Cause of ultrasonic oscillation in a push-pull output stage ?

[GrimJosef]

I'm currently working on a pair of Leak TL/50 Plus amps. The circuit is shown here http://www.44bx.com/leak/Leak/Circuits/tl50_ecc83.JPG. The only difference in my amps is that the KT88 cathode resistors are 600ohm ones, not 500ohms. This was a change that Leak themselves implemented in a slightly different version of the same amp.

One of my two amps is working reasonably well. There's a slight excess of output stage distortion but I think that can be cracked.

The other amp, when using exactly the same set of valves, is a 92kHz oscillator. I'd really like to know why. Here's what I do know:

1. If I disconnect the coupling capacitors from the KT88 grids, which isolates the output stage from the other valves and also breaks the amp's global negative feedback loop, the output stage oscillates at the same frequency on its own, at slightly larger amplitude than when it's connected to the rest of the amp. So I'm convinced the cause of the trouble is in the output stage circuitry.

2. The 92kHz waveforms measured at the two KT88 anodes are in phase i.e. the anodes reach their positive peaks at the same time and their negative troughs at the same time.

3. The oscillation amplitudes are very (frighteningly ?) large, but not equal - 540V pk-pk on one anode and 415V pk-pk on the other. The difference means that a 92kHz signal appears on the transformer's secondary, which I can monitor.

4. There is a somewhat distorted 92kHz signal at each of the KT88 cathodes but it's only about 40mV pk-pk with respect to ground (there's 50V DC on the cathode of course). The cathode resistor bypass capacitors are modern 47uF electrolytics.

5. I could stop the oscillation by making small changes at either of the KT88 control grids - e.g by touching the grid tag on the valve base with a probe connected to a digital multimeter on its voltage range (other probe connected to amp ground). Or by connecting a 100pF capacitor from the grid tag to ground. Both of these things worked when the whole amp was running i.e. the coupling capacitors were also connected to the KT88 grids. If those capacitors were disconnected then it took more like 200pF from the grid tag to ground to stop the oscillation.

6. Things I've tried which don't stop the oscillation include:

Changing the KT88s from some modern Svetlana ones to some old stock GEC ones.

Running the whole amp with or without the negative feedback link connected to the input valve cathode circuit.

Swapping between Erie ceramic 'high stab' resistors and their ceramic 'carbon comp' ones for the 330k KT88 grid leaks.

Adding some modern 680ks in parallel with the 330k grid leaks to drop them to 220k or so.

Swapping between Erie ceramic carbon comp and Allen-Bradley type little brown carbon comp ones for the 10k KT88 grid stoppers.

Adding 100R carbon comp screen grid stoppers very close to the KT88 bases.

Adding a 34uF modern electrolytic capacitor across the original 32uF HT smoothing cap C13 (which is perhaps suspect, since it only measures 22uF with 0.6ohm ESR). When the extra electrolytic didn't stop the oscillation (maybe it isn't really 34uF at 92kHz) I replaced it with a very large modern 30uF 630V polypropylene cap instead. This didn't stop the oscillation either.

So there we have it. I'm struggling with this one and would really appreciate any suggestions. I confess I was surprised to find that the oscillation was, if you like, 'common mode' i.e. both KT88 anodes going up and down in phase. Is that what we'd expect ?

Cheers,

GJ

[kalee20]

Well, it's a proven design, so there must be a build fault.

As it's common-mode, what happens if you hard-wire the anodes together with a short length of wire? As a thought experiment, this should not make any difference, it should still oscillate. But that's impossible, because the OP transformer is then short-circuited across its primary, so presents zero impedance to HT+ so you can't have any signal there.

So... is there really zero impedance to HT+? What happens if you connect a 0.1uF capacitor from CT to chassis? I note you seem to have added extra HT capacitance, but in the right place?

Is the O/P transformer at fault, maybe a section wound incorrectly and giving a not-quite-centre-tap, or extra leakage inductance? Stick mains across the whole primary and measure the voltages on the various taps, compare with the 'good' one.

As you've tried disconnecting the NFB loop, it isn't an overall loop oscillation, but more a local anomaly in the output stage, or so it seems to me!

[GrimJosef]

Originally Posted by kalee20

... So... is there really zero impedance to HT+? What happens if you connect a 0.1uF capacitor from CT to chassis? I note you seem to have added extra HT capacitance, but in the right place? ...

