Build A High Quality Audio Amplifier

[Whuffo]

Audio amplifiers aren't as easy as they look; they have to handle a wide frequency range and wide dynamic range and do it without adding or subtracting anything while driving a load that has varying reactance. Not to mention the human beings that use and abuse them.

So here's the schematic and parts list for one you'd be proud to own. All you need to add is a +/- 24 VDC power supply; I'd recommend a 36VAC center tapped transformer, a large bridge rectifier and some big filter caps. At least 5.000 uf per rail; double that won't hurt. If you're going to run multiples of this circuit, plan on supplying 2 amps per amplifier.  Heat sinks are critical; the driver transistors will be happy with finned hats, but the outputs need to be thermally coupled (and electrically isolated) from a heat sink of at least 60 square inches per amp circuit. Fins make this easier; using smaller heat sinks is risky. The power dissipation depends upon the output power, and at worst case it'll need all of that heat sink.

It's an elegant circuit; some of the design may be unfamiliar but it's been carefully optimized and if you build it exactly as designed, I guarantee it'll work great. Maybe some of the bright folks here can find ways to squeeze a little more out of the circuit? If you have questions or comments, I'd love to hear them.

Sorry about the rough looking diagram; I only had it on paper and didn't spend much time redrawing it to post here.

EDIT: See updated schematic on page two

[Tube_Dude]

Hi

- First of all, your input pot wiring is incorrect. The way you have it, when the pot is at minimum the input source will see a dead short, and it will not enjoy it. The center pin of the pot must be connected to the C1 input capacitor, and the point that was connected to C1 will become the input node.

- A 1KOhm resistor placed between C1 and the op-amp + input will help to not occur oscillations when the pot is at minimum, or maximum (in this last case with a cable connected at the input but not connected to the source).

- A electrolytic capacitor, say 100uf, across the biasing arrangement (between the bases of Q1 and Q2) will help the driver transistors (Q1/Q2) to have the same signal at each base.

- A electrolytic capacitor, say 100uf, in series with R1 will make the offset voltage lower, without loosing anything, in the usual audio band.

Cheers

[Whuffo]

Hi

- First of all, your input pot wiring is incorrect. The way you have it, when the pot is at minimum the input source will see a dead short, and it will not enjoy it. The center pot must be connected to the C1 input capacitor, and the point that is now connected to C1 will become the input node.

- A 1KOhm resistor placed between C1 and the op-amp + input will help to not occur oscillations when the pot is at minimum, or maximum (in this last case with a cable connected at the input but not connected to the source).

- A electrolytic capacitor, say 100uf, across the biasing arrangement (between the bases of Q1 and Q2) will help the driver transistors (Q1/Q2) to have the same signal at each base.

- A electrolytic capacitor, say 100uf, in series with R1 will make the offset voltage lower, without loosing anything, in the usual audio band.

Cheers

You're absolutely correct about the gain control, it should be connected as you say (and as it is on my paper drawing). I was in too much of a hurry when I transcribed the schematic. Swap the wiper and the high end of the pot and you're golden. Actually, for most purposes deleting everything before C1 would be fine.

The resistor to ground at C1 is not 1K, it's 100K - and it's there to properly set the differential input to the opamp without loading the input too much. It also acts as part of the input voltage divider.

Adding an electrolytic to the biasing chain is valueless; the diodes in the chain are in saturation at all times. They're used to develop the correct bias voltage and have no effect on the AC performance. And as far as the electrolytic in series with R1 goes (make the offset voltage lower), you're overlooking R2, which guarantees zero offset as well as setting the gain for the amplifier. The desire to add reactive components (like capacitors) to the signal chain is very common, but it's not advisable.  This design has one and only one capacitor in the signal chain - the input coupling capacitor. It's there because it's impossible to guarantee that the input will average zero volts DC. Beyond that, the amp is DC coupled all the way through. The fuse is there to protect the speaker - just in case.

I tested one of these on a Rhode and Schwartz distortion analyzer. At any output level, the R&S showed nothing more than its residual noise level. I've also (inadvertently) applied a dead short across the output before the fuse - at full output or several minutes. No damage resulted.

The design has been optimized many times; many of the components serve multiple purposes. Since no component is perfect, purity is best served by keeping the number of components to a minimum. The "fancy" biasing arrangement is needed to compensate for manufacturing variations in the transistors; there's no "one size fits all" value here. Once it's set, you won't need to touch it again unless you replace parts.

The circuit is a bit subtle; there's not one part in there that isn't absolutely required. Take another look, and try building one - hide it behind some glowing tubes and you'll fool the audiophools every time.

[ElectroIrradiator]

A few additional comments:

- VR1 should probably be 10K log. Otherwise the presence of R3 will skew the potentiometer curve to a noticeable degree.

- No high frequency compensation? Maybe a tiny cap across R2, or use a NE5534. The latter would also offer a convenient location for an output null adjustment, if needed. I would worry about potential high frequency oscillations at certain loads without one.

- C4 should be two 0.1 uF caps, one from each rail to ground, in parallel with the 5 (4.7?) uF 'lytics. As C4 is currently located, it will not be able to help supply high frequency current spikes to the output. (Or rather, it will draw them through the 'lytic on the opposing power rail).

- Would using the classic 'active' BJT bias network in place of D1-D3 + VR2 reduce thermal drift of the DC operating point?

[Tube_Dude]

- A 1KOhm resistor placed between C1 and the op-amp + input will help to not occur oscillations when the pot is at minimum, or maximum (in this last case with a cable connected at the input but not connected to the source).

The resistor to ground at C1 is not 1K, it's 100K - and it's there to properly set the differential input to the opamp without loading the input too much. It also acts as part of the input voltage divider.

Please reread what I have typed about the 1k resistor. It have nothing to do with the 100k biasing input resistor.

