push pull 6L6 amp, oscillates even with just one output tube and nothing else.

[ELS122]

I am building a push pull 6P3S (6L6 equivalent) amp, and I cant find a way to fix what I asume is it oscllating.
problem is this:
the amp sounds fine with at lower volumes, but once it gets loud enough, the output tubes rapidly increase (negatively) in bias, they go into cutoff.
when this happens, the audio becomes super muddy, and looses trebble, and the B+ supply increases, it doesnt necrease, it increases, kinda expected since the bias becomes more negative but a weird problem anyway...
the bias supply is stable, at around -20V.
I've probed the supply rails with a signal tracer and they have practically no interfeering signals on there,
I've tried removing all the preamp tubes, and the PI tube, which didnt change it,
I removed one of the 6P3S output tubes and that reduced the magnitude but it's still there
when I remove both output tubes the oscillation stops.

what Ive tried:
twisting the plate wires, keeping them as far away from the grids as I could
twisting the grid wires, keeping them as far away from anything else.
adding 1.6k grid stoppers for the control grid
adding 680 ohm grid stoppers for the screen grid
putting snubber capacitors on the heater lines
shorting the control grid directly to the bias supply
putting a shield between the transformers and output tube
shorting the NFBL to ground

nothing fixed it, well the 680ohm screen resistors reduced it slightly, but I cant just go and put like 22k screen resistors on the tubes, that's just not the right way to fix it.

my PC sound card input only goes up to 96khz bitrate, so the highest frequency I can scope is only ~45khz,
but there seems to be a ton of spurs at around 20khz, that are only there when the amp is turned on, only when the tubes have warmed up.

I've attached pictures of the FFT chart and scope. (note the file names)
the probe is connected to the output of the amp.

I also attached a schematic of the output section of the amp


oh and one thing to note, the transformers are oriented in the same direction, one is not turned 90 degrees, I cant easilly fix that, I'm just working with the chassis that I have at the moment... but maybe that could be the cause of this? even tho the B+ rail doesnt have any of those spurs on it...
and the OPT is shielded.

[TimFox]

Have you tried plate stopper resistors, somewhere between 10 and 100 ohms, directly at the tube sockets?
Have you looked for the oscillation with an oscilloscope to estimate its frequency?
Have you tried disconnecting the 3.3 k feedback resistor as a test?
The feedback network is suspicious:  often, there is a pull-down resistor on the end opposite the output to form a voltage divider, and a small capacitor across your 3.3 k resistor to compensate the phase at high frequencies.

[ELS122]

Have you tried plate stopper resistors, somewhere between 10 and 100 ohms, directly at the tube sockets?
Have you looked for the oscillation with an oscilloscope to estimate its frequency?
Have you tried disconnecting the 3.3 k feedback resistor as a test?
The feedback network is suspicious:  often, there is a pull-down resistor on the end opposite the output to form a voltage divider, and a small capacitor across your 3.3 k resistor to compensate the phase at high frequencies.

no, but many other amps operate fine without any, it's only a 2 tube amp, and with only 1 tube it has this problem... I dont see how it would be mandatory for plate stopper resistors for this.
no, I dont have a scope. but I can say that my cat doesnt mind it so it's either above 70khz, or it is what I'm seeing on the sound card scope...
I shorted the junction of the 100 and 3.3k resistors to ground, that would shunt any and all feedback to ground.
3.3k and 100 ohm resistors form the divider, and again many amps are fine without this... I dont like adding parts to fix what is probably just a cause of sloppy layout on my part.

those spurs at around 20khz disapear when the amp does the loose bias and go into cutoff thing., that's ofc with all tubes in place then.
and they pop back in as soon as the input signal drops.

also, before I had the amp configured for cathode biased output, and then it worked fine, but maybe the cathode resistor just was hiding the problem.

what is weird is that those spurs dont seem like oscillation, it just seems like random noise... but they show convincing patterns in relation to the problem so idk

maybe you have some recommended articles on tube amp layout, because I just dont know what to suspect.
I'll now try removing the OPT entirely from the chassis, and putting it far away from the amp, with the leads as short as possible, see what it does.

