Stability issues with my original 1966 JTM 50

[g1]

I would have thought that original KT66s are more valuable?

I believe they are. You have some of those in GEC too?

[trobbins]

I tried different compensation methods and could stop the oscillation on the load resistor.
BUT anything that increases stability with the resistor seems to cause instability with the speaker load. I saw just the opposite with the Vibrolux.

Will update when I have further results.

The most sensitive method I found to detect instability is to feed a square wave of maybe 400Hz to the PI input and vary signal level, as oscillation often only shows at specific levels..

What compensation methods did you try? I'm guessing you added some different small value caps across the feedback resistor and observed the ringing response on a squarewave in to a resistive load and chose the 'optimum' value of added cap that gave the least level of over-shoot ringing but not causing noticeable leading edge change? And that added cap made the onset of instability with a speaker load worse? Did you similarly try to identify a value of added cap that suppressed the onset of instability with a speaker load?

Is the feedback tap off the same tapping as the speaker you nominally connect to?

There is one other possibly surreptitious way of adding/augmenting compensation for stability. It is not a well known compensation technique, but I can confirm it works well either by its own or along with the added cap technique (where it typically reduces the value of cap needed) as a way to suppress the instability you appear to be inducing. It only affects the frequency response (ie. phase shift) around the ringing frequency, as does the parallel cap technique, so should not in any way modify the sonic performance of the amp (unless you have bat ears). The technique inserts a low uH inductor in the 0 tap of the OT secondary, and Patrick Turner (RIP) used it a lot - I can link to it if you want to assess further how that could be done.

[Helmholtz]

I believe they are. You have some of those in GEC too?

Yes.
All original KT tubes up to the 80s were made by the same UK company though owners changed, originally labeled OSRAM, then M-OV (Marconi-OSRAM valves) later GEC (General Electric Company - not identical to the US GE).

https://en.wikipedia.org/wiki/Marconi-Osram_Valve

[Helmholtz]

What compensation methods did you try?

Thanks for chiming in.
I tried both lag and lead compensation.

Will elaborate tomorrow.
Too late for me now..

[Helmholtz]

What compensation methods did you try? I'm guessing you added some different small value caps across the feedback resistor and observed the ringing response on a squarewave in to a resistive load and chose the 'optimum' value of added cap that gave the least level of over-shoot ringing but not causing noticeable leading edge change? And that added cap made the onset of instability with a speaker load worse? Did you similarly try to identify a value of added cap that suppressed the onset of instability with a speaker load?

Is the feedback tap off the same tapping as the speaker you nominally connect to?

There is one other possibly surreptitious way of adding/augmenting compensation for stability. It is not a well known compensation technique, but I can confirm it works well either by its own or along with the added cap technique (where it typically reduces the value of cap needed) as a way to suppress the instability you appear to be inducing. It only affects the frequency response (ie. phase shift) around the ringing frequency, as does the parallel cap technique, so should not in any way modify the sonic performance of the amp (unless you have bat ears). The technique inserts a low uH inductor in the 0 tap of the OT secondary, and Patrick Turner (RIP) used it a lot - I can link to it if you want to assess further how that could be done.

Ok, first I tried different ways of lag compensation.
The most effective one was a small cap (47p to 100p) in series with a 22k resistor between the power tube grids (or between the PI plates).

Then I tried lead compensation (as you describe) with a 47p cap across the 27k NFB series resistor and that worked even better.
Best results I achieved with a combination of both methods.

In principle I proceeded as you described.
But I noticed that it's essential to vary the signal level, as the instability may vanish at some level.
Also optimum compensation required somewhat different component values at low and high levels.

That was with a resistive load.
WIth a speaker load I didn't see signs of instability without compensation (except for the original 47p cap between the PI plates, which I left untouched for all tests.).
There is considerable square wave overshoot but little ringing.
Any of the compensation measures above introduced instability with speaker. Here the oscillation often developed after some delay.
I did not try to compensate the effect of compensation.

The feedback is taken from 16 Ohm secondary and both resistor and speaker load were 16 Ohm.

I'm interested in the inductor technique, please post the link.


Readable schematic for reference:
http://www.marstran.com/1963.gif

[trobbins]

Most squarewave testing to adjust high frequency (ie. circa your 100kHz) resonance response would typically use a resistor load first to optimise the parallel cap value, and once that suppresses the ringing but doesn't slow down the rising edge too much (given this technique is about subtle phase adjustment, not gross change), then the amp squarewave response is checked with open-circuit, and a few capacitor only loads (typ. between 10nF and 470nF) to confirm no unstable operation (although ringing can get gross) with presence control at worst-case end.