[[ Adding an electrolytic to the biasing chain is valueless; the diodes in the chain are in saturation at all times. They're used to develop the correct bias voltage and have no effect on the AC performance]]

You are forgetting the variable resistor in series with the diodes. Obviously it will work the way you have it, the cap will just help the symmetry of the output stage.

[[  And as far as the electrolytic in series with R1 goes (make the offset voltage lower), you're overlooking R2, which guarantees zero offset as well as setting the gain for the amplifier. The desire to add reactive components (like capacitors) to the signal chain is very common, but it's not advisable. ]]

What determine the DC offset voltage of the amplifier is the relation between R1//R2 and R3 (being R1 much lower than R2, is R1 the dominant between the two).

Curious to know how many mV offset voltage you get with your values... 

[Gall]

The quality of this circuit depends solely upon the quality of the opamp. It compensates everything except for high frequency phase delay that may cause RF oscillations if not compensated properly. A small capacitor in parallel to R1 will help if the circuit oscillates. Try to get the best existing opamp. Especially important are open-loop gain and good common mode and power rejection ratio. A large common-mode input voltage would make the circuit more RF stable. In any case, do not try to make the gain ((R1+R2):R1) less than 100. Anything like 1k:100k should be fine. An extra resistor between the opamp output and its inverting input will increase stability. This should be in order of 100xR2 or more, resulting in open-loop gain of 10000.

[Whuffo]

- A 1KOhm resistor placed between C1 and the op-amp + input will help to not occur oscillations when the pot is at minimum, or maximum (in this last case with a cable connected at the input but not connected to the source).

The resistor to ground at C1 is not 1K, it's 100K - and it's there to properly set the differential input to the opamp without loading the input too much. It also acts as part of the input voltage divider.

Please reread what I have typed about the 1k resistor. It have nothing to do with the 100k biasing input resistor.

[[ Adding an electrolytic to the biasing chain is valueless; the diodes in the chain are in saturation at all times. They're used to develop the correct bias voltage and have no effect on the AC performance]]

You are forgetting the variable resistor in series with the diodes. Obviously it will work the way you have it, the cap will just help the symmetry of the output stage.

[[  And as far as the electrolytic in series with R1 goes (make the offset voltage lower), you're overlooking R2, which guarantees zero offset as well as setting the gain for the amplifier. The desire to add reactive components (like capacitors) to the signal chain is very common, but it's not advisable. ]]

What determine the DC offset voltage of the amplifier is the relation between R1//R2 and R3 (being R1 much lower than R2, is R1 the dominant between the two).

Curious to know how many mV offset voltage you get with your values...

The DC offset is indeed determined by those resistors as well as the quality of the opamp. Note that R1 and R3 are connected to ground; the feedback resistor R2 nulls the output offset to zero. And that's what I've measured - this circuit works fine.

The variable resistor in series with the diodes is a fine tuning adjustment for the bias current; the bulk of the bias voltage is developed across those germanium diodes and the additional amount across the potentiometer is minimal. There's 48 volts across the bias chain, and less than 200 millivolts of it is across that potentiometer. Is the drive perfectly symmetrical? Not quite - but notice that the entire amp is enclosed in the feedback loop. The feedback takes care of any tiny errors here.

And while I understand what you're saying about that 1K resistor - it's not necessary here. This amp has been tested with both 535 and 5535 opamps and was found to be completely stable. There's some subtle bandwidth limiting in the circuit - it's not obvious, but it keeps the amp from doing strange things at frequencies we're not trying to amplify. Build one and give it a try; you'll be impressed.

[Whuffo]

A few additional comments:

- VR1 should probably be 10K log. Otherwise the presence of R3 will skew the potentiometer curve to a noticeable degree.

- No high frequency compensation? Maybe a tiny cap across R2, or use a NE5534. The latter would also offer a convenient location for an output null adjustment, if needed. I would worry about potential high frequency oscillations at certain loads without one.

- C4 should be two 0.1 uF caps, one from each rail to ground, in parallel with the 5 (4.7?) uF 'lytics. As C4 is currently located, it will not be able to help supply high frequency current spikes to the output. (Or rather, it will draw them through the 'lytic on the opposing power rail).

- Would using the classic 'active' BJT bias network in place of D1-D3 + VR2 reduce thermal drift of the DC operating point?

The selection of VR1 and associated input components is part of the bandwidth limiting and the component values are correct for their intended purpose. Is the curve of the pot skewed? A little, but it's a "set and forget" control in most applications.

Yes, there's no compensation capacitor. It's not necessary here due to bandwidth limiting. And C4 is intentional; it's there to stabilize the power inputs to the opamp. During operation, the current input varies dymamically and individually; C4 keeps the DC level on the output stable during these excursions. Supply of peak power to the opamp comes from the electrolytics - but keep in mind that the opamp doesn't supply much output power, that's developed in the transistor stages. If you want to add capacitors to ground, that won't hurt. But don't delete C4; it's needed to keep the amp stable.

The bias network is intended to provide very fine control of the idling current in the output devices in a simple / inexpensive way. As far as thermal drift of the DC operating point goes, please note that the feedback network through R2 controls this voltage and keeps it at ground potential. The "classic active" biasing solution is good for mass produced items - my design does require that each amp be individually "tuned". But this biasing method provides far lower distortion than those mass produced amps.

Regarding high frequency stability; consider the phase shift from input to output on this amp. Then take a look at the Zoebel network on the output and you'll see how it's stabilized. That's not according to "the book" but it's very effective.

[Whuffo]

The quality of this circuit depends solely upon the quality of the opamp. It compensates everything except for high frequency phase delay that may cause RF oscillations if not compensated properly. A small capacitor in parallel to R1 will help if the circuit oscillates. Try to get the best existing opamp. Especially important are open-loop gain and good common mode and power rejection ratio. A large common-mode input voltage would make the circuit more RF stable. In any case, do not try to make the gain ((R1+R2):R1) less than 100. Anything like 1k:100k should be fine. An extra resistor between the opamp output and its inverting input will increase stability. This should be in order of 100xR2 or more, resulting in open-loop gain of 10000.

