English is currently available.
Zigbee, Thread & Matter DISCUSSION
My Zigbee network has a USB coordinator on a small server and about 25 devices: bulbs, smart plugs and battery sensors. The sensors furthest from the coordinator go offline every few days, and some never rejoin until I re-pair them. The coordinator is on the default channel 11 and the Wi-Fi router sits about a metre away.
Does Wi-Fi really interfere with Zigbee, which channel should I use, and what else makes a Zigbee mesh unstable?
They share the band, so yes. Zigbee channels 11 to 26 are 2 MHz wide with centres at 2405 + 5 × (k - 11) MHz: channel 11 is 2405 MHz, 15 is 2425, 20 is 2450, 25 is 2475 and 26 is 2480. A 20 MHz Wi-Fi channel covers about ±11 MHz around its centre: channel 1 at 2412, channel 6 at 2437 and channel 11 at 2462 MHz. Zigbee channel 11 therefore sits inside Wi-Fi channel 1.
If the neighbourhood uses Wi-Fi 1, 6 and 11, the gaps are at Zigbee 15, 20, 25 and 26. Channel 26 is often avoided because some devices transmit at reduced power there or do not support it, so 15, 20 or 25 are the usual picks. Fix the router to one 20 MHz channel rather than automatic selection so the gap stays where you put it.
A metre from the router is too close regardless of channel. A Wi-Fi transmitter is often 10 dB or more stronger than a Zigbee device, and at that distance its energy outside the nominal channel still desensitises the coordinator's receiver. Move the dongle away on a USB extension cable of a metre or two.
The extension cable solves a second, well-documented problem: USB 3.0 ports and devices radiate broadband noise in the 2.4 GHz range. A dongle plugged directly into a USB 3.0 port, or sitting next to a USB 3.0 disk, can lose a large part of its range. Use a USB 2.0 port and keep the stick clear of the machine.
Look at the mesh structure. Only mains-powered devices act as routers; battery sensors are end devices that talk to a single parent and never relay. A sensor far from the coordinator depends on a chain of routers, and if its parent disappears it has to find a new one. Smart bulbs are routers too, so when someone switches a bulb off at the wall, every sensor that used it as a parent is orphaned. That matches devices vanishing every few days.
Add a few always-powered routers (smart plugs are good for this) between the coordinator and the far rooms, and do not cut power to routing devices. Coordinators also have a limit on direct children that depends on the firmware, so with 25 devices the routers are doing real work.
About changing the channel of an existing network: the coordinator can announce a channel change, and mains-powered devices usually follow, but sleepy battery devices often miss the announcement and have to be woken and rejoined or re-paired. Plan the change for a time when you can walk around to the sensors, and give the mesh a day to settle before judging it.
When you pair a sensor, do it in its final location, or pair it through the router it will use, instead of next to the coordinator. Some end devices are slow to look for a better parent after being moved and hold on to a weak link.
To confirm rather than guess, use the link quality figures your coordinator software reports (LQI per link) and its network map. Links with consistently low LQI, or sensors whose only route is through one weak router, show where to add a repeater. Some coordinator firmware can also run an energy scan that measures the noise on each of the 16 channels, which is a better basis for choosing a channel than theory alone.
Battery state matters as well. A coin cell near the end of its life can power the sensor while it sleeps but sag during the transmit burst needed to rejoin, which looks like a device that falls off and never comes back.
