Coupling caps between gain stages

[Fleetdog]

I know it's very common to put coupling caps between gain stages and that never really struck me as odd until today.  I'm working on my own design of an op amp based circuit.  I want to run the op amps of ground and +9V and just bias the incoming signal up to be centered around 4.5V.  Certainly a coupling cap at the input and output is necessary to block my DC biasing from leaking out either end of the effect, but do I really need to put coupling caps between stages?  Assuming my input goes from -100mv to +100mv peak to peak, I would bias that to be 4.4 to 4.6V then put it through a gain stage (let's assume a gain of 10) so out of the first stage I would have 3.5 to 5.5V peak to peak.  It seems to me like if I then use a coupling cap, it would bring it to +/- 1V and I'd need to rebias that back to 3.5 to 5.5V for the next op amp stage anyway.  Am I just missing something here today?  It's been a long time since I've done any actual circuit design or analysis.  Maybe I'm just too rusty. 

[DougH]

Well, I don't know what your 2nd stage looks like but if a 4.5v bias is sufficient for it, you are exactly right. (And it's good to see this kind of thinking here. ) If you can DC couple gain stages together, by all means do it. This makes extra bias networks and coupling caps superfluous. Check out Roger Mayer's work (Axis fuzz for one) for good examples of DC coupling. I incorporated DC coupling in a design I did last year and got maximal sound with minimal parts.

[soggybag]

JC has a page where he talks about coupling caps. The gist of the article is that he thinks they muddy the sound. He recommends not using them. This was all related to the TS.

http://www.lynx.bc.ca/~jc/pedalsSRVspecial.html

[DougH]

Can't comment on that but I do know if you have a lot of coupling caps you might accumulate enough phase shift in the signal to be noticeable. It's usually more of a concern for hi-fi than it is for 9v pedal circuits. And in a pedal the tonal effects may be considered more a "feature" than a problem.

[JDoyle]

DC coupling is a great idea for a majority of what we do, especially simple gain stages, just remember that any amplifier stage has an offset error so that the output of an opamp isn't REALLY biased at the same point as the input, but there are ways around this, such as servo biasing (which eliminates the advantage of fewer parts...).

For a lot of my transistor stages I directly couple a buffer onto the collector/drain of the amp stage and then bias the emitter/source of the buffer to where I want it.

Once you start to get out of basic gain stages and into distortions using diodes, it gets a bit trickier because the diodes generally have to be isolated from the DC bias for anything to work right.

But in your case, I'd go for it...

An op amp, after all, is just a bunch of transistor amplifier stages DC coupled together...

Regards,

Jay Doyle

[R.G.]

There is just so much nonsense on the web today, and such a small amount of reality.

Coupling caps allow you to couple signal from one stage to another without accurately equalizing the DC levels between them. There are penalties for using coupling caps (gosh! sorta like you pay for what you get with everything!) and the trick in using them is to know what matters, what doesn't, and what hurts. The devil is ALWAYS in the details.

If you can make your circuits operate DC coupling, you get the advantage of (a) not using a $0.05 capacitor and (b) not having your circuit limited by whatever the cap does as a side effect. You pay for that by having no means of isolating the DC levels, and being forced to accept signals varying all the way to DC, which is very much not wanted a lot of times.

For instance, opamps running from a common bias point. They're always tied to the same Vbias, why decouple them?

Well, opamps are not perfect. If you look at the datasheets of opamps, you'll see specs for the DC accuracy of them. In particular, the input offset voltage. This is much better these days than it was, but 10mv is common and 50mv is not that hard to find. What number means is that the opamp will "see" an input of that amount different from the real DC input level. Then there is bias current. We use a lot of 1M input resistors. A 1M resistor is the same as saying "1V per microamp", 100mV/100nA, 10mV/10nA, etc. So if you have a few nano amperes of input bias (again, read the datasheets) you're adding up the mV.

