Pacer-CfC: A Lightweight Timespan-Modulated Closed-Form Continuous-Time Network for Massive MIMO CSI Prediction
IEEE Communications Letters · 2026 · 30 · 2283-2287
DOI: 10.1109/lcomm.2026.3704293 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Communications Letters, 2026. Verfasst von Ziheng Wang und 2 weiteren Autor:innen. Die Arbeit nennt 16 Literaturangaben. Gefördert von: Xiongan New Area Science and Technology Innovation Program, National Natural Science Foundation of China.
Abstract (Originalsprache)
In mobile scenarios, channel state information (CSI) feedback delay causes channel aging, which significantly impairs the beamforming and link adaptation performance of extremely large-scale MIMO (XL-MIMO) systems. This letter presents Pacer-CfC, an incremental refinement of the closed-form continuous-time network (CfC) for CSI prediction, which augments the CfC backbone with a lightweight Timespan Pacer. This pacer extracts three types of low-overhead dynamic proxies (first-order temporal variation, latent space local variation, and second-order temporal variation) from the CSI latent space codeword sequence, generating an adaptive timespan signal to modulate the state evolution of the continuous-time units. Experiments demonstrate that at mobility speeds of 3, 30, and 120 km/h, Pacer-CfC outperforms baseline models, including Transformer, LSTM, GRU, RNN, STEM GNN, and Plain-CfC, in terms of Normalized Mean Square Error (NMSE) and system-level spectral efficiency. In the 120 km/h mobility scenario, Pacer-CfC achieves an MU-MIMO sum rate of 20.77 bps/Hz, while reducing the model parameter count and theoretical FLOPs by 69.7% and 85.5%, respectively, compared to the RNN baseline. Mechanism analysis further validates that the learned timespan signal adapts to the intensity of channel dynamics, thereby enhancing predictive performance in high-mobility scenarios.
Förderung
- Xiongan New Area Science and Technology Innovation Program
- National Natural Science Foundation of China
Zitieren
APA
Wang, Z., Xu, X., & Qin, X. (2026). Pacer-CfC: A Lightweight Timespan-Modulated Closed-Form Continuous-Time Network for Massive MIMO CSI Prediction. IEEE Communications Letters, 30, 2283-2287. https://doi.org/10.1109/lcomm.2026.3704293
Vancouver
Wang Z, Xu X, Qin X. Pacer-CfC: A Lightweight Timespan-Modulated Closed-Form Continuous-Time Network for Massive MIMO CSI Prediction. IEEE Communications Letters. 2026;30:2283-2287. doi:10.1109/lcomm.2026.3704293
BibTeX
@article{Wang2026,
author = {Wang, Ziheng and Xu, Xiaodong and Qin, Xiaowei},
title = {Pacer-CfC: A Lightweight Timespan-Modulated Closed-Form Continuous-Time Network for Massive MIMO CSI Prediction},
journal = {IEEE Communications Letters},
year = {2026},
volume = {30},
pages = {2283--2287},
doi = {10.1109/lcomm.2026.3704293}
}
EUVIMED