Method and device for updating lookup table of predistorter and predistortion processing system
By collecting observed values within the preset period and updating some feature matrices, the problem of high consumption of least squares calculation resources is solved, and the efficient update of the predistorter lookup table is realized, which avoids system task blockage and improves the RF transceiver chip. performance.
Patent Information
- Application Number
- CN202411019653.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-15
AI Technical Summary
In the prior art, when determining and optimizing the model coefficients of the predistortion model through the least squares method, computing resources are consumed greatly, affecting the embedded system performance of the RF transceiver chip, and easily causing system tasks to be blocked.
Observations are collected within the preset period, the eigenvectors are calculated and some feature matrices are updated through sub-matrixes, and the solutions of linear equations are finally updated to obtain model coefficients, and then the predistorter lookup table is updated.
It greatly saves computing resources, avoids the risk of system tasks blocking, and improves the performance of RF transceiver chips.
Smart Images

Figure CN118971810A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of signal processing, and in particular to a predistorter lookup table updating method, device and predistortion processing system. Background Art
[0002] Digital pre-distortion is used to improve the linearity and efficiency of RF power amplifiers and is often used in RF transceiver chips. Figure 1 As shown, the transmission channel of the RF transceiver chip is inserted into the predistorter, and the input signal u(n) of the RF transceiver chip is predistorted, and the predistorter outputs the signal x(n) to the power amplifier PA after digital / analog conversion, and the power amplifier PA outputs the signal y(n). Among them, the lookup table is a commonly used predistortion processing method in the predistorter. The predistorter lookup table corrects the nonlinear behavior of the power amplifier by storing a series of compensation coefficients, thereby achieving the purpose of compensating the nonlinearity of the power amplifier and reducing the distortion from the signal x(n) to the signal y(n).
[0003] The compensation coefficients stored in the predistorter lookup table are closely related to the model coefficients of the predistortion model used by the predistorter. In addition, to ensure the stability and accuracy of digital predistortion, the predistortion model must be updated in real time according to changes in the environment and working conditions. The model coefficients are constantly changing, and the compensation coefficients stored in the predistorter lookup table also need to be updated.
[0004] At present, for nonlinear pre-distortion models, adaptive algorithms are generally used to determine and optimize model coefficients, including least squares algorithm, recursive least squares algorithm, least mean square algorithm, etc. Among them, the least squares algorithm has high solution accuracy and is widely used in the field of digital pre-distortion. The key steps of the least squares method include matrix inversion and matrix multiplication operations. The size of the data matrix is determined by the product of the number of feature quantities and the length of the observation value. However, if the data matrix is large, on the one hand, it will consume a lot of CPU memory, and on the other hand, intensive matrix operations are also a challenge for the embedded system of the RF transceiver chip, which can easily cause system task blocking. Summary of the invention
[0005] In view of the deficiencies in the prior art, the present invention provides a method for updating a predistorter lookup table, which solves the problem in the prior art of determining and optimizing the model coefficients of a predistortion model by the least squares method, that is, the large consumption of computing resources required for updating the predistorter lookup table, which affects the performance of an embedded system using a radio frequency transceiver chip.
[0006] According to a first aspect of an embodiment of the present invention, a predistorter lookup table updating method is provided, which is applied to a predistortion processing system including a processor, a predistorter, and a power amplifier. The predistorter lookup table updating method includes:
[0007] At the beginning of each preset cycle, the processor obtains the predistorter output signal, the power amplifier output signal, and the model coefficients of the predistortion model in the previous preset cycle and constructs a linear equation group, obtains a data matrix of the linear equation group, and represents the solution of the linear equation group through a first intermediate matrix and a second intermediate matrix;
[0008] When a new preset cycle starts, observation values are collected; wherein the observation values collected at this time include the predistorter output signal collected at this time and the power amplifier output signal collected at this time;
[0009] After each acquisition in the current preset period, the feature vector corresponding to the observation value collected at that time in the data matrix is calculated according to the observation value collected at that time, so as to obtain the intermediate feature vector of the acquisition at that time;
[0010] Calculate the first sub-matrix and the second sub-matrix according to the intermediate feature vector collected at the time and the observation value collected at the time;
[0011] Update the first intermediate matrix and the second intermediate matrix by using the first submatrix and the second submatrix calculated by the current acquisition;
[0012] When the preset number of acquisitions is reached within the current preset period, the solution of the linear equation group is calculated using the first intermediate matrix and the second intermediate matrix that were updated most recently to obtain the model coefficients of the current preset period;
[0013] The predistorter lookup table is updated according to the model coefficients of the current preset period, wherein the updated predistorter lookup table obtained in the current preset period is used to instruct the predistorter to perform predistortion processing on the input signal of the predistortion processing system in the future preset period.