No biggie, right? What's a few millivolts? But we often then put those millivolts into an opamp stage with a gain of 10.05 times huge. You can start building up DC offsets that accumulate through your pedal. If you're lucky, they'll cancel. If you're not, you have an interesting debugging job ahead of you if you don't already know what I just typed in.

The task is harder with discretes where there are no neat datasheets telling you what the input offsets, leakage, and bias currents are.

Balance against that that if you do use coupling caps, you have some low frequency rolloff - which you might very well want - as well as the cork-sniffing side effects of capacitors like phase shift (only happens near the rolloff frequencies), ESR, ESL and dielectric absorption. In THIS arena, you have to fight your way through the hifi tweakos and others who think capacitors are fundamentally evil, even after they've been exorcized, and who will speak in riddles about transparent sound stages, inner clarity, and muddying the sound field. There is no amulet here. Unless you know the facts and remember them, the water-jacketed-speaker-cable people WILL get you.

Bottom line? LEARN what caps do and don't. LEARN what devices do and don't. LEARN to think about circuits and what happens inside.

[DougH]

Re. accumulating bias voltage offset- I'm not attempting to build 10 stage circuits with or without DC coupling. The circuit I built was a 2 stage circuit, and when I breadboarded it I honestly could not hear the difference between AC or DC coupling. So I went the simpler route.

I didn't read JC's article (most of the time I don't really understand him), which is where I suspect your reaction is coming from, R.G. But I certainly make no claims one way or the other about the inherent "evilness" of capacitors or any other discrete components. OTOH, I'm glad to see people on the forum thinking and questioning rather than just blindly throwing components in they don't need just because, well, hey, "that's what we always do".

[JDoyle]

No biggie, right? What's a few millivolts? But we often then put those millivolts into an opamp stage with a gain of 10.05 times huge.

Gotta disagree with you here R.G., I can't think of a single op amp based effect circuit (off the top of my head, I am completely prepared to be proved wrong) that amplifies all the way down to DC. Every case I can think of, like the TS for example, has the gain reduced to 1 at DC (via the cap on the 'lower' arm of the feedback resistance).

The offsets will accumulate, of course, but I can't think of a case where they are amplified.

But if the designer is building from scratch, they are going to have to use NONINVERTING op amp stages (as is the case with most FX). It is impossible to isolate a DC coupled INVERTING configuartion from amplifiying all the way down to DC.

Everyone is on their own with discrete parts, mostly because the aren't biased the same way, an op amp has it's input AND output biased to the same point - the offset is therefore easily measureable - discrete stages don't, so how do you even know how to measure the offset? What is it offset from?

[Plus, the tweekos aren't always off their rocker, though they most certainly often are. Walt Jung did an article about caps that showed that certain series of ceramic caps can change their capacitance value by as much as 5% with only a 5V change across them. Certainly enough for us to take notice to either avoid, or abuse for our purposes.]

[Mark Hammer]

Aw man, you scooped me, RG.

+1 on the recommendations.  There are plenty of instances where a bias can be "recycled" and shared across stages without penalty.  Given that so many of us here construct things with copious amounts of gain in a single stage, and that experimentation with op-amps is encouraged, blithely assuming that the bias fed to the input of an op-amp will necessarily be accurately reflected in the output, is false confidence.

There are also other instances where sharing a bias (and eschewing coupling caps) may not be the wisest course of action.  I am reminded of the Zombie Chorus, which originally used an asymmetrical Vref (i.e., 3/5 of V+ rather than 1/2) intended to serve as both the bias for the audio and LFO subcircuits, as well as the bias voltage for the MN3007 input.  While it made for an economical design, it resulted in headaches and ticks for lots of people.  As many have found, sonic delight was more likely to occur if the individual parts of the circuit were provided their own optimal bias, instead of finding some sort of compromise the share between them.  And of course, if every stage gets its own appropriate bias, then we will see coupling cap to "reset" the DC bias to zero.

Some 20 years ago, a guy I knew in my department who was a neuroscience researcher, was trying to put together an instrumentation amp for some EEG recording.  Keep in mind that a great deal of electrical brain activity of interest shows itself in signals below the audio range.  Naturally, the guy had to use DC coupling as much as possible.  But given that the electrical potential being recorded were on the order of microvolts, there had to be a lot of gain in the circuit.  The schematic he was working from was not particularly specific about op-amp choice (or perhaps was based on the limited choices of 1975 or thereabouts).  It was driving him nuts.  The DC offset was so great that he simply couldn't get a decent signal.  I was working on a mixer for myself in the departmental workshop and suggested that he try out some CA3140 Bi-MOS opamps that I happened to have and have an interest in.  He popped them in and bingo, the problem went away.  The guy was so thrilled to have the circuit finally work and so grateful he bought me the Weller iron I continue to use to this day.

As a, perhaps more relevant, aside, I will note that many of Craig Anderton's projects included info on omitting the input cap.  The vast majority of pedals will have a coupling cap on the input and output.  They do so chiefly because the manufacturer has no idea wht their pedal will see at its input or output.  If you could guarantee that the pedal would always be receiving an input from something that had a DC-blocking cap on the output, then you could probably omit the input cap.  Trouble is, there are no such guarantees unless you have made the entire rig yourself and know precisely what it contains.

At another level, as I keep harping on, there is a difference between multi-source, wide-bandwidth signals (like an orchestra), and single-source, modest-bandwidth signals, like a guitar.  The first type is depending on the near total absence of group phase delay for you to be able to psychologically pair up the harmonic content with the fundamentals well enough to be able to know that THIS harmonic content goes with the violins, and THAT harmonic content goes with the brass, while THAT harmonic content goes with the winds.  The analogy I sometimes use is that of taking a picture of a group, moving everyone's nose 3 inches to the left and their eyes 4 inches to the right, and THEN trying to find your best friend in that group picture.  Hard to do.  Consequently, there is an understandable urge amongst audiophiles to avoid AC coupling with a DC-blocking cap because big-value electrolytics can introduce group phase delay that staggers harmonic components from their fundamental roots.

When it comes to one guitar going through a pedal into an amp, though, the psychological challenge in grouping content together psychoacoustically is reduced considerably.  Not eliminated, mind you, just reduced; which is why the BBE, intended to address group delay issues, still manages to improve clarity in an electric guitar.

My point here is that those folks in the audiophile and sound reinforcement world are not nuts (well....some of them... ) in their recommendations to avoid AC coupling and rebiasing.  In their context, there is value added to following that urge.  In our world, it matters considerably less.  I would caution a 13 year-old not to smoke, because of the lifelong risk incurred.  I would feel little urgency to make the same recommendation to a 93 year-old who has smoked since turning 20, because whatever risk you were hoping to avoid is essentially water under the bridge..

[Fleetdog]

Great responses guys!  I didn't figure this question would spark so much insightful discussion.  From what I gather here, I shouldn't NEED coupling caps in my design but they may come in handy if things just aren't working quite right.  When I get time to breadboard this bad boy up I will post a schematic for it (assuming it doesn't sound like total rubbish) for enhancement suggestions and critiques.  But for now, I'm greatly enjoying the technical theory discussion.  This kind of makes me miss my electrical engineering classes.

[DougH]

From what I gather here, I shouldn't NEED coupling caps in my design but they may come in handy if things just aren't working quite right.

I doubt you'll have a problem with it, and you probably won't need the caps. If you do get into biasing problems you'll know soon enough and offset voltages can be easily measured.

[matchless]

Okay, I'm still a beginner, so I apologize for the newbiness of these questions:

In the case of several non-inverting opamp gain stages, is it an acceptable practice to skip the coupling caps but still tie the +input of each stage to Vbias with a resistor?

RG - is this what you meant by "a lot of 1M resistors?"

