I'm not concerned about that. I'm keenly interested in knowing what type of distortion and where it may come from so that I may try to avoid it on future projects.
Those 128 position relay switched resistor networks look very interesting....
I've played with this idea, but the BC5xx/BC3x7 transistors are somewhat special as you can see from my Kmultiplier results. Furthermore, thermal matching only reduces the error that the circuit already has, which is mainly Early and Hfe distortions. For instance if you make a current mirror with a 26mV (whether intentionally or by temperature difference) difference in Vbe, the output currents will be 100% mismatched but they will still track linearly due to the intrinsic nature of the BJTs - it will be a current-doubler-mirror. So I don't think there is much to gain from using matched transistors. I tried using BC8xx duals but they actually performed worse IIRC.
It is a simplified explanation. If Hfe is above 400, then there's not a lot of error from Ib. Furthermore, Ib errors are relatively benign, because Ib is proportional to Ic, just like Rm. The proportionality of Rm to Ic is the reason a current mirror is still linear even with a Vbe mismatch.
Vbe in-circuit or by multimeter? Vbes don't actually need to match for this to work.
Yes they differ, but it doesn't matter much because the circuit accounts for this. The emitter resistors are effectively part of the emitters in this schematic, and they cancel each other out. They are only needed to reduce the current in the input transistor.
Using a PNP current mirror won't have matching between Q1 and Q3 or Q2 and Q4 and this will be a significant source of distortion. Without the cancellation of Q1/R2 and Q3/R1, the nonlinearity of a forced (by degeneration) x10 current mirror ratio would cause failed cancellation.
It's in-phase cancellation of Vbe curves, log-antilog cancellation, as opposed to out-of-phase cancellation for instance in an EF or an LTP. This way ALL harmonics are cancelled, as well as the circuit can cancel them. For this reason the leftover residual has a benign monotonic profile, rather than a mix of canceled and uncanceled harmonics.
As I recall Joachim's circuit was never unstable, but his output cable was resonating in its series mode, which would happen with any high-speed buffer. This part of the Kuartlotron could be considered positive feedback, but positive feedback alone does not cause instability.
Last edited:
Keantoken,
I measured the Vbe's on the transistors. Both BC546C and BC556C read 0.763V. I dont know if they are suppose to match so well but that is what it read.
I think the manufacturer is Fairchild semiconductor. The 'C' grade is difficult to get and I got these from PMI. hFE is around 520 for npn and 420 for pnp.
Thanks
I measured the Vbe's on the transistors. Both BC546C and BC556C read 0.763V. I dont know if they are suppose to match so well but that is what it read.
I think the manufacturer is Fairchild semiconductor. The 'C' grade is difficult to get and I got these from PMI. hFE is around 520 for npn and 420 for pnp.
Thanks
Last edited:
Keantoken,
Can I use the inductor in the following link? The value is around 10uH.
09HCP-100M-50 Fastron | Mouser
Thanks
Can I use the inductor in the following link? The value is around 10uH.
09HCP-100M-50 Fastron | Mouser
Thanks
You can use almost any small inductor 2uH or over, but 2uH is the optimal value. It is easy to wind your own air coil of this size:
http://www.diyaudio.com/forums/anal...error-correction-superbuffer.html#post3823525
http://www.diyaudio.com/forums/anal...error-correction-superbuffer.html#post3823525
Hi,
sorry Keantoken for replying so late to #83, but a flu knocked me off a couple of days.
A kind of current scaling.
In that case, the greater the scaling factor, the higher the distortions in the upscaled CM-leg.
In which way should a complementary CM be advantageous?
From a guts feeling I´d say that a same gender CM should track better.
Referring to the basic circuit (4-transistor only) on Your website, my sims show indeed a tiny advantage for a same gender CM.
Vbe of Q1 and Q3 as well as Q2 and Q4 will differ due to differing Vces of the transistors.
In circuit the differences will be alot smaller than out-of-circuit though.
As my sims show a miniscule improvement with Q4 beeing a PNP also, I assume that the precision of the current mirror is rather more decisive for the THD-level of the circuit, than Q2/Q4 beeing as similar as possible.
It might be just a matter of how to look at the circuit.
Either as a half-Diamond with CM-support, or as a emitter-follower (Q1) feeding a CM (Q3,4) with cascode (Q2)
The trick beeing that the casodes base isn´t fed by a constant voltage but a signal dependent voltage
I tend to the latter, as the CM defines the current in the output leg.
