Threshold value optimization method for power amplifier model based on CPWL function
Through the amplifier model based on CPWL function and iterative learning control optimization threshold value, the problem of excessive resource consumption of Volterra series model is solved, and low power consumption and efficient predistortion performance is achieved, which is suitable for modern wireless communication systems.
Patent Information
- Application Number
- CN202111150860.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-29
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2041-09-29
AI Technical Summary
In modern wireless communication systems, the power amplifier model based on Volterra series consumes too much hardware resources due to the generation of high order terms, and it is difficult to reduce power and resource consumption while maintaining predistortion performance.
Using a power amplifier model based on segmented linear (CPWL) function, and by establishing a simplified second-order DDR power amplifier model expression, using symbolic functions to separate the threshold value, and optimizing the threshold value using least squares method and iterative learning control method to reduce the The number of segments and coefficients reduces the consumption of computing resources.
Without reducing the model accuracy, the resource consumption and power consumption of the amplifier model are reduced, the operating frequency of the predistortion system is improved, and the predistortion performance is improved.
Smart Images

Figure CN113868994B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of digital predistortion of radio frequency power amplifiers, and in particular to a threshold value optimization method suitable for a power amplifier model based on a CPWL function. Background Art
[0002] In modern wireless communication systems, in order to obtain a high communication rate, a large communication bandwidth and a modulation method with a high modulation order are often used, such as the widely used orthogonal frequency division multiplexing (OFDM) technology and quadrature amplitude modulation (QAM) technology. This will result in a high peak-to-average ratio (PARP) of the signal, which puts high demands on the linearity of the RF power amplifier. If the power amplifier is biased at the peak power of the signal when it is injected, the power amplifier is also in the linear region of operation. Since the average power is much lower than the peak power, this will cause the efficiency of the power amplifier to become very low, resulting in a large amount of power consumption and also posing a huge challenge to the heat dissipation of the communication system. Digital predistortion (DPD) technology is widely used to solve this problem. The basic principle of DPD is to pass a module called a predistorter with the opposite characteristics of the power amplifier before the signal is injected into the power amplifier, so that the cascade system of the predistorter-power amplifier becomes a linear system, thereby eliminating the nonlinear distortion of the final output signal of the RF power amplifier. Therefore, accurate modeling of the power amplifier characteristics is the core of DPD technology. For a long time, a variety of power amplifier models based on Volterra series have been proposed and widely used, such as the memory polynomial (MP) model. The basic idea of this type of model is to use an exponential function to fit the nonlinear characteristics of the power amplifier, which inevitably leads to the generation of high-order terms, so that this type of model needs to consume a lot of multiplier resources in hardware implementation. Recently, a new power amplifier model based on piecewise linear (CPWL) function has been proposed and has received extensive research and attention, such as the DVR model. The idea of this type of model is to segment the nonlinear characteristics of the power amplifier in amplitude, and then use a straight line to fit each segment. In general, a broken line is used to fit the curve corresponding to the nonlinearity. In CPWL, the absolute value operation is used to replace the high-order terms in the Volterra series, and the absolute value operation consumes far fewer resources than the multiplier in hardware implementation. Therefore, the power amplifier model of the CPWL function class has a very low complexity. Summary of the invention
[0003] The present invention aims to provide a threshold value optimization method suitable for a power amplifier model based on a CPWL function, so as to solve the technical problem that the resources and power consumption required by the model are too large while maintaining the pre-distortion performance.
