Pete,
Glad to see you continued to work on this and refine your thinking while everyone else was getting weird about what was up with you. A three pair MT-200 would seem to cover so many applications except for those with really inefficient multi-way systems with passive networks. Those people can stick with the 5 pairs boards. It would also be great if you could use some of the ideas that have been shown for the narrower boards where the output devices have been moved closer to the center line of the boards. Adding the input section to the board will save a bunch of wasted space and the connectors that no longer would be required. SMD on the input side would reduce the size even more if that is practical. I've just been sitting back and waiting for you to return, I had no concern you would just disappear off the grid. Great to have our leader back with some practical thinking.
Glad to see you continued to work on this and refine your thinking while everyone else was getting weird about what was up with you. A three pair MT-200 would seem to cover so many applications except for those with really inefficient multi-way systems with passive networks. Those people can stick with the 5 pairs boards. It would also be great if you could use some of the ideas that have been shown for the narrower boards where the output devices have been moved closer to the center line of the boards. Adding the input section to the board will save a bunch of wasted space and the connectors that no longer would be required. SMD on the input side would reduce the size even more if that is practical. I've just been sitting back and waiting for you to return, I had no concern you would just disappear off the grid. Great to have our leader back with some practical thinking.
Being in isolation for a while made me look at all these designs from a different perspective.
The symasui is just 1/2 of a spooky with a wilson mirrored VAS. One
would think they would perform equally. BUT .... vocals are better
on the symasui ??? Why ? Odd order cancellation and superior PSRR
seem to be something that is actually noticeable while listening.
While biasing my slew output stages with PC scope and other measurement
devices , I noticed the ND especially would give the lowest 3rd/5th all
the way down to a VERY low bias (20ma per device). What is actually
going on here ? The fastest IPS's are actually (error)
correcting the non linear output stage crossover region directly.
So , this is the "magic" of the CFA. Combined with the fast slew OPS , an
IPS like the ND will nearly eliminate the odd order class AB artifacts.
Another factor I explored is the load (speaker). A woofer presents quite
a "wild" impedance variation - especially in a ported enclosure. Most tweeters
will just be like my DC28 daytons (5-8R ). Very tame. Modelling these
drivers shows that a VFA is quite immune to these load variations.
So , a fast VFA is the best full bandwidth performer but a CFA will always
be the top HF performer.
I auditioned the ND's for the longest time - flawless on any speaker. A little
weaker on my sub.
(below picture) The ND even cancels odd order MORE at 35ma output stage bias !
The OPS "suckout" cap is another player (C114). The output Cob,base-stopper,
and C114's R/C will affect the crossover distortion in the OPS.
.68uf is ideal. .33 uf to.47uf will reduce H3/5 >10K a little - but increase
it at 1K.
But , whatever the AB "glitch' looks like .... ND's speed will still negate it.
BTW , symasui also has this ability - but needs a little more bias to do it.
I Have my test OPS and plenty of ND's to send to RN Marsh (soon).
This is why I'm quite sold on further developing the Sym for VFA and ND
for CFA.
OS
Attachments
Hey Pete,
Good to see you back in force. Now I am going to that to hook up some IPS's to the OPS's and see if I can hear what you are reporting. It has been a while since I have had both OPS set to compare the different IPS. I usually just compare them to the amps I have in the stack. I got too busy building stuff so I needed the bench space. I know I liked the ND a lot but I have really liked Spooky too. I'm curious to put them up sided by side and see if I can hear what you hear. I'll try to do that tomorrow.
Blessings, Terry
Good to see you back in force. Now I am going to that to hook up some IPS's to the OPS's and see if I can hear what you are reporting. It has been a while since I have had both OPS set to compare the different IPS. I usually just compare them to the amps I have in the stack. I got too busy building stuff so I needed the bench space. I know I liked the ND a lot but I have really liked Spooky too. I'm curious to put them up sided by side and see if I can hear what you hear. I'll try to do that tomorrow.