I'd also recommend looking for a high slew rate in your opamps intended for audio amplification.

Moving on to the other comments about stability: the extra resistor between the opamp output and its inverting input would probably make this amp unstable. Think of the entire amp as a big power opamp. If you do that, then the feedback resistor and other features will make better sense. Stability is guaranteed by bandwidth limiting; if you build this amplifier as per the schematic, it will not oscillate.

Your comments about minimum gain are appreciated, but they don't work well for this application. Taking a "line level" audio signal and boosting it up to 30 watts puts some real limits on the possibilities. The difference between the 0.707 volt AC input and the 48 volt AC output doesn't allow for the kind of opamp gains you suggest. And if it did, affordable opamps can't produce that kind of output level.

For what it's worth, this isn't some partly baked idea - this is a valid and proven design. Many people are listening to good music being amplified by amps built from this same circuit. If some wish to "roll their own" I wish them the best of luck and hope they're successful. For those who just want to build an amplifier that works very well, here's your design.

[dannyf]

if you build this amplifier as per the schematic, it will not oscillate.

I am not quite sure.

The use of a CFP pair here is necessary due to the opamp front end: you need additional voltage gain in the ops. Unfortunately, CFP ops is difficult to stablize, especially with a global feedback loop. Your saving grace here is the output devices used - they are very slow and that helped with the overall stability.

If someone had built it with a modern output device (1943/5200 or 15P/N), it sure will oscillate.

To be safe, I would put some miller capacitance there or follow the suggestions from one of the posters by using a 5534.

[Whuffo]

if you build this amplifier as per the schematic, it will not oscillate.

I am not quite sure.

The use of a CFP pair here is necessary due to the opamp front end: you need additional voltage gain in the ops. Unfortunately, CFP ops is difficult to stablize, especially with a global feedback loop. Your saving grace here is the output devices used - they are very slow and that helped with the overall stability.

If someone had built it with a modern output device (1943/5200 or 15P/N), it sure will oscillate.

To be safe, I would put some miller capacitance there or follow the suggestions from one of the posters by using a 5534.

That's OK, you can doubt all you wish - it won't make me unhappy or invalidate this design. The components weren't chosen because they're old, they're chosen because they're cheap. And it was designed to use a minimum number of inexpensive components; this is what good engineering is all about. The output devices were chosen on purpose; yes, they're slow and have low gain - and that isn't always a bad thing. Each component was chosen not for it's obvious characteristic, but also for the "less well known" characteristics. The whole is more than the sum of its parts.

Saying that it would certainly oscillate with a "more modern" output device only indicates that you didn't really consider the schematic. There's more going on than a quick glance would indicate. I made my living doing this kind of stuff, and this is one of my favorite designs. I was sufficiently successful at these things to retire at 50 to a tropical island. Build one; it'll only take a few hours. Then test it - after you've done this we can discuss the performance. Until that happens, there's an imbalance here; I know it works, and you're just speculating. Build one, get the experience, then let's talk about it.

[ElectroIrradiator]

Which opamp will substitute for the 5535? They don't seem to be available new any longer...

[c4757p]

It's a decent op amp but not particularly special. It does have quite a low input noise current. How about OPA2134? (That's a dual package, just like the 5535, but if you only need one there is the OPA134) Once you wade through the audiophool crap (where's the barfing emoticon?) it's pretty decent. Noise voltage is a bit high, but still less than the 5535, and the noise current is two orders of magnitude lower.

[dannyf]

Until that happens, there's an imbalance here; I know it works, and you're just speculating.

I think that's something that we can agree on: there is an imbalance: you claim that it works; and you have shown zero verifiable evidence that it works.

[c4757p]

It doesn't look particularly unstable to me. The voltage gain in the output section is small and controlled (about 2-4 depending on load). It might not care for a capacitive load, but I'm sure it can handle resistive and probably inductive as well.

I don't have the time to build the circuit right now, but I just tried it in SPICE with some quite fast parts (2N2222, 2N2907, D45H11, D44H11 and LM7171) and it had no problems. There are no anomalies in the frequency response, no hints of instability, nothing. Obviously YMMV in real life.

For noise purposes, you might want to consider reducing the 330k/10k divider to 33k/1k.

For (our) sanity purposes, you might want to consider putting component values in the schematic rather than off to the side.

[Gall]

Moving on to the other comments about stability: the extra resistor between the opamp output and its inverting input would probably make this amp unstable. Think of the entire amp as a big power opamp. If you do that, then the feedback resistor and other features will make better sense. Stability is guaranteed by bandwidth limiting; if you build this amplifier as per the schematic, it will not oscillate.

In most cases such a resistor increases stability. The idea behind this is simple: decrease the overall phase shift by compensating the opamp phase shift with a local feedback. This moves the pole on Bode plot away from working area. If the entire circuit is a "big opamp", the extra resistor is like an internal compensation circuit of every modern opamp.

Your comments about minimum gain are appreciated, but they don't work well for this application. Taking a "line level" audio signal and boosting it up to 30 watts puts some real limits on the possibilities. The difference between the 0.707 volt AC input and the 48 volt AC output doesn't allow for the kind of opamp gains you suggest. And if it did, affordable opamps can't produce that kind of output level.

Then it's better to attenuate the input. The usual stability problem is caused by negative feedback being too heavy, resulting in high common-mode signal at the opamp input. At higher frequencies due to phase shifts this turns into a positive feedback. The simpliest recipe against it is to make the gain higher. This also makes the opamp work at smaller signal and improves the overall linearity. An opamp-like circuit is likely to oscillate at gain=1. Many circuits are stable only at gain>=10. Since 0.7V is high enough, the noise is not an issue even if the input signal is attenuated.