I spent a lot of time playing with a 3 stage, discrete opamp distortion circuit a few months ago.  After hours and hours of experimenting with different coupling methods, I found that I always preferred the sound of the circuit with no coupling caps, but with each stage still tied to Vbias by a resistor.  Is this "good form" in circuit design?  It just sounded best to me this way.

I also found that I preferred the tone of the circuit when using smaller value bias resistors, say 10k-100k instead of 470K or 1M like I usually see.  This was without any coupling caps.  Is there a reason for the change in tone other than the difference in impedance (assuming the circuit uses an input buffer and each stage has a fairly low output impedance)?  Maybe I should rephrase - ignoring impedance could the value of this resistor possibly have an audible effect?  I would think that it shouldn't matter as long as the opamps have enough current, right?  So, the difference in impedance from the smaller resistors must have been the only audible effect.  Or am I missing something here?  Is impedance the only reason I see so many 470K and 1M bias resistors?

Finally, are the "improvements" in tone I heard (I swear, there was a difference!) most likely due to the fact that I was using discrete opamps which were probably not very DC accurate?  I guess I'll just have to order some nice accurate ICs and see if I can hear a difference, or as much of a difference between different coupling methods with those.

[JDoyle]

Okay, I'm still a beginner, so I apologize for the newbiness of these questions:

I wouldn't call anyone who breadboards a 'three stage, DC coupled, discrete op amp distortion' a beginner, but I am sure impressed with your modesty! 

In the case of several non-inverting opamp gain stages, is it an acceptable practice to skip the coupling caps but still tie the +input of each stage to Vbias with a resistor?

Sure, but understand that what you are doing has more to do with loading down the output of the previous stage than 'fixing' the bias. Without a cap completely eliminating DC altogether, the offset will still be part of the signal from the previous stage.

RG - is this what you meant by "a lot of 1M resistors?"

I think he meant that we use a lot of 1M resistors as bias resistors with op amps and any offset (or more accurately, bias current, remembering that any offset/signal has both voltage and current components) from one stage drops a larger voltage across a 1M resistor than across a 100k resistor. I don't like to use 1M resistors myself, more on that below...

...I found that I always preferred the sound of the circuit with no coupling caps, but with each stage still tied to Vbias by a resistor...Is this "good form" in circuit design?

In my opinion, who the hell cares if it is, IF:

It just sounded best to me this way.

That is ALWAYS the best judge of the 'right' way to do something with DIYFX...

I also found that I preferred the tone of the circuit when using smaller value bias resistors, say 10k-100k instead of 470K or 1M like I usually see.  This was without any coupling caps.  Is there a reason for the change in tone other than the difference in impedance (assuming the circuit uses an input buffer and each stage has a fairly low output impedance)?  Maybe I should rephrase - ignoring impedance could the value of this resistor possibly have an audible effect?  I would think that it shouldn't matter as long as the opamps have enough current, right?  So, the difference in impedance from the smaller resistors must have been the only audible effect.  Or am I missing something here?  Is impedance the only reason I see so many 470K and 1M bias resistors?

First you can't separate impedence from the effect of a resistor on the tone of a circuit. That is literally the only effect a resistor can have - impedence. But again, what I think you were hearing was the impedence of the biasing resistors loading down the previous stage and reducing the level of it's output signal - which many of us have found to be a useful tool in designing distortions. Sometimes you just don't want to smack a circuit with all you can, more like half-smack it, if you know what I mean.

The thing is, in a DC coupled situation, the ONLY option you have is to tie it to Vbias, if you tied it to V+ or Ground you would only pass the part of the signal that moved opposite to that bias (the negative swing in the case of V+, the positive swing in the case of ground). Vbias is close enough to the output of the previous stage, (which is (Vbias) + (gain*input signal)+/-(DC offset), keeping in mind that the signal swings both positive and negative, so for the negative swing, it subtracts from Vbias) that it doesn't affect the performace of the stages, but it DOES still load the previous stage. Not the most techinical of explinations, nor is it complete, but good enough for rock and roll, I think.