[0014] Optionally, calculating the first sub-matrix and the second sub-matrix according to the intermediate feature vector collected at the time and the observation value collected at the time includes:
[0015] The calculation part including the intermediate eigenvector in the first intermediate matrix is extracted to obtain a first submatrix; the calculation part including the intermediate eigenvector and the observation value collected at that time is extracted to obtain a second submatrix.
[0016] Optionally, the first intermediate matrix and the second intermediate matrix are updated by the first submatrix and the second submatrix calculated by the intermediate eigenvector collected this time, and the formula is:
[0017]
[0018] Among them, Q is the first intermediate matrix, Q i is the first submatrix calculated for the current acquisition, U is the second intermediate matrix, and Ui The second sub-matrix calculated for the current acquisition.
[0019] Optionally, before calculating the feature vector corresponding to the observation value collected at the time in the data matrix according to the observation value collected at the time, the method further comprises:
[0020] Perform time delay, phase and gain alignment processing on the observation values collected at that time.
[0021] Optionally, after calculating the feature vector corresponding to the observation value collected at the time in the data matrix according to the observation value collected at the time, and obtaining the intermediate feature vector collected at the time, the method further comprises:
[0022] The intermediate feature vector collected this time is written into the hardware accelerator via DMA.
[0023] Optionally, when the preset acquisition times are reached, solving the linear equations with the last updated first intermediate matrix and the second intermediate matrix to obtain the model coefficients includes:
[0024] Invoke the hardware accelerator to calculate and update the first intermediate matrix and the second intermediate matrix according to the intermediate feature vector written last time, and obtain the first intermediate matrix and the second intermediate matrix updated last time;
[0025] The linear equations are solved using the most recently updated first intermediate matrix and second intermediate matrix to obtain the model coefficients of the current signal acquisition period.
[0026] Optionally, updating the predistorter lookup table according to the model coefficients of the current signal acquisition period includes:
[0027] Decomposing the model coefficients of the current signal acquisition period using Cholesky to obtain compensation coefficients;
[0028] The predistorter lookup table is updated by the compensation coefficients.
[0029] A second aspect provides a predistorter lookup table updating device, which is applied to a predistortion processing system including a processor, a predistorter, and a power amplifier, wherein the predistorter lookup table updating device includes:
[0030] A model splitting module is used to obtain, at the beginning of each preset cycle, the predistorter output signal, the power amplifier output signal, and the model coefficients of the predistortion model in the previous preset cycle and construct a linear equation group, obtain a data matrix of the linear equation group, and represent the solution of the linear equation group through a first intermediate matrix and a second intermediate matrix;
[0031] An observation value acquisition module is used to collect observation values at the beginning of a new preset cycle; wherein the observation values collected at that time include the predistorter output signal collected at that time and the power amplifier output signal collected at that time;
[0032] The intermediate feature vector calculation module is used to calculate the feature vector corresponding to the observation value collected at that time in the data matrix according to the observation value collected at that time after each collection in the current preset period, so as to obtain the intermediate feature vector collected at that time;
[0033] A submatrix calculation module, used for calculating a first submatrix and a second submatrix according to the intermediate feature vectors collected at the time and the observation values collected at the time;
[0034] An intermediate matrix updating module, used for updating the first intermediate matrix and the second intermediate matrix by using the first sub-matrix and the second sub-matrix calculated by the current acquisition;
[0035] A model coefficient calculation module, used to calculate the solution of the linear equation group using the last updated first intermediate matrix and the second intermediate matrix to obtain the model coefficient of the current preset period when the preset number of acquisitions is reached within the current preset period;
[0036] The predistorter lookup table updating module is used to update the predistorter lookup table according to the model coefficients of the current preset period, wherein the updated predistorter lookup table obtained in the current preset period is used to instruct the predistorter to perform predistortion processing on the input signal of the input predistortion processing system in the future preset period.
