csi-har-dataset

Description

CSI-HAR-Dataset (Moshiri et al.) is a WiFi CSI human activity recognition dataset collected on a 5 GHz 802.11ac network. Each activity sample contains 4000 CSI packets recorded over a 20 second window with the labelled activity occurring in the middle 3 to 6 seconds. The dataset is released alongside an open Python pipeline and is used as a modern, 802.11ac-era complement to the older Intel-5300-based CSI HAR datasets.

Modality / size

• Modality: WiFi CSI from a 5 GHz 802.11ac network, 20 MHz bandwidth, channel 36.
• Subjects / scenarios: indoor activities; 4000 samples per activity recorded over 20 s windows.
• Labels: per-activity class label.

Used by (papers)

• Used by CSI HAR methods that want a 802.11ac-era benchmark instead of the older 802.11n Intel 5300 datasets.

Notes

• Public release; companion repository ships the data and reference pipeline.
• Distinct from ut-har; the alias CSI-HAR is ambiguous in the literature, prefer the full name CSI-HAR-Dataset when citing this one.