I'm interested in a emitter follower with active current source design to buffer op-amps with differential outputs. I've sketched a design:
If this seems reasonable, I probably will parallel the NPNs (4 per side) to obtain 80mA class A for each output.
Has anyone seen anything like this? I'm mostly interested in a transistor design non-complementary.
I don't mind dealing with heat issues. This amplifier is to drive headphones.
Comments welcome.
JF
An externally hosted image should be here but it was not working when we last tested it.
If this seems reasonable, I probably will parallel the NPNs (4 per side) to obtain 80mA class A for each output.
Has anyone seen anything like this? I'm mostly interested in a transistor design non-complementary.
I don't mind dealing with heat issues. This amplifier is to drive headphones.
Comments welcome.
JF
To me it looks basically like two half diamond buffers in
bridge coupling. Have you considered using two full diamond
buffers instead? In theory and in measurements the diamond
buffer has excellent distorsion figures. I am sure different
people have different opinions on which sounds best. I have
no opinion as of yet, but I am currently working on a board
design with opams followed by discrete diamond buffers, so
I expect testing it in the near future.
bridge coupling. Have you considered using two full diamond
buffers instead? In theory and in measurements the diamond
buffer has excellent distorsion figures. I am sure different
people have different opinions on which sounds best. I have
no opinion as of yet, but I am currently working on a board
design with opams followed by discrete diamond buffers, so
I expect testing it in the near future.
Because a full diamond buffer is also class A, if properly biased
and, at least in theory, it lacks even-order distorsion when
the output transistors are matched, while a half diamond
buffer has quite a lot of second-order distorsion. I do not
know which of the two has the lowest amount of odd-order
distorsion, but the full diamond buffer performs excellently
in this respect as can be confirmed by calculations or SPICE
simulations, and has also been confirmed by measurements
published by Walt Jung.
and, at least in theory, it lacks even-order distorsion when
the output transistors are matched, while a half diamond
buffer has quite a lot of second-order distorsion. I do not
know which of the two has the lowest amount of odd-order
distorsion, but the full diamond buffer performs excellently
in this respect as can be confirmed by calculations or SPICE
simulations, and has also been confirmed by measurements
published by Walt Jung.
Christer said:
I'd like to know why half a diamond would have much higher second order distortion, then a full diamond. I wonder if it is because it slews on faster than off. Of course, I expect some distortion cancellation (especially, with second order) because I'm using a differential bridged circuit.
I don't see much "half diamond" circuitry. The old BUF-03 used more of a non-complementery emitter follower out. Of course, newer buffers use complementery diamond topology.
Well, igonoring the fact that non-comp emitter followers are rare (only because of inefficiency?), I'd like to find a reference that indicates that the full diamond performs better than the half diamond. Again, efficiency aside.
Thanks for the replies.
JF
Bricolo said:
I let my CPU sweat and give me some value for the money it
cost me.
Basically, I run a transient analysis of desired source amplitude
and frequency and then do an FFT. I always make sure to do
an FFT also on the source signal as a reference, to see that
the simulation parameters chosen give none, or sufficiently
low distorsion on the source signal. Empirically, but testing
various trade-offs I have settled on the following parameters
in most cases. I run 21 cycles, starting to collect data after the
first one, ie. 20 cycles are analysed in the subsequent FFT.
I also set the minimum step-length to 10E-5 times the cycle
length, ie. for a 10 kHz source signal, the step length is 1ns.
Then I FFT those signal, including the source, that interest me,
choosing a large number of data points, usually 1Meg. I have
found this to produce a noise floor of around -120dB or better,
which is usually sufficient. you can improve of that, but it takes
quite some time already with these parameter values, so it
is hardly worth it.
Just don't forget, no simulation is better than the models used
and the simulator may produce strange effects in certain odd
cases.
johnferrier said:
The exponential transfer characteristic of a BJT causes a lot
of distorsion, both odd and even. In the SE (non-balanced)
case, you get all of this. With complementary output BJTs you
get a distorsion cancellation. I have done simulations on the
diamond buffer, and with perfectly matched BJTs the even
order distorsion seems to cancel out completely, but there is
still odd order distorsion. Changing either IS or BF of output
device, which corresponds to mismatch in Vbe or hfe respectively
will introduce also even order distorsion.
