yes
after building this toy, which is particular VFA but V to I amp
and after modifying some Creative 2.1 computer speakers set in the way that the satelites are driven from V to I amp. I started to enjoy excellent sound at my work.
Whatever the design you wont do 1ppm at high frequency
unless you re using some error correction as did Bob Cordell.
My #499 is <1ppm @20kHz with a very simple VFA circuit.
I can see that Diamond IPS CFA may have a limit that makes 1ppm impossible regardless of complexity. Can someone confirm Joachim's statement?
I'm not sure the simple CFA as VSSA in my #502 has such a limit limit though the complexity required may be more than I'm prepared to use. Presently, adding 2 BJTs to enhance the VASs brings it under 10ppm 20kHz but I don't think this is repeatable with different devices.
I'd be happy if it is reliably under 10ppm @ 20kHz and well behaved in other respects with this level of complexity. I'm disappointed but not surprised that adding the 2 extra devices improves stuff by less than 6dB.
Bonsai & mcd99k, which of mcd's circuits represents this?Bonsai said:
A problem with 'symmetrical' amps is that for DC bias reasons, CMs are complicated. So a VAS enhancer only has small benefit unlike in a Blameless type where it provides nearly the full extra Loop Gain.
In this discussion, I think using AFEC is cheating 🙂
Kgrlee wrote "In this discussion, I think using AFEC is cheating"
No, no I would use it just for the DC servo function. The 20 dB distortion reduction and c. 30 dB PSRR improvement would be purely incidental to that.
😀
No, no I would use it just for the DC servo function. The 20 dB distortion reduction and c. 30 dB PSRR improvement would be purely incidental to that.
😀
The input emitter followers which are outside the Negative FeedBack loop have the advantage to present a low impedance to the base of the NFB subtracting transistors. This is known to be necessary to benefit of the lowest distortion.
In other schemes, this low impedance has to be provided by a buffer adding a bit of complexity.
kgrlee posted up on another thread one of Jean Hiraga's very sparse CFA designs using a comp JFET input. This structure is great for creating these very simple and elegant designs that Hiraga's is known for, but I cannot help thinking that matching is the limiting factor - and by that I mean over the whole input operating range. Easier in BIP, but not with JFET comp pairs.
Current Mirrors can be OK, just not combined with a beta enhancer.
Edmond's SuperTIS has no current gain at all but very sensitive to stray capacitance and the like.
Seems just one beta would keep the impedance levels down without excessive sensitivity to bias problems. Kind of a happy medium.
It's my current study project. I have an attraction to symmetry, to call my comments "objections" as you did, is not quite accurate. I just wanted to point out the issue to Paul.
Best wishes
David
Finally time to respond...
Bonsai, I can't say from a performance point of view whether the bootstrapping in my version of the input diamond is better / worse than your version with the 15V zener reference. Certainly you version is less complex and also provides 15V supplies for your AFEC or DC servo. Would need to do some sims to come to any meaningful conclusions.
The rationale behind my choice of topology was simply to give the four transistors similar operating conditions. This was from a thermal point of view and Vce. Maybe the next reason is incorrect but I thought that if all the transistors were operating in the same conditions problems like quasi saturation would cancel out.
Regarding using the CCS instead of resistors from a purely THD point of view there was next to no difference in sims. Difference was in the order of ppbs. The main gain would be most likely be PSRR. Complexity, in my opinion, of this kind in an amp is no problem since it doesn't complicate PCB layout and parts cost is minimal.
Kgrlee, the circuit we are talking about is posted earlier in this thread. Its the input stage isolated. If I posted the whole circuit there would be many complaints about complexity.
http://www.diyaudio.com/forums/solid-state/240712-cfa-topology-audio-amplifiers-68.html#post3611008
David, I would like to thank you for pointing this problem out with symmetrical compensation. I was aware of bias issues with a symmetrical amp but not of compensation issues. Another of Edmond's ideas is the CMCL (Common mode loop control) which deals with VAS vs VAS bias issues. Since my current main interest is in CFA compensation it has added a new requirement.
Regarding the enhanced VAS I found that the performance of my circuit in sims is destroyed when it is changed to a single transistor. There must be more factors involved here rather than just the CFA topology.
That was a bit of a ramble.
