AB-class amplifier schematic analysis & optimization

[ratatax]

Hello
For my sound generator project I want to add headphones amplifier to it.
I planned initially to use a digital amplifier IC but since we want the ability to drive high impedance headphones (600 ohms) with studio-quality someone told me it would be better to use a classic approach with a ~1W class-AB amplifier, powered with +15V/-15V rails.

EDIT : See more info about my needs & project on my reply post : https://www.eevblog.com/forum/beginners/ab-class-amplifier-schematic-analysis-optimization/msg2425455/#msg2425455

We found this schematic, from http://sound.whsites.net/project113.htm :



The original design is very old-school, using through-hole components and big 100uF caps.



We'd like to modernize it to integrate into the product we're making. No problems, resistors diodes and small caps are easy, we found SMD transistors equivalents to BD193/BD140, but big caps are expensive and take a lot of board space.

I understand the role of some of them in the circuit but I have no idea what C2L, C3L & C4L do.

Also I feel this design is old and may use some overly-big capacitors to compensate for a slow linear power supply, but now it's full SMPS, much faster.

C2L seems in reverse polarity to me. I'd like to avoid any cap bigger than 10uF to stay with small ceramic ones, and maybe modernize other things if you find flaws... Can you help me ?


Edit: after looking at different class-AB amplifiers on google, most of them don't have the caps equivalent to C2L, C3L & C4L. Maybe they are meant to optimize something ?

[Audioguru]

C2 has a very small input offset voltage across it. It is used so that the amplifier does not amplify this DC offset voltage which would cause some DC voltage at the amplifier output. Headphones move the cone to one side if there is DC on them.

The capacitors parallel with the diodes reduce distortion a little.

[dietert1]

The circuit you show is something like a school example. For a real product there need to be:

a) Bandwidth limiting
If there is a SMPS nearby, for example 1 nF caps to Gnd on both inputs of the OpAmp. Then a 39pF parallel to R4 to get a compensated divider. Pulse response needs to be checked for stability.
Also there should be a 20 or 30 Hz high pass somewhere. Modern media include subsonic audio down to 5 or 10 Hz and sound quality will suffer if you pass that to the headphones. C1 could be 470 nF instead and C2 may be reduced to 10 uF as well to get something like a second order filter. That's a matter of taste.

b) Short circuit protection
For example sensing the voltage on the two 10 Ohm output resistors R7 and R8.

c) Turn-on and turn-off transient suppression (load/listener protection)
This can be done with a pair of low Rdson mosfets per channel and a supply voltage supervisor including a small turn on delay.

Two years ago i tested 150 uF multilayer smd caps. You can use them for C3 to C4, which only see small voltages.

Regards, Dieter

[David Hess]

C2 lowers the gain of the operational amplifier at low frequencies to 1 so it does not amplify its own offset.  Because of AC coupling from C1, it never sees significant negative voltages.

C3 and C4 provide a low AC impedance path between the operational amplifier output and the bases of the transistors so it can better control them.

[ratatax]

Ooh I see ! The polarity of C2 confused me since it will see positive and negative voltages, but it's not the first time I see electrolytics used like that, it doesn't seems to matter when the current/voltage is limited enough with a resistor. I was planning to replace them all with 22uF ceramic caps.

@dietert1 how do you calculate the value of 39pF for the capacitor to be used with R4 ?

a) Bandwidth limiting
the DAC output already has a lowpass output filter and the datasheet mentions a DC blocking cap isn't needed, so I guess there is already a high-pass integrated but I'd need to check its cutoff frequency.

b) Short circuit protection => you're right. Since the circuit is ready to drive already high-impedance headphones (600+ ohms), isn't the simplest solution simply to replace the 10 ohm resistors by a higher value ?

c) Turn-on and turn-off transient suppression (load/listener protection) => we need to think about that. I think the easiest solution would be to have mosfets controlled by a pin of our microcontroller, so there isn't any need for a hardware-based delay

[Yansi]

This is a B class amplifier to begin with.

Decoupling capacitor C1 placed wrong.  Should be split against GROUND.  This configuration of decoupling will not help much.

[Alex Nikitin]

C2L seems in reverse polarity to me. I'd like to avoid any cap bigger than 10uF to stay with small ceramic ones, and maybe modernize other things if you find flaws... Can you help me ?