In the simulation I run, where I used a
rather high input signal and heavy load to see distorsion effects
clearly, I got -81dB 3rd order and no 2nd order distorsion.
mismatching BF by a factor 4 between the output devices gave
the same 3rd order distorsion, but introduced -86 dB 2nd order.
Mismatching IS by a factor 2 also gave the same 3rd order but
introduced -114dB 2nd order. BTW, these tests were done at
10kHz and using the procedure explained above.
Unfortunately, I do not remember the exact load conditions etc.
Since I was only interested in the effects of mismatching when
I did that analysis, I only jotted down the distorsion figures
on a paper.
After your recent question, I decided to try the SE (non-balanced)
version. I am not sure if the bias current and load resistance
are the same as in the previous simulation, but I think they
are about the same. The result I got was -60dB 2nd order
distorsion and still -80dB 3rd order distorsion. It thus seems
that the diamond buffer cancels even order distorsion only.
I also tried your balanced setup, but for some reason I cannot
figure out right now it didn't work. I think the reason is
that I took your variant with only series resistors on the inputs,
which does not bias the amp properly if the input signal is
differential. I would expect this setup to cancel distorsion
also, perhaps better in practice than a single, non-balanced
diamond buffer set-up, since you have two NPNs cancelling
each other, not complementary devices. That's just a guess
though.
As for referring to it as a half diamond buffer. Well, I thought
is sounded appropriate, but others may disagree. It is no
important issue what to call it, and that easily just leads to
a philosophical discussion.
As I said, I suspect the bridged diamond buffer to have the
be the best performing of the alternatives discussed. Besides,
It will dissipate less power in the transistors, making cooling
easier, PSU req. slightly less demanding, at a slightly higher
component count and complexity. However, If you do not
absolutely need a bridged setup, I would suspect a single
diamond buffer performs well enough. I hope to experiment
with one soon myself.
BTW, unless you have already read it, I strongly recommend
reading Walt Jungs article on the discrete diamond buffer:
"Walts Tools and Tips, Op-Amp Audio, High performance
buffers (part II): Electronic design, Oct. 1, 1998".
I don't know if it is online still, though.
As for my effort, it was quite interesting and educational for
myself too. ´Trying to answer other peoples questions often
force you to think deeper on the issue which sheds new light
on it. Being a university teacher, I stronlgy believe in the
idea that you don't really know a course until you have taught
for a couple of years.
be the best performing of the alternatives discussed. Besides,
It will dissipate less power in the transistors, making cooling
easier, PSU req. slightly less demanding, at a slightly higher
component count and complexity. However, If you do not
absolutely need a bridged setup, I would suspect a single
diamond buffer performs well enough. I hope to experiment
with one soon myself.
BTW, unless you have already read it, I strongly recommend
reading Walt Jungs article on the discrete diamond buffer:
"Walts Tools and Tips, Op-Amp Audio, High performance
buffers (part II): Electronic design, Oct. 1, 1998".
I don't know if it is online still, though.
As for my effort, it was quite interesting and educational for
myself too. ´Trying to answer other peoples questions often
force you to think deeper on the issue which sheds new light
on it. Being a university teacher, I stronlgy believe in the
idea that you don't really know a course until you have taught
for a couple of years.
Christer said:
Okay, I now favor the diamond buffer (rather than the half diamond), but still plan to use a bridged (double-diamond) outputs. I'm willing to pay for a little more performance. Since, the symmetry-bridged topology does nothing for odd-order distortion, I will also see if I can find techniques to lower that too (does this mean MOSFETs?). I will search around.
Also, I did find Walt's article for op amp buffer, using the Wayback Machine--the internet archive.
And yes, I learn by trying to answer questions too (usually, I learn that I don't know quite as much as I thought).
Thanks and good luck with your circuitry!
JF
The constant-current "Class A" topology dissipates twice the power of a true push-pull Class A amp. Here you are dissipating 4.8 watts to get 0.95 watts in the load; push-pull Class A with the same drop-losses would be under 2.5 watts heat. This may not be too important at headphone levels.
With the constant current topology, device current swings from zero to 160mA; with push-pull it swings zero to 80mA. The lower swing tends to less distortion. The difference may be too small to worry about.
