Op-amp stacking

[Andi]

Wotcher gents (and ladies).

Having been short of day-job work the last couple of days I've been reading around a goodly number of FX sites. One thing that's cropped up a couple of times is the idea of stacking op-amps - either multiple chips or multiple stages in one chip. One of the claims that's oft made about this is a lowering of noise floor due to random cancellation, and that's what puzzles me.

Back when I did my EE degree, it was drummed into me that (a) information is the resolution of uncertainty, and (more relevantly) (b) noise is additive.

So who's right? Does stacking chips result in lower noise because (somehow) the noise turns out to be randomly more odd mode than even mode and thus likely to cancel out, or yer engineering standard that noise is, to put it colloquially, a git and does everything it can to ruin the party?

[Sir H C]

Noise will definitely go up.  This is not up for debate.

[R.G.]

Noise adds in an RMS sense. Always.

Paralleled amplifying devices may change the source/load impedances in a manner which is a better fit to the source/load impedances a circuit works in and thus may result in less audible expression of the noise energy. Source/load impedance matching is important for noise considerations.

If simply doubling the amplifying circuits in an opamp could reduce noise, all low noise opamps would already be double/quadruple/octuple/etc inside.

[Andi]

Hurrah for my not being entirely insane - ta muchly chaps.

[Ripthorn]

I can see how someone might think that the noise would be lowered, however.  If you have a Guassian distributed noise spectrum, then the phases are purely random and adding them together gives a statistical probability that phases would cause at least partial cancellation for at least some frequencies.  Whether this happens in chips, I don't know, but theoretically (meaning if the chip acts as a purely random noise generator) it could happen.  If RG says such is not the case, I believe him (I'm just a physicist, not an EE).

[Sir H C]

I can see how someone might think that the noise would be lowered, however.  If you have a Guassian distributed noise spectrum, then the phases are purely random and adding them together gives a statistical probability that phases would cause at least partial cancellation for at least some frequencies.  Whether this happens in chips, I don't know, but theoretically (meaning if the chip acts as a purely random noise generator) it could happen.  If RG says such is not the case, I believe him (I'm just a physicist, not an EE).

The math works that you do (IIRC) RMS addition, it will never on average reduce noise, as it is purely random, no correlation.

[composition4]

Grab an audio editor and generate say a 5 second sample of white noise (which is distributed evenly amongst all Fs).  Then generate another 5 second block of white noise at the same amplitude and mix it in with the first waveform.  The resultant waveform shows that noise does increase - if there were any significant phase cancellations that would bring the level down, the editor would show a decreased amplitude.

I think this displays that noise (no matter what type) added together will always increase average levels.  Noise is just too random over any significant period of time to have any chance of cancelling out another sample of noise.

Jonathan

[oskar]

I think this displays that noise (no matter what type) added together will always increase average levels.  Noise is just too random over any significant period of time to have any chance of cancelling out another sample of noise.

But this is true. Mix two sources of noise and the sum will increase. I think the point is that you have two signals mixed with noise mixed together. The two signals would suppress the noise just like in oversampling.
At least it's not in the realm of magic as the Hi-Fi folks who trim the edges of their CD's in order to increase sound quality.

But overall the noise will increase compared to not having an OP-amp stage...

[composition4]

Or the on where "audiophiles" run a green permanent marker around the edge of the CD to "trap the laser light" for better sound quality

[alfafalfa]

A happy new year for everybody !


http://www.barberelectronics.com/parallelopamps.htm

Why don't you look here ?

And try it yourself.