Paul
Bonsai, I can't say from a performance point of view whether the bootstrapping in my version of the input diamond is better / worse than your version with the 15V zener reference. Certainly you version is less complex and also provides 15V supplies for your AFEC or DC servo. Would need to do some sims to come to any meaningful conclusions.
The rationale behind my choice of topology was simply to give the four transistors similar operating conditions. This was from a thermal point of view and Vce. Maybe the next reason is incorrect but I thought that if all the transistors were operating in the same conditions problems like quasi saturation would cancel out.
Regarding using the CCS instead of resistors from a purely THD point of view there was next to no difference in sims. Difference was in the order of ppbs. The main gain would be most likely be PSRR. Complexity, in my opinion, of this kind in an amp is no problem since it doesn't complicate PCB layout and parts cost is minimal.
Kgrlee, the circuit we are talking about is posted earlier in this thread. Its the input stage isolated. If I posted the whole circuit there would be many complaints about complexity.
http://www.diyaudio.com/forums/solid-state/240712-cfa-topology-audio-amplifiers-68.html#post3611008
David, I would like to thank you for pointing this problem out with symmetrical compensation. I was aware of bias issues with a symmetrical amp but not of compensation issues. Another of Edmond's ideas is the CMCL (Common mode loop control) which deals with VAS vs VAS bias issues. Since my current main interest is in CFA compensation it has added a new requirement.
Regarding the enhanced VAS I found that the performance of my circuit in sims is destroyed when it is changed to a single transistor. There must be more factors involved here rather than just the CFA topology.
That was a bit of a ramble.
Paul
Bonsai
Thanks for the link providing the typical Prof. Sergio Franco's refreshing approach.
-Antonio
Again, up-down fold the schematic and look at the result. How is the "bootstrapping your input buffers to the complementary level shifter emitter" looking now?
Hint: it's a CFP, and the local feedback loop helps linearizing the response.
I tried that a few years ago and did no get any significant improvement over simply tying the collectors to the opposing rail (PPB) and using simple Zener regulation - hence the question to mcd99uk. Maybe you want to sim it and you might find an explanation you can share
Early voltage?
ru?
rc?
Quasi sat?
Early voltage?
ru?
rc?
Quasi sat?
Last edited:
Distortion cancellation Amp -
Before i forget, after this round with CFA and the next round with C-Conveyers and then the next round with distortion cancellation... and then combining it all together..... (sorry for the slight jump ahead).... see patent US5166637 A; App number US 07/852,994. And, then the design S.W. (SWOPA) put up in DIYAudio... Discrete Opamp Open Design forum.
Just had to drop this in somewhere for reference, later.
Thx-RNMarsh
Before i forget, after this round with CFA and the next round with C-Conveyers and then the next round with distortion cancellation... and then combining it all together..... (sorry for the slight jump ahead).... see patent US5166637 A; App number US 07/852,994. And, then the design S.W. (SWOPA) put up in DIYAudio... Discrete Opamp Open Design forum.
Just had to drop this in somewhere for reference, later.
Thx-RNMarsh
Last edited:
This design is not stable , your optimistic THD figures are due
to unreasonable GNFB ratios at high frequencies with gain
margin being barely 3dB if not 1dB with an accordingly phase margin.
Bonsai - I have tried to compare input stages but inexperience and a general confusion is occurring. I think I need some help to establish an LTspice setup that would allow an apples vs apples comparison to be made. Just changing the feeds to the collectors disturbed the whole setup of the amp and did not allow a fair comparison.
I've also been thinking about this 1ppm proposed limit. Firstly, just to confirm 1ppm = 0.0001%. Have done a 1kHz sim on my amp design at 100W/8R output and get 0.000011%. Is this suggesting that loop gain / feedback can be used to overcome this barrier at 20kHz and that the diamond input is not the limiting factor yet? It's more likely the output stage remains the bigger problem?
I've also been thinking about this 1ppm proposed limit. Firstly, just to confirm 1ppm = 0.0001%. Have done a 1kHz sim on my amp design at 100W/8R output and get 0.000011%. Is this suggesting that loop gain / feedback can be used to overcome this barrier at 20kHz and that the diamond input is not the limiting factor yet? It's more likely the output stage remains the bigger problem?
kgrlee said:
These are the Loop Gain & Nyquist for Blameless6 in #499 using the Tian probe I show. How did you measure gain & phase margin?
