My Meshtastic station could still hear the mesh as well as it ever had, and nobody outside my house could hear it. Receive stayed flat. Transmit went to zero. That one comparison cleared the antenna, the cable and every connector before I touched any of them, because a bad feedline hurts both directions equally. The community's collectors showed the rest: on one August morning, my broadcasts went from reaching more than a dozen outside gateways each to reaching none, inside a single hour. The cause was the radio's power amplifier, stuck in low-power mode. The fix was a new radio.
On August 26, 2026, I noticed my messages were not getting past the one other node of mine inside the house. Incoming traffic looked normal. So the first question was simple: is the station deaf, mute, or both?
On the receive side, nothing had changed. The signal-to-noise ratio the station measured from fixed neighbors, nodes that never move, was flat across the whole week, within about a decibel.
That matters because the antenna, the coax and the connectors are one shared path. A corroded connector or a crushed cable costs you the same decibels going out as coming in. If receive from fixed neighbors hasn't moved, the passive RF path hasn't moved either. Swapping the antenna or the cable would have been time spent on the one part of the system the data had already cleared.
It wasn't software or config either: no restart in a week, no config change, nothing installed. The radio's airtime counter was steady, so as far as the chip knew, it was still transmitting.
The mesh community runs collectors: gateways that upload what they hear to a shared service, which shows which gateways heard any given packet. I counted, for each day, the share of my station's broadcasts that at least one gateway outside my own gear had picked up.
Four normal days at 81 to 92 percent, then 25 percent, then 4. The daily numbers look like a decline, but packet by packet it was a switch. That morning, each broadcast was still reaching more than a dozen outside gateways. The next one, an hour later, reached none, and so did every broadcast after it, except one in the middle of the night that briefly reached a few. A dirty connector or water in a cable fades. This was something that stopped.
The cheap test came first: a full power cycle. The radio came back with a clean startup log, every power setting accepted, and the same result. Those messages only confirm the chip accepted its commands. They can't see what leaves the antenna.
So I sent three test messages on purpose rather than waiting for routine traffic. The collectors' record for each one showed my own station's upload and no outside gateway at all.
To tell a dead amplifier from a fixed loss somewhere after it, I commanded two different output powers, 13 dB apart, and measured both at the same receiver in the same spot: my other node, a short distance away. A healthy chain shows a 13 dB difference. A fixed loss downstream still shows 13 dB, just lower. A stuck amplifier shows much less.
| Commanded output | Median signal at my other node |
|---|---|
| 9 dBm (low) | -64 dBm |
| 22 dBm (full) | -60 dBm |
Thirteen decibels commanded, about four delivered. My own manual reading of the first round agreed, at about two. The low setting landed right where the station's healthy numbers predicted. The full setting came out roughly 11 dB short. The radio worked at low drive and couldn't climb past it: the high-power amplifier path no longer engaged.
Before the failure, the outside gateways that heard my station were getting it at -93 to -120 dBm. That is right at the edge of what those receivers could decode, with no headroom to spare. Take 11 dB off a link with no margin and every one of them drops below the floor at the same moment. My node in the house started with about 30 dB of headroom, so it still heard the station, just more weakly.
The station had been working, but only just. That weakness was there before the amplifier failed.
My own dashboard misled me more than once. Twice during the swap, its counts made the new, healthy radio look broken: once from a stale list of connections, once from its own link dropping. A dashboard counts what reached the dashboard. To know what the radio is doing, ask the radio, and check transmit with a receiver you don't control.
A first pass also said the outside receivers' signal levels were unchanged across the failure. That was survivorship: after the step there was nothing to measure, so every reception in the sample predated it.
And none of this was fixable in software. Once the sweep came back, I chose a new node, and it went in the same day. By that evening, the gateways that used to hear the old station were hearing its replacement.
What it does: a repeatable check for a Meshtastic node that "can't be heard," which separates the antenna path from the radio before you take anything apart.
The parts: the node in question, a second node of yours a fixed distance away to use as a meter, and a community collector service that shows which gateways heard each packet.
How the pieces fit: read receive signal-to-noise from neighbors that don't move, over a week. Count, per day and then per packet, how often outside gateways heard you. Send deliberate test messages. Then command two output powers and read both at your meter node.
Time: the whole diagnosis fit inside one day.
If you have chased a bad antenna that turned out to be the radio, or the other way around, I would like to hear about it.