Alf

[Joe Hart]

Does anyone have any suggestions for stacked chips and what they sound like? Like I feel that 4558 is a bit brighter but less noisy than a 741 (which I find to be a bit warmer). The 4559 is a little more "hi-fi" sounding than the 4558. Etcetera... So, any opinions on stacks? I know that I could just do it myself, but it seems like I would be randomly stacking combos and I was never that good at math, but if I use 5 different IC's and do every combo of 2, 3, 4, and 5 IC's, I would end up with just shy of a kajillion different stacks. Just trying to narrow the search some!
-Joe Hart

[oskar]

...but it seems like I would be randomly stacking combos

I think actually that's the whole point... This isn't the way you're supposed to use OP-amps. Also we're probably discussing two different things in this thread. And please do correct me on this one...
1. Using multiple OP-amps in a gainstage to lower the noise in that stage. which I tried to simulate in Audacity by generating tones, white noise and adding them together and then repeating that maneuver in absurdum... End result... I feel like a complete idiot.
On an emotional plane I can't understand why it doesn't work. It just doesn't...       
I thought the noise would sum up like noise + noise = noise*2^0.5 and the original signal would go signal*2 and the endresult would be lower noise.
(2*(signal+noise))/2 = (2signal+2*2^0.5noise)/2 = signal + 0.707... noise     

2. Changing sound. Stacking amps like on the link to Barber Electronics, tying their outputs together without resistors makes them do things they don't necessarily like a lot. You just don't tie outputs together.
The outcome of it will depend on individual chips so there just ain't no telling which ones to combine.

[Joe Hart]

The outcome of it will depend on individual chips so there just ain't no telling which ones to combine.

So, do I understand this correctly? If I have three 4558 chips, then stack A+B, B+C, and A+C will all sound different? Hmm. Then I guess I stack a bunch of random chips, try to limit myself on how many, and just go with whichever sounds best (with the nagging feeling that maybe the next combo would be better). I don't like random chance -- I like black and white. If only the world was simpler!
-Joe Hart

[timmyo]

Grab an audio editor and generate say a 5 second sample of white noise (which is distributed evenly amongst all Fs).  Then generate another 5 second block of white noise at the same amplitude and mix it in with the first waveform.  The resultant waveform shows that noise does increase - if there were any significant phase cancellations that would bring the level down, the editor would show a decreased amplitude.

I think this displays that noise (no matter what type) added together will always increase average levels.  Noise is just too random over any significant period of time to have any chance of cancelling out another sample of noise.

Jonathan

That's just the noise covered then. If you also mixed in 5 identical signals, might they not also add, and then when you normalise the overall signal back down, might you have a better signal to white noise ratio (I have no idea - it just seems odd we're only talking about the noise here - but maybe I just don't understand properly)

[oskar]

That's what I was trying to reason about, if I've understood you correctly. But it doesn't seem to work the way I want it to.   
I just generated two tracks of white noise and mixed them together. The result was white noise twice the amplitude.
So far this quote sums it up best.

Noise is just too random over any significant period of time to have any chance of cancelling out another sample of noise.

Noise really is noise... Still, I haven't given up on stacking yet though.

[Sir H C]

Stacking op-amps does not increase the closed loop gain of the amplifier stage, that is set by the resistor ratios.  The noise is uncorrelated and adds so now you have the same signal level but more noise.  You might get different sounds at the limits as the two or more opamps fight to control the output, that is where things really get weird with stacking. 

[ayayay!]

Does anyone have any suggestions for stacked chips and what they sound like?

Hey Joe Hart, it depends on what application (pedal) they're in.  BUT I dig stacked TL072s in a Daddy O or a Tube Screamer, but only after both have been quite heavily modified.  It's "different" and can be cool, but some people hype it wayyyy too much. 

[psychotech]

I am working with menatone clones with no transistors. The design uses only one Op-Amp and sounds very clean and transparent. Using two op's cranked up the gain and beefed up the bottom end and tightened the bass somewhat. I am having some heat though and was wondering are the Op-amps supposed to heat up when stacked? They don't heat up when only a single one is used.

[Nasse]

http://www.elektor.com/magazines/2010/october/the-5532-opamplifier-part-1.1529745.lynkx

I have heard tight bass too when stackin semiconductors

[merlinb]

So who's right? Does stacking chips result in lower noise?

No, simply stacking opamps won't improve S/N, as other have pointed out. However, you can get a real noise improvement if you keep their feedback paths and outputs separate, and then mix the output signals together via 10 ohm resistors (usually).

In a guitar pedal it is a moot point, however, as the noise from everything else will totally swamp any improvement in parallel gain stages.