Calibration
Calibration. Most questions start here.
Automatic calibration needs the room empty and still for a continuous 2 hours. Most calibration questions trace back to that one condition.
What does calibration do?
It records the normal background vibration of the room while nobody is in it. After that, the sensor can separate a person from everything else that moves or vibrates in its view. By default it runs automatically and needs no input from the user. A faster manual option is described below.
When does it run, and how long does it take?
Automatic calibration starts on its own after installation. The sensor calibrates itself once the room has been empty and still for a continuous 2 hours, using the data from that window. If someone enters during that time, the calibration is cancelled and starts again. If the room is left empty right after installation, it can finish in as little as 2 hours, so it is safe to allow about 3 hours. The 2-hour window can be changed for a site if needed.
What happens if the room is never empty for 2 hours?
If people keep coming in and out without a 2-hour break, automatic calibration will not complete. Vital signs are still measured. What suffers is presence detection: the sensor may be slow to show the bed as empty, or may switch between occupied and empty more often than it should. For rooms like this, use manual calibration.
Does the room really need to be empty? A calibration ran with someone in it.
Yes. A calibration taken with a person in the room treats that person as background. The usual result is that presence drops out even though the person never left. The fix is to reset the sensor and calibrate again with the room empty.
What if the person stays in bed all day? Could the sensor mistake them for an empty room?
A person who lies very still for long periods can look like an empty room to the radar, so an automatic calibration could start with them in it. A calibration taken with a person present ends with a very low calibration index, usually -10 or lower. Sensors on firmware 12.15.11 or later guard against this: if an automatic calibration finishes at -10 or lower, its result is not applied and the previous calibration stays in use. On earlier firmware there is no such guard, and an automatic calibration is applied even when it ends below -10. A manual calibration is applied as it is on any firmware, whatever its index, so run it only when the room is empty.
Is there a faster way?
Yes, manual calibration. It is triggered after the room is vacated and takes roughly 5 to 20 minutes, depending on the environment. The room must stay empty until it is complete. It suits rooms that are rarely empty for 2 hours, or a site that needs to go live quickly. We normally rely on automatic calibration and use manual calibration to speed things up. After a manual calibration, automatic calibration is switched off by default and the manual result is kept. Automatic calibration can be switched back on afterwards if wanted.
Can calibration be done remotely?
Yes. A manual calibration can be triggered, checked and reset remotely from the sensor settings, and the detection distance (end range) can be changed the same way, with no need to visit the room. Changing the end range clears the previous calibration, so calibrate again afterwards. Run the calibration at a time when the room will stay empty for about 20 minutes.
Does the sensor need to be calibrated again later?
Not as a routine. With automatic calibration on, the sensor keeps recalibrating on its own whenever the room is empty, for example while the person is out. After a manual calibration it keeps that result until it is recalibrated. Recalibrate when:
- The sensor is moved, or its angle changes.
- The detection distance (end range) is changed. Changing it clears the previous calibration.
- Large furniture or equipment near the bed is moved or added.
Is calibration part of installation, and can our own technicians do it?
Your own technicians can calibrate: automatic calibration needs no input, and manual calibration is triggered from the sensor settings. On-site support or a site-specific calibration plan needs a separate discussion with us.
How do I know whether calibration has completed, and whether it went well?
Three values tell you, all visible in the sensor tool:
| Value | What it shows | How to read it |
|---|---|---|
| Calibration status | Whether and how often calibration has run | 0 = never calibrated. 1 or more = at least one calibration has completed. |
| Calibration index | How quiet the room was during calibration | See the index guide below. |
| Calibration progress | Progress of the current run | Complete at 100. |
The calibration index reflects how much background vibration the sensor found in the empty room:
| Calibration index | Room condition | What it means |
|---|---|---|
| Above 0 | Quiet | Ideal. Presence and vital signs are read reliably. |
| -2 to 0 | Normal | A typical room. No action needed. |
| -5 to -2 | Slightly noisy | Works, but check for vibration sources if presence is unstable. |
| Between -10 and -5 | Noisy | The sensor may miss a person who is present. Find and remove the vibration source, then recalibrate. |
| -10 or lower | Too noisy | On firmware 12.15.11 or later, an automatic calibration ending here is not applied and the previous calibration stays in use. On earlier firmware it is applied as it is. A manual calibration ending here is always applied, so find the cause and run it again. |
The calibration index is Noisy or Too noisy (-5 or lower). What should we check?
