Can Wi-Fi sense through walls? Physics, experiments and privacy

By Wu Jun · Updated 2026-10-29 · 4 min read

"Wi-Fi can see through walls" is a famous headline and an easy claim to misunderstand. A more precise statement is: under specific link and wall conditions, motion behind a wall can change multipath and therefore CSI. That is not an image, an identity or reliable position.

This page does not claim that a consumer router can image through a wall. It separates physics, research conditions, a reproducible experiment and privacy boundaries.

1. Separate five very different capabilities

"Detects a change" and "knows who or what it is" are separated by physical information, aperture, models and validation.

2. How a signal passes through building materials

The receiver observes a sum of:

• the direct path;
• reflections from walls, floor, furniture and people;
• attenuated paths through walls;
• strong reflections or dead zones from metal, water and mirrors.

When a person behind a wall moves, some path lengths, phases and amplitudes change, and CSI changes too. This is the physical basis of through-wall sensing; a CSI sensing survey provides a broader taxonomy ^[1].

But the router does not see a clean "person behind wall" component. It sees all paths added together. Indoor multipath can make a small target visible, and can also make doors, fans or pets look like targets.

3. "Through a wall" is not one condition

Penetration loss has been measured and modelled across common building materials above 2 GHz; those measurements do not guarantee a particular home because thickness, moisture, rebar, frequency and incidence angle matter ^[2].

4. Research capability is not router capability

"See Through Walls with WiFi!" used a custom MIMO array and controlled experiments to demonstrate through-wall motion detection and coarse localisation ^[3]. Its array, geometry, synchronisation and processing were part of the result, not an optional accessory.

CSI pose work such as DensePose From WiFi uses multiple antennas, controlled data and heavy models; the public evidence remains largely preprint-level ^[4]. Therefore:

• the paper can support "the signal may contain information";
• it cannot be cited for "any router can see a person through a wall";
• motion detection cannot be upgraded to medical, security or identity claims without new validation.

5. Run a through-wall experiment at home

This is a validation procedure, not product certification. Only collect in spaces you own or where participants have given informed consent.

5.1 Fix the conditions

• fix router, client, antenna direction, channel, bandwidth and TX power;
• change one variable at a time: wall count, distance, door state or person position;
• record RSSI, SNR, packet rate, maximum gap, subcarrier count and chain policy;
• label empty room, still person, walking and door/fan/appliance events;
• collect long enough sessions to include failures.

5.2 Four controls

1. same room, line of sight: strongest baseline;
2. same room, obstruction: furniture/door multipath;
3. one lightweight wall: does motion remain above empty-room noise?
4. two walls or metal/concrete: does the capability disappear?

Each group should include:

• 10 minutes empty;
• 10 minutes still;
• 10 walking trials;
• door, bag drop and appliance negatives.

5.3 Report

If one lightweight wall works only for a fast walk and fails for a seated person, the honest label is motion detection, not through-wall presence detection.

6. Why it differs from mmWave

Wi-Fi CSI is an opportunistic channel measurement that relies on communication traffic. FMCW mmWave radar actively transmits chirps and can process range, velocity and angle directly. Their frequency, waveform, arrays and penetration behaviour differ. Use the Wi-Fi vs mmWave guide rather than treating one as a scaled version of the other.

7. Privacy: no camera is still monitoring

Through-wall sensing affects shared, rented and third-party spaces. Even without images it can infer:

• whether anyone is present;
• routines and activity patterns;
• breathing, motion or anomalies;
• how often a room is used.

Controls should include:

1. sense only owned/consented spaces;
2. state purpose, retention and deletion;
3. provide an off switch and auditability;
4. avoid medical, security or identity claims for a research prototype;
5. consider CSI countermeasures such as jamming/masking research ^[5].

IEEE 802.11bf is standardising WLAN sensing measurements and interoperability ^[6]. A standard capability does not remove the privacy problem; it makes permissions, visibility and data governance more important.

8. Conclusion

• motion behind a lightweight wall changing CSI is physically plausible;
• a normal router cannot claim through-wall imaging, identity or reliable pose from that fact;
• still-person presence, localisation and pose need more stable links, arrays and ground truth;
• wall material, distance, multipath and empty-room baselines determine repeatability;
• every demo should show failure conditions, false alarms and privacy limits.

Related: Wi-Fi CSI explained · Wi-Fi sensing vs mmWave · respiration detection · dataset