Transcript Modelling of TPM noise problems

Modelling of TPM noise
problems
Greg, following discussions and
measurements with David and
Senerath
Ground rules
(ha-ha)
• Try to understand the behaviour of our setups for high frequencies
>1MHz
– Model signal ground and all power planes as a single point in all circuits
• Should be a good approximation
– Similarly model the 48V supply and return lines as a single point
• Probably a less good approximation?
– Model “short circuit” wires as inductance
• Crudely: L= m0 (length) to within an order of magnitude and m0=10nH per
cm
• This corresponds to 1 per cm at 100 MHz
– Ignore unwanted stray capacitance for now
• Since 0 is 0.1pF per cm, giving impedances >k at 100MHz unless overlap
area is very large
• Look at observed effects in GCT modules, with and without
modification, and in various test setups
• Work out what to do to verify the model and design power supply
layout for production GCT
The unmodified design
•
•
Noise generators modelled as current sources
Noise voltage measurements observed between signal and chassis grounds
–
•
Common mode caps provide loop for the return of internal noise currents
–
–
–
•
3-5 Volt swing
Lower impedance than internal stray capacitance (or perhaps internal bypass caps)
If one set of caps is removed, currents can flow through earth connections at the back of the crate
Observed voltage swing implies currents are O(Amps)
External noise currents flow mainly through the caps at the input
–
Not seen on signal planes
Modelling (I)
• Here’s a model of the
circuit that has some
of the right behaviour
5V
– Most components can
be seen on the sketch
– Added resistors to
damp resonances
– Tuned component
values to match
observed behaviour
What does this do to the links?
•
The IM looks similar to the TPM
–
–
•
Signal planes are connected through
the drain wires
–
–
–
•
Same chassis ground
It doesn’t have the common mode caps
to chassis
This is correct according to National
design note
Required to provide low impedance
return path for common mode currents
Nevertheless: common mode rejection
is limited and if voltages are too large
the links will be unreliable
Connecting the IM significantly
reduces observed noise amplitudes on
TPM
–
–
Factors of order 2
Implies an alternative path for noise
currents with impedance comparable to
our common mode caps
•
•
Through the drain wires (eight per cable
assembly) to the IM signal ground
Back through the rack metalwork to the
supply input
First modification
•
Direct connection from supply to signal ground
–
–
Initially suggested as a capacitor
But L dominates impedance
•
•
Observed noise voltage reduces by 2-3
–
–
•
Perhaps 3-5 times lower than paths to the back of the crate
Impedance of direct connection is lower than common mode caps, but still comparable
Some further improvements with “fatter” connection (Cu tape)
External noise now has a lower impedance path through the signal planes
–
–
Not observed in this configuration, internally generated noise dominates
But we didn’t look for the external noise
Modelling (II)
• Adding a small inductor to
the model reduces the
size of noise pulses
1V
– Note change of vertical
scale on the plot
Second modification
• Remove common mode caps between supply and chassis
• Internal noise voltage reduces by a further factor 2-3
– Return path through direct connection is now much lower impedance
than loop through chassis ground
• Noise voltage seen on scope now dominated by external noise
– This has got larger than before
– The path through signal planes is still there, now with higher impedance
External noise source?
•
Adding filtering to the 48V supply has little effect
–
–
•
Noise from this source is already quite low
Our input filtering is doing its job here
The source of the external noise turns out to be the 5V crate supply on the VME
backplane
–
Injects noise directly into the signal ground with no filtering
•
–
•
Design error; the CM has a filter
The common mode caps were doing a job here
Not so sure what the loop path is, perhaps something like this
Modelling (III)
200mV
• Removing the caps
reduces the noise
amplitude further
• The unwanted
currents now flow
through stray
capacitance to the
supply input
Conclusion
•
We have reduced the noise amplitude in the existing TPMs by more than a
factor 10
– Provide low impedance path between input and output supply
• A solid plane is needed
– Break loops through the “chassis ground”
• Input filters should achieve this as long as there is no path to chassis between the filter
and the DC-DC converter
• Do we understand what the path is in the modified TPMs?
– The second problem comes from the common mode caps
• Or rather: would be much less obvious without them
• Perhaps if less obvious it would just have come to bite us later
– It is particularly bad because one supply has a filter and the others don’t
• So noise generated by the 3V3 converter hits the filter and meets a high impedance in
the loop through chassis
• But the common mode caps act on all supply planes (since they are capacitatively
coupled together)
• And the noise generated by the other converters doesn’t meet any large series
impedance
•
Modelling reproduces qualitative features of the observed behaviour