Which means, that You can exchange Q2 by basically any NPN transistor and still get low THD.
Referring to the Kuartlotron schematic now:
It certainly looks like a feedforward structure at first glance.
The collector signal of Q1 contains also a error signal.
This signal voltage modulates the emitter-follower Q4, that feeds this signal in antiphase and lowohmic into the output-leg of the circuit.
Thereby cancelling a great deal of the distortions.
A antiphase behaviour is measurable at "TP".
As the degree of modulation can be varyied with R5, the cancellation effectiveness varies also.
At the same the antiphase signal modulates Q3´s base, hence its collector currrent, which then feeds into Q1s emitter.
One could regard Q1 therefore as a quasi-differential input stage with negative feedback taken from "TP" and fed as as a correcting current into Q1s emitter.
Digging deeper into the circuit this explanation seems at least in part wrong.
I wondered about the role of C2, as it seemed counter intuitive, as it´d try to buck any voltage change at Q1s collector, thereby reducing the feedforward effect.
Surprisingly omitting with C2 didn´t change anything in the signal waveforms at Q4s base and the output in the transient sims.
So I opened the trace from Q1s collector to R5, omitted with C2 and connected R5 to gnd.
The THD is not only lower than with the simple theortical circuit, but also lower than the original Kuartlotron circuit.
Imho this result defeats the feedforward theory alltogether.
It hints us to the real distortion cancellation mechanism, which is the modulation of Q4s base due to the signal dependent current through R3.
If You take this effect out in that You open the trace from R3 to L1/C1 and connect a 1k resistor from L1/C1 to +10V the distortions are back and at the highest level.
As I understand it, with the Kuartlotron -or more precise the introduction of R3- You give up on Vbe cancellation and CM action, at least in the way You described.
Still though, introducing R3 and using a NPN for Q4 seems a clever move to create a THD-reducing modulative input by use of Q4´s base and to allow for matched Dual Transistors.
jauu
Calvin
sorry Keantoken for replying so late to #83, but a flu knocked me off a couple of days.
Isn´t that the same as having identical Vbes but using different emitter resistors?
A kind of current scaling.
In that case, the greater the scaling factor, the higher the distortions in the upscaled CM-leg.
In which way should a complementary CM be advantageous?
From a guts feeling I´d say that a same gender CM should track better.
Referring to the basic circuit (4-transistor only) on Your website, my sims show indeed a tiny advantage for a same gender CM.
Rm is which parameter? Couldn´t find it in Your or Hawksfords docs.
It doesn´t matter if You take Ib into account or not, nor does in-circuit or off-circuit measurement count.
Vbe of Q1 and Q3 as well as Q2 and Q4 will differ due to differing Vces of the transistors.
In circuit the differences will be alot smaller than out-of-circuit though.
I can only detect a noteable possible mismatch between Q2 and Q4 as those would be complementary.
As my sims show a miniscule improvement with Q4 beeing a PNP also, I assume that the precision of the current mirror is rather more decisive for the THD-level of the circuit, than Q2/Q4 beeing as similar as possible.
It might be just a matter of how to look at the circuit.
Either as a half-Diamond with CM-support, or as a emitter-follower (Q1) feeding a CM (Q3,4) with cascode (Q2)
The trick beeing that the casodes base isn´t fed by a constant voltage but a signal dependent voltage
I tend to the latter, as the CM defines the current in the output leg.
Which means, that You can exchange Q2 by basically any NPN transistor and still get low THD.
I tend to disagree with this explanation
Referring to the Kuartlotron schematic now:
It certainly looks like a feedforward structure at first glance.
The collector signal of Q1 contains also a error signal.
This signal voltage modulates the emitter-follower Q4, that feeds this signal in antiphase and lowohmic into the output-leg of the circuit.
Thereby cancelling a great deal of the distortions.
A antiphase behaviour is measurable at "TP".
As the degree of modulation can be varyied with R5, the cancellation effectiveness varies also.
At the same the antiphase signal modulates Q3´s base, hence its collector currrent, which then feeds into Q1s emitter.
One could regard Q1 therefore as a quasi-differential input stage with negative feedback taken from "TP" and fed as as a correcting current into Q1s emitter.
Digging deeper into the circuit this explanation seems at least in part wrong.
I wondered about the role of C2, as it seemed counter intuitive, as it´d try to buck any voltage change at Q1s collector, thereby reducing the feedforward effect.
