Digital predistortion method of improved memory polynomial model based on fft convolution correlation function

Abstract

The application belongs to the technical field of digital pre-distortion, and particularly relates to a digital pre-distortion method of an improved memory polynomial model based on an FFT convolution correlation function. The digital pre-distortion method adopts an improved memory polynomial model, adds a direct current term to correct errors caused by partial direct current bias, and adds a conjugate term to correct distortion caused by an image frequency. In a delay alignment module, a convolution correlation function method of fast Fourier transform (FFT) is used as a loop delay estimation algorithm. Compared with other correlation function loop delay estimation algorithms, the efficiency is greatly improved, and oversampling is not needed to improve accuracy.

Description

Technical Field

[0001] This invention belongs to the field of digital predistortion technology, specifically relating to a digital predistortion method based on an improved memory polynomial model using the FFT convolution correlation function. Background Technology

[0002] A power amplifier (PA) is used to amplify a modulated signal to the required power and transmit it through an antenna. It is an indispensable key component of modern wireless communication systems. When the power of the input signal is low, the PA usually operates in the linear region; however, when the peak value of the input signal approaches the PA's saturation point, the PA's linearity deteriorates, leading to an increase in the error vector amplitude (EVM) and spectral regrowth (broadening).

[0003] With the development of the wireless communication industry, modern communication systems such as LTE, 5G NR, and WLAN widely adopt modulation techniques with higher spectral efficiency, such as Quadrature Amplitude Modulation (QAM). When these signals pass through a power amplifier (PA), the nonlinearity of the PA causes intermodulation distortion (IMD) and spectral spread, leading to an increase in the adjacent channel power ratio (ACPR) and resulting in severe adjacent channel interference (ACI). In-band distortion further reduces bit error rate (BER) performance, increasing the error vector amplitude. Therefore, PA linearization technology has become a hot research area in the modern communication industry.

[0004] Current power amplifier linearization technologies mainly include power back-off technology, feedforward technology, Cartesian negative feedback technology, nonlinear device method, envelope elimination and recovery technology, and digital predistortion (DPD) technology.

[0005] Power back-off technology reduces the efficiency of power amplifiers (PAs). The need for higher-power PAs significantly increases equipment costs, sometimes reaching up to one-third of the total base station cost. Power back-off also reduces power utilization and increases heat dissipation. Furthermore, after a certain point, further back-off fails to significantly improve PA linearity. Therefore, in modern wireless communication systems with high linearity requirements, power back-off is no longer sufficient. Feedforward linearization provides good linearity but is generally inefficient, and the analog hardware is expensive, complex, and difficult to integrate and maintain. Besides feedforward, Cartesian feedback and nonlinear device linear amplification techniques, similar to feedforward and its variations, either involve extensive analog hardware or require difficult-to-control nonlinear components. Envelope cancellation and recovery techniques achieve high efficiency at different output powers, but the phase itself changes during envelope recovery to the carrier signal, causing a stretching of the useful signal spectrum and weakening the RF power amplifier's linearity. Therefore, achieving the required constant envelope phase-modulated signal synchronized with the envelope signal is very difficult to implement.

[0006] Among all linearization techniques, digital predistortion (DPD) is gradually replacing various analog linearization techniques and becoming a key technology for commercial applications in wireless communication.

[0007] The working principle of DPD (Distortion-Proofing) technology is to pre-generate inverse distortion of the baseband signal, which is the opposite of the power amplifier's characteristics, so that the cascaded response of the DPD and PA can achieve the desired linear response. DPD technology pre-distorts the transmitted signal by acquiring in-band and out-of-band output data of the PA with nonlinear distortion. Generally, 3 to 5 times the signal bandwidth is required to achieve a good correction effect. DPD technology allows the PA to operate linearly near the saturation region, achieving optimal PA efficiency. Furthermore, DPD implementation is low-cost and low-complexity, can be implemented in software on FPGAs or DSPs, has good reconfiguration capabilities, and is well-suited for various application environments.

[0008] Currently, most common digital predistortion (DPD) technologies employ memory amplifier models and correlation function loop delay estimation algorithms. Moreover, research on DPD remains largely at the simulation level, and there are still many problems in actual engineering implementation. For example, different amplifier paths may have different amplifier models, and the model parameters are related to factors such as input power, signal bandwidth, and signal type. Summary of the Invention

[0009] To address the aforementioned technical problems, this invention provides a digital predistortion method based on an improved memory polynomial model using the FFT convolutional correlation function. The improved memory polynomial model incorporates a DC term to correct errors caused by DC bias and a conjugate term to correct distortion caused by the mirror frequency. In the delay alignment module, the Fast Fourier Transform (FFT) convolutional correlation function method is used as the loop delay estimation algorithm. Compared to other correlation function loop delay estimation algorithms, this method significantly improves efficiency and eliminates the need for oversampling to enhance accuracy.

