I have designed and built a simple Class A amplifier to drive headphones. The output impedance of a Class A amp is too high, and I am trying to transform it to something much lower – on the order of 1Ω – so that I can drive low-impedance headphones (e.g. 32Ω Grado SR-80).
I tried using a common collector (emitter follower) as an impedance transformer, but even that output impedance was too high – I measured around 60Ω. If that could be lowered, I'd love to use that design and keep everything BJTs, but I could not find a way.
I hit upon the idea of using a unity-gain op-amp (OPA275) to act as a buffer (see attachment), similar to how the O2 works, except with a biased input. It simulates great, but when I built it on a protoboard it is EXTREMELY noisy – there is a very loud hiss when the op-amp is connected to the output of Q1. I expected it to be dead silent, since the output from Q1 alone is dead silent, and the op-amp has unity gain and is also dead silent without any input connected.
So, what gives? Why is the op-amp so noisy? Is there a better impedance transformer design I could use?
(Notes: VCC is 15V. R10 and R11 (10Ω) are just there to make LTspice happy. They're not in the actual circuit.)
I tried using a common collector (emitter follower) as an impedance transformer, but even that output impedance was too high – I measured around 60Ω. If that could be lowered, I'd love to use that design and keep everything BJTs, but I could not find a way.
I hit upon the idea of using a unity-gain op-amp (OPA275) to act as a buffer (see attachment), similar to how the O2 works, except with a biased input. It simulates great, but when I built it on a protoboard it is EXTREMELY noisy – there is a very loud hiss when the op-amp is connected to the output of Q1. I expected it to be dead silent, since the output from Q1 alone is dead silent, and the op-amp has unity gain and is also dead silent without any input connected.
So, what gives? Why is the op-amp so noisy? Is there a better impedance transformer design I could use?
(Notes: VCC is 15V. R10 and R11 (10Ω) are just there to make LTspice happy. They're not in the actual circuit.)
Attachments
Last edited:
The standard way to lower the output impedance of a class A transistor stage is to increase the bias current. The transconductance 'g' is roughly equal to 40 * Iq, and the output impedance is roughly 1/g. So, crank up the bias, and you lower the output Z.
Re. the "OPA275" buffer - is it really an OP275? ;-)
It's not clear to me what the DC bias values are at the input of the OP275. I'm too lazy to simulate, but the DC value at the collector of the transistor could be as low as 0.6V or so above ground. According to my datasheets, the input common mode voltage range of the OP275 is a lot narrower than "right next to the rails", around 4.5V away from the plus or minus rail. So, if you stuff 0.6V into the input stage, and it's not going to be happy with anything less than 4.5V, it may well be horribly noisy.
To solve this, you could use bipolar supplies, bias the transistor stage "up", so the collector ends up at 5V or higher, or use some sort of coupling cap arrangement into the OP275 and a pair of resistors to bias the OP275 input halfway between the supplies.
Honestly, a different circuit would work a lot better. Nobody uses a common emitter stage to drive a low impedance load - they're typically used for gain, loaded into a high-Z. Emitter follower stages are typically used to drive low-Z loads. One could add an emitter follower stage onto your common emitter stage, instead of the op amp, and it could work pretty well. By choosing a suitably low value for the emitter follower's emitter resistor, you can select the bias of that stage, and dial in exactly the output conductance you want. I'd also consider using a different bias scheme for the first transistor than just the 470K to the base - use a two resistor divider so you can stay away from the negative rail and end up with some voltage swing, after you get all the rest sorted out.
Best of luck!
Re. the "OPA275" buffer - is it really an OP275? ;-)
It's not clear to me what the DC bias values are at the input of the OP275. I'm too lazy to simulate, but the DC value at the collector of the transistor could be as low as 0.6V or so above ground. According to my datasheets, the input common mode voltage range of the OP275 is a lot narrower than "right next to the rails", around 4.5V away from the plus or minus rail. So, if you stuff 0.6V into the input stage, and it's not going to be happy with anything less than 4.5V, it may well be horribly noisy.