[0037] A third aspect provides a predistortion processing system, including:
[0038] A processor, configured to obtain, at the beginning of each preset cycle, a predistorter output signal, a power amplifier output signal, and a model coefficient of a predistortion model in a previous preset cycle and construct a linear equation system, obtain a data matrix of the linear equation system, and represent a solution of the linear equation system by a first intermediate matrix and a second intermediate matrix;
[0039] When a new preset cycle starts, observation values are collected; wherein the observation values collected at this time include the predistorter output signal collected at this time and the power amplifier output signal collected at this time;
[0040] After each acquisition in the current preset period, the feature vector corresponding to the observation value collected at that time in the data matrix is calculated according to the observation value collected at that time, so as to obtain the intermediate feature vector of the acquisition at that time;
[0041] Calculate the first sub-matrix and the second sub-matrix according to the intermediate feature vector collected at the time and the observation value collected at the time;
[0042] Update the first intermediate matrix and the second intermediate matrix by using the first submatrix and the second submatrix calculated by the current acquisition;
[0043] When the preset number of acquisitions is reached within the current preset period, the solution of the linear equation group is calculated using the first intermediate matrix and the second intermediate matrix that were updated most recently to obtain the model coefficients of the current preset period;
[0044] updating the predistorter lookup table according to the model coefficients of the current preset period;
[0045] a predistorter, instructing the predistorter to perform predistortion processing on a target input signal input to the predistortion processing system in a future preset period according to the updated predistorter lookup table obtained in the current preset period, and outputting a target predistortion signal;
[0046] A power amplifier, configured to receive the target predistortion signal and amplify the target predistortion signal
[0047] The technical principle of the present invention is: first, a linear equation group is constructed for the predistortion model used by the predistorter, and the linear equation group includes the predistorter output signal, the power amplifier output signal, and the model coefficient of the predistortion model, wherein the predistorter output signal and the power amplifier output signal are observation values collected in each preset period, and the model coefficient of the predistortion model is the solution of the linear equation group. The present invention also uses the least squares method to calculate the solution of the linear equation group, and therefore, the solution of the linear equation group is represented by the first intermediate matrix and the second intermediate matrix. However, in the solution process, each time an observation value is collected, the first intermediate matrix and the second intermediate matrix are updated once, which is equivalent to updating a part of the characteristic matrix, obtaining a part of the calculation results of the first intermediate matrix, and a part of the calculation results of the second intermediate matrix, and when the preset number of collections is reached in the current preset period, the first intermediate matrix and the second intermediate matrix finally updated are the same as the first intermediate matrix and the second intermediate matrix used in the traditional least squares method calculation. The first intermediate matrix and the second intermediate matrix finally updated can solve the linear equation group to obtain the model coefficients of the current signal collection period, and then can update the predistorter lookup table according to the model coefficients of the current preset period.
[0048] Compared with the prior art, the present invention has the following beneficial effects: the pre-distorter lookup table updating method provided by the embodiment of the present invention calculates the eigenvector after collecting the observation values once, and splits the first intermediate matrix and the second intermediate matrix originally used for solving the model coefficients by the least squares method according to the observation values collected at that time and the calculated intermediate eigenvector, that is, calculates the first sub-matrix and the second sub-matrix, and gradually updates the first intermediate matrix and the second intermediate matrix through the first sub-matrix and the second sub-matrix. The first intermediate matrix and the second intermediate matrix finally updated are equivalent to the first intermediate matrix and the second intermediate matrix originally used for solving the model coefficients by the least squares method. In the above process, the intensive matrix operations are dispersed. At the same time, only the eigenvector is calculated instead of the entire eigenmatrix for each collection, which greatly saves computing resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] Figure 1 This is a schematic diagram of the composition structure of the radio frequency transceiver chip;
[0050] Figure 2 Schematic diagram of the implementation flow of the predistorter lookup table updating method according to an embodiment of the present invention;
[0051] Figure 3 Schematic diagram of the structure of the pre-distortion processing system according to an embodiment of the present invention;
[0052] Figure 4 Schematic diagram of the algorithm logic for solving the model coefficients using the traditional least squares method;
[0053] Figure 5 A schematic diagram of an algorithm logic for calculating model coefficients in a predistorter lookup table updating method according to an embodiment of the present invention;
[0054] Figure 6 A schematic diagram of the structure of a predistorter lookup table updating method and device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0055] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0056] See also Figure 2 and Figure 3 ,like Figure 2 As shown, the embodiment of the present invention proposes a predistorter lookup table updating method, which is applied to Figure 3 The predistortion processing system 30 shown includes a processor 31, a predistorter 32 and a power amplifier 33. The embodiment of the present invention optimizes the steps of solving the model coefficients by the least squares algorithm through the processor, reduces the consumption of computing resources, disperses the intensive matrix operations, and avoids the risk of blocking the system tasks.
[0057] It is understandable that the predistortion processing system in the embodiment of the present invention is not limited to the above-mentioned processor, predistorter and power amplifier, but also includes other functional devices, such as Figure 1 The digital-to-analog converter and low-pass filter in the RF transceiver chip are shown.