[0010] This invention is achieved through the following technical solution:

[0011] A digital predistortion method based on an improved memory multinomial model using the FFT convolution correlation function, the method comprising the following steps:

[0012] (1) Input the acquired baseband signal x(n) into the predistortion module and output the predistortion signal z(n);

[0013] (2) The predistorted signal z(n) passes through the digital-to-analog converter DAC and the power amplifier PA in sequence to obtain the power amplifier output signal y(t);

[0014] (3) Input the power amplifier output signal y(t) into the analog-to-digital converter (ADC), and collect the PA output signal y(n) at the output terminal of the ADC;

[0015] Determine whether the PA output signal y(n) satisfies linearization. If it does not satisfy linearization, proceed to step (4); if it satisfies linearization, stop the iteration.

[0016] (4) Use y(n) / G as the feedback signal of the power amplifier PA, where G represents the desired amplitude gain of the power amplifier;

[0017] A data preprocessing module is used to preprocess y(n) / G to obtain preprocessed y(n) / G. The data preprocessing includes truncating the broadband periodic signal so that the truncated signal contains the complete number of cycles.

[0018] (5) Input the preprocessed y(n) / G into the delay alignment module. The delay alignment module performs delay alignment on the preprocessed y(n) / G and the predistorted signal z(n) to obtain the aligned PA output signal, denoted as y. D (n);

[0019] (6) Align the PA output signal y D (n) serves as the input to the predistortion training module, and after processing by the predistortion training module, it outputs the desired predistortion signal. The error function is z(n) is the predistortion signal;

[0020] By minimizing the error using Newton's method, the polynomial coefficients corresponding to the minimum value of e(n) are obtained, and the polynomial coefficients corresponding to the minimum value of e(n) are updated to the coefficients of the predistortion training module.

[0021] (7) Copy the updated coefficients of the predistortion training module directly to the predistortion module, update the coefficients of the predistortion module, and complete one iteration update; after the coefficients of the predistortion module are updated, proceed to step (1).

[0022] Furthermore, the predistortion module in step (1) and the predistortion training module in step (6) use the same improved memory polynomial power amplifier model.

[0023] Furthermore, in step (1), the improved memory polynomial power amplifier model used by the predistortion module is:

[0024]

[0025] Where n = 1, 2, 3, ... N, N is the number of sampling points; h dc The term represents the DC term and is a constant term; z mp (n) represents the memory delay term, z conj (n) represents the conjugate term. and Let these represent the advance delay term and the lag delay term, respectively. The formula is expanded as follows:

[0026]

[0027]

[0028]

[0029]

[0030] In the above formula, K represents the highest order of the polynomial, and the value of k ranges from 1, 3, ..., 2l+1, ..., K; Q represents the memory delay depth, and the value of q ranges from 0, 1, 2, ..., Q; Q m Indicates the depth of pre-memory delay, q m The value range of is 0, 1, 2, ..., Q m Q l q represents the depth of delayed memory. l The value range of is 0, 1, 2, ..., Q l h dc The constant is very small, on the order of 10. -5 Left and right are used to correct errors caused by DC bias.

[0031] h k,q Let k be the coefficients of the polynomial, k = 2l + 1, l = 0, 1, 2, 3, ...;

[0032] x(nq) means that x(n) is delayed by q units, and zeros are padded from x(1) to x(q);

[0033] x'(nq) represents taking the conjugate of x(nq);

[0034] h' k,q Let k be the coefficient of the conjugate term, k = 2l + 1, l = 0, 1, 2, 3, ...;

[0035] The coefficient for the delayed term in advance memory is given by k = 2l + 1, where l = 0, 1, 2, 3, ...;

[0036] Let k be the coefficient of the delayed term in delayed memory, k = 2l + 1, l = 0, 1, 2, 3, ...

[0037] Furthermore, in step (6), the improved memory polynomial power amplifier model used by the predistortion training module is:

[0038]

[0039] Where n = 1, 2, 3, ... N, N is the number of sampling points; h dc This represents the DC term, which is a constant term. Indicates a term indicating memory delay. Indicates the conjugate term. and Let these represent the advance delay term and the lag delay term, respectively. The formula is expanded as follows:

[0040]

[0041]

[0042]

[0043]

[0044] In the above formula, K represents the highest order of the polynomial, and the value of k ranges from 1, 3, ..., 2l+1, ..., K; Q represents the memory delay depth, and the value of q ranges from 0, 1, 2, ..., Q; Q m Indicates the depth of pre-memory delay, q m The value range of is 0, 1, 2, ..., Q m Q l q represents the depth of delayed memory. l The value range of is 0, 1, 2, ..., Ql ;

[0045] h k,q Let k be the coefficients of the polynomial, k = 2l + 1, l = 0, 1, 2, 3, ...;

[0046] y D (nq) represents y D (n) Delay by q units, y D (1) to y D (q) zero padding; y D '(nq) represents y D (nq) takes the conjugate; h dc The constant is very small, on the order of 10. -5 Left and right are used to correct errors caused by DC bias.