There is something odd going on with the HT smoothing capacitance. The TL/50 Plus's wiring does go round the houses somewhat - both HT and ground. In the end I monitored the HT voltage between the output transformer CT and the bottom of the two 600ohm KT88 cathode resistors (this is a Standee two-resistors-in-one-package unit). I divided the HT down with a couple of 270ks in series with a 27k at the bottom of the chain and put a 1M scope probe across the 27k. I measured a lot more HT ripple than I could account for, given the smoothing capacitance. I moved my 30uF polyprop around between different HT and ground points and moving it did change the HT ripple. But even when it was hard-wired directly across the divider chain I could still see 410mV pk-pk at 92kHz across the 27k, indicating 9V pk-pk across the 30uF cap. I measured less than 0.33ohm ESR for that cap at 92kHz. So to get 9V ripple would require 27A of ripple current !! Something's got to be wrong there.

But the fact is that the great big cap didn't stop the oscillation.

I'll have a go at some of the other stuff in the morning. Thanks for the input.

Cheers,

GJ

[peter_sol]

If the output transformer is not the original you may need to fit a series cap and resistor from each anode to the cantre tap.Try 10K and 0.005 or 0.01
Obviously that will reduce the HF response audibly if the caps are too large.

[snowman_al]

Off the wall,
U.L taps the right way round?
Tried strapping it in triode mode?
Ground resistance / fault?

[GrimJosef]

The output transformer is the original Peter, and although the amp had been seriously 'got at' before I saw it I think the wiring connections between transformer and output valves haven't been touched.

I have been concerned about the grounding Al. Having traced it through (a nuisance as there are a lot of wire links on the underside of the tagboard) it looks visually OK and measures OK at least at low voltage and current.

Cheers,

GJ

[kalee20]

I've looked at the circuit.

Your point 1: disconnecting the coupling caps to the output grids. (I was going to ask what happens if you pull the ECC83, but disconnecting the capacitors is even more extreme!) Do you disconnect at the grid end (so there's almost nothing on the grids, just the grid stopper and the grid leak) or at the anode end, so there's still the body of the capacitor to pick up any stray field? If the former, then there can't be much coupling to the grid externally to the valves, so it's either internal via g2 to g1 (as snowman al thinks) or via dodgy earth connections to the grid leaks.

I'd still suggest strapping the anodes. It might kill the oscillation, but you could put a common-mode signal from a sig gen to both grids (also strapped) and monitor the anodes. You could look for significant signal at the anodes as you swing the frequency around, which would indicate trouble.

[Techman]

I remembered what I thought was a very recent thread on this same problem and if you'd asked me when it was I would have said it was within the last six months. I've just done a search and and unless there was another more recent thread, it was in fact back in 2017. The thread didn't seem to come to a final conclusion other than the output transformer was suspected. The amplifier was the one from one of the later versions of the Pye Black Box. As it happens, I've just done a repair/restoration on one of these in this last week, fortunately with no oscillation trouble.

Here's the thread from two years ago regarding the oscillating output stage on a Black Box - linked to below:-
https://www.vintage-radio.net/forum/...d.php?t=136602

[Chris55000]

Hi!

Have you tried series grid stopper resistors (22k) AND ferrite beads on the grid, screen–grid & anode leads of the KT88 valves as close as possible to the valveholder tags of each o/p valve?

Chris Williams

[Herald1360]

Leaky/dirty valveholder? Megger between pins can't hurt.

[Radio Wrangler]

Removing the coupling caps to the KT88 grids leaves the grids bereft of the damping created by the output Z of the preceding stage and renders the grid stopper resistors ineffective. Even a good amplifier might not like this.

Anodes in phase, common mode oscillation... That suggests leakage inductance in the OP transformer might be involved.

27 amps of ripple current in a decoupling capacitor..... I suspect that if you monitor the true Vgc of the KT88s and look up the KT88 characteristic curve, even with two valves in phase, you won't see that much current. This, if verified, would suggest a resonance in the HT end of things. The big polypropylene capacitor may have low ESR, but this may involve ESL and the current may be somewhat lower than Vripple adXc would suggest. Have the other HT capacitors been 'improved'? The ESR of good old electrolytics serves as a damper killing resonances in a lot of cases where capacitors are effectively paralleled.

The KT88 is basically a rather lively device. With a top cap anode connection and beefed up grid/cathode connections, MO-V flogged them as TT21 which saw use in transmitters up to 30MHz. There is plenty of high frequency gain available, which means oscillation can happen if not by stray capacitance coupling, then by creation of negative resistance.

