A Novel Design Method for Power Amplifier

Matching Network Based on Simplified Real

Frequency Technique
Zhengyu Lu, Zhijiang Dai, Shengdong Hu, Ke Liu,
School of Microelectronics and Commmunication Engineering, Chongqing University, Chongqing
Email:[email protected]

Abstract—This paper briefly introduces new developments goal of topology optimization is to automatically generate the
for simplifying the design of broadband power amplifier overall architecture of the circuit through AI algorithms,
matching circuits using simplified real frequency technique ensuring that it meets specific performance requirements.
(SRFT). The core method of this paper is to combine SRFT with Parameter optimization, on the other hand, adjusts the specific
MATLAB algorithms to construct a suitable matching network
that optimizes the performance of the matching network on the
values of the components within the circuit using AI
basis of meeting the optimal impedance matching conditions algorithms to achieve optimal performance given the
within the frequency band. The design based on SRFT uses load- predefined circuit topology. In [1], the author employs deep
pull and source-pull data of the transistor, employing genetic learning to predict circuit topologies and optimize circuit
algorithm (GA) to optimize the matching network for better component parameters, and [2] similarly employs AI-driven
matching performance. Finally, the overall performance of the optimization methods for designing MMIC circuits.
matching network is demonstrated. Meanwhile, numerous optimization algorithms have been
proposed for use in optimizing. [1] presents an automated
Keywords—broadband, simplified real frequency technique
optimization-oriented strategy for designing high power
(SRFT), matching network, genetic algorithm (GA)
amplifiers (HPAs) using deep neural networks (DNNs), and
I. INTRODUCTION [3] utilized a simulated annealing particle swarm
optimization (SA-PSO) method for the design of high-
The rapid advancement of wireless communication
efficiency PA.
technologies has spurred an increased demand for high-radio
In this paper, a method is proposed to solve the optimal
frequency power amplifiers. RF PAs are critical components
broadband matching problem. By deriving the optimal
in wireless communication systems, majorly responsible for
impedance from waveform engineering or load-pull and
amplifying signals to the necessary power levels while
combining SRFT with a genetic optimization algorithm, a
maintaining efficiency and signal integrity. As a result, the
broadband matching network with better performance is
increasing demand for achieving a wider operational
achieved.
bandwidth simultaneously in modern communication systems
necessitates the development of broadband RF PAs . II. DESIGN METHODOLOGY
However, designing broadband RF PAs is a task of
considerable complexity, characterized by several critical A. Model Construction
aspects. Many design techniques and optimization methods The model establishment process mainly includes three
are acquired through prolonged practical experience. aspects: the mathematical expression representation of
Furthermore, Simulation and testing are integral to the process, matching networks, the introduction of Richards
necessitating the use of various simulation tools (such as transformation, and network synthesis to directly obtain
HFSS, ADS) for electromagnetic field and circuit distributed parameter matching networks. This method can
performance simulations, alongside frequent empirical testing effectively improve design efficiency.
and adjustments. Broadband RF PAs operate under high- For the first aspect, the design of a matching network using
frequency and high-power conditions, where nonlinear effects, SRFT starts with its mathematical representation [4]. A
such as nonlinear distortion and intermodulation distortion, lossless reciprocal two-port LC matching network can be
are pronounced, requiring intricate compensation and represented using its S-parameters. For a given source
linearization techniques to optimize performance. The design impedance, the goal is to adjust the S-parameters so that the
of matching circuits presents particular challenges due to the load impedance of 50Ω is transformed to the desired input
need to achieve effective impedance matching across a wide impedance. The simplified matching network block diagram
frequency range, manage high-power signals, attain precise is shown in Fig. 1.
tuning, and address coupling effects and parasitic parameters
at high frequencies. These rigorous requirements and inherent
complexities render matching circuit design one of the most
demanding aspects of the entire design process. Therefore,
utilizing artificial intelligence (AI) to optimize and assist in
the design of RF circuits has become an important research
direction nowadays
The current state of using AI in optimizing the design of RF
PAs can be broadly categorized into two main approaches:
Fig. 1. The simplified matching network.
circuit topology optimization and parameter optimization. The

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

 The S-parameters of a lossless reciprocal two-port amplifier's output power and efficiency metrics. For this
matching network can be represented as: purpose, the error function is used as follows:

𝑆11 𝑆12 1 ℎ(𝑠) 𝜎𝑓∗ (𝑠) 𝑁

𝑆=[ ]= [ ] (1)
𝑆21 𝑆22 𝑔(𝑠) 𝑓(𝑠) −𝜎ℎ∗ (𝑠) Δ𝑍 = ∑|𝑍𝑖𝑛 (𝜔𝑖 ) − 𝑍𝑜𝑝𝑡 (𝜔𝑖 )| (3)
𝑖=1
Where h(s) and f(s) are arbitrary real coefficient In practical design, it is challenging to match each
polynomials, and g(s) is a Hurwitz polynomial with all poles impedance within the broadband to the optimal impedance
in the left half-plane. simultaneously. Therefore, the goal is to minimize the overall
Richards transformation is introduced to convert distributed error function. The closer the impedance function is to the
parameter elements into rational polynomial representations. ideal impedance, the better the performance of the matching
This allows for the synthesis of networks using components network.
like cascaded transmission lines and parallel open/short-
circuit transmission lines. The impedance function for the load
matching network can be written as a rational polynomial
form:

𝑎𝑛 𝑠 𝑛 + 𝑎𝑛−1 𝑠 𝑛−1 … + 𝑎1 𝑠 + 𝑎0
𝑍𝑖𝑛 = (2)
𝑏𝑚 𝑠 𝑚 + 𝑏𝑚−1 𝑠 𝑚−1 … + 𝑏1 𝑠 + 𝑏0

where 𝑎𝑛 and 𝑏𝑚 are coefficients of the impedance

function’s numerator and denominator polynomials,
respectively.
To achieve this, the synthesis method utilizes a combination
of cascade and parallel transmission lines. If the transmission Fig. 3. ADS and MATLAB Functions and Data Exchange
zero is located at infinity, a parallel transmission line can be
removed. When the transmission zero is located at zero, the
C. Optimization Algorithm
series element is adjusted accordingly. By using Richards
transformation, transmission lines and their respective In the design of matching circuits, network function
impedances are synthesized to form the matching network, optimization is a key technology for achieving better
and the matching network structure is typically composed of matching networks. This paper employs a hybrid
a cascade of transmission lines, parallel open-circuit optimization algorithm that combines GA and Levenberg-
transmission lines, and parallel short-circuit transmission lines Marquardt Algorithm (LMA). Initially, GA is used for global
as shown in the Fig. 2. search [5], followed by LMA for local optimization to
achieve the target optimization strategy. The combination of
GA and LMA leverages GA's global search and LMA's fast
convergence characteristics, enhancing the efficiency and
accuracy of finding the optimal solution.

Fig. 2. The structure of the matching network

B. Definition of the Objective Function

As previously described, the circuit modeling process has
been detailed. To enable optimization of circuit parameters
based on SRFT, the algorithm must work in tandem with
electronic design simulation software. As illustrated in the
Fig. 3.the design process involves two software tools:
MATLAB and ADS. MATLAB is responsible for executing
the algorithm to optimize circuit parameters, while ADS
provides the initial design data and performance data, and
conducts a series of simulations.
In the optimization process, the optimal impedance
provided by ADS is used for load-pull or source-pull.
Optimization is performed in MATLAB, with the cost
function characterized in terms of the impedance form of the
matching network. Therefore, the error of the impedance
function can directly reflect the performance of the power Fig. 4. Match network optimization implementation process
 The framework of the optimization algorithm is presented Due to the physical realizability constraints of the matching
In Fig. 4. Firstly, a random initial population is generated, network, the function cannot completely approximate the
where each individual represents a set of design parameters. optimal impedance. However, the achieved error is within an
The population is then sorted based on the objective function acceptable range, and the matching network essentially meets
(optimal impedance), forming multiple non-dominated sets. the requirements. The final theoretical and simulation curves
Superior individuals are selected for crossover and mutation obtained are shown in Fig. 7. With this, the design of the
to generate the next generation population. Subsequently, the matching network is completed.
GA is run multiple times to obtain a relatively optimal
solution, which is then locally optimized using the LMA. The
relatively optimal solution h obtained by the GA is locally
optimized to yield the optimal solution h and the error value.
The current optimal solution h and error value within iteration
k are compared: if the solution in the current k is better than
the global optimal solution, the global optimal solution is (a) (b)
updated. Finally, an optimal result is output. Fig. 6. Comparison of the impedance function values obtained by SRFT with
the optimal impedance
III. CASE AND RESULT