Edit: after looking at different class-AB amplifiers on google, most of them don't have the caps equivalent to C2L, C3L & C4L. Maybe they are meant to optimize something ?

If the NE5532 is used, as suggested, the polarity of C2 is wrong, it will be biased by the opamp input current (about -0.3uA for NE5532) x 22K feedback resistor or about -6.6mV on pins 2(6). It would be better to reverse the C2 polarity with NE5532.

This is a B class amplifier to begin with.

Decoupling capacitor C1 placed wrong.  Should be split against GROUND.  This configuration of decoupling will not help much.

The C1 capacitor is essential for the NE5532 or other opamps with a similar topology (NE5534, NJM2114) and should be placed as close to the supply pins 4 and 8 as possible. Decoupling to ground caps are C4 and C5.

The amp is technically Class AB, as there would be a small idle current present through the output transistors.

Cheers

Alex

[drussell]

I planned initially to use a digital amplifier IC but since we want the ability to drive high impedance headphones (600 ohms) with studio-quality someone told me it would be better to use a classic approach with a ~1W class-AB amplifier, powered with +15V/-15V rails.

That is sound reasoning and a good idea.  I love my old Sennheiser HD400s (the original 1970/80s ones, with the yellow foamies) but many modern devices' headphone amplifiers simply won't drive them to any reasonable level with their 600 ohm impedance without an additional external amplifier.

We found this schematic, from http://sound.whsites.net/project113.htm :

There is nothing even remotely "studio quality" about that amplifier schematic.

I would start with something better rather than trying to modify that design.  Almost anything would better than an op-amp driving a couple transistors. 

That kind of simple design is great for learning about the basics of amplifiers and makes it very easy to experiment with things like amplifiers where you take the feedback from the current through the voice coil to smooth out the impedance curve of the driver, etc. but is really only appropriate for experimentation or general purpose rather than a high-sound-quality headphone amplifier.

If you're looking for something simple and inexpensive, I would try starting from something more like this:

https://www.audioxpress.com/article/Low-Cost-Headphone-Amp

...or start working backwards from one of the really good quality circuits, explore real class A amplifiers to remove crossover distortion (we're only talking about a headphone amp here, so unless you're working battery-powered-only, the bit of power consumed in true class A stages including the output stage really shouldn't be too much of a concern...)   etc.

If you REALLY want to make one from ICs and don't care as much about the budget, there is also always this approach: 

https://www.audioxpress.com/assets/upload/files/HighQualityPreampAXJun2013.pdf

[Yansi]

You say there is nothing studio quality about the simple amp, then you slap here a discrete one that will be likely even worse due to having rather lower open loop gain. And then I see C4: "C4 has been added to provide a greater internal gain." ... making the transfer function of the voltage amplifier stage inherently highly nonlinear. No real measurements of that amplifier provided.

I will not comment on the other one with a dozen of paralleled buffers, that is just one typical audiophool right there. I have seen this done even with a dozen of paralleled current output audio DACs.
Also, no real measurements of that amplifier provided.


If you want fancy pancy, look for a diamond buffer, these work decent. Not all of them, but here you get an idea:

[drussell]

You say there is nothing studio quality about the simple amp, then you slap here a discrete one that will be likely even worse due to having rather lower open loop gain. And then I see C4: "C4 has been added to provide a greater internal gain." ... making the transfer function of the voltage amplifier stage inherently highly nonlinear. No real measurements of that amplifier provided.

Being a beginners' thread, my point was to direct the OP to a type of design that allows more in-depth tweaking and experimentation with the various aspects of the building blocks of the discrete circuit than the one originally provided, rather than provide a "build it this way and be done" circuit.

If the goal is not to learn how to design, modify and tweak a circuit for an individual application, then the OP would probably be best to go with a module or essentially a decent power-op-amp.

Without more information on precisely what they're trying to accomplish, what their target specifications are and what measurement equipment the OP might have available to them, I can't provide much more specific information or pointers.

I will not comment on the other one with a dozen of paralleled buffers, that is just one typical audiophool right there.

Most audiophoolery is based on at least a grain of truth.

