[rjm003]

Please Download the schematic.

I've got as far as finishing the circuit values and designing the board, but I'm sure there are still a few things I've missed that should best be sorted out before getting a prototype up and running.

The circuit is designed to run off a 100k volume pot, and be either a line stage or a headphone amplifier. It is a standard non-inverting gain stage (A=15dB) followed by discrete copy of the old LH0002 buffer IC.

The basic idea is that the four transistors in the buffer will be carefully matched to ensure low output offset voltages and low distortion, so that it will not be necessary to include it in the op amp feedback loop.

The rather primitive buffer is inefficient as the driver transistors must pull as much current as the output devices, so the total current is about 3x more. It also means the damping resistors on the output emitters can only be very tiny otherwise the output stage pulls no static current and reverts to class B operation. I am drawn to the minimalist design however.

Comments and opinions sought.

 

[FallenAngel]

I don't think you need BD139/BD140 on the input of that buffer, small-signal like BC327/BC337 would work. You also really don't want those 47R output resistors.

 

[Nemo de Monet]

I've got an almost identical amp here, complete with BD139/BD140 devices, even. (Rather fewer power capacitors, though; mine only has a single 1500uf cap per rail, bypassed by film caps, and one power supply section serves both L+R.)

I strongly suspect you'll like yours.

Line stage? Mine powers bookshelf speakers...

I'd add small 0,1uf decoupling caps on the power pins of each opamp, just because. Also, while you're clearly really enamored of physical symmetry, I strongly suspect you can move R13/R15/R17/R19/R21/C29, and their corresponding parts on the other half, and substantially reduce trace lengths. Oh, and I'd try and play around with Q9/Q11 et al, and move them closer to Q7/Q5 et al, to minimize trace lengths again, especially the one between Q9/Q5/R23...

 

[TzeYang]

by adding two resistors to the emitters of the input transistors, you can move them to Class AB or even Class A.

I personally think starving them from Class A biasing is bad to the diamond buffers. You only need two extra resistors per channel so I don't think it goes against your principles of minimalism.

 

[rjm003]

Thanks for all the helpful ideas and suggestions.

@fallenangel and tzeyang

I had considered adding two resistors to the emitters of the input transistors to increase the bias on the output transistors. I also considered using a smaller set of input transistors so that the BE diode drop in the input transistors naturally pulled a larger current in the output transistors.

Both those ideas are still in play, though for what I want to accomplish (5-10 mA of bias current in the output devices) the efficiency gain from reduced currents in the input stage doesn't feel like it's worth chasing.

Bear in mind that having all the devices the same (including the capacitance multiplier in the power supply) keeps the BOM simple and makes matching devices that much easier, too.

The 47R resistor in series with the output is for when operated as a line stage, it could be reduced or replaced by a jumper when driving headphones.

@Nemo de Monet

And here I was thinking I had a pretty efficient layout! I appreciate your point re the spacing of the transistors, they could be moved up next to each other ... I had deliberately spaced them a little apart to allow for small clip on heatsinks.

Less sure about the resistors. As you can see there is no open space left on the board, so to tighten the traces joining the resistors and op amp would mean bringing the resistors closer the the op amp and pushing the bypass capacitors further away...

I will take another look, however.

The large number of capacitors shown are more about versatility than maxing out the capacitance. The idea was for some of those pads to be substituted with film or tantalum or left out altogether.

Further info on the amp you built would be welcome...

 

[diditmyself]

With an open loop buffer and class B (or AB) biasing, this amp will suffer from very high distortion when driving low impedance phones. The distortion spectrum will be nasty with high levels of odd order harmonics.

 

[rjm003]

Please elaborate. My understanding was that harmonic distortion arises from mismatched devices, and that crossover distortion is not an issue in class AB...

 

[TzeYang]

^P devices will always be mismatched from N devices.

They are not perfect.

And there are plentiful other of parameters to match other than just BETA alone.

 

[bidoux]

Why are you using this way of regulating your power supply ? I think that LM317 and LM337 would be better and easier to build while they are not that much expensive.

 

[peranders]

rjm, do you get any idle current in the last stage? The emitter at Q5 needs some resistance. 2 ohms is also a very low value. Hard to get a temperature stable output stage.

Why don't you include the buffer in the feedback loop?

 

[Nemo de Monet]

Quote:

Originally Posted by rjm003 /img/forum/go_quote.gif And here I was thinking I had a pretty efficient layout! I appreciate your point re the spacing of the transistors, they could be moved up next to each other ... I had deliberately spaced them a little apart to allow for small clip on heatsinks.

I don't think they're even going to get warm, honestly.