A. Matching Network Design

The previous section introduced the establishment of the
circuit model and the optimization process. In this section, we
present the design of a broadband power amplifier matching
circuit optimized based on SRFT. Before running the program,
we set the optimization target frequency range from 1.0 to 3.0
GHz. The order of the polynomial h is set to be optimized
within the range of 3 to 8, and the initial coefficients of h are
set to random numbers between -1 and 4. Within the frequency
band, multiple frequency points are selected at intervals for
load-pull and source-pull at the ADS end to obtain the optimal
impedance for the corresponding frequency points. The case
uses the load-pull results of the CGH40010F transistor to Fig. 7. Simulation results of the matching network
design the matching network. The load-pull results are shown
in Table 1. After importing the data into MATLAB, the
IV. CONCLUSION
optimization is performed to obtain the expressions for h(s)
and g(s) as follows: It can be easily understood that the proposed design
method for matching circuits based on the SRFT requires
𝑔(𝜆) = 1.1794𝜆 𝜆 + 5.2514𝜆 𝜆 + 13.2563𝜆 𝜆 + 12.0444𝜆 𝜆 + 9.5138𝜆 + 1 (4)
significantly less time in the broadband matching circuit
ℎ(𝜆) = -0.6254𝜆5 − 0.9519𝜆4 − 7.6743𝜆3 − 3.6917𝜆2 − 7.8373𝜆 design process as compared to traditional methods that rely on
From h(s) and g(s), we can derive the expression for the design experience. This is because the proposed method can
impedance function. Subsequently, the function can undergo quickly search for the optimal solution using GA based on the
results of load-pull or source-pull, thereby obtaining a
network synthesis to construct the matching network as
matching network while maintaining high efficiency.
shown in the Fig. 5. Consequently, this simple yet effective matching circuit
TABLE 1 The optimal impedance at discrete frequency points within the
design method may find its place in the broadband microwave
operating frequency range" design field as a very fast and efficient numerical solver.
Frequency( REFERENCES
1.0 1.4 1.8 2.1 2.6 3.0
GHz)
Optimal [1] L. Kouhalvandi, O. Ceylan and S. Ozoguz, "Automated Deep Neural
19.5+ 16.0+ 17.1+ 19.2+ 15.0+ 12.2+ Learning-Based Optimization for High Performance High Power
impedance
j*24.0 j*17.5 j*14.5 j*13.9 j*4.4 j*4.0 Amplifier Designs," in IEEE Transactions on Circuits and Systems I:
(Ω)
Regular Papers, vol. 67, no. 12, pp. 4420-4433, Dec. 2020.
[2] L. Xue, H. Fan, Y. Ding and B. Liu, "A Design Methodology of MMIC
Power Amplifiers Using AI-driven Design Techniques," 2023 19th
International Conference on Synthesis, Modeling, Analysis and
Simulation Methods and Applications to Circuit Design (SMACD),
Funchal, Portugal, 2023, pp. 1-4.
[3] C. Li, F. You, T. Yao, J. Wang, W. Shi, J. Peng, and S. He, “Simulated
Annealing Particle Swarm Optimization for High-Efficiency Power
Am- plifier Design,” IEEE Trans. Microw. Theory Techn., vol. 69, no.
5, pp. 2494–2505, May 2021.
Fig. 5. Structure of the matching network [4] B. S. Yarman and H. J. Carlin, “ A Simplified ‘ Real Frequency ’
Technique Applied to Broad-Band Multistage Microwave Amplifiers, ”
IEEE Trans. Microw. Theory Tech., vol.30, pp.2216-2222, Dec 1982.
B. Simulation and Results [5] B. Raj, I. Ahmedy, M. Y. I. Idris and R. M. Noor, "A Hybrid Sperm
Based on the optimized impedance function, we can Swarm Optimization and Genetic Algorithm for Unimodal and
synthesize the matching circuit. As shown in the Fig. 6, the Multimodal Optimization Problems," in IEEE Access, vol. 10, pp.
109580-109596, 2022.
impedance function approximates the optimal impedance.