latency-test

hi, since I read that the servo closed loop with bare bridge work better with higher thread servo I wanted to do some tests.
the nuc i3 with 7i80 db 16 kernel rtprempt to 2 khz --see photo
and intel atom n330 with 5i23 kernel rtai 8khz --- see photo
as far as I understand, jitter must not exceed 10% of the signal ... which is the safest combination?
for now I've only tried the standard 7i80 at 1 khz in motion.
regards
mauro

A couple of things here

1. The latency test is quite optimistic because doesn't actually do anything,
it really just tests dispatch latency, but since it does no I/O, many additional causes
of latency in real systems are not shown.

2. Ethernet latency will be quite a bit higher than indicated by the latency test. One way to check
overall servo thread latency (including all I/O and all LinuxCNC component times)
is by monitoring the servo-thread.tmax parameter (this parameter is in CPU clocks)

3. The main bad effects of varying latency on servo systems (and step/dir systems where the
stepgenerator is basically a velocity mode servo) is that the sampling of the encoder
position (or stepgen position) will have random errors because linuxCNC assumes they were
sampled at precisely the servo thread interval.

4. Problem 3 can be mostly eliminated by using the DPLL module present in most Ethernet FPGA configurations
the DPLL acts as a "jitter filter" and re-times various operations (like encoder sampling and stepgen position sampling)
and removes most of the sample time jitter (down to less than .5 usec in most cases)

By using the DPLL on Ethernet cards you can normally tolerate latency of about 1/2 of the servo period
(peak latency measured via servo-thread.tmax, _not_ the latency test)

thanks Peter I understood.
as I have both the 7i80 and the 5i23 I was wondering if it is appropriate to use the 5i23 to 2 or more khz for the closed loop with the 7i40. if there is a real benefit.
the 7i80 I would use it in the plasma I'm doing.
just to play because it already goes very well.
regards
Mauro

If you need a high speed servo thread then the 5I23 would normally be better,
but Ethernet is generally better for plasma since it has better ground loop noise immunity

In general with PCI cards and LinuxCNC possible servo rates are 1 KHz to maybe 8-10 KHz
For Ethernet, servo thread rates are more like 500 Hz to 4 KHz (depending on host)