The OT frequency response in post #6 can depend on how the OT windings are driven and configured - if you are seeing 100kHz ringing on the squarewave test with a matched resistor load then that may indicate your frequency response test setup was not quite valid as it shows a resonance at circa 60-70kHz. But the main outcome of that frequency response test does indicate that any subsequent resonance is out near 200kHz, and less likely to interact.

The speaker loading is likely to imitate an open-circuit out at 100kHz, and an RC zobel on the 16 ohm tap may be a simple effective way to avoid unstable operation with a speaker load - certainly worth checking. The R value is usually made close to the speaker nominal impedance (so something like 15 to 27 ohm), and the C provides a corner frequency out towards the resonance frequency - eg. 15 ohm and 220nF (if I've done my maths right) is circa 50kHz corner.

The inductor technique is not for the faint hearted, and would require effort on your part, so I'd suggest not going this path unless the zobel provides no benefit. Patrick Turner's website has gone - there is a mirror copy website (link below or http://public.digicom.bg/) but I don't know the origin of that mirror. Patrick had a webpage '100w-monobloc2-2004' where he describes his form of global series current NFB, and another webpage 'Dynaco-mkIV-reformed' where he provides an easier schematic to follow on where to add the inductor and on the design of the inductor value needed. I have pdf's of those pages if the link doesn't work. I haven't yet come across another reference for that technique.

I have tested this technique in a Williamson amp and can indicate that the inductor part may not be simple to implement for a few reasons. One issue is that an appropriate inductance value requires swap-and-compare testing (same as identifying the parallel cap value). Another issue is that the value is likely down at the 1-2uH range, and off-the-shelf parts may be difficult to obtain - I had a batch of small 22uH wirewound inductors (http://www.ecmelectronics.co.uk/pdf/pk.PDF) and was able to remove turns to get down to step values around 1uH by using one-turn increments to test for squarewave ringing. I also had a way to measure inductance at circa 1uH level using my soundcard and REW software - which may be difficult for some. Another issue is that although the inductor is minute, it has to be located within the existing wiring, and ground wiring needs to be subtly modified. Another issue is that the ferrite core material can affect performance, so using another inductor of the same inductance may give a slightly different squarewave ringing response - I had some 'wire through ferrite tube' parts used for emi suppression that also had approx 1uH value - there are also slip-on ferrite tubes that could allow incremental inductance increases for the purpose of optimising a squarewave response.

[Helmholtz]

then the amp squarewave response is checked with open-circuit, and a few capacitor only loads (typ. between 10nF and 470nF)

Do you really mean without load? I wouldn't want to risk that.

I'm not getting the point of a capacitor load. On the one hand speaker impedance is inductive at HF.
On the other hand the OT secondary already provides an effective/reflected capacitance around 50nF.

The OT frequency response in post #6 can depend on how the OT windings are driven and configured

OT frequency response was measured according to class B operation with pentodes, i.e. only one primary side driven with a 22k source impedance, primary CT and correct end of secondary as well as core grounded. One curve for each side.

I don't think an amp necessarily oscillates exactly at an OT resonant peak. The peak just means a local open loop gain maximum and its contribution to phase shift at the peak is zero. It seems more likely that the oscillation frequency lies somewhere on the descending slope above the peak, where phase turns capacitive and OL gain still is large enough.
Also speaker inductance will shift the resonance. Probably more realistic to measure OT response with a speaker load.

I haven't tried a Zobel yet.

What is the theory behind the inductor method?
I have some professional experience with designing HF inductors and can measure low values up to 100kHz.

[trobbins]

Using open-circuit or capacitive loading has been found to often impose the worst-case stability conditions on a wide variety of valve amps with global negative feedback (around the output transformer). The input signal is low for these tests as the intent is to observe squarewave response, so yes if the output stage does go into oscillation then quickly remove the input signal - imho there is no more risk than cranking the output with a heavy strum, but of course there is a risk with anything.

Speaker impedance certainly goes inductive above the audio band, but any inductor will have its first resonance at a relatively high impedance level and then go capacitive for a while - I haven't seen any data on when that happens, but will add it to my testing list next time I do some impedance testing out to 96kHz.

