Mystery Errors
Posted by
Mariss Freimanis
on 2004-01-13 19:31:30 UTC
Hi,
This problem (and successful solution) has come up 3 times in the
past 2 weeks during technical support calls so I thought I would post
it here.
In all cases the complaint was axies not positioning accurately; in
one case an axis that was supposed to be stopped was creeping while
another axis was in motion.
In all cases the cable that carried step and direction signals was
long (over 6' to 20') and was a randomly oriented multi-conductor
cable.
The problem was the cable, or more properly, cross talk.
Cross talk is where two wires are adjacent two each other for a
considerable distance and the signal on one wire injects itself into
the adjacent wire. This results in step signals being injected into
other axis step and direction lines, causing all sorts of mysterious
and irritating errors.
The solution is to use twisted-pair cables or ribbon cables.
If twisted-pair cable (like CAT-5) is used, send the signal down 1
wire and connect the other wire of the pair to ground at the PC end.
Do not terminate the drive end of the ground wire. Never send two
signals down the same pair.
If ribbon cable is used, have every alternate wire grounded at the PC
end. Never send signals down adjacent wires.
My guess is the guy that posted message 69264 is also having this
same problem.
-----------------------------
Techical stuff: The problem is caused by capacitive coupling. If you
figure on 50pF per foot, a 20' foot cable could have 1,000pF of
parasitic capacitance between two adjacent wires. The copper in the
wires form the plates of a capacitor, the wire insulation forms the
dielectric. Instant capacitor, right where you don't want to have one.
Digital logic output impedance is assymetrical. It is much higher for
a logic "1" as it is for a logic "0". If you have a stopped axis step
line sitting at a logic "1", an adjacent active axis step line is
perfectly happy to inject a cross talk glitch into it, particularly
on the "srong" 1 to 0 edge. The shape of the injected pulse can be
modelled as an RC differentiated pulse. If the logic "1" impedence is
1K then the RC time constant may be 1uS or more for a 20' cable. This
width pulse is more than enough to cause a spurious step on what is
supposed to be a stopped drive.
Grounding one wire of a twisted pair forms a very effective
capacitive voltage divider (>10:1). The injected signal will now have
a much smaller amplitude, small enough to where it can't cross the
logic threshold of the victim drive. End of problem, end of mystery
steps.
Mariss
This problem (and successful solution) has come up 3 times in the
past 2 weeks during technical support calls so I thought I would post
it here.
In all cases the complaint was axies not positioning accurately; in
one case an axis that was supposed to be stopped was creeping while
another axis was in motion.
In all cases the cable that carried step and direction signals was
long (over 6' to 20') and was a randomly oriented multi-conductor
cable.
The problem was the cable, or more properly, cross talk.
Cross talk is where two wires are adjacent two each other for a
considerable distance and the signal on one wire injects itself into
the adjacent wire. This results in step signals being injected into
other axis step and direction lines, causing all sorts of mysterious
and irritating errors.
The solution is to use twisted-pair cables or ribbon cables.
If twisted-pair cable (like CAT-5) is used, send the signal down 1
wire and connect the other wire of the pair to ground at the PC end.
Do not terminate the drive end of the ground wire. Never send two
signals down the same pair.
If ribbon cable is used, have every alternate wire grounded at the PC
end. Never send signals down adjacent wires.
My guess is the guy that posted message 69264 is also having this
same problem.
-----------------------------
Techical stuff: The problem is caused by capacitive coupling. If you
figure on 50pF per foot, a 20' foot cable could have 1,000pF of
parasitic capacitance between two adjacent wires. The copper in the
wires form the plates of a capacitor, the wire insulation forms the
dielectric. Instant capacitor, right where you don't want to have one.
Digital logic output impedance is assymetrical. It is much higher for
a logic "1" as it is for a logic "0". If you have a stopped axis step
line sitting at a logic "1", an adjacent active axis step line is
perfectly happy to inject a cross talk glitch into it, particularly
on the "srong" 1 to 0 edge. The shape of the injected pulse can be
modelled as an RC differentiated pulse. If the logic "1" impedence is
1K then the RC time constant may be 1uS or more for a 20' cable. This
width pulse is more than enough to cause a spurious step on what is
supposed to be a stopped drive.
Grounding one wire of a twisted pair forms a very effective
capacitive voltage divider (>10:1). The injected signal will now have
a much smaller amplitude, small enough to where it can't cross the
logic threshold of the victim drive. End of problem, end of mystery
steps.
Mariss