Hi Jan,
Your point here is spot-on. The output stage is often such a limiting factor in ULGF and distortion performance that it probably should be taken out of the picture by using an ideal buffer so as to highlight differences in the VFA and CFA topologies themselves.
Cheers,
Bob
I am sorry also. This latest excursion about cascoding is a further example and reason. If I were an employer, heads would roll for lack of focus. I like running a tight ship. cant do that here, unfortunately.
When I was CEO of an Asean electronics business, I was told by the president that i could not talk so directly to the employees. It would offend them. I said, fine... i'll tell you and the managers and you guys finese it with them.
I have many fun and interesting projects to work on -- like checking out my new distortion analyzer and new WIN 8 PC and doing T&M work.
Maybe another time and place. My hope is Bob Cordell will start a chapter or new book about it..... it is sorely needed and long over due.
Thx-RNMarsh
Last edited:
Jan, I don't think you can make 'meaningful' comparisons by leaving out the OPS. If you use an 'ideal buffer', you are just comparing gain stages into HiZ.
The IPS & VAS of a power amp is MUCH more than that. It has to drive a wonky load, the OPS.
Also IMHO, the real advances of VFAs over simple Blameless has been the understanding and adoption of clever compensation methods like 'pure Cherry' & TMC.
This is what gives my #499 design SOA performance from an old fogey circuit. These methods require integration of the OPS with the VAS & other stuff.
For me, the ease in which a topology can use and benefit from such advanced methods is an important factor. But there are important subtleties if we allow a 'real life' OPS.
I would very much like to use VSSA with its lateral MOSFETs. These would allow greater freedom in the use of 'pure Cherry' as the OPS only introduces 'one' extra pole. But the lack of EKV SPICE models means THD analysis is seriously flawed.
The EF2 OPS I use in #500 & 502 at least has known performance which I can relate back to 'real life'. (I'm hoping when Guru Cordell comes up with his promised EKV models, I'll be able to relate them to 'real life' too. 🙂
VSSA's MOSFET OPS & my EF2 output stage have very different drive requirements. It may be that my simple CFA2 in #502 with its push pull 'symmetrical' topology has advantages over the 'Blameless' topology in the #500 VFA for one or the other OPS. This are all hidden if we use an 'ideal OPS'.
Hence, I think we need to level the playing field with other criteria. My choice is complexity measured by no. of active devices in a fully working 'real life' ready design
_______________
The choice of VSSA's CCSs to bias the IPS is in line with this. Using resistors in a sim shows no degradation but in 'real life', the power rails will sag changing the bias conditions. It's not so much for PSR & THD. In the 'real life' #500 amp, C1 & C2 will be large, dealing with PSR & THD. But rather, the CCSs ensure 'real life' is closer to sim cos important bias conditions don't change.
LC has a complicated procedure for setting the CCSs which I'm not sure is sensible for even small scale production but is OK for one-offs.
_______________
At the risk of blowing my own trumpet on VFAs, I'll point out that in the same thread, I describe what some have called JFET990 with equivalent performance to SWOPA in most respects but simpler, with less current and less HO harmonics.RNMarsh said:
http://www.diyaudio.com/forums/analog-line-level/218373-discrete-opamp-open-design-206.html#post3238467
Definitely stone-age but with SOA performance.
There are slighly improved versions further along the thread but after this initial effort, I was more concerned to meet Marshy's requirement for even less active devices. Linuxguru is about to make 'real life' versions of both SWOPA & JFET990 so we might have some 'real life' comparisons soon.
The #499 VFA is based on this work.
Sorry about this VFA wanking.
_____________
But please, can we have less noise about stuff that doesn't really bear on the thread .. cascodes, ULGF bla bla. In da 21st century, with LTspice, just try these things on your pet design. It either gives an improvement or it doesn't. No need for religious fervour.
In the 'Discrete Opamp ..' thread I cascode the VAS for some versions and not in other. In each case, the problem that needs to be solved is the same; modulated Cob feedback to the base. Cascoding is just one way of solving it.
