Circuit review?

[paulca]

Can I be so bold as to ask for a quick review of this circuit?

Obviously it's a basic cmoy style headphone amp with the OPA551.

Is the thermal via stitching under the opamps overkill, likely to cause issues or fine?

The OPA551 layout is my best attempt at copying their layout from the datasheet, which is a bit confusing.

I've done my sums and simulations, but just checking the thermal shutdown LED should work given the 50nA OFF and 120uA ON current source from the OPA?

I expect this will be quite loud, but not dangerously so.  At full volume with a domestic signal of around 1Vpp it should produce around 200mW and 7.6Vpp.  Of course it shouldn't run at full volume very often, that's the point of having a volume knob.

[madires]

Doesn't the voltage divider based virtual ground hop around with the load?

[paulca]

Doesn't the voltage divider based virtual ground hop around with the load?

There are more complex solutions, but the divider based one is very common in cmoys.  I think it sort of relies on the audio "load" being kind of symmetrical and loads both rails more-or-less equally.  Putting a DC load on one of the rails would probably offset it, but I have avoided doing that.

[paulca]

Hmm..  An AC analysis showed the 22u into the 71 Ohm load formed a band pass filter with roughly a -3db roll off at 100Hz.

I could live with that, but I'd rather not, I like my low end.  So the choice is to sacrifice power output and increase the output load series resistance or increase the size of the output capacitor and risk upsetting the opamp.  To target -3dB at 20Hz look like I need to push the load resistance closer to 600Ohm.  However simply changing the output capacitor to 220u and leaving the 71Ohms alone works.

600 Ohm load on the 12V single supply available only provides (ball park figures) 26mW output power and the speaker will lack response causing other filtering issues I'm sure.

However, these are all details I can change after the circuit is printed.

Details I cannot change (easily) would be the addition of a 100pF cap in parallel with the feedback resistor which add a -3dB roll off on the high end of around 100Khz while leaving 20Khz pretty untouched, flat(ish) 20/20k band pass.  The ponderment I have however is whether to bother.  I won't hear it, the headphones probably won't respond to it and this is an end-of-line amplifier so nothing else has to see it. The only thing it would affect would be the OPA551 itself.

[madires]

There are more complex solutions, but the divider based one is very common in cmoys.  I think it sort of relies on the audio "load" being kind of symmetrical and loads both rails more-or-less equally.  Putting a DC load on one of the rails would probably offset it, but I have avoided doing that.

Let's assume a simple test setup with a 1kHz sine wave input signal. What will happen to the virtual ground during the positive half of the sine wave? And during the negative half?

[StillTrying]

It's been years since I looked at these things, but I'm sure the purpose of the +/- supply is so that the outputs can be DC coupled, if the outputs are AC coupled I think the +/- supply is going to cause more problems that it solves.

[paulca]

There are more complex solutions, but the divider based one is very common in cmoys.  I think it sort of relies on the audio "load" being kind of symmetrical and loads both rails more-or-less equally.  Putting a DC load on one of the rails would probably offset it, but I have avoided doing that.

Let's assume a simple test setup with a 1kHz sine wave input signal. What will happen to the virtual ground during the positive half of the sine wave? And during the negative half?

The 220uF caps will supply the current to hold the voltage should any of the three rails try and move.  I expect it will even do a 10Hz wave without the ground shifting.  A DC load will probably offset it though.

My current headphone amp has exactly the same circuit.  I'm fairly sure the Pro-ject Headbox £200 botique headphone amp I have uses the same circuit.  The only difference here is that my amp has a good bit more output current potential.  I "could" make it more complex by adding a buffer opamp as voltage follower to sabilize the v-ground.

[paulca]

It's been years since I looked at these things, but I'm sure the purpose of the +/- supply is so that the outputs can be DC coupled, if the outputs are AC coupled I think the +/- supply is going to cause more problems that it solves.

A true +/- supply relative to external/signal ground allows the inputs and outputs to be DC coupled, but nobody ever does, because it immediately makes you device incompatible with anything which produces a DC offset on it's outputs.

