Following your advice i came up with the below schematic. The FB network i will adjust when i finish the actual amp.
http://s16.postimg.org/k9gsvq23p/amplcfb.jpg
So whaddaya guys think? Is it junk or could work? I simulated it using an ordinary tube amp power transformer model, and the triode low impedance helps A LOT, at least in theory. So it should be stellar with my real transformers.
Could you please post the picture directly? I already tried three different browsers, none of them display the full size, here is a screen shot from IE:
An externally hosted image should be here but it was not working when we last tested it.
As you wish
An externally hosted image should be here but it was not working when we last tested it.
I'd put a passive Rf filter at the input (4.7K in series and a 1nF to ground will put you down 3dB at 34kHZ). I'd get rid of all miller caps and make sure the "grid stopper" R's are right at the grid pin of the tube sockets. The EL34's should be running about 35 - 45mA at quiesence. Monitor that as soon as you apply power as soon as the tubes are warmed up. The problem with added miller caps is that they depend on the operation of the tube. If Rf energy is coming in with the input signal, the tube will detect it (like an AM radio tuner) and cause audio band noise and slewing, because the non-linearity will be asymetric. Unless you know what you're doing with "phase margin", I'd be hesitant to involve global feedback.
Last edited:
That's a great idea actually. Thanks! Of course i'll have to place that *before* the volume potentiometer (not shown here) to have a constant -3dB frequency.
Yes i'm going for around 40mA, if they will stay put there, but they should since i partially use automatic bias.
I wouldn't worry that much since i've done this before, and it appears to me it's less of a problem with triodes, at least the computer thinks so. But also that's why i have dc coupled PI there. But i was able to get another amp stable at 26dB FB which is quite insane. (at that level it sounded horrible but it was stable) I might oversimplify things, however it's first of all a question of how much FB i want to introduce. Once i work out the resistor value i can start checking capacitors of increasing value until the amp's gain becomes too low at the "phase shifted" frequencies to oscillate. Of course this method is tedious but it always works. What bugs me is that often times the channels will not both be stable with the same bandwidth but if that happens i prefer to limit them both so they will sound the same.
Putting feedback around four stages plus an OPT is asking for trouble. Good luck! Simulation will not model all the parasitic components which affect HF phase shift and hence HF stability. The simulator doesn't know where you are going to put your wires.
I admit defeat. I resorted to GNFB because i've used it before and because i've read many complaints about the Local Hero topology.
But what do i do then... I CAN of course limit the bandwidth at the input like Bob Richards suggested, but what if the amp picks up RF at some point in it's circuit not at the input? What then? I must also mention there is a powerful AM station at the city edge... so i must be very cautious.
But what do i do then... I CAN of course limit the bandwidth at the input like Bob Richards suggested, but what if the amp picks up RF at some point in it's circuit not at the input? What then? I must also mention there is a powerful AM station at the city edge... so i must be very cautious.
If the circuit is in a grounded metal chassis, I wouldn't worry about it picking up Rf from within the circuit. Make sure you are using a star ground technique. The way I test for phase margin is by driving the circuit with a 10kHZ squarewave, and looking on an oscilloscope for ringing. It should be very minimal. I also look closely at how the waveshape changes as the amp is pushed slowly into clipping and then slowly brought back out. That's a common place for spurious oscillations.
The way phase margin works is that the difference between open loop gain and closed loop gain needs to roll off to zero, before phase shift causes the negative feedback to become positive feedback. Each roll-off mechanism contributes phaseshift that extends way beyond its frequency response rolloff frequency. This requires there to be a "dominant pole" rolloff mechanism that rolls off this "loopgain" before any of the other poles (caused by any other devices in the circuit) cause significant phase shift.
Since transformers usually present the lowest frequency pole, it's common to design around that, calling that the dominant pole. All other poles or rolloff mechanisms within the feedback loop need to have much more bandwidth (10X +) so their phaseshift contribution won't cause the feedback to become positive before the "loop gain" gets below 1, as you go up in frequency. I hope this makes sense. It's hard to explain without a drawing in front of us. I find that about 12dB of negative feedback gives my speaker good enough damping, and maintains a good phase margin in the tube hi-fi amp I built.
The way phase margin works is that the difference between open loop gain and closed loop gain needs to roll off to zero, before phase shift causes the negative feedback to become positive feedback. Each roll-off mechanism contributes phaseshift that extends way beyond its frequency response rolloff frequency. This requires there to be a "dominant pole" rolloff mechanism that rolls off this "loopgain" before any of the other poles (caused by any other devices in the circuit) cause significant phase shift.
