A Model-Driven Deep Learning-Based Channel Estimation Method for CPM
IEEE Communications Letters · 2026 · 30 · 1935-1939
DOI: 10.1109/lcomm.2026.3690494 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Communications Letters, 2026. Verfasst von Le Wang und 1 weiteren Autor:innen. Die Arbeit nennt 11 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Beijing Natural Science Foundation, R&D Program of Beijing Municipal Education Commission.
Abstract (Originalsprache)
Continuous phase modulation (CPM) offers significant advantages in power-efficient transmission systems due to its constant envelope. Under wideband transmission, an accurate channel estimation is critical for subsequent processing. Traditional channel estimation methods suffer from either high complexity or poor performance under limited pilots. Compressed sensing (CS) techniques have been widely adopted in channel estimation to exploit the inherent sparsity of multipath channels, thereby enabling accurate channel recovery with a reduced pilot overhead. Among them, the approximate message passing (AMP) algorithm is particularly attractive due to its low complexity. However, it is prone to divergence when measurement matrices are non-i.i.d. Gaussian. While vector AMP (VAMP) improves stability, it incurs high computational costs due to matrix inversions and remains sensitive to prior mismatches. Data-driven learning-based methods are attractive for their robustness but are unexplainable and unpredictable. To address these issues, we propose CPM-NN-AMP, a complex model-driven deep learning (MDDL) framework. We unfold the VAMP algorithm into a deep network and replace the linear estimator with a learnable non-linear module. Additionally, we introduce a unified complex radial gain formulation (UCRGF) to design complex-valued shrinkage functions that effectively preserve phase information. Simulation results demonstrate that the proposed method achieves nearly 7 dB of gain over conventional CS-based methods and outperforms other learning-based approaches across various complex multipath channels.
Förderung
- National Natural Science Foundation of China
- Beijing Natural Science Foundation
- R&D Program of Beijing Municipal Education Commission
Zitieren
APA
Wang, L., & Li, J. (2026). A Model-Driven Deep Learning-Based Channel Estimation Method for CPM. IEEE Communications Letters, 30, 1935-1939. https://doi.org/10.1109/lcomm.2026.3690494
Vancouver
Wang L, Li J. A Model-Driven Deep Learning-Based Channel Estimation Method for CPM. IEEE Communications Letters. 2026;30:1935-1939. doi:10.1109/lcomm.2026.3690494
BibTeX
@article{Wang2026,
author = {Wang, Le and Li, Junhao},
title = {A Model-Driven Deep Learning-Based Channel Estimation Method for CPM},
journal = {IEEE Communications Letters},
year = {2026},
volume = {30},
pages = {1935--1939},
doi = {10.1109/lcomm.2026.3690494}
}
EUVIMED