A method for design optimization of a radio frequency power amplifier

Abstract

The application discloses a design optimization method of a radio frequency power amplifier, and relates to the technical field of design optimization. The method comprises the following steps: collecting the working frequency and the circuit topological structure of the radio frequency power amplifier, establishing a digital twin simulation model, and generating an initial impedance value; simulating to obtain a simulation performance index of the amplifier under an input signal; adjusting matching network parameters according to a dynamic reconfiguration matching network algorithm to generate a dynamic optimization impedance set; controlling a tunable element to reconfigure the actual impedance value of the radio frequency power amplifier, and collecting the actual performance index of an output signal; comparing the error between the simulation performance index and the actual performance index, if the error exceeds a preset error threshold, iteratively updating the parasitic parameters of the digital twin model; if the error is within the threshold, locking the actual impedance value as an optimal configuration, and executing input signal amplification and output. The application improves the working efficiency of the radio frequency power amplifier.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of design optimization, in particular to a design optimization method of a radio frequency power amplifier. BACKGROUND

[0002] The radio frequency power amplifier is an important component of various wireless transmitters. In the front-end circuit of the transmitter, the radio frequency signal power generated by the modulated oscillation circuit is very small, and needs to pass through a series of amplification and buffer stages, intermediate amplification stages and final power amplification stages to obtain sufficient radio frequency power, which can then be fed to the antenna for radiation. In modern wireless communication, radio frequency devices are quite popular, and the radio frequency power amplifier plays a crucial role in the devices, so the design optimization of the radio frequency power amplifier is beneficial to improving the quality of wireless communication. However, the existing radio frequency power amplifier has parasitic parameters in operation. The parasitic parameters refer to unnecessary elements existing in electronic components or circuits, which will adversely affect the normal operation of the components or circuits. The parasitic parameters will cause efficiency decline, signal distortion and stability problems, and affect the efficiency of the radio frequency power amplifier. SUMMARY

[0003] The application aims to provide a design optimization method of a radio frequency power amplifier to solve the problems in the background.

[0004] The application provides a design optimization method of a radio frequency power amplifier, which adopts the following technical scheme:

[0005] The working frequency and the circuit topology structure of the radio frequency power amplifier are collected, a digital twin simulation model is established, and an initial impedance value is generated according to the digital twin simulation model;

[0006] The input signal is loaded to the digital twin simulation model, and a simulation performance index of the amplifier under the input signal is obtained through simulation;

[0007] The initial impedance value is taken as a search starting point, the matching network parameters are iteratively adjusted in a preset impedance range by combining a dynamic reconstruction matching network algorithm, and a dynamic optimization impedance set is generated;

[0008] Based on the dynamic optimization impedance set, the actual impedance value of the radio frequency power amplifier is reconstructed by controlling the tunable element, and the actual performance index of the output signal is collected;

[0009] The simulation performance index and the actual performance index are compared to determine whether the error exceeds a preset error threshold, and if the error exceeds the error threshold, the parasitic parameters of the digital twin simulation model are iteratively updated;

[0010] If the error is within the threshold, the actual impedance value is locked as the optimal configuration, the optimal configuration is solidified to the radio frequency power amplifier, input signal amplification is performed and a target power signal is output.

[0011] Preferably, the working frequency and circuit topology of the acquisition radio frequency power amplifier are used to establish a digital twin simulation model, and the initial impedance value is generated according to the digital twin simulation model.

[0012] The working frequency range and circuit topology of the radio frequency power amplifier are obtained, a physical model of the radio frequency power amplifier is established by a three-dimensional electromagnetic simulation software, and the parasitic parameters are extracted;

[0013] The parasitic parameters are imported into a circuit simulation tool to build a digital twin simulation model, a load-pulling test environment is set in the digital twin model, an output impedance plane is scanned, an impedance region meeting a preset standard is selected, and a candidate initial impedance value set is obtained.

[0014] The power matching is used as an optimization target, and a genetic algorithm is used to select an impedance value meeting a preset requirement from the candidate initial impedance value set to generate an initial impedance value.

[0015] Preferably, the parasitic parameters are imported into a circuit simulation tool to build a digital twin simulation model, a load-pulling test environment is set in the digital twin model, an output impedance plane is scanned, an impedance region meeting a preset standard is selected, and a candidate initial impedance value set is obtained.

[0016] A programmable impedance tuner module is inserted into the digital twin model, an adjustment resolution of a reflection coefficient is dynamically adjusted according to an adaptive step algorithm, and a load-pulling test environment is set according to the adjustment resolution.

[0017] Load-pulling tests are performed according to the load-pulling test environment, and an impedance plane is obtained according to test results.

[0018] A cloud computing platform is used to deploy distributed simulation tasks, the impedance plane is divided into multiple sub-regions, and each sub-region is allocated an independent computing node for parallel simulation.

[0019] Performance indicators of each impedance point are collected in real time to form a corresponding impedance-performance mapping curve.

[0020] Performance filtering criteria are set, and impedance regions meeting the performance filtering criteria are selected in combination with the impedance-performance mapping curve.

[0021] Impedance points meeting preset conditions are searched from the impedance regions to form a candidate initial impedance set.

[0022] Preferably, the adjustment resolution of the reflection coefficient is dynamically adjusted according to the adaptive step algorithm.

[0023] Set the initial load reflection coefficient, and extract the real part R and the imaginary part I in the initial load reflection coefficient complex form;

[0024] The resolution of the grid division in the scanning range is denoted as a step size, and an initial step size Δ0 is set;

[0025] According to the gradient threshold formula The gradient threshold G is calculated th , where N is the reference efficiency threshold, is the derivative of the PAE change rate with respect to the reflection coefficient;

[0026] The power added efficiency formula is: , where P out is the radio frequency output power, P in is the radio frequency input power, and P DC is the direct current input power;

[0027] The real part direction gradient formula is:

[0028]

[0029] , where G R is the real part direction gradient, PAE(R+Δ0, I) represents the power added efficiency value calculated under the condition that the real part of the reflection coefficient is increased by an initial step size and the imaginary part remains unchanged, and PAE(R-Δ0, I) represents the power added efficiency value calculated under the condition that the real part of the reflection coefficient is decreased by an initial step size and the imaginary part remains unchanged;

