High-efficiency broadband Doherty power amplifier optimization design method

Abstract

The invention belongs to the technical field of wireless communication, and particularly relates to a high-efficiency broadband Doherty power amplifier optimization design method, which is characterized in that according to the characteristics of a matching network and a power amplifier, proper parameter values are selected as optimization targets, and the power, the efficiency and the gain of the power amplifier are integrally optimized. By using a hybrid optimization algorithm based on the combination of a genetic algorithm and a neural network, when the overall optimization of the Doherty power amplifier is carried out, the input matching network and the output matching network of the main path and the auxiliary path are used as a whole to carry out performance simulation and optimization. The Doherty power amplifier is integrally optimized through the hybrid optimization algorithm based on the combination of the genetic algorithm and the neural network, the optimization efficiency can be effectively improved, and the design difficulty of the Doherty power amplifier is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communication, and particularly relates to a method for optimizing the design of a high-efficiency broadband Doherty power amplifier. Background Art

[0002] The power amplifier is one of the key components of a radio frequency wireless transceiver. As the main energy-consuming component of a wireless transmitter, its performance directly determines the energy efficiency performance of the entire communication system and affects the bandwidth and efficiency of the wireless communication system. With the continuous development of wireless communication technologies, new communication standards have emerged continuously, and the spectrum resources have also been continuously expanded. This trend has increased the demand for high-performance Doherty power amplifiers.

[0003] In the past, the optimization design of Doherty power amplifiers had limited effects, was time-consuming and laborious, and had low efficiency, making it difficult to meet the increasingly strict optimization standards. To solve this problem, researchers have proposed various optimization methods. Among them, Bayesian optimization is an effective method that can automate the design process of Doherty power amplifiers and has good performance. In addition, multi-objective optimization methods have also been used to solve the complex optimization problems of the broadband matching network of Doherty power amplifiers to achieve better performance. In addition, researchers have also proposed a multi-objective optimization method based on the particle swarm optimization algorithm for the design of Doherty power amplifiers. This method can not only reduce the optimization time but also provide higher power added efficiency.

[0004] Therefore, the optimization design of Doherty power amplifiers has become an important issue in the current research field. Facing this problem, the present invention proposes a design method for optimizing the overall Doherty power amplifier using a hybrid optimization algorithm combining a genetic algorithm and a neural network, and optimizing the matching network within the working frequency band. The present invention can improve the efficiency, power, and gain of the Doherty power amplifier within the working frequency band, providing strong support for the development of wireless communication technologies. Summary of the Invention

[0005] The object of the present invention is to provide a method for optimizing the design of a high-efficiency broadband Doherty power amplifier. By optimizing the overall Doherty power amplifier, the optimization efficiency can be effectively improved, and the design difficulty of the Doherty power amplifier can be reduced.

[0006] To achieve the above object, the specific technical solution adopted by the present invention is as follows: A method for optimizing the design of a high-efficiency broadband Doherty power amplifier, comprising the following steps:

[0007] S1 Select the operating frequency band, and determine parameters such as the saturated output power P_out, saturated efficiency DE_sat, saturated gain Gain_sat, 6dB back-off efficiency DE_6dB, 6dB back-off gain Gain_6dB, and gain Gain_30 at 30dBm power of the Doherty power amplifier, and use them as the target values for the overall optimization of the power amplifier;

[0008] S2 Design the input matching network and output matching network of the carrier power amplifier;

[0009] S3 Design the input matching network and output matching network of the peak power amplifier;

[0010] S4 Design the combiner matching network;

[0011] S5 Determine the optimization variables, number, and variable range of the Doherty power amplifier;

[0012] S6 Determine the optimization objective function according to the optimization requirements of the high-efficiency broadband power amplifier;

[0013] S7 Determine that the optimization algorithm is a hybrid optimization algorithm combining genetic algorithm and neural network;

[0014] S8 Use the hybrid optimization algorithm combining genetic algorithm and neural network to update the parameters of the matching network, and simulate the efficiency, power, gain and other parameters of the Doherty power amplifier, and calculate the optimization objective function;

[0015] S9 Judge whether the optimization objective function meets the requirements. If not, jump to step S8; if so, execute step S10;

[0016] S10 Obtain a power amplifier that meets the design requirements.