[0058] See also Figure 2 The predistorter lookup table updating method provided by the embodiment of the present invention includes but is not limited to the following steps:
[0059] S1. At the beginning of each preset cycle, the processor obtains the predistorter output signal, the power amplifier output signal, and the model coefficients of the predistortion model in the previous preset cycle and constructs a linear equation system, obtains a data matrix of the linear equation system, and represents the solution of the linear equation system through a first intermediate matrix and a second intermediate matrix.
[0060] In the embodiment of the present invention, the preset period is set according to the input signal of the input predistortion processing system.
[0061] It can be understood that if the preset period of the above step S1 is the first preset period, the predistorter output signal, the power amplifier output signal, and the model coefficients of the predistortion model are directly recorded.
[0062] Assuming that the predistorter uses the GMP DPD model, the GMP DPD model is expressed as:
[0063]
[0064] Where x(n) represents the predistorter output signal, y(n) represents the power amplifier output signal, i represents the memory term, j represents the cross-memory term, k represents the nonlinear order, and c represents the ijk Represents the model coefficients.
[0065] Based on the GMPDPD model and the above step S1, the following linear equations can be constructed:
[0066]
[0067] Among them, F ij is a eigenvector satisfying F ij =f(x i , c j ), use F X The characteristic matrix of the data matrix of equation (2) above is:
[0068] Y=F X c(3)
[0069] in,
[0070]
[0071] In the above equations (1) to (4), the predistorter output signal and the power amplifier output signal are obtained by observation, so x i and i As known numbers, the solution of the above linear equation system (2) represents the model coefficient, that is, c in the above equation (3).
[0072] The key steps of solving the model coefficients by the traditional least squares method include matrix inversion and matrix multiplication operations. The embodiment of the present invention also calculates the solution of the linear equation system by matrix inversion and matrix multiplication operations, that is, the solution of the linear equation system is represented by the first intermediate matrix and the second intermediate matrix. Assuming that the first intermediate matrix is Q, the second intermediate matrix is U, and the model coefficient is c, the mathematical expressions of the first intermediate matrix and the second intermediate matrix are:
[0073] Q=F X H F X (5);
[0074] U=F X H Y(6);
[0075] The mathematical expression of the least squares method for solving the model coefficients is:
[0076] c=Q -1 U(7).
[0077] Among them, F X represents the characteristic matrix of the data matrix of the above formula (2), and Y represents y1, ..., y i ,...,y n .
[0078] The predistorter lookup table updating method provided by the embodiment of the present invention optimizes the process of solving the model coefficients by using the least square method through the following steps S21 to S26.
[0079] S21. When a new preset cycle begins, collect observation values.
[0080] In the above step S21, observation values are collected multiple times in each preset period until the preset number of collections is reached. In the embodiment of the present invention, the observation values collected at that time include the predistorter output signal collected at that time and the power amplifier output signal collected at that time.
[0081] S22. After each acquisition in the current preset period, the feature vector corresponding to the observation value collected at that time in the data matrix is calculated according to the observation value collected at that time, so as to obtain the intermediate feature vector of the acquisition at that time.
[0082] In the above, Fij represents the eigenvector of the data matrix based on the linear equation system (2), F ij =f(x i , c j ). Assume that in the above step S22, the observation values collected at this time include a predistorter input signals x i (i=1,2,3,...,a), then we get a eigenvectors F ij (i=1, 2, 3, ..., a), the intermediate feature vectors collected this time include a feature vectors F ij , (i=1, 2, 3, ..., a), the characteristic matrix F of the data matrix belonging to the above formula (2) X part of.
[0083] S23, calculating the first sub-matrix and the second sub-matrix according to the intermediate feature vectors collected this time and the observation values collected this time.
[0084] In the above step S23, the first sub-matrix is a part of the first intermediate matrix, and the second sub-matrix is a part of the second intermediate matrix. The implementation steps of the above step 23 are:
[0085] The calculation part including the intermediate eigenvector in the first intermediate matrix is extracted to obtain a first submatrix; the calculation part including the intermediate eigenvector and the observation value collected at that time is extracted to obtain a second submatrix.
[0086] A preferred solution is to extract the calculation part including the intermediate eigenvector and the power amplifier output signal collected at that time from the second intermediate matrix to obtain the second sub-matrix.
[0087] S24: Update the first intermediate matrix and the second intermediate matrix by using the first sub-matrix and the second sub-matrix calculated during the current acquisition.
[0088] The above step S24 is expressed by the formula:
[0089]
[0090]
[0091] Among them, Q is the first intermediate matrix, Q i is the first submatrix calculated for the current acquisition, U is the second intermediate matrix, and U i The second sub-matrix calculated for the current acquisition.
