Conference2026IEEE

REAL-CSI: Benchmarking DL Architectures for Wi-Fi CSI Human Activity Recognition

Erdem Bera Emiroğlu, Yiğit Can Çelik, Sajjad Baghaee, Ilkay Ulusoy

2026 34th Signal Processing and Communications Applications Conference (SIU)

Official publicationDOI

Abstract

This paper presents the REAL-LAB CSI HAR Dataset, a publicly available Wi-Fi Channel State Information (CSI) dataset for human activity recognition (HAR), together with a controlled deep learning benchmark. We contribute to two key gaps in prior work: the lack of synchronized, balanced datasets and the limited understanding of how architecture and model capacity influence performance. Using a microsecond-synchronized acquisition pipeline, we collect a multi-class dataset and evaluate seven neural architectures across four parameter scales under a multi-seed protocol. The results show that architectural inductive bias dominates model capacity: 2D convolutional networks achieve the best accuracy-efficiency trade-off (96.40%), while larger models often yield diminishing or negative returns. These findings highlight that effective CSI-HAR design depends more on architecture than scale. The dataset and benchmark provide a reproducible foundation for future research.

Selected figures

Figure 2 from “REAL-CSI: Benchmarking DL Architectures for Wi-Fi CSI Human Activity Recognition”: CSI data preprocessing pipeline
Fig. 2 · CSI data preprocessing pipeline.
Reproduced from: E. B. Emiroğlu et al., “REAL-CSI: Benchmarking DL Architectures for Wi-Fi CSI Human Activity Recognition,” 2026 34th Signal Processing and Communications Applications Conference (SIU), 2026. View the original publication.
Figure 1 from “REAL-CSI: Benchmarking DL Architectures for Wi-Fi CSI Human Activity Recognition”: 2D layout of the experimental room
Fig. 1 · 2D layout of the experimental room.
Reproduced from: E. B. Emiroğlu et al., “REAL-CSI: Benchmarking DL Architectures for Wi-Fi CSI Human Activity Recognition,” 2026 34th Signal Processing and Communications Applications Conference (SIU), 2026. View the original publication.
Figure 3 from “REAL-CSI: Benchmarking DL Architectures for Wi-Fi CSI Human Activity Recognition”: Test accuracy vs. trainable parameters
Fig. 3 · Test accuracy vs. trainable parameters.
Reproduced from: E. B. Emiroğlu et al., “REAL-CSI: Benchmarking DL Architectures for Wi-Fi CSI Human Activity Recognition,” 2026 34th Signal Processing and Communications Applications Conference (SIU), 2026. View the original publication.

Cite this work

Plain text

E. B. Emiroğlu, Y. C. Çelik, S. Baghaee and I. Ulusoy, "REAL-CSI: Benchmarking DL Architectures for Wi-Fi CSI Human Activity Recognition," in 2026 34th Signal Processing and Communications Applications Conference (SIU), pp. 1-4, 2026. doi: 10.1109/SIU71813.2026.11636516.

BibTeX 
@inproceedings{emirolu2026realcsi,
  author    = {Emiroğlu, Erdem Bera and Çelik, Yiğit Can and Baghaee, Sajjad and Ulusoy, Ilkay},
  title     = {REAL-CSI: Benchmarking DL Architectures for Wi-Fi CSI Human Activity Recognition},
  booktitle = {2026 34th Signal Processing and Communications Applications Conference (SIU)},
  pages     = {1-4},
  publisher = {IEEE},
  year      = {2026},
  doi       = {10.1109/SIU71813.2026.11636516},
  url       = {https://ieeexplore.ieee.org/abstract/document/11636516}
}

All publications