[0017] Furthermore, step S6 includes the following steps:

[0018] S6.1 Select the optimization frequency points within the operating frequency band of the power amplifier;

[0019] S6.2 According to the carrier power amplifier matching network designed in step S2, the peak power amplifier matching network designed in step S3, and the combiner matching network designed in S4, simulate the power amplifier performance indicators at the designed frequency points, and calculate the saturated output power optimization objective function F of the Doherty power amplifier according to the following formula P_out , saturated efficiency optimization objective function F DE_sat , saturated gain optimization objective function F Gain_sat , 6dB back-off efficiency optimization objective function F DE_6dB , 6dB back-off gain optimization objective function F Gain_6dBGain optimization objective function F at 30 dBm power Gain_30 ;

[0020]

[0021]

[0022] Wherein, O_P_out is the target value of the saturated output power, O_DE_sat is the target value of the saturated efficiency, O_Gain_sat is the target value of the saturated gain, O_DE_6dB is the target value of the 6 dB back-off efficiency, O_Gain_6dB is the target value of the 6 dB back-off gain, and O_Gain_30 is the target value of the gain at 30 dBm power. P_out(f pi ) is the saturated output power at the design frequency point f pi , DE_sat(f pi ) is the saturated efficiency at the design frequency point f pi , Gain_sat(f pi ) is the saturated gain at the design frequency point f pi , DE_6dB(f pi ) is the 6 dB back-off efficiency at the design frequency point f pi , Gain_6dB(f pi ) is the 6 dB back-off gain at the design frequency point f pi , and Gain_30(f pi ) is the gain at 30 dBm power at the design frequency point f pi , which can be obtained by simulation. The above objective functions are all expected to be less than or equal to 1. From the formula of the objective function, it can be seen that when the minimum values of indicators such as power, efficiency, and gain reach or exceed the target values, the objective function value can be less than or equal to 1;

[0023] The final objective function F is the maximum value of the above objective functions, that is, when the maximum value in the objective function is less than or equal to 1, the remaining objective functions are all less than or equal to 1, and the corresponding indicators such as power, efficiency, and gain reach or exceed the target values:

[0024] F = max(F P_out , F DE_sat , F Gain_sat , F DE_6dB , F Gain_6dB , F Gain_30 ).

[0025] Furthermore, step S7 includes the following steps:

[0026] S7.1 Use the Latin hypercube sampling method to set the initial population to ensure uniform distribution in each dimension. For each input variable d, its dimensional interval [Ld , U d is divided into N small intervals, and then a sample point x is randomly selected from each small interval. For each sampled individual x, calculate its fitness value f(x) and store it.

[0027] Among them, L d is the lower bound of the input variable, and U d is the upper bound of the input variable;

[0028] S7.2 Construct a neural network model, design a neural network structure including an input layer, a hidden layer, and an output layer, as shown in the following formula;

[0029] f(x) = σ(W3·σ(W2·σ(W1·x + b1) + b2) + b3)

[0030] Among them, W1, W2, W3 are weight matrices; b1, b2, b3 are bias vectors; σ(·) is an activation function, including a batch normalization layer to accelerate convergence;

[0031] S7.3 When training the neural network, use an optimizer to adjust the model parameters, and minimize the mean squared error (MSE) as the error function, as shown in the following formula;

[0032]

[0033] Among them, W is the weight parameter of the model; b is the bias parameter of the model; N is the number of training samples; y i is the true value of the i-th sample; f(x i ) is the predicted value of the model for the i-th sample;

[0034] S7.4 Adopt a dynamic mutation strategy to generate the next offspring and adjust the mutation intensity;

[0035]

[0036] Among them, σ0 is the initial intensity, g is the current generation number, and G max is the maximum generation number;

[0037] S7.5 Implement crossover and mutation operations to generate an offspring c from parental individuals p1 and p2;

[0038] c d = β·p 1,a + (1 - β)·p 2,a

[0039] Among them, β ∈ [0, 1] is the crossover coefficient;

