An Efficient and General Automated Power Amplifier Design Method Based on Surrogate Model Assisted Hybrid Optimization Technique

• Published in: IEEE transactions on microwave theory and techniques Vol. 73; no. 2; pp. 926 - 937
• Main Authors: Liu, Bo, Xue, Liyuan, Fan, Haijun, Ding, Yuan, Imran, Muhammad, Wu, Tao
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.02.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Amplifier design; Bayes methods; Bayesian neural network (BNN); Computational modeling; Computer simulation; Data models; Design optimization; Doherty power amplifier (PA); evolutionary algorithm; Layouts; Machine learning; Neural networks; Optimization; Power amplifiers; Prediction algorithms; Predictive models; Semiconductor device modeling; surrogate modeling; Topology; Uncertainty; wideband
• ISSN: 0018-9480; 1557-9670
• DOI: 10.1109/TMTT.2024.3518913

In layout-level optimization-oriented power amplifier (PA) design, the need for a good quality initial design and the high computational cost of electromagnetic (EM) simulations are remaining challenges. To address these challenges, a new method called efficient and general Bayesian neural network (BNN)-assisted hybrid optimization algorithm for PA design (E-GASPAD), is proposed. The key innovations of E-GASPAD include the introduction of BNN to model the PA design landscape and a new hybrid optimization algorithm co-working with BNN prediction for efficient PA design optimization. The performance of E-GASPAD is demonstrated by a 27-31 GHz class-AB PA and a 24-31 GHz wideband Doherty PA. Considering around 30 design variables with wide search ranges, the complete set of PA performance specifications, and full-wave EM simulations, layout-level high-performance designs are obtained automatically within a few hundred simulations (i.e., less than 72 h).