There may be a shift in frequency response for class A operation, which is certainly the operating condition for full sinewave operation at low signal level. How much that changes response I'm not sure - although I didn't appreciate that the two curves were a curve for each side (I had puzzled over what the two curves were for), and the difference in response quite interesting. Different levels of leakage inductance for the half-primary windings could well be quite significant, and as you indicate, the onset of instability could be very dependant on the signal level and when operation starts to traverse in to class B (and on a scope may only show as oscillation during a portion of a low frequency sinewave).

In squarewave testing I have done with hi-fi amps, without feedback applied then the damped oscillatory response has typically been observable at the OPT first resonance frequency, but closed loop feedback instability typically occurs out well beyond 100kHz at the next OPT resonance (due to better winding interleaving, and often class A only operation), and so phase compensation may typically be targeted at a much higher frequency. It seems like your OPT is having sufficient gain/phase disturbance at circa 100kHz to cause an unstable feedback loop condition.

The inductor method provides some increasing phase shift out where the unstable condition occurs, in the same manner as the parallel cap does - both methods oppose the phase shift as resonance is approached to improve the phase margin where loop gain falls through zero - more phase margin reduces the damped oscillation response on the top of the squarewave. The theoretical treatment sees the capacitor induced phase shift as being caused by the output voltage signal, where as the inductor induced phase shift is caused by the output current signal. Out at the resonance frequency I'd guess that not everything is a simple voltage source or current source, and theory typically only goes in to simpler 'just' voltage or 'just' current based feedback analysis, and so people like to pigeon hole a particular type of feedback during discussion. Most audio valve amp feedback is effectively voltage driven, especially where the feedback point is in to a low value resistor like 100 ohm. The Williamson amp uses a 470 ohm resistor which is starting to mix the level of voltage and current induced feedback. A 5k presence pot may be getting even more current influence. But the outcome is that a mix of the two ways to generate that phase shift can typically provide a better stability outcome - which is why Patrick probably liked that idea. Sadly Patrick passed away last year, and I didn't get the chance to ask him to elaborate on that point during earlier correspondences.

When I used the inductor method I wanted to know what inductor values were being inserted, to better associate the change in squarewave ringing response to what I had observed when changing the parallel cap for that method. I used a 22uH ferrite open core (Panasonic ELC08D) which had 24.5T of ~0.44mmD copper wire. Reducing the number of turns I prepared samples with 6.5T = 2.4uH ; 5T = 2.0uH ; 4T = 1.6uH ; 3T = 1.3uH ; 2T = 1.05uH. The ferrite tube on wire part I had measured 0.9uH to 1uH (20mm lead lengths), which increased to 1.55uH with short ferrite bead slipped on one leg. I also had some small ferrite beads, and one of those with 2T (ie. 1 loop) = 2.3uH. The squarewave response wasn't exactly the same when using a ELC08D core compared to the emi ferrite tube part, for effectively the same measured part inductance. Below is link to three squarewave responses when using 2T, 4T and 6T (ie. 1uH, 1.6uH, 2.4uH) in a Williamson amp I have that had its phase margin concern out at 200kHz. (Problem with attaching link below and also to inserting the following link so you may have to copy the full phrase https://www.dalmura.com.au/static/2T 4T 6T inductor feedback responses.pdf)

[Helmholtz]

Using open-circuit or capacitive loading has been found to often impose the worst-case stability conditions on a wide variety of valve amps with global negative feedback (around the output transformer). The input signal is low for these tests as the intent is to observe squarewave response, so yes if the output stage does go into oscillation then quickly remove the input signal - imho there is no more risk than cranking the output with a heavy strum, but of course there is a risk with anything.

My concern is that with open circuit or pure capacitive loading the pentodes won't provide much resonance damping.
Also I don't see how a capacitor load could emulate a speaker load.

My speaker measurements show inductive phase between 1kHz and 100kHz.
Speaker impedance phase is capacitive just above the bass resonance up to a few hundred Hertz. Maybe the capacitor load is meant to test for LF stability?

Anyway, from what I've seen I won't rely on any test not using a real speaker load (or a very good "reactive " dummy load).
Of course the drawback using a speaker load is that you need earplugs and tolerant, deaf, distant or absent neighbors .

the onset of instability could be very dependant on the signal level and when operation starts to traverse in to class B

Good point!

[pdf64]

No experience with 6CA7s in a Marshall. Do you like them?
...

I asked because I'd recently found a pair of GE 6CA7 in my 'pulls' box, back from when I was doing a lot of repairs in the 80s.
This past couple of weeks I tested them, they're really good and strong, can achieve slightly higher peak current than even my best Mullard EL34s, and not too bad of a match
In use at high volume, I've not been able to detect much tonal difference with them though.

