Am I crazy??

[Chuck H]

I never checked it with a speaker load. My "bench" doesn't even exist right now (new house, haven't put it together yet). I'll look into this but I have a hunch that with my amp (and maybe some other el84 amps) the Zobel IS of benefit even with speakers. My amp never had any audibly apparent stability issues. Nor do many other el84 amps that employ a filter like this. From what I've read this "ringing" is much more common to el84 tubes and the transformers they use. Many exhibit this on the scope but sound just fine anyway. Can't say why because I haven't looked into it for myself. But I will test with a speaker whenever my bench goes up.

[g1]

According to Smith's sources, these are the snubber values I should be using. I don't know how the hell I came up with the values for the RC network I'm using now, but they ain't right (2.7k & 2.2nF??).

Might you have taken the values from a practical example somewhere?
For some of his EL34 amps, Garnet used a series arrangement containing two 3nF and a resistor. 6K for pair EL34 (Pro), 12K for quad (BTO). The resistor was 10W, the caps were 1600V, with the 2 caps in series to double the voltage handling I guess.

[Chuck H]

This ^^^^^^^^^^^^

As I mentioned, I take some care in selecting parts for this circuit because I've blown them up a couple of times. One was a shorting capacitor!!! Which means the OT primary was shorting across it's ends! Which didn't "sound" bad at all actually (other than reduced volume) but was definitely wrong.

The last parts I used were from Mouser and I'm sorry, but I can't remember the specifics. The resistor was advertised as a "high energy" resistor (?). I chose it for it's voltage specs. The capacitor was easy enough because "snubber" caps are readily available in higher voltage specs. No trouble reported with three amps that have these parts in the filter circuit. One of them my own. Which I have abused adequately

[Helmholtz]

One of the sources cited by Smith is the highly regarded RDH4 (http://www.tubebooks.org/books/rdh4.pdf).
Langford-Smith doesn't give a formula but rather a table of recommended RC pairs (see page 588).
E.g. for a Zaa of 6k he recommends 6.8k in series with 0.03µF from plate to plate or 3.3k + 0.05µFacross each half of primary.
These cap values are about 30 times larger than what the formula suggests.

I don't trust people who cite their sources wrong(ly?).

I haven't tried such large cap values.
The RDH4 RC combinations seem to be chosen to effectively compensate the inductive rise of speaker impedance and phase angle above 1kHz.

[Helmholtz]

One was a shorting capacitor!!! Which means the OT primary was shorting across it's ends!

A shorted cap places the series resistor directly across the primary. So no shorted primary but ~halved load impedance.

[Chuck H]

A shorted cap places the series resistor directly across the primary. So no shorted primary but ~halved load impedance.

Yes. It was plainly audible. I didn't even have to look for the problem. I knew where it was.

[SoulFetish]

One of the sources cited by Smith is the highly regarded RDH4 (http://www.tubebooks.org/books/rdh4.pdf).
Langford-Smith doesn't give a formula but rather a table of recommended RC pairs (see page 588).
E.g. for a Zaa of 6k he recommends 6.8k in series with 0.03µF from plate to plate or 3.3k + 0.05µFacross each half of primary.
These cap values are about 30 times larger than what the formula suggests.

I don't trust people who cite their sources wrong(ly?).

I haven't tried such large cap values.
The RDH4 RC combinations seem to be chosen to effectively compensate the inductive rise of speaker impedance and phase angle above 1kHz.

I looked up that info on page 588 (RDH4), and for a 10k primary, he recommends a 5k6 resistor in series with a .05uF cap across each half of the winding. I actually removed the existing RC network currently installed and used values he suggested. The OT im using has a 10k2 primary impedance, so I used a .047uF cap in series with a resistance of 5820 ohms across each half of the primary. Initially, when you posted the listed cap values that Smith recommends, it seemed quite large and I was worried it might adversely roll off too much high end and affect the response.
I just got done giving it a quick audio play test, and (subjectively) my quick assessment is this might be....

(sorry to cut it off there, but I'm exhausted, Lemme get some sleep and I'll give you the full right up tomorrow.)

[SoulFetish]

I'm back! Now I can follow up on the previous assessment.

(this might be...) the best thing I've done to that amp in a year.
One of the persistent troubles was that there seemed to be some instability in the high frequency response that was difficult to treat. It would manifest by sounding harsh, like almost painful at higher volumes. Solutions were difficult to find because some of the normal fixes (ie, change the tone stack curves, plate bypass caps, grid stoppers,etc) would affect the touch response, or normal treble range too much. I would compromise using a combination of all them.
Look, that's the short story. I don't want to go through the sweep tests and square wave results that have already been done. The only point I'm making now is that the above made a significant difference in how the amp sounded and responded in that there was none of the harshness in the normal treble range of the guitar - even with the treble and presence at the highest setting.
(Dare I say it)This amp might actually be done! but probably not

[Helmholtz]

(this might be...) the best thing I've done to that amp in a year.
t

So you're speaking of a 5k8 resistor in series with a .047uF cap across each primary half?
(This would correspond to the network recommended by the RDH4, i.e. Langford-Smith. The other author Smith recommends much smaller cap values.)
The network will lower open loop gain starting from around 600Hz.
Actual closed loop response depends on the amount of NFB used.

Could you ever see HF oscillation on the scope with a speaker load?

[SoulFetish]

So you're speaking of a 5k8 resistor in series with a .047uF cap across each primary half?
(This would correspond to the network recommended by the RDH4, i.e. Langford-Smith. The other author Smith recommends much smaller cap values.)

Yes, a 5k8 resistor in series with a .047uF cap across each primary half

The network will lower open loop gain starting from around 600Hz.
Actual closed loop response depends on the amount of NFB used.

Could you ever see HF oscillation on the scope with a speaker load?

I didn't check it with a speaker load alone. I did square wave test the open & closed loop gain under a resistive load, as well as test the phase angle up to 100kHz a while back. I don't recall what the bode plot looked like from memory at this point.

[Helmholtz]

I didn't check it with a speaker load alone. I did square wave test the open & closed loop gain under a resistive load, as well as test the phase angle up to 100kHz a while back. I don't recall what the bode plot looked like from memory at this point.

Instability often shows with speakers only.

Speaker load changes most everything: Square wave response, gain/phase frequency response, resonances, feedback ratio and consequently stability margin.

[SoulFetish]

Instability often shows with speakers only.

Speaker load changes most everything: Square wave response, gain/phase frequency response, resonances, feedback ratio and consequently stability margin.

yeah, you're right. We often see it in our repairs here. Sometimes even when a speaker shares a resistor load in parallel to match the output impedance is enough to cause the oscillation to disappear on our scope.

[Helmholtz]

yeah, you're right. We often see it in our repairs here. Sometimes even when a speaker shares a resistor load in parallel to match the output impedance is enough to cause the oscillation to disappear on our scope.

That's why I doubt the usefulness of compensation procedures with a resistive load as propagated by a number of authors.

I think it's essential to find out how to stimulate (signs of) instability with a speaker load.
Once the problem shows on the scope, it's easy to verify the effect of countermeasures.