As noted, you will need a 4-wire headphone connection, not the usual 3-conductor plug, but I assume you saw this.
What are all those resistors??? Why have you taken a Diamond and cut one side off? Those resistors are to make both sides of the diamond play nice together. The 10 and 1Ω Emitter resistors set the currents, but here you have set current with current sources so you do not need bias resistors. The 10Ω emitter resistor adds impedance to the drive of the output transistor which we really don't want. The 1Ω emitter and and 10Ω load resistors just waste power, you don't need either one. The 1K gives problems when the two sides of the diamond do not cancel bias current, and is worse when there isn't any cancellation at all.
Why even have two stages? For 80 or 160mA peak output current, any good little transistor needs only a couple mA base drive, well within the ability of any op-amp you are likely to use, even if you yank the opamp into class A. The B-E voltage variation does not cancel between first and second stage, so distortion of a one-stage design is not necessarily higher. Using one stage does mean, if you do not take DC feedback around the buffer, that there will be a 0.7V offset, but it will be the same at both sides of the headphone so there is no DC in the cans.
Then you have:
I've added a small resistor in the collector, to put some limit on fault current without much/any effect on sonic performance.
That will work quite well. Full push-pull both sides can be cooler and lower distortion, but now you need those emitter resistors back.
Interestingly, the Diamond can give lower THD if carefully trimmed in class AB than when working Class A. However the distortion spectrum shifts from mostly-low-order (Class A) to high-order (Class AB). You can make the 3rd vanish, but the 5th is unchanged, and the 7th rises a bit. The ear is the only judge of this.
With the constant current topology, device current swings from zero to 160mA; with push-pull it swings zero to 80mA. The lower swing tends to less distortion. The difference may be too small to worry about.
As noted, you will need a 4-wire headphone connection, not the usual 3-conductor plug, but I assume you saw this.
What are all those resistors??? Why have you taken a Diamond and cut one side off? Those resistors are to make both sides of the diamond play nice together. The 10 and 1Ω Emitter resistors set the currents, but here you have set current with current sources so you do not need bias resistors. The 10Ω emitter resistor adds impedance to the drive of the output transistor which we really don't want. The 1Ω emitter and and 10Ω load resistors just waste power, you don't need either one. The 1K gives problems when the two sides of the diamond do not cancel bias current, and is worse when there isn't any cancellation at all.
Why even have two stages? For 80 or 160mA peak output current, any good little transistor needs only a couple mA base drive, well within the ability of any op-amp you are likely to use, even if you yank the opamp into class A. The B-E voltage variation does not cancel between first and second stage, so distortion of a one-stage design is not necessarily higher. Using one stage does mean, if you do not take DC feedback around the buffer, that there will be a 0.7V offset, but it will be the same at both sides of the headphone so there is no DC in the cans.
Then you have:
An externally hosted image should be here but it was not working when we last tested it.
I've added a small resistor in the collector, to put some limit on fault current without much/any effect on sonic performance.
That will work quite well. Full push-pull both sides can be cooler and lower distortion, but now you need those emitter resistors back.
Interestingly, the Diamond can give lower THD if carefully trimmed in class AB than when working Class A. However the distortion spectrum shifts from mostly-low-order (Class A) to high-order (Class AB). You can make the 3rd vanish, but the 5th is unchanged, and the 7th rises a bit. The ear is the only judge of this.
In favour of SE operation:
Since music contains odd and even harmonics I think care needs to be taken in using configurations that favour one over the other. Push-pull looks attractive due to second harmonic cancellation but may not be the best in terms of sound quality.
The single ended design will produce higher total distortion but probably with a more even distribution of odd and even harmonics.
Since music contains odd and even harmonics I think care needs to be taken in using configurations that favour one over the other. Push-pull looks attractive due to second harmonic cancellation but may not be the best in terms of sound quality.
The single ended design will produce higher total distortion but probably with a more even distribution of odd and even harmonics.
Richard C said:
With the disclaimers I stated that it is only simulation and that
there might have been slightly different load and bias
conditions, the result I posted indicated that the 3rd order
distorsion was the same for SE and diamond, while the latter
lacked 2nd order. It's difficult to predict from theory what the
ear will like, and many seem to think 2nd order distorsion
make wonders, but will 0.01% 3rd order distorsion really be
less disturbing if you "mask" it by a 10 times higher level
of 2nd order distorsion? It would be another story, perhaps,
if we could decrease the 3rd order distorsion at the cost of
increasing the 2nd and get a higher total distorsion figure,
but that seems not to be the case here.
The analysis is simplified and non-conclusive of course. PRRs
remark on biasing the diamond buffer to class AB was interesting.
PRR,
First, thanks for your long post (discussing transistors) over on the Jung Regulator thread. I need more time to absorb what you wrote.
As far as heat, I've got relatively large heatsinks. And I plan to use 4 wires to the headphones--no problem.
The resistors, he...he...he... I wondered if anyone would say anything. I used those values (1 and 10 ohms) only because I read John Curl write that 15-18mV was an optimum value across Re (I used 20mV). Now that was for the emitter resistors.
I have the input (1k) and output resistor there so that I could parallel the circuit (probably only the output transistors though).
As far as two stage, well the diamond circuit uses two stages. However, the first stage is used to bias the output. I've got current sources to do that.
Richard C,
Yes, I'll try to balance the odd and even THD. You're suggesting SE achieves that.
Thanks.
JF
First, thanks for your long post (discussing transistors) over on the Jung Regulator thread. I need more time to absorb what you wrote.
As far as heat, I've got relatively large heatsinks. And I plan to use 4 wires to the headphones--no problem.
The resistors, he...he...he... I wondered if anyone would say anything. I used those values (1 and 10 ohms) only because I read John Curl write that 15-18mV was an optimum value across Re (I used 20mV). Now that was for the emitter resistors.
I have the input (1k) and output resistor there so that I could parallel the circuit (probably only the output transistors though).
As far as two stage, well the diamond circuit uses two stages. However, the first stage is used to bias the output. I've got current sources to do that.
Richard C,
Yes, I'll try to balance the odd and even THD. You're suggesting SE achieves that.
Thanks.
JF
> Since music contains odd and even harmonics I think care needs to be taken in using configurations that favour one over the other
Not the same.
The harmonics in amplifier distortion are exactly 2X, 3X, etc the fundamental and always phase-locked.
The harmonics in music are (generally) not exactly 2X or 3X the frequency of the fundamental, and not phase-locked. (Synthesizers are an exception; electric guitars are a complex mix of string inharmonicity and amplifier harmonics.)
The ear may be fooled, mistaking amplifier low-order distortion for "richer music". However only tin-eared rockers (like me) should be fooled by high-order "harmonics", which in normal musical instruments are NOT multiples of the fundamental but in an amplifier are perfect harmonics.
And we use single-tone harmonic distortion just because it is an easy test/simulation. If you go to two-tone: two clarinets on one stage will not intermodulate each other (to detectable degree). Two clarinets through an amplifier always will IM. The IM on complex milti-tone signals (real music) can be astonishing even when THD is low.
I think that higher-order THD means more IM for the same percent THD; however that's not what we measure on current amplifiers. Maybe just because the THD is so low that we are not really measuring harmonic production. On old low/no-feedback amps, the 1:4 60:7KHZ IM number was pretty sure to be four times the THD number; but as we supress THD this old guide becomes invalid.
> configurations that favour one over the other.
Neither one reliably "favors" musical harmonics, since the real musical harmonics are not perfect harmonics. If they were, then it might be possible for a high-2nd-THD amp to cancel the natural 2nd harmonic of an instrument, or boost it. In real life, the natural harmonic will go in and out of phase with the amplifier harmonic, "throbbing". The 3% or 10% THD numbers we accept on simple no-feedback amps limit "throbbing" to a small piece of a dB, ignorable.
Opinions and theory aside: we see happy music lovers with topologies from SE Triode to hyper-feedback. The ear must have a very complicated weighting rule for artificial harmonics. It clearly ignores 1% or more 2nd, yet complains painfully about a trace of 13th, yet another amp with slightly different trace distortion is well-loved.