[0004] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows:
[0005] A threshold value optimization method applicable to a power amplifier model based on a CPWL function comprises the following steps:
[0006] Step 1: Establish a simplified second-order DDR power amplifier model expression based on the CPWL function:
[0007]
[0008] In formula (1), n∈[0, N-1] represents the index position of the sampling point, and there are N sampling points in total; i
[0009] is the memory depth and i∈[0,M], M is the maximum memory depth; is the input complex signal of the power amplifier, is the complex output signal of the power amplifier, || is defined as the modulus operation of complex numbers and the absolute value operation of real numbers, () * Defined as the conjugate operation on complex numbers; K is the number of segments for the nonlinear fitting interval, k∈[1, K] represents the kth segment; β k is the kth segment threshold; is the model coefficient corresponding to the first term, is the model coefficient corresponding to the second term, is the model coefficient corresponding to the third term, is the model coefficient corresponding to the fourth term; the subscript ki represents the kth segment, and the memory depth is the model coefficient corresponding to i; the model is flexible in that it uses one or a combination of terms according to the power amplifier used and the requirements for the linearization of the power amplifier; in order to simplify the expression without loss of generality, all the following derivations are based on the scenario of using only the first-order terms of the model. When using multiple terms, it is only necessary to add the corresponding terms and coefficients in formula (1) to formulas (3)-(6), which will not be elaborated here; it can be expanded to obtain the case of using multiple terms;
[0010] Step 2: Derive model coefficients The solution method is to express the model in matrix form as
[0011] Y=Hc (2)
[0012] Where Y represents the vector of output signals, H is the model matrix, and c is the model coefficient vector, which are defined as H=[h 10 ,…,h KM ], (·) T Represents the transpose operation of a vector; where h ki It is defined as the model item corresponding to the kth segment and the memory depth i, which is expressed as
[0013]
[0014] The coefficients are solved using the least squares method, and the corresponding solution is expressed as
[0015]
[0016] In formula (4) represents the coefficient vector (·) H represents the conjugate transpose of a complex matrix, (·) -1 represents the inverse of a matrix;
[0017] Step 3: Derivation of threshold optimization method:
[0018] Step 3.1: Expressed as the sum of the real and imaginary parts
[0019]
[0020] in[·] r represents the real part of a complex number, [·] i represents the imaginary part of a complex number; j is the unit imaginary number;
[0021] Step 3.2: Define Φ ki (n)
[0022]
[0023] but The real part of
[0024]
[0025] where sgn(·) is the sign function, defined as
[0026]
[0027] Step 3.3: If the coefficient is the optimal solution, and the optimized threshold is expressed as The expected output corresponding to the optimized threshold is Its actual part is expressed as
[0028]
[0029] make
[0030]
[0031] definition Current Output and expected output Subtracting the real part of the error gives us the real part of the error, which is defined as
[0032]
[0033] Formula (11) can be expressed in matrix form as
[0034] E r = -ΨΔβ (12)
[0035] E in formula (12) r is the error vector, Ψ is the threshold matrix, Δβ is the threshold update vector, which are defined as Ψ=[[ψ 10 (0),…,ψ 10 (N-1)] T ,…,[ψ KM (0),…,ψ KM (N-1)] T ], (·) T represents the transpose operation of a vector; Δβ=[Δβ1,…,Δβ k ] T ; Solve for the least squares solution of Δβ, defined as
[0036]
[0037] Step 3.4: By iterating multiple times to approximate the optimal threshold, the difference between the current threshold value and the optimal threshold value is solved, thereby indirectly obtaining the optimal value of the threshold. The multiple iterations to approximate the optimal threshold are expressed as
[0038]
[0039] In the formula It represents the number of iterations;
[0040] Step 4: Update the predistorter with the obtained optimization threshold and model coefficients to complete the entire predistortion process.
[0041] Furthermore, the specific iterative process in step 3.4 is as follows:
[0042] Step 3.4.1: Use iterative learning control method to search for the ideal output of the predistorter Select the initial threshold as uniform distribution, β1 = [1 / K, 2 / K, ..., 1] T , set the variable representing the number of iterations
[0043] Step 3.4.2: Output the ideal Threshold and input Substitute into formula (4) to solve the model coefficient corresponding to this iteration, denoted as
[0044] Step 3.4.3, put and input Substitute into formula (2) to calculate the model output corresponding to the current coefficient and threshold, denoted as calculate And ideal output If the error meets the requirement, the modeling is completed and the iteration ends; otherwise, proceed to the next step.
[0045] Step 3.4.4, and Substitute into formula (11) to (14) to calculate the updated value of the threshold and record it as Set the number of iterations Then go to step 3.4.2 to start the next iteration process.