I'm concerned that my old pair of Groove Tube KT66 are getting a bit tired and weak, my JTM45 build with them hasn't been 'cutting the mustard' like it used to, though everything still seems to test ok on the bench.

[Helmholtz]

I asked because I'd recently found a pair of GE 6CA7 in my 'pulls' box, back from when I was doing a lot of repairs in the 80s.

As the 1960 Tung-Sol datasheet shows the 6CA7 as a pentode design, could you verify that your GEs have beam confining plates instead of G3?

https://frank.pocnet.net/sheets/127/6/6CA7.pdf

[pdf64]

As the 1960 Tung-Sol datasheet shows the 6CA7 as a pentode design, could you verify that your GEs have beam confining plates instead of G3?

https://frank.pocnet.net/sheets/127/6/6CA7.pdf

In the absence of a destructive autopsy, I'm pretty sure
Photo positioning in a post is beyond me, sorry.
1st 2 photos are the GE 6CA7, next 2 are Mullard EL34.
Viewed through the holes in the side of the anode, the 6CA7 certainly looks to have a large plate structure under the anode.
And there's only 2 sets of grid support struts below the lower mica.

Whereas the EL34 has grids underneath the anode, and has 3 sets of grid support struts.

[Jazz P Bass]

I thought that the 6CA7 was a beam tetrode.
"After the Phillips patent on the suppressor grid had expired, many beam tetrodes were referred to as "beam power pentodes". In addition, there were some examples of beam tetrodes designed to work in place of pentodes. The ubiquitous EL34, although manufactured by Mullard/Phillips and other European manufacturers as a true pentode, was also produced by other manufacturers (namely GE, Sylvania, and MOV) as a beam tetrode instead. The 6CA7 as manufactured by Sylvania and GE is a beam tetrode drop-in replacement for an EL34, and the KT77 is a similar design to the 6CA7 made by MOV."
Beam tetrode - Wikipedia

[Helmholtz]

I thought that the 6CA7 was a beam tetrode.
"After the Phillips patent on the suppressor grid had expired, many beam tetrodes were referred to as "beam power pentodes". In addition, there were some examples of beam tetrodes designed to work in place of pentodes. The ubiquitous EL34, although manufactured by Mullard/Phillips and other European manufacturers as a true pentode, was also produced by other manufacturers (namely GE, Sylvania, and MOV) as a beam tetrode instead. The 6CA7 as manufactured by Sylvania and GE is a beam tetrode drop-in replacement for an EL34, and the KT77 is a similar design to the 6CA7 made by MOV."
Beam tetrode - Wikipedia

Maybe, but the 1960 6CA7 Tung-Sol datasheet clearly shows a true pentode.

BTW, the European Philips (only one L) of the Netherlands which owned Mullard and the North American Phillips were different companies, both making tubes in the past.

[pdf64]

I thought that the 6CA7 was a beam tetrode.
"After the Phillips patent on the suppressor grid had expired, many beam tetrodes were referred to as "beam power pentodes". In addition, there were some examples of beam tetrodes designed to work in place of pentodes. The ubiquitous EL34, although manufactured by Mullard/Phillips and other European manufacturers as a true pentode, was also produced by other manufacturers (namely GE, Sylvania, and MOV) as a beam tetrode instead. The 6CA7 as manufactured by Sylvania and GE is a beam tetrode drop-in replacement for an EL34, and the KT77 is a similar design to the 6CA7 made by MOV."
Beam tetrode - Wikipedia

The term 'beam tetrode' didn't seem to get used much by valve manufacturers in the 40s-60s. GEC then revived it in the 70s, perhaps to differentiate their KT series from Philips competiton.
I think of the major valve manufacturers, only RCA never used the term 'beam pentode', or just 'pentode', to describe the valve types that get referred to on the internet as 'beam tetrodes', eg the 6L6.
And even they referred to the 6L6 as a beam power valve, rather than a beam tetrode.
My view is that tetrode here is a misnomer, as there's 5 electrode structures.

https://tubedata.altanatubes.com.br/.../049/6/6L6.pdf
https://tubedata.altanatubes.com.br/...7/6/6L6WGB.pdf
https://tubedata.altanatubes.com.br/...93/6/6L6GC.pdf
https://tubedata.altanatubes.com.br/...126/k/KT66.pdf
https://tubedata.altanatubes.com.br/...086/k/KT66.pdf