LED lights up when a certain voltage across R6 is exceeded.
Therefore I guess the LED indicates when the output tubes start to draw grid current.
It operates as an indicator of A2 threshold.
Therefore I guess the LED indicates when the output tubes start to draw grid current.
It operates as an indicator of A2 threshold.
The usual NFB scheme requires infinite loop gain to reach zero distortion. Of course, diminishing returns occur along the way there.
An interesting alternative approach comes to mind from an extension to an old xfmr linearization scheme (patented in the 1940s), now offered as the LEM (linearity enhancement module) by Custom Analogue Audio (Joe Rasmussen).
This scheme puts a low level HF (above audio) signal, of constant amplitude, into the input LTP of the amplifier, with the intention of improving the low signal permeability of the OT. Much like the HF bias used for tape recording. (and for removing xfmr core hysteresis)
If the level of the (constant) HF is monitored (via synchronous demodulation) at the final tube plates, this indicates the forward gain of the amplifier. If that level is used to control the tail current of the input LTP (altering the LTP gain), then the forward amplifier gain can be servo controlled to be constant. Only finite gain is required in this control loop to achieve constant amplifier gain (within the audio frequency range). (within the linear capabilities of the control loop devices themselves, ie the variable CCS tail and sync. demodulator)
Of course starting out with linear gain devices in the amplifier makes for easier work for any linearity control loop. Like Super Linear tubes, below:
An interesting alternative approach comes to mind from an extension to an old xfmr linearization scheme (patented in the 1940s), now offered as the LEM (linearity enhancement module) by Custom Analogue Audio (Joe Rasmussen).
This scheme puts a low level HF (above audio) signal, of constant amplitude, into the input LTP of the amplifier, with the intention of improving the low signal permeability of the OT. Much like the HF bias used for tape recording. (and for removing xfmr core hysteresis)
If the level of the (constant) HF is monitored (via synchronous demodulation) at the final tube plates, this indicates the forward gain of the amplifier. If that level is used to control the tail current of the input LTP (altering the LTP gain), then the forward amplifier gain can be servo controlled to be constant. Only finite gain is required in this control loop to achieve constant amplifier gain (within the audio frequency range). (within the linear capabilities of the control loop devices themselves, ie the variable CCS tail and sync. demodulator)
Of course starting out with linear gain devices in the amplifier makes for easier work for any linearity control loop. Like Super Linear tubes, below:
Attachments
Last edited:
I am trying to understand your schematic so I have a question. I noticed most folks around here use a cascode with depletion mode n-channel devices (DN2540) but you seem to use a P-channel enhancement mode device. Is this so you don't have to use two devices because the CCS isn't going to be modulated by the signal because the current set resistor is R4? (It's between the transistor and supply voltage instead of between the transistor and tube).
Indeed. But in practise we work with the difference in OL gain and CL gain which is the factor that lowers doistortion. I don't think anybody is seriously trying for zero distortion.
Of course, you would need infinite loop gain in this case as well. Because it is just an ordinary feedback scheme, only much more convoluted - a 'regular' NFB loop will be much simpler, easier and give better results.
If you fancy, you can move the 'infinite gain' into the feedback loop of a 'regular' NFB loop as well. I think it is called 'active feedback' as it would require an active element in the feedback arm.
Jan
Last edited:
I think it only requires unity loop gain for the modified linearization scheme. This is because it is dealing with gain directly. For an x% error in gain, only x% change in gain at the LTP is needed. Nothing infinite needed here. Essentially like a bias V control loop. But controlling gain directly in this case.
For the tiny variations in gain of the active amplifier chain devices (with nice linear ones at least), the control loop has almost nothing to do, lazy, easy work. Very tiny variations of the control signal needed, so extreme linearity should be the result. (the control devices would be very linear over the small control V excursions needed.)
For the tiny variations in gain of the active amplifier chain devices (with nice linear ones at least), the control loop has almost nothing to do, lazy, easy work. Very tiny variations of the control signal needed, so extreme linearity should be the result. (the control devices would be very linear over the small control V excursions needed.)
Last edited:
As I noted, once you move the 'NFB gain' into the feedback loop, you can use any forward OL gain you like, even lower than 1.
There have been many, many ingenious schemes for setting up a feedback system. In concept they are ALL the same, and ALL follow regular feedback and control system theory.
You can package simple things in complex boxes but when you open the box it is immediately recognizeable as a well-known concept that does not give better results within the same constraints. The scheme you describe has to fit into the same loop gain phase response and stability criteria as any other NFB scheme. The ONLY way to get better results is to somehow get more loop gain out of it all other things being equal). There are several smart solutions like two-pole compensation, output inclusive compensation, lead-lag, etc.
That is were you win or lose, not by introducing some weird scheme to hide that it is 'just' an NFB scheme.
Jan
BTW If you really think about it hard enough, it is clear that you cannot get to zero error for exactly the same reason why regular feedback cannot get to zero error. Nulling an error to zero requires infinite gain *somewhere* in the loop, however you slice it.
Jan
Not here. You will have to attenuate the forward amplifier gain out to get the required unity loop gain, for a direct gain control system.
You can turn it into a feedforward or 'error cancellation' scheme, where you do not need infinite loop gain. BUT (there's always a but, as Mother Nature is willing to trade but not give out freebies) here a few new constraints appear:
In this case you need an exact gain block with gain 1.00000000... to exactly cancel the error. This is of course not possible as perfection does not exist in physical systems. But, how close can you get?
As an example, if you want to decrease the error by 60dB, your gain block must be accurate to 60dB or 1/1000 or 0.1%.
Then also, if your error cancellation needs to be successful over the audio band, the bandwidth of the gain block must have the same X 1000 factor, iow it must have a really flat freq response out to 20MHz.
Normally these things can only be reached with opamp-like structures. And, surprise, surprise, you again find the infinite gain requirements in those auxiliary circuits.
100's, maybe 1000's of engineers and designers much, much smarter than me, have spend countless years to find this holy grail. There is a lot documented but it requires study time and effort to get down to what it all really means.
A good start would be Giovanni Stochino's article series on the relative merits and issues with feedback, feedforward and error correction in Linear Audio.
If, despite all this, you DO succeed to circumvent these iron laws, the Nobel Committee is waiting for you. And I will pay for your trip 😎
I'll leave it at this. I gave you enough points to get started on. Up to you.
Jan
In this case you need an exact gain block with gain 1.00000000... to exactly cancel the error. This is of course not possible as perfection does not exist in physical systems. But, how close can you get?
As an example, if you want to decrease the error by 60dB, your gain block must be accurate to 60dB or 1/1000 or 0.1%.
Then also, if your error cancellation needs to be successful over the audio band, the bandwidth of the gain block must have the same X 1000 factor, iow it must have a really flat freq response out to 20MHz.
Normally these things can only be reached with opamp-like structures. And, surprise, surprise, you again find the infinite gain requirements in those auxiliary circuits.
100's, maybe 1000's of engineers and designers much, much smarter than me, have spend countless years to find this holy grail. There is a lot documented but it requires study time and effort to get down to what it all really means.
A good start would be Giovanni Stochino's article series on the relative merits and issues with feedback, feedforward and error correction in Linear Audio.
If, despite all this, you DO succeed to circumvent these iron laws, the Nobel Committee is waiting for you. And I will pay for your trip 😎
I'll leave it at this. I gave you enough points to get started on. Up to you.
Jan
Last edited:
Using P-channel device I sacrificed PSRR, but such topology gave me lover capacitances in a low current stage. To load a stage with at least 5 times higher current I would use a N-channel device.
Cascode CCS? For purists who are in love with SIMs. But why not? ;-)
It is an optocoupler, with a LDR on input (attenuator, optical peak limiter)
- no matter how graceful an amp is clipping, compression instead of it sounds even less nasty. ;-)
Actually, 10-12V above the A2 threshold, when enough is enough. ;-)
Last edited:
So lower capacitance (Cds?) will give better high frequency performance but traded off for PSRR.
Many thanks.
No, reference to B+ instead of to ground results in worse PSRR. However, we deal with it in choke and transformer loaded stages, so it is not so horrible. ;-)
Cgd VS Cgs.
Black's feedforward is something different than negative feedback, but so far the two schemes seem to cover the territory
http://www.diyaudio.com/forums/solid-state/240712-cfa-topology-audio-amplifiers-229.html#post3723022
collects a few of my posts, scan of a few good pages of I. M. Horowitz "Synthesis of Feedback Systems", 1963
I spent considerable time and effort in the http://www.diyaudio.com/forums/solid-state/89023-bob-cordell-interview-error-correction.html thread trying to show the identity of "Hawksford Error Correction" and Horowitz' negative feedback equivalents
Bob endorsed the view of HEC as equivalent to negative fedback as correct
Last edited:
I need to go back through my transistor reference literature as you can tell. I deal with tubes much more than transistors.
Thanks again for explaining.
I agree with almost everything he wrote.
As a student in the early 1990's I did a literature study to figure out where the "feedback is bad" story originated from and to find out whether it was correct. I found a 1966 article by Daugherty and Greiner. It was actually a rather good article, but with one mistake in it: they found a sufficient condition to prevent slewing, but presented it as if it were a necessary and sufficient condition.
Their sufficient (but certainly not necessary) condition is to make the open-loop bandwidth of the amplifier greater than or equal to the bandwidth of the incoming signal, or equivalently, to put a low-pass filter in front of the amplifier with a bandwidth smaller than or equal to the open-loop bandwidth. Because of stability and bandwidth reasons, this automatically leads to a large open-loop bandwidth and limited negative feedback.
It seems that this started the "low feedback" movement that later changed into a "no feedback" movement.