[0030] The imaginary part direction gradient formula is:

[0031]

[0032] , where G I is the imaginary part direction gradient, PAE(R, I+Δ0) represents the power added efficiency value calculated under the condition that the imaginary part of the reflection coefficient is increased by an initial step size and the real part remains unchanged, and PAE(R, I-Δ0) represents the power added efficiency value calculated under the condition that the imaginary part of the reflection coefficient is decreased by an initial step size and the real part remains unchanged;

[0033] The total gradient G is calculated according to the total gradient formula , and the gradient threshold G th is combined to obtain the real-time step size Δ new , which is the adjustment resolution of the reflection coefficient. The real-time step size equation is:

[0034]

[0035] Preferably, the power matching optimization target is used to screen the impedance values meeting the preset requirements from the candidate initial impedance value set by using a genetic algorithm, and the initial impedance value generating step is specifically:

[0036] The fitness function is established to calculate the fitness F(Z) of each candidate initial impedance value in the candidate initial impedance value set {Z1, Z2, Z3…Z N};

[0037] The source end reflection coefficient Γ S and the load end reflection coefficient Γ L of each impedance are calculated, and the calculation formulas are as follows: Wherein, Z0 is the system characteristic impedance, Z L is the load impedance;

[0038] The fitness function is as follows: The impedance value with the fitness F(Z) reaching the preset fitness threshold is screened as the initial impedance population;

[0039] The parent impedance is selected from the initial impedance population according to the roulette wheel selection mechanism, and the impedance value with the highest fitness is selected as the initial impedance value after the parent impedance is crossed and mutated.

[0040] Preferably, the step of loading the input signal to the digital twin simulation model to simulate the simulation performance index of the amplifier under the input signal is specifically:

[0041] The actual parameters of the actual radio frequency power amplifier are collected, the input signal is loaded to the digital twin simulation model, and the model parameters are configured to be synchronized with the actual parameters;

[0042] The simulation is started, the node data of the internal nodes of the radio frequency power amplifier are collected in real time, the transient response and steady state response time series data are captured, and the simulation data are obtained;

[0043] The simulation performance index value of the amplifier under the input signal is calculated based on the simulation data.

[0044] Preferably, the step of taking the initial impedance value as the search starting point, combining the dynamic reconstruction matching network algorithm, and iteratively adjusting the matching network parameters in the preset impedance range to generate a dynamic optimization impedance set is specifically:

[0045] According to the simulation performance index output by the digital twin model and in combination with the preset performance index threshold, it is determined whether the dynamic reconstruction matching network algorithm is triggered;

[0046] If the dynamic reconstruction matching network algorithm is triggered, the initial impedance value is taken as the search starting point to set the impedance range;

[0047] The matching network parameters are iteratively adjusted in the impedance range, so that the output impedance Zout approximating the target impedance Z target ;

[0048] The adjusted matching network parameters are verified in real time through the digital twin model, the circuit stability and heat distribution state are detected, it is judged whether the circuit is stable, and if the circuit is not stable, the matching network parameters are re-iterated and adjusted;

[0049] If the circuit is stable, the impedance values that make the fitness function F(Z) reach the preset fitness threshold are screened according to the fitness function, and a dynamic optimization impedance set {Z y1 , Z y2 , Z y3 … Z yN} is generated.

[0050] Preferably, the matching network parameters are iteratively adjusted in the impedance range, so that the output impedance Z out approximates the target impedance Z target .

[0051] The target impedance formula is: Z targ et = argmin(|Γ S Γ L | + λ·|P loss |), wherein P loss is the loss power of the matching network, and λ is a weight factor.

[0052] The calculation formula of the loss power P loss of the matching network is: P loss = |I out | 2 · DR, wherein I out is the current at the output end of the matching network, and DR is the parasitic resistance.

[0053] Preferably, the step of verifying the adjusted matching network parameters in real time through the digital twin model, detecting the circuit stability and heat distribution state, and judging whether the circuit is stable is specifically:

[0054] A stability factor K is calculated according to a stability function, and the stability function is:

[0055]

[0056] When K>1, it is judged that the circuit has no self-excited oscillation, and the circuit satisfies the first stability condition.

[0057] The transistor junction temperature T j is calculated, and the transistor junction temperature calculation formula is: T j = T c + P j · rR j , wherein T cis the transistor package surface temperature, P j is the transistor power consumption, rR j thermal resistance of the junction to the environment;

[0058] a transistor junction temperature threshold T max , if T j ≤ T max , it is judged that the heat distribution is qualified, and the circuit meets the second stability condition;

[0059] If the circuit meets the first stability condition and the second stability condition at the same time, it is judged that the circuit is stable, otherwise it is judged that the circuit is unstable.

[0060] Preferably, the step of controlling the actual impedance value of the tunable element to reconstruct the radio frequency power amplifier based on the dynamically optimized impedance set, and collecting the actual performance index of the output signal, specifically comprises:

[0061] Based on the dynamically optimized impedance set, an application requirement is obtained, and the best impedance is selected from the dynamically optimized impedance set according to the application requirement;

[0062] A tunable element array is constructed and integrated in the output stage of the radio frequency power amplifier. The actual output impedance of the radio frequency power amplifier is adjusted to be consistent with the best impedance by dynamically adjusting the tunable element, so as to obtain the actual impedance value;

[0063] Real-time working data of the radio frequency power amplifier is collected, and the actual performance index of the radio frequency power amplifier is obtained according to the real-time working data.

[0064] In summary, the present application includes at least one of the following beneficial technical effects:

[0065] 1. By adopting digital twin simulation technology and dynamic impedance matching strategy, the intelligent design and significant optimization of the performance of the radio frequency power amplifier are successfully realized. The dynamic matching network can adjust the impedance parameters in real time, effectively reduce the reflection loss and nonlinear distortion, and significantly improve the output power and working efficiency. The introduction of the digital twin model replaces the traditional trial-and-error design method, greatly reduces the cost and debugging period of physical experiments. In addition, the dynamically reconstructed matching network has excellent adaptability and can meet the diversified needs of different frequencies and signal modulation, not only expanding the application range of the radio frequency power amplifier, but also further improving the working efficiency of the radio frequency power amplifier.

