Multi-Tone Harmonic Balance Optimization for High-Power Amplifiers through Coarse and Fine Models Based on...

Area of Science:

• Electrical Engineering
• Computational Electromagnetics
• Machine Learning Applications

Background:

• Radio frequency (RF) high-power amplifiers (HPAs) are critical components in modern wireless communication systems.
• Optimizing HPAs for simultaneous power gain, output power, efficiency, and linearity is challenging due to their inherent nonlinearity.
• Existing design methodologies often struggle with accurately characterizing and optimizing these complex circuits.

Purpose of the Study:

• To develop an automated optimization design methodology for RF HPAs.
• To concurrently trade off between power gain, output power, efficiency, and linearity specifications.
• To leverage deep neural networks (DNNs) for accurate HPA modeling and design.

Main Methods:

• A novel 'coarse and fine modeling' approach using DNNs is proposed.
• Transistor modeling using X-parameters in the fine phase and S-parameters in the coarse phase.
• A classification DNN is employed to accelerate the selection of optimal HPA topology and configuration.
• Automated generation of post-layouts with multi-tone harmonic balance specifications optimized concurrently.

Main Results:

• The proposed strategy achieves highly accurate nonlinear HPA designs.
• Demonstrated drain efficiency exceeding 54% and linear gain over 12.5 dB for a 10 W L-band HPA.
• Achieved adjacent channel power ratio (ACPR) better than 50 dBc after digital pre-distortion (DPD).

Conclusions:

• The coarse and fine DNN-based modeling approach effectively optimizes complex HPA designs.
• This methodology enables automated, efficient, and accurate design of high-performance RF HPAs.
• The results validate the effectiveness of DNNs in addressing nonlinearities and achieving multi-objective optimization in RF power amplifier design.