Switching Regulator for Audio Amplifier

[Eeyore]

Never.

Depends how carefully you listen I suppose.

D'ya know of any serious hi-fi equipment that uses them? There would
certainly seem to me market pressure in that direction, and I've seen
switching regulators and amplifiers that are targeted to very low-EMI
environments do quite well.

I think there's some moderately serious Denon stuff that does and doubtless
others but I'm more a pro-audio man, where you will find them in amps
especially designed for touring where weight counts.

Of course, when you get into audio you run smack into the "it sounds
better because it's dipped in LN2" crowd, so it can be hard to sort the
rational but subtle claims from the pure wackiness, and my golden ear
tops out at boom-boxes from Wall-Mart -- beyond that I'm just wasting my
money -- so I'm no judge.

The LN2 brigade should be either shot or prosecuted for fraud. Maybe both.

Graham

 

[MooseFET]

The higher is the switcher frequency, the smaller is the power supply
ripple rejection in the amplifier.

Yes but it rarely rises as fast as the filtering improves.

The higher is the switcher frequency, the higher are the losses.

Perhaps not if you stay "within reason" as I suggested.

The properly designed servo loop will have the cutoff frequency around
0.1 of the switcher frequency.

Yes give or take a little. It will certainly be a fraction of the
swithcing frequency.

The reasonable frequency for the medium or high power audio SMPS is
somewhere in the range of 30...40kHz.

That is too low IMO. 150KHz would be more reasonable.

The OP's 50W power requirement is just nothing.

It increases the size of heat sink the product needs.

So, there will be a huge power ripple and the surge current through that FET.

I did say "large" on the MOSFET. It will see the same sort of duty as
the rectifiers. It is only a 50W supply.

How about the 60Hz?

The magnetic amplifier will create a little 120Hz etc but these will
be nothing compared to what the rectifiers cause.

 

[MooseFET]

I must acknowledge this is probably the wisest solution to using this
chip: Just settle for 40 watts.

This is what the audio engineers at National said. They also said
getting 50 watts out of the chip "was difficutl" without regulating
the supply, and that goes against all their engineering effort to make
the device have as high ripple rejection as it does.

They were aghast when I told them I put 50,000uF on each rail. And
you know what...

It actually sounds better without all that reserve capacity

So basically, I guess what I want is simply a more powerful amplifier.

.... or ...

If you use more of the 40W amplifiers you can get more power than a
50W amplifier that has a regulator in it for about the same
complexity.

 

[MooseFET]

Ah, that's another technique some claimed to have mastered although I see no
product in the market.


You'd have slower looop response due to the large C but I guess you're saying
it wouldn't matter because the C makes up for it ?

You don't really end up with a much slower loop. As you make the
output filters larger the amount of noise and ripple etc decreases.
This means you can apply a large proportional gain to the error in the
output voltage without troubles from the ripple. If your switcher's
core section looks like a transconductance, the loop stability remains
constant if you increase the capacitor and the gain proportionally.

Another way to think about it is to make the output capacitor just
short of infinite. You then can charge it up on Monday and play the
music on Tuesday.

 

[Phil Allison]

"Fred"

I have been looking at thoses 1kW rack mount PA amplifiers, and
wondering how in the world do they get that power in such a small
place. I know SMPSs have something to do with it.

** In nearly all examples, the switching PSUs used are non-regulated - so
the DC output voltage varies with the AC supply voltage and has maybe 10%
100/120 Hz ripple voltage under load.

I have also seen the terms maginetic amplification

** No such thing in audio.

You may be thinking of Carver " Magnetic Field " amplifiers, which has got
to be the most misleading title ever printed on an audio power amplifier.
These amps used conventional, laminated iron transformers - just rather
small ones with a TRIAC pre-regulator in the AC supply.

and digital amplifaction

** Another marketing department invented misnomer that refers to PWM - a
purely analogue technique.



...... Phil

 

[Bob Eld]

Designing a discrete amp to beat the LM3886 needs the kind of skills and
experience I have. It's NOT a hobby task.

The typical 'gainclone' circuit the OP is likely usung has PUNY reservoir caps.
Around 1500uF IIRC. Make them 4700uF and use a larger transformer in
'under-rated mode' to get better supply regulation.



