Machine-Learning Assisted Optimisation of Free-Parameters of a Dual-Input Power Amplifier for Wideband Applications.

Area of Science:

• Electrical Engineering
• Signal Processing
• Wireless Communications

Background:

• Power amplifiers (PAs) face a trade-off between power efficiency and linearity.
• Dual-input PA topologies offer additional degrees of freedom for optimization.
• Existing methods struggle to simultaneously enhance efficiency and linearity for complex signals.

Purpose of the Study:

• To develop an auto-tuning approach for dual-input PAs.
• To leverage global optimization and adaptive linearization for improved performance.
• To enhance power efficiency across wide bandwidths and high peak-to-average power ratios (PAPR) while maintaining linearity.

Main Methods:

• Utilized global optimization search algorithms (simulated annealing, adaptive Lipschitz Optimization).
• Employed adaptive linearization techniques, including PA biasing, signal calibration, and digital predistortion.
• Tested the approach on a load-modulated balanced amplifier using Long-Term Evolution (LTE) signals.

Main Results:

• Achieved maximum power efficiency for LTE signals with up to 200 MHz bandwidth and >10 dB PAPR.
• Demonstrated a mean output power of ~33 dBm with 22.2% mean power efficiency.
• Met stringent linearity requirements: in-band error vector magnitude <1% and out-of-band adjacent channel leakage ratio < -45 dBc.

Conclusions:

• The auto-tuning method effectively optimizes dual-input PAs for enhanced efficiency and linearity.
• Global optimization algorithms successfully mitigate the linearity-efficiency trade-off.
• The approach is validated for demanding wireless communication standards like LTE.