Appearance
The calibration bench
/admin/manage/device/tag-offsets/bench — Device → Tag Power → Measure
/admin/sysop/calibration-bench — SYSOP → Calibration Bench (platform operators only)
Two radios of different make do not read the same at the same distance. A wristband and a card tag lying side by side are heard a few decibels apart by every gateway, and two gateways of the same model on one shelf hear the same tag a few decibels apart. Omaya turns signal strength into distance, so a constant difference like that becomes a distance that is wrong by a multiple: six decibels makes every range about 1.7 times too long, thirteen makes it three times.
Those differences are properties of the hardware. Nothing can look them up and nothing can work them out from a datasheet; they have to be measured once. The calibration bench is the page that measures them, guides the person doing it, and refuses to save a measurement that does not agree with itself.
An example. Northwind Tower walked every floor with a staff card. Visitors get wristbands. On the Live Monitor, staff land in the right rooms, and visitors always seem to be a room further from every gateway than they are. That is the wristband transmitting quieter than the card: every gateway hears it a few decibels down, and Omaya reads "quieter" as "further away". Daniel Lim (Technician) measures the wristband on the bench once, in about twenty minutes; after that, every wristband is ranged as if it were a card.
What it measures
Tag power, per tag model. How much louder or quieter a model of tag transmits than the reference tag: the tag your floor's calibration walk was carried with. Every gateway's "signal at one metre" was measured with that tag, so every other model has to be compared against it. Once a model is measured, positioning takes the difference off every reading from tags of that model before it works out a distance.
Gateway hearing, per gateway (platform operators only). How much louder or quieter a gateway hears than other gateways standing beside it. Presence compares gateways with each other to decide which room a tag is in, so a gateway that hears a few decibels hot pulls tags towards its own room.
Where to find it
| Who | Where | What it can measure |
|---|---|---|
| An organisation's administrator (anyone who can manage devices) | Device → Tag Power, then Measure | Tag power, for their own organisation only |
| A platform operator | SYSOP → Calibration Bench | Tag power and gateway hearing, for any organisation |
The Tag Power page can also be reached from a single tag: open the tag under Device → Sensor, expand its Tag Power card under Calibration, and press Tag Power calibration.
Both are the same page. The organisation's version is locked to that organisation and to tags. Measuring gateways means taking them off the ceiling and the result feeds a switch that is off by default, so that half is kept for whoever installs the hardware.
If the menu entry is missing after an update, reload the page once: the menu is built when the page first loads.
Before you start
You need:
- The reference tag. The bench picks it for you: the tag the most recent calibration walk was carried with. If no walk names one, pick the tag model your organisation uses most.
- One tag of each model you want to measure. Up to eight at a time. Two tags of the same model are fine and useful, see Checking one tag against its siblings.
- A spot near one gateway, with nothing between the tags and the gateway: no metal cabinet, no wall, no person. One to three metres away is ideal.
- About twenty minutes, mostly waiting.
- The tags awake. In the tag list a green dot means the tag was heard in the last ten minutes; a grey dot means it was not. A tag that is asleep, or out of battery, will show "not heard" in the result.
Each tag and gateway is listed with its name, its model and its MAC address, so two tags of the same model can be told apart. Match the MAC with the label on the back of the tag. Each gateway is also listed with its floor and the room it is assigned to, so you can find the one in the room you are standing in.
With many tags or gateways, use the search box above each list. The tag boxes match a name, a model or a MAC address (with or without the colons); the gateway box also matches a room or a floor. Searching only hides the rows that do not match: anything already ticked stays ticked, and a dropdown keeps its current choice, so typing in a search box never changes what is selected.
Measuring tag power, step by step
1. Lay the tags out
Put the reference tag and every tag you want to measure side by side, touching, the same way up, on a table or a shelf.
Side by side is the part that matters. Two tags lying together reach every gateway by exactly the same path, through the same walls and off the same surfaces, so whatever is different between them is the tags themselves. The room cancels out.
How far away the gateway is matters much less. Close gives clean readings: at one to three metres the gateway hears every advertisement, and a minute is plenty. Further away still works, but the gateway misses some frames, the readings scatter, and you need longer rounds to get the same confidence.
2. Pick what is on the bench
Reference tag. Already selected. Leave it unless you know better.
Tags under test. Tick the tags lying beside it, at most eight. Left empty, a round simply reads the reference tag on its own.
Gateways that count. Tick one gateway: the one nearest the tags, in the room you are standing in. The bench ticks one for you when the page opens or the reference changes: the gateway that heard the reference tag loudest in the last thirty seconds. A line under the list says which one it picked and why. If that is not the gateway beside you, untick it and tick the right one. If no gateway has heard the reference tag in the last thirty seconds, nothing is ticked and the line says so: check the tag is awake, then tick the gateway yourself. A round will not start until a gateway is ticked.
Do not tick gateways in other rooms. They hear the tags through walls, their readings scatter, and that can turn a good capture into one the bench refuses to save.
