Have you considered interference. Having a lot of devices I’m sure you have looked into it at some point, however it might behave differently on the coordinator. Also USB3, maybe you changed the port?
I’m using a sonoff dongle with Z2M, and it is stable.
That is taken into account in LQI. It’s not just signal strength. LQI incorporates error rate as well as rssi (for each hop, incoming and outgoing).
Hi Tom, Been reading on the LQI part, and there is no standard. I do belive the LQI for a Sonoff is just a mathematical value between the RSSI and the LQI. There is never a difference between the 2. On my Conbee it was also like this.
I know the deficition should include a lot of other stuff, just not sure they include it. Much easier just to make a set difference.
1 Like
LQI is a mathematical value between RSSI and LQI?
That’s rather recursive.
Putting the dongle on a usb extension cord is worthwhile though.
USB3 ports are known to cause interference. Plus you get to put the dongle up higher away from metal objects.
Hi Tom, I did put in “I do belive”
When I was using ZHA (I have moved to Z2M and like it better) the ZHA card was giving a simple overview, showing both LQI and RSSI for all devices, besides each other. There were never a difference in the sequence of the devices, hence I belive the implementation is much simpler than what should be expected.
My lowest LQI is currently 43. If you say that is good enough, I’ll just leave those be unless they malfunction.
I have 5 APs around the house so all my 2.4GHz wifi spectrum is used (Ch 1, 6, 11). Neighbors are not close enough to have strong signals, if anything, I would be the cause for my own interference. I keep my 2.4GHz SSIDs to 40MHz and set them all to 16dB. I only put IoT or low bandwidth / old devices on 2.4GHz and exclusively use 5GHz for high bandwidth devices. My devices when on the ConBee II never had any issues with wifi interference as the mesh was rock solid… but I would not exclude it as I use Unifi hardware and am always on their latest beta firmware. I could not figure out what Zigbee channel I am on.
The sticks are on top of a metal networking rack, on a magnetic stand, connected to a USB3 port via a shielded extender cable. The ConBee II was there too and never had issues.
I am aware that sitting the radios on top of a metal rack will likely introduce some signal blockage in certain directions but the mesh with the ConBee II never seemed to mind. This room is on the upper floor, so the omnidirectional antenna should do a good job on the upper floor, and a poor job (worse) right under and to the left as the lobe is likely blocked to some extent by the rack. But… that is where the mesh should come in… and I have routers all over the house evenly spread.
With the ConBee II the zha-card showed both values. I discovered after switching to the Sonoff coordinator that the latter does not provide RSSI. As for LQI it is still a head scratcher… my highest LQI is 185 and most of the devices with that LQI are in the same room as the coordinator with minimal blockage.
I forgot to mention that I have routers do the same thing. I have some that surely have way better routes back but they ignore them. At a workbench in my office I have 7 plug in switchable sockets and 1 in-wall socket, all zigbee, and their LQIs vary between 43 and 109 even though they are on the same power strip. 
When you re-paired the devices did you force the device to pair thru a specific router?
I have had to do that a few times since the device seems to want to pair directly with the coordinator but another routing device is actually closer and would have a better signal.
you can do that thru the ZHA integration config panel for the device you want to pair with.
At first I thought that they would figure out the best path, then I learnt it is not the case, so now I am using that method. I am watching the mesh to see if the same devices keep dropping off even after using that method. Using that method I have noticed that sometimes a sensor appears to have trouble connecting via one router and less via another. Last night I was trying to re-connect a Smartthings Samjin Button which dropped off several times. Right below it is a zigbee in wall outlet but the pairing was seemingly stalling so I tried with a switch one wall away and it worked 
EDIT: I suspect that given the higher gain antenna on the Sonoff coordinator, devices are trying to pair with the coordinator and then struggle to communicate (pointless to have a huge antenna only on one end… it is a two way communication) as they have a comparatively poor antenna. Maybe by using “Add via this device” and picking a router nearby I can remedy this issue
I have never done anything special with my zigbee mesh except for pairing thru specific routers (no USB extension, etc) and I haven’t had any issues using a HUSZBZ-1 controller.
I was quite happy with my raspi4 but now that I see yours it feels so… I don’t know… inadequate. 
5 Likes
Me neither, up until I moved to the Sonoff coordinator, my mesh was great.
Why did you move?
I guess I would probably move back. 
Home Assistant is running on a Lenovo m920Q (i7 8700; 32GB) in a Proxmox VM so yeah a bit overkill. I am a bit of a geek so ‘overkill’ is fun
1 Like
You’re not using increased tx power for the Sonoff, are you?
When I updated to 2022.4 my zigbee network blew up. While I know that in part it was likely due to sw issues, I had run into this before. A simple reboot or power outage could leave me with a dead mesh until it would start working again. I decided to ditch it in favor of a more modern chipset too that extends the devices that can link to the coordinator directly from 20-30 to 100. A direct connection, if good, would result in faster response time, less traffic, more reliability, etc.
1 Like
i’ve been reading that a lot.
It’s really got me nervous about updating. I’d rather wait till the dust settles on that but It’s hard to know for sure if/when things get resolved in the point releases to make ZHA stable again.
No, but I was considering reducing its rx and tx gains to see if it helps. Sounds backwards but it is like with wifi where blasting the highest wifi signal you can is counterproductive given the client needs to be able to reach your AP using its tiny little poor gain antenna so the best setup is usually with more APs using less power. In the case of Zigbee the APs would be the routers.
That’s why I asked: for a lot of people it’s counterintuitive that decreasing gain can actually result in a better network, especially with lower power devices.
FWIW, I moved from deCONZ to z2m last week, and from Conbee II to Sonoff, and so far so good. I have about 60 devices. I did have to re-pair some Aqara devices after the mesh settled, because they weren’t using the best router, but now that I’ve run it for about a week it looks like the mesh is in much better shape than it was with deCONZ (perhaps because z2m uses source routing which I hadn’t turned on in deCONZ).
I will have to give it a try…
What is that? And do all coordinators have that feature?
Not at all. I don’t know how the zigbee routing algorithms work, but the coordinator’s default +9db may be making it look stronger to the device and therefor preferred.