VHF,
Lavardin's technique for reducing 'memory distortion' in the VAS was simply to cascode, removing Early effect in the lower device; but I found years ago that unless you inserted a resistor between the lower and upper device between collector/emitter the clip performance became very ugly.... some mention is made of this by peufeu here where he addresses it as 'the magic resistor'.
Notice the ultimate version cascodes the cascode. I'm not convinced this is the best way to do it, too much cost of rail efficiency, but a simple cascode reduces memory distortion almost ten fold, obviously worth it.
This is a guy who has it all; deep maths insights, clearly a gifted engineer, but one who listens, is not didactic or opinionated, and is not afraid to describe the sound in heavily subjective terms.
I guess this is the engineer's equivalent of dancing.....
Good to see, restores the faith.....
Cheers,
Hugh
Lavardin's technique for reducing 'memory distortion' in the VAS was simply to cascode, removing Early effect in the lower device; but I found years ago that unless you inserted a resistor between the lower and upper device between collector/emitter the clip performance became very ugly.... some mention is made of this by peufeu here where he addresses it as 'the magic resistor'.
Notice the ultimate version cascodes the cascode. I'm not convinced this is the best way to do it, too much cost of rail efficiency, but a simple cascode reduces memory distortion almost ten fold, obviously worth it.
This is a guy who has it all; deep maths insights, clearly a gifted engineer, but one who listens, is not didactic or opinionated, and is not afraid to describe the sound in heavily subjective terms.
I guess this is the engineer's equivalent of dancing.....
Good to see, restores the faith.....
Cheers,
Hugh
If you increase the CM's emitter resistors, they become less sensitive to temperature variations - at least when simulated. The "memory effect" in the CM is quite large with no or "normal" resistor values.
There's almomst no memory effect in the output stages. Most of it is in the voltage amplifying stages. I find this interesting since my experience tells me that the voltage amplifying stages, especially the input stage, colors the sound very much more than the current amplifying stage. If OLG is reduced - the memory effect is reduced.
There's almomst no memory effect in the output stages. Most of it is in the voltage amplifying stages. I find this interesting since my experience tells me that the voltage amplifying stages, especially the input stage, colors the sound very much more than the current amplifying stage. If OLG is reduced - the memory effect is reduced.
Hi Hugh,
I am not sure Perrot (Lavardin's engineer) used a cascode VAS.
For the low power stages, Peufeu used a classical structure à la Self (differential input, CM, VAS) but with sophistication in the input diff and VAS. As far as it can be deduced from his writings in L'Audiophile, Perrot used three low power stages (as Ottala's circuit based Electrocompaniet) using no less than five of his patented very linear diff circuits.
I think Perrot's work is very original in his criticism of DC feedback modifying the operating point of the input stage in almost any solid state amplifier.
Don't you think the ugly clip you see in a cascode is due to the voltage of the common base device being disturbed by the saturation ? I think a small value resistor in series with the base is then beneficial.
I am not sure Perrot (Lavardin's engineer) used a cascode VAS.
For the low power stages, Peufeu used a classical structure à la Self (differential input, CM, VAS) but with sophistication in the input diff and VAS. As far as it can be deduced from his writings in L'Audiophile, Perrot used three low power stages (as Ottala's circuit based Electrocompaniet) using no less than five of his patented very linear diff circuits.
I think Perrot's work is very original in his criticism of DC feedback modifying the operating point of the input stage in almost any solid state amplifier.
Don't you think the ugly clip you see in a cascode is due to the voltage of the common base device being disturbed by the saturation ? I think a small value resistor in series with the base is then beneficial.
AKSA said:
Hugh,
Thanks for this. A slightly different perspective but it's a nice and simple way to banish most BJT nasties. The cascoded VAS is well known but this is timely for the thread as discussions move to invlude VAS.
Yes I can see the merits of the magic resistor for a more stable CE stage collector voltage. I actually use a cascoded VAS in my DIY amp so the modification is not too difficult for me to implement and to report back on. FWIW my common emitter stage is also slighly degenerated and is actually a 2 transistor CFP which turbo charges the VAS beta. This also increases the VAS LF impedance and fits in with the MD theory (albeit in a limited way) but the CFP driver is also further buffered from VAS current excursions so I have also extended the elimination of MD beyond a single cascode without that being an objective.
And yes, as you do, I also place listening session results ahead of purely technical merit but WRT cascoding of the VAS I have always found a strong correlation even if sims are marginal.
The trade of rail voltage is really not that serious and in any case you can always use a dedicated LTP / VAS psu with higher rails to compensate which is something you might do anyway in a high end design.
Cheers and thanks.
VHF
"vas-3" looks like the one I use, do I win a price? ;-) ,cascoded cascoded, hello I'm trying to build minimalistic here(gain in quality / nr. of components)
hmm skämt åtsido.. hmm anyway nice page that!! I too discovered (something that multisim did'nt) that without the resistor it sometimes just works bad, to put it in non scientific terms..
hmm skämt åtsido.. hmm anyway nice page that!! I too discovered (something that multisim did'nt) that without the resistor it sometimes just works bad, to put it in non scientific terms..
VHF, Forr,
Pierre Frederic's work is very interesting, and well expressed. I like his approach..... empirical.
The wildly oscillating temperature variations across the 0.2 gram die of a VAS semiconductor certainly seems plausible in the production of ugly artefacts.
The fact this is compounded by base parasitics to me suggests that the ideal VAS would be common base, although the usual C/B compensation is then not appropriate.
Yes, Forr, I do agree that a saturated common base could be causing the ugly clip. Typically 200mV is all that needs to be dropped across the 'magic resistor'.
VHF, I agree about CFP VASs. They work well.... The chief advantage is the high Zin, and very low Zout, which makes driving the output stage a walk in the park and eases the load on the global negative feedback loop.
I have not found output stages add as much coloration as the VAS. However, there is a low level 'mist' caused by switching noise in Class AB stages. Some means of removing the minority charge carriers from the base/emitter junction is necessary for the inactive device; Self's cap/resistor across the bases works well, but the choice of resistor and cap is critical, with low dielectric absorption caps performing best here.
Another option is to ensure the drivers are always on; Blomley did some work on this in the sixties.
There is almost nothing new. It is a matter of putting the pieces together. Modern semis are remarkable devices, now very refined.
Nikwal, everything should be as simple as possible, but definitely no simpler. Some complexity is inevitable.
Cheers,
Hugh
Pierre Frederic's work is very interesting, and well expressed. I like his approach..... empirical.
The wildly oscillating temperature variations across the 0.2 gram die of a VAS semiconductor certainly seems plausible in the production of ugly artefacts.
The fact this is compounded by base parasitics to me suggests that the ideal VAS would be common base, although the usual C/B compensation is then not appropriate.
Yes, Forr, I do agree that a saturated common base could be causing the ugly clip. Typically 200mV is all that needs to be dropped across the 'magic resistor'.
VHF, I agree about CFP VASs. They work well.... The chief advantage is the high Zin, and very low Zout, which makes driving the output stage a walk in the park and eases the load on the global negative feedback loop.
I have not found output stages add as much coloration as the VAS. However, there is a low level 'mist' caused by switching noise in Class AB stages. Some means of removing the minority charge carriers from the base/emitter junction is necessary for the inactive device; Self's cap/resistor across the bases works well, but the choice of resistor and cap is critical, with low dielectric absorption caps performing best here.
Another option is to ensure the drivers are always on; Blomley did some work on this in the sixties.
There is almost nothing new. It is a matter of putting the pieces together. Modern semis are remarkable devices, now very refined.
Nikwal, everything should be as simple as possible, but definitely no simpler. Some complexity is inevitable.
Cheers,
Hugh
Hi Nelsonvandal,
--- There's almomst no memory effect in the output stages. Most of it is in the voltage amplifying stages. I find this interesting since my experience tells me that the voltage amplifying stages, especially the input stage, colors the sound very much more than the current amplifying stage. ---
Self shows that, in the ouput stage, variations of the power supply voltage has some effects on the quiescent current due the Early effect. This cannot be compensated with the bias circuit.
---If OLG is reduced - the memory effect is reduced.---
Memory effects are related to thermal behaviour. The less you have OLG with conventionnal circuits, the more the input stage is submitted to voltage variations, hence power and thermal variations.
--- There's almomst no memory effect in the output stages. Most of it is in the voltage amplifying stages. I find this interesting since my experience tells me that the voltage amplifying stages, especially the input stage, colors the sound very much more than the current amplifying stage. ---
Self shows that, in the ouput stage, variations of the power supply voltage has some effects on the quiescent current due the Early effect. This cannot be compensated with the bias circuit.
---If OLG is reduced - the memory effect is reduced.---
Memory effects are related to thermal behaviour. The less you have OLG with conventionnal circuits, the more the input stage is submitted to voltage variations, hence power and thermal variations.
Jam do you mean in a current mirror or the resistor in the vas. The resister in the vas is interesting, I havent tried it, but in general I have disliked the sound using cascode in vas and ltp in self type amps. I use very low cob , high speed trannies, and I like the warmer sound without cascode. Maybe this resistor does something to the sound, anyone here that can comment on this ???
Cascoding Jfets works good but now I got hold of some sanyo Jfets and in a headamp I tried them in, Ive gotton rid of the cascode too.
Jam take a look at the buffer thread for my comments on the sound of mirrors with that buffer design of mine.
Cascoding Jfets works good but now I got hold of some sanyo Jfets and in a headamp I tried them in, Ive gotton rid of the cascode too.
Jam take a look at the buffer thread for my comments on the sound of mirrors with that buffer design of mine.
AKSA said:
Memory distortion is an interesting matter, but bear in mind that the right measurements will often bring it out. Even very low frequency THD can be useful in this regard. Some take THD at 20 Hz for granted and don't even measure it, as it should be "easy" for an amplifier to faithfully reproduce such a slowly changing signal (especially with all of the loop gain often available at those low frequencies). But don't take it for granted, and do be suspicious if it rises above, say, 1 kHz THD. Such "memory distortion" even exists in the feedback resistor as a result of its thermal variations when it is subjected to substantial power.
However, some of these effects are buried in amplifiers that have otherwise high levels of THD.
Another interesting test is to do SMPTE with a low frequency component much lower than the usual 60 Hz.
Another interesting test would be low-frequency CCIF, where 19Hz +20Hz signals are combined, resulting in a 1 Hz signal power envelope.
Cheers,
Bob
AKSA said:
Depending on the thermal time constants of the elements involved, I suppose thermally-induced memory distortion could have an influence even when high frequencies are involved, since the time variation of the power envelope of the high-frequency signals could play a role. Consider, for example a test tone consisting of 20 kHz and 19,990 Hz, with a power envelope changing at a 10Hz rate.
Not sure about DA by itself, but I do believe in using a high-quality polypropylene capacitor for input coupling.
Cheers,
Bob
Reviving an old thread gone quiet 
But I have a quick question on a current mirror improved with the extra "helper" transistor as in the one posted in post #8. (3 trannies)
In the context of a symmetric topo with the 2 ltps with their current source, if I need the helper transistor's collector not to be referenced to ground, how is it correctly done?
I would think that just linking up the collectors directly between the 2 opposing helpers would not be a good thing. So what would be the right way?
But I have a quick question on a current mirror improved with the extra "helper" transistor as in the one posted in post #8. (3 trannies)
In the context of a symmetric topo with the 2 ltps with their current source, if I need the helper transistor's collector not to be referenced to ground, how is it correctly done?
I would think that just linking up the collectors directly between the 2 opposing helpers would not be a good thing. So what would be the right way?