[0046] The threshold value optimization method of the power amplifier model based on the CPWL function of the present invention has the following advantages:
[0047] The present invention separates the threshold of the CPWL function from the absolute value sign by using a sign function, so that the updated value of the threshold value and the output can be approximately regarded as a linear relationship and the updated value of the threshold can be solved by the least square method. Then, the optimized threshold is gradually searched out by an iterative method, and after optimization, the number of CPWL segments can be reduced without reducing the accuracy of the model. The number of coefficients and the consumption of computing resources are reduced, the scale of the model is reduced, and the power consumption is reduced. At the same time, since the scale of the model is reduced, the requirements for the timing of the digital circuit are relaxed, and the pre-distortion system can be operated at a higher frequency. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 It is a structural schematic diagram of a power amplifier digital predistortion device of the present invention;
[0049] Figure 2 is a flow chart of the digital predistortion modeling method of the present invention;
[0050] Figure 3 It is a schematic diagram of the power spectrum after digital pre-distortion using the initial threshold and the optimized threshold for a 5G-NR signal with a bandwidth of 100 MHz when the system sampling frequency is 500 MHz. DETAILED DESCRIPTION
[0051] In order to better understand the purpose, structure and function of the present invention, a threshold value optimization method of a power amplifier model based on a CPWL function of the present invention is further described in detail below with reference to the accompanying drawings.
[0052] like Figure 1As shown, the power amplifier digital predistortion device of the present invention includes a digital predistorter based on a simplified second-order DDR model of a CPWL function, an analog-to-digital converter, a digital-to-analog converter, an orthogonal modulator, an orthogonal demodulator, a radio frequency power amplifier, a coupler, iterative learning control, coefficient estimation and threshold update.
[0053] The overall flow chart of the present invention is as follows Figure 2 As shown, the expected output of the predistorter is first searched using iterative learning control, then the model terms, number of thresholds and memory depth used are set, and then the threshold optimization method proposed in the present invention is used to iterate the optimized thresholds and coefficients, and finally the optimized thresholds and coefficients are used to model the predistorter. Specifically, the baseband signal without predistortion is first used. The analog RF signal is obtained by sequentially passing through a digital predistorter, a digital-to-analog converter, an orthogonal modulator, and an RF power amplifier based on a simplified second-order DDR model of a CPWL function. The analog RF signal is sequentially passed through a coupler, an orthogonal demodulator, and an analog-to-digital converter to obtain a distorted digital baseband signal corresponding to the output of the power amplifier. Then, the power amplifier is searched through iterative learning control to output a linear signal, that is, The input signal corresponding to the power amplifier That is the ideal output signal of the predistorter.
[0054] The input end of the digital predistorter based on the CPWL function is externally connected to a digital baseband complex input signal And the ideal output All are known, and the optimized threshold and coefficient are iteratively searched according to the following steps and the predistorter is updated:
[0055] Step 1: Establish a simplified second-order DDR power amplifier model expression based on the CPWL function:
[0056]
[0057] In formula (1), n∈[0, N-1] represents the index position of the sampling point, and there are N sampling points in total; i is the memory depth and i∈[0, M], where M is the maximum memory depth. is the input complex signal of the power amplifier, is the complex output signal of the power amplifier, || is defined as the modulus operation of complex numbers and the absolute value operation of real numbers, () * Defined as the conjugate operation of complex numbers. K is the number of segments for the nonlinear fitting interval, and k∈[1, K] represents the kth segment. β k is the kth segment threshold; is the model coefficient corresponding to the first term, is the model coefficient corresponding to the second term, is the model coefficient corresponding to the third term, is the model coefficient corresponding to the fourth term; the subscript ki represents the kth segment, and the memory depth is the model coefficient corresponding to i. When using this model, one or a combination of terms can be used according to the actual power amplifier used and the requirements for power amplifier linearization, which has high flexibility. In order to simplify the expression without loss of generality, all the following derivations are based on the scenario of using only the first-order terms of the model. When using multiple terms, it is only necessary to add the corresponding terms and coefficients in formula (1) to formulas (3)-(6), which will not be repeated here; it is easy to expand it to the case of using multiple terms.
