A Comparison of Surrogate Behavioral Models for Power Amplifier Linearization under High Sparse Data.

Area of Science:

• Electrical Engineering
• Signal Processing
• Machine Learning

Background:

• Accurate power amplifier (PA) behavioral modeling is crucial for mitigating nonlinearities, especially with digital predistortion (DPD) for linearization.
• Validating PA models under signal conditioning and transmission restrictions requires a robust framework.

Purpose of the Study:

• To develop a framework for validating PA modeling figures of merit under signal conditioning and transmission restrictions.
• To introduce a feature extraction concept for heuristic target approach modeling, addressing sparse data in amplitude-to-amplitude (AM-AM) and amplitude-to-phase (AM-PM) models.

Main Methods:

• An FPGA-based testbed was developed to measure wide-band PA behavior using LTE-based 64-QAM OFDM signals.
• A feature extraction concept was introduced to manage sparse data issues in AM-AM/AM-PM model extraction.
• Models were compared against regression tree (RT), random forest (RF), and cubic-spline (CS) for accuracy and complexity.

Main Results:

• Experimental results demonstrated model performance in high sparse data regimes.
• The proposed models showed reliable validation and low complexity compared to RT, RF, and CS.
• Figures of merit included PAPR, CCDF, coefficients extraction, NMSE, and execution time.

Conclusions:

• The presented models, including machine learning (ML)-based and CS interpolated, offer effective solutions for high sparse data scenarios.
• The CS interpolated models demonstrated efficacy in prediction methods with lower convergence times and complexities.
• The study provides a reliable validation framework for PA modeling with reduced hardware complexity.