[0047] h' k,q Let k be the coefficient of the conjugate term, k = 2l + 1, l = 0, 1, 2, 3, ...;

[0048] The coefficient for the delayed term in advance memory is given by k = 2l + 1, where l = 0, 1, 2, 3, ...;

[0049] Let k be the coefficient of the delayed term in delayed memory, k = 2l + 1, l = 0, 1, 2, 3, ...

[0050] Furthermore, in step (7), each time the coefficients of the predistortion training module are updated, the updated coefficients (including h) will be updated. k,q ,h' k,q , h dc Copy the corresponding coefficients to the predistortion module one by one to complete one iteration update.

[0051] Furthermore, in step (1), the initial coefficients of the predistortion module are [1,0,0,…], that is, when k=0 and q=0, h k,q =1, and the remaining coefficients (including h) k≠0,q≠0 ,h' k,q , h dc All are 0. After the first pass through the predistortion module, z(n) = x(n).

[0052] Furthermore, in step (3), during the acquisition of the PA output signal y(n) at the output of the analog-to-digital converter (ADC), the sampling rate is at least 3 to 5 times the bandwidth of the baseband signal x(n), and the length of the acquired data of the PA output signal y(n) is longer than the length of the baseband signal x(n). For WLAN broadband signals, the length of y(n) should be at least twice the length of x(n) to ensure that y(n) contains at least one complete cycle of the WLAN signal.

[0053] Further, in step (3), determining whether the PA output signal y(n) satisfies linearization specifically involves comparing the PA output signal y(n) with the amplitude characteristics (AM / AM), phase characteristics (PM / AM), normalized mean square error (NMSE), and error vector magnitude (EVM) of the baseband signal x(n) to evaluate the linearization performance.

[0054] Further, in step (5), the delay alignment module uses three methods—the convolution correlation function method of fast Fourier transform, phase compensation, and fractional alignment—to perform delay alignment on the preprocessed y(n) / G and the predistorted signal z(n). The specific methods are as follows:

[0055] (1) Pad the preprocessed y(n) / G with zeros so that the length of the preprocessed y(n) / G is consistent with that of z(n). Then perform Fast Fourier Transform (FFT) operations on the preprocessed y(n) / G and z(n) respectively.

[0056] Y(n) = FFT(y(n) / G);

[0057] Z(n) = FFT(x(n));

[0058] Taking the conjugate of Z(n):

[0059] Z conj (n) = Conj(Z(n));

[0060] Convolution followed by inverse Fourier transform (IFFT) operation, taking the absolute value, yields the correlation function F. corr (n) is:

[0061] F corr (n)=|IFFT(Y(n)·Z conj (n))|

[0062] Calculate the index n of the maximum value max Max() represents the maximum value operation;

[0063] F corr (n max ) = Max(F corr (n))

[0064] The delay length τ, which is an integer multiple of the data length of z(n), is equal to the data length of z(n) minus the index value n. max The integer-aligned signal is denoted as y. I (n):

[0065] y I (n)=y(n-τ) / G

[0066] (2) Phase compensation technology is used to correct the phase deviation caused by PA nonlinearity:

[0067] Integer-aligned y I Both z(n) and z(n) are represented by the product of the magnitude function and the phase function:

[0068]

[0069]

[0070] Where, Φ y (n) and Φ z (n) represent y I The phase functions of z(n) and z(n), A y (n) and A z (n) represent y I The magnitude functions of z(n) and z(n), where i is the imaginary unit;

[0071] Using Φ y (n) and Φ z The average phase difference of (n) To compensate y I (n) signal, average phase difference The calculation formula is as follows:

[0072]

[0073] n = 1, 2, 3, ... N, where N is the number of sampling points;

[0074] The phase-compensated signal is denoted as y. P (n):

[0075]

[0076] (3) The phase-compensated data y is processed using fractional alignment technology. P (n) Perform a 10x interpolation operation, and denot the interpolated data as y. P (n 10 ), n 10 =1,2,3,…,10N;

[0077] For the interpolated y P (n10 Delay -9 to 9 units, y P (n 10 A delay of 1 unit is equivalent to y P (n) Calculate y after each delay of 0.1 units. P The NMSE (Normalized Mean Square Error) index of (n) is used, and the data with the smallest NMSE value is used as the decimal-aligned data to obtain the aligned PA output signal, denoted as y. D (n). Decimal alignment technology solves the sampling bias caused by sampling randomness, avoids potential jump points at the beginning and end, and makes the aligned data smaller in NMSE and more accurate.