A virtual ground is used when you only have a single supply.  So you need to create an artificial ground of (usually) 1/2 Vcc so that the amplifer can amplify both + and - signals.  There for you need an AC coupling cap and a ground reference resistor on the inputs... and potentially the outputs, but I'm relying the headphone being a resistive load to ground to pull the output down.

[Twoflower]

One additional point would I would avoid polarized caps at the input and the output. If possible I would also skip non-polarized electrolytics.

Not sure about the virtual GND. My first thought: No way. But the more I think about it I think it might work. As it's only used as reference for the OPAs. But for my taste the 2x 4.7kOhm is a bit too high to help stabilize it at the center. At power on it rely on identical cap-values as voltage divider. In real world they differ +/-20% (or +/-10%) so does the GND until the resistive voltage divider equalize it. The result will be a offset at the output during power on.

Wrt the thermal protection indicator. The datasheet works GND-Based and not V- I'm not sure if your circuit will do. Plus the 3.43V is fairly low for the 2N7002. Probably still OK, but if you have an alternative go for it.

Layout:
* The +12V trance close to the OPA551 seems very weak. In general I would increase the power traces a bit.
* Since you're already SMD, why do you use through hole caps?
* If possible place the parts a bit cleaner. For example R2 close to C3; R3 to C4. With the current design that doesn't make a big difference. but it looks cleaner. In general try to clean up the placement (e.g. swap R2 and R3)
* Why the strange layer jumping between C3/C4->R12? You're already at the top layer going to R2.
* Do you really want to have the LEDs spread over the board? No connectors for front panel mounting?
* Depending on the way you plan to solder the board and your solder experience: The pad-size seems rather small.

[paulca]

One additional point would I would avoid polarized caps at the input and the output. If possible I would also skip non-polarized electrolytics.

Not sure about the virtual GND. My first thought: No way. But the more I think about it I think it might work. As it's only used as reference for the OPAs. But for my taste the 2x 4.7kOhm is a bit too high to help stabilize it at the center. At power on it rely on identical cap-values as voltage divider. In real world they differ +/-20% (or +/-10%) so does the GND until the resistive voltage divider equalize it. The result will be a offset at the output during power on.

Wrt the thermal protection indicator. The datasheet works GND-Based and not V- I'm not sure if your circuit will do. Plus the 3.43V is fairly low for the 2N7002. Probably still OK, but if you have an alternative go for it.

Layout:
* The +12V trance close to the OPA551 seems very weak. In general I would increase the power traces a bit.
* Since you're already SMD, why do you use through hole caps?
* If possible place the parts a bit cleaner. For example R2 close to C3; R3 to C4. With the current design that doesn't make a big difference. but it looks cleaner. In general try to clean up the placement (e.g. swap R2 and R3)
* Why the strange layer jumping between C3/C4->R12? You're already at the top layer going to R2.
* Do you really want to have the LEDs spread over the board? No connectors for front panel mounting?
* Depending on the way you plan to solder the board and your solder experience: The pad-size seems rather small.

Thank you.  This helps.

I've tried SMDs electryltics a few times, decided to go back to through hole.  The SMDs are usually just the THT versions with the leads mounted in a plastic base.  I found them trickier to solder too, found to reflow them with hot air the bases melt while the case soaks up the heat.

I could probably go 0805 on the 10uF, maybe, but I have 10uF THTs in stock.

I had the thermal protect mosfet on the GND, but I moved it to V- for some reason.  I can move it back to AGND and pick a different resistor to give me closer to 5V.  Thanks, this is my first circuit with a mosfet and I know they can be tricky.  I did simulate it in Spice with a 2N7002 model, but ... that doesn't always equate to the real world.

This is a prototype for part of a much larger project, so some of the design decisions are a bit weird.  In the full project I have +-15V rails to play with and melt headphones, but as I was making the prototype I considered...

1.  Test the general amp works as expected.
2.  As I need to make it anyway, it could realistically replace my current amp, so I need it to run on a single 12V rail.

On that last point, I will take another pass based on your comments and a general tidy up, given it should be a decent looking device in it's own right, no need to be messy.  Having all the LEDs in one corner in a neat row would be cool.  I had already laid it out before I picked a case and thus a board shape/size.  I can now let things spread out a bit more while keeping the important stuff close.