Sometimes the best way to pick a configuration is to look at cost. In a loudspeaker amp, JF's plan would be mighty expensive (at headphone levels and with personal labor, this is no big deal). Instead of paying for oversized transformer and heatsink, you fancy-up a basic Class B stage to hold it in AB, reduce Gm nonlinearity, and probably (not always) add gain so you can use feedback.
Going back to JF's situation: as long as he's using op-amps anyway, an "absurd" yet perfectly workable plan is to get a couple dozen op-amps (cheaper in full-tubes) and parallel them (with appropriate connection so they don't fight). Most modern opamps JF would use for audio are loafing in 1KΩ. He could strap two quad-packs and drive 150Ω (one side of his bridge). That's 8 or 16 chips for stereo, but a simple design and perhaps a "simpler" distortion by omitting a booster stage. (IM multiplies with number of hard-working stages.)
Not the same.
The harmonics in amplifier distortion are exactly 2X, 3X, etc the fundamental and always phase-locked.
The harmonics in music are (generally) not exactly 2X or 3X the frequency of the fundamental, and not phase-locked. (Synthesizers are an exception; electric guitars are a complex mix of string inharmonicity and amplifier harmonics.)
The ear may be fooled, mistaking amplifier low-order distortion for "richer music". However only tin-eared rockers (like me) should be fooled by high-order "harmonics", which in normal musical instruments are NOT multiples of the fundamental but in an amplifier are perfect harmonics.
And we use single-tone harmonic distortion just because it is an easy test/simulation. If you go to two-tone: two clarinets on one stage will not intermodulate each other (to detectable degree). Two clarinets through an amplifier always will IM. The IM on complex milti-tone signals (real music) can be astonishing even when THD is low.
I think that higher-order THD means more IM for the same percent THD; however that's not what we measure on current amplifiers. Maybe just because the THD is so low that we are not really measuring harmonic production. On old low/no-feedback amps, the 1:4 60:7KHZ IM number was pretty sure to be four times the THD number; but as we supress THD this old guide becomes invalid.
> configurations that favour one over the other.
Neither one reliably "favors" musical harmonics, since the real musical harmonics are not perfect harmonics. If they were, then it might be possible for a high-2nd-THD amp to cancel the natural 2nd harmonic of an instrument, or boost it. In real life, the natural harmonic will go in and out of phase with the amplifier harmonic, "throbbing". The 3% or 10% THD numbers we accept on simple no-feedback amps limit "throbbing" to a small piece of a dB, ignorable.
Opinions and theory aside: we see happy music lovers with topologies from SE Triode to hyper-feedback. The ear must have a very complicated weighting rule for artificial harmonics. It clearly ignores 1% or more 2nd, yet complains painfully about a trace of 13th, yet another amp with slightly different trace distortion is well-loved.
Sometimes the best way to pick a configuration is to look at cost. In a loudspeaker amp, JF's plan would be mighty expensive (at headphone levels and with personal labor, this is no big deal). Instead of paying for oversized transformer and heatsink, you fancy-up a basic Class B stage to hold it in AB, reduce Gm nonlinearity, and probably (not always) add gain so you can use feedback.
Going back to JF's situation: as long as he's using op-amps anyway, an "absurd" yet perfectly workable plan is to get a couple dozen op-amps (cheaper in full-tubes) and parallel them (with appropriate connection so they don't fight). Most modern opamps JF would use for audio are loafing in 1KΩ. He could strap two quad-packs and drive 150Ω (one side of his bridge). That's 8 or 16 chips for stereo, but a simple design and perhaps a "simpler" distortion by omitting a booster stage. (IM multiplies with number of hard-working stages.)
> I read John Curl write that 15-18mV was an optimum
For what?
28-30mV is a theoretical optimum for a Class AB push-pull emitter-follower amplifier with a very low impedance drive.
In Class A, that works (though bias stability is tough and higher is not bad).
Where the driver is not very-low-impedance, the optimum shifts. My impression for the Diamond Buffer with emitter resistors is that the optimum may be half of the above case. That may be Curl's 15mV-18mV, though without knowing where he wrote this I am just blind-guessing.
And 15mV or 30mV is all wrong for many other topologies.