[0092] The embodiment of the present invention further illustrates the relationship between the first sub-matrix and the first intermediate matrix, and the relationship between the second sub-matrix and the second intermediate matrix through detailed data.
[0093] For example, formula (5) is expanded to obtain
[0094]
[0095] Expanding formula (6), we get
[0096]
[0097] make
[0098]
[0099] get
[0100]
[0101] make
[0102]
[0103] get
[0104]
[0105] In practical applications, the traditional square multiplication method for solving the model coefficients needs to calculate the eigenvector through all the observation values obtained in the period to obtain a complete eigenmatrix. Through the above steps S23 and S24, the embodiment of the present invention solves the first submatrix and the second submatrix for each group of observation values and updates the first intermediate matrix and the second intermediate matrix, which significantly reduces the consumption of system memory during the solution process and disperses the dense matrix operations on the time axis, thereby avoiding the risk of system task blocking.
[0106] It can be understood that if it is the first collection, the first intermediate matrix updated by the observation values collected at that time and the intermediate feature vectors collected at that time is equal to the first submatrix, and the second intermediate matrix updated by the observation values collected at that time and the intermediate feature vectors collected at that time is equal to the second submatrix.
[0107] In one embodiment, the above steps S23 and S24 can be implemented by a hardware accelerator, and the above step S22 calculates the feature vector corresponding to the observation value collected at the time in the data matrix according to the observation value collected at the time, and after obtaining the intermediate feature vector collected at the time, includes:
[0108] The intermediate feature vector collected this time is written into the hardware accelerator through DMA (Direct Memory Access).
[0109] S25. When the preset number of acquisitions is reached within the current preset period, the solution of the linear equation group is calculated using the first intermediate matrix and the second intermediate matrix that were updated most recently to obtain the model coefficients of the current preset period.
[0110] In the above steps S21 to S24, each time an observation value is collected, the first intermediate matrix and the second intermediate matrix are updated once, which is equivalent to updating part of the characteristic matrix, obtaining part of the calculation results of the first intermediate matrix, and part of the calculation results of the second intermediate matrix. Therefore, when the preset number of collections is reached within the current preset period, the first intermediate matrix and the second intermediate matrix finally updated are the same as the first intermediate matrix and the second intermediate matrix used in the traditional least squares calculation.
[0111] After the above step S22, the updating of the first intermediate matrix and the updating of the second intermediate matrix of step S23 and step S24 are implemented in the hardware accelerator. When the preset acquisition times are reached, the hardware accelerator outputs the last updated first intermediate matrix and the second intermediate matrix. When the preset acquisition times are reached, the linear equations are solved with the last updated first intermediate matrix and the second intermediate matrix to obtain the model coefficients, including:
[0112] Invoke the hardware accelerator to calculate and update the first intermediate matrix and the second intermediate matrix according to the intermediate feature vector written last time, and obtain the first intermediate matrix and the second intermediate matrix updated last time;
[0113] The linear equations are solved using the most recently updated first intermediate matrix and second intermediate matrix to obtain the model coefficients of the current signal acquisition period.
[0114] S26. Update the predistorter lookup table according to the model coefficients of the current preset period.
[0115] The updated predistorter lookup table obtained in the current preset period is used to instruct the predistorter to perform predistortion processing on the input signal of the predistortion processing system in the future preset period.
[0116] It should be noted that the model coefficient is not equal to the compensation coefficient of the predistorter lookup table. The embodiment of the present invention uses Cholesky decomposition of the model coefficient of the current signal acquisition period to obtain the compensation coefficient; the predistorter lookup table is updated using the compensation coefficient.
[0117] In one embodiment, before updating the predistorter lookup table by using the compensation coefficient, the method includes:
[0118] Verify the model coefficients of the current signal acquisition cycle;
[0119] The verified model coefficients are stored in the predistorter lookup table.
[0120] Exemplary, validation model coefficients include:
[0121] calculating a coefficient of determination of the predistortion model using the model coefficients;
[0122] When the determination coefficient is within a preset range, the model coefficient verification is passed.
