Comparison of Feature Selection Techniques for Power Amplifier Behavioral Modeling and Digital Predistortion...

Area of Science:

• Electrical Engineering
• Signal Processing
• Telecommunications

Background:

• Power amplifiers (PAs) are crucial for linearity and efficiency.
• Digital predistortion (DPD) mitigates nonlinearities near saturation.
• Volterra series models in DPD lead to high coefficient counts and overfitting.

Purpose of the Study:

• To benchmark various dimensionality reduction techniques for DPD models.
• To compare modeling and linearization performance against runtime.
• To identify optimal methods for efficient and robust DPD.

Main Methods:

• Categorization of techniques into greedy pursuits (OMP, DOMP, SP, RF), regularization (Ridge, LASSO), heuristic local search (HC, DMS), and global optimization (SA, GA, adaLIPO).
• Comparative analysis based on modeling accuracy, linearization performance, and computational runtime.
• Evaluation of techniques for reducing the dimensionality of Volterra series models in DPD.

Main Results:

• Greedy pursuits demonstrate strong performance in dimensionality reduction.
• Doubly Orthogonal Matching Pursuit (DOMP) shows a superior trade-off between execution time and linearization robustness.
• Other techniques like regularization and global optimization present different performance characteristics.

Conclusions:

• Dimensionality reduction is essential for practical DPD implementation.
• DOMP emerges as a highly effective technique for balancing efficiency and performance in DPD.
• The benchmark provides valuable insights for selecting appropriate DPD order reduction methods.