Transformer-Based SNR-Adaptive Channel Estimation With IRS Phase Optimization for Multi-User MIMO-ISAC Systems
IEEE Transactions on Wireless Communications · 2026 · 25 · 25680-25691
DOI: 10.1109/twc.2026.3746851 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Wireless Communications, 2026. Verfasst von Saumya Thakor und 3 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben. Gefördert von: Department of Telecommunications (DoT), Government of India, through the Telecom Technology Development Fund.
Abstract (Originalsprache)
Accurate channel estimation in intelligent reflecting surface (IRS)-assisted multi-user MIMO integrated sensing and communication (ISAC) systems is challenged by the high-dimensional cascaded channel structure, heterogeneous fading across BS–IRS and IRS–user links, and wide SNR variations. This paper proposes a unified SNR-adaptive CNN–Transformer framework for uplink channel estimation in a $32\times 32$ massive MIMO-ISAC system with a 64-element IRS at 28 GHz. The proposed architecture combines convolutional feature extraction for local spatial representation with multi-head self-attention for global dependency modeling across the antenna array. Feature-wise linear modulation (FiLM) is incorporated into every encoder layer to condition intermediate representations on the instantaneous SNR, enabling adaptive denoising across a 40 dB dynamic range without retraining. IRS phase shifts are optimized using block coordinate descent to generate physically consistent training data under 3GPP-compliant pilot signaling, while a dual-path residual reconstruction head improves estimation accuracy. Unlike existing approaches that rely on local convolution, implicit SNR adaptation, or model-based inference, the proposed framework jointly integrates global attention, explicit SNR conditioning, and optimized IRS configurations. Simulations against six representative baselines demonstrate an average NMSE of $\mathbf {-17.92}$ dB, outperforming LS by $\mathbf {8.08}$ dB, MMSE by $\mathbf {5.62}$ dB, and CNN-based methods by $\mathbf {4.47}$ dB across the evaluated SNR range, demonstrating accurate and scalable channel estimation for large-scale IRS-assisted MIMO-ISAC systems.
Förderung
- Department of Telecommunications (DoT), Government of India, through the Telecom Technology Development Fund
Zitieren
APA
Thakor, S., Shukla, M. K., Trivedi, A., & Pandey, O. J. (2026). Transformer-Based SNR-Adaptive Channel Estimation With IRS Phase Optimization for Multi-User MIMO-ISAC Systems. IEEE Transactions on Wireless Communications, 25, 25680-25691. https://doi.org/10.1109/twc.2026.3746851
Vancouver
Thakor S, Shukla MK, Trivedi A, Pandey OJ. Transformer-Based SNR-Adaptive Channel Estimation With IRS Phase Optimization for Multi-User MIMO-ISAC Systems. IEEE Transactions on Wireless Communications. 2026;25:25680-25691. doi:10.1109/twc.2026.3746851
BibTeX
@article{Thakor2026,
author = {Thakor, Saumya and Shukla, Mahendra K. and Trivedi, Aditya and Pandey, Om Jee},
title = {Transformer-Based SNR-Adaptive Channel Estimation With IRS Phase Optimization for Multi-User MIMO-ISAC Systems},
journal = {IEEE Transactions on Wireless Communications},
year = {2026},
volume = {25},
pages = {25680--25691},
doi = {10.1109/twc.2026.3746851}
}
EUVIMED