[0123] The embodiment of the present invention also compares the provided predistorter lookup table updating method with the traditional least squares method for solving the model coefficients. Figure 4 FIG. 1 is a logic diagram of a predistorter lookup table updating method provided by an embodiment of the present invention, showing the steps of solving the model coefficients in a preset cycle, namely, the above-mentioned steps S21 to S26. Figure 5 As shown, it is a logical schematic diagram of the traditional least squares method for solving the model coefficients, which also shows the steps of solving the model coefficients in one cycle. Figure 5 In the above process, the feature matrix is updated after each observation is collected. Assuming that the observation is collected N times in a period, the feature matrix will be updated N times, and finally a feature matrix of a larger magnitude is obtained. Then, the first intermediate matrix and the second intermediate matrix are calculated based on this feature matrix of a larger magnitude, so as to calculate the model coefficients. In the above process, a large memory is required to store the feature matrix, and a large amount of computing resources will be concentrated in the calculation of the first intermediate matrix and the second intermediate matrix. Figure 4 As shown, the pre-distorter lookup table updating method provided by the embodiment of the present invention calculates the eigenvector after collecting the observation values once, and splits the first intermediate matrix and the second intermediate matrix originally used for solving the model coefficients by the least squares method according to the observation values collected at that time and the calculated intermediate eigenvector, that is, calculates the first sub-matrix and the second sub-matrix, and gradually updates the first intermediate matrix and the second intermediate matrix through the first sub-matrix and the second sub-matrix. The first intermediate matrix and the second intermediate matrix finally updated are equivalent to the first intermediate matrix and the second intermediate matrix originally used for solving the model coefficients by the least squares method. In the above process, the intensive matrix operations are dispersed. At the same time, only the eigenvector is calculated instead of the entire eigenmatrix for each acquisition, which greatly saves computing resources.
[0124] As shown in Table 1 below, Figure 5 The computational resources required for the traditional least squares method of solving the model coefficients are shown.
[0125]
[0126]
[0127] It can be seen that dense matrix operations have high requirements on the CPU's computing power. At the same time, embedded systems need to take into account other real-time tasks at the same time. Therefore, the dense matrix operations required for traditional least squares solution may cause system task blocking. Figure 2 and Figure 4In the predistorter lookup table updating method provided by the embodiment of the present invention, when solving the model coefficients, if the predistortion model is fixed (the number of feature quantities remains unchanged), the required memory does not increase with the increase of the length of the observation data, and only the currently collected observation values are used to calculate the feature vector, and only the row of intermediate feature vectors corresponding to the current observation value needs to be saved in the feature matrix. For example, Figure 2 and Figure 4 In the method shown, the memory allocation of the feature matrix changes as shown in Table 2:
[0128] Memory requirements Before optimization After optimization <![CDATA[Feature matrix F X > M×N M×1
[0129] Taking M=96, N=32k as an example, the memory required by the optimized feature matrix F will be reduced by 3040k (98.9%).
[0130] like Figure 6 As shown, the embodiment of the present invention further provides a predistorter lookup table updating device 60, which can be applied to Figure 3 As shown, the predistortion processing system 30 includes a processor 31, a predistorter 32, and a power amplifier 33. In a preferred implementation, the predistorter lookup table updating device 60 is a virtual device of the processor 31, which includes:
[0131] The model splitting module 61 is used to obtain the predistorter output signal, the power amplifier output signal, and the model coefficients of the predistortion model in the previous preset cycle at the beginning of each preset cycle, and construct a linear equation system to obtain a data matrix of the linear equation system, and represent the solution of the linear equation system through a first intermediate matrix and a second intermediate matrix;
[0132] The observation value acquisition module 62 is used to acquire observation values at the beginning of a new preset cycle; wherein the observation values acquired at that time include the predistorter output signal acquired at that time and the power amplifier output signal acquired at that time;
[0133] The intermediate feature vector calculation module 63 is used to calculate the feature vector corresponding to the observation value collected at the time in the data matrix according to the observation value collected at the time after each collection in the current preset period, so as to obtain the intermediate feature vector collected at the time;
[0134] A submatrix calculation module 64, used to calculate a first submatrix and a second submatrix according to the intermediate feature vectors collected at the time and the observation values collected at the time;
[0135] An intermediate matrix updating module 65, configured to update the first intermediate matrix and the second intermediate matrix by using the first sub-matrix and the second sub-matrix calculated by the current acquisition;
[0136] A model coefficient calculation module 66 is used to calculate the solution of the linear equation group using the last updated first intermediate matrix and the second intermediate matrix to obtain the model coefficient of the current preset period when the preset number of acquisitions is reached within the current preset period;
[0137] The predistorter lookup table updating module 67 is used to update the predistorter lookup table according to the model coefficients of the current preset period, wherein the updated predistorter lookup table obtained in the current preset period is used to instruct the predistorter to perform predistortion processing on the input signal of the predistortion processing system in the future preset period.