On soldering, this board is sparse compared to my last one.  I use a microscope, paste and hot air, so as long as I can get the teasers into drop the components, it's fine.

[StillTrying]

A virtual ground is used when you only have a single supply.  So you need to create an artificial ground of (usually) 1/2 Vcc so that the amplifer can amplify both + and - signals.  There for you need an AC coupling cap and a ground reference resistor on the inputs... and potentially the outputs, but I'm relying the headphone being a resistive load to ground to pull the output down.

I'm afraid that you've convinced me so much that at a virtual ground is needed, that I'm sticking by my previous post.

[paulca]

A virtual ground is used when you only have a single supply.  So you need to create an artificial ground of (usually) 1/2 Vcc so that the amplifer can amplify both + and - signals.  There for you need an AC coupling cap and a ground reference resistor on the inputs... and potentially the outputs, but I'm relying the headphone being a resistive load to ground to pull the output down.

I'm afraid that you've convinced me so much that at a virtual ground is needed, that I'm sticking by my previous post.

The interesting difference in that circuit is that the feedback loop is biased to actual ground via a cap to block DC.  I've seen both variations.

The other thing I have seen is biasing the feedback to "VGND", but also putting a decouling cap on VGND to GND near the amp to absorb and noise.

[StillTrying]

Not sure about the virtual GND. My first thought: No way. But the more I think about it I think it might work. As it's only used as reference for the OPAs. But for my taste the 2x 4.7kOhm is a bit too high to help stabilize it at the center. At power on it rely on identical cap-values as voltage divider. In real world they differ +/-20% (or +/-10%) so does the GND until the resistive voltage divider equalize it. The result will be a offset at the output during power on.

I agree the 4k7s seem a bit high, but I can't see an uneven current anywhere, and with the VGND's decoupling caps being 10X the value of the output coupling caps. I think the VG will only ever move about 5-10%, - in theory. 

Edity random thought: I don't think the input 22u caps are needed.

[BrianHG]

This will be the only 1 comment I will make on this thread.  Paulca, you do realize that having headphone jacks right next to the left and right of the volume knob not only means it will be difficult to turn, even considering a bit of height difference, you wont be able to place a knob on the volume knob, or even use professional high quality headphone jacks which may have a larger diameter that typical earphone bud jacks.

[paulca]

or even use professional high quality headphone jacks which may have a larger diameter that typical earphone bud jacks.

I haven't seen a pair of headphones in a long time that had only a 1/4" jack plug.  Most have been 3.5mm with a 1/4" screw over.

This is desktop audio, not a mixing desk.  Size matters.

I have to get all of things I am prototyping over the past month onto one single board and case that I hope will be under 6" across!  So I need to practice with tight layouts.

Going panel mount on the pots and jacks is starting to seem like the right approach for the final all-in-on though.  Even for part selection, it gets a lot easier when you aren't mounting it onto the PCB to pick those things.

I noted the pot is very close to the jacks.  It makes a lot less sense when you add plugs with mouldings, making a knob over the shaft of the pot not likely to work.

Also as I found on my DAC, getting 2 PCB mounted jacks and a PCB mounted USB to work when it comes to a case requires... inovative and fairly messy solutions.  Like jack holes big enough for the plastic moulded barrel of the plugs!

For litlte small prototypes I'd prefer they were PCB mount, all in one units though.

[paulca]

So the board arrived, soldered it together.  Only the left channel works.  The left is absolutely fine.  The right is completely dead.

Probing the board shows what I believe is the issue, but I'm not exactly sure what to do about it.

R4 and R5 are the virtual ground biasing resistors.  On the working left channel the DC offset is 4.4V (which is a bit strange for 1/2Vcc of 12V), but that's a separate issue.  The right hand channel however has a DC offset of only 2.4V.  Yet the feedback on the right channel is biased to 4.4V, so the right op amp sees it's inverting input consistently higher than it's non-inverting input and thus nothing comes out of it.

R4 and R5 are 22K.  I am wondering if I made them smaller, say 2.2K to ensure they pull the inputs up harder.  Although that doesn't explain why one works and one doesn't.

