bipolar supply summing amp DC on output?

[ashcat_lt]

I have no idea where to start to troubleshoot this.  I've built a simple non-inverting opamp summing amplifier.  Actually, it's a stereo amp using a TL072 (for now, but socketed for possible future upgrades) and a dual gang pot for the gain.  It uses a split +/- 12 V supply which measures out just about right - from ground to the positive rail is 11.75V, and to the negative rail is -11.75V or so.  I've the input AC coupled via a cap, but the didn't have a big enough cap for the output.  Figured I could leave it DC coupled since it's a bipolar supply, and I'm fairly confident I can trust the things to which I'm going to attach it to not leak much DC.  

The gain control consists of a 10K pot wired as a variable resistor (parallel to an 18pf cap) from out to -in with a 680R resistor from -in to ground.  With this turned all the way up, I get a full 10V across the output.  Turned all the way down this comes down to about 5V.  This is true on both of the opamp stages.  I have confirmed that they seem to be working independently in that when I connect audio to the left input I hear sound only on the left side, and when I connect to the right input, it only comes out that side.  I have checked for inadvertent solder bridges and things, but I can't see any, and the way it's laid out there's just no way such a thing could affect both sides.  I'm at a loss on where to even start.  Is there something really basic that I'm missing or?

Edit - here's the basic schematic for one side of this thing.  Doesn't show the power supply section, but that reads fine at the opamp.  This looks super simple and that's all I've got on this thing!  (The 1000M resistor isn't actually there, it's just here to fool 5spice into running its analysis, seems it kind of hates series caps.

[slacker]

You need a resistor from the + input to ground to bias the opamp around ground, without it the + input is floating, which will cause all sorts of weirdness.
You might also want to add a cap between the 680 Ohm resistor and ground, to set the roll off frequency in the audio range. At the minute the opamp will amplify DC so any DC offset will get amplified giving you a larger amount of DC on the output.

[ashcat_lt]

Thanks for your  reply.

I had tried a 1M resistor to ground at the input, but I put it "outside" the coupling cap, and that didn't help.  Guess I need to move it over to the other side?

The problem I've got with the cap to ground is that all the big caps I have are polarized electros with 16V rating.  I know that I can put these back to back and fake a non-polarized cap, but I still worry about the max voltage.  Could I be lucky enough that the back to back arrangement lets me add their V ratings?  I've got a RadioShaft close by if I must, but I'd rather use what I've got.

[ashcat_lt]

Well anyway, it works now.  Thanks for the input.  I put 1M pulldowns at the inputs and found a couple 10uf/35V nonpolar electrolytics at RS.  They were physically HUGE and axial leaded, but I managed to squeeze them on the board.  5spice told me the -3db point is somewhere around 20Hz.  I might lose a tiny bit of the super lows at very high gain levels, but I don't think I've got any speakers which actually reproduce that low anyway.  Now the DC output at all gain levels is within hundredths of mV of 0.

For those who might be interested, it's got 20 stereo pairs of inputs (hardwired to the board with approximately 3 ft of pigtail - this will terminate at a patchbay), a single passive output in case I ever want to use something else (tube pre or whatever) for a recovery amp, and four active outputs with independent Volume controls to feed multiple sets of monitors.  Currently my studio's got two pair working and a headphone amp, so there's a little room for expansion. 

The TL072 sounds fine to me, but it's socketed in there in case I ever decide to go cork sniffing.  There's a lot of gain here.  Somebody told me that you lose 6db for each doubling of the inputs.  10 inputs per side should come out somewhere around 20, so the max gain is right around there.  Testing with just one input, though, it's super loud.  I guess maybe when i have all the inputs connected I'll start to see the volume drop.  Of course, any time you mix two signals which sit at the same level the output will be 6db higher, so I think that should sort of even itself out.  I don't mind much running it at unity most of the time anyway, since it's flat as far as the eye can see (like Nebraska!) with no gain. 

With the 24V power supply it's got piles of headroom.  My testing so far has used my iPhone's headphone out as input.  Even with that and the gain on the mixer cranked I couldn't hear any distortion if I turned the output volume down far enough not to crush the amp which followed.  Not exactly scientific, I know, but good enough for me. 

My use for this is mostly for monitoring during tracking.  I mix ITB exclusively, but I do have my eye toward a day when I might try this analog summing thing that everybody seems to swear by.

[PRR]

> 1M pulldowns ....10uf ... 5spice told me the -3db point is somewhere around 20Hz.

Then you didn't ask the question correctly. (And you shouldn't need to ask a machine...)

0.01uFd against 1Meg is 17Hz.

10uFd is 1,000 times bigger. 17/1000 or 0.017 Hz.

> I don't think I've got any speakers which actually reproduce that low anyway.

You don't have speakers which are honest to 50Hz; such brutes are very rare and not usually worth the size/cost.

If you are in doubt, then you don't have a room which is honest even to 50Hz. That requires a VERY large (>20'x20'x20') room or extensive (expensive) analysis and treatment, including absurdly heavy walls.

This stage is not affecting your bass response.

> The 1000M resistor isn't actually there, it's just here to fool 5spice into running its analysis, seems it kind of hates series caps.

It is sometimes necessary to toss 1T into a series-CAP affair so that SPICE can compute the DC voltage everywhere (even where unimportant).

But SPICE was also complaining that U1 +in had no DC path. Each input to an ampifier *MUST* have a known DC voltage. Either right to a voltage reference, or (for -in) from feedback, either through a reasonably small resistor. If you don't tell it where to be, Murphy's Law ensures it will find an unhappy place to be.

In this case, TL072 has very-very-small input current. Teeny, but NOT "zero". Teeny current against infinite bias-leak resistance drifts to infinite voltage. Assuming just N-JFET inputs and +/-12V supply, it probably drifted up to 11V and "stuck" there. (It may drift the other way because there's other stuff on the chip inputs.)

Because '072 input current is teeny, 1Meg is "nothing" to the input current. At room temp, '072 sucks 100pA input. 100pA times 1Meg is 0.000,1 Volts. There's also a  voltage error, typ 0.003V, so the error of 1Meg is negligible. (If the chip were as hot as boiling water, '072 inputs can suck 10nA, 0.01V error, still small though sometimes significant.)

[ashcat_lt]

> 1M pulldowns ....10uf ... 5spice told me the -3db point is somewhere around 20Hz.

Then you didn't ask the question correctly. (And you shouldn't need to ask a machine...)

0.01uFd against 1Meg is 17Hz. 

10uFd is 1,000 times bigger. 17/1000 or 0.017 Hz. 

But the 10uf I added is nowhere near the 1M.  The 1M is at the + input as suggested above.  The 10uf is in series with the 680R to ground from the feedback path also as suggested above.

Somebody on another forum has got me a little worried that the 2K mixing resistors might be too low for the recorder output to which it will be connected, but this was recommended in a couple different places.  Sure hope I don't end up having to replace all 20 of those suckers!