[TimFox]

The 6L6 and its big brother the 807 were often used in ham radio transmitters years ago to above 20 MHz, and they were prone to "parasitic oscillation" at even higher frequencies.  The plate stopper was the normal fix for this.  Even though the oscillation frequency is much higher than the audio range, there is nothing to stop its amplitude and it totally screws up the DC values and the AF amplification, as you have noticed.  When it occurs, the parasitic oscillation happens with the single tube, regardless of the circuit around it.
If the frequency of oscillation is too high for the sample rate of your sound-card system, you will see it as an alias (q.v.).
I missed the 100 ohm pull-down resistor in your drawing, but if the entire circuit is oscillating (not a parasitic in one tube), a small capacitor across the 3.3 k might still be needed.  Try 470 pF as a test and see what happens (that would cause the closed-loop hf response to roll off above about 100 kHz).  If your feedback-removal test did not stop the oscillation, then I suspect a hf parasitic oscillation.

[ELS122]

The 6L6 and its big brother the 807 were often used in ham radio transmitters years ago, and they were prone to "parasitic oscillation" at quite high frequencies.  The plate stopper was the normal fix for this.  Even though the oscillation frequency is much higher than the audio range, there is nothing to stop its amplitude and it totally screws up the DC values and the AF amplification, as you have noticed.  When it occurs, the parasitic oscillation happens with the single tube, regardless of the circuit around it.
If the frequency of oscillation is too high for the sample rate of your sound-card system, you will see it as an alias (q.v.).

but the 5F6A fender bassman ran 5881's with not but just 470 ohm screen resistors, in fixed bias. and those amps were fine even when the output stages were pushed really hard.
but maybe my OPT is to blame, unfortunetly I didnt take any notes of the amplifier before I tore it down and put in my own circuit into it, but it seems to have been a PA amp before.
and the OPT has weird windings it seems, the windings seem to be interleaved, I read a bit into tube amp transformers and apparently this is done to increase frequency response, so maybe it was a hifi amp and the output transformer has a wide frequency response, and thus oscillates much easier...

oh and I also tried putting EL34's into the amp and it still had those spurs when the tubes heated up, but the EL34 is quite similar to 6L6's so maybe that's no surprise there...

[TimFox]

There are all sorts of circuit-wiring details that may not be important at 20 kHz but cause problems at 20 MHz.
Look up "parasitic oscillation".  Here is a forum thread for guitar amplifiers:  https://www.antiqueradios.com/forums/viewtopic.php?t=90967

[TimFox]

A modern (somewhat idiosyncratic) book about vacuum-tube amplifiers is Morgan Jones, "Valve Amplifiers, Fourth Edition", Newnes (2012).
He mentions parasitic oscillations and anode stoppers on pp 468-9.  The ham trick that he cites is to wind a 10-turn coil around a 100 ohm, 2W carbon resistor.
Fundamentally, this type of tube is capable of parasitic oscillation when connected with long wires to an arbitrary transformer with parasitics.
If you can't get access to an oscilloscope (he suggests you need 100 MHz bandwidth for these problems), you are in the dark and can only proceed by trying individual suggestions such as adding resistors.
A few years back (before retirement), we had problems using a very high frequency GaAs device (capable of frequencies up to 3 GHz, if I remember correctly) at around 20 MHz in a low-noise amplifier.  Unfortunately, we did not have any oscilloscopes or spectrum analyzers that worked up to GHz frequencies.  As in your amplifier, the prototypes sometimes oscillated out-of-band.  The only way that I could detect the oscillation when it occurred was to connect a microwave detector (diode) package rated up to several GHz RF input (with an N connector at the input and BNC at the "video" output) to the N connector on the prototype box and look for DC at the video output.

[ELS122]

I'll check out that book!

