wvarty
commented
3 months ago Test 1 (comparison data point):
- Ranger TX flashed with
3.x.x-maintenance - Connected to the PC via USB
- Stats measured in MP for a couple of mins
- Around 1% packets dropped vs received
Test 2:
- Ranger TX flashed with
3.x.x-maintenance - Backpack flashed with the PR above
- Backpack connected via home wifi, to my netgear router, which my PC is also connected to via ethernet
- UDP connected in MP on the default port
- Stats measured in MP for a couple of mins
- Around 0.5% packets dropped vs received
Test 3:
- Ranger TX flashed with
3.x.x-maintenance(also triedmaster) - Backpack flashed with
master - Retesting master on backpack, I expected to see the failure case, where loss is ~20% to 50%
- Instead, I have 100% in MP.
Not sure my test results actually mean anything now, as I cannot reproduce the "broken" state 😐 Only thing Ive changed between "broken" master and "working" master is doing a full chip erase on the ESP32.
I'm going to approve this PR anyway, as it doesnt seem to do any harm from what Ive seen in my testing, and from reports from other users in discord, this did help.
hyp0dermik-code
For unknown reasons, it seems that some PC / WiFi card / Backpack combinations struggle with high packet rate broadcasts and drop 20-50% of packets sent this way. This change instead unicasts MAV-to-GCS packets to the IP address from which the last MAVLink GCS-to-MAV packet was received.
Downsides: Can't connect multiple GCSes (but this would have been bad anyway) Upsides: 100% quality.