The amount of GNFB that an amp will tolerate is governed primarily by its OPT.
Interaction between input stage and OPT determines the stability of an amp when GNFB is applied.
So I insist to adjust the value of C in the feedback loop.
As I said before, an RC between cascode anodes, may also help.
Edit: Put C=2uF in the output, probably both channels will oscillate.
Interaction between input stage and OPT determines the stability of an amp when GNFB is applied.
So I insist to adjust the value of C in the feedback loop.
As I said before, an RC between cascode anodes, may also help.
Edit: Put C=2uF in the output, probably both channels will oscillate.
Last edited:
As my friend popilin said, first remove the cap in // with the R in the FB loop. The amplifier may be not unity gain stable at higher frequencies. Doing it (Remove C) you are warranting the same amount of feedback to any frequency. Then, when the amp is stable, try increasing this cap in small amounts, example, 100, 120, 150 pf or place a variable trimmer. Then, tune the amp using square wave as it has been said.
I've been through almost this exact same experience recently. Removing the cap in parallel with the FB resistor cleared up the oscillation. Adding small amounts of capacitance back in let me know where it became unstable. In my case, 330 pF made my amp oscillate (at about 1.5 MHz), but 220 pF was solid as a rock.
with or without feedback capacitor the amp is not stable.
I've added RC between the two anode of the 12AX7, 33k + 100pF and no oscillation occour...
from the music I'm listening there is no kind of problem, excelent resolution as the twin amp...
the question is: how much is "legal" this solution on an hi-fi amp?
is the first time I build an hi-fi amp so I don't have any comparison with other project
I've added RC between the two anode of the 12AX7, 33k + 100pF and no oscillation occour...
from the music I'm listening there is no kind of problem, excelent resolution as the twin amp...
the question is: how much is "legal" this solution on an hi-fi amp?
is the first time I build an hi-fi amp so I don't have any comparison with other project
It is not too legal. What you are doing is attenuating, or better, reducing gain at high frequencies of the entire amplifier.
Lets look at this: the plate of bot cascodes are to be supposed at about the same ac voltage, but in antiphase between them. The resistor and the capacitor added, are lowering the gain at high frequencies, as they shunt together two signals in antiphase. The impedance at the plate at the cascodes is about the same value of the plate resistors.
Imagine that the 33K can be seen as two 16.5K in series, and the two 100pf are two in series too.
Then the net is about equal to have a 16.5K in series with 200pf to ground, as in the imaginary middle point of them there is a virtual ground. Then, at very high frequencies, the 200pf shunts the plate resistors with the 16.5K, and this is the load impedances at such frequencies.
Going down in frequency domain, there is a frequency at which the impedance of 200pf and the 16.5K is equal to the plate load, this is the high frequency corner of the amplifier, -3dB.
At still lower frequencies, the capacitor has too high reactance, the it no longer shunts the plate load, and at frequencies sufficiently lower the only impedance in the plate load is the loading resistors.
So, a simple calculation of f = 1 /(2 * pi *16.5K * 200pF) = 48.3Khz is the cutoff frequency of the amplifier. So anything in the order of 50KHz to infinite is not good designed/implemented.
I wonder if 50KHz BW is sufficient to the use of the amplifier.
Lets look at this: the plate of bot cascodes are to be supposed at about the same ac voltage, but in antiphase between them. The resistor and the capacitor added, are lowering the gain at high frequencies, as they shunt together two signals in antiphase. The impedance at the plate at the cascodes is about the same value of the plate resistors.
Imagine that the 33K can be seen as two 16.5K in series, and the two 100pf are two in series too.
Then the net is about equal to have a 16.5K in series with 200pf to ground, as in the imaginary middle point of them there is a virtual ground. Then, at very high frequencies, the 200pf shunts the plate resistors with the 16.5K, and this is the load impedances at such frequencies.
Going down in frequency domain, there is a frequency at which the impedance of 200pf and the 16.5K is equal to the plate load, this is the high frequency corner of the amplifier, -3dB.
At still lower frequencies, the capacitor has too high reactance, the it no longer shunts the plate load, and at frequencies sufficiently lower the only impedance in the plate load is the loading resistors.
So, a simple calculation of f = 1 /(2 * pi *16.5K * 200pF) = 48.3Khz is the cutoff frequency of the amplifier. So anything in the order of 50KHz to infinite is not good designed/implemented.
I wonder if 50KHz BW is sufficient to the use of the amplifier.
