UL PVC Cable,PVC electronic wire, flexible electronic wire,Power battery cables Jiangyin City Weicheng Special Cable Co.,Ltd , https://www.weichengcable.com
A method for modeling RF power amplifier circuits using a fuzzy neural network is introduced. Fuzzy neural networks are a relatively new type of network structure that possess the ability to function as universal function approximators. In this study, the self-adaptive neural network in MATLAB is employed. The ANFIS (Adaptive Neuro-Fuzzy Inference System) is used to model simulated data, and the resulting model is applied to calculate the spectrum, power compression curve, and power gain curve of the power amplifier. These results are compared with those from ADS simulations, and the outcomes demonstrate the effectiveness of the proposed modeling approach.
With the rapid development of communication technologies, RF circuits have become essential components in modern communication systems. The design and analysis of power amplifiers remain a critical area of research in this field. Recently, the use of fuzzy neural networks for modeling RF devices and circuits has yielded significant progress. This technique provides valuable insights into the modeling of large-scale integrated circuits and complex systems, making it an active area of current research. This paper explores the application of such a theory to model and analyze RF amplifiers.
In this paper, the first-order Sugeno fuzzy model is adopted. To implement the learning process of the Sugeno fuzzy inference system, it is transformed into an adaptive network known as the Adaptive Neuro-Fuzzy Inference System (ANFIS), as illustrated in Figure 1.
The adaptive network is a multi-layer feedforward network, consisting of five layers, each with parameterized nodes. The following sections describe each layer in detail.
[Figure 1: Structure of the Adaptive Fuzzy Neural Inference System]
The first layer performs the fuzzification process, calculating the degree of match between input variables and fuzzy sets. Assuming a Gaussian membership function is used, the output of the first layer can be expressed as:
$$ O_i = \exp\left(-\frac{(x - c_i)^2}{2\sigma_i^2}\right) $$
where $ c_i $ and $ \sigma_i $ represent the center and width of the Gaussian function, respectively, and are parameters that need adjustment during the rule premise phase.
The second layer calculates the firing strength of each rule by taking the product of the membership values of the antecedent parts of the rules:
$$ w_i = \mu_{A_i}(x) \cdot \mu_{B_i}(y) $$
The third layer normalizes the firing strengths:
$$ \bar{w}_i = \frac{w_i}{\sum_{j=1}^{n} w_j} $$
The fourth layer computes the output of each rule, which is the product of the normalized firing strength and the consequent part of the rule:
$$ y_i = \bar{w}_i \cdot (p_i x + q_i y + r_i) $$
The fifth layer aggregates the outputs of all rules to produce the final output of the fuzzy system:
$$ y = \sum_{i=1}^{n} \bar{w}_i \cdot (p_i x + q_i y + r_i) $$
This fuzzy logic system defines a mapping from inputs $ x $ and $ y $ to output $ z $. By carefully tuning the parameters in the fuzzy rules, the relationship between the variables can be accurately captured.
Modeling using fuzzy logic involves two main steps: establishing an initial model and then refining it through training. The initial model can be constructed based on a set of training data using a specific algorithm to determine the number of fuzzy sets, their shapes, and the associated fuzzy rules. Once the initial model is built, algorithms like backpropagation (BP) or Davidon-Fletcher-Powell (DFP) are used to adjust the parameters in the membership functions, reducing the error between the model's output and the actual system output.
The modeling process using ANFIS involves the following steps:
1. Simulate the designed power amplifier circuit in ADS. Input signals such as single-tone, dual-tone, and modulated signals are used to generate input and output voltage data for training.
2. Set up the ANFIS parameters, including the type of membership function, number of fuzzy rules, iterations, and fuzzy sets. Build the initial model and train it with the data.
3. Validate the model using test data and fine-tune the parameters using least squares and gradient descent methods.
4. Evaluate the model’s performance. If the error meets the required accuracy, the modeling process is complete; otherwise, further adjustments are made.
An initial model with three inputs and one output is used. The inputs are defined as $ V_{in}(k) $, $ V_{in}(k-1) $, and $ V_{out}(k-1) $, where $ V_{in}(k-1) $ is represented in differential form as $ V_{in}(k) - V_{in}(k-1) $, and $ V_{out}(k-1) $ accounts for the memory effect. The output is $ V_{out}(k) $, and the dynamic relationship is described by equation (7):
$$ V_{out}(k) = f(V_{in}(k), V_{in}(k-1), V_{out}(k-1)) $$
The model uses a Gaussian membership function with four fuzzy rules, determined via average segmentation.
An example of an RF power amplifier designed using SMIC technology is presented. Its design specifications include S11 < -15 dB, S21 > 20 dB, P1dB > 20 dBm, PAE > 30%, and PGAin > 20 dB. The circuit includes an NMOS transistor from the SMIC library, with other component parameters detailed in Tables 1–3.
The circuit operates at 2.45 GHz, with input power ranging from -20 dBm to 10 dBm. Time-domain sampling data from two cycles is used for training. Test data is selected within a similar range to evaluate the model’s performance.
The modeling results are shown in Figures 3–6. Figure 3 displays the steady-state output voltage at input powers of 6.5 dBm and -6.5 dBm. Figure 4 presents time-domain data for an input power of 7.5 dBm, with FFT results showing the spectrum of the fundamental and harmonic voltages. Figures 5 and 6 compare the power compression and gain curves calculated by the model with those from ADS simulations.
The results demonstrate that the fuzzy logic model closely matches the simulation results, validating its accuracy. The required output power and gain can be calculated using equations (8)–(10).
In conclusion, the fuzzy neural network approach effectively models RF power amplifiers, capturing nonlinear characteristics and providing accurate predictions. While the model performs well under various input conditions, some deviations exist, indicating room for improvement in accuracy. Further refinement of the model is recommended to enhance its performance across different scenarios.
August 31, 2025