[0078] Furthermore, for broadband signals, in the predistortion training module, an effective data truncation method is adopted for processing before predistortion training. Specifically, a segment of inherent noise floor signal at the end of a complete broadband periodic signal is truncated, making the parameters obtained by the predistortion training module more accurate. Because the amount of data is shortened, the efficiency of predistortion training is also higher.

[0079] Beneficial technical effects of the present invention:

[0080] The digital predistortion method provided by this invention can achieve the following: First, improved efficiency. It adopts the FFT convolutional correlation function method as the loop delay estimation algorithm, which greatly reduces the complex multiplication units in the estimation operation. Compared with the original loop delay estimation algorithm, the efficiency is greatly improved, and oversampling is not required to improve accuracy. Second, improved accuracy. The model provided by this invention is more in line with actual needs. It adds DC term and conjugate term. The DC term can correct the deviation caused by DC bias, and the conjugate term is used to correct the distortion caused by the image frequency.

[0081] The digital predistortion method provided by this invention employs an indirect learning structure based on a memory polynomial model. By minimizing the error function using Newton's method, it pre-generates an inverse distortion in the baseband signal, which is the opposite of the power amplifier's characteristics, thereby linearizing the system's cascaded response. Through this digital predistortion method, the power amplifier (PA) can operate linearly near the saturation region, thus achieving optimal PA efficiency. Furthermore, digital predistortion (DPD) has low implementation cost and complexity, can be implemented in software on FPGAs or DSPs, and possesses excellent reconfiguration capabilities, making it highly suitable for various application environments.

[0082] The digital predistortion method provided by this invention incorporates a data preprocessing module for broadband signal types such as WLAN, which improves the accuracy of the delay alignment module, avoids spectral distortion caused by incomplete data, and prevents interference from invalid data when training DPD parameters, resulting in higher accuracy.

[0083] This invention employs an improved memory amplifier model that better meets practical needs and a more efficient FFT convolutional correlation function delay estimation algorithm. Tests using actual signal sources have shown that it has a good linearization correction effect for broadband signals such as multi-tone and WLAN. Attached Figure Description

[0084] Figure 1 This is a schematic diagram of the digital predistortion method based on the FFT convolutional correlation function in an embodiment of the present invention.

[0085] Figure 2 This is a flowchart of a digital predistortion method based on an improved memory polynomial model using the FFT convolution correlation function, as described in an embodiment of the present invention. Detailed Implementation

[0086] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0087] Conversely, this invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of the invention as defined in the claims. Furthermore, to provide a better understanding of the invention, certain specific details are described in detail below. However, those skilled in the art will fully understand the invention even without these detailed descriptions.

[0088] This invention provides an embodiment of an improved digital predistortion method for memory multinomial models based on the FFT convolution correlation function, such as... Figure 1-2 As shown, the method includes the following steps:

[0089] (1) Input the acquired baseband signal x(n) into the predistortion module and output the predistortion signal z(n);

[0090] (2) The predistorted signal z(n) passes through the digital-to-analog converter DAC and the power amplifier PA in sequence to obtain the power amplifier output signal y(t);

[0091] (3) Input the power amplifier output signal y(t) into the analog-to-digital converter (ADC), and collect the PA output signal y(n) at the output terminal of the ADC;

[0092] Determine whether the PA output signal y(n) satisfies linearization. If it does not satisfy linearization, proceed to step (4); if it satisfies linearization, stop the iteration.

[0093] (4) Use y(n) / G as the feedback signal of the power amplifier PA, where G represents the desired amplitude gain of the power amplifier;

[0094] A data preprocessing module is used to preprocess y(n) / G to obtain preprocessed y(n) / G. The data preprocessing includes truncating the broadband periodic signal so that the truncated signal contains the complete number of cycles.

[0095] The purpose of the data preprocessing module is to extract a signal with a complete cycle. This module is primarily designed for broadband signal types such as WLAN. Specifically, it divides the signal into 100 consecutive blocks of sampling points and calculates the sum of the amplitudes of all blocks. The amplitude differences between these blocks are then used to estimate the start and end points of the signal, as well as the number of cycles. This approach offers two advantages: firstly, it reduces the computational load of the delay alignment process; secondly, it ensures the accuracy of the delay alignment results, avoiding spectral distortion caused by inconsistencies in the data. This also prevents interference from invalid data during DPD parameter training, resulting in higher accuracy.