Paralleling the outputs like that is an effective way to create a medium-power op-amp from individual building blocks while retaining the characteristics of the individual blocks.

Is it messy?  Sure!  Does it work and perform well?  Yes.

Would I build my headphone amplifier using exactly either of those circuits or the OP's?  Heck no.

If you want fancy pancy, look for a diamond buffer, these work decent. Not all of them, but here you get an idea:
https://www.atlhifi.com/wp-content/uploads/2016/02/headamp-2-bp.png

Indeed, however, I have a feeling that the OP is looking for something reasonably simple and straightforward while providing reasonable fidelity.  The OP will need to chime in with more information to be able make a rational decision... 

As for performance data, even small changes in component types can obviously make large differences in the circuit performance and sound quality.  The OP will need to be able to make some measurements on the actual unit to be able to quantify at least some of these aspects, regardless of the circuit topology chosen if they expect to be able to tweak for the desired performance level.

[drussell]

Ooh I see ! The polarity of C2 confused me since it will see positive and negative voltages, but it's not the first time I see electrolytics used like that, it doesn't seems to matter when the current/voltage is limited enough with a resistor. I was planning to replace them all with 22uF ceramic caps.

You should try to avoid using ceramic capacitors in the audio signal path.  Their capacitance tends to be highly non-linear with changes in the applied voltage.  Best to use film when possible if permitted by space / cost constraints.

For decoupling and stabilization type uses they're great.

b) Short circuit protection => you're right. Since the circuit is ready to drive already high-impedance headphones (600+ ohms), isn't the simplest solution simply to replace the 10 ohm resistors by a higher value ?

If you go too high, you then remove the ability to drive low impedance, low efficiency headphones to high level, although with your original circuit that can theoretically supply multiple watts, that may not be an issue.

On some older typical hi-fi stereo amplifiers / integrated receivers, the headphone jack was just run right off the main speaker power amplifier output with, say, 390-560 resistors.

[dietert1]

If you use 1 nF parallel to R3 then for a compensated divider C = 1 nF * 1 KOhm / 22 KOhm = 45 pF should be parallel to R4. By looking at the transient response you can determine whether a 47 pF or a 39 pF works better. The 1 nF close to the OpAmp inputs can make a big difference on sound quality in a RF poisoned environment.
Concerning the 20 or 30 Hz high pass, this is an problem i have seen in many audio solutions. Engineers think the lower Hz they handle the better, but this may just not be true, when the speaker/headphone system starts producing lots of intermodulation. I did not propose to use ceramic capacitors in the audio path, but only for C3 and C4, that are uncritical in this respect (effects eliminated by negative feedback). The 10 uF i proposed for C2 can be film, maybe 2x 4u7 50V.
Concerning short circuit safety: It's fairly easy to test how long your final design survives a short cicuit at reasonable audio levels and how it dies. Certainly you don't want a fire.

Regards, Dieter

[drussell]

Just a few quick additional notes...

Regarding the Elektor magazine example I posted,

No real measurements of that amplifier provided.

Indeed, there are no published curves or in depth measurements provided for the unit as-built, only the specifications table according to the original article:
https://archive.org/stream/ElektorMagazine/Elektornonlinear.ir2011-01#page/n47/mode/2up

Input impedance (without P1)	10kΩ
Bandwidth	3.4 Hz -	2.4 MHz
THD + Noise (1 kHz, 1mW/33Ω)	0.005%	(B = 22 kHz)
THD + Noise (20 Hz - 20 kHz, 1mW/33Ω)	0.01%	(B = 80 kHz)
Signal to noise ratio (ref. 1mW/33Ω)	89 dB	(B = 22 kHz)
	92 dBA
Max. voltage (into 33Ω)	3.3 V	(THD+N = 0.1%)
Max. input voltage	0.57 V	(with P1 set top maximum volume)

And, then to the OP, regarding this advice from:

The 1 nF close to the OpAmp inputs can make a big difference on sound quality in a RF poisoned environment.
...
I did not propose to use ceramic capacitors in the audio path, but only for C3 and C4, that are uncritical in this respect (effects eliminated by negative feedback). The 10 uF i proposed for C2 can be film, maybe 2x 4u7 50V.
...
Concerning short circuit safety: It's fairly easy to test how long your final design survives a short cicuit at reasonable audio levels and how it dies. Certainly you don't want a fire.

