Try reducing R15 from 150k to 68k.
With the hybrid/fet enhanced SRPP drive should be very strong with no chance of inadequate current.
Simple try!
HD
With the hybrid/fet enhanced SRPP drive should be very strong with no chance of inadequate current.
Simple try!
HD
While paralleling the GM70s, did u have individual grid stopper resistors for each GM70? If you been sharing the grid stopper, you may want to try using individual grid stopper, one for each of the GM70s.
Can also try to increase the grid stopper, which is currently 100ohm as shown in your schematic, to say 1K or even higher. Then if this stops the oscillation, you can try reducing the value until it begins to oscillate again
Operating in Fixed Bias requires attention to the resistance between the grid terminal and the filament. This is to enure that any grid leakage current does not add to the bias votage in the direction of increased anode current.
None of the GM-70 data sheets give aa maximum value for R(g→k) but that does not mean that high values are OK. Old Soviet tubes have plenty of gas, and consequently high values of grid leakage current. This leakage current can destabilise the anode current upward.
Other Large Power DHTs need LOW values of grid resistance, which reflect the potential effect of grid leakage current upon the DC anode current.
For example: the RCA-813 does not permit any value greater than 30kΩ in the grid circuit, for all classes of operation. Even the 300B allows no more than 50kΩ.
This limits the value of R15, as others have noted.
But the grid to filament circuit also includes the bias supply's network of resistors too - and these alone amount to >100kΩ.
My bias-regulators have low impedance from DC upward to maintain full DC stability, and very low distortion from signal impedance effects (without using large capacitors).
One solution:
Use P-channel FET followers to buffer the bias voltage and provide a low (DC) impedance to ground. High voltage P-channel parts are not plentiful, so some searching may be needed. FET followers present a lower voltage drop (a volt or so) after the amp is warmed up, so they are not perfect - but at least the return impedance is under control.
Now change R15 to 27kΩ. This will mean a much larger coupling capacitor, naturally.
Next, I don't believe that sharing a bias voltage between two GM-70s is a good idea. The spread of anode current values on these tubes is often too wide to allow it; and interaction between the tubes is also likely. Better to have a separate coupling cap, trim poti + follower for each GM-70, IMHO.
None of the GM-70 data sheets give aa maximum value for R(g→k) but that does not mean that high values are OK. Old Soviet tubes have plenty of gas, and consequently high values of grid leakage current. This leakage current can destabilise the anode current upward.
Other Large Power DHTs need LOW values of grid resistance, which reflect the potential effect of grid leakage current upon the DC anode current.
For example: the RCA-813 does not permit any value greater than 30kΩ in the grid circuit, for all classes of operation. Even the 300B allows no more than 50kΩ.
This limits the value of R15, as others have noted.
But the grid to filament circuit also includes the bias supply's network of resistors too - and these alone amount to >100kΩ.
My bias-regulators have low impedance from DC upward to maintain full DC stability, and very low distortion from signal impedance effects (without using large capacitors).
One solution:
Use P-channel FET followers to buffer the bias voltage and provide a low (DC) impedance to ground. High voltage P-channel parts are not plentiful, so some searching may be needed. FET followers present a lower voltage drop (a volt or so) after the amp is warmed up, so they are not perfect - but at least the return impedance is under control.
Now change R15 to 27kΩ. This will mean a much larger coupling capacitor, naturally.
Next, I don't believe that sharing a bias voltage between two GM-70s is a good idea. The spread of anode current values on these tubes is often too wide to allow it; and interaction between the tubes is also likely. Better to have a separate coupling cap, trim poti + follower for each GM-70, IMHO.
I am curious how this will turn out.
I know that combining chokes with big H and low R with small caps can cause ringing in the Power Supply, Merlin Blencowe covers this well in his Designing Power Supplies book. I simulated the PS in PSUD, some ringing indeed, but nothing crazy. Can be improved, but reason to cause wild oscillation?
Is the oscillation at RF, or is it motorboating? Motorboating is due to feedback through the PS, but the author uses different supplies for driver and output stage. Question there: are both driver stages always powered, or you only power the driver stage of the corresponding output tube? Maybe the driver PS works well with one driver, but has problems when powering both?