[0096] (5) Input the preprocessed y(n) / G into the delay alignment module. The delay alignment module performs delay alignment on the preprocessed y(n) / G and the predistorted signal z(n) to obtain the aligned PA output signal, denoted as y. D (n);

[0097] The delay alignment module employs three methods—the convolutional correlation function method using Fast Fourier Transform (FFT), phase compensation, and fractional alignment—to perform delay alignment on the preprocessed y(n) / G and the predistorted signal z(n). Specifically, one method in the delay alignment module utilizes the convolutional correlation function method using Fast Fourier Transform (FFT). This method significantly reduces the number of complex multiplication units in the estimation operation, greatly improving computational efficiency compared to existing loop delay estimation algorithms such as the amplitude difference correlation function method and the cross-correlation function method. The specific calculation method is as follows:

[0098] (1) Pad the preprocessed y(n) / G with zeros so that the length of the preprocessed y(n) / G is consistent with that of z(n). Then perform Fast Fourier Transform (FFT) operations on the preprocessed y(n) / G and z(n) respectively.

[0099] Y(n) = FFT(y(n) / G);

[0100] Z(n) = FFT(z(n));

[0101] Taking the conjugate of Z(n):

[0102] Z conj (n) = Conj(Z(n));

[0103] Convolution followed by inverse Fourier transform (IFFT) operation, taking the absolute value, yields the correlation function F. corr (n) is:

[0104] F corr (n)=|IFFT(Y(n)·Z conj (n))|

[0105] Calculate the index n of the maximum value max Max() represents the maximum value operation;

[0106] F corr (n max ) = Max(F corr (n))

[0107] The delay length τ, which is an integer multiple of the data length of z(n), is equal to the data length of z(n) minus the index value n. max The integer-aligned signal is denoted as y. I (n):

[0108] y I (n)=y(n-τ) / G

[0109] (2) Phase compensation technology is used to correct the phase deviation caused by PA nonlinearity:

[0110] Integer-aligned y I Both z(n) and z(n) are represented by the product of the magnitude function and the phase function:

[0111]

[0112]

[0113] Where, Φ y (n) and Φ z (n) represent y I The phase functions of z(n) and z(n), A y (n) and A z (n) represent y I The magnitude functions of z(n) and z(n), where i is the imaginary unit;

[0114] Using Φ y (n) and Φ z The average phase difference of (n) To compensate y I (n) signal, average phase difference The calculation formula is as follows:

[0115]

[0116] n = 1, 2, 3, ... N, where N is the number of sampling points;

[0117] The phase-compensated signal is denoted as y. P (n):

[0118]

[0119] (3) The phase-compensated data y is processed using fractional alignment technology. P (n) Perform a 10x interpolation operation, and denot the interpolated data as y. P (n 10 ), n 10 =1,2,3,…,10N;

[0120] For the interpolated y P (n 10 Delay -9 to 9 units, y P (n 10 A delay of 1 unit is equivalent to y P (n) Calculate y after each delay of 0.1 units. P The NMSE (Normalized Mean Square Error) index of (n) is used, and the data with the smallest NMSE value is used as the decimal-aligned data to obtain the aligned PA output signal, denoted as y. D (n). Decimal alignment technology solves the sampling bias caused by sampling randomness, avoids potential jump points at the beginning and end, and makes the aligned data smaller in NMSE and more accurate.

[0121] (6) Align the PA output signal y D (n) serves as the input to the predistortion training module, and after processing by the predistortion training module, it outputs the desired predistortion signal. The error function is z(n) is the predistortion signal;

[0122] By minimizing the error using Newton's method, the polynomial coefficients corresponding to the minimum value of e(n) are obtained, and the polynomial coefficients corresponding to the minimum value of e(n) are updated to the coefficients of the predistortion training module.

[0123] Specifically, the polynomial order of the improved memory polynomial model is generally set to 7, the memory length is generally set to 3, and the improved memory polynomial model includes DC terms, conjugate terms, and lag and advance memory terms can be selected.

[0124] (7) Copy the updated coefficients of the predistortion training module directly to the predistortion module, update the coefficients of the predistortion module, and complete one iteration update; after the coefficients of the predistortion module are updated, proceed to step (1).