[0066] 2. Extracting parasitic parameters (such as lead inductance, distributed capacitance) through three-dimensional electromagnetic simulation solves the model misalignment problem caused by traditional circuit simulation ignoring high-frequency parasitic effects, reduces the error between the digital twin simulation model and the actual physical device. Combined with load traction and genetic algorithm, the optimal impedance point can be quickly screened in the virtual environment, avoiding the time-consuming problem of traditional trial-and-error experiments, and shortening the research and development cycle. And by introducing an adaptive step algorithm, not only the accuracy is improved, but also the speed is accelerated, improving the speed of design and optimization simulation of radio frequency power amplifiers.

[0067] 3. The dynamic reconfiguration matching network algorithm significantly improves the efficiency, linearity and multi-scene adaptability of the radio frequency power amplifier (PA) by adjusting the impedance matching state of the radio frequency power amplifier (PA) in real time. The technology combines intelligent control mechanism and tunable elements to dynamically reconfigure the matching network parameters, so that the output impedance always approximates the target impedance. Dynamic adjustment of the output impedance improves the power added efficiency and reduces the total harmonic distortion (THD). Through thermal distribution state detection, the device overheating or self-oscillation caused by impedance mismatch is avoided. Combined with the K factor and the junction temperature threshold, the final stability conclusion is output, and double verification is ensured to ensure long-term reliable operation of the circuit, improving the stability and reliability of the design and optimization simulation of the radio frequency power amplifier. BRIEF DESCRIPTION OF DRAWINGS

[0068] Figure 1 is a specific step schematic diagram of an embodiment of a design optimization method of a radio frequency power amplifier. DETAILED DESCRIPTION

[0069] The embodiments of the present application will be described below in conjunction with the accompanying drawings and specific embodiments. Figure 1 The embodiments of the present application will be described below in conjunction with the accompanying drawings and specific embodiments.

[0070] The present application discloses a design optimization method of a radio frequency power amplifier, specifically comprising the following steps:

[0071] Step S1, collect the working frequency and circuit topology of the radio frequency power amplifier, establish a digital twin simulation model, and generate an initial impedance value according to the digital twin simulation model.

[0072] Step S2, load the input signal to the digital twin simulation model, and simulate to obtain the simulation performance index of the amplifier under the input signal.

[0073] Step S3, taking the initial impedance value as the search starting point, combining the dynamic reconfiguration matching network algorithm, iteratively adjusting the matching network parameters within the preset impedance range, and generating a dynamic optimization impedance set.

[0074] Step S4, based on the dynamic optimization impedance set, control the tunable elements to reconfigure the actual impedance value of the radio frequency power amplifier, and collect the actual performance index of the output signal.

[0075] Step S5, compare the simulation performance indicators with the actual performance indicators to determine whether the error exceeds the preset error threshold. If the error exceeds the error threshold, iteratively update the parasitic parameters of the digital twin simulation model.

[0076] Define the key performance indicator error threshold (such as third-order intermodulation distortion IMD3, power added efficiency PAE, gain error, etc.), for example, the threshold of IMD3 is set to -40dBc. Collect the frequency spectrum, power, and efficiency data of the actual output signal through an analog-to-digital converter (ADC), and compare them with the simulation results of the digital twin model item by item. Use the least squares error algorithm or weighted error algorithm to calculate the comprehensive error value and determine whether it exceeds the preset threshold. Performance indicators include but are not limited to power gain, bandwidth, linearity, etc. Parasitic parameters include lead inductance, distributed capacitance, package parasitic resistance, etc. For example, if the actual output power is lower than the simulation value, increase the simulation value of the drain lead inductance in the model and recalculate the impedance matching network. By comparing the simulation and actual performance, identify the error of the parasitic parameters (such as distributed capacitance, lead inductance), and update the digital twin model in reverse. This can improve the accuracy of the model and ensure the reliability of the subsequent optimization process, forming a closed-loop correction mechanism.

[0077] Step S6, if the error is within the threshold, lock the actual impedance value as the optimal configuration, and solidify the optimal configuration to the RF power amplifier to perform input signal amplification and output target power signals.

[0078] Write the optimal value in the dynamic optimization impedance set to the FPGA or microcontroller register through tunable elements such as varactor diodes and MEMS switches. Use non-volatile memory such as EEPROM to store the optimal impedance parameters to ensure that the configuration is not lost after power failure. When solidifying the impedance of high-frequency circuits, use laser trimming or pad jumpers to fix the capacitance / inductance values in the matching network. On the premise of ensuring performance, reduce the complexity of dynamic adjustment, and improve the stability of the RF power amplifier.

[0079] In actual use, through digital twin simulation and dynamic impedance matching technology, intelligent design and optimization of RF power amplifier performance is achieved. By dynamically adjusting the impedance through the dynamic matching network, the reflection loss and nonlinear distortion are reduced, and the output power and efficiency are optimized. The digital twin model replaces the traditional trial-and-error design, reducing the cost and debugging time of physical experiments. The dynamically reconstructed matching network can adapt to different frequency and signal modulation requirements, expanding the application range and improving the working efficiency of the RF power amplifier.

[0080] Collect the working frequency and circuit topology of the RF power amplifier, establish a digital twin simulation model, and generate initial impedance values based on the digital twin simulation model. The steps are as follows:

[0081] Step S11, the working frequency range and circuit topology of the radio frequency power amplifier are acquired, a physical model of the radio frequency power amplifier is established through three-dimensional electromagnetic simulation software, and parasitic parameters are extracted.

[0082] ANSYS Q3D Extractor can be used to model the packaging structure and bonding wires of the radio frequency power amplifier in three dimensions, and the parasitic parameters in the equivalent circuit network are extracted through the finite element method (FEM). The parasitic parameters directly affect the impedance matching and stability of high-frequency circuits. For example, the drain lead inductance and gate distributed capacitance can introduce resonance points, resulting in power backoff or self-excitation. The electromagnetic field simulation of the physical model can accurately capture the near-field effects of microstrip lines, vias and other structures, and make up for the shortcomings of traditional SPICE models.

[0083] Step S12, the parasitic parameters are imported into the circuit simulation tool, a digital twin simulation model is built, a load-pulling test environment is set in the digital twin model, the output impedance plane is scanned, and the impedance region that meets the preset standard is selected to obtain a candidate initial impedance value set.