There is no negative feedback, so how can one make it oscillate at RF? Some pictures for an assessment of the build layout would be nice. Driver uses depletion fets, 6N6P and EL84. I think it is hard to make a 6SL7 oscillate, but there devices all have considerable transconductance so precautions are in order.
I indeed agree that R15 is probably too high, for long term reliability a lower value is recommended, and indeed fed from a low impedance source (a follower, the P fet.. I use a LND150 as follower and a second as CCS to create low source bias supplies). My GM70 is powered from a source follower, so there is indeed no problem with G1 current, I actually drive it into A2.
The big question is that both channels work nicely independently, and the PS seems to be able to deliver clean 100mA DC. So individually there is no oscillation, no problems with grid resistor. Why can't the PS deliver the 100mA nicely over two valves (45mA at each makes it oscillate). Which interaction makes them oscillate?
I know that combining chokes with big H and low R with small caps can cause ringing in the Power Supply, Merlin Blencowe covers this well in his Designing Power Supplies book. I simulated the PS in PSUD, some ringing indeed, but nothing crazy. Can be improved, but reason to cause wild oscillation?
Is the oscillation at RF, or is it motorboating? Motorboating is due to feedback through the PS, but the author uses different supplies for driver and output stage. Question there: are both driver stages always powered, or you only power the driver stage of the corresponding output tube? Maybe the driver PS works well with one driver, but has problems when powering both?
There is no negative feedback, so how can one make it oscillate at RF? Some pictures for an assessment of the build layout would be nice. Driver uses depletion fets, 6N6P and EL84. I think it is hard to make a 6SL7 oscillate, but there devices all have considerable transconductance so precautions are in order.
I indeed agree that R15 is probably too high, for long term reliability a lower value is recommended, and indeed fed from a low impedance source (a follower, the P fet.. I use a LND150 as follower and a second as CCS to create low source bias supplies). My GM70 is powered from a source follower, so there is indeed no problem with G1 current, I actually drive it into A2.
The big question is that both channels work nicely independently, and the PS seems to be able to deliver clean 100mA DC. So individually there is no oscillation, no problems with grid resistor. Why can't the PS deliver the 100mA nicely over two valves (45mA at each makes it oscillate). Which interaction makes them oscillate?
As you said no max value for grid leak is specified in the datasheet so i used what i saw in other designs. Many of them use values between 100 and 150k, i think i've even seen one somewhere with 220k. I will be lowering it tho.
I do have separate bias supplies for each GM70, well separate apart from the transformer winding, rectifier and CRCRC filter.
The Driver stage is always fully powered (so both sides) regardless of which GM70 i have inserted.
If you look at my initial post there is a link on the bottom to the tube gallery thread in which i posted some pics of the amp, as you can see there the driver stage is on a PCB. Ofc this isn't optimal for debugging but from what i can see the issue does not point towards the driver stage.
I am now leaning towards the power supply being the issue like multiple people have mentioned. So what i'm planning on trying first is removing the 2nd choke and moving the first one one cap upwards (C-L-CCC). I did some simulating in spice at work today using what euro21 has provided and this does smooth out the fluctuations considerably in comparison to how the PSU is wired now.
Will be reporting back later with results.
I do have separate bias supplies for each GM70, well separate apart from the transformer winding, rectifier and CRCRC filter.
The Driver stage is always fully powered (so both sides) regardless of which GM70 i have inserted.
If you look at my initial post there is a link on the bottom to the tube gallery thread in which i posted some pics of the amp, as you can see there the driver stage is on a PCB. Ofc this isn't optimal for debugging but from what i can see the issue does not point towards the driver stage.
I am now leaning towards the power supply being the issue like multiple people have mentioned. So what i'm planning on trying first is removing the 2nd choke and moving the first one one cap upwards (C-L-CCC). I did some simulating in spice at work today using what euro21 has provided and this does smooth out the fluctuations considerably in comparison to how the PSU is wired now.
Will be reporting back later with results.
The layout looks quite good, very nice! I never had, or read, about a PS leading to wild oscillation at high frequency in a feedbackless amplifier, but there is a first time for everything, so I am following close.