[0125] In this embodiment, the predistortion module in step (1) and the predistortion training module in step (6) use the same improved memory polynomial power amplifier model. The predistortion training module z(n) and The calculation formula is the same, the difference is that the input variable for z(n) is x(n), while... The input variable is y D (n);

[0126] In step (1), the improved memory polynomial power amplifier model used by the predistortion module is as follows:

[0127]

[0128] Where n = 1, 2, 3, ... N, N is the number of sampling points; h dc The term represents the DC term and is a constant term; z mp (n) represents the memory delay term, z conj (n) represents the conjugate term. and Let these represent the advance delay term and the lag delay term, respectively. The formula is expanded as follows:

[0129]

[0130]

[0131]

[0132]

[0133] In the above formula, K represents the highest order of the polynomial, and the value of k ranges from 1, 3, ..., 2l+1, ..., K; Q represents the memory delay depth, and the value of q ranges from 0, 1, 2, ..., Q; Q m Indicates the depth of pre-memory delay, q m The value range of is 0, 1, 2, ..., Q m Q l q represents the depth of delayed memory. l The value range of is 0, 1, 2, ..., Q l ;

[0134] h k,q Let k be the coefficients of the polynomial, k = 2l + 1, l = 0, 1, 2, 3, ...;

[0135] x(nq) means that x(n) is delayed by q units, and zeros are padded from x(1) to x(q);

[0136] x'(nq) represents taking the conjugate of x(nq);

[0137] h' k,qLet k be the coefficient of the conjugate term, k = 2l + 1, l = 0, 1, 2, 3, ...;

[0138] The coefficient for the delayed term in advance memory is given by k = 2l + 1, where l = 0, 1, 2, 3, ...;

[0139] Let k be the coefficient of the delayed term in delayed memory, k = 2l + 1, l = 0, 1, 2, 3, ...

[0140] In step (6), the improved memory polynomial power amplifier model used by the predistortion training module is as follows:

[0141]

[0142] Where n = 1, 2, 3, ... N, N is the number of sampling points; h dc This represents the DC term, which is a constant term. Indicates a term indicating memory delay. Indicates the conjugate term. and Let these represent the advance delay term and the lag delay term, respectively. The formula is expanded as follows:

[0143]

[0144]

[0145]

[0146]

[0147] In the above formula, K represents the highest order of the polynomial, and the value of k ranges from 1, 3, ..., 2l+1, ..., K; Q represents the memory delay depth, and the value of q ranges from 0, 1, 2, ..., Q; Q m Indicates the depth of pre-memory delay, q m The value range of is 0, 1, 2, ..., Q m Q l q represents the depth of delayed memory. l The value range of is 0, 1, 2, ..., Q l ;

[0148] h k,q Let k be the coefficients of the polynomial, k = 2l + 1, l = 0, 1, 2, 3, ...;

[0149] y D (nq) represents y D (n) Delay by q units, y D (1) to y D (q) zero padding; y D'(nq) represents the expression for y D (nq) takes the conjugate;

[0150] h' k,q Let k be the coefficient of the conjugate term, k = 2l + 1, l = 0, 1, 2, 3, ...;

[0151] The coefficient for the delayed term in advance memory is given by k = 2l + 1, where l = 0, 1, 2, 3, ...;

[0152] Let k be the coefficient of the delayed term in delayed memory, k = 2l + 1, l = 0, 1, 2, 3, ...

[0153] In step (1), the initial coefficients of the predistortion module are [1,0,0,…], that is, when k=0 and q=0, h k,q =1, and the remaining coefficients (including h) k≠0，q≠0 ,h' k,q , h dc All are 0. After the first pass through the predistortion module, z(n) = x(n).

[0154] In step (7), each time the coefficients of the predistortion training module are updated, the updated coefficients (including h) will be updated. k,q ,h' k，q , h dc Copy the corresponding coefficients to the predistortion module one by one to complete one iteration update.

[0155] In step (3), during the acquisition of the PA output signal y(n) at the output of the analog-to-digital converter (ADC), the sampling rate is at least 3 to 5 times the bandwidth of the baseband signal x(n), and the length of the acquired data of the PA output signal y(n) is longer than the length of the baseband signal x(n). For WLAN broadband signals, the length of y(n) should be at least twice the length of x(n) to ensure that y(n) contains at least one complete cycle of the WLAN signal.

[0156] In step (3), determining whether the PA output signal y(n) satisfies linearization specifically involves comparing the amplitude characteristics (AM / AM), phase characteristics (PM / AM), normalized mean square error (NMSE), and error vector magnitude (EVM) of the PA output signal y(n) with those of the baseband signal x(n) to evaluate the linearization performance. Conventional techniques can be used for this linearization evaluation, and will not be elaborated further here.

[0157] For broadband signals, in the predistortion training module, an effective data truncation method is used to process the data before predistortion training. Specifically, a segment of inherent noise floor signal at the end of a complete broadband periodic signal is truncated, making the parameters obtained by the predistortion training module more accurate. Because the data volume is shortened, the efficiency of predistortion training is also higher.