Since transformers usually present the lowest frequency pole, it's common to design around that, calling that the dominant pole. All other poles or rolloff mechanisms within the feedback loop need to have much more bandwidth (10X +) so their phaseshift contribution won't cause the feedback to become positive before the "loop gain" gets below 1, as you go up in frequency. I hope this makes sense. It's hard to explain without a drawing in front of us. I find that about 12dB of negative feedback gives my speaker good enough damping, and maintains a good phase margin in the tube hi-fi amp I built.
It sounds just fine, and I've used it to good effect. However, there's a devil lurking in the details. This sort of parallel NFB causes a reduction if the Zi of the finals. This can cause an excessive load on the driver, especially if these are triodes. Triodes don't like heavy plate loads and complain by making more harmonic distortion.
That's why you don't see it more often. That, and there are some finals (like the 6BQ6GA) that don't really need the extra help to sound really good. Just enough gNFB to take the "edge" off and all is well.
I solved the problem with the low Zin of the "Schade" feedback by driving it with a p-channel fet follower in the amp I linked to above, then my driver just had a few pF in parallel with 1MOhm to drive, so the driver was very happy.
Seriously, that amp sounds and measures better than the amp I just recently finished that I spent much more money and effort on. I got .3% distortion at 10W and it was a waterfall distortion profile, 2nd harmonic dominant. A very pleasant sounding amp.
All that without any GNF.
Seriously, that amp sounds and measures better than the amp I just recently finished that I spent much more money and effort on. I got .3% distortion at 10W and it was a waterfall distortion profile, 2nd harmonic dominant. A very pleasant sounding amp.
All that without any GNF.
Thanks for the... feedback (pun intended) folks 🙂 Then if i am to use Schade feedback, i could convert the driver stages into cathode followers. Then i would have to go back to a 12AX7 however because the gain would be way too low. OR as SpreadSpectrum suggested i could use some mosfets that i have as followers, which would help a lot. Then again, i could do the right thing and experiment to see how it sounds and try various arrangements.
I would incline to agree with Miles Power, that Schade may not even be needed at all and once again i'm left with just the option to experiment and listen to the amp, and observe the response with the scope. Not much of a scope, it's 50 yrs old but does the job.
Thanks Bob Richards for the very enlightening explanation. Much of that i already knew, however i lack some of the experience to make it perfect i suppose.
Rest assured that all my stages have larger bandwidth (now i don't know for sure about the power tubes) than the transformers themselves. And rest assured i understood why that is necessary. And if i'll be using GNFB i will use just a little, just enough to smooth out possible small nonlinearities that may still exist after applying CFB.
I would incline to agree with Miles Power, that Schade may not even be needed at all and once again i'm left with just the option to experiment and listen to the amp, and observe the response with the scope. Not much of a scope, it's 50 yrs old but does the job.
Thanks Bob Richards for the very enlightening explanation. Much of that i already knew, however i lack some of the experience to make it perfect i suppose.
Rest assured that all my stages have larger bandwidth (now i don't know for sure about the power tubes) than the transformers themselves. And rest assured i understood why that is necessary. And if i'll be using GNFB i will use just a little, just enough to smooth out possible small nonlinearities that may still exist after applying CFB.
the main problem is to find a right OT trafo.
You can use a lot of circuit but if the trafo is not good you always have issue.
At this link:
http://www.multitask.it/kt120/2-Litz-10w-8ohm-NoFB.jpg
You can see a answer at 10w rms on 8 ohm of an amp with 4 KT120 channel, tetrode connectionl. No feedback. The secondary winding is LITZ wire with 150 strand
The result is very good. -3dB at 120 KHz
The driver circuit is also good, one ECC81 totem+ECC99 long tail; all were selected very fine by Sofia.
The final results comes mainly by trafo
Walter
You can use a lot of circuit but if the trafo is not good you always have issue.
At this link:
http://www.multitask.it/kt120/2-Litz-10w-8ohm-NoFB.jpg
You can see a answer at 10w rms on 8 ohm of an amp with 4 KT120 channel, tetrode connectionl. No feedback. The secondary winding is LITZ wire with 150 strand
The result is very good. -3dB at 120 KHz
The driver circuit is also good, one ECC81 totem+ECC99 long tail; all were selected very fine by Sofia.
The final results comes mainly by trafo
Walter
- Status
- Not open for further replies.