[0084] Step S13, taking power matching as the optimization target, the genetic algorithm is used to select the impedance value that meets the preset requirements from the candidate initial impedance value set to generate an initial impedance value.

[0085] In actual application, the parasitic parameters (such as lead inductance and distributed capacitance) are extracted through three-dimensional electromagnetic simulation, which solves the model inaccuracy problem caused by the traditional circuit simulation ignoring high-frequency parasitic effects, reduces the error between the digital twin simulation model and the actual physical device. Combined with load pulling and genetic algorithm, the optimal impedance point can be quickly selected in the virtual environment, avoiding the time-consuming problem of traditional trial-and-error experiments, and the research and development cycle can be shortened. The genetic algorithm optimization based on the power matching target balances efficiency, linearity and bandwidth, etc., so that the output power additional efficiency is improved by 15%-20%, and the harmonic distortion is suppressed.

[0086] The step of importing the parasitic parameters into the circuit simulation tool, building a digital twin simulation model, setting a load-pulling test environment in the digital twin model, scanning the output impedance plane, selecting an impedance region that meets the preset standard, and obtaining a candidate initial impedance value set is as follows:

[0087] Step S121, a programmable impedance tuner module is inserted into the digital twin model, the adjustment resolution of the reflection coefficient is dynamically adjusted according to the adaptive step algorithm, and the load-pulling test environment is set according to the adjustment resolution.

[0088] Integrating programmable impedance tuners (such as MEMS switch arrays or varactor diode matrices) in the digital twin model and dynamically adjusting the resolution of the reflection coefficient through an adaptive step algorithm can effectively reduce the time waste of the load-pulling test environment. Avoid overfitting or missed detection problems caused by traditional fixed step, such as achieving high-precision matching near the resonance point.

[0089] Step S122, performing load-pulling test according to the load-pulling test environment, and obtaining an impedance plane according to the test results.

[0090] For example, use Keysight ADS software to build a load-pulling environment, scan the frequency range of 2.4-5GHz (step size 10MHz), and compensate for the lead inductance of 0.5nH and the distributed capacitance of 0.2pF. Generate Smith circle impedance plane, mark the high-efficiency area (real part 10-20Ω, imaginary part -5j+5j) with PAE>45%.

[0091] Step S123, deploying distributed simulation tasks using a cloud computing platform, dividing the impedance plane into multiple sub-regions, and assigning independent computing nodes to each sub-region for parallel simulation.

[0092] Divide the impedance plane into grid sub-regions (such as 10x10), and perform parallel simulation through cloud computing clusters. For example, based on a Kubernetes cluster, schedule 100 computing nodes, each node is assigned 4-core vCPU to process a sub-region (such as 20% resources are allocated to the frequency band of 2.4-3GHz). Use Docker containers to encapsulate ADS simulation tasks, and use gRPC protocol to realize data synchronization between nodes.

[0093] Step S124, real-time collection of performance indicators of each impedance point and formation of corresponding impedance-performance mapping curve.

[0094] Real-time collection of performance indicators (gain, PAE, THD) of each impedance point, construction of multi-dimensional database and fitting of continuous curve.

[0095] Step S125, setting performance screening criteria, and screening impedance regions that meet the performance screening criteria in combination with the impedance-performance mapping curve.

[0096] Set dynamic screening criteria (such as PAE>45%, THD<-40dBc), and lock candidate regions through clustering algorithm.

[0097] Step S126, searching for impedance points that meet the preset conditions from the impedance regions to form a candidate initial impedance set.

[0098] Balance efficiency and bandwidth, for example, set the real part range and imaginary part range of the impedance point, and select the impedance points that meet the range of the real part and imaginary part as the candidate initial impedance points to form the set.

[0099] In actual application, the resolution of the reflection coefficient is dynamically adjusted through the programmable impedance tuner module and the adaptive step algorithm to solve the local optimal trap problem caused by the traditional fixed step. The impedance plane is simulated in a distributed and parallel manner using a cloud computing platform, the single simulation time is compressed from hours to minutes, and the rapid screening of millions of impedance points is supported. Based on the impedance-performance mapping curve, a dynamic screening standard (such as PAE>45%, THD<-40dBc) is set to realize the collaborative optimization of efficiency, linearity, bandwidth and other indicators. By dynamically dividing the impedance sub-region, the impedance matching requirements of 5G multi-band, millimeter wave and other high frequency / broadband scenarios are adapted, and the adaptability of the radio frequency front end is expanded.

[0100] The step of dynamically adjusting the adjustment resolution of the reflection coefficient according to the adaptive step algorithm is specifically:

[0101] In step S1211, an initial load reflection coefficient is set, and the real part R and the imaginary part I in the complex form of the initial load reflection coefficient are extracted.

[0102] For example, the initial reflection coefficient is set to 0.5+j0.3 (corresponding to impedance Z=50Ω) in the digital twin model. The real part R=0.5 and the imaginary part I=0.3 are extracted as the reference point for subsequent gradient calculation.

[0103] In step S1212, the resolution of the grid division in the scanning range is denoted as a step, and an initial step Δ0 is set.

[0104] The initial step needs to cover the key area of the impedance plane (such as the center area of the Smith chart).

[0105] In step S1213, the gradient threshold G is calculated according to the gradient threshold formula th where N is the reference efficiency threshold, is the derivative of the PAE change rate with respect to the reflection coefficient.

[0106] The gradient threshold is used to determine whether the step needs to be refined or coarsened in the current area, and the resolution is dynamically adjusted. For example, the reference efficiency threshold is 60%, the current PAE is 55%, and the derivative of the PAE change rate with respect to the reflection coefficient is 0.1, so the gradient threshold is 0.5.

[0107] In step S1214, the power added efficiency formula is: where P out is the radio frequency output power, P in is the radio frequency input power, and P DC is the direct current input power.

[0108] ​Quantify the efficiency of the amplifier, provide an optimization target for gradient calculation, PAE directly reflects the energy conversion efficiency, and is the core index of matching network optimization.

