I see in this model a 2.5% distortion. It is very much. To reduce the resulting distortion to less than 1 ppm, the required total feedback depth of 70 dB. It is unclear, what a bad conventional three-stage emitter-follower, which under the same conditions can make only 0,02...0,03% distortion, and for the same resulting distortion need feedback depth of 30 dB.
In your model supply voltage 2 x 30 volts. It follows that: the power output, undistorted limit of supply voltage reaches 52 W at the amplitude of the output voltage of 25 volts.To see the linearity of the amplifier at the limit of the linear operation zone, the original model was modified to achieve such power output, and was set to the appropriate input voltage. When the output voltage is 25 volts and output power 52 watt THD = 2.5%. Anyone who would do the same, get the same result.
I don't catch the logic of this reasoning (and I am neutral regarding the issue of triple CFP or darlington, or whatever), but it is wrong: with 30V rails, it is simply impossible to arrive at 2.5% distortion for the followers alone, irrespective of their flavor. Even if you deliberately sabotage them, like omitting a spreader, it will be difficult
Attachments
I didn't either. I know they changed a little between sims but the largest component was about 60dB down so - am I wrong? - that is 0.1%. It's one thousandth of the initial signal, is it not? The other peak could only double that at best and, though I can't see from where I am typing here, we were mostly looking at 20 odd dB down on that, so a twentieth or a tenth again at worst. 0.25% might be reasonable but not 2.5%. It's still not ideal.
I plan to look at this over the weekend with slightly grander transistors. Probably MJL3281 and MJL1302, or similar at a lower price. Bob Cordell's models are freely available on his site but I have taken the liberty of uploading them here so you and Pawel can be sure of having access to them. They also happen to be very good, though we won't get the full panoply of CFP problems in any sim. But we can probably get to a solution that is least likely to cause trouble.
It's getting late here in the UK so I'll pick this up sometime on Saturday.
All the best.
Hi!
Thank you, I have the models from Cordell, even plus
. Rechecked several times. The same: 2.5 per cent. Here, look.The difference with the discussed model of the single-stage push-pull follower with no bias in that there is applied the emitter follower. That is, a cascade with 100% local negative feedback. In the discussed example, the output stage performs the function of amplifier current and voltage, i.e., the NFB can not be 100%, but only 20%, judging by the increased voltage of about 5. In addition, the amplification voltage distort the signal much stronger than the followers, and especially - with CE.
To clarify the context: I just looked at the various follower options, and the D-in, CFP-out looked promising from a stability POV. That's all, as far I am concerned.
The context was as a pass element for a supply, which explains the resistor's values I included, for realism.
If the circuit is used in the context of an audio OP, some adaptations would be required, but basically I would expect it to perform very similarly to a conventional triple D, with some minor specificities, like HF stability, linearity, bias voltage.
If a triple D is used in the sim above (with due adaptations), it will certainly perform very similarly, maybe a little better or worse, but nothing overwhelming.
Anyway, the circuit needs to be explored more in depth than what I have done: I just opened a possibility, and it may be marginally better than conventional solutions, or complete crap......
The context was as a pass element for a supply, which explains the resistor's values I included, for realism.
If the circuit is used in the context of an audio OP, some adaptations would be required, but basically I would expect it to perform very similarly to a conventional triple D, with some minor specificities, like HF stability, linearity, bias voltage.
If a triple D is used in the sim above (with due adaptations), it will certainly perform very similarly, maybe a little better or worse, but nothing overwhelming.
Anyway, the circuit needs to be explored more in depth than what I have done: I just opened a possibility, and it may be marginally better than conventional solutions, or complete crap......
Crap - you can say this is the most correct word, when see in the output stage the cascade of a CE. At the time, how hard is it to mess up the standard three-stage follower, so that it distorted the signal more than 0.03%, even at low current bias, CE, as we have seen, will ruin the signal is much stronger.
Maybe crap, but any other triple variant including darlington will be exactly identical when tested in the same conditions, with the same components.
A better test would be a classical follower OP stage: there, differences could begin to mean something.
For a power supply ballast, all of that is mostly irrelevant: stability and output impedance are far more important. The D-in version seems to offer a good tradeoff in this regard (but it has to be thoroughly evaluated)
Did you do any more on this?
I looked at similar circuits a few years back, in the context of class B output section so I was more concerned about how the complementary halves fitted - not analysed in this thread.
But I did compare frequency response and phase of a triple EF with a CFP + EF, similar to Elvee's, and was a little disappointed in the CFP+EF.
The poor performance of the versions with a Common Emitter as the first transistor is explicable as feed-forward thru the base-emitter capacitance.
This can cause non-minimum phase behavior, visible in the red trace.
I never tried Elvee's last circuit, I expect it would be problematic in class B distortion, too sharp transition from one half to the other.
Looks nice for class A but my environmental conscience bothers me too much to use that.
It's possible he has been a bit lucky and that some of the internal capacitances just happen to be about the correct value for optimum compensation for that particular combination of transistor models.
I have seen that in another amp that performed anomalously well, but probably the circuit is more tolerant as well.
I am interested to see how it works with the Cordell models.