Bias stability in the Diamond is not good, unless you match parts. And you can NOT match a PNP to an NPN. The resistivity is higher in PNP. You can match Vbe, or you can match Hib, but unless you carve them both on one waver and adjust the area of the PNP you won't get a good 4-way match. That turns out to be the practical limit on Diamond Buffer performance. It can be very-very good, but with real parts it can't match theory. And with real parts, it tends to have significant offset (output sits about 70mV more positive than input).
As to whether single-ended, push-pull, constant current, or light-bulb sources sound "best": IMHO, a well-designed version of any of these should be "faultless". Often the "sound" is not in the topology but in a capacitor (or some trickery needed to "avoid capacitors"). Sometimes we "want fault" to "enhance sound". I've been called a pragmatist and you should have different opinions and preferences, but I think any of the above will work fine if you get all the small details right.
For what?
28-30mV is a theoretical optimum for a Class AB push-pull emitter-follower amplifier with a very low impedance drive.
In Class A, that works (though bias stability is tough and higher is not bad).
Where the driver is not very-low-impedance, the optimum shifts. My impression for the Diamond Buffer with emitter resistors is that the optimum may be half of the above case. That may be Curl's 15mV-18mV, though without knowing where he wrote this I am just blind-guessing.
And 15mV or 30mV is all wrong for many other topologies.
Bias stability in the Diamond is not good, unless you match parts. And you can NOT match a PNP to an NPN. The resistivity is higher in PNP. You can match Vbe, or you can match Hib, but unless you carve them both on one waver and adjust the area of the PNP you won't get a good 4-way match. That turns out to be the practical limit on Diamond Buffer performance. It can be very-very good, but with real parts it can't match theory. And with real parts, it tends to have significant offset (output sits about 70mV more positive than input).
As to whether single-ended, push-pull, constant current, or light-bulb sources sound "best": IMHO, a well-designed version of any of these should be "faultless". Often the "sound" is not in the topology but in a capacitor (or some trickery needed to "avoid capacitors"). Sometimes we "want fault" to "enhance sound". I've been called a pragmatist and you should have different opinions and preferences, but I think any of the above will work fine if you get all the small details right.
PRR said:
PRR,
I suppose he is referring to this post and some of the follow-up
discussion. Myself I have no opinion, not having read the
HP article.
http://www.diyaudio.com/forums/showthread.php?postid=244296#post244296
Further, thanks for your very interesting discussion of
instument vs. amplifier harmonics, as well as for other
recent interesting posts here and in the regulator thread.
Okay, here are sketches of my alternatives.
BRIDGED ACTIVE FOLLOWERS
JUNG DOUBLE DIAMOND BUFFER
I'm using a cross coupled OPA637s. (Concept found on last page of the AD8610 datasheet: http://www.analog.com/UploadedFiles/Data_Sheets/734705833AD8610_20_c.pdf)
Initially, I had thought a simple active follower was a bit basic. But, now sketching it out with the cross coupling, it seems like it may have just the right amount of sophistication.
The double diamond is still appealling and I don't mind the extra work involved in constructing it. Which circuit is best?
The 2SC3953 (2SA1538 w/datasheets http://www.datasheetcatalog.com/datasheet/2/2SC3953.shtml) seems like a super part to use.
Let me last add, my initial plan was to use BUF634s (which I have). However, I really like to use discrete circuitry at the output.
Anyone have further thoughts???
JF
BRIDGED ACTIVE FOLLOWERS
An externally hosted image should be here but it was not working when we last tested it.
JUNG DOUBLE DIAMOND BUFFER
An externally hosted image should be here but it was not working when we last tested it.
I'm using a cross coupled OPA637s. (Concept found on last page of the AD8610 datasheet: http://www.analog.com/UploadedFiles/Data_Sheets/734705833AD8610_20_c.pdf)
Initially, I had thought a simple active follower was a bit basic. But, now sketching it out with the cross coupling, it seems like it may have just the right amount of sophistication.
The double diamond is still appealling and I don't mind the extra work involved in constructing it. Which circuit is best?
The 2SC3953 (2SA1538 w/datasheets http://www.datasheetcatalog.com/datasheet/2/2SC3953.shtml) seems like a super part to use.
Let me last add, my initial plan was to use BUF634s (which I have). However, I really like to use discrete circuitry at the output.
Anyone have further thoughts???
JF
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.