Esperado in reply to post 3199, no sorry I haven't, but I am preparing a board where I will fit the different options, When working with thing like this i have the get handson to gain a physical understanding of the various compensation schemes. I have a sensation that the subjective result gets better the closer the loop is around the "issues" like local problem are fixed with local means.
I know This is very subjective, and probably more inspired from my work with speakers and speaker drivers. Better not to have to fix a problem in the crossover, when you in fact can fix it in the driver itself (or something like that)
Back to the amplifier of things i see the various feedback schemes as exactly that, fixing problems inside the gain-cell of the amplifier, anything that happens to signal in the return path is really bad and can't be fixed. I see the output stage as being the worst section of our amplifier, the part that requires the biggest hammer to get fixed, that is also why I see great promise in Cherry types of nested feedback as you close the loop closer to the issues. I know that by doing that you shift the issues up, and you potentially face stability problems. However as we have seen with CFA(ish) they seem to be easier to get unconditionally stable, so maybe nested feedback has it's place right here.
I know This is very subjective, and probably more inspired from my work with speakers and speaker drivers. Better not to have to fix a problem in the crossover, when you in fact can fix it in the driver itself (or something like that)
Back to the amplifier of things i see the various feedback schemes as exactly that, fixing problems inside the gain-cell of the amplifier, anything that happens to signal in the return path is really bad and can't be fixed. I see the output stage as being the worst section of our amplifier, the part that requires the biggest hammer to get fixed, that is also why I see great promise in Cherry types of nested feedback as you close the loop closer to the issues. I know that by doing that you shift the issues up, and you potentially face stability problems. However as we have seen with CFA(ish) they seem to be easier to get unconditionally stable, so maybe nested feedback has it's place right here.
And very easy to also make a two gain stage CFA - see Hiraga for example.
That's exactly why I called it a pointer and not one of the two tests I postulated.
There are lots of shades of gray here . . .
You will never have much higher currents (little parasitic cap ), in the feedback resistances, than the one that the rail voltages can inject in this impedance. And you need to fulfill this continuous power requirement for your amp safety.
Nobody like to heard his amp had burned on the benchmark table of some HIFI magazine.
Last edited:
This is never the case for volume or wired resistors. They need to be specified to withstand the average power, not the peak power. A 1:2 ratio for sine.
Film resistors need to be specified at at least twice the average power, say x3 for a good safety margin. You may not want to take the chances to leave your amp open loop because of a faulty resistor.
Last edited:
Hi David,
Indeed, way too little discussion about this topic.
Except three gurus, nobody seems interested in a decent CFA IPS. 🙁
BUT...... the Barney Oliver condition implies: RE = 1/Gm, while I stated: RE = 0.5/Gm. So that's another condition (the Edmond Stuart condition 🙄 ).
>what's your rationale?
Just the lowest distortion (only feasible with perfectly matched PNP/NPN trannies
)
(btw, I figured this out in the pre-simulator aera)
Best wishes,
Edmond.
Indeed, way too little discussion about this topic.
Except three gurus, nobody seems interested in a decent CFA IPS. 🙁
You are perfectly right, the Oliver condition does NOT apply to a CFA IPS. 😀
BUT...... the Barney Oliver condition implies: RE = 1/Gm, while I stated: RE = 0.5/Gm. So that's another condition (the Edmond Stuart condition 🙄 ).
>what's your rationale?
Just the lowest distortion (only feasible with perfectly matched PNP/NPN trannies
)(btw, I figured this out in the pre-simulator aera)
Best wishes,
Edmond.
If you got to peak into some CFA IC's you would find a lot of attention to making up for NPN to PNP differences, Is can be 3:1 or more different.
Anything else interesting, about design, apart a decent VAS and a not too pornographic OPS ? 🙂
Anything witch can avoid blind experiments save our lifes and is a great gift.
Oh, Scott, can-you explain us the problem of industrially producing matched PNP/NPN devices (preferably in one package). There is so much need for such devices.
PNP/NPN in one package is not the problem, it is the down on the silicon construction techniques that at the heart of the problem, in the old days vertical pnp's where all to be had, and that was a poor results. This is for the IC industry, not the discretes...
Hi Richard,
I'm sorry I wasn't clear enough on this point. Above condition for minimal distortion only applies to complementary CFA input stages (thus no 'singletons'), no matter whether it is a diamond IPS or whatsoever.
On the other hand, regarding PSRR optimization, you do need the diamond configuration. This is because the collector currents of the additional BJTs (Q1 & Q2) are used for cancellation of the noise (or rumble), induced by the constant current sources. Obviously, this is not possible without 'diamonds'.
Cheers, E.
I'm sorry I wasn't clear enough on this point. Above condition for minimal distortion only applies to complementary CFA input stages (thus no 'singletons'), no matter whether it is a diamond IPS or whatsoever.
On the other hand, regarding PSRR optimization, you do need the diamond configuration. This is because the collector currents of the additional BJTs (Q1 & Q2) are used for cancellation of the noise (or rumble), induced by the constant current sources. Obviously, this is not possible without 'diamonds'.
Don't trust me, only trust my sims. 😉
Cheers, E.
I have not the slightest idea of the industrial process of ICs or discretes. Two complementary devices on the same wafer witch can be produced close enough, or two different assembled in the same package after sorting ?
But it seems that the best place to sort transistors is the place where they are the most numerous. And i believe the matching could be automatized, so more economical.
Matching components in the manufacturer of semi-industrial audio equipment (it is not a big mass market) is so expensive and complicated that we tried anything else, design side, to avoid-it.
And, for a DIYer, he don't have enough samples to succeed-it.
An other way, why not more classes than A, B, C ?
Last edited:
That is what you can do. You can not AFAIK make a PNP and a NPN on the same substrate. Transistors are grown or deposited on the substrate and the techniques are different for each one. I left that industry 40 years ago, and things have changed. I still have some super matched pair LM394 from that era, but they never made a PNP.
I don't think one can make perfectly matched PNPs and NPNs, simply because the physics are fundamentally different. For example, electron mobility vs hole mobility are far from the same.
Cheers, E.
Cheers, E.
Member
Joined 2009
Paid Member
I used an integrated PNP-NPN input for my earlier CFA (post #37)
http://www.diyaudio.com/forums/soli...t-simple-symmetric-amplifier.html#post2730499
http://www.diyaudio.com/forums/soli...t-simple-symmetric-amplifier.html#post2730499
Tommy, I am not so sure about the NPN-PNP matching of those HFe vise, they are low gain types. They are probably the same as THAT 340, i have tried those and could not find any good match. I opted to find a 2x2 set of BC640/639 they are quite easy to find and to match (I thermally couple input stage transistors and the current-source transistors in m housekeeping to reduce thermal drift)
Bigun how did you find the NPN-PNP Hfe match in the MMDT parts..??
Bigun how did you find the NPN-PNP Hfe match in the MMDT parts..??
Last edited:
They state that those are for LNAs LOW NOISE AMPLIFIERs as in Satellite communications.