If you want to build this circuit "as is", then DO NOT change the opamp. Using a "better" opamp without changing anything else may be a very very bad idea. Its open-loop gain may be too high and so the bandwidth limit may be not enough for a certain phase shift. Also its internal bandwidth limit will not help at all, modern opamps are way too fast. This may have catastrophic result.

[Whuffo]

It doesn't look particularly unstable to me. The voltage gain in the output section is small and controlled (about 2-4 depending on load). It might not care for a capacitive load, but I'm sure it can handle resistive and probably inductive as well.

I don't have the time to build the circuit right now, but I just tried it in SPICE with some quite fast parts (2N2222, 2N2907, D45H11, D44H11 and LM7171) and it had no problems. There are no anomalies in the frequency response, no hints of instability, nothing. Obviously YMMV in real life.

For noise purposes, you might want to consider reducing the 330k/10k divider to 33k/1k.

For (our) sanity purposes, you might want to consider putting component values in the schematic rather than off to the side.

The capabilities of the program I used to generate that schematic weren't up to putting component values on the diagram. For what it's worth, it was done in Visio; not the most appropriate tool, but it was close at hand. Many schematics in my files are on paper (often on the back of a Denny's placemat) and I need to find a better way to put them online.

The selection of the 330K/10K resistances wasn't arbitrary - and if you use metal film resistors per the schematic the noise will be largely minimized. In the last one I built, I used 2N2905 and 2N2907 transistors for the drivers; 2N2222 / 2N2219 would run a little hotter than I'd like here. There's opportunity for different semiconductors and opamps here; as long as they're reasonably compatible the circuit will work as designed. Global feedback is a wonderful thing, you know.

I just hope that the people here who are interested in audio amplifiers give this circuit a chance. The performance is better than they think, and 30 watts per channel is enough for most home stereo systems. I used a stereo version of this at home for many years; these days I use a 100 watt version - but that design is encumbered and can't be shared. I've got seven channels of the 100w version - plus a subwoofer powered by the same amp in my home theater system.

[c4757p]

The capabilities of the program I used to generate that schematic weren't up to putting component values on the diagram. For what it's worth, it was done in Visio; not the most appropriate tool, but it was close at hand. Many schematics in my files are on paper (often on the back of a Denny's placemat) and I need to find a better way to put them online.

If you're going to continue to use this program, why can't you just use the same feature you used to do the "Test Point" label?

Your schematics are probably the tastiest thing to ever grace a Denny's placemat.

The selection of the 330K/10K resistances wasn't arbitrary - and if you use metal film resistors per the schematic the noise will be largely minimized.

Thermal noise from the resistors, yes, but the op amp's input current noise will be amplified by the high resistance. I figured this was a concern to you since you chose an op amp with such a low current noise. Nearby EM fields (think: mains hum) will also be picked up more by high impedance nodes, though that's pretty easy to avoid, so not much of a concern here if you do the layout right.

And if it wasn't arbitrary, then why was it chosen? The only reason I can see would be to match the impedance at the noninverting input, but you're not even close to having matched that (and using a modern FET-input op amp makes that largely pointless at low frequency anyway).

[Whuffo]

Moving on to the other comments about stability: the extra resistor between the opamp output and its inverting input would probably make this amp unstable. Think of the entire amp as a big power opamp. If you do that, then the feedback resistor and other features will make better sense. Stability is guaranteed by bandwidth limiting; if you build this amplifier as per the schematic, it will not oscillate.

In most cases such a resistor increases stability. The idea behind this is simple: decrease the overall phase shift by compensating the opamp phase shift with a local feedback. This moves the pole on Bode plot away from working area. If the entire circuit is a "big opamp", the extra resistor is like an internal compensation circuit of every modern opamp.

Your comments about minimum gain are appreciated, but they don't work well for this application. Taking a "line level" audio signal and boosting it up to 30 watts puts some real limits on the possibilities. The difference between the 0.707 volt AC input and the 48 volt AC output doesn't allow for the kind of opamp gains you suggest. And if it did, affordable opamps can't produce that kind of output level.

Then it's better to attenuate the input. The usual stability problem is caused by negative feedback being too heavy, resulting in high common-mode signal at the opamp input. At higher frequencies due to phase shifts this turns into a positive feedback. The simpliest recipe against it is to make the gain higher. This also makes the opamp work at smaller signal and improves the overall linearity. An opamp-like circuit is likely to oscillate at gain=1. Many circuits are stable only at gain>=10. Since 0.7V is high enough, the noise is not an issue even if the input signal is attenuated.

If you want to build this circuit "as is", then DO NOT change the opamp. Using a "better" opamp without changing anything else may be a very very bad idea. Its open-loop gain may be too high and so the bandwidth limit may be not enough for a certain phase shift. Also its internal bandwidth limit will not help at all, modern opamps are way too fast. This may have catastrophic result.

If you want to try the additional feedback, then go for it. I didn't find it useful, but maybe your implementation will be a win. Regarding gain; there's a fixed amount of input noise at the opamp. If you attenuate the input to keep the gain down, you'll be seeing more of that input noise in the final signal. Best to keep the input hot and the gain down. You can play with this if you want, but there's nothing to gain here.

And as far as changing opamps - as long as they're internally compensated then they'll work fine. The bandwidth limiting comes from the surrounding circuitry, and the gain is set by the feedback network. Make sure you've got a high slew rate opamp and that the voltages are compatible and you'll be fine. I've used 535 and 5535 opamps because they're cheap and readily available; if you want to spend more and get the very best it'll be OK. It might not improve the performance, but it'll be OK.

[Lightages]

As far as the schematics go, a very simple easy to use and free one is:
http://sourceforge.net/apps/mediawiki/tinycad/index.php?title=TinyCAD
It is easier to learn than the full blown schematic/circuit board program.

[c4757p]

I've used 535 and 5535 opamps because they're cheap and readily available

Do you have a large bin of NE5535 in your garage - that's what "cheap and readily available" means? Because from here, they look to be many years obsolete. The only place I could find them was eBay.

[Whuffo]

I've used 535 and 5535 opamps because they're cheap and readily available

Do you have a large bin of NE5535 in your garage - that's what "cheap and readily available" means? Because from here, they look to be many years obsolete. The only place I could find them was eBay.

Yeah, I've got a good supply of them in my parts cabinets. But don't despair; any old high slew rate opamp which is internally compensated will work. Low noise is a plus, but keep the slew rate high. It could be worse - I could have put a 741 on the schematic.

[ElectroIrradiator]

Do you have a large bin of NE5535 in your garage - that's what "cheap and readily available" means? Because from here, they look to be many years obsolete. The only place I could find them was eBay.

...and those are from our friends in the far east, who by magic are able to supply every obsolete components you could possibly want.

Hmm, doesn't look like I have any opamps, which beats a 5535 in every way. A TL072 is close, though the slew rate is slightly lower.

Edit: 1N914 are not Germanium diodes, just plain Silicon.

[madires]

Yesterday I fixed two old amps (40W with 2N3055s) with some bad caps and RF oscillation on the negative side. The oscillation was about 10MHz and started at a specific input level. The driver for the positive side got a small ceramic cap between base and collector of the end stage driver PNP but there was none for the negative side. A 470pF ceramic between the base of the lower end stage and negative rail fixed the oscillation problem. Must have been a design error? 

[dannyf]

I don't have the time to build the circuit right now, but I just tried it in SPICE with some quite fast parts (2N2222, 2N2907, D45H11, D44H11 and LM7171) and it had no problems. There are no anomalies in the frequency response, no hints of instability, nothing. Obviously YMMV in real life.

I sim'd it in LTSpice (and orcad) and both showed instability - the gain peaked at around 500Khz. With square wave output, considerable pinging on the output (slew rate limited to about 4.5v/us, but that's dependent on the devices used). A 220ohm/22p pair around the opamp cures that.

For noise purposes, you might want to consider reducing the 330k/10k divider to 33k/1k.

thermal noise really isn't an issue for audio amplifiers (line-level). 330k/10k is in my view the right combination as it allows the use of a 10k resistor from the non-inverting end to ground (the original schematic used a 100k resistor and that's clearly wrong).

I would point out a few inconsistencies (to be polite) in the design:

1) I would avoid the use of CFP at all costs. They are very difficult to handle, even for seasoned designers. A follower OPS is much easier in my view;
2) I would put a small resistor on the collectors of the output devices to provide some thermal stability, particularly when you are using diodes to set the bias. Adjust VR2 to drop about 25mv on that resistor;
3) The drivers idle at about 1.5ma - way too low for such devices, particularly if you are driving 3055/2955 at high current levels - those guys' hFE can droop to 10x or so. I would use at least BD139/140 or some Motorolla devices for that - but then they have much higher fT and that exposes the instability issue;
4) the thermal compensation portion is poorly designed: the output devices want the drivers idle at 1.5ma; The diodes want the drivers idle at 3ma at least, with VR2 shorted. The end result is that they idle current on the output devices will be pushed to a very high level to balance out the two. R9/R10 should be at least 1/2 of R8/R11;
5) R6/R7 drop considerable current and I would use 1/2 w resistors to be safe.
6) the peak current from the opamp can be as high as 15ma -> that limits your choices of opamps used here.
7) the max output power from one pair of output devices is ~30w into an 8ohm load - assuming no SOA issues with the ops. Because of the cfp configuration, it is harder to parallel output devices.
at the max current output, considerable voltage drop happens on R4/R5 (1.5k now), not counting the current needed for the zeners. I would use no more than 110ohm there.
9) if i were to design it, I would use the same transistors as Q1/Q2 for D1/D2 to match the thermal characteristics.
10) idling the OPS at 100ma is on the high side in my view and runs the risk of gm doubling. I would do no more than 75ma, corresponding to a resistor of 0.22 - 0.47ohm on the collectors of the output devices.

I think you will generally find that opamp based audio amps don't perform well.

and those are from our friends in the far east, who by magic are able to supply every obsolete components you could possibly want.

Those are typically remasked TL072 or 5532.

The oscillation was about 10MHz and started at a specific input level.

When clipping takes place, ...

3055's are quite helpful in those cases due to their poor fT. A 1943/5200 would have done some serious damage.

[dannyf]

Just to add one more point: I would definitely add a small resistor (<1k) in serial with the input capacitor - never let your base / gate see the input signal directly.

[GK]

1) I would avoid the use of CFP at all costs. They are very difficult to handle, even for seasoned designers. A follower OPS is much easier in my view;
2) I would put a small resistor on the collectors of the output devices to provide some thermal stability, particularly when you are using diodes to set the bias. Adjust VR2 to drop about 25mv on that resistor;
3) The drivers idle at about 1.5ma - way too low for such devices, particularly if you are driving 3055/2955 at high current levels - those guys' hFE can droop to 10x or so. I would use at least BD139/140 or some Motorolla devices for that - but then they have much higher fT and that exposes the instability issue;
4) the thermal compensation portion is poorly designed: the output devices want the drivers idle at 1.5ma; The diodes want the drivers idle at 3ma at least, with VR2 shorted. The end result is that they idle current on the output devices will be pushed to a very high level to balance out the two. R9/R10 should be at least 1/2 of R8/R11;

Resistors will have to be put in series with the emitters, not collectors, of the power output devices to have any significant effect on the bias current stability. Being a CFP "with gain" it's likely not as touchy (loop oscillation local to the power output stage) as a unity gain CFP and those large value (470 ohm) B-E resistors for the power output devices also likely aid stability as they form a compensatory pole (in conjunction with the transistor input capacitance) in the open loop response of the CFP. Quiescent current of the power output devices is essentially governed by beta multiplication of the driver transistor quiescent current. Beta of the power output devices varies significantly with junction temperature, so the dynamic quiescent current stability would be mediocre.