Blessings, Terry
Pete,
If I'm following you correctly you are saying at the port tuning frequency that the Spooky has an easier time with the load from the impedance at resonance and the phase angle of the same? It makes me wonder if there was a Zobel or LCR that flattened the impedance at resonance if you would still have the same concern or that both amplifiers would then sound virtually the same?
If I'm following you correctly you are saying at the port tuning frequency that the Spooky has an easier time with the load from the impedance at resonance and the phase angle of the same? It makes me wonder if there was a Zobel or LCR that flattened the impedance at resonance if you would still have the same concern or that both amplifiers would then sound virtually the same?
Finally !
I'm so glad you confirm what I try to preach since decades ! This, the reduced phase shift at HF, and the reduction of IM in the audio band produced by very high order harmonics. And the expansive behaviot of CFA topolgy witch help during fast high level transients.
Even for an old man like me, witch had the luck to met some very famous designers, you really impressed-me. You learned so fast, if, like I believe, you don't had so much experience with CFAs before to initiate this "rumble".
Not to forget your incredible expertise for designing...even complex amplifiers... so fast.
As long as I know, the study you just made, comparing all those amps with a common OPS, had never been done before.
Hats off!
Last edited:
Yup , A VFA makes a superior sub amp. If it weren't for the lower voltage PS in my sub ,
the spook would have been in there from the beginning.
After running my sub for a while , I did put one of my (now un-used) spooks
into the sub amp - perfect. Bass to die for.
The slew amp is such overkill for my Tang band 10" ... did not matter that
I lost 6V of headroom with the cascoded spook.
I have been testing both the Sym and ND on both my AB123 and Dayton/mission full range loudspeakers.
Sym can actually faithfully reproduce some of my 64 bit DSP hall reflections
with the ultimate clarity. ND ... the same but a little bit "brighter". When
I extend those same reflections >5Khz ,the ND exceeds the SYM.
Not to denigrate the spook and wolverine , but they audibly can not
reproduce these hall reflections like the ND/Sym.
Kind , I do know of the technique to tame woofer Z with a LCR network in
the X-over. You would have a more controlled Z ... but added
C (plus more Xover parts).
Up at the midbass/vocal region , Z is pretty flat on both my loudspeaker systems.
Except for those DSP reflections , spook/sym/ND - even kypton
C , sound the same (and exceptional).
The only weak link is the class AB OPS , the Xover region directly correlates
to those remaining few PPM of odd harmonics at higher power levels.
The IPS's that have a real low group delay and are (slewmasters
) caneven cancel this out. NO need for a error correcting OPS when feedback
in the nanosecond realm.
If you look at these AB artifacts while at LF , the suckout caps have plenty
of time to do their job.
Above a few Khz , Xover "glitch" gets sharper and adds more odd harmonics that a slower VFA can not correct.
The wolverine is the worst performer while under biased (as expected).
Also needs the highest OPS bias to achieve it's lowest THD.
ND is truly a "blameless killer". I don't say this lightly (I like VFA's).
OS
Thanks for positive comments.
Most people just consider a fixed input sine and calculate THD at output.
In the example (waveform below) , a 100hz waveform is crossing the AB
region with (imagine) ... even a more complex 5Khz carrier. On a real audio
signal , the 5K carrier might have many other frequencies (symbols/brushes ,
room reflections).
As this complex waveform crosses 0V (non-linear region) ,both the base
frequency and the carrier(s) must interact with this AB region.
So .... feedback must correct the base , carrier(s) , and any non-linearity
at 0V.
A (slow) VFA has a lot of phase variations from LF to HF in it's NFB ...
much harder to do the job above perfectly.
CFA's have almost
no phase variation across the audio band (and more speed).
They can correct even the most complex waveform as it crosses this
imperfect AB region.
In fact , a CFA can correct even a very underbiased (non-ideal) region.
Knowing all this also means the CFA has to be more sensitive to impedance
variations from the loudspeaker.
This should not be too bad as CFA NFB is usually dominated by the extremely
low output stage Z .... the loudspeaker would just be a small proportion
of the actual takeoff point impedance.
BTW , the "blameless" attempts this "EC" with the TMC feedback scheme.
It is only effective over a limited bandwidth and with a optimal OPS bias.