Surprisingly omitting with C2 didn´t change anything in the signal waveforms at Q4s base and the output in the transient sims.
So I opened the trace from Q1s collector to R5, omitted with C2 and connected R5 to gnd.
The THD is not only lower than with the simple theortical circuit, but also lower than the original Kuartlotron circuit.
Imho this result defeats the feedforward theory alltogether.
It hints us to the real distortion cancellation mechanism, which is the modulation of Q4s base due to the signal dependent current through R3.
If You take this effect out in that You open the trace from R3 to L1/C1 and connect a 1k resistor from L1/C1 to +10V the distortions are back and at the highest level.
As I understand it, with the Kuartlotron -or more precise the introduction of R3- You give up on Vbe cancellation and CM action, at least in the way You described.
Still though, introducing R3 and using a NPN for Q4 seems a clever move to create a THD-reducing modulative input by use of Q4´s base and to allow for matched Dual Transistors.
jauu
Calvin
Not really. At low CM currents, Rm will swamp the emitter resistors, and the CM ratio will be close to 1:1. At high currents, Rm will be low and swamped by the emitter resistors, so the current mirror will have the same ratio as the resistors. So what you get when you do this is an inherently flawed current mirror that transitions from a 1:1 ratio to whatever ratio you set it to.
Rm is 1/Gm. Doug Self and others like to talk about Gm so that no one understands them, but I think Rm is a bit more intuitive and less obfuscating. It is simply the resistance of the emitter. BJT emitter resistance decreases proportionally with forward current.
Rm = 27mV / Ie
Q4 is the current mirror input, not the output. A current mirror only has distortion if your current mirror is flawed by design or the devices are operating outside of their intrinsic regions.
A current mirror is just a current mirror. The current mirror alone being linear isn't necessarily what's best for the circuit, if there are distortion mechanisms that the current mirror can be volunteered to cancel. In theory, the complimentary arrangement should be better because Q4 matches Q2, rather than hoping that a PNP as Q4 will match Q2 well enough. The circuit depends on this matching to reduce non-intrinsic distortion mechanisms - which may or may not be present, depending on the transistor and operating point. So there may be some leeway.
First you say you don't know what Rm is, then you are correcting the paragraph where I used it to explain my point. Perhaps you should understand what I write before trying to correct it?
Even tiny mismatches will make a relatively huge difference in non-feedback circuits like this.
That's possible, but needs to be tested in real life. I went through versions like this in the prototype, but I should revisit this idea, since it according to the simulator it may be viable.
The circuit operates the same in reality, no matter how it looks to us. Therefore we should see it as it is, rather than making assumptions based on how it is connected. I notice many of us see circuits first as configurations or combinations of topology, but I think it's better first to make no assumptions about how the circuit operates, and to understand it in all its complexity and scope.
Either as a half-Diamond with CM-support, or as a emitter-follower (Q1) feeding a CM (Q3,4) with cascode (Q2)
The trick beeing that the casodes base isn´t fed by a constant voltage but a signal dependent voltage
I tend to the latter, as the CM defines the current in the output leg.
If the output current were not defined by the 1k resistor to the negative rail, then the output voltage would not stay zero when the value of the resistor was changed. The current mirror defines the current through Q1.
That is true, IF the transistors are operating in their intrinsic regions. The Kuartlotron tries not to rely solely on that, by using closest-match wherever possible.
I tend to disagree with this explanation
What is with this winking? I do not understand.
You didn't understand what I meant. Harmonics that are in-phase cannot cancel, they will just add. In-phase means one transistor does not turn off when the other turns on. Instead, both transistors have a proportional Ic no matter what. This way not only are the even harmonics cancelled, but the odd harmonics as well.
Q4 is the input of the current mirror, not the output. The most it can do is change Vbe in response to a current difference. Yes, the B and C are not shorted, but this doesn't matter because Q3 goes across the E and B.
R5 is decoupled by 470u caps on both sides. It cannot change anything except Vce, and therefore thermal and Early behavior.
If Q2's Ie increases, Q1's Ie increases, because Q3 mirrors the current of Q2. They are in phase. If it were any other way, only the even harmonics would be cancelled.
Q4's base is decoupled to ground through a 470uF cap; this is just a voltage supply point really. Furthermore, Vb itself is relatively arbitrary, it's Vbe that matters. In the simplified schematic, Vbe(Q4)=Veb(Q3).