[0138] The embodiment of the present invention also Figure 3 The predistortion processing system 30 shown in FIG. 1 is described in detail. The predistortion processing system 30 includes a processor 31, a predistorter 32 and a power amplifier 33, wherein:
[0139] The processor 31 is used to obtain, at the beginning of each preset cycle, the predistorter output signal x(n), the power amplifier output signal y(n), and the model coefficients of the predistortion model in the previous preset cycle and construct a linear equation system, obtain a data matrix of the linear equation system, and represent the solution of the linear equation system through a first intermediate matrix and a second intermediate matrix;
[0140] When a new preset cycle starts, observation values are collected; wherein the observation values collected at this time include the predistorter output signal collected at this time and the power amplifier output signal collected at this time;
[0141] After each acquisition in the current preset period, the feature vector corresponding to the observation value collected at that time in the data matrix is calculated according to the observation value collected at that time, so as to obtain the intermediate feature vector of the acquisition at that time;
[0142] Calculate the first sub-matrix and the second sub-matrix according to the intermediate feature vector collected at the time and the observation value collected at the time;
[0143] Update the first intermediate matrix and the second intermediate matrix by using the first submatrix and the second submatrix calculated by the current acquisition;
[0144] When the preset number of acquisitions is reached within the current preset period, the solution of the linear equation group is calculated using the first intermediate matrix and the second intermediate matrix that were updated most recently to obtain the model coefficients of the current preset period;
[0145] updating the predistorter lookup table according to the model coefficients of the current preset period;
[0146] The predistorter 32 instructs the predistorter to perform predistortion processing on a target input signal u(n) input to the predistortion processing system in a future preset period according to the updated predistorter lookup table obtained in the current preset period, and outputs a target predistortion signal x(n);
[0147] The power amplifier 33 is used to receive the target predistortion signal x(n), amplify the target predistortion signal, and output a signal y(n).
[0148] It should be noted that the predistortion processing system provided in the embodiment of the present invention is an embedded system, which can be applied to different signal processing requirements, such as signal transmission in intelligent communication systems, audio signal processing in audio equipment, etc. The embodiment of the present invention does not limit its application.
[0149] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0150] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.
Claims
1. A method for updating a predistorter lookup table, characterized in that: Applied to a predistortion processing system including a processor, a predistorter, and a power amplifier, the predistorter lookup table updating method includes: At the beginning of each preset cycle, the processor obtains the predistorter output signal, the power amplifier output signal, and the model coefficients of the predistortion model in the previous preset cycle and constructs a linear equation system to obtain a data matrix of the linear equation system, and represents the solution of the linear equation system through a first intermediate matrix and a second intermediate matrix; When a new preset cycle starts, observation values are collected; wherein the observation values collected at this time include the predistorter output signal collected at this time and the power amplifier output signal collected at this time; After each acquisition in the current preset period, the feature vector corresponding to the observation value collected at that time in the data matrix is calculated according to the observation value collected at that time, so as to obtain the intermediate feature vector of the acquisition at that time; Calculate the first sub-matrix and the second sub-matrix according to the intermediate feature vector collected at the time and the observation value collected at the time; Update the first intermediate matrix and the second intermediate matrix by using the first submatrix and the second submatrix calculated by the current acquisition; When the preset number of acquisitions is reached within the current preset period, the solution of the linear equation group is calculated using the first intermediate matrix and the second intermediate matrix that were updated most recently to obtain the model coefficients of the current preset period; The predistorter lookup table is updated according to the model coefficients of the current preset period, wherein the updated predistorter lookup table obtained in the current preset period is used to instruct the predistorter to perform predistortion processing on the input signal of the predistortion processing system in the future preset period.
2. The predistorter lookup table updating method according to claim 1, characterized in that: The first sub-matrix and the second sub-matrix are calculated according to the intermediate feature vector collected at the time and the observation value collected at the time, including: The calculation part including the intermediate eigenvector in the first intermediate matrix is extracted to obtain a first submatrix; the calculation part including the intermediate eigenvector and the observation value collected at that time is extracted to obtain a second submatrix.
3. The predistorter lookup table updating method according to claim 2, characterized in that: The first intermediate matrix and the second intermediate matrix are updated by the first submatrix and the second submatrix calculated by the intermediate eigenvector collected this time, and the formula is: Among them, Q is the first intermediate matrix, Q i is the first submatrix calculated for the current acquisition, U is the second intermediate matrix, and U i The second sub-matrix calculated for the current acquisition.
4. The method for updating the predistorter lookup table according to any one of claims 1 to 3, characterized in that: Before calculating the characteristic vector corresponding to the observation value collected at the time in the data matrix according to the observation value collected at the time, it includes: Perform time delay, phase and gain alignment processing on the observation values collected at that time.