To solve this, you could use bipolar supplies, bias the transistor stage "up", so the collector ends up at 5V or higher, or use some sort of coupling cap arrangement into the OP275 and a pair of resistors to bias the OP275 input halfway between the supplies.
Honestly, a different circuit would work a lot better. Nobody uses a common emitter stage to drive a low impedance load - they're typically used for gain, loaded into a high-Z. Emitter follower stages are typically used to drive low-Z loads. One could add an emitter follower stage onto your common emitter stage, instead of the op amp, and it could work pretty well. By choosing a suitably low value for the emitter follower's emitter resistor, you can select the bias of that stage, and dial in exactly the output conductance you want. I'd also consider using a different bias scheme for the first transistor than just the 470K to the base - use a two resistor divider so you can stay away from the negative rail and end up with some voltage swing, after you get all the rest sorted out.
Best of luck!
Driving a 32ohm load you only need a couple of volts swing - the bias current for your output transistor needs to be around 70mA. At that level of bias, the output impedance will easily be below 1ohm for an emitter follower.
Using an opamp won't be classA unless you add external biassing to it.
Using an opamp won't be classA unless you add external biassing to it.
First off, thanks for the great response. Yeah, it's OP275 – slip of the brain. It's just what I had lying around, and I figured it would be okay even though I'm not actually amplifying anything.
Good catch on the bias voltage being too low – it definitely is, it's only about 3.5V which is indeed less than the specified 4.5V input for the OP275. I will play with the biases both before and after the BJT and see how that helps.
I had tried an emitter follower but I couldn't get the output impedance low enough. I may give that another shot.
Good catch on the bias voltage being too low – it definitely is, it's only about 3.5V which is indeed less than the specified 4.5V input for the OP275. I will play with the biases both before and after the BJT and see how that helps.
I had tried an emitter follower but I couldn't get the output impedance low enough. I may give that another shot.
As Monte said, the output impedance of an EF has a fairly simple equation - 1/gm (gm here means 'transconductance' - a measure of output current(A) per input volt). The gm is calculated from 40*Iq - for 70mA this gives gm=2.8S and hence output impedance will be around 0.3ohm. If you then use the gain of the opamp (considerable) by including the EF in its feedback loop, the output impedance will drop to an almost immeasurably low level.
I reckon you're better off using the opamp to drive the transistor, rather than the other way around
I reckon you're better off using the opamp to drive the transistor, rather than the other way around
Last edited:
Have a look at the output stage here. This would transfer over to silicon with no problems other recalculating a couple of resistor values.
http://www.diyaudio.com/forums/head...rmanium-single-ended-class-headphone-amp.html
http://www.diyaudio.com/forums/head...rmanium-single-ended-class-headphone-amp.html
@OP: So what did you use for a power supply? Are you aware that the common emitter circuit has just about zero PSRR? If it's quiet on its own, that might be just because it has a very high output impedance (namely about 10k) - a 32 ohm load is quite a different story than an opamp input.
I wouldn't want to rule out oscillation either. There was no mention of power supply decoupling.
70 mA of emitter current would usually be a bit too much for a 2N3904, something a bit more beefy is recommended then. Usually a push-pull buffer is recommended for less inefficiency, within the feedback loop - then you can have the opamp do voltage gain, too.
I wouldn't want to rule out oscillation either. There was no mention of power supply decoupling.
70 mA of emitter current would usually be a bit too much for a 2N3904, something a bit more beefy is recommended then. Usually a push-pull buffer is recommended for less inefficiency, within the feedback loop - then you can have the opamp do voltage gain, too.
abraxalito: What is Iq? To me that means quiescent current, which I assume is the DC collector current on the output BJT when the input is grounded, is that correct?