Good luck with the further troubleshooting!
Good luck with the further troubleshooting!
Danke Erik!
Look what happened:
The Solution:
(hard to reach with the soldering iron 🙄 )
So yea it was the Grid Leaks being too large, with a second 150K in parallel the amp works with both GM70's at 100mA 🙂
Although some instability is also caused by the power supply. The more capacitance is behind the choke the more stable it seems to be so a re-design is in order. What i may end up doing as it works out well mechanically/space wise is 40µ - 1.5H - 5x40µ with the 2nd choke being replaced by another 2 Epcos caps.
Another thing i noticed is that there is definitely some switching noise present, i put 2200µ across the filament supply and it did improve it by about 50% but it's still audible when no music is playing.
Look what happened:
The Solution:
(hard to reach with the soldering iron 🙄 )
So yea it was the Grid Leaks being too large, with a second 150K in parallel the amp works with both GM70's at 100mA 🙂
Although some instability is also caused by the power supply. The more capacitance is behind the choke the more stable it seems to be so a re-design is in order. What i may end up doing as it works out well mechanically/space wise is 40µ - 1.5H - 5x40µ with the 2nd choke being replaced by another 2 Epcos caps.
Another thing i noticed is that there is definitely some switching noise present, i put 2200µ across the filament supply and it did improve it by about 50% but it's still audible when no music is playing.
There is no intentional feedback - but that's not the full story.
most devices can be caused to self-oscillate: transistors of all types, and valves. In all cases, the higher the gm [mA/V], the greater the chance of self-oscillation. Low gm triodes like GM-70s are the least likely to oscillate at first glance, but their large size, and the long wires associated with their construction make it possible.
Doesn't mean that it's happening in this example - but maybe you are interested in why it could happen!
The conditions for self-oscillation are the same: the feedback (unintentional in this case) finds its way from the anode circuit to the grid circuit by capacitive or electomagnetic coupling, having had its phase rotated by 180° at a frequency where the amplitude at the grid can sustain oscillation. The phase shift can occur because of stray and parasitic capacitance and inductance in the anode load, and anode wiring, which forms an LC tank at some high frequency; and sometimes the grid circuit is also tuned.
In all cases, the frequency of self-oscillation is much higher than the frequency of oscillation that occurs across two or more stages of gain - the case where closed-loop feedback is applied, but frequency-compensation has not been correctly implemented.
Self-oscillation of small all-glass valves can run at 100MHz or even higher; for bigger valves it might be 20-60MHz.
If the oscillation can be measured and found to be <1MHz, it is most likely to be a loop feedback problem; if >20MHz, self-oscillation is most likely.
Self oscillation is more likely if the anode wiring is long. The length of the anode wiring, for RF analysis, goes right back to the supply decoupling capacitor, and back from there to the cathode. Always place a decoupling cap such as to make a short loop around the anode-cathode circuit.
Running parallel-SE doubles the gm, and doubles the potential AF or RF signal current. It also increases the current-carrying loop area of the anode circuit, and increase the susceptible area of grid circuits...
I have not read what all others say but I would say you don't have enough power supply decoupling into the plate of the first stage. Maybe a shunt zener may be needed. i.e. motorboating.
Rod Colman
Good reply.
I have always been sceptical regarding CCS as an anode load with tubes having relatively high gm´s, be it interstage triodes or pentodes, as the tube in such working condition with an undamped load, provides all the frequency gain and transparency to oscillate, when compared having an anode resistor which acts as a damper. Worse, it is unscreened.
More thought; the EL84 (10mA/gm in triode) with such CCS as a load direct AC coupling to the GM70 is´s a sure way of creating an oscillatory circuit with the anode L&C parasitics of the output tube +Miller effect. My initial suss-out would be to replace the CCS in the EL84 anode citrcuit with an equivalent resistor and observe. An old amateur radio trick I always use when a circuit is behaving suspiciously is to put a battery AM/FM radio off-tuned nearby: it will certainly respond if a circuit is squegging which could be too high freq for a Scope. The other symptom is already indictated with increasing current, erratic meter measurement is a sure sign of instability as the needle of a moving coil meter will "kick".
