qhjqhj00 helelena-ce-eval
HELENA: High-Efficiency Learning-based channel Estimation using dual Neural Attention — Camelo Botero et al. (2025) (arXiv:2506.13408, 2025)
What this evaluates
Evaluates deep learning architectures for pilot-based channel estimation in 5G-NR OFDM systems. It probes the model's ability to reconstruct full Channel State Information (CSI) from sparse pilot measurements across varying SNR levels, Doppler shifts, and 3GPP TDL propagation profiles.
Datasets
- 5G Deep Learning Data Synthesis (MATLAB) — total 11264; splits: train (-1), val (-1), test (-1)
Metrics
accuracy(primary) — range: dB- Mean Squared Error (MSE) or Normalized MSE expressed in decibels (10*log10(MSE)), where lower (more negative) values indicate better channel estimation performance.
Input / output format
Input: LS estimates at pilot positions or LI-interpolated LS estimates.
Output: Full CSI (Channel State Information) estimates.
Scoring recipe
def compute_accuracy_dB(predictions, gold):
mse = np.mean((predictions - gold) ** 2)
return 10 * np.log10(mse)
Common pitfalls
- Input preprocessing differs across methods: HELENA/LSiDNN use raw LS estimates at pilot positions, while baselines use LI-interpolated LS estimates, complicating direct accuracy comparisons.
- Model parameter count does not strictly dictate inference latency; lightweight dual-attention designs can achieve significantly faster inference than heavier ViT-based baselines despite comparable accuracy.
Evidence (verbatim from paper)
It achieves 45% faster inference (0.175 ms vs. 0.318 ms) and comparable accuracy (−16.78 dB vs. −17.30 dB) to CEViT while using 8× fewer parameters (0.11M vs. 0.88M), demonstrating superior accuracy-efficiency trade-offs without interpolation, and challenges the assumption that model complexity directly correlates with inference latency.
Citation
@misc{camelobotero2025helenachannelestimation,
title={HELENA: High-Efficiency Learning-based channel Estimation using dual Neural Attention},
author={Camelo Botero et al. (2025)},
year={2025},
note={arXiv:2506.13408}
}
- arXiv: 2506.13408