Since the LM3886 will sustain +/-47V (obviously allow for AC line overvoltage
on top of this) where's his problem ?
http://www.national.com/mpf/LM/LM3886.html

Graham

I agree. I didn't mean that he should necessarily design the amp, maybe copy
one. But switching regulators are no simple hobby task either. Yes, looking
at the 94 volt range of the 3886 does beg the question, where is his
problem? I foolishly assumed he had one as he claimed.

My comments about the necessity of regulation still stand, however. There is
no reason what so ever that these amps will not work perfectly well with a
peak rectified, capacitor input voltage of +/- 30 volts under full load.
That would be over 50 watts per channel into 8 ohms. No regulators required
and a 17 volt over voltage margin on the raw voltage, each side.

 

[Phil Allison]

"Bob Eld"
"Eeysore"

overvoltage> on top of this) where's his problem ?

I agree.

** Foolish move to agree with any of the utter rubbish posted by the
Stevenson TROLL.

The fallacy of arguing from a position of 100% ignorance is his only talent.

I didn't mean that he should necessarily design the amp, maybe copy
one. But switching regulators are no simple hobby task either. Yes,
looking
at the 94 volt range of the 3886 does beg the question, where is his
problem?

** Read the specs more closely - the max safe supply voltage is actually
84 volts.

The 94 volt ( no drive) figure is meaningless for most applications.

I foolishly assumed he had one as he claimed.

** He does: " I have an audio amp based on the LM3886... "

My comments about the necessity of regulation still stand, however. There
is
no reason what so ever that these amps will not work perfectly well with a
peak rectified, capacitor input voltage of +/- 30 volts under full load.
That would be over 50 watts per channel into 8 ohms.

** Bad math.

The stated DC supply for 50 watt output at 8 ohms is +/- 35 volts minimum -
with no allowance for ripple voltage minima. This for 0.1% THD or just on
the clipping point.

No regulators required
and a 17 volt over voltage margin on the raw voltage, each side.

** Using the REAL figures paints a different picture:

The DC supply must not exceed +/- 42 volts under no load and highest
expected AC voltage - or the LM3886 can be expected to fail.

They are in fact NOTORIOUS for doing this !!!!

Makes the nominal, no load supply +/- 40 volts DC.

To get +/- 35 volts under full load, including ripple minima, means the
transformer must have better than 10 %regulation with a rectifier/capacitor
load combined with a ripple voltage on the DC of less than 2 volts p-p.

For a stereo pair of LM3886s, this equates to using a 400 - 500 VA tranny
with 10,000 uF caps on each rail.

Opting for 40 watts per IC is far more sensible.


...... Phil

 

[Eeyore]

Yes but it rarely rises as fast as the filtering improves.


Perhaps not if you stay "within reason" as I suggested.

Vlad's suggestions of 30-40 kHz are 2 decades old. I felt a bit old fashioned at
120kHz. Plenty of ferrites go to 500 or more kHz now.

Graham

 

[Eeyore]

You don't really end up with a much slower loop. As you make the
output filters larger the amount of noise and ripple etc decreases.
This means you can apply a large proportional gain to the error in the
output voltage without troubles from the ripple. If your switcher's
core section looks like a transconductance, the loop stability remains
constant if you increase the capacitor and the gain proportionally.

Another way to think about it is to make the output capacitor just
short of infinite. You then can charge it up on Monday and play the
music on Tuesday.

I see what you're getting at. I don't recall having seen that technique used which
is interesting.

The other side of the equation is that audio power amps don't need rock solid
regulated rails, in fact some people claim they sound more 'dynamic' when the
regulation is quite loose.

Graham

 

[Eeyore]

"Fred"


** In nearly all examples, the switching PSUs used are non-regulated - so
the DC output voltage varies with the AC supply voltage and has maybe 10%
100/120 Hz ripple voltage under load.

Absolutely true. Although of course you get, size for size of electrolytic
less ripple when it's stored at 320V due to the V2 term in energy storage. The
ripple is coming from the primary side.

Graham

 

[Eeyore]

I agree. I didn't mean that he should necessarily design the amp, maybe copy
one.

There are VERY few good designs out there that'll beat an LM3886 and tons of
crappy ones. Doug Self's designs are probably the best bet readily available
(use Google). It's a whole new ball game using discretes though. Think layout !

But switching regulators are no simple hobby task either. Yes, looking
at the 94 volt range of the 3886 does beg the question, where is his
problem? I foolishly assumed he had one as he claimed.

I don't know the tolerances on your supplies ( I assume it's the USA) so design
for 94V with a 126V supply or whatever is typical high line.

My comments about the necessity of regulation still stand, however. There is
no reason what so ever that these amps will not work perfectly well with a
peak rectified, capacitor input voltage of +/- 30 volts under full load.
That would be over 50 watts per channel into 8 ohms. No regulators required
and a 17 volt over voltage margin on the raw voltage, each side.

This suggests he is indeed using the popular 'gainclone' design which as I said
has PSU caps that are absurdly small. Up them to 4700/6800 uF and the rails
won't drop the way they clearly are in his case. Plus the transformer issue.
That's why I have transformers very custom wound for audio amps.

Graham

 

[Eeyore]

"Bob Eld"



** Read the specs more closely - the max safe supply voltage is actually
84 volts.

The 94 volt ( no drive) figure is meaningless for most applications.

You don't understand semiconductors in that case. Either they go into avalanche
? failure or they don't.

It's either 84V or 94V, maybe a misprint somewhere. Ridiculous data sheet. I
have NEVER heard of no signal / with signal Max voltage difference. That's NUTS
! It makes no sense whatever. I'd want to have words with the Nat Semi app guys
over that.


Better still use an LM4780. It's even marginally better.
http://www.national.com/mpf/LM/LM4780.html

Graham

 

[Fred]

Well, mine does a bit more, never had 'regulator problems' as it has none:
http://panteltje.com/panteltje/amplifier/index.html
Of course your speaker's impedance determines how many watts you can deliver at a given supply voltage.

The advantage of a switch mode supply that I see is
1) less weight (mains transformers are heavy),
and
2) perhaps easily adapted to both 230 V / 50 Hz and European 110V / 60 Hz US supply.

If the amp has good supply rejection (it should), then a simple rectifier would be enough.

I used the LM3886 it has better than excellent rejection. It was
designed to run off an unregulated supply. the problem came in trying
to get 50 watts into 4 ohms. it requires slightly more then +/-24V at
5A. My 24 volt transformers would rise to +/-42 during no-load
conditions. That is the absolute maximum limit of the chip. I
destroyed two chips, and they have all sorts of built in protections,
driving them to full output with those transformers.

By the way, Nation recommended to avoid the use of a choke in the
power supply but omitted elaboration about why. Presumably, it has
negative affects upon the device.

Transformers I used are old Signal units, probably 15% regulation. So
if if got a larger, perhaps toroidal, transformer with 5%-6%
regulation, I might be able to slip underneath the voltage limit.

The device you used looks interesting, and you got 75W

Fred

 

[Fred]

Designing a discrete amp to beat the LM3886 needs the kind of skills and
experience I have. It's NOT a hobby task.

The typical 'gainclone' circuit the OP is likely usung has PUNY reservoir caps.
Around 1500uF IIRC. Make them 4700uF and use a larger transformer in
'under-rated mode' to get better supply regulation.


Since the LM3886 will sustain +/-47V (obviously allow for AC line overvoltage
on top of this) where's his problem ?http://www.national.com/mpf/LM/LM3886.html

Graham

I think you mean +/-42V. And that is exactly what my PS was supplying
at no-load. It was under heavy load that two of the devices failed.
Nation said the only way to destroy the chips was wtih excessive
supply voltage combined with low load resistance.

Fred

 

[Fred]

"Bob Eld"
"Eeysore"





** Foolish move to agree with any of the utter rubbish posted by the
Stevenson TROLL.

The fallacy of arguing from a position of 100% ignorance is his only talent.


** Read the specs more closely - the max safe supply voltage is actually
84 volts.

The 94 volt ( no drive) figure is meaningless for most applications.


** He does: " I have an audio amp based on the LM3886... "


** Bad math.

The stated DC supply for 50 watt output at 8 ohms is +/- 35 volts minimum -
with no allowance for ripple voltage minima. This for 0.1% THD or just on
the clipping point.


** Using the REAL figures paints a different picture:

The DC supply must not exceed +/- 42 volts under no load and highest
expected AC voltage - or the LM3886 can be expected to fail.

They are in fact NOTORIOUS for doing this !!!!

Makes the nominal, no load supply +/- 40 volts DC.

To get +/- 35 volts under full load, including ripple minima, means the
transformer must have better than 10 %regulation with a rectifier/capacitor
load combined with a ripple voltage on the DC of less than 2 volts p-p.