The power output stage also has no implicit protection circuitry. Beta of the power output transistors specified drops of rapidly as Ic increases. This, combined with the weak and under biased driver stage, is what stops this thing from immediately blowing up when the output is shorted - the driver stage just doesn't have enough guts to force the power output transistors to conduct heavily enough to blow up straight away. I do not think that this makes the design a notably sophisticated one however. The open loop linearity of the CFP is in part a product of the (grossly non-linear) power output transistor beta. The low Iq driver stage will not have adequate capability to rapidly switch off the output transistors. The output stage will therefore suffer from significant cross conduction at 20kHz and generally poor HF linearity. A simpler, unity gain "double emitter follower" power output stage would offer much better linearity than the output stage circuitry presented here, but that wouldn't be compatible with the op-amp front-end as the output voltage swing would be limited to that of the op-amp.
Even so 30W into 8 ohms isn't a very demanding specification the THD claims ("well under 0.01%") for the circuit as presented are definitely not credible - particularly so at 30kHz. For THD performance targets that are achievable with CFP power output stages using single pairs of old and slow power transistors like the 2955/3055 pair, the early editions of Douglas Selfs "Power Amplifier Design Handbook" are good references.


 

[Whuffo]

Here's an updated copy of the schematic; the gain control wiring has been corrected, and the diode part number has been corrected. Those were napkin to jpg transcription errors; sorry about that.

I've been reading through the comments posted and I've got to say it's been very informative. I see opinions masquerading as facts being spouted by people who clearly do not understand how this amplifier works.

It is what it is; a very clean amplifier that's simple enough to put together on the kitchen table. Old school parts in places - because they do the job and are cheap. Enough output power for a home stereo system. If you want to substitute the opamp or driver transistors, feel free - the last of these I built used 2N2905 / 2N2907 for the drivers. I'd keep the same output transistors, though - they're very well suited for the job.

For those who have batted the input resistance issue around the block a few times - electrolytic capacitors in the signal path are evil. By keeping the input resistance high, a 0.1uf input coupling capacitor can be used. Does this increase the opamp's input noise? Yes, but it's swamped by the input signal so it's not really a problem at all. And as a bonus, the input capacitor / resistance provides a HPF at the input.

Those who say opamp based audio amplifiers aren't much good are mostly correct. There's been far too many misbegotten examples, but that doesn't mean they're all bad. Sometimes things just work and the whole is greater than the sum of the parts. This is an example of what happens when it goes right. No, it didn't just happen by chance, many hours / days were spent testing various variations and permutations until a stable, minimum cost design was finally arrived at.

[Whuffo]

It's a decent op amp but not particularly special. It does have quite a low input noise current. How about OPA2134? (That's a dual package, just like the 5535, but if you only need one there is the OPA134) Once you wade through the audiophool crap (where's the barfing emoticon?) it's pretty decent. Noise voltage is a bit high, but still less than the 5535, and the noise current is two orders of magnitude lower.

Sounds like a great choice!

[Whuffo]

Do you have a large bin of NE5535 in your garage - that's what "cheap and readily available" means? Because from here, they look to be many years obsolete. The only place I could find them was eBay.

...and those are from our friends in the far east, who by magic are able to supply every obsolete components you could possibly want.

Hmm, doesn't look like I have any opamps, which beats a 5535 in every way. A TL072 is close, though the slew rate is slightly lower.

Edit: 1N914 are not Germanium diodes, just plain Silicon.

Thanks for catching that diode part number error. 1N34 is the right number; I just posted an updated schematic. You might be OK with a TL072; it'd increase the high frequency distortion a bit, though.

[Whuffo]

Just to add one more point: I would definitely add a small resistor (<1k) in serial with the input capacitor - never let your base / gate see the input signal directly.

Look at the updated schematic; there was a transcription error around the input gain control.

[c4757p]

If the volume is turned up all the way, the op amp's input still sees the signal directly. That 100nF capacitor is damn near a short for anything above 20kHz or so. Add a resistor so that ESD on the input can be gently discharged by the input clamp diodes.

[dannyf]

electrolytic capacitors in the signal path are evil.

Poorly designed amplifiers are evil. Electrolytiic capacitors aren't.

Look at the updated schematic;

The same issue persists: you don't seem to understand why it is needed.

As to input impedance, 47K is probably near the upper end: anything above that may introduce more trouble than it is worth.

No, noise isn't the issue: output DC offset is.

[dannyf]

I will give you two examples of beautifully used electrolytic capacitors: the bootstrap capacitors in JLH1969 and AKSA amps. In the case of the former, an ideal capacitor would have meant loads of troubles.

[c4757p]

electrolytic capacitors in the signal path are evil.

Poorly designed amplifiers are evil. Electrolytiic capacitors aren't.

Indeed - in fact, I have found electrolytics in general to be rather handy and decent little things, as long as you understand how they behave. They are, of course, not ideal capacitances, but you're not going to get very far in electronics assuming all your components are physical manifestations of ideal passive properties anyway. And they're just fine for audio apps, again, if you use them correctly. The only real problem that they may cause in the signal path is temperature-dependent leakage, causing bias points to drift - easy to counter in many places by reducing the impedance of the bias point, or even using an active bias servo if you really must.

