Power amplifier design method based on multi-objective particle swarm optimization

Abstract

This invention discloses a power amplifier design method based on multi-objective particle swarm optimization. This method first uses ADS software to load-pull and source-pull transistors to obtain the optimal source and load impedances. The method then uses the impedance corresponding to the center frequency of the frequency band to obtain the parameter values ​​of the Chebyshev low-pass topology component. Latin hypercube sampling is used to sample the initial microstrip line parameter values ​​of the matching network to obtain two particle swarms. The impedance matching performance of the particle swarms is evaluated using a fitness function. The overall performance of the power amplifier is evaluated using output power and efficiency. Multiple sets of microstrip line parameters are obtained using a Pareto front based on a reference point. Through continuous iteration, a broadband, high-efficiency power amplifier is ultimately achieved. This method, the first of its kind, proposes a power amplifier design method based on multi-objective particle swarm optimization. It designs the power amplifier by comprehensively considering impedance matching and power amplifier performance. Microstrip line parameter selection is achieved through reference points, accelerating the efficiency of power amplifier simulation design.

Description

Technical Field

[0001] The present invention belongs to the field of machine learning and radio frequency power amplifier design, and specifically relates to a power amplifier design method based on multi-objective particle swarm optimization. Background Art

[0002] With the continuous increase in data transmission volume, the increasing complexity of communication system application scenarios, and the scarcity of spectrum resources, market demand is continuously driving the development of wireless communication systems. As the core functional module of wireless communication systems, the performance of RF power amplifiers directly affects the quality of wireless communication systems. For example, improving the efficiency of power amplifiers can reduce heat loss in communication base stations, while increasing the output power of power amplifiers can expand the signal coverage range of base stations. Therefore, the design of power amplifiers has become a research hotspot.

[0003] However, power amplifier performance indicators often conflict with each other, making the design and production of power amplifiers that balance high power and high efficiency more complex. This is a typical multi-objective optimization problem, typically addressed through simulation and optimization using software platforms such as the Advanced Design System (ADS). Currently, many intelligent algorithms, such as gradient algorithms, genetic algorithms, and simulated annealing algorithms, have been embedded in the ADS optimization tool to accelerate the simulation and optimization process.

[0004] Although the ADS built-in optimization tool is widely used, it still has three major shortcomings. First, for multi-objective optimization problems, the gradient algorithm, genetic algorithm or simulated annealing algorithm built into ADS needs to perform a lot of calculations during the optimization process, which takes a long time to optimize. In addition, the impedance matching theory of the power amplifier design is not taken into account during the optimization process, which is somewhat blind. Second, compared with the design of passive circuits such as filters, power amplifier circuits need to consider factors such as harmonic modulation and node voltage. Therefore, power amplifier circuits have strong nonlinear characteristics, which also leads to the problem of non-convergence when ADS simulates power amplifiers. If this problem occurs during the optimization process, the optimization will stop, affecting the progress of the project. Third, with the continuous expansion of commercial application scenarios, more and more More and more power amplifier architectures have been proposed, such as Doherty power amplifiers. The debugging process of these power amplifiers is extremely difficult. Designers often use the method of "independent design of hierarchical power amplifiers + overall integration" for debugging. In this process, a large number of design parameters need to be continuously combined and screened, and ADS can only obtain one set of design parameters each time it is optimized. If you want to obtain multiple sets of design parameters, you have to perform multiple optimizations. However, multiple optimizations are ineffective for the gradient algorithm, genetic algorithm, or simulated annealing algorithm embedded in ADS. Because when the optimization starting point is the same, these algorithms will only get fixed results regardless of the number of optimizations. If you want to obtain multiple sets of design parameters by changing the optimization starting point multiple times, you will have to face the problems of long optimization time and non-convergence errors. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this paper proposes a power amplifier design method based on multi-objective particle swarm optimization. This method incorporates a Pareto frontier iteration method for multi-objective optimization, as well as a particle swarm optimization method based on impedance matching. By comprehensively considering the relationship between output performance and impedance, a novel particle optimal position iteration mechanism is established. This method continuously optimizes power amplifier circuit parameters and ensures convergence of the optimization method, ultimately yielding multiple sets of design parameters.