A very low index almost always means something in range is vibrating or moving. Check, in order:
- Equipment near the bed or in the sensor's view: fans, air conditioners, oxygen concentrators, ventilators, speakers, analogue clocks.
- Anything that sways: curtains, pull strings.
- Vibration carried through the wall or from pipes.
- The sensor itself: it must be fixed firmly, with no wobble. A sensor clipped to a stand or pole picks up its own movement.
Once the source is moved or switched off, reset and recalibrate. If the source cannot be moved, shortening the end range so it falls outside the sensor's reach usually helps.
Does a fan or air conditioner need to be turned off?
Usually not. A fan matters only when it is closer to the sensor than the person is, for example a ceiling fan next to a ceiling-mounted sensor. For a sensor beside the bed, a fan elsewhere in the room is rarely an issue, and calibration generally absorbs it.
Can a clock interfere?
Only in particular positions. The sensor reads the closest source of movement, so a clock farther away than the person does not interfere. It can be picked up when the room is empty, or when it sits within about 20 cm behind the sensor through a thin wall. A digital clock, or moving the clock out of the sensor's view or range, solves it.
Is SUHO calibrated the same way?
Yes, the process is the same. SUHO is tuned to find very faint breathing, so it is more affected by small vibration sources than the XK300. Placement and a clean calibration matter even more for SUHO.
Installation and positioning
Placement. Where most reading problems start.
Mount the sensor firmly, aimed at the chest of the person lying down, with the centre of the bed within 3 m.
How high should the sensor be, and how far from the bed?
For the XK300 Essence, about 1 m above pillow height, on the wall at the head of the bed or beside it, facing the chest. The centre of the bed should be within 3 m. The heartbeat signal is clearest when the chest is about 1 to 1.5 m from the sensor, and it weakens with distance. Mounting much higher can push the chest out of the beam, and mounting lower than about 1 m makes it more likely to pick up a partner. The XK300 is mounted on the wall at the head of the bed or on the ceiling, and its radar board is aimed at the chest by the installer.
How far can the sensor see?
The reach depends on what is being measured:
| Distance from sensor | What can be measured |
|---|---|
| Up to 3 m | Movement, breathing and heart rate |
| 3 to 7 m | Movement and breathing |
| 7 to 9 m | Movement only |
The sensor ignores everything beyond its end range. The end range is 3 m by default and can be changed remotely, for example shortened so a partner falls outside it. Calibrate again after changing it.
Why does the distance value read about 70 cm when the person is closer?
The distance value does not go below about 70 cm, so when the person is closer than that it stays at about 70 cm. This affects only the distance value: heart rate and breathing are still measured normally when the person is closer than 70 cm. The distance value is averaged over about 40 seconds, so it changes slowly. A distance that jumps between two values means another, stronger source of movement is in range.
Two people share the bed. Which one is measured?
The sensor measures the nearest person who is lying still. Mount it on the side of the person being monitored and set the end range short enough that the partner falls outside it. The end range can be shortened remotely from the sensor settings, with no need to visit the room. A short overnight reading of the partner, for example while the user is in the bathroom, does not distort the data as a whole.
Can staff or visitors entering the room affect the data?
The radar does not tell people apart, so presence shows anyone in the room. Vital signs are taken only when a person is almost completely still, and staff are usually moving, so their readings are rarely captured.
Can the sensor be clipped to an IV pole or bed frame?
No. Fix it to the wall with the supplied double-sided tape or screws, angled toward the bed. Anything that lets the sensor move or wobble adds its own movement to the measurement.
What can the signal pass through?
Wood, drywall and thin glass. Not concrete or metal. Metal matters only when it blocks the line between the sensor and the chest, for example a steel upper bunk. A cover over the sensor must not be metal, and recessing the sensor deep into a wall narrows its 130 degree field of view.
Can two sensors be used in the same room?
Yes. Sensors more than about 30 cm apart do not interfere with each other. Set each one's end range so it covers only its own bed.
Is the sensor waterproof?
No. It is for indoor use only.
Can it be used in a bathroom?
The XK300 and XK300 Essence are built for beds. For bathrooms, where someone who stops moving may need help, SUHO is the product to use. Plan bathroom placement carefully: in a small bathroom the sensor can pick up movement outside the room.
Readings
Readings. Taken only when the person is still.
Heart rate and breathing are measured only while the person is still, so gaps in the readings are expected and usually mean the person was moving.
Why are there gaps in heart rate and breathing?
A value is taken only when the whole body is still: not talking, turning over or reaching for something. We do not guarantee a value every 5 minutes. A correctly installed sensor collects many valid readings over a day, mostly during sleep and rest.
Heart rate is missing but breathing comes through. Why?
Heart rate needs a cleaner signal than breathing, so placement problems show up in heart rate first. Check that the bed is within 3 m and that the sensor faces the chest. Readings are also less reliable on days when the breathing signal itself is weak, since the sensor uses it as its reference. The reverse can also happen: heart rate can be shown while breathing is not, because each value is qualified separately.
What is the stability score?
It shows how settled each reading is, from 0 to 100. The dashboard and the aggregated API show only readings scoring 80 or above. The real-time WebSocket feed sends every reading, so if you use it, keep only heart rate where its stability is 80 or above, and the same for breathing. The movement index can be used as it is.
How is each value produced?
The sensor follows about 20 seconds of signal to produce each heart rate and breathing value, and needs at least 30 seconds of stillness before the first one. The sensor produces a new reading every 3 seconds. History is shown as 5-minute averages. The underlying 3-second data can be downloaded with the download button in the dashboard, or by a separate request. Because raw data is costly to prepare, the number of raw-data downloads may be limited.
Heart rate sometimes looks about half or double the real value. Why?
The radar reads the movement of the heart through the chest wall, not an electrical signal. For some people, the way the heart moves produces a strong component at twice or half the beat rate, and the reading can follow it for a while. This is a property of that person's cardiac motion, not a fault in the sensor or the algorithm. It is uncommon, and isolated half or double readings do not move the person's daily baseline much. People who show it tend to switch back and forth between normal and half or double readings over hours of monitoring, so a short run of them is expected rather than a sign of a problem.
The readings look lower than my chest strap or watch. Which is right?
Wearables are not a reliable reference for this comparison. For a meaningful check, compare against a hospital ECG for heart rate and end-tidal CO2 capnography for breathing, side by side in real time.
How is this different from a wearable?
First, nothing is worn. Readings are taken automatically, with nothing for the person to put on, charge or remember.
Beyond that, the sensor looks at the whole body, not one spot on it, and takes a reading only when the whole body is still, that is, when the person is truly at rest. Because every reading comes from the same resting state, a change from the person's own normal stands out clearly, and a meaningful deviation can be told apart from the ordinary ups and downs of a day.
A wearable reads from one part of the body, such as the wrist or the chest, and keeps measuring while the person is talking, moving or otherwise not at rest. Its readings mix resting and non-resting moments, which makes a meaningful deviation from the person's baseline comparatively harder to find.
What is POBC?
POBC (probability of baseline change) compares the last 24 hours with the same person's previous 7 days, on a scale from -100 to 100. Near 0 is an ordinary day; the further from 0, the more unusual. It needs enough data to be meaningful: about 700 valid readings in a continuous 24 hours, and at least 2 days of history. A value of 123 means there was not enough data.
How is the sleep score calculated?
Mainly from the breathing waveform: steadier breathing scores higher. Sleep and wake are judged from movement and vital-sign trends over several minutes. It is a monitoring measure, not a substitute for a sleep study (PSG), so sleep and wake times can differ from a PSG.
What does presence show, and what counts as empty?
Presence shows whether anyone is in the sensor's range. After a person leaves, the sensor waits a short hold time before marking the room empty, to avoid false alerts when someone lies very still. A room that keeps flipping between occupied and empty usually points to calibration (see Calibration).
Does the sensor alert when breathing stops?
No. To the radar, a person who has stopped breathing looks like an empty room, and the product is designed for early warning of change, not for acute events. For a person who has collapsed and stopped moving in a space where movement is expected, SUHO is the product.
Does SUHO alert when someone falls but keeps moving?
Not with the Motionless alert. SUHO looks for a person who is breathing but not moving at all, so a person who has fallen and is still moving, for example shifting on the floor or trying to get up, does not trigger it, and any movement starts the count again. For that case the Long Stay alert is the safeguard: it fires when the space stays occupied without a break, after 8 minutes by default.
Connectivity and devices
Connectivity. Power, network, then the LED.
When data stops while our servers are running, the cause is almost always local power or network, and the LED tells you which.
Which Wi-Fi does it support?
2.4 GHz only. 5 GHz networks are not supported, so check that the router has a 2.4 GHz band switched on.
Do LTE units need a local SIM?
LTE units ship with an IoT SIM already fitted and configured, so the site needs only power and mobile coverage. In some countries roaming on that SIM is limited to a few months; long-term use there needs a SIM from a local operator. Tell us the country before ordering.
What do the LED colours mean?
| Where | Colour | Meaning |
|---|---|---|
| Sensor (beside the USB-C port) | Green for a moment when power is connected, then off | Normal. The sensor LED stays off during use so it does not light the room at night (see below). |
| Sensor | Yellow or red | Needs attention. Send us the serial number and the colour, or a photo. |
| Communication unit | Steady, not blinking | Connected. On LTE units, green is the correct state. |
| Communication unit | Light blue, then yellow | Booting, then waiting for a network. Expected during first setup. |
| Communication unit | Rainbow or white blinking | Firmware or setup issue, usually fixable remotely or with a firmware update from a PC. |
The LED on the sensor is off. Is it broken?
Usually not. By design, the sensor LED turns green only briefly when power is connected, then stays off, so it does not create light in the room at night. An unlit sensor LED during normal use is expected.
When reporting an LED problem, please tell us which LED you mean: the one on the sensor, or the one on the communication unit (gateway). The communication unit LED stays on while it is powered. If the sensor shows no green light even at the moment power is connected, and the communication unit is powered, contact support with the sensor serial number.
A sensor shows as offline. What should we check?
In order: power to the communication unit, the sensor cable, then the network (Wi-Fi signal or LTE coverage). If several units drop at the same moment, the cause is usually the network rather than the devices. A power cycle of the communication unit fixes most cases. If it stays offline after that, contact support with the serial number.
How many sensors per gateway?
The current communication unit (LGW) pairs with one sensor. The earlier gateway (XK-GW) takes up to four. With the earlier XK-GW on a wired network, plan two to three sensors per gateway where rooms are close together, and one per room where they are not.
Can the data go straight to our own server instead of the XK cloud?
Yes. The vital-sign processing runs on the sensor itself, so the communication unit can send directly to your server over MQTT, and the data never passes through our cloud. You then aggregate the raw data yourselves. This works with the XK300 and XK300 Essence and their communication unit. If you use our API instead, the data goes through our cloud.
How are firmware updates done?
Communication units check for updates automatically every 30 minutes, so no action is needed. For a large fleet, updates can also be pushed remotely in batches, and units can be ordered with your server endpoint and certificate pre-loaded so they connect to your system on first power-up.
Dashboard, alerts and data access
Data access. One source, three ways in.
The dashboard, the API and the real-time feed all draw on the same data, so anything on the dashboard can also be pulled through the API.
What is the difference between the XKare app and the Insights dashboard?
XKare is for a person at home and the family they invite, and updates about every 5 minutes. The Insights dashboard is for facilities and care teams watching many beds at once, and updates in near real time, every 3 to 30 seconds depending on the view. The two run separately and do not share accounts.
How do real-time alerts decide when to fire?
They work on fixed 3-minute blocks. An alert fires when the average of the reliable readings in a block crosses the threshold the facility set, and enough readings were collected in that block. Because the blocks are fixed, an alert can fire less than 3 minutes after someone enters the room.
What does the API provide?
| Channel | What it carries | How often |
|---|---|---|
| REST API (API key over HTTPS) | Heart rate, breathing, movement, POBC, sleep score and duration, awakeness | 5-minute history by default; daily metrics after processing |
| Webhooks | Alerts: vital-sign thresholds, POBC, presence changes | As they happen |
| Real-time feed (Socket.IO) | Heart rate, breathing, presence, movement, with stability scores | Every 3 seconds per sensor |
The REST API returns only readings with stability 80 or above. The real-time feed sends everything, so filter on stability yourself. See the Kardian API Guide for the details.
How do we connect to the real-time feed?
Use a Socket.IO version 4 client, not a plain WebSocket, and authenticate with your API key. Join the room for your building ID; data then arrives with the sensor serial number as the event name. Delivery is best effort, so after a reconnect, join your buildings again.
How do we get API access?
After an NDA and a sensor order, we issue the API documentation and a key tied to your account. A test building with sample data can be arranged for development.
Is sleep data in the API?
Yes, through the REST API's daily metrics, which return the sleep score and sleep duration for a given date. It is not in the real-time feed.
How long is data kept, and can we export it?
Data stays in the working database for 2 years, then in cold storage for 4 more. From the dashboard history view you can pick a date range and download CSV of the 5-minute data and daily metrics. The 3-second data is a separate download (see How is each value produced?).
Can XK send data to our EHR?
EHR integration, for example with Epic, is not a standard feature. It needs a separate discussion with us.