[time-nuts] need power trans. for HP3325
Didier Juges
didier at cox.net
Mon Mar 26 00:53:34 EDT 2007
Dr Bruce Griffiths wrote:
> Didier
>
> SCR's are still used in very high power (Megawatt) high voltage (0.5
> Megavolts or more) inverters where they replaced grid controlled mercury
> arc devices (some of these have been in operation for 40 years or more).
>
> I am sceptical that any affordable switcher has an output noise as low
> as a well designed linear regulator (<10uV pp DC -20MHz).
>
> Bruce
>
They are not, but system designers, as far as I can tell, have overcome
this problem by providing local filtering/decoupling/regulation at the
most sensitive points and therefore do not require the entire power
budget to a system to be low noise.
Most of our power supplies (we only make custom military supplies) have
many outputs (one we make right now has 34 outputs, all independently
regulated with floating grounds), sometimes several outputs of the same
voltage, so that the customer can isolate various sections of the
system, and provide as much filtering at the point of load as needed for
each subsystem. Also, being a 100% military house, few customers call
our supplies affordable :-)
For instance, on the 5370, a lot of the power is used to drive logic
circuits, TTL and ECL. For the most parts, these are not very sensitive
to noise, certainly not as much as the OCXO, the VCOs and the input
comparators, which probably (I have not checked the manual yet) run from
a different voltage.
Modern high integration chips have advanced power/thermal management
functions which turn off parts of the chip when not in use. This results
in very severe load transients. Often, the noise (periodic and random
deviation, PARD) on the 5V, 3V and lower rails is driven by the very
fast load transients of the logic chips more than by the switcher
itself. A specification we are reviewing right now has a load change
requirement from 55 to 110A on a 3V rail in less than 1uS at the end of
a 56 inches bus bar. With that kind of load transient, the ripple does
not matter much compared to the transient response through the bus bar,
even assuming a perfect power supply with infinite output capacitance.
Didier
More information about the time-nuts
mailing list