Building an Aging-in-Place Care System with Three 60GHz Radars: Fall, Sleep, Presence

Each of our three product guides describes one radar on its own: the fall radar, the sleep radar, and the presence radar. What none of them covers is the view a system designer actually needs: how three radars work together in one home, on one broker, feeding one alerting path.

This article is that reference architecture for aging-in-place, home care, and assisted-living deployments.

This is a safety and wellness system, not a medical one. The radars do not diagnose, treat, or prevent a condition, and they are not a substitute for human care, a call system, or emergency services. Nothing here is a medical claim.

Why three sensors in one dwelling

A single radar answers one question well. A caring adult asks several questions in different rooms, at different hours:

Two themes repeat across them. The bathroom is where cameras are unacceptable and a call button is often out of reach. The bedroom and hallway answer the slower questions — did they get up at 3 a.m., did they reach the kitchen, have they been unnaturally still for an hour?

Reference architecture

 Fall radar ─┐
 Sleep radar ┼──MQTT──>  Broker  ──MQTT──>  Home Assistant / your backend ──> caregiver alert
 Presence ───┘          (1883)                    (automation + escalation)

Keep three properties fixed from day one:

1. One broker, one set of credentials per device. See the Artemis setup guide for a self-hosted option; Mosquitto and managed brokers work identically from the radar's point of view.
2. A unique ClientID per radar. Two devices sharing one ID will boot each other off indefinitely.
3. Topic namespaces per model — /Radar60FL/{DeviceID}/…, /Radar60SP/{DeviceID}/…, /RadarHP/{DeviceID}/…. Give each device an ID that encodes the room (fl-bathroom-01), because that ID is what you will group on later.

Integration

Declare the entities per area in Home Assistant, following the patterns in the integration overview and the per-model guides. Then build automations around combinations, which is where the real value sits:

• Fall → notify. A bathroom fallStatus change is the highest-priority event. Set fallDuration long enough to ignore someone who stumbled and recovered — see the fall radar guide for the timing trade-off.
• Still too long → check. Combine the bathroom fall radar's dwell alarm with the presence radar's occupancy: someone present in a room and completely still beyond your threshold is worth a nudge.
• Night-time absence. If the sleep radar reports out-of-bed and the presence radar reports no occupancy anywhere for a set window overnight, escalate.
• Poor night, repeated. End-of-session sleep score and stage durations trend badly several nights running — that is a conversation, not an alarm.

Keep notifications in your own application: step 5 in the fall guide's alarm chain is yours. The radar does not place a call and does not contact anyone by itself.

Escalation design

Radar signals are noisy by nature; escalate in stages rather than on one measurement.

1. Soft signal — the module's own states help here: the fall radar can be configured with a fall duration so a recovered stumble never becomes an alarm, and its local voice call asks the person whether they are all right. A spoken reply can stand the alarm down before anyone is contacted. See the fall radar guide for the voice phrases and their limits.
2. Confirm on the network — your backend correlates a second signal (presence in another room, absence elsewhere, time of day).
3. Contact — push, SMS, or a call placed by your own service, with a documented fallback when nobody acknowledges.

An acknowledge button matters as much as the alarm. Without it, carers learn to ignore the alert.

Privacy by design

Radar output is the reason this is acceptable in bedrooms and bathrooms: distance, presence, motion, and state — no video, no audio, no image. With a broker you run yourself, data stays on the LAN by default; retention becomes a policy choice you control rather than a vendor's. If you are deploying in the EU, treat that local-only default as the starting point for your own compliance work and confirm details with your own advisor — this article is engineering guidance, not legal advice.

Per-room deployment checklist

• Bathroom / shower. Ceiling-mount above the fixture and the open floor, not above the door; coverage for a small room is on the order of 3 m × 2 m, height ≤ 2.8 m. Small mistakes here dominate false alarms — see fall detection failure modes.
• Bedroom. Side-wall at 1.8–2.2 m, or above the bed head tilted 30°–45°; breathing degrades first if you exceed roughly 1.5 m. The sleep radar mounting article has the geometry.
• Living room / hallway. Ceiling-mounted, long side toward the door, within about 3° of level.
• Everywhere. Keep fans, air-conditioner internal motors, swaying metal curtains, and pets out of the main lobe; the occupancy relay article lists the same interference sources from the module datasheet.
• Keep to 2.4 GHz Wi-Fi for the finished units.

Limitations you should design around

• No person counts. Treat multi-occupancy as ambiguous; nobody should assume these modules resolve several people reliably.
• Derived values. Sleep stage and sleep score are computed, not measured. The sleep-sensing article explains what is measured versus inferred.
• False alarms are a tuning problem, not a defect. Differences between a small bathroom and a large bedroom change the right sensitivity and dwell settings.
• Test in situ. Ceiling height, layout, and reflections differ per home; a few nights of local data beats any default.

Where to go next

• The three radars in detail: fall, sleep, presence
• Broker you own: Apache Artemis setup
• Device configuration: setup guide
• Certifying your own product: certification support

Evaluation units and modules are on Tindie.