The reason you see so many 470k and 1Meg resistors is because that has just become 'standard'. Personally, I try to not go above 470k, and normally stick with 220k. My reasons for this are simply to reduce the additional noise from a 1M resistor and to LOWER the input impedence of a stage. Sometimes you really CAN have too high an impedence. The higher the impedence, the easier it is for smaller, similiarly high-impedence signals, such as stray radio signals, to get into the input and be amplified by the stage.

For example, if you directly couple a MOSFET voltage amplifer (not an easy feat considering the way a MOSFET biases, hence CMOS) you now have a node with an input impedence of almost a TERRAOHM, so even if you have a random radio signal in the air with an impedence of 100Megohms, the input to the MOSFET would easily pick that up and the loading would only reduce the radio signal by 1/10th, and your MOSFET would amplify that along with everything else you send to it...

Finally, are the "improvements" in tone I heard (I swear, there was a difference!) most likely due to the fact that I was using discrete opamps which were probably not very DC accurate?

Again, I think the improvements you were hearing had more to do with the padding down of the signal, bringing you out of the square-wave range, than any affect of the rather large offsets in a discrete op amp. Of course you do have the effect of the biasing resistors on any paracitic capacitance, as well as the paracitic capacitance of the resistors themselves, but relative to the signal, I think that would be a small part of the tonal change.

Remember that a square wave multiplied by a gain of, well, anything, is still a square wave. (Until it's slew limited...  )

I hope this helps,

Jay Doyle 

[matchless]

Wow, thanks!  That helps a lot, JD.  I do stick with what sounds best to me for my personal effects, but I always wonder whether engineers would laugh at my piddly efforts.    

[R.G.]

Engineers get a bad rap, probably because so many of them can't understand or communicate with real people.

The overall Engineer's Creed might  be something like  "If it works, do it. The theoretical physicists will figure it out in a few decades."

Not knowing exactly why something works or doesn't work has never been a great hold-up on engineering. As a group, though, we are very fond of knowing how and why things work, having run into a few surprises - like singed eyebrows - along the way.

What drives engineers nuts is people who casually presume that the several millenia of technical background is and must be wrong because they don't know it. The engineers won't laugh at your piddly efforts, not least of all because they're not piddly, but also because you're trying to learn and build on what you have found. You aren't trying to teach us about some alternate beautiful new reality that the techies missed for centuries, nor preaching about things that flatly aren't so or are unprovable. Or worse trying to sell such things to us.

So keep digging. You're doing fine.

[MikeH]

This thread gets an A+ for the mention of "psychoacoustics".

[kvb]

I also found that I preferred the tone of the circuit when using smaller value bias resistors, say 10k-100k instead of 470K or 1M like I usually see.  This was without any coupling caps.  Is there a reason for the change in tone other than the difference in impedance (assuming the circuit uses an input buffer and each stage has a fairly low output impedance)?  Maybe I should rephrase - ignoring impedance could the value of this resistor possibly have an audible effect?  I would think that it shouldn't matter as long as the opamps have enough current, right?  So, the difference in impedance from the smaller resistors must have been the only audible effect.  Or am I missing something here?  Is impedance the only reason I see so many 470K and 1M bias resistors?

First you can't separate impedence from the effect of a resistor on the tone of a circuit. That is literally the only effect a resistor can have - impedence. But again, what I think you were hearing was the impedence of the biasing resistors loading down the previous stage and reducing the level of it's output signal - which many of us have found to be a useful tool in designing distortions. Sometimes you just don't want to smack a circuit with all you can, more like half-smack it, if you know what I mean.

Jay Doyle 

I still can't discuss any of the math or theory related stuff, but I noticed in Aron's "Insanity Box" that in between each gain stage was a small value trim pot that was letting some of the signal out - before the next stage would boost it back up.