The amp meets Bode & Cordell recommendations for PM but not GM. But GM & PM show only 2 points on the Nyquist plot. Looking at this, which is the true measure of 'linear' stability, shows that both points are roughly equidistant from (-1,0) ie stability is 'well balanced'.
As 20kHz THD is at 0.6 ppm, you can increase these stability margins a lot and still stay below 1ppm. The amp will then easily meet Bode & Cordell recommendations and fall well within Bonsai's arbitrary 3MHz limit.
There are subtleties to do with the correct use of R22 & C9 which I won't go into here. You can see from the shape of the LG, this isn't fully optimised. It was in fact an artifact from when I used lesser output devices which still met 1ppm @ 20kHz
More important than these 'linear' measures of stability is to try all sensible and stupid loads with .TRANS to see if you can provoke instability. For a power amp, pure reactive loads from 100p to 100n are easy to do and important.
Even more important is to do all this in 'real life' too including a big guitar speaker at different levels & frequencies.
I have a formal procedure to do 'real life' capacitive loads quickly but not for the guitar speaker. Fortunately, a power amp is unlikely to burst into flames with the guitar speaker while an unstable amp is likely to do so on capacitance 😀
___________________
WHAT WOULD I DO IN REAL LIFE
Layout, decoupling and earthing are likely to be MUCH more important than circuit well before we reach 1ppm. If there is only a small handful of people who can design 1ppm SPICE stuff, there is an even smaller number who can do layout, decoupling and earthing such that 'real life' approaches SPICE world in this respect.
I don't think I can measure 1ppm 20kHz this Millenium (though I might take Marshy up on his offer) For a commercial amp (or even a 1 off) I'm likely to abandon stuff that few, if any, people can see. Top of the list would be the input cascodes which finally pushed THD 20kHz below 1ppm and maybe even the VAS enhancer. 🙂
This is justified as even Guru Wurcer's SWOPA only shows some harmonics (not THD) at 1ppm level and then only for stuff like voltage followers.
Sorry for all dis off topic VFA BS
PS I can post the .ASC for those who want to check eg the inner Cherry loop too. All models are Cordells.
___________________
mcd, I'm not proposing 1ppm as a limit or even a target. Just pointing out that I my eyes glaze over if stuff more complex than my #499 circuit dun achieve 1ppm 20kHz.
If your circuit does this with with an even simpler circuit and meets other important stuff like stability with wonky loads .. then I will grovel at your feet. 😱
To me, it looks like simple CFAs do better than simple VFAs. Can't see how to get better complex CFAs at the moment .. but that's what I'm hoping to get from this thread.
And all this ignores Listening Tests ... except my contention that simpler with the same performance usually sounds better in Double Blind bla bla ..
___________________
There are several myths about CFA vs VFA which should be addressed. One of them is that CFAs are noisier. This is only possibly true with evil Diamond IPS.
Simple CFA IPS like with VSSA and my #500 have less Ein cos at least 2 theoretical & 'real life' reasons.
Attachments
Last edited:
kgrlee,
Quick question, when simulating these loads what do you set series resistance to? When you say pure this suggest 0R, is this correct? I've always used 0.1R.
Edit: The grovelling will be done by me. I don't see much danger of anyone grovelling at my feet. Sometimes, I feel out of my depth when participating in these sort of threads.
Last edited:
I use 0R but it really doesn't matter in the sims or 'real life' if you use 0R1. If there are problems, you will see them in both. Have some sort of signal running eg near clipping 20kHz. This exercises the amp at various parts of its operating cycle.
Also good to see what happens with mild and severe clipping with wonky loads.
I side with Bob in thinking that overload and recovery behaviour may be the most audible characteristic between amps. Many (all?) Golden Pinnae amps show signs of oscillation on real speakers when overloaded. 😱
No wonder they sound different. 🙂
Thank you for the clarification. I happen to agree with amps having to be very good under overload conditions. I would also add simulating with both voltage clipping occurring at the same time as overload (SOA) protection kicking in. Unrealistic in reality but in my very limited experience can show
problems in the design e.g. oscillations can occur in an otherwise stable amp.
Remove the output LR R19/L1 network , replace it by a short circuit
and re do your sims.... 😉