[0058] Step 2: Derive model coefficients The solution method is to express the model in matrix form as
[0059] Y=Hc (2)
[0060] Where Y represents the vector of output signals, H is the model matrix, and c is the model coefficient vector, which are defined as H=[h 10 ,…,h KM ], (·) T Represents the transpose operation of a vector. ki It is defined as the model item corresponding to the kth segment and the memory depth i. It can be expressed as
[0061]
[0062] Using the least squares method to solve the coefficients, the corresponding solution can be expressed as
[0063]
[0064] In formula (4) represents the coefficient vector (·) H represents the conjugate transpose of a complex matrix, (·) -1 represents the inverse of a matrix;
[0065] Step 3, deriving a threshold optimization method;
[0066] Step 3.1: Expressed as the sum of the real and imaginary parts
[0067]
[0068] in[·] r represents the real part of a complex number, [·] i represents the imaginary part of a complex number; j is the unit imaginary number;
[0069] Step 3.2: To simplify the expression, define Φ ki (n)
[0070]
[0071] but The real part of
[0072]
[0073] where sgn(·) is the sign function, defined as
[0074]
[0075] Step 3.3: Assume that the coefficients at this time are It is already the optimal solution, and the optimized threshold is expressed as The expected output corresponding to the optimized threshold is Its actual part is expressed as
[0076]
[0077] Generally speaking, the difference between the optimization threshold and the current threshold is small, so we can make the following assumptions:
[0078]
[0079] definition Current Output and expected output Subtracting the real part of the error gives us the real part of the error, which is defined as
[0080]
[0081] Formula (11) can be expressed in matrix form as
[0082] E r = -ΨΔβ(12)
[0083] E in formula (12) r is the error vector, Ψ is the threshold matrix, Δβ is the threshold update vector, which are defined as Ψ=[[ψ 10 (0),…,ψ 10 (N-1)] T ,…,[ψ KM (0),…,ψ KM (N-1)] T ],
[0084] Δβ=[Δβ1,…,Δβ k ] T ; The least squares solution of Δβ can be solved, which is defined as
[0085]
[0086] Step 3.4: From the above derivation process, it can be seen that although the optimal value of the threshold cannot be directly obtained, the difference between the current threshold value and the optimized threshold value can be solved, thereby indirectly obtaining the optimal value of the threshold. When actually implementing this process, it is necessary to approximate the optimized threshold through multiple iterations, which can be expressed as
[0087]
[0088] In the formula It represents the number of iterations. The specific iteration process is as follows:
[0089] Step 3.4.1: Use iterative learning control (ILC) to search for the ideal output of the predistorter Select the initial threshold as uniform distribution, β1 = [1 / K, 2 / K, ..., 1] T , set the variable representing the number of iterations
[0090] Step 3.4.2: Output the ideal Threshold and input Substitute into formula (4) to solve the model coefficient corresponding to this iteration, denoted as
[0091] Step 3.4.3, put and input Substitute into formula (2) to calculate the model output corresponding to the current coefficient and threshold, denoted as calculate And ideal output If the error meets the requirement, the modeling is completed and the iteration ends. Otherwise, proceed to the next step.
[0092] Step 3.4.4, and Substitute into formula (11) to (14) to calculate the updated value of the threshold and record it as Set the number of iterations Then go to step 3.4.2 to start the next iteration process.
[0093] After the above iterative process, the corresponding optimized threshold value and the corresponding model coefficient can be obtained, and the obtained optimized threshold and model coefficient are used to update the predistorter to complete the entire predistortion process.
[0094] The baseband signal is an I / Q dual-tone complex signal. The test uses a 100MHz bandwidth 5G-NR signal with a peak-to-average ratio of 9.3dB and a sampling rate of 500MHz. This can well reflect the pre-distortion processing capability of the pre-distortion device for broadband signals. The specific parameter settings are as follows:
[0095] 1. Use the first and third terms in a simplified second-order DDR predistortion model based on the CPWL function.
[0096] 2. Select the model's memory depth M=4 and the threshold numbers K=4 and 10 respectively.
[0097] The tests were conducted without pre-distortion, with 4 unoptimized thresholds, 4 optimized thresholds, and 10 unoptimized thresholds. The corresponding adjacent channel leakage ratio (ACLR) results are summarized in Table 1. The power amplifier output power spectrum in each case is plotted and compared, as shown in Table 1. Figure 3 As shown. By comparison, it can be seen that when the same 4 thresholds are used, the pre-distortion performance is greatly improved when the optimized threshold proposed by the present invention is used, and is even slightly better than the performance when 10 non-optimized thresholds are used. Compared with using 10 non-optimized thresholds, the number of thresholds and coefficients of the present invention is reduced by 60%, which greatly reduces the calculation amount and power consumption of the pre-distorter.