Because you break the current mirror and ruin the concept of the circuit.
No, the idea is to improve Vbe cancellation by giving Q2 and Q4 matching Vce, and therefore equal temperature.
The Kuartlotron is still a work in progress. The current version is very good.
After you made your first post I went back and did simulations to retrace my steps. There are modifications that decrease distortion. However they create a larger thermal mismatch between Q2 and Q4. Is this counterproductive? I don't know. Simulations are one thing. The current version of the Kuartlotron has been tested in real life and shown to work well. I will not post "improved" circuits until they have been vetted in real life.
That said, the simulator is a very good hint at what the real life behavior of a circuit will be. That's less true for this circuit but still fairly true. I am going to test different versions.
Last edited:
Hey, I woke up this morning feeling absolutely terrible, so I think I was a bit irritated in responding - that wasn't your fault.
I understand about headaches. I've been battling with lifestyle and health for years trying to reduce mental fog and confusion in general (debugging the human CPU). We could compare notes if you'd PM me.
I understand about headaches. I've been battling with lifestyle and health for years trying to reduce mental fog and confusion in general (debugging the human CPU). We could compare notes if you'd PM me.
Hi,
well then, best wishes and recover soon
I admit I always had difficulties with those error correction concept and am mostly better with things I like or trust in.
Had built the Hawksford Vbe corrected I/V stage a couple of years ago and compared that to a very simple Jocko-style grounded base stage.
While the Hawksford sounded very well, the Jocko sounded in direct comparison more real to me.
Iirc the THD of the Hawskford was no better, as I managed to achieve a THD of -90dB@fs (PCM1795 DAC) with just 1 NPN as grounded base and a NJFET as CCS.
The maybe lower input impedance of the Hawksford circuit didn´t pay off sonically in that case.
The simpler circuit won sonically over the more complicated and technically more advanced circuit and this seems to be the case all so often.
jauu
Calvin
well then, best wishes and recover soon
I admit I always had difficulties with those error correction concept and am mostly better with things I like or trust in.
Had built the Hawksford Vbe corrected I/V stage a couple of years ago and compared that to a very simple Jocko-style grounded base stage.
While the Hawksford sounded very well, the Jocko sounded in direct comparison more real to me.
Iirc the THD of the Hawskford was no better, as I managed to achieve a THD of -90dB@fs (PCM1795 DAC) with just 1 NPN as grounded base and a NJFET as CCS.
The maybe lower input impedance of the Hawksford circuit didn´t pay off sonically in that case.
The simpler circuit won sonically over the more complicated and technically more advanced circuit and this seems to be the case all so often.
jauu
Calvin
Hi Kean, hi Joachim
I just post the last messurement we made whilst I could use the audioprecision during my stay in Vienna
Greetings!
Martina
View attachment KeanToken SMPTE.pdf
Just tu specify SMPTE
from AP High Performance Audio Analyzer & Audio Test Instruments : Service & Support
SMPTE (DIN)
SMPTE IMD is a technique for measuring IMD (intermodulation distortion) according to the SMPTE RP120-1983 standard. The DIN intermodulation distortion technique uses a similar method.
The stimulus is a strong low-frequency interfering signal (f1) combined with a weaker high frequency signal of interest (f2). f1 is usually 60 Hz and f2 is usually 7 kHz, at a ratio of f1:f2=4:1. The stimulus signal is the sum of the two sine waves. In a distorting DUT, this stimulus results in an AM (amplitude modulated) waveform, with f2 as the “carrier” and f1 as the modulation.
In analysis, f1 is removed, and the residual is bandpass filtered and then demodulated to reveal the AM modulation products. The rms level of the modulation products is measured and expressed as a ratio to the rms level of f2. The SMPTE IMD measurement includes noise within the passband, and is insensitive to FM (frequency modulation) distortion.
The APx500 implementation of SMPTE IMD provides the capability to vary the stimulus frequencies and to choose a 1:1 stimulus ratio.
I managed to set it up again today and still couldn't measure more than .0003% THD (-90db) across the spectrum. This is with no input pot, and using Fairchild transistors from China, with 600mV RMS input.
I'm not sure how this AP metric relates to THD, so I don't know how to interpret your results. I guess I will do some comparative simulations.
I'm not sure how this AP metric relates to THD, so I don't know how to interpret your results. I guess I will do some comparative simulations.
Last edited: