Efficient Signals Denoising Under Imperfect Channels: A Noise-to-Noise Learning Approach
IEEE Transactions on Communications · 2026 · 74 · 16345-16357
DOI: 10.1109/tcomm.2026.3747966 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Communications, 2026. Verfasst von Zhiyi Tian und 4 weiteren Autor:innen. Die Arbeit nennt 21 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Guangdong Basic and Applied Basic Research Foundation, Scientific Research Foundation of UESTC.
Abstract (Originalsprache)
Denoising complex communication signals remains challenging for conventional mathematical algorithms. Deep learning approaches offer a promising solution via nonlinear fitting. However, supervised learning requires paired clean-noisy data, which is impractical due to rapidly varying wireless channels and non-Gaussian hardware noise. Moreover, perfect channel state information is unavailable, leading to estimation errors and consequent signal degradation. To address these issues, we propose an unsupervised noise-to-noise denoising (N2ND) framework. Existing Noise2Noise (N2N) denoising studies mainly assume additive zero-mean noise. In this work, we prove that the N2N principle remains valid under a composite signal model that includes both non-Gaussian hardware noise and multiplicative channel distortion from imperfect channel estimation. Based on this result, the N2ND framework employs a convolutional neural network (CNN) and gated recurrent unit (GRU) encoder-decoder trained without clean labels, together with an entropy-guided dynamic encoding mechanism powered by reinforcement learning (RL) that adapts the model complexity to heterogeneous signals under a computational constraint. Experimental results demonstrate that the N2ND framework achieves effective and robust denoising performance across various SNRs, modulation schemes, and imperfect channel estimation conditions.
Förderung
- National Natural Science Foundation of China
- Guangdong Basic and Applied Basic Research Foundation
- Scientific Research Foundation of UESTC
- Natural Science Foundation of Sichuan Province
Zitieren
APA
Tian, Z., Ji, Z., Yang, K., Huang, W., & Ye, P. (2026). Efficient Signals Denoising Under Imperfect Channels: A Noise-to-Noise Learning Approach. IEEE Transactions on Communications, 74, 16345-16357. https://doi.org/10.1109/tcomm.2026.3747966
Vancouver
Tian Z, Ji Z, Yang K, Huang W, Ye P. Efficient Signals Denoising Under Imperfect Channels: A Noise-to-Noise Learning Approach. IEEE Transactions on Communications. 2026;74:16345-16357. doi:10.1109/tcomm.2026.3747966
BibTeX
@article{Tian2026,
author = {Tian, Zhiyi and Ji, Zelin and Yang, Kuojun and Huang, Wuhuang and Ye, Peng},
title = {Efficient Signals Denoising Under Imperfect Channels: A Noise-to-Noise Learning Approach},
journal = {IEEE Transactions on Communications},
year = {2026},
volume = {74},
pages = {16345--16357},
doi = {10.1109/tcomm.2026.3747966}
}
EUVIMED