Wi-Fi CSI radar: what it really measures (and where the marketing stops)
2026-10-11 · 2 min read
Search for "Wi-Fi CSI radar" and you will find everything from research papers to claims of seeing through walls. Here is a sober mapping between radar concepts and what CSI can actually deliver.
Radar concepts vs CSI reality
| Radar concept | With CSI | Notes |
|---|---|---|
| Motion detection | Yes | amplitude/phase variance across subcarriers |
| Presence (still person) | Partial | needs respiration micro-motion; gated |
| Range | Coarse | time-of-flight / phase slope; resolution limited by bandwidth (80 MHz ≈ 3.75 m bins, refined by multipath) |
| Velocity / Doppler | Yes | STFT on phase/amplitude over time |
| Angle of arrival | Yes, multi-antenna | needs calibration; our hardware has 2 Tx × 4 Rx |
| Imaging / identification | No (as commonly claimed) | you get channel changes, not images or identity |
Why "radar" is a fair-but-limited analogy
Classic radar sends a known waveform and measures reflections. Wi-Fi sensing uses existing traffic and measures how the channel changes between transmitter and receiver. Both infer motion from reflections, but CSI is:
- opportunistic — no dedicated waveform, so resolution depends on the Wi-Fi frame structure;
- multi-path rich — indoor reflections help and hurt at the same time;
- receiver-measured — you need a station that keeps sending traffic (taxonomy and trade-offs: CSI sensing survey [1]).
Doppler in practice
import numpy as np
# phase series for one subcarrier (after unwrap + CFO/SFO removal)
win = 256 # ~2.7 s at 95 records/s
Z = np.stack([np.fft.fftshift(np.fft.fft(ph[i:i+win] * np.hanning(win)))
for i in range(0, len(ph) - win)])
# |Z| is a Doppler-time map: moving limbs produce moving ridges
In our 10-minute capture you can see this as slow ridges that drift when a person walks between the router and the client.
What you should not claim
- "Radar-like imaging" — you cannot reconstruct a body or a face (coarse pose is a controlled-setting research result, currently preprint-level [3]);
- "Works through any wall" — 5 GHz attenuates heavily through concrete/metal; drywall + multipath is the realistic case;
- "Identity recognition" — gait-based identification exists in papers but is environment- and dataset-specific.
Where this is going
IEEE 802.11bf (WLAN Sensing) standardises sensing measurements, which means future chips will expose cleaner sensing data [2]. Until then, projects like this one build on firmware-patched CSI.
Related: Wi-Fi CSI explained · respiration detection · through-wall sensing: what is real
References and data sources
Every external claim carries a source and an evidence grade (grading scheme). Grade E marks first-party, reproducible measurements from this project.
- Grade A · standard/review
WiFi Sensing with Channel State Information: A Survey
W. Ma, W. Zhou, X. Wang · ACM Computing Surveys · 2019
doi:10.1145/3310194 · citations 857 · verified: CrossRef API, 2026-10-11 · The field's most-cited CSI sensing survey (857 citations, verified via CrossRef). - Grade A · standard/review
IEEE 802.11bf — WLAN Sensing (standard project)
IEEE Standards Association · IEEE 802.11 Working Group · 2020
standards.ieee.org/ieee/802.11bf/10331/ · verified: standards.ieee.org HTTP 200, 2026-10-11 · Official project page for turning sensing into a WLAN standard capability (approved September 2020). - Grade D · preprint/wiki
DensePose From WiFi
J. Geng, D. Huang, F. De la Torre · arXiv preprint arXiv:2301.00250 · 2023
arxiv.org/abs/2301.00250 · verified: arXiv abs page HTTP 200, 2026-10-11 · Preprint: coarse human pose reconstruction from multi-antenna CSI, a controlled-setting study that does not imply consumer capability.
This page contains affiliate links (rel="sponsored"). Buying through them costs you nothing extra and earns this site a small commission.