Quickly browsing over the circuit, I wondered if the double Mos CCS in the anode load is correctly drawn with very short gate resistors, pref carbon or ferrite beads. Two mos seems an over-kill.
I´ve come across amps with dreadfully performing power supplies without circuit displaying any instability. As mentioned, not to rule out proximity issue with the tube layout ? A quick screen on the outside of the EL84 tube will show changes.
All hints of course.
BB
Well the oscillation was fixed by decreasing the GM70 grid leak resistors to 75K. The remaining fluctuations in the HV supply are mostly gone too after changing the PSU layout but more capacitance wouldn't hurt.
Driver stage B+ is not high enough to replace the CCS with a resistor (about 300V), EL84 Plate voltage would be pretty low. Also not such an easy thing to do as the driver stage is on a PCB.
The driver stage was tested before it was installed in the amp and hooked up to the GM70s, it showed no oscillation or any other strange behavior. Ofc now that a big triode is attached this may have changed so i will hook up the scope to the driver stage tomorrow and see how it behaves.
Driver stage B+ is not high enough to replace the CCS with a resistor (about 300V), EL84 Plate voltage would be pretty low. Also not such an easy thing to do as the driver stage is on a PCB.
The driver stage was tested before it was installed in the amp and hooked up to the GM70s, it showed no oscillation or any other strange behavior. Ofc now that a big triode is attached this may have changed so i will hook up the scope to the driver stage tomorrow and see how it behaves.
Good.
Another question. Is the 10R cathode resistor in the GM70 cathode a metal film type ? Start square wave testing under load and one might start noticing lower stage current and power as these resistor types are not good in keeping their value with fast pulse handling. Standard replacement is a wirewound type. This is often overlooked as the resistor will not show any signs of discolouration.
Keep at it!
BB
Another question. Is the 10R cathode resistor in the GM70 cathode a metal film type ? Start square wave testing under load and one might start noticing lower stage current and power as these resistor types are not good in keeping their value with fast pulse handling. Standard replacement is a wirewound type. This is often overlooked as the resistor will not show any signs of discolouration.
Keep at it!
BB
It does not affect the fluctuation, but 6N6P operating point isn't optimal.
Near 200V, 10mA, -9V is on the right side of curves (on the curved section), so distortion can be measured even at low swing.
I would use it around 120-140V, 15-20mA, -4V.
Near 200V, 10mA, -9V is on the right side of curves (on the curved section), so distortion can be measured even at low swing.
I would use it around 120-140V, 15-20mA, -4V.
Thanks for the extensive reply, Rod,
Oscillation is an ongoing thing in my latest experiments at the lab, every now and then I open a new thread about it. But there is all intentional feedback there, necessary for those crazy OTL builders... but has been a very interesting learning journey.
https://www.diyaudio.com/community/threads/nfb-in-circlotron-more-nfb-higher-thd-why.389662/
https://www.diyaudio.com/community/threads/beam-power-pentode-otl.391888/
https://www.diyaudio.com/community/threads/oscillation-with-cathode-feedback.381803/
https://www.diyaudio.com/community/threads/beam-power-otl-shifting-bias-levels.403061/
I agree, I also wrote that all active devices in the driver are prone to oscillation, at least more than the low gm output tube.
And indeed, there is a lot of "bad" power supplies there, or at least that swing and ring a lot. A well known German amp builder likes chokes with low RDC and high H combined with small caps - put that in your simulator of choice and see it ringing and swinging.
The operation points of the amp were the next thing I wanted to discuss after they were working. The GM70 will have about -80V at the grid, requiring 160V PP. It looks like the EL84 is biased at circa 30mA, so the plate can swing down to about 80V (G1 at 0V). Add 160V to that, we are at 240V, so there is 60V margin left for the CCS. It is still doable, but one must check that over the lifetime the plate really is and stays at about 180V for the symmetric swing. Some 400V would make things easier, I also started to like the gyrator circuit very much, as it allows for a constant plate voltage of the driver and to drive the output valve from a low source impedance (and going a bit more complicated, with negative supplies, even DC coupled drive of the output stage, which I think is best).
Step by step by step... I joined diyaudio 20 years ago... wow!