David

[Diabolical Artificer]

Not many idea's why it's oscillating, but you said THD is different between amps, this might be indicitive. Is it the oscillating amp that has higher THD, what happens when you test at higher frequencies? Also what happens if you disconnect the common cathode R grounds and ground it using a wire right back to the RC filter?

To kill it try a 10r 5w and 0.1u , cap towards ground, on the OPT secondary or similar value's and maybe a 2200p and 1k5 5w from anodes to ground, cap to anode.

Andy.

[GrimJosef]

Lots of helpful thoughts since I went to bed, thanks !

1. Disconnecting the coupling caps to the grids: I did this at the grid end of the caps, kalee, at the tagboard tags in fact, marked 'x' in the attached pic (not the amp I'm working on, but pretty much identical to it). This doesn't mean that there's no wiring left attached to the grids though. The TL/50 Plus's wiring does go round the houses somewhat. I've sketched in the wiring runs which are still attached to the grid stoppers, much of them underneath the tagboard. The black lines show part of the ground wiring. At the right hand end is the connection to the Standee KT88 cathode resistors and at the far left hand end, in the corner of the amp, is the connection to the chassis. You can see the output transformer, numbered 8013, in the bottom left corner of the amp. The five wires pink/purple/red/white/yellow are connected to the primary then loomed together and run under the tagboard. Just beyond the centre of the tagboard the loom turns 90 degrees right and continues to run along with individual wires being branched out to the valve sockets and the HT smoothing cap (red wires, bottom centre).

David, I agree that disconnecting the grid caps is likely to make the oscillation worse, and indeed it does. But the amp was oscillating pretty badly before I disconnected the caps. The damping from the previous stage is substantial, but in the end it is only an ECC83, which has a high ra, in parallel with its 100k or 91k anode loads, so it doesn't help as much as it would if there was a more substantial drive valve.

2. Strapping the anodes: interesting idea. At the moment they're not oscillating with the same amplitude (25% difference or so) so strapping would push the amp away from its current fault point. I might try it. But I think I'll do some AC voltage checks on the OP transformer first.

3. More stoppers: I did try 120ohms in the screen grids Chris, but to no avail (I wrote 100ohms in the first post, but they actually measured 120). This is a threshold thing though, and GEC recommend 270ohms, so maybe my test doesn't tell us much. The more important point is that Leak made enough of these amps, all of them without any screen or anode stoppers or ferrites, and all of which didn't oscillate. So rather than manage the broken lintel by putting Acrow props here and there I think the amp's owner would sooner I just find which lintel it is that's worn out and replace it .

4. Dirty valveholder: the underside of this amp is quite grubby Chris. I'll have a closer look and give it as good a clean as I can.

5. Pye Black Box oscillation: I will make another cuppa and read that Techman, thanks.

6. 27A of ripple current. I agree that can't be right David. There has to be a large anomaly (either measurement or understanding of the cap's properties) there. It was a suck-it-and-see exercise really, as the original electrolytic smoother does measure too low in capacitance. But adding neither the great big polyprop nor some small modern electrolytics in the same place made any difference to the oscillation, so that points away from the power supply as the primary issue. I'm still hoping it is though, as a failed output transformer would be a very expensive tragedy. I didn't work on any of the other HT rail caps as the downstream ones are modern, and isolated by a 22k resistor in any case, and the upstream reservoir is isolated by the 2H choke. At 92kHz the smoother has to do all the heavy lifting.

7. Distortion: when I suppressed the oscillation in this amp and tried to measure audio distortion it did look significantly worse than the other amp Andy. But to be honest I couldn't be certain that the oscillation wasn't re-starting then so interpreting what that might have meant would have required some more careful measurements. I might try those after the other stuff. I do (sort of) have a way of doing the suppression now and although extra components might help with that I can't really use them as a permanent crutch. The cathode resistor ground is quite easy to move though. The problem is I'm not sure where it originally went as both these amps had had the Standees replaced and their current owner has sourced a pair and got me to put them back in.

Onwards and upwards !

Cheers,

GJ

[peter_sol]

I think you have eliminated the valves as they work OK in the other amp.

[Boater Sam]

I have a gut feeling that it may still be the power supply, can you feed it from the other amp or another source?

[GrimJosef]

Oh dear.

Dear oh dear.

