ADMEM Power Supply Noise Tests

Noise Measurement Methodology
- The problem: The ADMEM/Café card response to noisy power supplies represents a largely unknown dynamic load due to regulators, capacitances of traces and planes, inductances like ferrite beads, etc.
- The solution (Hao Wei): Use the time resolution of the VME system itself to determine the power supply noise response of the ADMEM. This allows use to measure up to 1/(132 ns)=7.6 MHz noise components.
- Use RMS(Ped(t1) - Ped(t2)) where t1, t2 are seperated by N x 132 ns, and plot this RMS(ped) vs. Time. This provides a "pantogram" of the actual noise distribution as seen by the Café card: if the time interval is in the same as a major frequency component, the RMS is small; if it's a half wavelength the RMS is large.
- Have to avoid other known problems with the ADMEM: second half of card (channels 10-19) have appreciably larger sigmas, adding capacitive load increases sigma. These are being addressed in Charlie Nelson and Terri Shaw's next revision of the the ADMEM board and the Café cards. Our standard is a channel in the first half of the card using the input (as opposed to connecting to the DC current calibrator) with no cables attached to the transition card.
Noise tests using Wiener (switcher/linear regulator hybrid) modular power supply
- Pleasant surprise: Electronic loads in parallel with power lugs on backplane of VME crate added no detectable noise to ADMEM measurements at any load. Originally thought these used transistor switchers, but schematics show they are just transistors at less than unity gain.
- Could apply 100% load of central ADMEM crate on all power supplies except +15 V where we could only reach 80% @ 10 A
- Interesting surprise: pedestal RMS remained at 3 counts as long as no extra load was put on the -15 V power supply, even if all other supplies were at 100% (except for +15 V at 80%). Load on -15 V power supply is only 1.72 A at 100% (6 ADMEMs), but pedestal RMS rose to 10 counts. At 2.6 A, pedestal RMS rose to 15 counts, and (RMS difference) vs. delay time shows typical time structure.
- Interesting surprise #2: Looking at AC signal across -15 V power lugs on VME crate with all power supplies at 100% load (but +15), we see 240 mV P-P noise! Main frequency is about 300 kHz, presumably the switching frequency, but there are many frequency components as usual. Watching AC signal as electronic load is turned on and off, these is no discernable difference at any time scale.
- Charlie's plot shows sensitivity to noise in 100 kHz - 1 MHz range for even 1 mV noise at inputs to Café card. Presumably, these is much filtering on ADMEM before Café card inputs, but perhaps not enough on -15 V inputs?
Noise tests using Condor (linear) individual power supplies
- Could apply 100% load of central ADMEM crate on all power supplies
- Behavior of Condor linears was same as Wiener switchers, but slightly better. At 100% load on all power supplies (1.72 A on -15 V), pedestal RMS was 8 counts instead of 10.
- Also, tried swapping +15 and -15 V electronic loads just to make sure load itself was not source of noise. It was not.
Noise tests using Vicor (switcher/linear regulator hybrid) modular power supply (400 Hz MegaPac)
- Almost as good performance as linears without load.
- Works without problems under full load! Better performance under full load than any other switcher; close to linear performance.

Plans for immediate future

Add L, C filtering configuration alá Binkley to Vicor power supply and remeasure.
Measure PMI (Pioneer Magnetics) power supply prototype when it arrives.

Editorial: I really like the Vicor design. It's extremely compact (less than shoebox size for all 5 power supplies). Even combining with remote monitoring (which is not provided) and big filter caps and coils, it should still take up only 3 U of rack space per "analog" crate. It also runs very cool compared to other supplies we have looked at.

ADMEM Noise Tests with Electronic Loads / FNAL /hahn@fnal.gov