Wi-Fi sensing vs mmWave radar for the home

By Wu Jun · Updated 2026-10-29 · 5 min read

Wi-Fi CSI and millimetre-wave radar are often placed in the same "camera-free sensing" table. Both avoid cameras, but they measure different physical quantities and fail in different ways. Copying a performance number from one to the other usually produces the wrong conclusion.

This project has real Wi-Fi CSI measurements but no mmWave module benchmark. The evidence is therefore separated: the Wi-Fi side uses our capture and a survey ^{[1] [2]}; the radar side uses authoritative FMCW and penetration-loss material ^{[3] [4]}.

0. Bottom line

• For low-cost coarse whole-home presence, reuse a CSI-capable AP first;
• for a single critical room with range/velocity/angular information, evaluate an mmWave module;
• do not assume one radar covers a whole home or one AP is a reliable sensor everywhere;
• use multi-sensor state, hysteresis and timeouts instead of one naive AND/OR rule.

1. What each system measures

Wi-Fi CSI

A Wi-Fi transmitter sends ordinary data frames and the receiver estimates a complex channel per subcarrier and antenna chain. Human movement changes multipath, so CSI contains motion, respiration, posture and environmental changes ^[2].

The advantages come with constraints:

• another station must keep sending traffic;
• rate control, bandwidth, antennas and appliances affect the channel;
• one AP has limited angle/range resolution for a given array and bandwidth;
• multipath both provides information and creates ambiguity.

mmWave radar

FMCW radar actively transmits a linear chirp. Beat frequency maps to range, phase across chirps maps to velocity, and phase across antennas supports angle estimation ^[3]. The device can therefore form a range-velocity-angle observation, whereas Wi-Fi CSI is primarily a channel measurement.

That does not mean every module automatically provides accurate range, velocity and angle. Performance depends on:

• carrier frequency and sweep bandwidth;
• chirp/frame timing and processing window;
• antenna count, aperture and mounting angle;
• whether vendor firmware exposes presence, range or tracks;
• occlusion and field of view.

2. Home selection table

3. Overlooked limits

3.1 mmWave is not better through walls

At higher frequency, building materials usually introduce more penetration loss. Walls, metal, water and furniture can weaken the echo ^[4]. A radar normally belongs inside the room it monitors, with a usable field of view, rather than behind a thick wall.

3.2 Wi-Fi is not automatically whole-home accurate

AP coverage is a networking claim, not a sensing claim. Distance, walls, metal cabinets, microwave/Bluetooth interference, client power save and packet rate all change CSI quality. Our 10-minute capture is specific to one channel and link, not every room ^[1].

3.3 Neither system is identity recognition

Wi-Fi research can reconstruct coarse pose with multiple antennas, controlled data and heavy models. mmWave can also study pose, but requires data and ethical controls. Turning "motion detected" into "we know who it is" crosses the evidence boundary.

4. Fair same-room comparison

Before selecting hardware, run a small A/B in the actual room.

4.1 Fix variables

• fix AP/radar position, height, orientation and power;
• fix furniture, door state, walking paths and test times;
• record Wi-Fi channel/width/packet rate and radar carrier/bandwidth/chirp/frame rate;
• record fans, HVAC, pets and other periodic sources;
• collect both systems simultaneously, without letting mounts block one another.

4.2 Scenarios

4.3 Metrics

• event precision, recall, misses and false alarms per day;
• occupied-to-detected and leave-to-release latency;
• empty-room occupancy rate and still-person miss rate;
• subgroup results by distance, angle, wall, subject and posture;
• unavailable time, gaps, temperature drift and reboots;
• timestamp error between raw sensor data and ground truth.

Do not report only one successful screenshot. Include failed sessions and the distance/posture/occlusion that caused them.

5. A sensible Home Assistant fusion

An interpretable combination is:

1. mmWave for fast room-level presence/micro-motion;
2. Wi-Fi CSI for low-frequency whole-home occupancy trends and fallback;
3. expose unknown, unavailable and a quality/confidence state from both;
4. use 5-30 second hysteresis, timeout and human confirmation;
5. require more evidence for critical alerts and less for lighting;
6. retain raw events so automations can be debugged.

A simple AND works only when both sensors are reliable; a simple OR amplifies any one sensor's false positive. The fusion rule should depend on room, time and sensor availability.

6. Privacy and compliance

No camera does not mean no privacy impact:

• Wi-Fi CSI can infer motion and routines from shared wireless channels;
• mmWave can infer presence, respiration, trajectory or posture;
• shared homes, hotels, offices and care spaces require informed consent and purpose limitation;
• recording, cloud upload and third-party models belong in the privacy policy;
• IEEE 802.11bf is bringing WLAN sensing into a formal standards process, so privacy must be designed alongside it ^[5].

7. Decision checklist

Choose Wi-Fi CSI first when:

• a flashable AP already exists;
• whole-home coarse presence/activity is enough;
• periodic calibration and some false alarms are acceptable;
• low-frequency, slow-change features are the main target.

Choose mmWave first when:

• one specific room is the product boundary;
• range, micro-motion or velocity information is valuable;
• a clear field of view and controlled blockage are possible;
• one powered module per room is acceptable.

If neither fits, redefine the problem with door, PIR, pressure or UWB sensors rather than collecting more CSI or radar data without a measurable target.

8. Conclusion

• Wi-Fi CSI and mmWave are different physical measurements, not a simple low/high model pair;
• FMCW is designed for range/velocity/angle processing; Wi-Fi is attractive because it reuses communications;
• mmWave penetrates walls poorly, while Wi-Fi CSI is multipath- and traffic-sensitive;
• selection needs a same-room, ground-truth, session-split comparison;
• medical, identity and through-wall imaging claims require much stronger evidence and compliance.

Related: Wi-Fi CSI explained · the physical limits of through-wall sensing · Home Assistant integration