[0040] S7.6 Perform Gaussian mutation on the offspring;

[0041]

[0042] where c' d is the value after the offspring mutates in the current dimension d, and is a random variable of a normal distribution;

[0043] S7.7 Use the trained neural network to predict the fitness F' of the offspring;

[0044] S7.8 Perform a true fitness evaluation on individuals with lower predicted fitness and randomly sampled individuals, and update their fitness values;

[0045] S7.9 Combine the offspring and the parents, adopt the elitist retention strategy, and screen the top 50% of the individuals Elite g in the current population P g directly as the next generation;

[0046] Elite g = {x ∈ P g | f(x) ≤ f α}

[0047] where f α is the fitness threshold, taken as the minimum α quantile;

[0048] S7.10 Store the parameters of each individual in the new generation population into the historical population;

[0049] S7.11 Determine whether the end condition is met. If not, return to step S7.3 and input the historical population into the neural network as the training set; if so, verify the true fitness of the best solution. If the predicted best fitness is inconsistent with the true value, re-evaluate the entire population to find the true best solution, and finally output the best individual and its fitness.

[0050] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0051] (1) The present invention can achieve the design of a high-efficiency broadband Doherty power amplifier. Aiming at the problems in the design of traditional Doherty power amplifiers, such as complex matching networks that may lead to efficiency degradation and narrow operating frequency bands, the present invention proposes a design method, that is, for the optimization objective function design problem, the minimization idea is adopted. The main meaning is: when the maximum value of various performance indicators optimized in the operating frequency band has been minimized and can meet the optimization objective, the performance indicators within the entire operating frequency band can reach the optimization objective. This design method enables the Doherty power amplifier to maintain high-efficiency output in a wider frequency band and can meet the requirements of modern wireless communication systems.

[0052] (2) The present invention can effectively reduce the design difficulty of the Doherty power amplifier. The present invention abandons the practice of only focusing on a single parameter in the traditional design and instead adopts the black box concept, taking the entire Doherty power amplifier system as the optimization object for overall optimization. This approach comprehensively considers the overall performance of the power amplifier, especially in terms of efficiency and power output, enhances the flexibility and accuracy of the design, and at the same time reduces the investment of human and time resources, thereby improving the optimization efficiency of the Doherty power amplifier. Description of the Drawings

[0053] Figure 1 It is a schematic structural diagram of the high-efficiency broadband Doherty power amplifier of the present invention.

[0054] Figure 2 It is a flowchart of the design method of the high-efficiency broadband Doherty power amplifier of the present invention.

[0055] Figure 3 It is a flowchart of the hybrid optimization algorithm based on the combination of genetic algorithm and neural network in the present invention.

[0056] Figure 4 It is the overall circuit structure diagram of the Doherty power amplifier in the example of the present invention.

[0057] Figure 5 It is a result diagram showing the change of the algorithm fitness with the number of iterations during the optimization process of the optimized design of the 1.8GHz - 2.7GHz high-efficiency broadband Doherty power amplifier in the example of the present invention.

[0058] Figure 6 It is a result diagram of the efficiency and saturated output power of the optimized design of the 1.8GHz - 2.7GHz high-efficiency broadband Doherty power amplifier in the example of the present invention.

[0059] Figure 7 It is a result diagram of the gain of the optimized design of the 1.8GHz - 2.7GHz high-efficiency broadband Doherty power amplifier in the example of the present invention.

[0060] Figure 8 It is a result diagram showing the change of the gain and efficiency with the output power at 1.8GHz, 2.3GHz, and 2.7GHz of the optimized design of the high-efficiency broadband Doherty power amplifier in the example of the present invention. Detailed Embodiments

[0061] The following further describes the present invention in conjunction with the drawings and specific embodiments. It should be noted that only a preferred technical solution is used here to elaborate in detail the technical solution and design principle of the present invention, but the protection scope of the present invention is not limited thereto.

[0062] The described embodiments are the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Without departing from the essence of the present invention, any obvious improvements, substitutions or variations that those skilled in the art can make fall within the protection scope of the present invention.

[0063] Figure 2 It is a flow chart of the design method of the high-efficiency broadband Doherty power amplifier of the present invention, including:

[0064] S1 Select the working frequency band, and determine parameters such as the saturated output power P_out, saturated efficiency DE_sat, saturated gain Gain_sat, 6dB back-off efficiency DE_6dB, 6dB back-off gain Gain_6dB, and gain Gain_30 at 30dBm power of the Doherty power amplifier, and use them as the target values for the overall optimization of the power amplifier;

[0065] S2 Design the input matching network and output matching network of the carrier power amplifier;

[0066] S3 Design the input matching network and output matching network of the peak power amplifier;

[0067] S4 Design the combining matching network;

[0068] S5 Determine the optimization variables, number and variable range of the Doherty power amplifier;

[0069] S6 Determine the optimization objective function according to the optimization requirements of the high-efficiency broadband power amplifier;

[0070] S7 Determine that the optimization algorithm is a hybrid optimization algorithm combining genetic algorithm and neural network;

[0071] S8 Use the hybrid optimization algorithm combining genetic algorithm and neural network to update the parameters of the matching network, and simulate the parameters such as the efficiency, power and gain of the Doherty power amplifier, and calculate the optimization objective function;

[0072] S9 Judge whether the optimization objective function meets the requirements. If not, jump to step S8; if so, execute step S10;

[0073] S10 Obtain a power amplifier that meets the design requirements.

[0074] As a preferred embodiment of the present invention, step S6 includes the following specific contents:

[0075] S6.1 Select the optimization frequency points within the working frequency band of the power amplifier;

[0076] S6.2 Based on the carrier power amplifier matching network designed in step S2, the peak power amplifier matching network designed in step S3, and the combining matching network designed in S4, simulate the power amplifier performance indicators at the designed frequency points, and calculate the saturation output power optimization objective function F of the Doherty power amplifier according to the following formula P_out , the saturation efficiency optimization objective function F DE_sat , the saturation gain optimization objective function F Gain_sat , the 6dB back-off efficiency optimization objective function F DE_6dB , the 6dB back-off gain optimization objective function F Gain_6dB and the gain optimization objective function F at 30dBm power Gain_30 ;

[0077]

[0078]

[0079] where O_P_out is the target value of the saturation output power, O_DE_sat is the target value of the saturation efficiency, O_Gain_sat is the target value of the saturation gain, O_DE_6dB is the target value of the 6dB back-off efficiency, O_Gain_6dB is the target value of the 6dB back-off gain, and O_Gain_30 is the target value of the gain at 30dBm power. P_out(f pi ) is the saturation output power at the designed frequency point f pi , DE_sat(f pi ) is the saturation efficiency at the designed frequency point f pi , Gain_sat(f pi ) is the saturation gain at the designed frequency point f pi , DE_6dB(f pi ) is the 6dB back-off efficiency at the designed frequency point f pi , Gain_6dB(f pi ) is the 6dB back-off gain at the designed frequency point f pi , and Gain_30(f pi ) is the gain at 30dBm power at the designed frequency point f pi , which can be obtained by simulation. The above objective functions are all expected to be less than or equal to 1. From the formula of the objective function, it can be seen that when the minimum values of indicators such as power, efficiency, and gain reach or exceed the target values, the objective function value can be less than or equal to 1;

[0080] The final objective function is the maximum value among the above objective functions, that is, when the maximum value in the objective function is less than or equal to 1, the remaining objective functions are all less than or equal to 1, and the corresponding power, efficiency, and gain and other indicators reach or exceed the target values:

[0081] F = max(F P_out , F DE_sat , F Gain_sat , F DE_6d , F Gain_6dB , F Gain_30 ).

[0082] As a preferred embodiment of the present invention, step S7 includes the following specific contents:

[0083] S7.1 Use the Latin hypercube sampling method to set the initial population to ensure uniform distribution in each dimension. For each input variable d, divide its dimensional interval [L d , U d into N small intervals, and then randomly select a sample point x from each small interval. For each sampled individual x, calculate its fitness value f(x) and store it.