What I found was that by decreasing the emitter resister I could indeed reduce the output impedance, but it was still being loaded down by a low-impedance load (i.e. the overall gain was noticeable less with 32Ω load versus an open). See attachment, with 70 mA of bias current running through Q4.
@Mooly: Cool, thanks. It's interesting looking at an "upside down" design. If I understand correctly, Q3 is your "emitter follower" (sort of), and Q4 and Q5 combine to create a constant current source, which would deter a low-impedance load from drawing so much current away from the output transistor. Is that true, or an oversimplification?
@sgrossklass: Hm, I did not know that about PSRR. I am using a simple full-wave rectifier with an LM7815 (see attachment). I will probably want to double up on C4. Right now I'm actually using an LM1086 fixed at 15V because that's what I had handy. Simulated ripple in the output is 360 µV, but I guess the regulator could add noise too, huh?
I agree, 2N3904 is wimpy. It's a placeholder for now so I can understand the basics and what specs are important before picking a real part.
@AndrewT: Yes, I was referring to adding an emitter follower as an output stage. Sorry if that was not clear. A gain of 10 is pretty excessive for a headphone amp, I agree – I will probably dial it down.
Attachments
Here's the latest and greatest. I replaced the output stage with a Darlington pair. I haven't breadboarded it out yet, but Q5 will be bumped up to something beefier and heat-sinked (I have a TIP48 handy), since it draws 100 mA just sitting at DC. The resistors R10, R11, R15, and R16 can consume up to 250 mW apiece, so those will have to be 0.5W resistors I suspect. Seems pretty inefficient.
I biased up the amplification stage and reduced the gain to 6X (15 dB). I also added C7 to the input to form a low-pass filter with R3, which has a 3 dB cutoff at 75 kHz.
THD is less than 0.15% for loads from 32Ω - 300Ω at 0.5 VRMS output, which is more than sufficient for headphones. That's the best THD I could find with this topology. At higher drive levels the distortion from the phones themselves probably dominates, anyway.
Thanks for all the feedback so far. I'm learning a lot very quickly and appreciate the help. What are some potential pitfalls I have missed?
I also have a delayed turn-on delay circuit with a relay, but by increasing C2 to 1000 µF, the DC offset on the output takes a long time to settle down (< 10 mV) – more than 10 seconds. At what level of DC offset is it safe to turn on the output, provided that the initial "thump" is skipped?
I biased up the amplification stage and reduced the gain to 6X (15 dB). I also added C7 to the input to form a low-pass filter with R3, which has a 3 dB cutoff at 75 kHz.
THD is less than 0.15% for loads from 32Ω - 300Ω at 0.5 VRMS output, which is more than sufficient for headphones. That's the best THD I could find with this topology. At higher drive levels the distortion from the phones themselves probably dominates, anyway.
Thanks for all the feedback so far. I'm learning a lot very quickly and appreciate the help. What are some potential pitfalls I have missed?
I also have a delayed turn-on delay circuit with a relay, but by increasing C2 to 1000 µF, the DC offset on the output takes a long time to settle down (< 10 mV) – more than 10 seconds. At what level of DC offset is it safe to turn on the output, provided that the initial "thump" is skipped?
Attachments
What are you trying to achieve?.
Presumably you are thinking class A is higher quality?, and it can be, but the quality of this is extremely low - if you're happy with how it sounds, then fair enough (but don't do any proper tests on it
).
Member
Joined 2009
Paid Member
This was my approach,
http://www.diyaudio.com/forums/headphone-systems/275180-grasshopper.html
Simple, Class A and works just fine with Grado SR-60
What are you trying to achieve?.
Presumably you are thinking class A is higher quality?, and it can be, but the quality of this is extremely low - if you're happy with how it sounds, then fair enough (but don't do any proper tests on it
My ulterior motive here is to learn about Class A amps by designing, building, and testing one. I understand the basic functionality of what a Class A amp is and how it works. It's details like impedance matching, power supplies, etc. that are learned best by doing. Then I can use this experience as a springboard for moving on to other amp designs.
