PSU for precision audio opamps

[Dirk Bruere at NeoPax]

I need +/- 12V from a 5V or 12V supply (PC PSU) to drive precision
audio stuff (LM4562).
Thinking of using a DC-DC converter eg XP Power or C&D.
The regulation on these isn't brillant at around 10%, but that
shouldn't be a problem.
Ripple is typically 0.5%
Any advice?

Dirk

 

[Dirk Bruere at NeoPax]

Umm, what does "precision audio" mean?

S/N > 120dB

Dirk

 

[Dirk Bruere at NeoPax]

You better have stable power supply and filter to a high degree,
as I do not see power supply rejection in the data sheet,
but'average input offset drift vs power supply oltage is 110dB' worst case.
That is referenced to input, so any supply voltage will appear 100dB
attenuated at the input if I understand this right.
So from that you can calculate some thing about the supply ripple / fluctuation.

Forget about monster cables, just build the whole thing in a Faraday cage ;-)

Well, the only PSU I have is that of a PC.
So how do audio cards like the Auzentech that boast 120dB s/n ratios
handle their supplies?

Dirk

 

[Joerg]

Op amps have really amazing supply rejection around DC, and much less
good supply rejection up near switching power supply frequencies.

If you use a capacitance multiplier (basically a one- or two-stage RC
lowpass with a Darlington emitter follower on it) you can get upwards of
100 dB rejection in the tens to hundreds of kilohertz, at the price of a
couple of diode drops on each supply--which you generally don't care
about unless you're really supply-constrained (a Darlington on a 3V
supply would be fairly painful).

Something like this (use a MPSA64 on the negative supply)




+15V 0--*------------- /--------------0 +13.5V
crappy | \ A clean
| ----------
| |
+-RRR--*--RRR----*
| |
C C
C C
C C
| |
gnd gnd

If you make R=100K and C = 1 uF, at high frequency your filter function
will be

H(f) = (1.6/f)**2 which is down by 60 dB at 1.6 kHz and 100 dB at 160 kHz.

This trick really really works. There's another approach, which is to
return the first capacitor to the emitter, thus making a Sallen-Key
lowpass, but that's inferior because the sneak path through the RC
network limits the attenuation at high frequencies.

Add a couple diodes. One from the +13.5V back to the +15V and another
from the base to +15V. Just in case the +15V goes down hard and
instantly for some reason, such as a short.

 

[MooseFET]

Umm, what does "precision audio" mean?


PSRR is bad on most all opamps, so the ripple may shoot through. And
the dc/dc guys often make spikes, which can get into the opamp
front-end. If you use a converter brick, it would be prudent to add LC
filtering to the input and the output.

Yes.
Also beware that the PSRR for the minus supply is often worse than for
the positive one. The lower of the two numbers is the one you really
need to worry about.

You should also be very careful about where the DC-DC converter's
input side ripple can flow to and the capacitively driven AC current
that is forced between its input side and output side.

DC-DC converters also radiate AC magnetic fields. The shield over the
top of them never fully stops it. Ideally the DC-DC should be on the
far side of some of the sheet metal.

 

[MooseFET]

You better have stable power supply and filter to a high degree,
as I do not see power supply rejection in the data sheet,
but'average input offset drift vs power supply oltage is 110dB' worst case.
That is referenced to input, so any supply voltage will appear 100dB
attenuated at the input if I understand this right.
So from that you can calculate some thing about the supply ripple / fluctuation.

Forget about monster cables, just build the whole thing in a Faraday cage ;-)

No, not a cage, a metal box perhaps with double walls.

 

[Eeyore]

I need +/- 12V from a 5V or 12V supply (PC PSU) to drive precision
audio stuff (LM4562).
Thinking of using a DC-DC converter eg XP Power or C&D.
The regulation on these isn't brillant at around 10%, but that
shouldn't be a problem.
Ripple is typically 0.5%

I wouldn't. I *always* use linear supplies for high performance audio,
even if that means linear post-regulating a switcher.

A really good audio rail should have perhaps as low as 30uV of audio
band noise, certainly no more than a few 100 uV.

Failure to do this all well will introduce noise current into the audio
ground(s) though any bulk decoupling used. The PSRR itself of the IC's
is unlikely to be the big problem, but polluting the ground is far
worse, because that's one half of the signal unless your design is fully
differential throughout.

Why 12V anyway ? For pro-gear I use 17V typically to get max headroom
although I note that the 4562 has Vs = +/- 17V anyway so 16V would make
sense there.

ALSO, just think of the EM radiation from the chokes / transformers in
those switchers that'll couple into every loop in sight ! That'll bugger
stuff up good and proper. Seen it happen.

Now having said all this it's not impossible to do it and get it right
but it's an R&D task in its own right.

Graham

 

[Eeyore]

Umm, what does "precision audio" mean?

Better than what your ears are probably used to ? ;~p

Certainly not TV, or AM / FM radio, boomboxes and the like.

Graham

 

[Eeyore]

The data sheet claims 120dB PSRR,

At what frequency ?

Graham

 

[Eeyore]

Monster power supply cables will certainly help, but the real improvement in
the precisionicity of the sound will come from painting the leads of the
opamp with yellow highlighter, hanging small pieces of wood in the inside of
the enclosure, and plugging a Tice digital alarm clock into a nearby ac
outlet.

All fascinating stuff, but reducing loop areas will do far, far more.

Graham

 

[Eeyore]

Check the graphs: -75 at 10 KHz, and getting worse.

At least somone else here has his head fairly well screwed on !

One could get potential heterodyning against power supply spikes, too,
maybe.

This probably doesn't matter for audio, but might affect important
stuff.

Why do you think it might not matter for audio ? What other human sense has
such a wide dynamic range ?

Graham

 

[Eeyore]

S/N > 120dB

So no resistor value above about ~ 470 ohms ? (sorry just guessing there,
I didn't fancy doing the sums).

You could quote them the 4562's THD figure. That'll blow 'em sideways.

Why the 4562 btw as opposed to the newer LME497x0 range ?

Graham

 

[Eeyore]

Well, the only PSU I have is that of a PC.

It has to go in a PC ? Is -12V still available on the PCI bus ? Or ANY negative
voltage ?

So how do audio cards like the Auzentech that boast 120dB s/n ratios
handle their supplies?

Interestingly perhaps. I could hazard a few guesses. I'd also suspect the use of
'weighted' s/n ratios. It's amazing what an 'A filter' can do to bad numbers. And I'll
bet the specs don't say.

If you seriously want this done well, I can (for some blue/brown/red folding) help you
a LOT.

Graham

 

[Eeyore]

Add a couple diodes. One from the +13.5V back to the +15V and another
from the base to +15V. Just in case the +15V goes down hard and
instantly for some reason, such as a short.

LOL !

Been there, done that.

It IS damn quiet too but I'd use a bigger device than an MPSA.

Graham

 

[Eeyore]

On a sunny day it happened Dirk Bruere at NeoPax wrote


Likely filter them,

You bet !

also there is +12 and -12 available in a PC, so you could add a LM317 in the +12
to a clean make +8, and something similar in the -12.
The PC supplies are junk, especially the 5 V is full of spikes in mine.

His biggest problem by the proverbial country mile will be a noisy (PC) ground with
associated leakage currents from the PSU's 'Y caps' that'll seriously screw up the
noise floor of most of the equipment it's likely to get connected too as some portion
of that nasty leakage current flows down the audio cables.

Yes, I know how to fix that too.

Graham

 

[Eeyore]

Good point. Hmm, what happens if you reverse a Darlington?

If it were a plain BJT, it would immediately be in normal bias as an
inverted transistor...but you'd zener the BE junction, which with a big
bypass on the output would turn it to lava. Check.

If the Darlington were really just two BJTs, then when it inverted the
driver stage would find itself connected from the base to the 'emitter'
of the inverted transistor...which should make it turn off fine, except
that you'd still zener the output stage. Lava again.

Assuming the MPSA14 has a built-in pulldown resistor on the base of the
output device, that would be trying very hard to turn on the inverted
output stage, and there'd still be the BE breakdown. More lava.

So okay, it needs some diodes. If one were to use 1N4005s or the
equivalent, though, their capacitance would be a problem--it's 15 pF,
typical, even at -4V, whereas this is only -1.5V. (It's the low Cce of
the transistor that makes this circuit work so well.)

To protect the CB junction, a diode from output to input would work, as
long as the output bypass cap was at least 10**6 times bigger than the
diode capacitance.

To protect the base, I'd probably put another smaller resistor in series
with the base lead to limit the fault current. If you wanted to use
diodes to protect the base, you could put one diode across each of the
resistors, so the sneak path would be heavily bypassed in the middle.

Yup, I did those things on my example. You just can't be too careful.

Come to that you should never forget the reverse diode across an 780x etc
either.

Graham

 

[Eeyore]

Hearing is limited by brownian motion

Tell me about it. I hear it all the time.

and internal noise sources; and
as a practical matter, by ambient noise, so it's about as good as it
needs to be. After a while in an anechoic chamber, you hear the blood
rushing through your head and neck. And the higher end of the dynamic
range isn't all that useful.

How loud have YOU gone ?

Easily 130dB here (at v. low THD).

Is there any other human sense that can cover that much range ?

Graham

 

[Eeyore]

Where would you get a signal that good? Or a quiet-enough place to
listen to it?

Do you stop breathing when you listen to stuff like that

If you need to ask ...... You've never lived.

Graham

 

[Eeyore]

You can get a serious amount of noise magnetically induced from fans,
AC or DC brushless. One has to worry about loop areas, yet another
interesting constraint in PCB layout.

Too damn right.

And yes, all the other stuff you mention too.

Graham

 

[Eeyore]

No, not a cage, a metal box perhaps with double walls.

Uhuh.

Faraday cage type thing. Very effective.

raham