This is embarrassing . Still, they say you learn by your mistakes ...

I've spent this morning putting AC through the output transformer. I haven't done any detailed analysis yet, but it looks fine. My final measurements were of the secondary voltages, so I could get an idea of the turns ratio. There are three secondary terminals underneath the transformer - one at the bottom of the winding, one which can be switched between the 4ohm, 8ohm or 16ohm tappings by a little plug on the transformer's top, and one which is permanently wired to the 16ohm tapping and is used to supply the negative feedback signal. First I measured the voltage between the first and second terminals. Then between first and third. What ? The voltages were the same ! It turns out I've spent a day testing this amp with the output transformer plugged for a 16ohm load but running into my standard 8ohm resistor .

When I put the little plug into the 8ohm socket the power oscillations stopped. Panic over.

The resonance hasn't gone away completely. You can see from the scruffy phone pic (please excuse the screen reflections) that it's still there (actually at 91kHz now). That's a 10-shot average taken over a second or two and on some traces (most perhaps) it's not there at all. Presumably it only appears when noise excites it. But it's 70dB down compared to yesterday and I'm no longer worried about it.

I suppose people might accuse Leak of sailing too close to the wind if underloading the valves by just a factor of two can cause really worrying oscillation. But of course he never expected his customers to connect the amp to a low-inductance resistor, and the wrong resistor at that. No speaker looks like a few ohms at 92kHz (actually having said that I'm not sure what the ESL57's transformer looks like there).

Anyway, all is now calm and peaceful. I'm sorry for wasting your time and genuinely grateful for all the help. When you're banging your head against a problem there's nothing like having smart, constructive people with good ideas on hand .

Cheers,

GJ

[Radio Wrangler]

Nothing to be embarrassed about. We all do this sort of thing. Sometimes worse!

Does the other amplifier misbehave similarly into the same mis-load, just to check that the two really are similar. It would be interesting to find that "they all do that" because general knowledge may be missing out in this area.

It is a fair indication of just how difficult it is to deploy large amounts of feedback in a transformer coupled valve amp.

David

[kalee20]

Great that you found the fault! I was thinking last night and going to suggest, what about scoping the OP transformer CT to see if it's solid, or has 92kHz (which would point towards decoupling capacitor resonance). No need now!

However, if I'd designed an amplifier that oscillated when loaded with a pure resistance - especially one which was only different from nominal by a factor of 2 - I'd consider it not ready for design release! It's more than a bit embarrassing from a design point of view,it's really not good! Something of a black mark for Leak!

[Ed_Dinning]

Hi GJ, a few black marks to Leak for the OPT as well. That design of sec taps will not use all of the sec winding to give good coupling at the various speaker tappings.

I did some later work on the Pye BB which used a bifilar wound sec in the centre of the anode windings. This also oscillated at about 30Khz.
The output trans was rewound with the sec in 3 series connected sections which fell between each anode screen junction and the centre tap.
This gave a total cure.

Ed

[GrimJosef]

Well I spoke too soon I'm afraid. I thought I was out of the woods with this amp, but I'm not.

The fact is that it's only marginally stable with an 8ohm resistive load on the 8ohm output transformer plugging. I was running it earlier this afternoon with a longish speaker lead connection (a little over 4m of thickish copper figure-of-eight, nothing audiophoolish !) and I think it started oscillating intermittently. I haven't been able to reproduce it. But it still goes straight into hard oscillation with an 8ohm load on the 16ohm tap and, worse still, it does the same with a 16ohm load on the 16ohm tap, even with a short lead to the load and a hard short-circuit of the input socket to ground

I have tried the other amp using the same valves and external leads, load etc and it doesn't misbehave. Even in the worst case - 16ohm output transformer plugging, 8ohm load, long speaker lead - there's no sign of trouble. Things get a little ultrasonically spiky if I connect an audio input source (600ohm Zout). But not worryingly so. At the output of the amp none of the spikes is above the millivolts level. I can't be sure this isn't just switchmode noise getting into the external wiring (the amp has a great deal of gain - 50W out for 125mV in).

I did try substituting a proper 32uF modern electrolytic (actually two 32+32uF cans in series-parallel, so I didn't have to worry about the 500V rating) for the smoothing cap. This made no difference at all. I then added 0.24ohm of series resistance (two 0.47ohm 'cement block' wirewounds in parallel) to mimic some ESR. Again this changed nothing.

Cheers,

GJ