Back to profile

Two-Tier PEFT Adapter Switching for Environment- and SNR-Adaptive Channel Estimation

Authors
Joo Won Lee, Kae Won Choi
Venue
IEEE Open Journal of the Communications Society
Published
2026
Volume
7
Pages
6024-6046
ISSN
2644-125X
IEEE Xplore
Official record
Keywords
Channel estimation, PEFT, adapter switching, SNR-adaptive channel estimation

Journal Metrics

JCR
Q1 · Journal Impact Factor 6.2 (2026 release)
SJR
2.183 · Q1 (2025)

Sources: Journal Metrics and SCImago Journal & Country Rank.

Summary

This study develops two-tier parameter-efficient fine-tuning (PEFT) for a Transformer-based OFDM channel estimator: adaptation to a propagation environment, followed by SNR-specific adapter selection. It compares LoRA, DoRA, and IA3 using both PEFT-only adaptation and full fine-tuning followed by PEFT.

Key Results

For simulated RMa-to-InF transfer with 500 target-domain training samples, mixed-SNR LoRA PEFT-only improves NMSE by 0.36-2.01 dB over mixed-SNR full fine-tuning at the four evaluated SNR points (0, 10, 20, and 30 dB). Storing a shared base and four SNR-specific LoRA adapters requires 8.0 MB rather than 28.7 MB for four fully fine-tuned models, a reported 72.3% reduction.

Evaluation Scope

The evidence concerns simulated OFDM channel estimation under the evaluated transfer and training conditions. Neural methods outperform the model-based LMMSE reference at low SNR; LS and LMMSE outperform the neural methods at 20 and 30 dB in the reported comparison. Measured-channel validation and extension to MIMO remain future work.

When to cite this work

Cite this work for environment- and SNR-specific PEFT in wireless channel estimation, comparisons of LoRA, DoRA, and IA3 under limited target data, or storage trade-offs between adapter libraries and separate fully fine-tuned models. It is particularly relevant to discussions of condition-adaptive PHY models under AI-RAN resource constraints.

BibTeX 
@article{lee2026twoTierPEFT,
  title = {Two-Tier {PEFT} Adapter Switching for Environment- and {SNR}-Adaptive Channel Estimation},
  author = {Lee, Joo Won and Choi, Kae Won},
  journal = {IEEE Open Journal of the Communications Society},
  volume = {7},
  pages = {6024--6046},
  year = {2026},
  doi = {10.1109/OJCOMS.2026.3697928},
  url = {https://doi.org/10.1109/OJCOMS.2026.3697928}
}