[0084] Among them, L d is the lower bound of the input variable, and U d is the upper bound of the input variable;

[0085] S7.2 Construct a neural network model, design a neural network structure including an input layer, a hidden layer, and an output layer, as shown in the following formula;

[0086] f(x) = σ(W3·σ(W2·σ(W1·x + b1) + b2) + b3)

[0087] Among them, W1, W2, W3 are weight matrices; b1, b2, b3 are bias vectors; σ(·) is an activation function, including a batch normalization layer to accelerate convergence;

[0088] S7.3 When training the neural network, use an optimizer to adjust the model parameters, and minimize the mean square error (MSE) as the error function, as shown in the following formula;

[0089]

[0090] Among them, W is the weight parameter of the model; b is the bias parameter of the model; N is the number of training samples; y i is the true value of the i-th sample; f(x i ) is the predicted value of the model for the i-th sample;

[0091] S7.4 Adopt a dynamic mutation strategy to generate the next generation and adjust the mutation intensity;

[0092]

[0093] Among them, σ0 is the initial intensity, g is the current generation number, and G max is the maximum generation number;

[0094] S7.5 Perform crossover mutation operation on parental individuals p1 and p2 to generate offspring c;

[0095] c d = β·p 1，d +(1 - β)·p 2,d

[0096] where β ∈ [0, 1] is the crossover coefficient;

[0097] S7.6 Perform Gaussian mutation on the offspring;

[0098]

[0099] where c′ d is the value of the offspring after mutation in the current dimension d, is a random variable of the normal distribution;

[0100] S7.7 Use the trained neural network to predict the fitness F′ of the offspring;

[0101] S7.8 Evaluate the true fitness of individuals with lower predicted fitness and randomly sampled individuals, and update their fitness values;

[0102] S7.9 Merge the offspring and the parents, and adopt the elitist retention strategy to select the top 50% of the individuals Elite g in the current population P g directly as the next generation;

[0103] Elite g = {x ∈ P g |f(x) ≤ f α}

[0104] where f α is the fitness threshold, taken as the minimum α quantile;

[0105] S7.10 Store the parameters of each individual in the new generation population into the historical population;

[0106] S7.11 Determine whether the end condition is satisfied. If not, return to step S7.3 and input the historical population into the neural network as the training set; if satisfied, verify the true fitness of the best solution. If the predicted best fitness is inconsistent with the true value, re-evaluate the entire population to find the true best solution, and finally output the best individual and its fitness.

[0107] The following uses a specific embodiment to further illustrate the present invention.

[0108] In this embodiment, the Wolfspeed CGH40010F GaN HEMT power amplifier tube is used to design a high-efficiency broadband Doherty power amplifier operating at 1.8 GHz - 2.7 GHz. The dielectric substrate used is Rogers 4350B with εr = 3.5 and h = 30 mil.

[0109] First, as described in step S1, for the high-efficiency broadband Doherty power amplifier operating at 1.8 GHz - 2.7 GHz, taking a frequency interval of 100 MHz, there are 10 optimized frequency points within the entire operating frequency band. Then, determine the optimized target values of the Doherty power amplifier, as shown in Table 1. Second, as described in steps S2 to S4, design the matching network of the Doherty power amplifier. Then, perform the overall optimization of the high-efficiency broadband Doherty power amplifier according to steps S5 to S10. Among them, for the optimization algorithm selected in step S8, the embodiment of the present invention adopts a hybrid optimization algorithm based on the combination of genetic algorithm and neural network (as Figure 3 shown), to Figure 4 update the width and length of the microstrip line in Figure 5 Figure is the result graph of the algorithm fitness changing with the number of iterations during the optimization process, indicating that the convergence trend of the algorithm is basically stable around the 10th generation and reaches the minimum fitness value.

