Can WiFi See Through Walls? What Sensing Can Detect

A practical test plan before trusting the result

Begin with a written claim that can be proven or rejected. For example: the system should distinguish an empty room from one walking person through one documented drywall partition at a fixed distance. Avoid starting with broad goals such as seeing people through walls, because there is no clear success condition and almost any visualization can look convincing.

Record the transmitter and receiver model, antenna arrangement, firmware, frequency band, channel width, sampling rate, wall material, wall thickness, room dimensions, device height, and distance from the wall. Photograph or sketch the placement so the same geometry can be repeated. If the system uses a trained model, record which rooms, people, activities, and wall types appeared in the training data.

Collect an empty-room baseline long enough to observe normal drift and interference. Then capture repeated trials for movement, stillness, entry, exit, and any target activity. Randomize the order so the operator cannot unconsciously tune the result. Include negative controls such as movement on the transmitter side, a fan, a closing door, another WiFi network, or a pet when those conditions are realistic.

Separate calibration data from evaluation data. A model that performs well on the same recordings used to choose thresholds has not demonstrated generalization. Reserve different times, people, placements, and ideally another room for the final test. Report the number of trials, not only the best examples, and keep failures visible.

Finally, decide what happens when confidence is low. A safe interface should show unknown, insufficient signal, or recalibration required rather than forcing a human-shaped output. Define a review process before deployment and repeat the baseline after furniture, router, firmware, or occupancy patterns change.

• Write one measurable claim and one failure condition.
• Document geometry, hardware, radio settings, and wall construction.
• Use empty-room controls and realistic interference tests.
• Hold out unseen people, times, placements, or rooms.
• Publish confusion counts, latency, and low-confidence behavior.

How WiFi signals get through a wall

A transmission reaches the receiver along many paths. Some energy passes through drywall or wood, while other parts reflect from concrete, metal, furniture, doors, and people. The receiver observes the combined result rather than a clean ray.

CSI exposes amplitude and phase across subcarriers. Movement changes some paths, and a model compares those changes with labeled examples. It estimates a target; it does not reconstruct a photograph.

What through-wall WiFi sensing may detect

Research systems have demonstrated presence, motion, respiration patterns, activity classes, localization, and coarse pose estimation. Each result is task-specific; a walking classifier does not automatically identify a person or work in every building.

• Room or zone occupancy with a stable baseline.
• Movement or stillness inside a tested range.
• Coarse activity or pose classes with suitable training data.
• Breathing-related periodic changes under controlled conditions.
• Approximate location with enough links or antennas.

What WiFi cannot reliably show

A consumer router does not produce a live picture behind a wall. Fine facial details, clothing, identity, exact health status, and intent are not directly visible in CSI.

A crisp skeleton is a model visualization and may use live data, replay, simulation, or illustration. Ask what hardware produced the input and whether the result survives a new room or participant.

• No photographic or X-ray image.
• No dependable identity from ordinary CSI.
• No guaranteed detection through every material.
• No medical or emergency decision without validated specialist systems.

How to evaluate a through-wall demo

Confirm the chipset, firmware, channel, receiver, sampling process, model, and target label. Test an empty room, known movements, stillness, changed placement, and an unseen participant.

Report false positives, false negatives, latency, and low-confidence periods. A credible system should sometimes say it does not know. Repeatability across days matters more than one striking clip.

Privacy, consent, and safe use

Camera-free sensing is not privacy-free. Occupancy, routines, sleep, movement, and location can still be sensitive. People should know what is measured, why it is collected, who can access it, and how to disable it.

Do not use experimental WiFi sensing as the sole basis for medical, emergency, policing, employment, or safety decisions. Prefer local processing, minimal retention, access controls, visible uncertainty, and human review.

• Use the minimum sensing resolution needed.
• Obtain consent and provide an off switch.
• Separate research demos from production claims.
• Document unsupported uses and retention limits.

Sources and technical references

Through-Wall WiFi Sensing FAQ