If you have specific suggestions, please let me know. On paper a Darlington pair meets many of my design goals.
My "soft" engineering design goals are:
- Class A
- Sufficiently drive headphones from 32-300Ω impedance to listenable levels
- Low distortion
- Low gain
- Low output impedance
- Relatively simple design (few parts)
- Delayed turn-on relay
- Integrated DC power supply
- Final design using SMT components
I read the whole thread – good stuff, thanks. Learned about complementary feedback pairs and that I need to figure out how to make nice plots in LTspice like you have done (PSRR, FFTs, etc).
I may copy some or all of your design. I'd still like to have some gain, so I should be able to feed your unity-gain CFP stage with the output from the gain stage.
Constant current sources are still a bit of a mystery to me, but I'll do some research.
Why don't you look at existing class A designs?, yours is VERY high distortion.
Yep, you're there. I suppose I could have written Ie or Ic.
One I like, assuming I'm not running over 12V, is FZT688B. I haven't found a part with higher current gain (typically over 500). On occasion I do run over 12V I'll cascode it.
<afterthought> If you take a look at my blog I have a couple of outline designs for headamps, the most recent one is SE classA (in my estimation the best kind). Its using a bootstrapped darlington loaded by a cascoded current source (to get the highest possible output impedance i.e. lightest loading). In your case the input transistor of the darlington would probably benefit from having a bleed resistor from its emitter to the output emitter.
Last edited:
One of the simplest circuits that's actually any good would be the JLH (Circuit 1). It's a bit part-specific, as these simple circuits tend to be (it prefers older ones), and in simulation the pole/null introduced by C5/R8 seemed to degrade rather than improve stability, but it's quite a decent performer - there are two stages performing voltage gain, which gives it a decent amount of loop gain. There's no reason why you couldn't run it on a single positive supply with minimal adaptation, in fact its output DC offset tends to be a bit drifty anyway. There's a JLH thread in the headphones section.
Generally speaking, complexity, power draw and performance of a circuit are interrelated. So for a given level of performance, you can always trade off between the other two. The JLH, for example, is a moderately complex circuit that runs a fairly generous amount of Class A push-pull output stage quiescent current.
If you want to go with a CFP output buffer à la Grasshopper amp, I would recommend going with another two transistors for the voltage amp. Douglas Self discusses a few inverting versions here, though I'd be more after a noninverting variety. This is the voltage amp from a Grundig MXV100 pre (you'll typically find circuits like these executed to perfection in mid-late '70s gear):
Note several advanced tricks like filtered input bias (the circuit shown here achieves not only decent PSRR but also high input impedance). They probably needed C350 for compensation as closed-loop gain is rather low (these Grundigs already had a pre-gain stage in front of the volume control so as to keep output noise levels low).
And yes, all other things being equal, higher supply voltage = lower distortion in these kinds of circuits. Thermals and device limits must be kept in mind, of course.
Generally speaking, complexity, power draw and performance of a circuit are interrelated. So for a given level of performance, you can always trade off between the other two. The JLH, for example, is a moderately complex circuit that runs a fairly generous amount of Class A push-pull output stage quiescent current.
If you want to go with a CFP output buffer à la Grasshopper amp, I would recommend going with another two transistors for the voltage amp. Douglas Self discusses a few inverting versions here, though I'd be more after a noninverting variety. This is the voltage amp from a Grundig MXV100 pre (you'll typically find circuits like these executed to perfection in mid-late '70s gear):
Note several advanced tricks like filtered input bias (the circuit shown here achieves not only decent PSRR but also high input impedance). They probably needed C350 for compensation as closed-loop gain is rather low (these Grundigs already had a pre-gain stage in front of the volume control so as to keep output noise levels low).
And yes, all other things being equal, higher supply voltage = lower distortion in these kinds of circuits. Thermals and device limits must be kept in mind, of course.
Attachments
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.