[0110] Table 1 Optimized target values of the power amplifier

[0111] Optimization objective Optimization objective value P_out (dBm) 43 DE_sat (%) 60 DE_6dB (%) 55 Gain_sat (dB) 9 Gain_6dB (dB) 12 Gain_30 (dB) 12

[0112] Input the optimal optimized variable values into ADS for simulation. The power, efficiency, and gain of each optimized frequency point of the high-efficiency broadband Doherty power amplifier at 1.8 GHz - 2.7 GHz are shown in Table 2. It can be Figure 6 seen that the saturation efficiency of the optimized Doherty power amplifier within the 1.8 GHz - 2.7 GHz frequency band is greater than 63%, the 6 dB back-off efficiency is greater than 56%, meeting the optimization requirements of high efficiency and broadband, and the saturation output power is close to 44 dBm. It can be Figure 7 seen that the saturation gain is greater than 9 dB, the 6 dB back-off gain is greater than 12 dB, and the small-signal gain is greater than 14 dB, meeting the optimization requirements. In addition, Figure 8 Figure is the result graph of the gain and efficiency changing with the output power of the high-efficiency broadband Doherty power amplifier at 1.8 GHz, 2.3 GHz, and 2.7 GHz, indicating that as the output power increases, the gain generally shows a downward trend and the efficiency generally shows an upward trend.

[0113] Table 2 Power, efficiency, and gain of each optimized frequency point of the power amplifier

[0114] Frequency (GHz) 1.8 1.9 2.0 2.1 2.2 P_out (dBm) 43.6 43.5 43.4 43.7 43.5 DE_sat (%) 64.7 68.1 69.4 66.9 64.3 DE_6dB (%) 60.6 61.4 61.1 59.9 58.1 Gain_sat (dB) 9.7 9.5 9.6 9.8 9.5 Gain_6dB (dB) 14.4 13.8 13.2 12.9 12.7 Gain_30 (dB) 15.8 15.2 14.7 14.4 14.3 Frequency (GHz) 2.3 2.4 2.5 2.6 2.7 P_out (dBm) 43.7 43.8 43.9 43.7 43.5 DE_sat (%) 63.9 65.2 67.5 69.2 68.5 DE_6dB (%) 56.9 57.1 58.5 60.0 60.8 Gain_sat (dB) 9.8 9.8 10.0 9.8 9.6 Gain_6dB (dB) 12.8 13.0 13.1 13.1 12.6 Gain_30 (dB) 14.3 14.5 14.5 14.5 14.3

Claims

1. A high-efficiency broadband Doherty power amplifier optimization design method, characterized in that: The following steps are involved: S1 selects the working frequency band, determines the parameters of the Doherty power amplifier, such as the saturated output power P_out, saturated efficiency DE_sat, saturated gain Gain_sat, 6dB back-off efficiency DE_6dB, 6dB back-off gain Gain_6dB and gain Gain_30 at 30dBm power, and uses them as the target values ​​for the overall optimization of the power amplifier; S2 designs the input matching network and output matching network of the carrier power amplifier; S3 designs the input matching network and output matching network of the peak power amplifier; S4 designs the combiner matching network; S5 determines the optimization variables, number and variable range of the Doherty power amplifier; S6 determines an optimization objective function according to optimization requirements of a high-efficiency broadband power amplifier; S7 determines that the optimization algorithm is a hybrid optimization algorithm based on a combination of a genetic algorithm and a neural network; S8 uses a hybrid optimization algorithm based on a combination of genetic algorithm and neural network to update the parameters of the matching network, simulate the efficiency, power and gain of the Doherty power amplifier, and calculate the optimization objective function; S9 determines whether the optimization objective function meets the requirements, if not, jumps to step S8; If yes, execute step S10; S10 obtains a power amplifier that meets the design requirements.