[Whuffo]

1) I would avoid the use of CFP at all costs. They are very difficult to handle, even for seasoned designers. A follower OPS is much easier in my view;
2) I would put a small resistor on the collectors of the output devices to provide some thermal stability, particularly when you are using diodes to set the bias. Adjust VR2 to drop about 25mv on that resistor;
3) The drivers idle at about 1.5ma - way too low for such devices, particularly if you are driving 3055/2955 at high current levels - those guys' hFE can droop to 10x or so. I would use at least BD139/140 or some Motorolla devices for that - but then they have much higher fT and that exposes the instability issue;
4) the thermal compensation portion is poorly designed: the output devices want the drivers idle at 1.5ma; The diodes want the drivers idle at 3ma at least, with VR2 shorted. The end result is that they idle current on the output devices will be pushed to a very high level to balance out the two. R9/R10 should be at least 1/2 of R8/R11;

Resistors will have to be put in series with the emitters, not collectors, of the power output devices to have any significant effect on the bias current stability. Being a CFP "with gain" it's likely not as touchy (loop oscillation local to the power output stage) as a unity gain CFP and those large value (470 ohm) B-E resistors for the power output devices also likely aid stability as they form a compensatory pole (in conjunction with the transistor input capacitance) in the open loop response of the CFP. Quiescent current of the power output devices is essentially governed by beta multiplication of the driver transistor quiescent current. Beta of the power output devices varies significantly with junction temperature, so the dynamic quiescent current stability would be mediocre.

The power output stage also has no implicit protection circuitry. Beta of the power output transistors specified drops of rapidly as Ic increases. This, combined with the weak and under biased driver stage, is what stops this thing from immediately blowing up when the output is shorted - the driver stage just doesn't have enough guts to force the power output transistors to conduct heavily enough to blow up straight away. I do not think that this makes the design a notably sophisticated one however. The open loop linearity of the CFP is in part a product of the (grossly non-linear) power output transistor beta. The low Iq driver stage will not have adequate capability to rapidly switch off the output transistors. The output stage will therefore suffer from significant cross conduction at 20kHz and generally poor HF linearity. A simpler, unity gain "double emitter follower" power output stage would offer much better linearity than the output stage circuitry presented here, but that wouldn't be compatible with the op-amp front-end as the output voltage swing would be limited to that of the op-amp.
Even so 30W into 8 ohms isn't a very demanding specification the THD claims ("well under 0.01%") for the circuit as presented are definitely not credible - particularly so at 30kHz. For THD performance targets that are achievable with CFP power output stages using single pairs of old and slow power transistors like the 2955/3055 pair, the early editions of Douglas Selfs "Power Amplifier Design Handbook" are good references.


 

I like Doug Self's work - but there's other ways to do things and he's got his opinions like the rest of us do. Anyway, the "mediocre quiescent current stability" isn't quite as bad as you fear, and the small variations here don't affect the performance of the circuit. My "solution" to this problem was to run a little extra quiescent current so that any possible variations would still be in the linear region of the transistor curve.

I appreciate your respect for the difficulty of CFP designs; you're right, they can be a problem. That is the reason I'm so proud of this little circuit; good power from few and inexpensive parts, stable and clean. And while you say there's no explicit output transistor protection, you follow by describing how they're protected. It's not a bug, it's a feature. For fun, consider how if the amp is pushed to max without adequate cooling, the decrease in beta in the outputs will reduce the power consumption automatically. People do some strange things to their amplifiers; short output terminals, hook up multiple speakers in parallel, etc. A good design should be able to handle these misadventures without letting the magic smoke out. Doing it without added circuitry is a win.

There was a lot of time spent working out how to accomplish all the design goals using as little circuitry as possible. So things like the output protection, bandwidth limiting, etc. aren't obvious - but they are in there.  And it does meet the quoted specs easily - try one and see.

[Whuffo]

Yesterday I fixed two old amps (40W with 2N3055s) with some bad caps and RF oscillation on the negative side. The oscillation was about 10MHz and started at a specific input level. The driver for the positive side got a small ceramic cap between base and collector of the end stage driver PNP but there was none for the negative side. A 470pF ceramic between the base of the lower end stage and negative rail fixed the oscillation problem. Must have been a design error?

Probably so. Hanging capacitors around the outputs and drivers to suppress oscillations isn't the optimum design solution.

[c4757p]

Yesterday I fixed two old amps (40W with 2N3055s) with some bad caps and RF oscillation on the negative side. The oscillation was about 10MHz and started at a specific input level. The driver for the positive side got a small ceramic cap between base and collector of the end stage driver PNP but there was none for the negative side. A 470pF ceramic between the base of the lower end stage and negative rail fixed the oscillation problem. Must have been a design error?

Probably so. Hanging capacitors around the outputs and drivers to suppress oscillations isn't the optimum design solution.

Huh? You're going on about how good it is that you picked slow transistors, etc., because their "inferior" properties make them a better fit for the application, allowing you to make something that works well without putting lots of effort into stability and such. So, how is that any different from adding a capacitor to slow down the transistors and get the same effect?

[Whuffo]

If the volume is turned up all the way, the op amp's input still sees the signal directly. That 100nF capacitor is damn near a short for anything above 20kHz or so. Add a resistor so that ESD on the input can be gently discharged by the input clamp diodes.

R3 on the schematic is strangely suggestive...

[c4757p]

R3 on the schematic is in parallel with the input, and is quite large. An ESD event will fuck the diodes in the op amp and pass right over the resistor. You need one in series.

[Whuffo]

electrolytic capacitors in the signal path are evil.

Poorly designed amplifiers are evil. Electrolytiic capacitors aren't.

Look at the updated schematic;

The same issue persists: you don't seem to understand why it is needed.

As to input impedance, 47K is probably near the upper end: anything above that may introduce more trouble than it is worth.

No, noise isn't the issue: output DC offset is.

Electrolytic capacitors are non-linear devices by design and construction. Leaving them out of the signal path is always a good idea.

And as to input impedance, look at the input circuit again. 100K pot and a 100K resistor - both connected to ground and as you note, C1 is essentially a short at audio frequencies. So the input impedance is 50K; the general design spec for line level circuits specifies a 47K impedance. This is close enough. Also note that as the input frequency drops below the passband, the Xc of C1 increases. This helps keeps "rumble" out of the amp; it's a Pi section RC HPF.

