I continued to play around with the Darlingtons and I got it..
simply awesome.. I had to model 2 transistors toemulate a KSA 700/800 pair (fairchild describes the internals
but no stinking model
)or bd 677/678.... 3db more gain ,same unity gain/phase. plot..MUCH better LTP balance (1-2 uA).
But my concern is the massive Cob of these devices..100pf!!
All in all .. this is afrugal mod as the BD's and the KSA
darlingtons cost $.67c Vs $.40c for the 1381's..
OS
Attachments
Hi OS,
Just a word of warning about combining a Darlington and cascoding with a VAS.
The Miller compensation capacitor makes the VAS a feedback amplifier by itself. Its output voltage is sampled and the feedback signal is the current through the Miller cap. When you replace a CE amp with a Darlington in a cascoded VAS, the loop gain of this local Miller loop is increased, and this can lead to oscillation in some cases. If it occurs, it will be in the range 10 MHz - 50 MHz or so. There's a way to look at the loop gain of this local Miller loop in SPICE, determine its phase margin, etc. I can post how to do this if you're interested.
Just a word of warning about combining a Darlington and cascoding with a VAS.
The Miller compensation capacitor makes the VAS a feedback amplifier by itself. Its output voltage is sampled and the feedback signal is the current through the Miller cap. When you replace a CE amp with a Darlington in a cascoded VAS, the loop gain of this local Miller loop is increased, and this can lead to oscillation in some cases. If it occurs, it will be in the range 10 MHz - 50 MHz or so. There's a way to look at the loop gain of this local Miller loop in SPICE, determine its phase margin, etc. I can post how to do this if you're interested.
Maybe carlos has been using the darlingtons for the wrong purpose or the wrong way, I wouldnt dream of using them in small signal areas without a cascode nor would i dream of using something like a bd140 as a vas. Carlos also says class A amps sound muffled, this is purely preferencial taste, me thinks he likes the high frequencies cross over distortion artifacts which I can hear too and is percieved as more detail and high frequency which is not necessarily that bad but it puts a edge on human voices I would prefer not to have. Yet again one mans muffled is the others normal and ones bright is anothers normal. OOps I said I wasnt going to mention anything subjective anymore.
He is a guy though, this forum needs more people like him, I love watching his clips on youtube, sometimes brings a huge smile to my face.
Even though the cascode should take care of cob issues I find it better to have these already as low as possible too, well its obvious, internal nonlinear capacitances are detrimental everywhere, havent found any suitable complementary models yet, Ill wait for the factory. The cob issues offsets some of the gains achieved with the higher gain obtained from the darlington.
That darlington you built up there will surely be on steriods
But its an extra part, if you going that route its mandotory trying a cfp too. I will do some comparisons as soon as I get some reliable darlington models from japanese factory. I think the cfp might have the legs on this one. Try it, I will first have to make a complete schem on LT. This amp has very high gain, beats my blameless but its not so linear, YET. Try using higher degeneration resitors and decrease vas miller slightly.
Some say amps with this topology sounds ^%$#
Again subjective preference. In the end you must listen and decide.
BTW I notice you use trannies like mje340 for current sources,
. This messes up the PSRR, a good read on this is Walt s papers on current sources. Maybe only a couple of cents difference to a 1381, lowest cod high gain device should be used, this for mirrors too.
He is a
guy though, this forum needs more people like him, I love watching his clips on youtube, sometimes brings a huge smile to my face. Even though the cascode should take care of cob issues I find it better to have these already as low as possible too, well its obvious, internal nonlinear capacitances are detrimental everywhere, havent found any suitable complementary models yet, Ill wait for the factory. The cob issues offsets some of the gains achieved with the higher gain obtained from the darlington.
That darlington you built up there will surely be on steriods
But its an extra part, if you going that route its mandotory trying a cfp too. I will do some comparisons as soon as I get some reliable darlington models from japanese factory. I think the cfp might have the legs on this one. Try it, I will first have to make a complete schem on LT. This amp has very high gain, beats my blameless but its not so linear, YET. Try using higher degeneration resitors and decrease vas miller slightly.
Some say amps with this topology sounds ^%$#
Again subjective preference. In the end you must listen and decide. BTW I notice you use trannies like mje340 for current sources,
. This messes up the PSRR, a good read on this is Walt s papers on current sources. Maybe only a couple of cents difference to a 1381, lowest cod high gain device should be used, this for mirrors too.
ostripper said:
Exactly. And that's one of the problems. I'll summarize the problems below.
1) The Miller loop is inside the global feedback loop. If you inject a signal inside the Miller loop, it will go around this loop and the global loop. The global loop feedback path will cause the simulated Miller loop gain to be completely wrong. We need to block the global feedback path, yet still allow global DC feedback to set the operating point. See the attached picture for how to do this.
2) For the case of a complementary VAS, there are two Miller loops. If a loop gain probe is put inside only one of these loops, completely wrong simulated loop gain will result. For a fix, you must artificially join the loops together per the attached picture.
3) The impedance interactions within the Miller loop make the single-voltage-source method of measuring loop gain (from the LTspice audioamp.asc example) totally inaccurate. Instead, the full-up loop gain probe from the LTspice loopgain2.asc example must be used. See this post for details.
So, to solve these, do the following:
a) Block the global loop gain for all frequencies except DC using the LC network as shown in the attached picture.
b) Artificially join the Miller loops together per the attached picture.
c) Use the full-up loop gain probe, place in the artificially-joined path as shown in the attached picture.
Remember to turn off all other AC voltage sources before simulating. If the results are correct, the Miller loop gain magnitude should look something like a bandpass filter with gain (since the Miller loop by itself has no DC feedback).
Edit: In the picture, the inductor should be 1e10 Henries. It's kind of hard to read. Also, the cap should be 1 Farad, but if you specify it as 1F, that's one femtofarad!
Attachments
yes, that is my high Hfe/low Cob "guinea pig" ,10-15pf/10k Hfe.
the other test was with the "dogs" BC546/MJE350.
I modeled the 2 trannies after the KSA700 ,which is more
like the 1381/992 combo.
you were right, a darlington brought the distortion down
another notch..(see below) 5-6 DB to be exact.
That is at 100V p-p output .. I did the same test before
and got 87/92/95 (H2/3/5).
OS
Attachments
I've been playing around myself, and after much fiddling I came to the conclusion whoever it was that designed the Cyrus One/Two knew what they were doing!
www.darkmatter.myby.co.uk/newamp (yes, a highly original name)
Check this out:
This isn't one for the people who don't belive in gain, though!
It's a bit of a pig to stabilise but C1 does the trick.
In case youre wondering why D2 and R3, I intend to use some salvaged 2SA1360/2SC3423, and i have only 4x of the 2SAs and 2x of the 2SCs (the Teac amp from which they were salvaged also used the diode and resistor method). If I go to PCB with this I'll be leaving pads on the PCB for adding the extra transistors.
www.darkmatter.myby.co.uk/newamp (yes, a highly original name)
Check this out:
An externally hosted image should be here but it was not working when we last tested it.
This isn't one for the people who don't belive in gain, though!
An externally hosted image should be here but it was not working when we last tested it.
It's a bit of a pig to stabilise but C1 does the trick.
In case youre wondering why D2 and R3, I intend to use some salvaged 2SA1360/2SC3423, and i have only 4x of the 2SAs and 2x of the 2SCs (the Teac amp from which they were salvaged also used the diode and resistor method). If I go to PCB with this I'll be leaving pads on the PCB for adding the extra transistors.
That single voltage source probe is only an approximation for loop gain measurements, and that approximation completely falls apart when trying to measure the Miller loop gain. That technique assumes the impedance from one side of the probe to ground is zero. This is a good approximation when simulating the global loop gain when one side of the probe is the amp output, but it falls apart in the Miller loop case.
Please read this post for how to use the full-up probe from the loopgain2.asc example in the LTspice Educational directory. If you use the simplified single-voltage-source method you've been using, the results will be very inaccurate.
Sorry about the diagram. Once I shrunk it down small enough for the posting software to accept it, it became hard to read.
Please read this post for how to use the full-up probe from the loopgain2.asc example in the LTspice Educational directory. If you use the simplified single-voltage-source method you've been using, the results will be very inaccurate.
Sorry about the diagram. Once I shrunk it down small enough for the posting software to accept it, it became hard to read.
In terms of distortion, the diode vs a diode-connected transistor doesnt seem to make much difference. I suspect that where it really matters is thermal compensation.
The CFP input stage as Cyrus have it seems to be the key - with this in place, H2 disappears and H3 practically disappears. Oddly enough, attempting to add degeneration resistors on Q1/Q2 ruins it!
This thing has crazy gain though. Careful compensation is definitely needed. The original Cyrus One amp also had a triple stage, although it's also quasi-complimentary. This seems to allow them to use 2SA872/2SC1775 in all stages apart from drivers and outputs, but if memory serves the rail voltage isn't very high.
The CFP input stage as Cyrus have it seems to be the key - with this in place, H2 disappears and H3 practically disappears. Oddly enough, attempting to add degeneration resistors on Q1/Q2 ruins it!
This thing has crazy gain though. Careful compensation is definitely needed. The original Cyrus One amp also had a triple stage, although it's also quasi-complimentary. This seems to allow them to use 2SA872/2SC1775 in all stages apart from drivers and outputs, but if memory serves the rail voltage isn't very high.
Nice Jaycee and MJ
One can see why the cyrus sounds as good as it does. The liitle bump in analisys might be cureable, had same small problem with my blameless design. BTW run the sim again but at high power and at 20 Khz and look at the figures again regarding the two current mirrors.
Tonight I put part of my blameless on simm too, to play around with maybe a cascode in vas or ltp but Ive had mixed results before.
Anyway I atach plot of what is possible with such a simple design, only couple trannys more than blameless, simm at 20KHZ and power shown. Its a little handicapped because I have fets and mosfet in the circuit and only simming it with a cfp driver ef. I use this but with some modifications that lower THD some more but only significant at 20KHZ and higher powers. Before you comment, no its not roenders outputstage, its been used since 1966, do your homework it is shown in a book I cant recall now and JLH has it in a paper dating from 80s. The amp is stabilized with just 10pf miller cap but I use alternative compensation real life, no difference in measured performance but the simm is having some problem with this other compensation, there was no such problem with the test on a scope back when I designed this.
Yep, this is not for the anti GNFB crowd, they would
Hope the pic comes out.
One can see why the cyrus sounds as good as it does. The liitle bump in analisys might be cureable, had same small problem with my blameless design. BTW run the sim again but at high power and at 20 Khz and look at the figures again regarding the two current mirrors.
Tonight I put part of my blameless on simm too, to play around with maybe a cascode in vas or ltp but Ive had mixed results before.
Anyway I atach plot of what is possible with such a simple design, only couple trannys more than blameless, simm at 20KHZ and power shown. Its a little handicapped because I have fets and mosfet in the circuit and only simming it with a cfp driver ef. I use this but with some modifications that lower THD some more but only significant at 20KHZ and higher powers. Before you comment, no its not roenders outputstage, its been used since 1966, do your homework it is shown in a book I cant recall now and JLH has it in a paper dating from 80s. The amp is stabilized with just 10pf miller cap but I use alternative compensation real life, no difference in measured performance but the simm is having some problem with this other compensation, there was no such problem with the test on a scope back when I designed this.
Yep, this is not for the anti GNFB crowd, they would
Hope the pic comes out.
Attachments
Then I replaced the CFP input stage which I dropped earlier as a waste of parts ,(boy ,I was wrong)the CFP eliminated H3 fully.
Seems what little bit of loading occurs from VAS/Cdom has
much less effect on CFP stages. I next put the darlington
into play and that only reduced H2 by a measly DB , but since
this mod is a board direct replacement.. no biggie..
I also tried NO Re on my LTP .. no instability and 88 DB gain..
(can't do 100+ )
I guess I will "readd" my CFP .. only 2 trannies for no H3
is a good deal..
As good as it gets for this..(below)
OS
Seems what little bit of loading occurs from VAS/Cdom has
much less effect on CFP stages. I next put the darlington
into play and that only reduced H2 by a measly DB , but since
this mod is a board direct replacement.. no biggie..
I also tried NO Re on my LTP .. no instability and 88 DB gain..
(can't do 100+
)I guess I will "readd" my CFP .. only 2 trannies for no H3
is a good deal..
As good as it gets for this..(below)
OS
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.