Quote:

Less sure about the resistors. As you can see there is no open space left on the board, so to tighten the traces joining the resistors and op amp would mean bringing the resistors closer the the op amp and pushing the bypass capacitors further away...

What bypass capacitors? The (six) 100uf electrolytics? You can very easily lose a couple of those, IMO, and add in some very small ceramic caps instead. I mean, hell, they're just opamp power supply bypass caps, use 0805 SMD parts and you'll free up a huge amount of real estate, allowing you to shorten some of those leads dramatically.

Quote:

Further info on the amp you built would be welcome...

I don't have the schematic anymore, but I basically took the old A47 design, replaced the power supply with a discrete (BD139/BD140) virtual ground, used single opamps for the gain stage, and used this design as the buffer (outside the feedback loop, BTW), using BD139/BD140 devices biased at around 10ma. So, not quite identical to what you're doing, but pretty darned close. I mainly use it as a headphone amp, but it drives small speakers quite well, from +/-9ishV rails.

It's not especially elegant, but it works quite well, IMO.

 

[diditmyself]

Quote:

Originally Posted by rjm003 /img/forum/go_quote.gif Please elaborate. My understanding was that harmonic distortion arises from mismatched devices, and that crossover distortion is not an issue in class AB...

Like stated above I think you should include the buffer inside the loop of the opamp. Also like stated above I think you should have emitter resistors on the input transistors. Crossover distortion is very much an issue in class AB amps. Since this is a wallpowered amp I can't see the reason not to have it class A biased.

I've found that diamond buffers sound better with constant current sources than with resistors. This will increase complexity though.

 

[rjm003]

The only thing that is off the table is putting the buffer inside the feedback loop.

If the diamond buffer sounds far worse than a MOSFET follower for example, then I'd switch to a MOSFET (Szekeres, open loop) stage before I'd consider putting the diamond buffer in the feedback loop to correct for it's lousy performance.

"Bias the buffer into class A."

For a push pull stage, I suppose you mean increase the bias current until the amp runs out of voltage headroom before it leaves class-A operation... I see no need to go that far. All that's required is that the output devices run in class A most of the time, and 5-10 mA is enough to do that. (for a line stage certainly, and for most headphones also)

The LH0002 is rated for 0.1 % open loop distortion at 5V input, +\-12V supplies, and 2 mA bias current - though it is unclear under what load. Match the devices, raise the bias a bit, and it should be good enough open loop, the inherent symmetry of the circuit sees to that.

LM317/LM337? Been there, done that. Have the T-shirt in the closet somewhere. I can do better. But I've been hearing good things about the old capacitance multiplier / zener regulated circuits like this, and I've been wanting to try it out.

Peranders brings up a more serious problem. The resistors on the emitters of the output devices lower the BE voltage, reducing the current in the output stage substantially. Any value more than the ~2 ohms shown is likely to turn off the output transistors completely. Larger values would be desirable for stability, however.

Adding the resistors to the emitters of the driver transistors as well, and increasing the value to about 22 or 47 ohms, would seem like a good idea... the attached image (from John Broskie's Tubecad.) suggests that a additional capacitor might be needed as well in this case.

 

[peranders]

Quote:

Originally Posted by rjm003 /img/forum/go_quote.gif Peranders brings up a more serious problem. The resistors on the emitters of the output devices lower the BE voltage, reducing the current in the output stage substantially. Any value more than the ~2 ohms shown is likely to turn off the output transistors completely. Larger values would be desirable for stability, however. Adding the resistors to the emitters of the driver transistors as well, and increasing the value to about 22 or 47 ohms, would seem like a good idea... the attached image (from John Broskie's Tubecad.) suggests that a additional capacitor might be needed as well in this case.

Please notice that you must have resistors at the driver transistors also in order to get a stable idle current at the output transistors. Right now your design will be very sensitive against variations of different parameters.

You can find inspiration here

Check the resistor R9 in my design!

 

[rjm003]

Hi P-A,

I browsed your schematic a couple of times as I was researching this circuit. All told I've seen half a dozen variants of the basic LH0002. As I see it the "improvements" fall into three categories:

1. Addition of JFETs to increase the input impedance.
2. Revising the input stage to allow more flexibility in setting the bias current in the output stage.
3. Replacing the resistors in the input stages with current sources of various levels of complexity to improve linearity.

I read your design as being mostly "3", with a two nested current mirrors instead of just one...

As well as your design, there's Sijosae's, the PPA, the BUF634, LH0033, revised Millett, ... and more. It's hard to tell sometimes whether the project author has actually developed the circuit themselves or just cut and paste the buffer section from somewhere else.

Can you tell I'm getting confused yet?