I'd like to see some arguments for Diamond input over the simple VSSA inputs I favour. So far only Waly's tome on the subject has been useful but I still don't see any advantage for a Power Amp. Some benefits for a 'general purpose' OPA.
which is the core of my argument that CFA isn't necessarily the be-all end-all audio power amplifier design approach
it is the limitations of the power output stage that is most important so it doesn't matter if CFA or some other technique made for "perfect" preceding differencing and gain stage
the limits of global or output inclusive loop gain that can be applied are limited by the output stage speed, distortion by AB output nonlinearity, output stage driving impedance nonlinearity
if Richard Marsh has too little patience for typical forum give and and take, diversions – perhaps he should Blog his cm/cfa storyline instead http://www.diyaudio.com/forums/blogs/all/
Last edited:
Hi KGRLEE -- I am very aware of your contribution to that forum and my original challenge there and conditions to meet. My challenge and criteria was set forth in the beginning of that. Scott took the challenge but a more complex path/circuit for the reasons he gave.
Until there are designs put to physical test, i have to suspend judgement. i know the SWOPA was being built. As there and here... it is the as-built of the design that ultimately matters. It is the proof required. The distortion is said to be at the -120dB level (?) That level is the bench mark I referred to. I would give 'bonus' points for the CMA with the least number of active devices - all else being approx equal.
Thx-RNMarsh
Until there are designs put to physical test, i have to suspend judgement. i know the SWOPA was being built. As there and here... it is the as-built of the design that ultimately matters. It is the proof required. The distortion is said to be at the -120dB level (?) That level is the bench mark I referred to. I would give 'bonus' points for the CMA with the least number of active devices - all else being approx equal.
Thx-RNMarsh
Last edited:
Mr. Marsh, I'm also waiting for LinuxGuru's real life comparisons between SWOPA & my poor efforts at JFET990.
As I've said many times, the only reason we do sims, is we believe they have some resemblance to real life. The only reason I believe my JFET990 will compare favourably with SWOPA is that in each case, the sim improvements have been in line with my 'real life' experience.
Using the same conditions & models, (mostly Scott's & Bob's) SWOPA & my JFET990 appear to have very similar sim performance down to the -120dB products.
I've concentrated on performance at very high 40dB gain cos there is a chance mere mortals might be able to confirm this in 'real life'. It's likely only a literal handful of people in da known universe are equipped with the know how & gear to reliably measure -120dB THD .. let alone the difference in harmonic structure that the sims show as the main difference at 20kHz.
Alas, this beach bum is unlikely this Millenium to assist in the important practical verification .. but he HAS obtained the carcases of 3 large PA amps hoping the transformers & PSU caps are still in good order. Now for some dosh for other bits. 🙂
______________
Back to the main programme ..
I've tried manso's suggestion of using my extra 2 devices to enhance the VAS(s) in my simple CFA2 from #500. As expected, the improvement is slightly less than 6dB.
Ju..ust gets under 10ppm so counts as a win over Blameless11 from #502. But I don't think its repeatable with different devices.
Using triples instead gets a little more improvement but still not even 10dB
______________________
I'm still trying to figure out how to use more sophisticated compensation methods with CFA.
Someone mentioned DTMC. Any links?
Last edited:
Waly
All that is being said is that a cascode has the potential (depending on circuit details) to increase low frequency loop gain.
Indeed with an ideal output buffer the cfa's input linearity "can" result in better thd than an ltp.
Half the time in this thread, even when all agree it appears as an argument.
So many times a post is not read thoroughly enough and replies are less informed than the original post, or rephrased to appear superior
(this is my first negative post ever) .
Hope this helps
-Antonio
All that is being said is that a cascode has the potential (depending on circuit details) to increase low frequency loop gain.
Indeed with an ideal output buffer the cfa's input linearity "can" result in better thd than an ltp.