[0109] Step S1215, the real part direction gradient formula is:

[0110]

[0111] Where G R is the real part direction gradient, PAE(R+Δ0, I) represents the calculated power added efficiency value under the condition that the real part of the reflection coefficient is increased by an initial step and the imaginary part remains unchanged, and PAE(R-Δ0, I) represents the calculated power added efficiency value under the condition that the real part of the reflection coefficient is decreased by an initial step and the imaginary part remains unchanged.

[0112] Evaluate the impact of real part changes on PAE to guide impedance adjustment direction. A positive gradient indicates that increasing the real part can improve efficiency, and vice versa.

[0113] Step S1216, the imaginary part direction gradient formula is:

[0114]

[0115] Where G I is the imaginary part direction gradient, PAE(R, I+Δ0) represents the calculated power added efficiency value under the condition that the imaginary part of the reflection coefficient is increased by an initial step and the real part remains unchanged, and PAE(R, I-Δ0) represents the calculated power added efficiency value under the condition that the imaginary part of the reflection coefficient is decreased by an initial step and the real part remains unchanged.

[0116] Evaluate the impact of real part changes on PAE to guide impedance adjustment direction. A positive gradient indicates that increasing the real part can improve efficiency, and vice versa.

[0117] Step S1217, according to the total gradient formula Calculate the total gradient G, combine the gradient threshold G th Get the real-time step size Δ new , as the adjustment resolution of the reflection coefficient, the real-time step size equation is:

[0118]

[0119] In practical application, the resolution of reflection coefficient is dynamically adjusted by the adaptive step algorithm, and the precise tuning and efficiency optimization of the RFPA impedance matching network are realized. The step is automatically adjusted according to the gradient sensitivity (such as 0.05Ω in the high sensitivity area and 0.2Ω in the low sensitivity area), which solves the local optimum or calculation redundancy problem caused by the traditional fixed step. Taking the power added efficiency (PAE) as the optimization target, combined with the gradient threshold judgment, the PAE is improved, and the total harmonic distortion (THD <-40dBc) is suppressed. The optimal impedance point is quickly converged through the gradient direction search, and the matching network tuning time is shortened. The real and imaginary parts of the gradient are combined to dynamically adjust the step to optimize the search efficiency. The step is refined in the high gradient area to improve the precision, and the step is coarsened in the low gradient area to speed up the search.

[0120] Taking the power matching as the optimization target, the genetic algorithm is used to select the impedance values meeting the preset requirements from the candidate initial impedance value set to generate the initial impedance value step, which is specifically:

[0121] Step S131, the fitness function is established to calculate the fitness F(Z) of each candidate initial impedance value in the candidate initial impedance value set {Z1, Z2, Z3…Z N}.

[0122] Step S132, the source end reflection coefficient Γ s and the load end reflection coefficient Γ L of each impedance are calculated, and the calculation formulas are as follows: Where Z0 is the system characteristic impedance, and Z L is the load impedance.

[0123] Step S133, the fitness function is: The impedance value with the fitness F(Z) reaching the preset fitness threshold is selected as the initial impedance population.

[0124] Step S134, the parent impedance is selected from the initial impedance population according to the roulette selection mechanism, and the impedance value with the highest fitness after the parent impedance is crossed and mutated is selected as the initial impedance value.

[0125] The probability of selecting a highly adaptable individual is generated by generating a more optimal offspring impedance. The natural evolution process is simulated to avoid local optimization and achieve global search. For example, the candidate impedance set is {50Ω, 75Ω, 100Ω}, and the corresponding fitness is F(50) = 0.8, F(75) = 1.06, and F(100) = 0.5. The total fitness is 0.8 + 1.06 + 0.5 = 2.36, and the individual probabilities are P(50) = 0.8 / 2.36 ≈ 34%, P(75) = 1.06 / 2.36 ≈ 45%, and P(100) = 0.5 / 2.36 ≈ 21%. A random number (such as 0.5) is generated, which falls in the 45% interval (corresponding to 75Ω), and is selected as parent 1. A random number (such as 0.8) is generated again, which falls in the 21% interval (corresponding to 100Ω), and is selected as parent 2. The parent impedance is 75Ω (binary: 1001011) and 100Ω (binary: 1100100).

[0126] Random crossover point (such as the 4th bit):

[0127] Offspring 1: 1000100 → 1000100 (68Ω)

[0128] Offspring 2: 1101011 → 1101011 (107Ω)

[0129] Mutation operation (non-uniform mutation):

[0130] Offspring 1 (68Ω) randomly mutates the 3rd bit: 100 → 101, becoming 72Ω.

[0131] Offspring 2 (107Ω) remains the same.

[0132] Fitness screening:

[0133] Calculate offspring fitness:

[0134] F(72) = 1 / |1-(0.18×0.22)| 2 ≈1.15

[0135] F(107) = 1 / |1-(0.36×0.03)| 2 ≈1.12

[0136] Select the highest fitness offspring (72Ω, F = 1.15) to join the new generation population.

[0137] In practical application, the genetic algorithm and the impedance matching optimization driven by the reflection coefficient are used to realize the rapid design and performance improvement of the RFPA impedance network. The source end and the load end reflection coefficients are combined to balance the power transmission efficiency and the standing wave ratio, so that the output power additional efficiency is improved. The crossover and mutation operations of the genetic algorithm are used to avoid the local optimum of the traditional trial-and-error method and cover a wider impedance matching area. The roulette wheel selection mechanism is used to select the parent impedance, reduce the invalid search, and optimize and shorten the time.

[0138] The step of loading the input signal to the digital twin simulation model to obtain the simulation performance index of the amplifier under the input signal is specifically as follows:

[0139] In step S21, the actual parameters of the actual RFPA are collected, the input signal is loaded to the digital twin simulation model, and the model parameters are configured to be synchronized with the actual parameters.

[0140] In step S22, the simulation is started, the node data of the internal nodes of the RFPA are collected in real time, the transient response and steady-state response time series data are captured, and the simulation data are obtained.

[0141] For example, the ADS transient simulation is started, the sampling rate is set to 10 GSa / s, the probe is inserted at the drain node, the current overshoot peak value is 2.3 A (the steady-state value is 1.8 A), the rise time is 1.2 ns, the output power is 38.2 dBm, and the efficiency is 43%. The simulation data can be adjusted autonomously according to user requirements and can include current, voltage, etc.

[0142] In step S23, the simulation performance index value of the amplifier under the input signal is calculated based on the simulation data.

[0143] Different simulation performance indexes have their corresponding performance index calculation methods. For example, when the input power is 30 dBm, the gain is Gain=38.2 dBm-30 dBm=8.2 dB, and the power additional efficiency is calculated by The third-order intermodulation component is extracted by FFT to be -25 dBc.

[0144] In practical application, the digital twin technology is used to realize the virtual and real synchronous simulation of the RFPA, solve the problems of low efficiency and high cost in traditional physical debugging, map the bias voltage, load impedance and other parameters of the actual PA to the twin model in real time, improve the simulation accuracy, optimize the whole life cycle, capture the transient response characteristics (such as rise time and overshoot), predict the power amplifier thermal failure risk, and replace the physical experiment with virtual debugging to shorten the research and development cycle.

[0145] The step of taking the initial impedance value as the search starting point, combining the dynamic reconfiguration matching network algorithm, and iteratively adjusting the matching network parameters in the preset impedance range to generate a dynamic optimized impedance set is specifically as follows:

[0146] Step S31, according to the simulation performance index output by the digital twin model, combined with the preset performance index threshold, determine whether to trigger the dynamic reconstruction matching network algorithm.

[0147] According to the simulation performance index and the preset threshold, it is judged whether the impedance optimization needs to be started. Avoid invalid tuning, trigger the algorithm only when the performance does not meet the standard, save the computing power. For example, the digital twin model shows that PAE=35%(threshold is 40%) and gain G=8dB(threshold is 10dB), so it does not reach the performance index threshold, and the dynamic reconstruction matching network algorithm needs to be triggered.

[0148] Step S32, if the dynamic reconstruction matching network algorithm is triggered, set the impedance range with the initial impedance value as the search starting point.

[0149] With the initial impedance value as the search starting point, define the tuning range. It can limit the search space and accelerate convergence (such as reducing from the full impedance plane to the local sensitive area).

[0150] Step S33, iteratively adjust the matching network parameters in the impedance range, so that the output impedance Z out approximates the target impedance Z target .

[0151] Step S34, verify the adjusted matching network parameters in real time through the digital twin model, detect the circuit stability and thermal distribution state, and judge whether the circuit is stable. If the circuit is not stable, re-iterate the adjustment of the matching network parameters.

[0152] Through the digital twin model to detect whether the circuit is stable, it can avoid hardware damage and ensure long-term reliable operation.

[0153] Step S35, if the circuit is stable, according to the fitness function, filter the impedance values that reach the preset fitness threshold, and generate a dynamic optimization impedance set {Z y1 , Z y2 , Z y3 …Z yN}.

[0154] In actual application, through the digital twin model and the dynamic reconstruction matching network algorithm, the closed-loop optimization of the impedance parameters of the radio frequency power amplifier (PA) is realized, and the circuit stability and performance are improved. Dynamically adjust the output impedance to improve the power added efficiency and reduce the total harmonic distortion (THD). Through the thermal distribution state detection, avoid the device overheating or self-oscillation caused by impedance mismatch. Replace the traditional trial-and-error method to shorten the matching network design cycle. In the case of stable circuit, according to the actual situation, filter the impedance values that meet the preset fitness threshold, provide multiple groups of alternative impedance, and adapt to the dynamic environment (such as temperature drift and frequency band switching).

[0155] Iteratively adjust the matching network parameters within the impedance range to make the output impedance Z out Approximating the target impedance Z target The steps are as follows:

[0156] Step S331, the target impedance formula is: Z targ et =argmin(|Γ S Γ L |+λ·|P l |), where P loss It represents the power loss of the matching network, and λ is the weighting factor.

[0157] Define an optimization objective that balances the weighting relationship between the reflection coefficient and power loss. This is achieved by minimizing |Γ... S Γ L |+λ·|P loss This achieves an optimal trade-off between efficiency and stability. The source-end reflection coefficient can be measured using a Vector Network Analyzer (VNA). The `arg min` function is a minimum value solving function, primarily used to find the minimum value of a set of numbers.

[0158] Step S332, Match network loss power P loss The calculation formula is: P loss =|I out | 2 ·DR, where I out DR is the parasitic resistance used to match the current at the network output.

[0159] Quantifying the energy loss of the matching network provides data for optimization, and reducing losses can improve overall efficiency. The output current of the matching network can be measured using a current probe, and the parasitic resistance can be measured using an LCR meter.

[0160] In practical applications, by iteratively adjusting the matching network parameters, the output impedance is brought close to the target impedance, maximizing power transmission efficiency and minimizing losses. Power amplifier efficiency is improved by balancing the reflection coefficient and power loss. The risk of parasitic oscillations and thermal failure caused by impedance mismatch is suppressed, and dynamic weighting factor adjustment is supported to adapt to different frequency bands and power level requirements. In scenarios such as 5G base stations and millimeter-wave radar, matching network losses can be reduced. Simulation accuracy errors are reduced through closed-loop calibration using virtual and real data. For example, PAE is optimized in the 3.5GHz band to reduce losses. S =0.5, Γ L =0.6, λ=0.8, I out =2A, DR=0.5Ω, then Z targ et =argmin(|0.5×0.6|+0.8·|2) 2×0.5|)=arg min(1.9), through genetic algorithm search, the target impedance is locked at 48+j2Ω, and PAE is increased from 40% to 46%. In the 28GHz band, when the target impedance is 52-j3Ω, the temperature rise is reduced from 25℃ to 18℃.

[0161] The steps for verifying the adjusted matching network parameters in real time using a digital twin model, detecting circuit stability and thermal distribution, and determining circuit stability are as follows:

[0162] Step S341: Calculate the stability factor K based on the stability function, which is:

[0163]

[0164] When K > 1, it is determined that the circuit has no self-excited oscillation and the circuit meets the first stability condition.

[0165] The stability factor K is used to determine whether the circuit is likely to oscillate (K>1 indicates stability), thus avoiding circuit instability caused by reflection coefficient mismatch.

[0166] Step S342, calculate the transistor junction temperature T j The formula for calculating transistor junction temperature is: T j =T c +P j ·rR j T c It is the surface temperature of the transistor package, P j It refers to transistor power consumption, rR j The thermal resistance of the junction to the environment.

[0167] Junction-to-environment thermal resistance refers to the total thermal resistance of heat transfer from the junction region inside the transistor to the surrounding environment. Quantifying the thermal distribution of the transistor and preventing performance degradation or burnout due to overheating is crucial; junction temperature is a core indicator of device lifespan and reliability. Thermistors can be used to measure the surface temperature of the transistor package, power meters to measure transistor power dissipation, and datasheets to determine the thermal resistance.

[0168] Step S343: Set the transistor junction temperature threshold T max If T j ≤T max If the heat distribution is satisfactory, the circuit meets the second stability condition.

[0169] Step S344: If the circuit simultaneously satisfies the first stability condition and the second stability condition, then the circuit is determined to be stable; otherwise, the circuit is determined to be unstable.

[0170] In actual application, the digital twin model is used to verify the matching network parameters in real time, detect the circuit stability and thermal distribution state, and ensure the reliable operation of the radio frequency power amplifier. Through the dual judgment of the stability factor and the junction temperature, the self-oscillation of the circuit and the overheating failure are avoided. The risk of device damage caused by self-oscillation is reduced, and the service life of the power amplifier is prolonged through thermal distribution optimization.

[0171] Based on the dynamically optimized impedance set, the step of controlling the tunable element to reconstruct the actual impedance value of the radio frequency power amplifier and collecting the actual performance index of the output signal is specifically:

[0172] In step S41, based on the dynamically optimized impedance set, the application requirement is obtained, and the best impedance is selected from the dynamically optimized impedance set according to the application requirement.

[0173] According to the application requirement (such as efficiency priority or linearity priority), the optimal impedance is selected from the dynamically optimized set, blind tuning is avoided, and the target impedance is matched with the application scenario. For example, the 5G base station PA needs to achieve high efficiency at 3.5 GHz frequency band, and the impedance with the maximum PAE is selected as the best impedance.

[0174] In step S42, the tunable element array is constructed and integrated in the output stage of the radio frequency power amplifier. The tunable element is dynamically adjusted to adjust the actual output impedance of the radio frequency power amplifier to be consistent with the best impedance, and the actual impedance value is obtained.

[0175] The PA output impedance is adjusted to the target value by the tunable element (such as MEMS switch, varactor diode), the impedance is accurately matched, and the reflection loss is suppressed. For example, the adjustable matching network topology: π-type network (2 adjustable capacitors c1, c2, 1 adjustable inductor L), the best impedance Z = 47 + j1 Ω. The matching network parameters are calculated: At 3.5 GHz (ω = 2π × 3.5e9), c1 = 2.1pF, L = 3.5nH, and c2 = 1.8pF are obtained, where J is the imaginary unit. The bias voltage of the varactor diode is controlled by the FPGA to adjust c1 and c2 to the target value.

[0176] In step S43, the real-time working data of the radio frequency power amplifier is collected, and the actual performance index of the radio frequency power amplifier is obtained according to the real-time working data.

[0177] The actual performance (efficiency, linearity, stability) of the tuned radio frequency power amplifier is measured, the optimization effect is verified, the closed-loop feedback ensures the reliability of the tuning result, and dynamic re-optimization is supported.

[0178] In actual application, the output impedance of the radio frequency power amplifier is adjusted in real time by dynamically optimizing the impedance set and the array of tunable elements, so as to realize the collaborative optimization of efficiency, linearity and stability. The optimal impedance is selected according to application requirements (such as frequency band, power level), and the 5G / 6G multi-scene requirements are adapted.

[0179] The above are preferred embodiments of the present application, and do not limit the protection scope of the present application, so that: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method of design optimization of a radio frequency power amplifier, characterized by, The method comprises the following steps: Collecting the working frequency and circuit topology of the radio frequency power amplifier, establishing a digital twin simulation model, and generating an initial impedance value according to the digital twin simulation model; Loading the input signal into the digital twin simulation model to obtain the simulation performance index of the amplifier under the input signal; Taking the initial impedance value as the search starting point, combining the dynamic reconfiguration matching network algorithm, and iteratively adjusting the matching network parameters within the preset impedance range to generate a dynamic optimization impedance set; Based on the dynamic optimization impedance set, the actual impedance value of the radio frequency power amplifier is controlled by the tunable element, and the actual performance index of the output signal is collected; Comparing the simulation performance index with the actual performance index to determine whether the error exceeds the preset error threshold. If the error exceeds the error threshold, iteratively update the parasitic parameters of the digital twin simulation model; If the error is within the threshold, the actual impedance value is locked as the optimal configuration, and the optimal configuration is solidified to the radio frequency power amplifier to amplify the input signal and output the target power signal; The step of taking the initial impedance value as the search starting point, combining the dynamic reconfiguration matching network algorithm, and iteratively adjusting the matching network parameters within the preset impedance range to generate a dynamic optimization impedance set is specifically: According to the simulation performance index output by the digital twin model, and combining the preset performance index threshold, it is determined whether the dynamic reconfiguration matching network algorithm is triggered; If the dynamic reconfiguration matching network algorithm is triggered, set the impedance range with the initial impedance value as the search starting point; The matching network parameters are iteratively adjusted over a range of impedances to cause the output impedance to approach the target impedance ; Real-time verify the adjusted matching network parameters through the digital twin model, detect the circuit stability and thermal distribution state, and determine whether the circuit is stable. If the circuit is not stable, iteratively adjust the matching network parameters; If the circuit is stable, impedance values of F(Z) reaching a preset fitness threshold according to the fitness function are screened to generate a dynamic optimization impedance set .

2. The method of claim 1, wherein the design optimization of the radio frequency power amplifier is performed by a computer. The step of collecting the working frequency and circuit topology of the radio frequency power amplifier, establishing a digital twin simulation model, and generating an initial impedance value according to the digital twin simulation model is specifically: Obtain the working frequency range and circuit topology of the radio frequency power amplifier, establish a physical model of the radio frequency power amplifier through a three-dimensional electromagnetic simulation software, and extract the parasitic parameters; Import the parasitic parameters into a circuit simulation tool, build a digital twin simulation model, set a load-pulling test environment in the digital twin model, scan the output impedance plane, select the impedance region that meets the preset standard, and obtain a candidate initial impedance value set; Taking power matching as the optimization target, use a genetic algorithm to select impedance values that meet the preset requirements from the candidate initial impedance value set to generate an initial impedance value.

3. The method of claim 2, wherein the step of determining the optimum load impedance of the RF power amplifier is performed by: The step of importing the parasitic parameters into a circuit simulation tool, building a digital twin simulation model, setting a load-pulling test environment in the digital twin model, scanning the output impedance plane, selecting the impedance region that meets the preset standard, and obtaining a candidate initial impedance value set is specifically: Insert a programmable impedance tuner module in the digital twin model, dynamically adjust the adjustment resolution of the reflection coefficient according to the adaptive step algorithm, and set the load-pulling test environment according to the adjustment resolution; According to the load-pulling test environment, perform load-pulling test, and obtain the impedance plane according to the test result; The distributed simulation task is deployed by using a cloud computing platform, and an impedance plane is divided into multiple sub-regions, and each sub-region is allocated an independent computing node for parallel simulation; Real-time performance indicators of each impedance point are collected to form a corresponding impedance-performance mapping curve; Performance screening criteria are set, and impedance regions meeting the performance screening criteria are screened based on the impedance-performance mapping curve; Impedance points meeting preset conditions are searched from the impedance regions to form a candidate initial impedance set.

4. The method of claim 3, wherein the step of determining the optimum load impedance of the RF power amplifier is performed by: The step of dynamically adjusting the resolution of the reflection coefficient according to the adaptive step algorithm is specifically: An initial load reflection coefficient is set, and the real part R and the imaginary part I in the complex form of the initial load reflection coefficient are extracted; The resolution of the grid division in the scanning range is denoted as a step size, and an initial step size is set ; According to the gradient threshold formula The gradient threshold is calculated where N is the reference efficiency threshold, is the PAE rate of change derivative of the reflection coefficient; The power added efficiency formula is: wherein Pout is the radio frequency output power, Pin is the radio frequency input power, Pdc is the direct current input power; The real part direction gradient formula is: ； wherein is the real part direction gradient, represents the calculated power added efficiency value under the condition that the real part of the reflection coefficient is increased by an initial step size and the imaginary part remains unchanged, represents the calculated power added efficiency value under the condition that the real part of the reflection coefficient is decreased by an initial step size and the imaginary part remains unchanged; The imaginary part direction gradient formula is: ； wherein, is the imaginary part direction gradient, represents the power added efficiency value calculated under the condition that the imaginary part of the reflection coefficient is increased by an initial step and the real part is kept unchanged, represents the power added efficiency value calculated under the condition that the imaginary part of the reflection coefficient is decreased by an initial step and the real part is kept unchanged. According to the total gradient formula The total gradient G is calculated, combined with the gradient threshold Get real-time step As the adjustment resolution of the reflection coefficient, the real-time step equation is: 。 5. The method of claim 2, wherein the step of determining the optimum load impedance of the RF power amplifier is performed by: The step of taking power matching as an optimization target, using a genetic algorithm to screen impedance values meeting preset requirements from the candidate initial impedance value set to generate an initial impedance value is specifically: establishing a fitness function to calculate a set of candidate initial impedance values the fitness F(Z) of each candidate initial impedance value; The source-end reflection coefficient of each impedance is calculated and the load-end reflection coefficient , the calculation formulas are respectively: , wherein is the system characteristic impedance, is the load impedance; The fitness function is: The impedance values with the preset fitness threshold are screened as the initial impedance population. A parent impedance is selected from the initial impedance population according to a roulette selection mechanism, and after the parent impedance is crossed and mutated, the impedance value with the highest fitness is selected as the initial impedance value.

6. The method of claim 1, wherein The step of loading the input signal to the digital twin simulation model to simulate the simulation performance indicators of the amplifier under the input signal is specifically: Actual parameters of an actual radio frequency power amplifier are collected, the input signal is loaded to the digital twin simulation model, and model parameters are configured to be synchronized with actual parameters; Simulation is started, node data of internal nodes of the radio frequency power amplifier are collected in real time, transient response and steady state response time series data are captured, and simulation data are obtained; Simulation performance indicator values of the amplifier under the input signal are calculated based on the simulation data.

7. The method of claim 1, wherein the design optimization of the radio frequency power amplifier is performed by a computer. The step of iteratively adjusting the matching network parameters within the impedance range to make the output impedance approximate the target impedance , specifically comprises: The target impedance formula is: wherein is the matched network loss power, is a weight factor; The matching network loss power The calculation formula is: Wherein is the current at the output end of the matching network, and DR is the parasitic resistance.

8. The method of claim 1, wherein the design optimization of the radio frequency power amplifier is performed by a computer. The step of verifying the adjusted matching network parameters in real time through the digital twin model, detecting circuit stability and thermal distribution state, and judging whether the circuit is stable is specifically: A stability factor K is calculated according to a stability function, and the stability function is: ； When K>1, it is judged that the circuit has no self-oscillation, and the circuit meets the first stability condition; Computing transistor junction temperature The transistor junction temperature is computed by the formula: where is the transistor package surface temperature, is the transistor power dissipation, is the junction-to-ambient thermal resistance; Setting transistor junction temperature threshold , if , it is determined that the heat distribution is qualified, and the circuit satisfies the second stability condition; If the circuit meets the first stability condition and the second stability condition at the same time, it is judged that the circuit is stable, otherwise it is judged that the circuit is not stable.

9. The method of claim 1, wherein the design optimization of the radio frequency power amplifier is performed by a computer. The step of controlling a tunable element to reconstruct an actual impedance value of the radio frequency power amplifier based on the dynamically optimized impedance set, and collecting actual performance indicators of an output signal is specifically: Based on the dynamically optimized impedance set, an application requirement is obtained, and the best impedance is selected from the dynamically optimized impedance set according to the application requirement; A tunable element array is constructed and integrated in an output stage of the radio frequency power amplifier, the tunable element is dynamically adjusted, the actual output impedance of the radio frequency power amplifier is adjusted to be consistent with the best impedance, and an actual impedance value is obtained; Real-time working data of the radio frequency power amplifier are collected, and actual performance indicators of the radio frequency power amplifier are obtained according to the real-time working data.

Images