Inaudible distortions

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
G

Guest

Guest
In the trhread about source follower I showed part of amp's schematic and got lot of questions in e-mail and in private messages.

A first, what means inaudible? For me, it means distortions are not unexpected. For example, if a cat barks everybody will see the fact, even being scared. If a dog barks, few people will pay attention.

The same, if distortions increase when the sound fades out (guitar strings, reverberation), and spector of them is wide, it is well audible.

But if level of distortions decrease with fading and order of harmonics became lower and lower we don't pay attention: it is natural, ordinary, "clean".

Here is the example of mentioned in that thread amplifier, this version is designed to drive headphones. In SPICE simulator the amp gives 8V PP on 32 Ohm load with 0.02 percent of 2'nd order harmonics, higher order harmonics nearly don't exist. In reality, it sounds very clean and natural, as if just an amplifying wire is connected between a player and headphones.

Feel free to build it for personal usage, no usage of output stage for profit without my permissions!
It contains in this particular case paralleled source and emitter follower (emitter follower is degraded - one part of error correction), loaded by a voltage to current convertor. Iddle current is half of maximal peak current for given load, this is another part of error correction. Bottom right transistor (Q5) with diode and resistor in base is a current mirror implementation. R1: set 4V on output (Vpp-Vgs)/2, R7 - to set iddle current.


Anatoliy

headphone_amp_from_wavebourn.gif
 
[quote author="bcarso"]I like your heatsink symbols!
[/quote]
It is not mine, it is from the guy who SPICED it and found no distortions of order higher than 2'nd. :cool:
Is D1 thermally coupled to Q5?

Not necessary, but if to couple measured distortions will be less. For the same sonical quality the diode may be omitted, resistor increased, but measured THD would be less impressive.
 
It's interesting. By coincidence I did an all-bipolar headphone amp design recently for a client which bears a distant resemblance. It is also class A although rather lower power, and at that I'm a bit worried about the typical earbud user blasting their skulls off.

I would share it but it doesn't belong to me as it were. It did sound good when I auditioned it.
 
Q1 and Q2 are a basic gain of 4 amplifier. Since Q6 follows this to the output node, I would assume Q2 collector sits between +4V and +6V, Q1 emitter about 1/4 of that, Q1 base about 0.5V higher.

Why do we need R1 trim? Why does it go to zero, shorting the input and slamming the output to ground? Counting on my thumbs instead of SPICE, I see Q1 base at 1.7V, I see R1 should be 43K, probably not much under 40K nor much over 50K. The 2mV/degC drift of Q1 will be a 4mV/degC drift at the output, so even 20degC ambient changes won't "break" the circuit. And if my room shifts 20degC, I'm not going to rush to the trim-pot, I'm going to rush for the hot coffee or cold beer.

Q4 seems to do nothing.

Like R1, R7 has so much range that it can get the amp in trouble. And not just "won't work" trouble, but burning-parts trouble. It is imposible to derive the exact relations between two different random devices like "1N4148" (which is a sloppy spec) and "SK95-2M3" (whatever that is?), but assuming similar doping (which is not a good assumption between diodes and BJTs) and a 100:1 junction area ratio(?), we need like 1mA in R7 to get 100mA in Q5. There is 6V or 5V across R7. Why would its resistance ever need to be less than about 5K? If we turn it near zero, Q5 current wants to be infinite, though somewhere around R7=100 ohms R7 will just burn-up.

I also don't see thermal stability for Q5. Without any emitter resistance, long before Q1's drift is interesting Q5's current with near-fixed base voltage will run-away toward infinity (unless it has the infinite heatsink). But if R9 is as high as 510 ohms, and Q5 is a typical low-Hfe power device, then Q5's emitter resistor should be (by Shea's stability criteria) about 50 ohms. And that would ruin output in 32 ohm loads. Hmmmm. Ya know, Q3 should probably run a lot more than 1mA to drive a 100mA power device, which suggests R9 a lot less than 500 ohms.

It almost seems that 5K and 510 ohms must be typos. With those values, I don't see a way Q5 won't burn-up at turn-on. Maybe something was mis-copied?

I do think I see your point about inaudible distortion. It isn't new. Your output stage reminds me a lot of JLH's early designs. For guys like me who grew up on Complementary Symmetry, JLH's output stage looks like it "can't work". In fact it does, and sounds sweet, but bias stability is tricky.

> SPICED it and found no distortions of order higher than 2'nd.

SPICE is a liar. And it can be tricky to get "good" distortion numbers out of SPICE, even if your models are realistic (many are over-simplified: if the nonlinearity is represented as a first-order formula, you will only get low-order "answers" from SPICE). The default analysis in most SPICE rigs will not interpolate low THD. I do agree that THD of your plan may be low and simple.... The Q1 Q2 loop can be quite clean and simple-2nd, and Q6 can swallow the huge distortion in Q5 quite well, at least until Q6's current becomes much less than Q5 current.
 
PRR, here is my original. ;)

kubeek.gif



D1 was suggested to be added to impress SPICElowers. However, a transistor in the current mirror would look more like in a schoolbook. ;)

Similar design I use for 200-300W power range. No thermal problems at all. Combination of FET and BJT follower, like the whole idea of loading follower by U/I converter, looks odd, non-trivial, but it works excellent. The single drawback is power consumptions, but if everything was made of tubes the amp of such quality would consume much more.

Why do we need R1 trim?
Because MOSFETs have very different Vgs. However, I personally would rather trim R3, but as drawn will work as well.

Like R1, R7 has so much range that it can get the amp in trouble.