[0158] This invention employs a delay alignment module, which utilizes the convolution correlation function method of Fast Fourier Transform (FFT). First, Fourier transforms are performed on both the original and delayed signals, followed by convolution and then inverse Fourier transform to obtain the correlation curve. This method significantly reduces the number of complex multiplication units in the estimation operation, greatly improving computational efficiency compared to existing loop delay estimation algorithms such as the amplitude difference correlation function method and the cross-correlation function method. The phase compensation technique in the delay alignment module corrects the phase deviation caused by PA nonlinearity. The improved fractional alignment technique in the delay alignment module solves the sampling deviation caused by sampling randomness, avoids jump points at the beginning and end of the signal due to interpolation, resulting in a smaller NMSE and higher accuracy in the aligned data.

[0159] The effective data extraction method in the predistortion training process, which performs DPD learning on effective data, avoids some learning errors caused by noise background signals, making the obtained parameters more accurate and more efficient.

[0160] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital predistortion method for an improved memory polynomial model based on the FFT convolution correlation function, characterized in that, The method includes the following steps: (1) Input the acquired baseband signal x(n) into the predistortion module and output the predistortion signal z(n); (2) The predistorted signal z(n) passes through the digital-to-analog converter DAC and the power amplifier PA in sequence to obtain the power amplifier output signal y(t); (3) Input the power amplifier output signal y(t) into the analog-to-digital converter (ADC), and collect the PA output signal y(n) at the output terminal of the ADC; Determine whether the PA output signal y(n) satisfies linearization. If it does not satisfy linearization, proceed to step (4); if it satisfies linearization, stop the iteration. (4) Use y(n) / G as the feedback signal of the power amplifier PA, where G represents the desired amplitude gain of the power amplifier; A data preprocessing module is used to preprocess y(n) / G to obtain preprocessed y(n) / G. The data preprocessing includes truncating the broadband periodic signal so that the truncated signal contains the complete number of cycles. (5) Input the preprocessed y(n) / G into the delay alignment module. The delay alignment module performs delay alignment on the preprocessed y(n) / G and the predistorted signal z(n) to obtain the aligned PA output signal, denoted as . ; (6) Align the PA output signal As input to the predistortion training module, the desired predistortion signal is output after processing by the predistortion training module. (n); the error function is e(n) = z(n) - z(n) and z(n) are the predistortion signals; By minimizing the error using Newton's method, the polynomial coefficients corresponding to the minimum value of e(n) are obtained, and the polynomial coefficients corresponding to the minimum value of e(n) are updated to the coefficients of the predistortion training module. (7) Copy the updated coefficients of the predistortion training module directly to the predistortion module, update the coefficients of the predistortion module, and complete one iteration update; after the coefficients of the predistortion module are updated, proceed to step (1). The predistortion module in step (1) and the predistortion training module in step (6) use the same improved memory polynomial power amplifier model; In step (1), the improved memory polynomial power amplifier model used by the predistortion module is as follows: Where n = 1, 2, 3, ... N, N is the number of sampling points; This represents the DC term, which is a constant term. Indicates a term indicating memory delay. Indicates the conjugate term. and Let these represent the advance delay term and the lag delay term, respectively. The formula is expanded as follows: ； ； ； ； In the above formula, K represents the highest order of the polynomial, and the value of k ranges from 1, 3, ..., 2l+1, ..., K; Q represents the memory delay depth, and the value of q ranges from 0, 1, 2, ..., Q; Q m Indicates the depth of advance memory delay, q m The value range of is 0, 1, 2, ..., Q m ; Q l Indicates the depth of delayed memory. q l The value range of is 0, 1, 2, ..., Q l ; Let k be the coefficients of the polynomial, k = 2l + 1, l = 0, 1, 2, 3, ...; This indicates that x(n) is delayed by q units, and zeros are padded from x(1) to x(q); Indicates to Take conjugate; Let k be the coefficient of the conjugate term, k = 2l + 1, l = 0, 1, 2, 3, ...; The coefficient for the delayed term in advance memory is given by k = 2l + 1, where l = 0, 1, 2, 3, ...; Let k be the coefficient of the delayed term in delayed memory, k = 2l + 1, l = 0, 1, 2, 3, ...