Davisson's overdrives use a largish 470K value resistor and a small value cap to control the amount of what is going into the next stage.

I've use both ways on a couple of things.

What I do know from listening while working on the breadboard (besides the usual stuff about what caps and resitors limit/allow) is that if you want the signal to go to the next stage and retain its sound, you would use the low value resistor to ground or Vref way. And if you want to change the signal somehow as it enters the next stage, you would use the series caps and/or resistance.

[pee-j]

- - - - off - - - -     a forum technical suggestion...

Such an incredibly beautiful thread here! Again!

- - - - -
the suggestion:

We should be able to kind of mark these / "tag" them, so they could be found easily in the future...
We could also set up an ever forming liquid book of knowledge -- of the forum posts (OPs and posts), and sometimes entire threads, or parts of threads (when they are 20+ "pages" in length)

A thread like this one is extremely valuable but not just practically... Historically, cultural-historically too...
Imagine that 10 years from now people won't necessarily even have a clue how a cool and elegant conversation used to be carried out... "back then"...
There might be lots of books and publications, lie 9+1 things you needa know about caps -- evil or not?
But a conversation like this is a different thing...
This is like an introduction to a culture...

Maybe the future mission of this forum is not only to provide space for new conversations but also to provide access
to past conversations... which they could be labelled, marked, tagged, and linked in various contexts...
available in a library like structure...

:::::
If only we could bookmark threads... in a special way too... like, "promote thread to the collective bookmark library"...
and then we could select some categories for it... (multiple categories allowed) .... etc etc...

[GibsonGM]

A 'forum wiki', which at one time (and may still be out there, I'm just not seeing the links) was part of 'the group' here.  +1. 

There are entire books written on this subject, so it could be difficult to condense big convo's like this or ensure they're accurate.  And I think, tho, that a simple bookmark might not do it - I have 100s of 'bookmarks' in my browser, and still find myself doing net searches for topics like this. When i read the 'newly found source' and go to bookmark IT, I often see "manage bookmark" instead of "Bookmark current tab", meaning it's already in my collection!     But the idea is great, just need a logical way to collect the info!

[amptramp]

The reason you see so many 470k and 1Meg resistors is because that has just become 'standard'. Personally, I try to not go above 470k, and normally stick with 220k. My reasons for this are simply to reduce the additional noise from a 1M resistor and to LOWER the input impedence of a stage. Sometimes you really CAN have too high an impedence. The higher the impedence, the easier it is for smaller, similiarly high-impedence signals, such as stray radio signals, to get into the input and be amplified by the stage.

The noise of a resistor is not as important as it seems in a lot of cases because if you have a situation like a guitar where you have a pickup with approximately a 15 K resistance (usually less) in parallel with a tone control of 500 K and a volume control of 500 K, these resistances are all in parallel at frequencies where you can ignore the capacitive reactance of the tone capacitor and the inductive reactance of the pickup.  This would be the midband of guitar response.  If you turn the guitar volume control down, which adds resistance in series with the signal, you still have less then the 500 K in parallel with a 1 megohm input resistor.  If you turn the tone control down, this further reduces the input impedance because this resistance is in parallel with the input resistor.

If you have a 1 megohm resistor at 25°C with a bandwidth of 10000 Hz, the thermal noise output is 12.832 µV.  There is a calculator at:

https://www.omnicalculator.com/physics/resistor-noise

If you have a design (like a Tillman amplifier) where there is no input capacitor, the effect of parallel resistance goes down to 0 Hz for the guitar volume control and pickup winding resistance, where the 15 K pickup dominates.  In the case of a 15000 ohm resistor, the noise is 1.5716 µV.  This isn't much noise even after a lot of amplification.  Where you have capacitive coupling, this effect remains throughout the passband of the coupling cap, so an input cap doesn't isolate the noise from a resistor, it just isolates it below the coupling frequency, which is relatively low because it must pass the 82 Hz open low E string signal.