Did you try removing the 39p cap that Wavebourne suggested?
I never understood why people add those caps either TBH.
But there again my general comment about your amp is 'Good God how many tubes, CCS's, capacitors and transformers does that feedback loop travel through?!
I'd scrap the global feedback entirely and just use localised ones instead.
No, I am suggesting to add one more, internal loop, through 240K 2W resistors from anodes of output tubes to anodes of driver tubes. Short parallel feedback by voltage across output tubes only. It shifts up pole caused by output resistance of driver and Miller capacitance of output triode. And it does not involve parasitic loss inductance of transformer.
There's a general procedure that allows you to actually see what you're doing, rather than just shooting in the dark, and it requires a dual-trace scope. Don't say you don't have one - you need one to do the level of work you're wanting to do.
The feedback network is disconnected at the input cathode, then loaded with a resistor equal to the load it would normally see (cathode resistor in parallel with 1/gm). One scope probe on amp input, the other on the new feedback output. Observe phase shift at various frequencies and various loads while varying lead and lag options inside the amp. Some people can work with a square wave of about 10 to 20 KHz (easily made by any computer sound card) but that takes practice and luck. Best learning and realest results come from actually plotting frequency vs. phase shift. It's easy, fun and answers your questions FOR REAL.
All good fortune,
Chris
The feedback network is disconnected at the input cathode, then loaded with a resistor equal to the load it would normally see (cathode resistor in parallel with 1/gm). One scope probe on amp input, the other on the new feedback output. Observe phase shift at various frequencies and various loads while varying lead and lag options inside the amp. Some people can work with a square wave of about 10 to 20 KHz (easily made by any computer sound card) but that takes practice and luck. Best learning and realest results come from actually plotting frequency vs. phase shift. It's easy, fun and answers your questions FOR REAL.
All good fortune,
Chris
Is it easy to damp between U56 and the cathodes with a small resistor just in case the 334 chip is acting up?
The fact that it was stable with a dummy load is significant.
Problem seems to be related to the speaker load impedance going inductive at high frequencies. You can make the speaker "kinder" to the amp by addressing that directly.
Anyway its a quick thing to try.
Try a 0.22uF + 15 Ohm Zobel (48kHz) or even a 0.47uF +15 Ohms (23kHz) wired permanently across the output (speaker) terminals to supress this impedance shift a bit. Use a 5W "Rats Coffin" power resistor and a film cap, ordinary polyester cap would be fine, you can upgrade it to polypropylene later if it works.
Cheers,
Ian
Problem seems to be related to the speaker load impedance going inductive at high frequencies. You can make the speaker "kinder" to the amp by addressing that directly.
Anyway its a quick thing to try.
Try a 0.22uF + 15 Ohm Zobel (48kHz) or even a 0.47uF +15 Ohms (23kHz) wired permanently across the output (speaker) terminals to supress this impedance shift a bit. Use a 5W "Rats Coffin" power resistor and a film cap, ordinary polyester cap would be fine, you can upgrade it to polypropylene later if it works.
Cheers,
Ian
Excellent! What you'll be plottting are called Bode diagrams, which sounds fancier than it really is. If you rough-guesstimate the 45, 90, and 135 degree phase shifts by hairy eyeball, you can quickly make six data points on two curves, both with frequency as the (conventionally) x axis: gain all the way around the amplifier plus feedback network and phase shift at the same locations. Three data points each is enough. Change something and make a new plot - it's that simple.
The circuit will oscillate if gain is equal to 1 or more when phase shift hits 180 degrees. A minimum margin of 45 degrees at all load conditions will suffice, but you may want more if you can have it without too bad side effects.
All good fortune,
Chris
"This nonsense" (a small cap across the feedback resistor) is a standard way to improve loop stability by adding a little phase advance to counteract the phase lags in the forward path. It won't make a bad loop design good, but it can improve phase margin on a stable loop and so reduce ringing if the loop gain locus gets too close to the (1,0) point in the Bode plot.Wavebourn said:
"This nonsense" is best omitted during loop debugging, then added at the end if necessary. Splitting the feedback resistor, as I suggested, can help too as the cap then has less effect at the highest frequencies.
My guess is that the output stage has too much phase shift, probably caused by the OPT HF resonance, so a reactive load is tipping you over the edge. Snubbers from anodes to g2 may help, as UL often needs these. Make sure the output stage is completely stable whatever the load, with no sudden phase shifts, then close the loop and tweak the compensation network (but not randomly - measure as has been suggested by Chris).
- I know am missing something here. Wouldn't the feedback network see R10 in series with R5?
- As far as measuring the phase goes, my scope has a math function so can you use that to measure phase? I am asking this cause someone showed me the math function to perform something, but I cannot remember if it was related to measuring phase difference or not.
- Why are you mentioning square waves? Do you measure phase difference using square waves or sine waves? I would think you would use a sine wave at various frequencies.
I think your speaker wires may act as antenna injecting RF noise to the input (pre-driver) stage.
Can you please disconnect speaker box (not speaker wires), and instead of them connect your dummy load resistor, and then test again? With long wires attached, we could at least trace possible "hot spots".
Also, try to connect let's say 500 - 1000 pF capacitor across R5 in order to suppress injection of RF noise.
Injected RF does not usually cause oscillations.
The feedback network is R10, which sees a resistance at the junction of R4 and R5. Disconnecting R10 means you have to temporarily replicate this resistance. The resistance seen here consists of R5 in parallel with (R4 plus cathode impedance of U61). The cathode impedance of U61 is (1/gm plus (anode load)/mu)). The anode load of U61 is (U60 anode impedance plus mu x 1k). When you work it all out you end up with U61 cathode looks like (2/gm + 1k). So we end up with R5||(R4 + 2/gm + 1k). This is the resistance you have to put to ground from the end of R10.
The feedback network is R10, which sees a resistance at the junction of R4 and R5. Disconnecting R10 means you have to temporarily replicate this resistance. The resistance seen here consists of R5 in parallel with (R4 plus cathode impedance of U61). The cathode impedance of U61 is (1/gm plus (anode load)/mu)). The anode load of U61 is (U60 anode impedance plus mu x 1k). When you work it all out you end up with U61 cathode looks like (2/gm + 1k). So we end up with R5||(R4 + 2/gm + 1k). This is the resistance you have to put to ground from the end of R10.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.