Rather the person with a scredriver, not the R7. But he knew what he were putting there, so I believe he would not. :)

I also don't see thermal stability for Q5.

R9 is of the value that brings voltage drop caused by ditry currents of both hot transistors (Q3 and Q5) on negligible level.

It almost seems that 5K and 510 ohms must be typos. With those values, I don't see a way Q5 won't burn-up at turn-on. Maybe something was mis-copied?

Yes, you forgot about a diode in series. ;)
Without the diode 510 Ohm is indeed too low.
5K means 1.6 mA idle current (about 2.4mA max)
 
Hi again PRR, if you look at the input capacitance of MOSFET it may be seen as a varicap between driver and load. It may be shunted by resistor for less impact. Resistor may also provide clean zero-crossover current from driver to load, if it's value is small, and a driver is powerful. Also, the current may be provided by biased in class A emitter follower working in parallel with FET... This emitter follower may be degraded such a way with a resistor in emitter so it may supply current up to the region of beta droop, working always in small region of currents defined by Vgs of the FET. I call such couple "Tango", as if FET and BJT emitter followers dance helping each other...
 
And here is an example of 2x200W amp that uses the same topology of output stage.
First opamp is an amplifier for the input signal limiter, the second one is a servo feedback amp.

tower-II.gif
 
The guy for whom I've designed this headphone amp schemo reports that he uses it to drive shelf speakers. Go figure. :cool:
 
[quote author="Wavebourn"]I call such couple "Tango", as if FET and BJT emitter followers dance helping each other...[/quote]

I wonder how it sounds. the idea is the mosfet does the big current swings and bjt is more or less constant current and fills in the "gaps" for the fet? one way to look at it is sort of like an old quad circuit where a resistor connects from driver to output node, except you buffered it with a bjt.

I did some simulatory investigations of this topology myself. I stripped it down to just a mosfet, a degenerated bjt and an ideal current sink. I dont know if I am better or worse with the Spice than your friend, but I see alot more than just second harmonic. 3rd 4th 5th 6th 7th... falling off at about the same rate as you would expect from other simple followers. there may be some cancellation of distortion products between the two transistors, but I would'nt think that their nonlinearities would perfectly cancel one another, and my simulation shows that they dont. the degree of cancellation, or ratio of bjt/mosfet contribution, and weather it tends to make things more even or odd depends, among other things, on standing current. so the bias thermal-instability may actually be audible and at very least a moving target to trim. perhaps a small degeneration R in the Fet source is a good idea? I imagine this being more advantagous with a class-b mosfet + single bjt config, but now we are getting even closer to the quad circuit. at least that master's poop has expired by now.

mike p
 
[quote author="mikep"][quote author="Wavebourn"]I call such couple "Tango", as if FET and BJT emitter followers dance helping each other...[/quote]

I wonder how it sounds. the idea is the mosfet does the big current swings and bjt is more or less constant current and fills in the "gaps" for the fet? one way to look at it is sort of like an old quad circuit where a resistor connects from driver to output node, except you buffered it with a bjt.
[/quote]

No. Walker's patent was about A+C approximation (I did such thing at the end of 70'th not knowing that it was already patented), while here we have 2 devices working in class A.

I did some simulatory investigations of this topology myself. I stripped it down to just a mosfet, a degenerated bjt and an ideal current sink. I dont know if I am better or worse with the Spice than your friend, but I see alot more than just second harmonic. 3rd 4th 5th 6th 7th... falling off at about the same rate as you would expect from other simple followers. there may be some cancellation of distortion products between the two transistors, but I would'nt think that their nonlinearities would perfectly cancel one another, and my simulation shows that they dont. the degree of cancellation, or ratio of bjt/mosfet contribution, and weather it tends to make things more even or odd depends, among other things, on standing current. so the bias thermal-instability may actually be audible and at very least a moving target to trim. perhaps a small degeneration R in the Fet source is a good idea? I imagine this being more advantagous with a class-b mosfet + single bjt config, but now we are getting even closer to the quad circuit. at least that master's poop has expired by now.

The trick is in loading it on a modulated current source so current through a voltage follower is nearly constant. The same trick I used in PP amp with SE truiode sound ("Alligator" project) where one shoulder was made of triode connected GU-50, another one was a modulated CCS with mosfet out.

Cool. As soon as that English poop has expired I'm going to use A+C in woofer amps, using class A + class C Tango.
 
[quote author="Wavebourn"]
The trick is in loading it on a modulated current source so current through a voltage follower is nearly constant. [/quote]

IME, in general waggling the current sink has the effect of increasing the headroom of a follower for a given standing current and possibly reducing the even harmonic content a little bit. If you modulate the current such that the follower current is constant, you are in effect driving the load with the sink. imperfections in the current "sink" may well dominate overall performance. you will never accomplish this with a feed-forward scheme like you showed unless the load impedance is constant and well-known. if you sense the actual output current, or perhaps try to keep constant the drop in a resistor from the positive rail to the drain/collector of the follower, this could be achieved. another way to look at it is that the follower transistor(s) will think that there is no load. even with infinite load impedance, a follower loaded by constant current sink has distortion. I havent checked to see what Tango does in this case.

mike p
 
You are right, real speakers differ a lot from constant resistive load, but any way amount of distortions and their pattern are significantly different. Modulated current sink's distortions can't dominate because a voltage follower has much smaller dynamic resistance. It correlates well with listening tests.

Why I used a feed-forward approach, because feedback inevitably increases order of transfer function (read order of harmonic distortions) and the higher is frequency the more distortions you get, that is audible.
 

Latest posts

Back
Top