Mesa Ethernet, dropped packets and potential noise issues

Hello,

I have a setup with a Mesa 7i94, 3 smart serial remotes, and 3 servo motors (2x 750W + 1x 400W).

During testing I noticed occasional weird errors (like joint following errors during homing/jogging) and hostmot read times spiking to 400-800us (servo thread period is 1ms).
I started suspecting dropped packets due to EMI noise from the servo drives.  By adding a debug log in hm2_eth.c: record_soft_error() I could monitor dropped packets in the console.

I noticed that as soon as the 750W servo drive were enabled, packet errors started happening. Sometimes one every couple of seconds, sometimes many per seconds, but never enough to trigger a hard error. 

I tried replacing the Ethernet cable with another one (both were shielded) going straight from the PC to the Mesa card, keeping it away from the servo drives/cables and I think it reduced somewhat the frequency of the packet errors but they still happened.

Now it gets weird: once I manually jogged a tiny bit the 750W motors, no more packet errors occured!
And I can reproduce it every time: enable the servo drives: packet errors. Jog the motors: no more errors!

Also I noticed that I never got any packet loss on any of the 3 SmartSerial remotes, and their cables aren't even shielded.

What is going on?! Please help!
 

That's weird, it's like the drives are in some different mode before the first motion

I would look at adding common mode chokes (a giant ferrite bead over U,V,W)
on the motor leads and perhaps a line filter on the driver power source.

Thanks for your reply.

Yes it is weird... I did not expect the Ethernet link to be the weak link...
Also all my motor cables are shielded and terminated at both ends.

I don't have any chokes on hand but I'll try adding some line filters.

Yeah, the problem with motor wiring is you have ~350V square waves (for 220V drives)
with microsecond or sub microsecond risetimes driving into cabling/winding
capacitance. Unless you have something to limit the peak current, like a
common mode choke at the drive, you can easily induce peak ground currents
of many amps at MHz frequencies. At these frequencies all ground wiring is an autotransformer...