Best wishes
David
I didn't get as far as doing any significant analysis but I did look into who else had used compound triples. One interesting one was Crimson Electrik, whose amps are well regarded, and they used the sandwiched CFP. There were some interesting comments from bepowell (who is on here and who designed them) about having a fast enough main transistor. He also used resistor values closer to what I had in mind and, on the opening transistor, a little bit of rolloff. I haven't looked into how significant that was, or whether it's just a stability tweak.
The output of these things seems to be the geometric mean of the the transistors so the little ones would be working quite hard in practice.
Another place CFPs were used is in the Linn LK100 and Klout. These seem to be very bandwidth limited being >1dB down at 20kHz and still drooping at 12kHz. One of the contributors amusingly described it as "state of the ark". One interesting thing was the Klout using 3xZTX653s in parallel as the driver, connected together by three 12 ohm resistors. Whether this counts as opening the loop I haven't looked into but it would be nice to have that quality and speed of transistor in there. In a similar vein I also want to look at treating the main compound pair as a single and then paralleling them up, driven by the opening transistor.
Like you I wouldn't dream of doing a Class A. I have an A370 here and it's capable of keeping the front room at a very nice temperature with no central heating. I don't know of there being any objections to CFP aside from those that Doug Self has already investigated and there are some advantages. I still can't see how this other fellow is getting 2.5% distortion but I'll put it together in an ordinary push pull and see what happens.
I will come back to this soon. I have some IT issues to deal with and one of my own amplifiers to repair.
I hope the schematics for the Crimson have uploaded
Attachments
First glance that seems to be the Darlington out which Philips are saying has lower distortion but which seems for us to model with a huge peak (well we can look into that. I have to nip out for an hour or so but this needs more examination to see if the circuits are in fact the same. I couldn't see anything different right away but in each of the four or five that I have seen, they have all been drawn differently - and none of them the way that I would.
I'll read the whole patent if it's all available on Google. Thanks for that.Nice joke but poor numbers, even for the time.
Similarly, B. Powell seems nice and helpful but the Crimson also shows it's period with a sort of quasi-complementary output.
Crossover distortion is the main distortion in a competent class B amp, so the main issue is how well the two halves "fit" at the crossover point.
And CFP doesn't seem to help here, it just pushes the non linearities into a narrower band and makes bias points more critical.
So that's what I would look at first then, if you can find a combination that looks worthwhile, look at frequency/phase response.
Obviously different for a regulator.
Luck with the IT and amp repairs, both on my list too.
Best wishes
David
I had a quick look at the patent last night and it seems that figures 6, 7 and 8 are the ones they are claiming for. It gets very complicated with the patent language and I reckon I'd need a full clear hour to be sure of what they after really doing.
In the circuit above we are getting something of a current source with the Vbe drop of T3 going across that resistor, though that will change with the voltage at its base. On the patent there seems to be some sort of bootstrapping going on with a capacitor going around it. I really would need to draw these out by hand to see what's really going on. And without an explanation I bet I won't be sure even then!
I got the full patent with just a search with the patent number and I think it was Freepatentsonline that was at the top and did give me the whole thing. Sometimes they make it difficult to get at the drawings. I may have a chance to have another look at it tonight but that won't be a good concentrated session so any pointers as to what is going on would definitely help.
Edit PS. Incidentally I saw this one as effectively a speeded up driver, with T3 just making T1 better behaved.
Last edited:
Hi,
just attached a own sim, based on the schematic of #19.
First I used same models but switched to other transistors I´d prefer in a power output stage.
2nd pic shows the amplitude response ithout compensation caps, the 3rd pic with comp. caps.
The caps values depend on the chosen transistors and might become larger or may be omitted with alltogether with differnet types.
But obviously it doesn´t require much effort to tame the amplitude response.
Pics 4 to 6 show the THD @1kHz with 100mVrms, 1Vrms and 10Vrms input voltage.
It seems that as soon as he output transistors approach class-A limit and the stage slides into class-AB that the higher harmonics shoot up.
That comes a bit disappointing to me as I had expected better from my experience with simple 2-stage CFPs.
jauu
Calvin
ps: @ padamiecki:
the file is write protected, the ops contain doubles -that stop tran from working, V(out) is not placed where one would expect and the dimensioning is not well -check the currents with .tran simu)
just attached a own sim, based on the schematic of #19.
First I used same models but switched to other transistors I´d prefer in a power output stage.
2nd pic shows the amplitude response ithout compensation caps, the 3rd pic with comp. caps.
The caps values depend on the chosen transistors and might become larger or may be omitted with alltogether with differnet types.
But obviously it doesn´t require much effort to tame the amplitude response.
Pics 4 to 6 show the THD @1kHz with 100mVrms, 1Vrms and 10Vrms input voltage.
It seems that as soon as he output transistors approach class-A limit and the stage slides into class-AB that the higher harmonics shoot up.
That comes a bit disappointing to me as I had expected better from my experience with simple 2-stage CFPs.
jauu
Calvin
ps: @ padamiecki:
the file is write protected, the ops contain doubles -that stop tran from working, V(out) is not placed where one would expect and the dimensioning is not well -check the currents with .tran simu)
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.