For a stereo pair of LM3886s, this equates to using a 400 - 500 VA tranny
with 10,000 uF caps on each rail.

Opting for 40 watts per IC is far more sensible.

..... Phil

Hey Phil,

The old signal Transformer I used have a stated regulation of 10%, but
the actual PS rose more like 14%.

Thanks,

Fred

 

[Fred]

Vlad's suggestions of 30-40 kHz are 2 decades old. I felt a bit old fashioned at
120kHz. Plenty of ferrites go to 500 or more kHz now.

Graham

The LM5116 has a reference design for 5V @ 5A that runs at 250kHz and
over 90% efficiency at 5A.

Fred

 

[Jan Panteltje]

I used the LM3886 it has better than excellent rejection. It was
designed to run off an unregulated supply. the problem came in trying
to get 50 watts into 4 ohms. it requires slightly more then +/-24V at
5A. My 24 volt transformers would rise to +/-42 during no-load
conditions.

Thats transformer is really really bad.
I do admit I use a 200W toroid (once I wanted to make a stereo amp),
but, for a fraction of what this home build did cost me, I also bought a 2 x 180 W in 4 Ohm
19inch stereo amp ...... You cannot beat German design manufactured by a Chinese company
where they work for peanuts, or maybe M&Ms.
Lookup PA3000 at conrad.nl...

I work the other way around,
If 42V is max, then subtract 10% for mains variation makes 37.8V DC nominal peak.
This is 'top of the ripple'.

You drop 1.4V in a bridge, so 39.2V peak out of the transformer.

At 50 W into 8 Ohm you need a voltage swing of P = U^2 / R so
U^2 = 50 * 8 = 400, that makes U = 20V eff.
The peak would be 20 x sqrt(2) = 28.28 V.

How much does the chip drop of this? I dunno. but say 4V (just for the example).
So that makes 28.28 + 4 = 32.28.

We have 37.8 - 32.28 = 5.52V for the ripple.
At 50 W the peak current into 8 Ohm is 28.8 / 8 = 2.54 A
At 50 Hz mains, and a full bridge you have 10 mS between mains tops,
Q = C . I = U . t
So 5.52 V at 2.54 A in 10 mS makes C = (5.52 x .01) / 2.54 = .022 F or 22 mF or
22 000 uF for the filter caps.

Not counting any internal resistance of the transformer.
I'd say: You cannot even do 50 W into 8 Ohm with a 'perfect' transformer with that chip.

Maybe I messed up the math....

Anyways, in audio the average loading is much less (unless you want to play continuous sine waves),
and 30% power rating would be safe perhaps for the transformer for music use.
Then you run into the larger internal resistance (thinner wire) of that transformer again...
And somebody WILL do sinewaves for hours on end (I know somebody that did, but IIRC they
told him to move his workshop after that).

I see my Chinese amp simply uses a _huge_ torroid, 250W? ...... and normal transistors,
a temp controlled fan, stereo, all in a nice 19 inch rack, with separate volume controls
for left and right, with clipping warning indicator LEDs, with power switch, cinch AND XLR
connectors on the back....
And I payed 65 Euro (about 91 dollars) for it..... some years ago, see they are now 87.95 Euro...
with more LEDs
Lookup 'PA3000' at www.conrad.nl
















That is the absolute maximum limit of the chip. I

 

[Eeyore]

By the way, Nation recommended to avoid the use of a choke in the
power supply but omitted elaboration about why.

Modern transistor designs never have CLC or LC filters. Partly because of the impracticality of the choke
size I reckon.

Graham

 

[Eeyore]

I think you mean +/-42V. And that is exactly what my PS was supplying
at no-load. It was under heavy load that two of the devices failed.
Nation said the only way to destroy the chips was wtih excessive
supply voltage combined with low load resistance.

How low ? Might make a difference ?

Sounds like there's a **** up on that chip though in that case.

As mentioned before by Speff IIRC and myself the difference between 40 and 50 W is
barely audible though (about 1 dB). If it was 40W vs 100W then yes, I'd be bothered.

Graham

 

[Eeyore]

The old signal Transformer I used have a stated regulation of 10%, but
the actual PS rose more like 14%.

Presumably the ripple adding to it.

Use an oversize transformer with 5% regulation and maybe 10,000 uf caps and you'll
get that 50W I bet.

Graham