Last edited:
thank you for the explanation, I'm not new to electronic so I've understand what's appen with this solution 
from the measure taken in laboratory with the other amp BW is about 50kHz...
the first of this amplifier has been built for my bachelor degree so I have measured distortion and output power, gain and phase... in a second moment I decide to build a twin amplifier to have a stereo tube amp in my home and I have this trouble....
now I'm waiting the listening test of the stereo setup... maybe when I got enough money I'll change the output transformers.
from the measure taken in laboratory with the other amp BW is about 50kHz...
the first of this amplifier has been built for my bachelor degree so I have measured distortion and output power, gain and phase... in a second moment I decide to build a twin amplifier to have a stereo tube amp in my home and I have this trouble....
now I'm waiting the listening test of the stereo setup... maybe when I got enough money I'll change the output transformers.
I could also only look at the schematic thumbnail. We seem to forget, if I read correctly, that one other such amplifier is working correctly - which would lead me to consider differences in OPT at such a high frequency. Also, you may try to group down the anode and screen leads fron the EL34s; they seem to rather fly over everything; are they in different posisions in the other amplifier?
You mention access to a lab. That will make matters easier; input (low level) a say 5kHz square wave and look at the output with a 'scope. I imagine you would possibly find overshoot at several frequencies, which will complicate diagnosis.
To ease that, do a frequency run to several 100kHz without NFB and try notice any peaks in frequency response. You could also look at the output stage alone to see where the OPT goes reactive. Any such peaks would indicate a reactive response (phase angle involved) and could lead to oscillation with nfb. (On the thumbnail I notice G2-resistors of 270 ohm; the EL34 works rather best with 1K resistors there.) You could also try some 27pFs or so from anode - g2 of the EL34s, depending on the OPT. Sort that out first.
As Osvaldo indicated, an R.C between the 12AX7 anodes will provide a first h.f. pole. Using a 100pF there will make that at some 3,5kHz; you could try 33pF. Also as said, the cap across the feedback resistor (100K?) is critical in achieving a final acceptable result. But get the output stage in order first.
Then, permanently placing a say 470 ohm resistor over the output (loudspeaker) will provide some load under no load conditions. (Some circuits dislike the Zobel because of the temporary phase angle imparted as it comes into play.) Stable tube amplifiers will not oscillate even without a load, with such a resistor in place.
Good fortune!
You mention access to a lab. That will make matters easier; input (low level) a say 5kHz square wave and look at the output with a 'scope. I imagine you would possibly find overshoot at several frequencies, which will complicate diagnosis.
To ease that, do a frequency run to several 100kHz without NFB and try notice any peaks in frequency response. You could also look at the output stage alone to see where the OPT goes reactive. Any such peaks would indicate a reactive response (phase angle involved) and could lead to oscillation with nfb. (On the thumbnail I notice G2-resistors of 270 ohm; the EL34 works rather best with 1K resistors there.) You could also try some 27pFs or so from anode - g2 of the EL34s, depending on the OPT. Sort that out first.
As Osvaldo indicated, an R.C between the 12AX7 anodes will provide a first h.f. pole. Using a 100pF there will make that at some 3,5kHz; you could try 33pF. Also as said, the cap across the feedback resistor (100K?) is critical in achieving a final acceptable result. But get the output stage in order first.
Then, permanently placing a say 470 ohm resistor over the output (loudspeaker) will provide some load under no load conditions. (Some circuits dislike the Zobel because of the temporary phase angle imparted as it comes into play.) Stable tube amplifiers will not oscillate even without a load, with such a resistor in place.
Good fortune!
Last edited:
So you didn't get the OPTs at the same time?
There could be small differences between them. I have a set of transformers that were ordered at the same time but delivered separately as there was only one in stock. One has the lams stacked in alternating orientation, whereas the other has all the 'E's on one side and all the 'I's on the other. Who knows what else is different?
did you simulate it? cos' from what I can measure (now I have a signal generator and an oscilloscope, no access to lab) the gain of the amp is around 18 from 20Hz to 15kHz, my SG max frequency is 15kHz...
yess is taken with two years of difference
cos' no money!!! I'm an engineer student!!!I'm off topic so if you want, reply me by pvt, how can I add orcad pspice lybrary? I remember there was a procedure by the principal controll panel...
Ah I forget to mention that the other amp don't want the feedback capacitor, any capacitor from 10pF to 470pF make it oscillate!!!
Last edited:
My very good friend,
Getting shenanigans in the h.f. region, you will have to get (beg, borrow or ....) a signal generator that can at least give you a square wave at some 5 kHz, or as I suggested, a sine wave range into the hundreds of kHz. It is just simply risky to just get rid of oscillations only, not knowing what the origin is. Even without literal oscillation there might be a peak lurking somewhere, ready to give you grief with an unfavourable load. There is no way of conquering this satisfactorily without knowing what you are doing - in this case seeing what you are doing, with scope and generator.