[0098] Table 1
[0099]
[0100] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.
Claims
1. A threshold value optimization method for a power amplifier model based on a CPWL function, characterized in that: The following steps are involved: Step 1: Establish a simplified second-order DDR power amplifier model expression based on the CPWL function: In formula (1), n∈[0, N-1] represents the index position of the sampling point, and there are N sampling points in total; i is the memory depth and i∈[0,M], M is the maximum memory depth; is the input complex signal of the power amplifier, is the complex output signal of the power amplifier, || is defined as the modulus operation of complex numbers and the absolute value operation of real numbers, () * Defined as the conjugate operation on complex numbers; K is the number of segments for the nonlinear fitting interval, k∈[1, K] represents the kth segment; β k is the kth segment threshold; is the model coefficient corresponding to the first term, is the model coefficient corresponding to the second term, is the model coefficient corresponding to the third term, is the model coefficient corresponding to the fourth term; the subscript ki represents the kth segment, and the memory depth is the model coefficient corresponding to i; when the model is used, one or a combination of terms is used according to the power amplifier used and the requirements for the linearization of the power amplifier; all the following derivations are based on the scenario of using only the first-order terms of the model. Adding the corresponding terms and coefficients in formula (1) to formulas (3) to (6) can expand it to the case of using multiple terms; Step 2: Derive model coefficients The solution method is to express the model in matrix form as Y=Hc (2) Where Y represents the vector of output signals, H is the model matrix, and c is the model coefficient vector, which are defined as H=[h 10 ,…,h km ], (·) T Represents the transpose operation of a vector; where h ki It is defined as the model item corresponding to the kth segment and the memory depth i, which is expressed as The coefficients are solved using the least squares method, and the corresponding solution is expressed as In formula (4) represents the coefficient vector (·) H represents the conjugate transpose of a complex matrix, (·) -1 represents the inverse of a matrix; Step 3: Derivation of threshold optimization method: Step 3.1: Expressed as the sum of the real and imaginary parts in[·] r represents the real part of a complex number, [·] i Represents the imaginary part of a complex number; j is a unit imaginary number; Step 3.2: Define Φ ki (n) but The real part of where sgn(·) is the sign function, defined as Step 3.3: If the coefficient is the optimal solution, and the optimized threshold is expressed as The expected output corresponding to the optimized threshold is Its actual part is expressed as make definition Current Output and expected output Subtracting the real part of the error gives us the real part of the error, which is defined as Formula (11) can be expressed in matrix form as E r =-ΨΔβ (12) E in formula (12) r is the error vector, Ψ is the threshold matrix, Δβ is the threshold update vector, which are defined as Ψ=[[ψ 10 (0),…,ψ 10 (N-1)] T ,…,[ψ KM (0),…,ψ KM (N-1)] T ], (·) T represents the transpose operation of a vector; Δβ=[Δβ1,…,Δβ k ] T ; Solve for the least squares solution of Δβ, defined as Step 3.4: By iterating multiple times to approximate the optimized threshold, the difference between the current threshold value and the optimized threshold value is solved, thereby indirectly obtaining the optimized value of the threshold. The multiple iterative approximation to the optimized threshold is expressed as Where l represents the number of iterations; Step 4: Update the predistorter with the obtained optimization threshold and model coefficients to complete the entire predistortion process.
2. The threshold value optimization method for a power amplifier model based on a CPWL function according to claim 1, characterized in that: The specific iteration process in step 3.4 is as follows: Step 3.4.1: Use iterative learning control method to search for the ideal output of the predistorter Select the initial threshold as uniform distribution, β1 = [1 / K, 2 / K, ..., 1] T , set the variable l representing the number of iterations to 1; Step 3.4.2: Output the ideal Threshold β l and input Substitute into formula (4) to solve the model coefficient corresponding to this iteration, denoted as Step 3.4.3, put β l and input Substitute into formula (2) to calculate the model output corresponding to the current coefficient and threshold, denoted as calculate And ideal output If the error meets the requirement, the modeling is completed and the iteration ends; otherwise, proceed to the next step. Step 3.4.4, β l , and Substitute into formula (11) to (14) to calculate the updated value of the threshold, denoted as β l+1 , set the number of iterations l = l + 1; Then go to step 3.4.2 to start the next iteration process.