And output DC offset is a non issue. Really; both inputs of the opamp are referenced to ground. The global feedback is also referenced to ground. The only thing that could drive the output DC point off of zero would be a DC input - and those are blocked by C1. Rather than try to force the opamp / bias chain to be perfect, the solution is to let the global feedback correct any small errors. That's proven to be a very successful solution.

[madires]

Yesterday I fixed two old amps (40W with 2N3055s) with some bad caps and RF oscillation on the negative side. The oscillation was about 10MHz and started at a specific input level. The driver for the positive side got a small ceramic cap between base and collector of the end stage driver PNP but there was none for the negative side. A 470pF ceramic between the base of the lower end stage and negative rail fixed the oscillation problem. Must have been a design error?

Probably so. Hanging capacitors around the outputs and drivers to suppress oscillations isn't the optimum design solution.

But it was quite obvious where the oscillation started and the scope confirmed that. First I tried to use a cap with the same value as used for the PNP driver (220pF) and it wasn't enough to suppress the oscillation. With 470pF the oscillation was gone and I've also checked if high input frequencies (30-50kHz) pass the amp without major distortions.

[Whuffo]

Yesterday I fixed two old amps (40W with 2N3055s) with some bad caps and RF oscillation on the negative side. The oscillation was about 10MHz and started at a specific input level. The driver for the positive side got a small ceramic cap between base and collector of the end stage driver PNP but there was none for the negative side. A 470pF ceramic between the base of the lower end stage and negative rail fixed the oscillation problem. Must have been a design error?

Probably so. Hanging capacitors around the outputs and drivers to suppress oscillations isn't the optimum design solution.

Huh? You're going on about how good it is that you picked slow transistors, etc., because their "inferior" properties make them a better fit for the application, allowing you to make something that works well without putting lots of effort into stability and such. So, how is that any different from adding a capacitor to slow down the transistors and get the same effect?

The parts were chosen to do the job they are required to do; there's no need to use a sledgehammer to drive tacks. If using those appropriate components means I don't have to deal with problems caused by that curious "bigger / faster is better" mindset, then that's a win. And you have no idea how much effort went into achieving stability - and doing it with a simple and cheap design. That's what real engineering is all about; designs that fulfill the goals and do so as cheaply and simply as possible.

Going back after the design is complete and tacking capacitors in here and there is a sign of lazy design; they should have gotten it right before they went to production.

[Whuffo]

R3 on the schematic is in parallel with the input, and is quite large. An ESD event will fuck the diodes in the op amp and pass right over the resistor. You need one in series.

If you have an ESD event in a line level connection between the preamp / mixer and this amp - then you're going to have much worse problems than a popped opamp. Rather than include meteorite shielding "just in case", remember the old saying about "horses for courses" and try not to overdesign things.

[c4757p]

Why? Never heard of someone touching a connector after walking across the room? And 1k or so in series is hardly overdesigning.

[Whuffo]

I will give you two examples of beautifully used electrolytic capacitors: the bootstrap capacitors in JLH1969 and AKSA amps. In the case of the former, an ideal capacitor would have meant loads of troubles.

I'm not saying that electrolytic caps are useless - and I use them quite often. They have some problems, though - short lifespan, non-linear response, and the tolerances are a bit on the wide side. That said, I'll still avoid putting them into signal paths whenever possible; it's an easy way to improve THD on an amp and cheap to implement. That input cap is in a critical place; any errors there due to an electrolytic cap would not be corrected by feedback. Signal to noise is important, and in this amp design anything in front of the opamp is out of the feedback loop.

[Whuffo]

Yesterday I fixed two old amps (40W with 2N3055s) with some bad caps and RF oscillation on the negative side. The oscillation was about 10MHz and started at a specific input level. The driver for the positive side got a small ceramic cap between base and collector of the end stage driver PNP but there was none for the negative side. A 470pF ceramic between the base of the lower end stage and negative rail fixed the oscillation problem. Must have been a design error?

Probably so. Hanging capacitors around the outputs and drivers to suppress oscillations isn't the optimum design solution.

But it was quite obvious where the oscillation started and the scope confirmed that. First I tried to use a cap with the same value as used for the PNP driver (220pF) and it wasn't enough to suppress the oscillation. With 470pF the oscillation was gone and I've also checked if high input frequencies (30-50kHz) pass the amp without major distortions.

I'm glad you found a solution. And I'm wondering what brand / model that amp is; getting 3055 series transistors to oscillate at 10 Mhz is difficult to do.

[Whuffo]

Why? Never heard of someone touching a connector after walking across the room? And 1k or so in series is hardly overdesigning.

OK, when you build yours you can include that 1K resistor; it's only a penny or two. Then, when someone shuffles their feet across the room, pulls out a cable and zaps the inner conductor you'll - well, nevermind. Line level runs through shielded cables, you know.

If people want to improve the design, then they can. If you build what the schematic shows, it'll work fine. Do that, then try your improvements. You just might find something better. Or you may find that you're happy with the base design.

[madires]

I'm glad you found a solution. And I'm wondering what brand / model that amp is; getting 3055 series transistors to oscillate at 10 Mhz is difficult to do.

Me too It's an old circuit from an electronics magazine (late 80ies). One of both stereo amps was worse than the other, but both channels were always the same.

[dannyf]

the solution is to let the global feedback correct any small errors.

You are so not getting it. Pick up an intro-level book on opamp would be helpful.

And 1k or so in series is hardly overdesigning.

It is less about ESD protection but more about signal conditioning: the serial resistor + input capacitance forms a low-pass filter. I learned it from Pass.

The general design philosophy here is so 1980s: high open loop gain and let negative feedback take care of everything.

We have evolved a couple generations from that.

[dannyf]

BTW, whoever has a real life picture of this thing, how it performs on a bench and how it performs driving real life loads and some tough loads, I am all ears (and eyes).

If no, ask yourself why.