These are all very good points.  I agree wholeheartedly! 

[ratatax]

Thanks for all those informations !

Now I have more for you.

Yeah I'm looking for something 'quite' simple, it's for my product which is a synthesizer I plan to build for my upcoming business, so I can't go crazy on the costs. I'm only in the design phase currently, but the budget for heaphones amplifier is only a small part of the whole product so I think I need to keep it under $5 (this is not including the +15V/-15V power supply and connectors. Price is for buying components in qty ~100 on LCSC).

That's why I'm looking either for an amplifier IC or a quite simple but good enough solution with opamps (TL072 series) and transistors. Amplifier IC are almost always specified for 8ohms loads in their datasheet, I lack the knowledge here to understand if they would work well enough for higher impedance loads. I thought it was only a power thing, but I heared frequency response may also be affected if the amplifier isn't designed for high impedances ? (even if it is powerful enough)

I'm not searching for the Audiophile thing, if it has no audible/obvious distorsion or noise it's fine

I routed the schematic to see how it looks on the board :



yup it's a TL074 with two unused modules in this implementation... !

22uF ceramic caps here but I may have to switch back to aluminium caps @drussell

current transistors are MMBT100 / MMBT200

I have seen opamps with decoupling caps between V+/V-, some others between V+/GND and V-/GND. Best practice would be the latter ?

[Audioguru]

Will the tiny little surface mounted output transistors survive the heating in this amplifier?

[ratatax]

I have no idea, those puppies are rated for 300mA : https://datasheet.lcsc.com/szlcsc/1902210930_ON-Semicon-ON-MMBT100_C274690.pdf

RMS power will be way below 1W (maybe peaks could reach that).

I can go for the BCP56/BCP53 which comes in SOT-223 package with a small heatsink, but those are rated for 80V 1A, it seems crazy overkill : https://datasheet.lcsc.com/szlcsc/Nexperia-BCP56-115_C22239.pdf

[D Straney]

Be careful about reading too much into the "300 mA" or "1A", as the current spec on transistors can be very misleading: I've seen people assume that because a bipolar (or MOSFET, or diode, or LED...) says "5A" in big letters on the datasheet, that they can put 5A through it under any condition - smoke usually ensues!

Think of the current rating more as a way to compare transistors, not as an absolute metric.  Having blown up plenty of power devices myself, and hopefully learned something from it, the things that actually matter are:
1. How easily the transistor can get rid of the heat: this is more about the package than anything else.  I don't know how much power you're planning on allowing your output transistors to dissipate, but I'm going to make up some numbers here, just for an example.  Let's say the amplifier is putting out 1W at 50% efficiency: this means it's dissipating 1W too, mostly in the output transistors, 0.5W for each one.  Now looking at the datasheet for the SOT-23 parts, it shows 357 C/W junction-to-ambient thermal resistance - that means that the temperature of the transistor's die is going to be 0.5W*357 C/W=179 C above ambient temperature...if the ambient air temperature around the transistor is maybe 60 C inside a box on a hot summer day, that bit of silicon (rated only to 150 C, by the way) is going to be at 239 C!  Instant explosion.

On the other hand, if you use the other transistor, because the package has a big metal tab its junction-to-ambient thermal resistance is about half of the other one (192 C/W), which means that the temperature rise above ambient will be only half as big.  That's only some of the benefit: the metal tab also provides a good place to conduct even more heat away either into an external heatsink, or to a copper area on the PCB as a cheap heatsink.  Check out the thermal conductivity list on page 6, and notice that the thermal resistance (junction to ambient) drops from 192->125->93 C/W as you go from no extra copper -> 1 cm^2 extra copper -> 6 cm^2 extra copper.

Again, most of this is applicable to pretty much any transistor in a SOT-23 package vs. a SOT-223 package, not just these two particular part numbers.  There's some variation between manufacturers, etc. but one package type is usually pretty consistent.  Be careful about footnotes and test conditions: the BCP56 datasheet has very good info about thermal resistance, while some "too good to be true" thermal resistance numbers I've seen on other datasheets for some tiny SOT-323 package turn out to be only when using a special ceramic (high-thermal-conductivity) PCB with 4 cm^2 pads on every pin, once you dig into the footnotes.