5. The method for updating the predistorter lookup table according to any one of claims 1 to 3, characterized in that: After calculating the characteristic vector corresponding to the observation value collected at the time in the data matrix according to the observation value collected at the time, and obtaining the intermediate characteristic vector collected at the time, the method includes: The intermediate feature vector collected this time is written into the hardware accelerator via DMA.
6. The predistorter lookup table updating method according to claim 5, characterized in that: When the preset acquisition times are reached, solving the linear equations with the last updated first intermediate matrix and the second intermediate matrix to obtain the model coefficients includes: Invoke the hardware accelerator to calculate and update the first intermediate matrix and the second intermediate matrix according to the intermediate feature vector written last time, and obtain the first intermediate matrix and the second intermediate matrix updated last time; The linear equations are solved using the most recently updated first intermediate matrix and second intermediate matrix to obtain the model coefficients of the current signal acquisition period.
7. The predistorter lookup table updating method according to claim 1, characterized in that: The predistorter lookup table is updated according to the model coefficients of the current signal acquisition period, including: Decomposing the model coefficients of the current signal acquisition period using Cholesky to obtain compensation coefficients; The predistorter lookup table is updated by the compensation coefficients.
8. A predistorter lookup table updating device, characterized in that: Applicable to a predistortion processing system including a processor, a predistorter, and a power amplifier, the predistorter lookup table updating device comprising: A model splitting module is used to obtain, at the beginning of each preset cycle, the predistorter output signal, the power amplifier output signal, and the model coefficients of the predistortion model in the previous preset cycle and construct a linear equation group, obtain a data matrix of the linear equation group, and represent the solution of the linear equation group through a first intermediate matrix and a second intermediate matrix; An observation value acquisition module is used to collect observation values at the beginning of a new preset cycle; wherein the observation values collected at that time include the predistorter output signal collected at that time and the power amplifier output signal collected at that time; The intermediate feature vector calculation module is used to calculate the feature vector corresponding to the observation value collected at that time in the data matrix according to the observation value collected at that time after each collection in the current preset period, so as to obtain the intermediate feature vector collected at that time; A submatrix calculation module, used for calculating a first submatrix and a second submatrix according to the intermediate feature vectors collected at the time and the observation values collected at the time; An intermediate matrix updating module, used for updating the first intermediate matrix and the second intermediate matrix by using the first sub-matrix and the second sub-matrix calculated by the current acquisition; A model coefficient calculation module, used to calculate the solution of the linear equation group using the last updated first intermediate matrix and the second intermediate matrix to obtain the model coefficient of the current preset period when the preset number of acquisitions is reached within the current preset period; The predistorter lookup table updating module is used to update the predistorter lookup table according to the model coefficients of the current preset period, wherein the updated predistorter lookup table obtained in the current preset period is used to instruct the predistorter to perform predistortion processing on the input signal of the input predistortion processing system in the future preset period.
9. A predistortion processing system, characterized in that: include: A processor, configured to obtain, at the beginning of each preset cycle, a predistorter output signal, a power amplifier output signal, and a model coefficient of a predistortion model in a previous preset cycle and construct a linear equation system, obtain a data matrix of the linear equation system, and represent a solution of the linear equation system by a first intermediate matrix and a second intermediate matrix; When a new preset cycle starts, observation values are collected; wherein the observation values collected at this time include the predistorter output signal collected at this time and the power amplifier output signal collected at this time; After each acquisition in the current preset period, the feature vector corresponding to the observation value collected at that time in the data matrix is calculated according to the observation value collected at that time, so as to obtain the intermediate feature vector of the acquisition at that time; Calculate the first sub-matrix and the second sub-matrix according to the intermediate feature vector collected at the time and the observation value collected at the time; Update the first intermediate matrix and the second intermediate matrix by using the first submatrix and the second submatrix calculated by the current acquisition; When the preset number of acquisitions is reached within the current preset period, the solution of the linear equation group is calculated using the first intermediate matrix and the second intermediate matrix that were updated most recently to obtain the model coefficients of the current preset period; updating the predistorter lookup table according to the model coefficients of the current preset period; a predistorter, instructing the predistorter to perform predistortion processing on a target input signal input to the predistortion processing system in a future preset period according to the updated predistorter lookup table obtained in the current preset period, and outputting a target predistortion signal; A power amplifier is used to receive the target predistortion signal and perform power amplification on the target predistortion signal.
Citation Information
Cited By
Multistage power amplifier linearization system based on broadband communication
CN121923603A
Multistage power amplifier linearization system based on wideband communication
CN121923603B