2. The digital predistortion method for the improved memory polynomial model based on the FFT convolution correlation function according to claim 1, characterized in that, In step (6), the improved memory polynomial power amplifier model used by the predistortion training module is as follows: Where n = 1, 2, 3, ... N, N is the number of sampling points; This represents the DC term, which is a constant term. Indicates a term indicating memory delay. Indicates the conjugate term. and Let these represent the advance delay term and the lag delay term, respectively. The formula is expanded as follows: ； ； ； ； In the above formula, K represents the highest order of the polynomial, and the value of k ranges from 1, 3, ..., 2l+1, ..., K; Q represents the memory delay depth, and the value of q ranges from 0, 1, 2, ..., Q; Q m Indicates the depth of advance memory delay, q m The value range of is 0, 1, 2, ..., Q m ; Q l Indicates the depth of delayed memory. q l The value range of is 0, 1, 2, ..., Q l ; Let k be the coefficients of the polynomial, k = 2l + 1, l = 0, 1, 2, 3, ...; express (n) Delay by q units, (1) to (q) Zero padding; Take conjugate; Let k be the coefficient of the conjugate term, k = 2l + 1, l = 0, 1, 2, 3, ...; The coefficient for the delayed term in advance memory is given by k = 2l + 1, where l = 0, 1, 2, 3, ...; Let k be the coefficient of the delayed term in delayed memory, k = 2l + 1, l = 0, 1, 2, 3, ...

3. The digital predistortion method based on the improved memory polynomial model using the FFT convolution correlation function as described in claim 2, characterized in that, In step (7), each time the coefficients of the predistortion training module are updated, the updated coefficients are copied to the corresponding coefficients of the predistortion module to complete one iteration update.

4. The digital predistortion method for the improved memory polynomial model based on the FFT convolution correlation function according to claim 1, characterized in that, In step (1), the initial coefficients of the predistortion module are [1,0,0,…], that is, when k=0 and q=0, All other coefficients are 0. After the first pass through the predistortion module, z(n) = x(n).

5. The digital predistortion method for the improved memory polynomial model based on the FFT convolution correlation function according to claim 1, characterized in that, In step (3), during the process of acquiring the PA output signal y(n) at the output end of the analog-to-digital converter ADC, the sampling rate is at least 3 to 5 times the bandwidth of the baseband signal x(n), and the length of the acquired data of the PA output signal y(n) is longer than the length of the baseband signal x(n).

6. The digital predistortion method for the improved memory polynomial model based on the FFT convolution correlation function according to claim 1, characterized in that, In step (3), determining whether the PA output signal y(n) satisfies linearization specifically involves comparing the amplitude characteristics, phase characteristics, normalized mean square error, and error vector amplitude of the PA output signal y(n) with those of the baseband signal x(n) to evaluate the linearization performance.

7. The digital predistortion method for the improved memory polynomial model based on the FFT convolution correlation function according to claim 1, characterized in that, In step (5), the delay alignment module uses three methods—convolution correlation function method of fast Fourier transform, phase compensation, and fractional alignment—to perform delay alignment on the preprocessed y(n) / G and the predistorted signal z(n). The specific method is as follows: (1) Pad zeros into the preprocessed y(n) / G so that the length of the preprocessed y(n) / G is equal to that of the preprocessed y(n) / G. Consistent, then for the preprocessed y(n) / G and Perform Fast Fourier Transform (FFT) operations on each. ； ； right Take conjugate: ； Convolution followed by inverse Fourier transform, and then taking the absolute value, yields the correlation function. for: Calculate the index n of the maximum value max Max() represents the maximum value operation; Delay length in multiples of an integer equal Data length minus index value n max The integer-aligned signal is denoted as : (2) Phase compensation technology is used to correct the phase deviation caused by PA nonlinearity: Integer alignment Both z(n) and z(n) are represented by the product of the magnitude function and the phase function: Where, Φ y (n) and Φ z (n) respectively represent The phase function of z(n), A y (n) and A z (n) respectively represent And the magnitude function of z(n), where i is the imaginary unit; Using Φ y (n) and Φ z The average phase difference φ of (n) is used to compensate The formula for calculating the average phase difference φ of the signal is as follows: n = 1, 2, 3, ..., N, where N is the number of sampling points; The signal after phase compensation is denoted as : ； (3) Use fractional alignment technique to adjust the phase-compensated data Perform a 10x interpolation operation; the interpolated data is denoted as... n 10 =1,2,3,…,10N; After interpolation Delay -9 to 9 units, A delay of 1 unit is equivalent to Delay by 0.1 units, calculate the result after each delay. The NMSE index uses the set of data with the smallest NMSE value as the decimal-aligned data, thus obtaining the aligned PA output signal, denoted as . .

8. The digital predistortion method for the improved memory polynomial model based on the FFT convolution correlation function according to claim 1, characterized in that, For broadband signals, in the predistortion training module, an effective data truncation method is used for processing before predistortion training. Specifically, a segment of inherent noise floor signal at the end of a complete broadband periodic signal is truncated to make the parameters obtained by the predistortion training module more accurate.

Images