I think it is better to have a single FB loop that natively corrects (everything).
OS
Attachments
A VFA does this better, moreover if some TMC is used.
Of course the residuals will be mainly odd harmonics, but if there s enough NFB around the OS these will be marginal, as in the exemple below with a dual pair at 8mA each.
Attachments
Indeed since that s the only stage wich switch off at some point.
Provided the current is a few mA at idle those odd harmonics can be kept at -80dBV if there s still enough NFB, the odd content will be dispersed in a wider bandwith, wich will reduce its amplitude at audio frequencies.
If we decrease the impedance of the NFB in a VFA, we can achieve bandwidth performances very close to the equivalent CFA. (But not slew rates).
But there is two issues. The first one is we loose the symmetry of the long tailed pair, means its PSRR advantages. The second is, because the added pole of the - input transistor, even fast, faster phase change at the HF limit of the bandwidth. Means a reduction of the stability margin.
I believe you've already had a look at this ? : http://www.esperado.fr/temp/VSSA/vssa-vs-vfa.html
One of the arguments I read often about VFAs is they can allow more feedback ratio (better distortions numbers at 1kHz). Well, we can only do this by reducing the open loop bandwidth, while it is better to have a feedback ratio as constant as possible all along the audio bandwidth (phases again).
On my side, I will prefer VFAs when i will find tranies 5 or 10 time faster than the ones we can get for the moment ;-)
As I don't hesitate to use them for multi-ways systems up to 2KHz ;-)
The VFA example in your link looks like my failed "infidel" input stage.
That extra pole it has made for instability that the simulator did not
foretell.
While what you say about the higher OLG VFA is generally true if you
compare the VSSA to a blameless , the ND don't show this with its 66db
OLG (below THD) .
A CFA needs a few "helpers" to match VFA PSRR and ultra low 1-10Khz
THD. The mirrors and zener's , cap multipliers for the PSRR and that
"fancy" super pair cascoded VAS are necessary to beat a good VFA.
It's good that these CFA's had these "weaknesses" .
This triggered the subsequent VAS and power supply enhancements.
Now we beat all the previous VFA and CFA's (In just about every metric).
OS
Attachments
I use a trick to see what really happens in the loop (OLG) in simulation. I look at the bandwidth right at the input of the VAS ;-)
No need for a prob, and it is very revealing: you see the increase of level of the response curve there, when phases begin to turn.
Yes. Looking from high altitude, the PSRR CFA weakness at low frequencies (audio) is easy to cure, rails side. Because the difference is only at the low current input stage (auto cancel of the LTP). And, yes, same causes and effects for THD.
BTW, i just realize we had similar approach, in order to compare both topologies, I was using the same VAS and the same OS, just the IPS change, (with the same OLG and CLG gain in my case, as it was for educational purpose).
Last edited:
You can make DC feedback separate from AC feedback, like AKSA style. It can achieve higher ULGF, it mean higher bandwidth.
Last edited:
It makes a useful point, although the noise comparison should be removed - any noise in LTSpice that was not added by the user is computational noise, due to rounding and bit depth limitations. LTSpice's .noise analysis will be based on reality, assuming your BJT models have Rb set correctly.
bimo, as i said, this article is for educational only. just to compare, no attempt to get best results. I use servos in real life.
Here one i have build, to avoid caps in feedback:
First the study: www.esperado.fr - VSSA with Diamond input and DC servo
Then optimisation:
http://www.esperado.fr/vssa-diamond/dvssa.php
Nothing to compare with OS work, i wanted to keep them as simple as possible.
The only real interesting thing, on my point of view is low pass the filter after the diamond that both isolate the Fc from impedance changes of the sources, and kill high HF instability (What I call "nervous character"), coupling the diamond.
This said, I will probably build the best CFA of OS. Just to make my mind about the benefit of complexity VS Simplicity, sonic side, and to see if I can point some signatures differences.
About bandwidth, VS OLG, HD and FB, we face a wall. Apart Cascode, no way to add stages without adding poles. I look with a lot of interest to OS choices.
Last edited: