CS-GA-XGBoost-Based Model for a Radio-Frequency Power Amplifier under Different Temperatures.

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Machine learning (ML) methods like Support Vector Regression (SVR) and gradient boosting are used for power amplifier (PA) modeling.
• XGBoost offers high-precision modeling but requires optimal hyperparameter tuning.
• Traditional hyperparameter searches (e.g., grid search) are inefficient and time-consuming.

Purpose of the Study:

• To develop an efficient and accurate PA modeling method.
• To address the limitations of traditional hyperparameter optimization.
• To improve both modeling accuracy and speed for PAs.

Main Methods:

• Proposed a novel PA modeling method using a hybrid Cuckoo Search (CS)-Genetic Algorithm (GA) optimized XGBoost (CS-GA-XGBoost).
• Integrated GA's crossover operator into CS to leverage global search and fast convergence.
• Validated the method using measured data from a 2.5-GHz GaN class-E PA across various temperatures (-40 °C, 25 °C, 125 °C).

Main Results:

• CS-GA-XGBoost improved modeling accuracy by over one order of magnitude compared to XGBoost, GA-XGBoost, and CS-XGBoost.
• CS-GA-XGBoost reduced modeling time by over one order of magnitude compared to XGBoost, GA-XGBoost, and CS-XGBoost.
• Outperformed gradient boosting, random forest, and SVR by three orders of magnitude in accuracy and two orders of magnitude in speed.

Conclusions:

• The CS-GA-XGBoost method demonstrates superior performance in PA modeling accuracy and speed.
• This hybrid optimization approach effectively addresses hyperparameter tuning challenges.
• The CS-GA-XGBoost model shows significant potential for application in radio-frequency/microwave device and circuit modeling.