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0082] Specific embodiments are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A method for simplifying a digital predistortion model based on the Greville method, characterized in that: The following steps are involved: (1) Collecting the input baseband signal and output baseband signal of the power amplifier; (2) Convert the model coefficient simplification problem into a signal reconstruction problem; (3) constructing a measurement matrix, which is a model basis function matrix composed of input signals; (4) Complete the model coefficient simplification process; The specific steps include: (401) Input the observation matrix X, observation vector y, and termination condition, and set the output model coefficient estimate to Index set index; (402) Initialize the residual r0 = y, support set Index Set Iteration number k = 1; (403) Calculate the inner product u of the observation matrix and the residual = X H ·r0, take out the maximum value in u, the vectors at the corresponding position in X form the set S1, and the vectors at the corresponding positions form the set J1; the superscript H indicates the conjugate transpose; (404) Update support set F1 = F0∪S1, index set index1 = index0∪J1; (405) Update the generalized inverse matrix T1 = (F1 H ·F1) -1 ·F1 H ,coefficient Residual The number of iterations k = 2; represents the least squares solution at the first iteration, the superscript -1 represents matrix inversion, and the superscript ^ represents an estimated value; (406) Determine whether the termination condition is met, if so, exit, if not, enter the iteration process; (407) Iteration starts: Calculate the inner product u=X of the observation matrix and the residual H ·r k-1 , take out the maximum value in u, and the vectors at the corresponding position in X form a set S k , the corresponding positions constitute a set J k ; (408) Update support set F k =F k-1 ∪S k , index set index k =index k-1 ∪J k ; (409) Set the intermediate variable d k =T k-1 ·S k , c k =S k -F k-1 ·d k ; (410) If c k =0, then update the intermediate variable b k H =(1+d k H ·d k ) -1 ·d k H ·T k-1 , if c k ≠0, then update the intermediate variable b k H =(c k H ·c k ) -1 ·c k H ; (411) Update the generalized inverse matrix Find the coefficient estimates Update residual represents the least squares solution at the kth iteration; (412) When the termination condition is met or the number of coefficient estimates is greater than the maximum coefficient of the model, the iteration is exited and the output is Index set index, otherwise update k=k+1 and continue iterating.