piclist 2002\11\28\032512a >
www.piclist.com/techref/index.htm?key=15v+psu
BY : Russell McMahon email (remove spam text)

> I'm assuming AC mains in, so I need a step-down, etc yes.
> I don't need variable but I also will be using 12V, -12V .1A regs on it
> as well, so I prefer the 18V, -18V transformer Chris suggested.
> I would prefer to keep noise down to <5 mV.
> Load is varying anything from 3ma to 3A.
> Efficiency isn't really important but I'd like to get away with smaller
> heat sinks, but oh well if not.  :)

Chris's suggestion is typical of conventional solutions.
If you are getting a transformer custom wound (can be as cheap sometimes
depending on what's available) you could consider getting as close to
optimum as possible at full load. But 18-0-18 OK.
That peak rectifies to say 18*1.414 = 25 volts say or somewhere over 20 on
load. So at 3A per side you'll dissipate 3A x (20-15) = 15 watts per
regulator. Add another 1A x (20-12) = 8w for the each of the 1A supplies. So
if these are all on the same heatsink you have about 50 watts of dissipation
and 110 watts odd of load power. That's a moderately large heatsink at
reasonable temperatures. Depending on ripple and dropout of the regulators
you can probably halve that with a lower DC voltage. One way which is a
little brute force is to have a multitapped mains winding. A 15-0-15
transformer will give you the better part of 20 volts DC and may be enough
for this application depending again on ripple (which depends on filter
caps). .

5mV noise is much easier with a linear regulator than a switcher. One trick
I have seen used which is quite effective is to use an SCR pre-regulator-
two of the 4 diodes in the main rectifier are SCRs and these are phase angle
controlled to keep the DC bus level as low as reasonably possible. Not  a
totally noise free solution but worth a thought.

Anything used as a linear regulator will dissipate the same power, whether
hexfet or IC or bipolar transistor. A switcher(which the SCR pre-regulator
is arguable a version of) avoids this by duty cycle modulation the input in
some manner.

Russell McMahon

--