Half the time in this thread, even when all agree it appears as an argument.
So many times a post is not read thoroughly enough and replies are less informed than the original post, or rephrased to appear superior
(this is my first negative post ever) .
Hope this helps
-Antonio
kgrlee --
I have a fair sized investment in test equipment that I will offer to test anything you would like. Down to at least -150dB levels. -140dB is accurate and recently verified by Demian Martin... using similar equipment. We both live in Northern California. Where is OZ land?
Never-the-less, they are not Current-Mode Amps. And, not power amps. Though the SWOPA has some features of both VF and CF which is interesting to me, personally. It shows a thorough understanding of both techniques.
-Richard Marsh
I have a fair sized investment in test equipment that I will offer to test anything you would like. Down to at least -150dB levels. -140dB is accurate and recently verified by Demian Martin... using similar equipment. We both live in Northern California. Where is OZ land?
Never-the-less, they are not Current-Mode Amps. And, not power amps. Though the SWOPA has some features of both VF and CF which is interesting to me, personally. It shows a thorough understanding of both techniques.
-Richard Marsh
Last edited:
The LF gain is certainly increased. I think if you look at the cascode on its own (but buffer it !) and not buried in the middle of an amp with Cdom wrapped around it, you will find it does enhance gain at HF because it mitigates Cob in the amplifier transistor, effectively killing the Miller feedback.
Another thing to look at here is the loop gain inside the cascode circuit - agsin, on its own and buffered - I think at HF it will be higher than without, and as a result, more linear.
Last edited:
Kgrlee wrote "But please, can we have less noise about stuff that doesn't really bear on the thread .. cascodes, ULGF bla bla. In da 21st century, with LTspice, just try these things on your pet design. It either gives an improvement or it doesn't. No need for religious fervour."
I think we need to discuss these here Kevin.
The cascode may be a little off the track, and perhaps only required once you start to want eke out the last few ppm in distortion. However, in the context of the CFA - VFA comparison, ULGF is absolutely part of the discussion. It's very easy, in a simulator, to close the loop around a CFA (or VFA for that matter) at 10 MHz and claim one is more superior than the other. Hence, my proposal a few pages back to set the ULGF at 3 MHz. When one topology, using whatever comp schemes we care to deploy or new ones we cook up, consistently shows better overall performance then we can declare it the 'winner'.
Don't hold your breath for that though.
I think we need to discuss these here Kevin.
The cascode may be a little off the track, and perhaps only required once you start to want eke out the last few ppm in distortion. However, in the context of the CFA - VFA comparison, ULGF is absolutely part of the discussion. It's very easy, in a simulator, to close the loop around a CFA (or VFA for that matter) at 10 MHz and claim one is more superior than the other. Hence, my proposal a few pages back to set the ULGF at 3 MHz. When one topology, using whatever comp schemes we care to deploy or new ones we cook up, consistently shows better overall performance then we can declare it the 'winner'.
Don't hold your breath for that though.
Kgrlee wrote "I'm still trying to figure out how to use more sophisticated compensation methods with CFA."
I've used TPC on a CFA and gotten very good distortion at 20 kHz at 350 Watts - and it gets very good below that as would be expected. This is in a sim, but I think some of the well tried comp techniques we've been using in VFA are also applicable to CFA.
I've used TPC on a CFA and gotten very good distortion at 20 kHz at 350 Watts - and it gets very good below that as would be expected. This is in a sim, but I think some of the well tried comp techniques we've been using in VFA are also applicable to CFA.
Andrew, this is exactly the sort of stuff I'm after to guide me on my CFA journey. I think this sort of stuff should be what this thread is about; how to make better CFAs.
I've just downloaded SX, NX & AFEC from your websites but these appear to be simple (??) MIC.
Is 'TPC on CFA' on your website?
Some of my problem is more to do with how to apply 'advanced' compensation to a symmetrical amp rather than just CFA.
BTW, its Richard, not Kevin.
______________
Mr. Marsh, thank you for your kind offer. I'm doing some very casual design work on a fancy mike and may also convert the carcases of 3 PA amps into one good one before the end of the Millenium. Supa dupa discrete OPAs don't really figure in either plan.
May I pass on your offer to Linuxguru who is presently making up examples of both SWOPA & JFET990?
______________
No swearing please 🙂wahab said:
That s right.
The only valuable contribution to the CFA debate has been brought
by Manso who produced a functionnal schematic with good caracteristics.
As for the disgression about cascodes , well , it could had been
easy for people in the know to reduce the length of this debate
by stating the obvious, but it didnt occur, for some resons....
Nice. So I'm talking about a cascoded gain stage properties (in isolation), while I am told about the loop gain of a blameless amplifier, including a cascoded VAS with a current source load, and then wondering why the LF gain and the PSRR increases. Talking about an apple to apple comparison... This thread deserves to be renamed to "ADD effects in audio".
Yes manso did, but see my comments about closing the loop at 3 MHz.
I've done a lot of breadboarding and testing with cap loads on both VFA and CFA, and as a number of other practitioners on this forum have noted, the O/P stage sets the upper ULGF. There is the not too insignificant problem of layout parasitics as well, and these become especially troublesome in fast, wide band amps unless you take a lot of care.
Ignore the 3 MHz upper limit at your peril. If it were not a problem, there would not be all ths effort over the last 20 years or so to come up with new, sophisticated comp schemes. We'd all just close the loop at 7 MHz or higher and be done with it.
Besides optimizing the front end for best linearity, I see comp design and testing as the other big thing that needs to be explored in CFA. Most designs I have looked at use shunt comp or Miller (Homemodder was right on that one - it does work).
I've started playing with TPC, then there's TMC to look at next. If CFA holds the promise of very high slew rates, wide bandwidth and lower IMD, then we are going to have to decide how we will deal with the slow output stage, and that is likely to lead to nested feedback schemes with suitable frequency transitions. I think Edmond has done quite some work here.
I've done a lot of breadboarding and testing with cap loads on both VFA and CFA, and as a number of other practitioners on this forum have noted, the O/P stage sets the upper ULGF. There is the not too insignificant problem of layout parasitics as well, and these become especially troublesome in fast, wide band amps unless you take a lot of care.
Ignore the 3 MHz upper limit at your peril. If it were not a problem, there would not be all ths effort over the last 20 years or so to come up with new, sophisticated comp schemes. We'd all just close the loop at 7 MHz or higher and be done with it.
Besides optimizing the front end for best linearity, I see comp design and testing as the other big thing that needs to be explored in CFA. Most designs I have looked at use shunt comp or Miller (Homemodder was right on that one - it does work).
I've started playing with TPC, then there's TMC to look at next. If CFA holds the promise of very high slew rates, wide bandwidth and lower IMD, then we are going to have to decide how we will deal with the slow output stage, and that is likely to lead to nested feedback schemes with suitable frequency transitions. I think Edmond has done quite some work here.
Last edited:
There isn't a lot out there about DTMC. It is the something that Edmond used on a couple of his amps.
I have "stolen" his idea to use on my current prototype CFA design.
Here's a link to one of Edmond's amps:
MCP Front-End
The DTMC is made from C3, C4, C5, C9, C10, R20, R28, R29. R28+R29 is to allow for the fact that his output stage has gain.
There is also some shunt (lead lag) at the TIS outputs C7, C8, R35, R36.
Last edited:
Here is a link to a copy of the Alexander amplifier. It's good general reading on a unique CFA implementation. I am posting it up because at the back of the app note he discusses the comp scheme in quite some depth.
http://hifisonix.com/wordpress/wp-content/uploads/2012/09/The-Alexander-Amplifier.pdf
http://hifisonix.com/wordpress/wp-content/uploads/2012/09/The-Alexander-Amplifier.pdf