Instead of a tap on the primary of the opt, why not use a second trans at the output and feed the screen with it ? It could take into account the signal at the real outout, not at the primary of the opt.
M.
M.
Some folks have done it with a 2nd HV wdg on th OPT, should there be a very large difference in Plate & G2 volts. For example PP 6146s. A few amps like that were built in the ancient world.😀
Those were my conclusions several years ago when I did those "pseudo UL" experiments using mosfets.
re: Aboulafia
"Instead of a tap on the primary of the opt, why not use a second trans at the output and feed the screen with it ? It could take into account the signal at the real output, not at the primary of the opt."
The problem with that would be the excessive phase shift of going through 2 OTs. (output tube oscillation)
------------------------------
"why would you waste time on UL, when Schade does better?"
Exactly!
------------------------------
One un-speakable UL option would be to use an Op. Amp. to compare the actual output with the signal input, and derive a grid2 signal to minimize the difference. (aka N Fdbk loop) I suppose one could use a tube Op. Amp. for that. Might work TOO well.
Alternatively, (and no extra parts needed), just return the attenuated output(s) to the driver stage cathodes as N Fdbk. Driver stage provides the loop gain.
"Instead of a tap on the primary of the opt, why not use a second trans at the output and feed the screen with it ? It could take into account the signal at the real output, not at the primary of the opt."
The problem with that would be the excessive phase shift of going through 2 OTs. (output tube oscillation)
------------------------------
"why would you waste time on UL, when Schade does better?"
Exactly!
------------------------------
One un-speakable UL option would be to use an Op. Amp. to compare the actual output with the signal input, and derive a grid2 signal to minimize the difference. (aka N Fdbk loop) I suppose one could use a tube Op. Amp. for that. Might work TOO well.
Alternatively, (and no extra parts needed), just return the attenuated output(s) to the driver stage cathodes as N Fdbk. Driver stage provides the loop gain.
Last edited:
if the trafo came with taps - use 'em, if not - go Schade. Simple.
These improvemizers work even better altogether on none-to-sound-application tubes IME.
These improvemizers work even better altogether on none-to-sound-application tubes IME.
Point of clarification:
Schade is preferable to genuine UL, or
Schade is preferable to psuedo UL?
Win W5JAG
Schade is preferable to genuine UL, or
Schade is preferable to psuedo UL?
Win W5JAG
Complex answer.
UL in general transitions the tube toward a triode characteristic, with some residual 2nd harmonic (due to the mis-match between g2 and g1 characteristics, same for triodes) (and toward reduced power output typical of a triode). It would approach 0 distortion more quickly (less tube gain needed), and accurately, if the g2 and g1 non-linear characteristics were identical. But real world mis-matched characteristics essentilly lock in the difference (leaving 2nd H).
The pseudo UL schemes that restore the FULL screen current to the plate/OT connection should do this well. (ie, looking like a triode with higher Mu than the internal g2/g1 Mu, like intMu/ULfract.) (FULL versions: typically a bootstrapped Mosfet (follower) drain load with a cap to the tube plate, to return the full AC g2 current component) Using an oversized (hard to find small HV Mosfets now) Mosfet may incur additional gate capacitance dist. Somewhat reduced power output, just like real UL.
The "real" UL tap version does not return all the screen current (40% typ), and so has additional screen current distortion components in the plate current with higher harmonics. (the wigglies in typical UL curves) Some tubes (low kink types) will still work quite well.
The usual "Schade" plate to grid resistive N Fdbk provides true linear N Fdbk, so asymptotically approaches 0 distortion, but it only has limited loop gain to enforce that, so a tapering tail of residual harmonics results. (linear N Fdbk takes more loop gain to achieve similar results as -matched- non-linear N Fdbk) Schade type N Fdbk does maintain the full pentode power capability.
So all depends on what you like the sound of.
Of course, global N Fdbk is usually also applied, so these differences may shrink into insignificance.
..
UL in general transitions the tube toward a triode characteristic, with some residual 2nd harmonic (due to the mis-match between g2 and g1 characteristics, same for triodes) (and toward reduced power output typical of a triode). It would approach 0 distortion more quickly (less tube gain needed), and accurately, if the g2 and g1 non-linear characteristics were identical. But real world mis-matched characteristics essentilly lock in the difference (leaving 2nd H).
The pseudo UL schemes that restore the FULL screen current to the plate/OT connection should do this well. (ie, looking like a triode with higher Mu than the internal g2/g1 Mu, like intMu/ULfract.) (FULL versions: typically a bootstrapped Mosfet (follower) drain load with a cap to the tube plate, to return the full AC g2 current component) Using an oversized (hard to find small HV Mosfets now) Mosfet may incur additional gate capacitance dist. Somewhat reduced power output, just like real UL.
The "real" UL tap version does not return all the screen current (40% typ), and so has additional screen current distortion components in the plate current with higher harmonics. (the wigglies in typical UL curves) Some tubes (low kink types) will still work quite well.
The usual "Schade" plate to grid resistive N Fdbk provides true linear N Fdbk, so asymptotically approaches 0 distortion, but it only has limited loop gain to enforce that, so a tapering tail of residual harmonics results. (linear N Fdbk takes more loop gain to achieve similar results as -matched- non-linear N Fdbk) Schade type N Fdbk does maintain the full pentode power capability.
So all depends on what you like the sound of.
Of course, global N Fdbk is usually also applied, so these differences may shrink into insignificance.
..
Last edited:
If you are willing to sacrifice the gain of the output. Shunt (Schade) is a classic feedback that trades gain for distortion reduction, proportionally.
So-called "Schade" feedback linearises the output stage, but at the expense of de-linearising the driver stage due to increased loading. On balance it improves things. Better or worse than UL? I don't know.
Yeah, unfortunately, Schade like, shunt N Fdbk is somewhat un-natural to drive for tubes, due to the low input Z (high current drive) at the feedback node. Either a linearized driver pentode (degenerated, moderate current capable too) or a Mosfet follower with a series resistor are required. Low Rp triode drivers have been used too, although there are dist. concerns due to Rp variation affects on the N Fdbk.
The more natural approach would be to send the feedback resistor back to the driver cathode(s). More loop gain and better match that way. Only thing is it works so well that it starts to sound like some 0.001% dist SS amp. Also much more susceptible to oscillation if not properly freq. rolled off.
Another option is to send the plate N Feedback resistors back to the driver screen grids (crossed P-P for phasing). No impact on power output, reduced loop gain, and triode like dist. Could be made very low dist. too, if using a high quality driver pentode (more nearly matched g2 and g1 characteristics, 12HL7 for example). The driver screen grids do draw some current, so lower Z feedbacks needed, or followers, or design the stage so that the plate and screens track proportionately (drawing a constant % of plate current, to look like resistors). Or use high Z loads on the driver plates for high gain (so low screen V variation).
The more natural approach would be to send the feedback resistor back to the driver cathode(s). More loop gain and better match that way. Only thing is it works so well that it starts to sound like some 0.001% dist SS amp. Also much more susceptible to oscillation if not properly freq. rolled off.
Another option is to send the plate N Feedback resistors back to the driver screen grids (crossed P-P for phasing). No impact on power output, reduced loop gain, and triode like dist. Could be made very low dist. too, if using a high quality driver pentode (more nearly matched g2 and g1 characteristics, 12HL7 for example). The driver screen grids do draw some current, so lower Z feedbacks needed, or followers, or design the stage so that the plate and screens track proportionately (drawing a constant % of plate current, to look like resistors). Or use high Z loads on the driver plates for high gain (so low screen V variation).
Last edited:
I just meant in the context of the above discussion, fella's ....
i.e. if you don't have "UL" taps, then just go with plate to grid and be done with it. Don't bother trying to emulate UL. But the context was ambiguous to me, perhaps because I am following at a lower level of understanding than you guys.
Not trying to wrench open cans of worms about what is "best" among the various methods ...
Win W5JAG
i.e. if you don't have "UL" taps, then just go with plate to grid and be done with it. Don't bother trying to emulate UL. But the context was ambiguous to me, perhaps because I am following at a lower level of understanding than you guys.
Not trying to wrench open cans of worms about what is "best" among the various methods ...
Win W5JAG
That's the very question I asked a few days ago in "about pentode mode type" topic, unfortunately non-answered...
The next day I tried it on a breadboarded two-stage (6AD10, two dissimilar pentodes) PP amp, with SFB from output plates fed to the inputs' grid2, yeah, the crossed P-P necessary. Did not like it, lots of distortion. I tried to vary the SFB resistors and the plate loads of the input/driving stage, but due to the lack of practical experience what I did could be not the most senseful, so I dropped experimenting.
Though, I did not try the SFB resistors lower than 100K, seems could be the case.
For the starter, the pentode differential input stage of 6AD10 giving twice as much gain did not sound well to my ear (without SFB), with them triode-strapped I get just enough gain to apply some SFB and I want to stay off the GNFB.
6AD10 is a tricky tube to be getting any linear results out of. Section 2 is a dual control pentode. You would want to put around +5V on grid3 to get normal square knee pentode curves from section 2. I assume you were using two of these tubes to get a differential P-P amplifier with the same tube sections.
For N Fdbk to the driver screens, you would want a resistive divider on each (crossed over) output plate to driver screen grid N Fdbk path. So that a resulting low impedance could drive the (driver) screen currents. The low value divider resistor would then connect to the +Scrn supply V. (maybe +50 to +100V range) With the mid point of each of the divider resistors connected to the driver screen grid(s). The output tube plate V will pull the screen V up some thru the divider, so adjust the screen V down some to compensate.
Since the loop encloses the gain of both the input and output stages, a lot of gain is enclosed, and one does not want more than say 10% variation of the driver pentode screen V for linearity. So figure the divider ratio to attenuate the required output signal (which is less than the input x tube gains, due to the N Fdbk gain reduction) down to say 7.5V swing or so for the screen. (ie, set the overall gain of the amp, but from required output V variation to driver screen V variation. That screen V variation will be about 3x the input grid 1 signal variation, from the gm1/gm2 ratio for section 2: or 2500/850 ) One can play around with the divider ratio used for the N Fdbks to see if one can get driver screen grid V variation similar to driver plate V variation (or up to maybe 1/2 x plate V variation, internal tube design parameter dependent). Somewhere in that range, the screen grids will draw a constant % of plate current, so will look like resistive loads (instead of the usual non-linear screen grid loads, improve linearity some).
Note:
This is about as complex a design as it gets, I only know of one other member who has tried this!
https://frank.pocnet.net/sheets/123/6/6AD10A.pdf
For N Fdbk to the driver screens, you would want a resistive divider on each (crossed over) output plate to driver screen grid N Fdbk path. So that a resulting low impedance could drive the (driver) screen currents. The low value divider resistor would then connect to the +Scrn supply V. (maybe +50 to +100V range) With the mid point of each of the divider resistors connected to the driver screen grid(s). The output tube plate V will pull the screen V up some thru the divider, so adjust the screen V down some to compensate.
Since the loop encloses the gain of both the input and output stages, a lot of gain is enclosed, and one does not want more than say 10% variation of the driver pentode screen V for linearity. So figure the divider ratio to attenuate the required output signal (which is less than the input x tube gains, due to the N Fdbk gain reduction) down to say 7.5V swing or so for the screen. (ie, set the overall gain of the amp, but from required output V variation to driver screen V variation. That screen V variation will be about 3x the input grid 1 signal variation, from the gm1/gm2 ratio for section 2: or 2500/850 ) One can play around with the divider ratio used for the N Fdbks to see if one can get driver screen grid V variation similar to driver plate V variation (or up to maybe 1/2 x plate V variation, internal tube design parameter dependent). Somewhere in that range, the screen grids will draw a constant % of plate current, so will look like resistive loads (instead of the usual non-linear screen grid loads, improve linearity some).
Note:
This is about as complex a design as it gets, I only know of one other member who has tried this!
https://frank.pocnet.net/sheets/123/6/6AD10A.pdf
Last edited:
Yes, I use a pair of them with both cascades differential.
Thanks for the detailed explanations. If I got the idea right, for the starter it could be a series network of two 51K resistors and a ~60V Zener from the output plates to ground, per each. With the resistors' joint cross-connected to the drivers g2, correct?
I assume you suggested the swing of 7.5V for the output stage considering the cathode bias of 7.5V which would be the close to max of 39 mA idle current and 11 Wa plate dissipation. I keep it much lower, about 9 Wa with current of 30 mA, which makes it it 9.5-10 V.
Though it is not any close to high-end the amp sounds very pleasant and with lot of bass on low-sensitive speakers.
Question: that +5-6V on g3, would it make the things better for the section 2 triode-strapped? (g2 to plate with 10K)
Thanks for the detailed explanations. If I got the idea right, for the starter it could be a series network of two 51K resistors and a ~60V Zener from the output plates to ground, per each. With the resistors' joint cross-connected to the drivers g2, correct?
I assume you suggested the swing of 7.5V for the output stage considering the cathode bias of 7.5V which would be the close to max of 39 mA idle current and 11 Wa plate dissipation. I keep it much lower, about 9 Wa with current of 30 mA, which makes it it 9.5-10 V.
Though it is not any close to high-end the amp sounds very pleasant and with lot of bass on low-sensitive speakers.
Question: that +5-6V on g3, would it make the things better for the section 2 triode-strapped? (g2 to plate with 10K)
"would it make the things better for the section 2 triode-strapped? (g2 to plate with 10K)"
I don't know how good a triode section 2 of the 6AD10 makes, but you would lose the driver UL effect (N Fdbk loop to driver screens). Take a look at this for a triode driver with UL like N Fdbk:
(but only has the correct phase for use with a CFB output stage, or have to use the crossed P-P N Fdbks again. The bootstrap would have to be modified, or purely a CCS, for the non CFB output stage)
I don't know how good a triode section 2 of the 6AD10 makes, but you would lose the driver UL effect (N Fdbk loop to driver screens). Take a look at this for a triode driver with UL like N Fdbk:
(but only has the correct phase for use with a CFB output stage, or have to use the crossed P-P N Fdbks again. The bootstrap would have to be modified, or purely a CCS, for the non CFB output stage)
Attachments
Last edited:
- Status
- Not open for further replies.