2. Other effects like "secondary breakdown" (essentially non-uniform conductivity causing hot spots which cause thermal runaway) on bipolar transistors: check the device's "safe operating area" (SOA) diagram for this limit.

Apologies if you know all this already: just looked like a common misunderstanding was about to happen, and hoped to maybe save some unnecessary smoke

[ratatax]

You're right pointing that, I was forgetting that those voltages and current handling are absolute max in ideal cooling conditions (my developer background isn't helping here!). Since the sot-23 package probably can't dissipate even 0.1W without getting hot, we'll go for the SOT-223 with some pcb aera forming a heatsink.

[Kleinstein]

You're right pointing that, I was forgetting that those voltages and current handling are absolute max in ideal cooling conditions (my developer background isn't helping here!). Since the sot-23 package probably can't dissipate even 0.1W without getting hot, we'll go for the SOT-223 with some pcb aera forming a heatsink.

In the ear phone range there are quite some amplifier chips available. Depending on the power needed (too much is not good for the ears anyway), one could even consider 2 channels of the NE5532 in parallel. No real need to go for a discrete design at this power level. AFAIK it is not that uncommon to have some series resistor (100 Ohms range) at the output, so that a 600 Ohms and 32 Ohms speak could produce a similar sound level.

[David Hess]

This is a B class amplifier to begin with.

The diodes between the bases forward bias the base-emitter junctions and the emitter resistors limit the current.  Essentially it is a simplified diamond buffer with diodes replacing two of the transistors.  It is not as adjustable as replacing the two diodes with a Vbe multiplier but it still works out to class-AB.

If I was designing a simple headphone amplifier, I might do something like that shown below but modified for single supply operation.  One advantage of this configuration is a rail-to-rail output range.

[Yansi]

I am not sure if two diodes are enough for any bias to flow through the output stage, are they?  Such current will be mostly ill-defined.

[David Hess]

I am not sure if two diodes are enough for any bias to flow through the output stage, are they?  Such current will be mostly ill-defined.

They are, especially if the right diodes are selected, but you are right, the current will be ill-defined.  A 60 millivolt difference yields a 10 fold change in current.  An alternative for better matching is to use an extra set of output transistors as the diodes and if you do that, you might as well configure the output circuit as the full diamond buffer.

Diode connected transistors:

https://www.edn.com/design/analog/4363991/High-speed-buffer-comprises-discrete-transistors

Diamond buffer:

https://www.tubecad.com/2012/09/blog0244.htm

[dietert1]

The tubecad log is interesting but who wants to make a completely discrete circuit nowadays? I think the proposal to bypass the two 10 Ohm output resistors with a capacitor is valid, but it's application isn't really practical with SMD only.
I think the Rod Elliott circuit shown in the beginning is a little more rational. Depending on the OpAmp it will certainly outperform all those diamond circuits.
I would use a modern audio OpAmp like OPA2134 instead of a TL082. By the way, the OPA2134 is specified for a 600 Ohm load (+/- 35 mA output current). So for headphones you may not even need an extra output stage.

Regards, Dieter

[ratatax]

I agree I'm don't like very much to design a discrete circuit, as it's a bit like reinventing the wheel and taking so much extra risk in design flaws, but I'm stuck with all the confusion in the amplifier world... So much nonsense informations everywhere, I can't believe finding a modern headphones amplifier schematic is almost impossible, all searches always brings very old designs without any explanations about performances except "it sounds good" 

I've looked at the OPA2134 and the fact the datasheet mentions 600ohm load and +/-18V VCC is very interesting.

However it mentions the voltage output to be max +/-2.5V at 600 ohm, which is I=U/R = 4mA. 4mA @ 2.5V is about 10mW, which seems low for those purposes. In real world I don't know, a typical headphones with high impedance would be about 90dB/mW so it seems OK, but without much margin

[Alex Nikitin]

The best IC for a headphone amp I've tried is the LM7171 , however it is not an easy opamp to deal with, requires a good layout and P/S decoupling, plus some careful filtering on the input and the output.

Cheers

Alex