Signal Strength

Signal Strength

Signal is experimental. Everything here is recorded and reported, and nothing here raises an alert. Signal will not notify you until it has been shown to predict real failures, and that takes examples of meshes that actually misbehave. If yours does, see Help Map Signal Failures at the foot of this page.

Device Sentinel watches the radio link quality of every device that reports one, learning each device's normal range and watching for a link that sits unusually low against its own normal. This screen tunes how that judgment works and who is muted from it.

Judged Against Itself, Never a Chart

There is no universal number for a healthy signal. A sensor two rooms from the coordinator lives its whole life at a level that would mean trouble for one in the same room, and both can be perfectly reliable. So Device Sentinel never compares a device to a fixed threshold or to other devices: each is judged against its own history alone.

Two scales exist, LQI (higher is better) and RSSI in dB (closer to zero is better). Each device is judged on whatever it reports, and the two are never mixed.

Readings stuck at a rail value (255 LQI or -128 RSSI) are stale reports, not strong or weak links. They never feed the learning, and a device pinned there for three or more days is flagged as railed, which is its own finding.

How a Day Is Judged

Each night, at the fold, a device's readings for the day become two numbers: the time-weighted fifth percentile of its signal and the time-weighted median. The fifth percentile is what the judgment uses, because it is the low edge of the day rather than the single worst packet, and one bad packet is noise.

That day is compared against the days before it. The median of those baseline days is the device's normal, and their population standard deviation is its spread, which is how far that device wanders when nothing is wrong. A day is bad when the fall from normal clears both gates below: far enough in the device's own units, and far enough in the device's own spread.

A day goes unjudged rather than judged wrongly when the device has fewer than four usable baseline days, when the day's own reading is unusable, or when the baseline holds both positive and negative numbers, which means two scales landed in one series.

The result is on The Signal Report, which is written beside the daily brief.

The Four Sliders

They are listed here in the order they appear on the screen.

• Bad Day Drop, LQI (default 25, range 10 to 60): The fall in LQI points required to flag a bad signal day. The day must clear this gate and the sensitivity gate below, both. Raise it to flag fewer days.
• Bad Day Drop, RSSI (default 6, range 3 to 15): The same fall in decibels, for devices that report in dBm. The gate is in the device's own units rather than a percentage, because a percentage of an RSSI number means nothing.
• Bad Day Sensitivity (default 4, range 2 to 8): How many of the device's own spreads the fall must cover. This is what keeps a jittery device from being flagged for doing what it always does: at 4, a device that routinely swings 10 points needs a 40 point fall. Lower it to flag more days.
• Days of Signal History (default 7, range 4 to 14): How many past days form the baseline that today is compared against. Fewer days react quicker and forgive quicker; more days are steadier and slower to notice a decline.

Three values, Sensitivity, Line Lift and Anomaly Trim, are fixed rather than settings. They feed recorded history alone and nothing you see.

What Alerts and What Does Not

A railed signal becomes a problem list item and follows the normal notification path, because a device reporting a false value is a real finding.

A bad signal day is recorded and reported but never pushes. See Help Map Signal Failures below.

Signal Muting

Muting a device here stops signal judgment and reporting for that device only; its battery and freeze watching continue. For dropping a device entirely, use Exclusions and Muting.

The three lists run broad to narrow (integrations, labels, devices) and interact exactly like the global lists: a broader mute swallows the narrower entries it covers when you save.

Help Map Signal Failures

Signal is the one part of Device Sentinel that watches without judging, and it stays that way until it earns the right to wake you.

The reason is that nobody yet knows which signal pattern predicts a device actually failing. A link can sit low for months and never miss a report. Another can look healthy and drop out weekly. Turning on alerts before that relationship is mapped would mean a stream of warnings about devices that are fine, which teaches you to ignore the one that is not.

What settles it is failures: houses where devices really do drop out, with the signal history recorded either side of the drop. A healthy mesh cannot answer the question, however long it runs.

If your mesh is flaky, you can help. Open an issue at Issues titled contribute signal, and attach two things:

1. A diagnostics download: Settings, Devices and Services, Device Sentinel, the three-dot menu, Download diagnostics. It carries device names and no passwords or tokens.
2. Your signal report, at config/www/device_sentinel/signal_report.html.

Say which devices actually give you trouble and roughly how often. That last part is the piece no file contains, and it is the piece that turns a set of numbers into an answer.