You mentioned a lab and being a student; surely the powers-that-be will not begrudge you using the necessary equipment in the lab?
P.S: No I did not simulate; too many unknowns. Just gut feel that with trouble at all of 700 kHz you could probably get away with a first h.f. roll-off above audio, so as to give you full loop gain in the audio band.
Getting shenanigans in the h.f. region, you will have to get (beg, borrow or ....) a signal generator that can at least give you a square wave at some 5 kHz, or as I suggested, a sine wave range into the hundreds of kHz. It is just simply risky to just get rid of oscillations only, not knowing what the origin is. Even without literal oscillation there might be a peak lurking somewhere, ready to give you grief with an unfavourable load. There is no way of conquering this satisfactorily without knowing what you are doing - in this case seeing what you are doing, with scope and generator.
You mentioned a lab and being a student; surely the powers-that-be will not begrudge you using the necessary equipment in the lab?
P.S: No I did not simulate; too many unknowns. Just gut feel that with trouble at all of 700 kHz you could probably get away with a first h.f. roll-off above audio, so as to give you full loop gain in the audio band.
One of my buttons has been pushed. Doing Zobels properly.
The correct way to fix this is Zobels (series connected R + C) across each of the 100K anode loads on the 12AX7.
The object of a zobel network is to reduce the gain at high frequencies (amplitude response) WHILST LEAVING THE PHASE RESPONSE largely unaffected.
The resistor for the zobel therefore needs to be 1/10th the value of the anode load - in this case 10K. The capacitor is then choosen to give the required high freqency roll off.
The -3db amplitude response will be when Zcap + 10K = 100K (load is halved)
The R and C should be treated as vectors BUT since we chose the Rzobel = 1/10th the Rload its not really necessary.
Chose 100KHz as the roll off as a start point
Want Zcap = 90K at 100kHz => 18pF
For 18pF the 45 degree phase shift will be when Zc = Rzobel = 10K
calculating this gives a value of 884 kHz.
So we got a -3dB hf amplitude roll off at 100KHz with the "usual" R+C 45 degree phase shift >8 times higher at 884kHz - that is the way it should be done always.
Zobels are so often done so badly. Start with the Rzobel = 1/10th of Rload and you will get the best results.
When ever I see a say 100K load resistor with a zobel across it where the Rzobel is 27K or 33k or similar I just shake my head and conclude that the designer has no real idea what he/she is doing, or why.
Cheers,
Ian
The correct way to fix this is Zobels (series connected R + C) across each of the 100K anode loads on the 12AX7.
The object of a zobel network is to reduce the gain at high frequencies (amplitude response) WHILST LEAVING THE PHASE RESPONSE largely unaffected.
The resistor for the zobel therefore needs to be 1/10th the value of the anode load - in this case 10K. The capacitor is then choosen to give the required high freqency roll off.
The -3db amplitude response will be when Zcap + 10K = 100K (load is halved)
The R and C should be treated as vectors BUT since we chose the Rzobel = 1/10th the Rload its not really necessary.
Chose 100KHz as the roll off as a start point
Want Zcap = 90K at 100kHz => 18pF
For 18pF the 45 degree phase shift will be when Zc = Rzobel = 10K
calculating this gives a value of 884 kHz.
So we got a -3dB hf amplitude roll off at 100KHz with the "usual" R+C 45 degree phase shift >8 times higher at 884kHz - that is the way it should be done always.
Zobels are so often done so badly. Start with the Rzobel = 1/10th of Rload and you will get the best results.
When ever I see a say 100K load resistor with a zobel across it where the Rzobel is 27K or 33k or similar I just shake my head and conclude that the designer has no real idea what he/she is doing, or why.
Cheers,
Ian
Last edited:
Sorry to interrupt...but I don't like the way you grounded your output tubes. I would reccoment to move the black cathode wire to the yellow/green wire...right side of the picture. Thats closer to the star grounding and rectifier. The way you did it might work...maybe not as it draws heavy current.
Last edited:
One of my buttons has been pushed: giving things the correct name.gingertube said:
Not a Zobel: Zobels are at amp outputs, to isolate the output stage from load capacitance at RF frequencies.
It is a lead-lag compensation network, to stabilise the feedback loop at ultrasonic frequenicies. Different purpose, different frequency range, different design procedure, different name. Whether 1/10th of the anode load resistor is about right depends on the relationship between the anode load resistor and the anode impedance, which depends on triode or pentode and whether cathode degeneration has been used.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.