I was looking at layouts of various guitar amps, and many have the plate wires twisted, but the grid wires arent, would twisting the grid wires increase the likelyness of oscillations or no?

on the link you posted, someone suggested wraping 3 turns of wire connected to the plate, around the grid wire, but if the grid wires are twisted, there would be more capacitance between them, so could that then form an even worse oscillator, having positive feedback to the other tube, or no?

and would solid core wire in this case have a benefit, with the thought that stranded wire may act more like litz wire, while solid core wire would increase in impeadance at higher frequencies and so act as a plate/grid stopper by itself?
I guess you could achieve the same effect just by a small value inductor in series

[ELS122]

well I tried removing the choke, because it wasnt needed since I am using a buffered low pass filter in the power supply, and routed the plate wires on the other side (outside the chassis), then put the screen resistors right across the tube socket, now its even worse.
but oscillations in my experience are worse, the closer you are to fixing them, so I'll just ignore that.

but but but, adding a 5.6nF cap from grid to ground totally kills those spurs.
I'll try the other suggested methods like wraping the plate wire around the grid, and now that I have a clear path to go, I'll also try twisted grid wires vs not, and solid core vs stranded wires...

going a bit off topic, but another amp I have, a Tesla Mono 130, has suuuuuper thin wires going to everything, the plate wires are 0.5mm solid core hairs, so maybe that's the trick, because that amp is >100 watts, running 4 EL34's in push pull, fixed bias, with 600V on the plates, and the screen resistors are only 2, 680ohm that are shared between push pull pairs, having no snupping capacitors.


well I'll test it and post my results


...and they are:
nothing, didnt matter, at least not a lot.
however untwisted pair seemed to be worse every time than twisted.

[TimFox]

The wire type should have no effect on your problem, and litz wire requires that the strands be individually insulated and connected together only at the end.
The disappearance of the problem with 5.6 nF from grid to ground confirms to me that you have an RF parasitic oscillation in the tube itself, rather than with the entire circuit caused by the total feedback loop.
I have never seen before a suggestion to wrap the plate wire around the grid:  that would increase the plate-grid capacitance which is not normally a good thing.
I seriously suggest trying, say, 22 ohms directly at the plate pin of the tube socket before the wire to the primary.  The purpose is to decouple any hf resonances in the rest of the plate wire and transformer impedance from the tube itself.
What choke did you remove?  I didn't see one on your diagram.
Also, I do not see any screen bypass capacitors on your diagram.  I would suggest at least 10 nF directly at the tube socket, or from the other side of the stopping resistors, to ground or to the cathode.  In a tetrode amplifier (as opposed to an ultra-linear with the screen to a transformer tap), the screen should be at the cathode for AC to achieve the low plate-grid capacitance that helps avoid oscillation.

[ELS122]

The wire type should have no effect on your problem, and litz wire requires that the strands be individually insulated and connected together only at the end.
The disappearance of the problem with 5.6 nF from grid to ground confirms to me that you have an RF parasitic oscillation in the tube itself, rather than with the entire circuit caused by the total feedback loop.
I have never seen before a suggestion to wrap the plate wire around the grid:  that would increase the plate-grid capacitance which is not normally a good thing.
I seriously suggest trying, say, 22 ohms directly at the plate pin of the tube socket before the wire to the primary.  The purpose is to decouple any hf resonances in the rest of the plate wire and transformer impedance from the tube itself.
What choke did you remove?  I didn't see one on your diagram.
Also, I do not see any screen bypass capacitors on your diagram.  I would suggest at least 10 nF directly at the tube socket, or from the other side of the stopping resistors, to ground or to the cathode.  In a tetrode amplifier (as opposed to an ultra-linear with the screen to a transformer tap), the screen should be at the cathode for AC to achieve the low plate-grid capacitance that helps avoid oscillation.

well the plate is inverted from the control grid so that should add some negative feedback.
ok ok, I'll try that, but only after I've rerouted the wired 20 more times 
the choke was in the power supply, it's a center tapped power supply full wave rectified, into a 40uF capacitor, that feeds the OPT center tap, then goes to a 1.5k resistor (and before also a 8.4H choke), that goes into a simple buffered low pass filter, and then into a 10uF capacitor feeding the screens, then a 22k to a 10uF cap feeding the phase inverter and preamp.
the first node supplies around 250V now, but that's because I am running on reduced line voltage, since at higher voltage it goes into hard oscillation that skyrockets the current draw of the output tubes.
at full line voltage, B+ is around 400volts

hmm, I thought it was the other way around, that if I have a bypass cap for the screen it would increase gain.

[ELS122]

... I havent facepalmed harder in my life...

it WAS the power supply, my stupid f'in buffered low pass filter, 'get out of buzz free card' BS must've been the issue...