[0006] The power amplifier design method based on multi-objective particle swarm optimization includes the following steps:

[0007] Step 1: Determine the matching circuit structure

[0008] First, the transistors used in the power amplifier are source-pull and load-pull respectively within the frequency band to obtain the optimal source impedance Z source =Z sr +jZ si and the optimal load impedance Z load =Z lr +jZ li , where Z sr and Z si Represents Z sourceThe real and imaginary parts, Z lr and Z li Represents Z load The real and imaginary parts of the input and output matching networks are obtained by using the optimal source and load impedances at the center frequency to obtain the prototype component parameters of the Chebyshev low-pass topology. The prototype components are then converted into discrete components through inverse normalization. The discrete components are then converted into microstrip lines using the short-line approximation theory. The initial parameters X1 and X2 of the input and output matching networks are then obtained.

[0009] Step 2: Particle swarm initialization

[0010] In step 1, around the initial parameters X1 and X2, n values ​​are randomly selected to obtain the initial particle swarm X IMN and X OMN , and then randomly initialize the velocity V of the particle swarm IMN and V OMN .

[0011] Step 3: Performance Evaluation

[0012] Performance evaluation includes impedance matching and output performance evaluation, specifically:

[0013] S3.1, use ADS to analyze the particle swarm X after initialization in step 2 IMN and X OMN Simulation is performed in the frequency band [freq min ,freq max ] to set m simulation frequency points and obtain the actual source impedance R source =R sr +jR si and the actual load impedance R load =R lr +jR li , where R sr and R si R source The real and imaginary parts, R lr and R li R load The real and imaginary parts of X are evaluated by fitness functions f1 and f2 respectively. IMN and X OMN Impedance matching effect:

[0014]

[0015]

[0016] S3.2, the particle swarm X evaluated in S3.1 IMN and X OMN The individual optimal position P IMNand P OMN Then, the updated individual optimal position is simulated by ADS to obtain the output power Pout and efficiency PAE. Since the output power Pout and efficiency PAE are a simulation curve within the frequency band, the minimum value of the curve is selected, Pout min and PAE min , to evaluate the overall performance of the amplifier:

[0017] Pout min =min(Pout1, Pout2, ..., Pout m ) (3)

[0018] PAE min =min(PAE1,PAE2,…,PAE m ) (4)

[0019] Step 4: Update the Pareto front and global optimal position

[0020] After each iteration, n sets of Pout are obtained. min 、PAE min , after the tth iteration we will get These n sets of data are further screened to ensure that progress is made in each iteration and to set benchmarks for:

[0021]

[0022]

[0023] will be greater than and point Filter them out and get the Pareto frontier, then randomly assign the Pareto optimal solution corresponding to the Pareto frontier to each particle as the global optimal position G IMN , G OMN .

[0024] Step 5: Update particle swarm

[0025] Let particle swarm X IMN and X OMN Tracking individual optimal position (P IMN , P OMN ) and the global optimal position (G IMN , G OMN ) is updated. If the set number of iterations is reached after the update, the parameters of the optimized input and output matching networks are output; otherwise, return to step 3.

[0026] The present invention has the following beneficial effects:

[0027] Compared to traditional ADS optimization, this method considers impedance matching theory during the optimization process. By using an improved Pareto front, it balances high power and high efficiency. This allows for the simultaneous generation of multiple sets of high-quality microstrip line parameters, significantly enhancing parameter selection. Experiments have also demonstrated that this method can significantly shorten optimization time compared to ADS optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A power amplifier design method based on multi-objective particle swarm optimization;

[0029] Figure 2 Graph showing the optimal source impedance changing with frequency in the embodiment;

[0030] Figure 3 : This is a graph showing the optimal load impedance changing with frequency in the embodiment;

[0031] Figure 4 Schematic diagram of simulated output power and efficiency curve in the embodiment;

[0032] Figure 5 A schematic diagram of the Pareto front based on a reference point in an embodiment;

[0033] Figure 6 The output power and efficiency distribution diagram in the embodiment;

[0034] Figure 7 This is a simulation performance diagram of a broadband high-efficiency power amplifier in an embodiment;

[0035] Figure 8 A comparison diagram of actual source impedance and optimal source impedance in the embodiment;

[0036] Figure 9 A comparison diagram of actual load impedance and optimal load impedance in the embodiment; DETAILED DESCRIPTION

[0037] In order to make the advantages of the present invention more clear, the present invention is further explained below with reference to the accompanying drawings;

[0038] like Figure 1 As shown in FIG, the power amplifier design method based on multi-objective particle swarm optimization includes the following steps:

[0039] Step 1: Determine the matching circuit structure

[0040] In this embodiment, transistor CGH40010F is selected for power amplifier design. First, the ADS software of Keysight is used to perform source pull and load pull on the transistor at intervals of 0.1GHz in the frequency band from 2GHz to 3GHz to obtain the optimal source impedance Z source =Z sr +jZsi and the optimal load impedance Z load =Z lr +jZ li The trajectory of change, such as Figure 2 and Figure 3 The optimal source impedance and optimal load impedance at the center frequency are then used to obtain the prototype component parameters of the Chebyshev low-pass topology. The prototype components are then converted into discrete components through inverse normalization. The discrete components are then converted into microstrip lines using the short-line approximation theory. Finally, the initial parameters of the input matching network, X1 = {1.02, 3.47, 1.02, 3.47, 1.61, 6.91, 4.65, 4.53}, and the initial parameters of the output matching network, X2 = {1.02, 3.47, 1.02, 3.47, 2.76, 5.66, 6.21, 3.5}, are obtained.

[0041] Step 2: Particle swarm initialization

[0042] In the range of ±10% of the initial parameters X1 and X2 obtained in step 1, 10 values ​​are randomly selected using the Latin Hypercube Sampling (LHS) method to obtain the initial particle swarm X IMN and X OMN , and then randomly initialize the velocity V of the particle swarm IMN and V OMN .

[0043] Step 3: Performance Evaluation

[0044] Sample evaluation mainly includes impedance matching effect evaluation and output performance index evaluation. The specific steps include:

[0045] S3.1. Initialize the particle swarm X obtained in step 2 through ADS IMN 、X OMN Perform simulation and set 11 simulation frequency points in the frequency band [2GHz, 3GHz] to obtain the actual source impedance R source =R sr +jR si and the actual load impedance R load =R lr +jR li , R sr and R si R source The real and imaginary parts, R lr and R li R load The real and imaginary parts of X are evaluated by fitness functions f1 and f2 respectively. IMN and X OMN Impedance matching effect:

[0046]

[0047]

[0048] S3.2. After completing the evaluation of the impedance matching effect, update the particle swarm X IMN and X OMN The individual optimal position P IMN and P OMN , and then use ADS to find the optimal position of the individual (P IMN , P OMN ) to simulate and get the output power Pout and efficiency PAE. Since the output power and efficiency are a simulation curve within the frequency band, such as Figure 4 As shown, the minimum value of the curve is selected, Pout min 、PAE min To measure the overall performance of the amplifier:

[0049] Pout min =min(Pout1, Pout2, ..., Pout 11 ) (3)

[0050] PAE min =min(PAE1,PAE2,…,PAE 11 ) (4)

[0051] Step 4: Update the Pareto front and global optimal position

[0052] After each iteration, 10 sets of Pout are obtained. min 、PAE min , after the tth iteration we will get This data needs to be further filtered to ensure progress is being made with each iteration and to set benchmarks. for:

[0053]

[0054]

[0055] Putting performance above and point After screening, the Pareto front is obtained after traversal, and then the Pareto optimal solution corresponding to the Pareto front is randomly assigned to each particle as the global optimal position G IMN , G OMN . Figure 5 A Pareto front diagram based on the benchmark is shown. Data points with better performance than the benchmark are selected and traversed in one step to obtain the Pareto front solution, while the remaining points are discarded.

[0056] Step 5: Update particle swarm

[0057] Let particle swarm X IMN and X OMN Tracking individual optimal position (P IMN , P OMN ) and the global optimal position (G IMN , G OMN ) is updated. If the set number of iterations is reached after the update, the parameters of the optimized input and output matching networks are output; otherwise, return to step 3.

[0058] Figure 6 The distribution diagram of the minimum output power and minimum efficiency in the 2GHz-3GHz frequency band corresponding to the 10 sets of design parameters obtained in this example proves that this example can obtain multiple sets of available microstrip line parameters at one time, greatly improving the selectivity and flexibility of the design. The 10 sets of parameters obtained by this method can be directly used as alternative parameters for the Doherty power amplifier structure. To further demonstrate the advantages of this method, a set of parameters is randomly selected from the 10 sets of microstrip line parameters to draw the simulation curve in the entire frequency band, as shown in Figure 2. Figure 7 As shown in the figure, the results show that any set of design parameters of this method can be directly used as a broadband high-efficiency power amplifier, and has high performance in the entire frequency band. Figure 8 and Figure 9 The following are the source impedance matching and load impedance matching results obtained in this example, where "○" represents the optimal impedance and "△" represents the actual impedance. The results show that the actual source impedance and actual load impedance of the final output surround the optimal source impedance and optimal load, which proves that this method can achieve excellent impedance matching and its reliability, and is an effective supplement to existing optimization techniques. Table 1 shows a comparison of the cumulative optimization time and number of errors of this method and the traditional ADS method, under the same conditions of 10 sets of design parameters. The results show that the proposed algorithm can effectively avoid errors and has an absolute advantage in cumulative optimization time.

[0059] method Cumulative optimization time Number of errors Multi-objective particle swarm optimization algorithm 3m18s 0 ADS-Gradient Algorithm 32m43s 1 ADS-Genetic Algorithm 40m11s 4 ADS-Simulated Annealing Algorithm 35m04s 1

[0060] Table 1

Claims

1. A power amplifier design method based on multi-objective particle swarm optimization, characterized by: The method comprises the following steps: Step 1: Determine the matching circuit structure First, the transistors used in the power amplifier are source-pull and load-pull respectively within the frequency band to obtain the optimal source impedance Z source and the optimal load impedance Z load ; Design the initial parameters X1 of the input matching network and X2 of the output matching network according to the optimal source impedance and optimal load impedance of the center frequency; Step 2: Particle swarm initialization Randomly select n values ​​around the initial parameters X1 and X2 to obtain the initial particle swarm X IMN and X OMN , and then randomly initialize the velocity V of the particle swarm IMN and V OMN ; Step 3: Performance Evaluation S3.1, use ADS to analyze the particle swarm X after initialization in step 2 IMN and X OMN Perform simulation, set m simulation frequency points within the frequency band, and obtain the actual source impedance R source and the actual load impedance R load , respectively evaluate X by fitness functions f1 and f2 IMN and X OMN Impedance matching effect: Among them, Z sr and Z si Represents Z source The real and imaginary parts, Z lr and Z li Represents Z load The real and imaginary parts of R sr and R si R source The real and imaginary parts, R lr and R li R load The real and imaginary parts of S3.

2. Particle swarm X after evaluation of S3.1 IMN and X OMN The individual optimal position P IMN and P OMN Update, and then simulate the updated individual optimal position through ADS to obtain the output power Pout and efficiency PAE; select the minimum value Pout of the curve min and PAE min To evaluate the overall performance of the amplifier; Step 4: Update the Pareto front and global optimal position Setting the reference point for: After the tth iteration, Filter and select and point Filter them out and get the Pareto frontier, then randomly assign the Pareto optimal solution corresponding to the Pareto frontier to each particle as the global optimal position G IMN , G OMN ; Step 5: Update particle swarm Let particle swarm X IMN and X OMN Tracking individual optimal position (P IMN , P OMN ) and the global optimal position (G IMN , G OMN ) is updated. If the set number of iterations is reached after the update, the parameters of the optimized input and output matching networks are output; otherwise, return to step 3.

2. The power amplifier design method based on multi-objective particle swarm optimization according to claim 1, characterized in that: Within the range of ±10% of the initial parameters X1 and X2, 10 values ​​are randomly selected as the initial particle swarm.

3. The power amplifier design method based on multi-objective particle swarm optimization according to claim 1 or 2, characterized in that: Latin hypercube sampling was used for random selection.

4. The power amplifier design method based on multi-objective particle swarm optimization according to claim 1, characterized in that: The transistors used in the power amplifier are 10W GaN HEMT transistors.

Images