[time-nuts] Anybody want a Thunderbolt power supply?
csteinmetz at yandex.com
Wed Aug 31 04:48:41 EDT 2016
> I’m going to shoot for <= 35 mV P-P. If you need better than that, then it probably turns into a hybrid switching+linear system
It's been years since I played with Tbolts (as opposed to just using
them), but as I recall there was no possibility whatever of getting the
best results from any switching supply without at least one stage of
linear "polishing." In fact, everything I tried with initial switching
regulation benefitted from two levels of linear polishing. It was my
conclusion that starting with a switching supply of any sort is a bad idea.
I found that for best performance the -12v supply, which feeds the DAC,
needs to be about 1000 times better than what you are shooting for
(i.e., low tens of uVp-p). That surprised me, because I assumed there
must be some internal regulation of the DAC supplies -- but it was what
my testing showed.
Sure, a Tbolt will work with dirtier power than that, but for real
time-nuts performance I consider an all-linear supply with an
ultra-low-noise -12v rail to be absolutely necessary. Since that is not
very hard to provide, I see no reason to try to make do with less.
One low-cost solution is to wait for a cheap Power One HTAA-16W-A
triple-output power supply to show up on ebay, and do a very minor
modification to raise the +12v current limit (so it will handle the
first moments of a cold start without current limiting) as detailed here:
That approach achieves excellent results -- close to the best you can do
-- with very little fiddling. If one wants the very best performance,
the -12v supply can be adjusted down to -16v or so to provide the
headroom to add a ULN regulator or a capacitance multiplier. (The Tbolt
only draws a few mA from the -12v supply, so the ULN regulator can be a
suitable op amp with no external pass transistors.)
Last time I bought them, I paid $25 for a lot of three HTAA-16W-A's.
More information about the time-nuts