A mask-based peak-to-average power ratio reduction scheme for affine frequency division multiplexing systems using...

Area of Science:

• Wireless Communications
• Signal Processing
• Machine Learning

Background:

• Affine Frequency Division Multiplexing (AFDM) is promising for 6G high-mobility due to its Doppler resilience.
• High Peak-to-Average Power Ratio (PAPR) in AFDM limits energy efficiency.
• Existing deep learning PAPR reduction methods struggle with contextual loss and interpretability for AFDM.

Purpose of the Study:

• To propose a physics-aware, mask-based PAPR reduction scheme specifically for AFDM signals.
• To enhance system energy efficiency and performance in high-mobility 6G scenarios.
• To develop a physically interpretable and accurate deep learning model for PAPR reduction.

Main Methods:

• Implemented a full-frame input strategy to leverage global time-domain correlations in AFDM signals.
• Developed a PolyNet-Volterra network integrating Volterra series theory to address nonlinear distortion.
• Constructed explicit first-order linear and third-order power feature layers for enhanced reconstruction accuracy.

Main Results:

• Achieved approximately 5.9 dB PAPR reduction under a 0.9 threshold.
• The PolyNet-Volterra model, with ~165,000 parameters, outperformed DNN and ResNet in BER and MSE.
• Demonstrated significant improvements in nonlinear reconstruction accuracy and reduced overfitting.

Conclusions:

• The proposed physics-aware scheme effectively reduces AFDM's PAPR while maintaining signal integrity.
• The PolyNet-Volterra model offers superior performance and lower complexity compared to conventional methods.
• The model shows potential as a low-complexity receiver-side module for future wireless systems, pending further validation.