I measured the input caps, which I decide to up to 220uF, one reads 260uF the other reads 331uF!

Do you think it's worth replacing the 220uF with 22uF and replacing the 22Ks with 2.2Ks?

[paulca]

Dropping the caps from 220uF to 22uF fixed it.

The next issue is that it's incredibly sensitive to power rail noise.  It's totally silent running on a battery, but even when I connect a volt meter to the battery I can hear the switching hum of the LED driver. 

[Twoflower]

Is there any reason for the divider using 2x 22kOhm? That makes the virtual GND extremely sensitive. The current running through that resistors is 200µA. So any stray current around a few µA gets your setup out of balance. At least go to 2.2kOhm or even 1kOhm. This will probably stabilize your setup and reduces the noise-sensitivity.

[paulca]

Is there any reason for the divider using 2x 22kOhm? That makes the virtual GND extremely sensitive. The current running through that resistors is 200µA. So any stray current around a few µA gets your setup out of balance. At least go to 2.2kOhm or even 1kOhm. This will probably stabilize your setup and reduces the noise-sensitivity.

The virtual grounds divider uses 4k7s.  If I dropped those to 2k2 it would burn away 2mA / 32mW.  I suppose that not an awful lot.  1k's would waste 6mA / 72mW.

[Twoflower]

Ups, indeed 4.7kOhm. My mistake.

At least for a Lab-Try reducing that resistors would give an idea if that would be a cure for the problem. It might not be a permanent solution but it might rule out that this is the root cause.

Have you tried to add some 100nF parallel to the 22µF?

[paulca]

I'll have a play with it over the weekend.  For now it's working perfectly running off a 3 series LIon battery.  I got the gain just right, in that on full tilt it will drive my headphones to distortion.  For prolonged listening a setting of about 1/4 volume works.

[paulca]

So I couldn't get anything by poking at the circuit on the board.  So tonight I set up a breadboard with the same cmoy style amp in mono with NE5532s.

I could instantly replicate the power rail noise transfer.

Long story short, things had to get inivative and out of the box,  but I tracked it down to one component.  The upper virtual ground cap.  It is effectively and literally AC coupling power rail noise directly onto the virtual ground and thus straight to the output with gain.  When it is removed the noise stops.  I knew something was going on because when I got to 2000uF of decoupling on the power rails and STILL had noise, though each cap lessened it slightly, I knew something wasn't right.  No measly couple of 100Hz digital switch hum should survive that much decoupling.

Now obviously that cap serves a purpose (other than making noise).  So I used the other opamp channel as a buffer for the virtual ground to give it some grunt.

It's fine with a decoupling cap between the VGND and GND, but not VGND and VCC. 

So the circuit now has
An NE5532 as a buffer
A 470uF VGND to GND

I removed the 470uF from Vcc to GND as it was decoupling the power rail noise onto the GND, as you would expect, but that was then being AC coupled back onto the VGND. 

And the power rail noise has dropped from 200mV to 5mV and is no longer audible, unless you listen really really hard on max gain.  Good enough.  Also the drift on the VGND driving the NE5532 to it's max was only about 5mV, still absolutely fine.

I did forget to test making the divider resistors smaller, which is a bit dumb of me as that might well have helped too.  I'll try and test that tomorrow evening as I have dinner to make.  That said the power on the virtual ground now comes from the NE5532 buffer, so as long as it has enough grunt it should be moot.

The thing is... given that the actual amp uses OPA551s which put out a shed more current I do need a buffer that can sink/source the return current, so I need another OPA551 as a buffer?  I know there are high current buffers, but they aren't any cheaper than the 551.

[paulca]

Small follow up.  I think the CMoy circuit relies heavily on the fact that it run off a battery.  If you want to run it off a power supply, or indeed if you want to run anything else off the same battery the virtual ground becomes a major issue.

The one thing that still baffles me is that the cheap chinese Classic 47 headphone amp is a CMoy and it is immune to this power rail noise and I have no idea why.  It could simply come down to it have much less gain.  I believe it's gain is only 2.x.

Here is it's schematic.