I thought well... since I'm comparing everything to fender designs, why not copy the power supply too while I'm at it, and guess what, now all those spurs are GONE,

DAYS wasted on me chasing my own tail.

well, at least I can sleep at night now that I found the issue.
and thanks for the suggestions, if I have any problems in the future I'll try them, but it seems pretty clear to me that the power supply was the big problem, so I doubt I'll have big problems, anyway... yeah.

oh and btw if you want to tell me to just fix the buffered LPF circuit so it doesnt do that, NO, I wont have that thing anywhere near my amp anymore 

[ELS122]

here I attached an image of the amp being overdriven hard with a 1khz signal...
into an 8 ohm load.

see those slight oscillations there.. well the upper one isnt slight but that might be some weird bias excursion artifact maybe, I'm still working out the bias circuit right now...

but the amp acts weird, like at first it acts like a class A amp, the louder it is, the less current draw, but then when it's already overdriven to heck, if I increase the volume even more it finally acts like a class AB amp being overdriven, increasing in current draw.

but is this right? it makes me think maybe my tubes are biased wayyyyyy too hot, and yeah one of the tubes the GM is trough the roof, and thus it runs real hot, but I'm not convinced, I mean the tubes in the 5F6A bassmans ran so hot to bake the chassis until it was brown... so idk, is this right or no?

also, the upper half of the sinewave when it distorts, it has a slope, the leading edge is high up and then it goes down, looking at some waveforms taken from actuall 5F6A bassman circuits, that doesnt seem to be the case, and rather it has a rounded shape on both waveforms (which the lower one for my amp is flat)
so... what's up 

[ELS122]

here's the waveform

[TimFox]

Your attachment is missing.
A couple of comments: 
I presume that the downward slope when you clip the peak is due to the low-frequency response of the output transformer.  It's what you would see with a low-frequency square-wave drive to the transformer due to its finite inductance.
Oscillation or ringing on the waveform is usually something worth investigating, even if it doesn't affect the sound of the guitar.
The g_m of one tube doesn't make it run hot, but does change the bias voltage on that tube required to get the correct quiescent current.  If the two quiescent currents are far from equal, you have a net DC in the transformer primary that is not a good thing.
You might want to add 10 ohm resistors in each cathode lead, which will have negligible effect on the amplifier, to verify the bias currents with an external voltmeter.

[ELS122]

Your attachment is missing.
A couple of comments: 
I presume that the downward slope when you clip the peak is due to the low-frequency response of the output transformer.  It's what you would see with a low-frequency square-wave drive to the transformer due to its finite inductance.
Oscillation or ringing on the waveform is usually something worth investigating, even if it doesn't affect the sound of the guitar.
The g_m of one tube doesn't make it run hot, but does change the bias voltage on that tube required to get the correct quiescent current.  If the two quiescent currents are far from equal, you have a net DC in the transformer primary that is not a good thing.
You might want to add 10 ohm resistors in each cathode lead, which will have negligible effect on the amplifier, to verify the bias currents with an external voltmeter.

I added the attachment in a second post.
but would this be the case at 1khz?, and actually I probed the input of the phase inverted and it has that sharp slope on that, even more I found out that with my biasing of the stage connected to the cathode follower, I have super limited headroom, so that it probably why I have kind of flubby distortion and dont have that 'bite' of the output tubes distorting first. or whatever the actual scientific reason is 

yeah I meant hot relative to the other tube, I should really put adjustable bias for both tubes in the amp, I was just lazy and used a fixed resistor dropper and here I am swapping tubes to see which one doesnt red plate at full line voltage 

well 10 ohms I'd say would increase the amp's clean headroom quite I bit, I usually go for 1 ohm, but yeah I'll add them.

also, I had seen in a marshall amp schematic where they had 470 ohm grid stoppers for EL34 screens, and then those were joined and then had a 1k going to the actual supply rail... why was this done, what effects would this cause...
I also have seen many amps where they paralel the screens of the push pull pair, and have only one screen stopped for both... I'm asuming that's just to save cost, well it doesnt make sense in the marshall amp then, but what effects could this have?, maybe increased balance between the output tubes when overdriven?