If you want to know the details:
https://linearaudio.net/sites/linearaudio.net/files/volume1ltemvdg.pdf
Anatoly - Unfortunately, the only information I've got about your 4P10S amplifier was from the youtube video (BTW, I enjoyed the song). Now it seems like it disappeared from youtube, so I can only guess what's in your schematic.
I know 4 ways of applying local NFB in a pentode power stage: 1) through UL taps in the output transformer, 2) through cathode feedback winding in the output transformer, 3) by connecting plate to g1 a la Gary Pimm, and 4) by leaving cathode bias resistor unbypassed in a SE stage. 1 and 2 I don't care about, and 4 does not bring any sonic improvement. Gary's 47 amp sounds great indeed, and I did some putzing with his circuit about 10 years ago. The lesson learnt was to avoid MOSFET CCSes, which Gary was so fond of.
Your notions of "sound nice" and "high end" raise an interesting philosophical issue: what is the goal of our audio endeavors? "Sounds nice" is highly subjective, but in the end this is what really matters. One might say that good reproduced sound must be as close as possible to the original, but even this is not always true. For example, a good symphony record may better the actual concert hall experience. The record puts you in the midst of the orchestra, an experience quite different from that of 22nd row seat. With "high end" you've lost me. This is about vanity and nothing else.
I don't believe in amplifier competitions, so I don't accept your challenge. However, I am much interested in experiencing the real sound of your creations, and would welcome you as a guest to show off my stuff.
I described it in Photo Gallery:
http://www.diyaudio.com/forums/tubes-valves/71300-photo-gallery-390.html#post5000840
Different aspects of the design were discussed multiple times here on the forum, as well as what I call "High End". My High End Ideal is when amps and speakers disappear and not audible at all. When subconscious reaction on sounds happen before you realise that sounds are re-produced.
All videos are still on Facebook, nobody removed them:
https://www.facebook.com/Wavebourn/videos/
http://www.diyaudio.com/forums/tubes-valves/71300-photo-gallery-390.html#post5000840
Different aspects of the design were discussed multiple times here on the forum, as well as what I call "High End". My High End Ideal is when amps and speakers disappear and not audible at all. When subconscious reaction on sounds happen before you realise that sounds are re-produced.
All videos are still on Facebook, nobody removed them:
https://www.facebook.com/Wavebourn/videos/
Last edited:
No luck with photo gallery. There are photos of a 6CD6 amplifier - is 4P10S same circuit? I tried searching for 4P10S on the forum - no luck either. Facebook directs me to sign-up, but I am under wow of never using the social media. That video was on youtube, but no longer.
I get your "High End". You are a dreamer.
6CD6 is a sweep tube for civil TVs, 4П10С is a military sweep tube for radars. I.e. power pentode or ray tetrode with big cathode and dense screen grid. They are very similar tubes.
Do you expect me to give results of my work to a guy who insults me and by insulting my sanity tries to force me to do that?
You are a dreamer. ;-)
Do you expect me to give results of my work to a guy who insults me and by insulting my sanity tries to force me to do that?
You are a dreamer. ;-)
Hi,
I suggest to cleary state what kind of feedback is talked about as the parameters influenced by the feedback may be influenced oppositely, for example the impedance values of a circuit.
One can divide at least into four types:
- voltage derived voltage- or current-feedback
- current derived voltage- or current-feedback
Further one can divide by the number of encompassed stages between local feedback and global feedback.
It seems that the global methods are in the mids of the critique ... probabely since they are easier to ´recognize´ in most schematics.
Local loops are harder to spot and often fly fully below the radar ...
... and hey, putting a resistor somewhere in a circuit almost always introduces some kind of feedback anyway .... so if You can´t avoid FB, can it be bad?
I agree with BP that if You use FB, use as much as possible.
Interesting I find his claim, that the OL bandwidth be large also, as the typical very low OL-Bandwidth, resp. the falling-with-frequency feedback factor leads to a raising distortion amplitude response ... and a associated frequency dependent dynamic behaviour of the circuit.
jauu
Calvin
I suggest to cleary state what kind of feedback is talked about as the parameters influenced by the feedback may be influenced oppositely, for example the impedance values of a circuit.
One can divide at least into four types:
- voltage derived voltage- or current-feedback
- current derived voltage- or current-feedback
Further one can divide by the number of encompassed stages between local feedback and global feedback.
It seems that the global methods are in the mids of the critique ... probabely since they are easier to ´recognize´ in most schematics.
Local loops are harder to spot and often fly fully below the radar ...
... and hey, putting a resistor somewhere in a circuit almost always introduces some kind of feedback anyway .... so if You can´t avoid FB, can it be bad?
I agree with BP that if You use FB, use as much as possible.
Interesting I find his claim, that the OL bandwidth be large also, as the typical very low OL-Bandwidth, resp. the falling-with-frequency feedback factor leads to a raising distortion amplitude response ... and a associated frequency dependent dynamic behaviour of the circuit.
jauu
Calvin
- Status
- Not open for further replies.