Distance to the gateway, roughly (m). Optional when measuring tags; it only feeds the Implies at 1 m column.
Seconds per round. Sixty is a good default.
Save for. The whole organisation is right almost always. Pick a single floor (Only followed by its name) only when that floor was calibrated with a different reference tag from the others.
3. Run the rounds
Press Start round 1 and step back from the table. Bodies absorb signal: a person standing next to the tags changes the reading.
While a round runs, a progress bar fills and a live table shows, for every tag on every ticked gateway, the middle reading so far and how many frames counted, as "−52 · 20 of 60". That second figure is normal. Gateways re-send the last frame they heard from a tag over and over; the bench counts only frames the gateway actually heard again, so "20 real out of 60 reported" is what a healthy gateway looks like.
When the round ends, the bench waits a few seconds for the last frames to arrive, records the round as a chip (Round 1 · 10:42:05 – 10:43:05) and shows the verdict so far.
Then turn every tag a quarter turn and start the next round. A tag's antenna does not send equally in every direction; four rounds at four orientations average that out. Four rounds is a full set.
- Stop throws the running round away, for example if someone walked past.
- The × on a round chip removes that round, and the verdict is worked out again without it.
- Recalculate works the verdict out again from the rounds you have.
- Clear rounds starts the whole capture again.
The page keeps at most eight rounds of up to ten minutes each.
Once a round is recorded, the organisation and the mode are locked, so one capture never mixes two set-ups.
4. Read the result
For each tag under test:
| Column | What it means |
|---|---|
| Against (reference) | How many decibels louder (+) or quieter (−) this tag reads than the reference lying beside it. This is the number that is saved. |
| Disagreement | The gap between the highest and lowest difference seen across rounds and gateways. Small is good: two tags on the same spot should differ by the same amount whoever listens. |
| Readings | How many round-and-gateway pairs had enough frames from both tags to compare. |
| Verdict | See below. |
| What it means | In plain words, how wrong its distances were before: "Its distances were coming out 1.68 times the true distance." |
The verdicts:
| Verdict | Meaning | Can be saved? |
|---|---|---|
| clean | The readings agree within 3 dB. | Yes |
| usable | They agree within 6 dB. | Yes |
| not side by side | They disagree by more than 6 dB. The tags were not really together, or one was against a body, a wall or metal. | No |
| one reading only | Only one round on one gateway had enough frames. Run another round. | No |
| not heard | No ticked gateway heard both this tag and the reference well enough. Check the tag is awake and the right gateway is ticked. | No |
| same model | The tag is the same model as the reference. Shown so you can compare two tags of one model; never saved. | No |
Under the table, What Save will write lists each model with its new offset, what is saved for it now, and how many tags were measured. When two tags of the same model are measured, the model gets the middle of their results. The reference model is always written as 0.0 dB, so every other row says what it was measured against.
Plain readings over all rounds lists every tag on every gateway with its middle reading, its scatter and how many real frames counted. When you entered a distance of five metres or less, Implies at 1 m shows what that reading means at one metre with a clear line of sight. Comparing that figure with the gateway's own 1 m setting is only meaningful for the reference tag, because that setting was measured with the reference.
5. Save
Save tag offsets is enabled only when at least one model can be saved. It asks for confirmation and lists exactly what it will write. The bench works the result out again on the server from the recorded rounds before saving; nothing typed or shown in the browser is trusted.
Positioning starts using the new offsets within ten minutes. After saving:
- Tag Power shows the model as measured, with the offset, how it was measured and the disagreement.
- Opening a tag under Device → Sensor shows the offset in force in the Tag Power card under Calibration. The card is collapsed to one line that reads Reference model · 0.0 dB, Not measured, or the offset in force, with a warning sign when tags of that model advertise different power. Click it to see where the number comes from; the page remembers in your browser whether you left it open.
6. Check that it worked
- Back on Tag Power, the model's row now shows its Offset and what it was Measured by, and the line under it says how far wrong its ranges were coming out. The note above the table counts how many models are still Not measured.
- Wait ten minutes, then carry a tag of that model into a room you know and watch it on Monitor → Live Monitor. It should now sit as close to the gateways as a tag of the reference model does in the same spot.
- If it still lands a room away, open Why here? for it. A tag that is now ranged right but still in the wrong room has a different problem — see A tag shows in the wrong room.
The Tag Power page
Device → Tag Power. One row per tag model your organisation owns:
| Column | What it shows |
|---|---|
| Model | The tag model, the power its tags advertise (after Read from tags), and a warning when tags of this model advertise different power. |
| Offset | The saved difference from the reference tag, in dB, with Ranges were coming out …× — how wrong its distances were before. The reference model is marked Reference tag; a model nobody has measured is marked Not measured and is ranged as if it were the reference. A floor with its own offset shows as a chip under it. |
| Measured by | How the offset was set, with how many readings and the spread. |
| Tags | How many tags of that model you have. |
Read from tags records the transmit power each tag states in its own advertisement over the last six hours. Where two models state the same kind of figure, the page shows what their advertisements suggest (Its own advertisement suggests …). Treat that as a hint, not a measurement: many tags state a figure that cannot be compared with another make's — one states its power at the antenna, another the signal expected at one metre — so the difference between them is not the difference you would measure. It is most useful for the warning it produces when tags of one model state different powers.
Checking one tag against its siblings
Tags of one model are supposed to transmit the same, but one tag can be out of step, either because of how it was configured or because its radio is different. The Tag Power card on a tag warns when tags of the same model advertise different power.
To find out whether the difference is real, put that tag on the bench with another tag of the same model next to it, the first as reference. The result is shown with the verdict same model and is not saved. If they really differ, give the odd one its own offset in its Tag Power card under This tag's own offset, rather than changing the whole model.
Do this first for the reference tag itself. Every other model is measured against it, so if the reference is the odd one out, everything else inherits its error.
Measuring gateway hearing (platform operators)
SYSOP → Calibration Bench, then Gateways, side by side.
- Take the gateways down and stand them together on one table, antennas the same way, powered and connected.
- Put one tag about a metre in front of them, and enter that distance. It is required here.
- Pick that tag under Tag the gateways listen to, and tick only the gateways on the table, at least two.
- Run four rounds, turning the tag a quarter turn between them.
For each gateway the result shows its middle reading, how many decibels it hears above or below the group, the disagreement across rounds and a verdict.
The bench refuses to save when any ticked gateway reads too quietly to be standing at the distance you entered. Its verdict reads too quiet for this distance, and the page names it: "Gateway04, Gateway08 read too quietly to be standing 1.0 m from the tag." That check exists because a gateway still on the ceiling across the room agrees with itself perfectly round after round while measuring nothing but how far away it is. Untick it, clear the rounds and capture again.
Saved gateway offsets take effect only when the organisation's Allow for gateways that hear louder than others switch is on, under System → Configuration → Indoor Positioning, on the Rooms tab. Saving does not turn it on; the page says so when it is off. Turning it on changes how every tag in the organisation is assigned to a room, so check the result on a quiet day.
Every gateway capture is kept, whether you save it or not: each gateway's offset, how well the rounds agreed and when it was measured. A clean result (rounds that agree) from the last 30 days can then be applied later without measuring again — by the organisation's administrators through OAIA, which offers it on its review and applies exactly the measured number. Saving here applies it at once, as before.
Changing or undoing an offset
Tag Power lists every model. The pencil on a row opens it: change the offset by hand, set it per floor under Applies to, or Clear offset. A cleared model is ranged as if it were the reference tag, which is how every model behaved before it was measured. Measuring again on the bench simply replaces the saved value.
Questions
Do the tags have to be exactly one metre from the gateway? No. Side by side is what makes the measurement right. One to three metres from the nearest gateway is recommended because the readings are cleaner and the rounds can be shorter.
Should I tick one gateway or several? One: the nearest. A second nearby gateway that hears the tags clearly does no harm, but gateways in other rooms make the result scatter and may stop it from saving.
The reading says 20 of 60. Is the gateway losing packets? No. Gateways repeat the last frame they heard for up to two minutes. The bench counts only frames the gateway heard again, which is usually a third to half of what it reports.
Can I measure a model I only have one of? Yes. One tag of a model, beside the reference, is enough.
Why can I not save a tag that reads +2 dB? Look at its verdict. If it is the same model as the reference it is never saved. If it is one reading only, run another round.
How often do I need to do this? Once per tag model, and again only if the reference tag changes, for example after a calibration walk with a different tag. New tags of a model already measured need nothing.
Does this change the rooms people are shown in? Tag offsets change distances, which mostly affects where on the floor plan a tag is drawn and which room the room estimator picks for tags of that model. It does not touch tags of the reference model.
Every tag says "not heard". No ticked gateway heard both the tag and the reference often enough. Check the green dots in the tag list (heard in the last ten minutes), that the gateway ticked is the one beside the table, and that the gateway is online under Device → Gateway. A tag in a shipping or sleep mode may need a button press or a magnet to start advertising.
One tag reads "not side by side" every time. Its readings swing more than 6 dB between rounds. Check it is touching the reference and the same way up, that nobody is standing at the table, and that no metal shelf or wall is between the group and the gateway. If it still fails, compare it with a sibling of its own model — the tag itself may be the odd one out.
Can I measure without the reference tag? Not usefully. The offset is the difference from the reference, so without it there is nothing to subtract. If the reference tag is lost, walk the floor again with the tag you will use from now on, then measure every other model against that one.
Which floor's offset is used? A floor with its own offset uses it; every other floor uses the All floors value. A tag's own offset beats both.
Related
- Floor plans and positioning, for the calibration walk and why the tag it was carried with matters.
- Your things and your spaces, for the Sensor page and the Tag Power card.
- A tag shows in the wrong room, for how a model that was never measured shows up on the floor.