Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PAPR Reduction in OFDM Systems Using Neural Networks: A Case Study on the Importance of Dataset Generalization

Bianca S. de C. da Silva, Pedro H. C. de Souza, Luciano L. Mendes

TL;DR

The paper addresses the critical need to assess generalization of NN-based PAPR reduction for OFDM by correcting prior data leakage and pilot normalization issues. It demonstrates that with proper train/test separation, the NN generalizes well to unseen OFDM symbols, achieving substantial PAPR reductions with lower inference complexity than the MCSA baseline. The updated results, including larger subcarrier configurations and publicly available code/datasets, reinforce the method's practicality in realistic channel conditions and underscore the importance of dataset handling in data-driven communications research. Overall, the NN-based approach remains effective and computationally advantageous for PAPR control in OFDM systems, supporting its deployment in current and future wireless networks.

Abstract

In [1], we introduced a NN designed to reduce the PAPR in OFDM systems. However, the original study did not include explicit generalization tests to assess how well the NN would perform on previously unseen data, which prevented a comprehensive evaluation of the model's robustness and applicability in diverse scenarios. To address this gap, we conducted additional generalization assessments, the results of which are presented in this case study. These results serve both to complement and to refine the original analysis reported in [1]. Most importantly, the overall conclusions of the initial study remain valid: the NN is still able to reduce the PAPR level to a desired reference value, also with a lower computational cost, confirming the effectiveness and practical applicability of the proposed method across a more generalized setting.

PAPR Reduction in OFDM Systems Using Neural Networks: A Case Study on the Importance of Dataset Generalization

TL;DR

The paper addresses the critical need to assess generalization of NN-based PAPR reduction for OFDM by correcting prior data leakage and pilot normalization issues. It demonstrates that with proper train/test separation, the NN generalizes well to unseen OFDM symbols, achieving substantial PAPR reductions with lower inference complexity than the MCSA baseline. The updated results, including larger subcarrier configurations and publicly available code/datasets, reinforce the method's practicality in realistic channel conditions and underscore the importance of dataset handling in data-driven communications research. Overall, the NN-based approach remains effective and computationally advantageous for PAPR control in OFDM systems, supporting its deployment in current and future wireless networks.

Abstract

In [1], we introduced a NN designed to reduce the PAPR in OFDM systems. However, the original study did not include explicit generalization tests to assess how well the NN would perform on previously unseen data, which prevented a comprehensive evaluation of the model's robustness and applicability in diverse scenarios. To address this gap, we conducted additional generalization assessments, the results of which are presented in this case study. These results serve both to complement and to refine the original analysis reported in [1]. Most importantly, the overall conclusions of the initial study remain valid: the NN is still able to reduce the PAPR level to a desired reference value, also with a lower computational cost, confirming the effectiveness and practical applicability of the proposed method across a more generalized setting.
Paper Structure (12 sections, 1 equation, 4 figures, 1 table)

This paper contains 12 sections, 1 equation, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. PAPR Reduction in OFDM Systems
  3. The Importance of Dataset Generalization
  4. Brief Review of the NN-Based PAPR Reduction Method
  5. Summary of the MCSA Method
  6. The NN Method
  7. Computational Complexity Comparison
  8. Updated Numerical Results
  9. Training Loss and Validation
  10. Code and Datasets
  11. Conclusion
  12. Acknowledgment

Figures (4)

  • Figure 1: Illustration of OFDM symbols in the time domain silva2025. (a) OFDM symbol with $K = 32$ subcarriers and $\mathcal{P} = 10.52$ dB. (b) OFDM symbol with $K = 64$ subcarriers and $\mathcal{P} = 11$ dB. (c) OFDM symbol with $K = 256$ subcarriers and $\mathcal{P} = 11.79$ dB.
  • Figure 2: CCDF of the PAPR for an OFDM system with QPSK modulation and 15 subcarriers.
  • Figure 3: CCDF of the PAPR for an OFDM system with QPSK modulation and 30 subcarriers.
  • Figure 4: Training and validation loss curves using the MSE loss over 500 epochs.