2. A high-efficiency broadband Doherty power amplifier optimization design method according to claim 1, characterized in that: The step S6 comprises the following steps: S6.1 select an optimized frequency point within the operating frequency band of the power amplifier; S6.2 According to the carrier power amplifier matching network designed in step S2, the peak power amplifier matching network designed in step S3 and the combined matching network designed in step S4, the power amplifier performance index of the designed frequency point is simulated, and the saturated output power optimization objective function F of the Doherty power amplifier is calculated according to the following formula: P_out , saturation efficiency optimization objective function F DE_sat , saturation gain optimization objective function F Gain_sat , 6dB back-off efficiency optimization objective function F DE_6dB , 6dB back-off gain optimization objective function F Gain_6dB The gain optimization objective function F when the power is 30dBm Gain_30 ; Where O_P_out is the target value of saturated output power, O_DE_sat is the target value of saturated efficiency, O_Gain_sat is the target value of saturated gain, O_DE_6dB is the target value of 6dB back-off efficiency, O_Gain_6dB is the target value of 6dB back-off gain, and O_Gain_30 is the target value of gain at 30dBm power. pi ) is the design frequency f pi The saturated output power, DE_sat(f pi ) is the design frequency f pi Saturation efficiency, Gain_sat(f pi ) is the design frequency f pi The saturation gain, DE_6dB(f pi ) is the design frequency f pi 6dB back-off efficiency, Gain_6dB(f pi ) is the design frequency f pi 6dB back-off gain, Gain_30(f pi ) is the design frequency f pi The gain at 30dBm power can be obtained by simulation. The above objective functions all take less than or equal to 1 as the expected target. From the formula of the objective function, it can be seen that when the minimum value of indicators such as power, efficiency and gain reaches or exceeds the target value, the objective function value can be less than or equal to 1; The final objective function is the maximum value among the above objective functions. That is, when the maximum value in the objective function is less than or equal to 1, the remaining objective functions are less than or equal to 1, and the corresponding power, efficiency, gain and other indicators reach or exceed the target value: F=max(F P_out ,F DE_sat ,F Gain_sat ,F DE_6dB ,F Gain_6dB ,F Gain_30 )。 3. A high-efficiency broadband Doherty power amplifier optimization design method according to claim 1, characterized in that: The step S7 comprises the following steps: S7.1 uses Latin hypercube sampling to set the initial population to ensure uniform distribution in each dimension. For each input variable d, its dimension interval [L d , U d ] is divided into N small intervals, and then a sample point x is randomly selected from each small interval. For each sampled individual x, its fitness value f(x) is calculated and stored. Among them, L d is the lower bound of the input variable, U d is the upper bound of the input variable; S7.2 builds a neural network model and designs a neural network structure including an input layer, a hidden layer, and an output layer, as shown in the following formula; f(x)=σ(W3·σ(W2·σ(W1·x+b1)+b2)+b3) Where W1, W2, W3 are weight matrices; b1, b2, b3 are bias vectors; σ(·) is an activation function, which includes a batch normalization layer to accelerate convergence; S7.3 When training a neural network, an optimizer is used to adjust the model parameters by minimizing the mean square error (MSE) as the error function, as shown in the following formula; Where W is the weight parameter of the model; b is the bias parameter of the model; N is the number of training samples; y i is the true value of the i-th sample; f(x i ) is the predicted value of the model for the i-th sample; S7.4 uses a dynamic mutation strategy to generate the next generation and adjust the mutation intensity; Where σ0 is the initial intensity, g is the current generation, G max is the maximum algebra; S7.5 implements the crossover mutation operation to generate offspring c for the parent individuals p1 and p2; c d = β·p1,d+(1 - β)·p 2,d Where β∈[0,1] is the cross coefficient; S7.6 performs Gaussian mutation on the offspring; Among them, c′ d is the value of the offspring after mutation on the current dimension d, is a normally distributed random variable; S7.7 Use the trained neural network to predict the offspring fitness F′; S7.8 evaluates the true fitness of individuals with low predicted fitness and randomly sampled individuals, and updates their fitness values; S7.9 merges the offspring and the parent, adopts the elite retention strategy, and selects the current population P g Elite individuals in the top 50% of fitness g , directly as the next generation; Elite g ={x∈P g |f(x)≤f α } Among them, f α is the fitness threshold, which is taken as the minimum α quantile; S7.10 stores the parameters of each individual in the new generation population into the historical population; S7.11 determines whether the end condition is met. If not, return to step S7.3 and input the historical population into the neural network as a training set. If it is met, verify the true fitness of the best solution. If the predicted best fitness is inconsistent with the true value, re-evaluate the entire population to find the true best solution, and finally output the best individual and its fitness.

Images