Ok, I'm enjoying reading at this point and I'm going to keep it that way but I will throw a series of papers by Langford-Smith into the ring. This was written a little after RDH 4ed was published and is a fairly comprehensive look at UL, it certainly doesn't assume 43% as the be all and end all.
I'm not sure if it's any use here, but I've not seen it on the net before, so here we are. Apologies for the sometimes dodgy scan.
http://greygum.net/uploads/tronics/UL.pdf
I've got a few of these old things to go up when I get time...
I'm not sure if it's any use here, but I've not seen it on the net before, so here we are. Apologies for the sometimes dodgy scan.
http://greygum.net/uploads/tronics/UL.pdf
I've got a few of these old things to go up when I get time...
I noticed that there was 56 left when I ordered mine. They were quoting a 26 week lead time for more. Now there are still 56 left, but the lead time is now 8 weeks. I assume the lead time changed because they reordered some more.
It seems that I have no problem finding tubes that haven't been made in 50 years, but getting modern semiconductors in small quantities is always a gamble. The IXYS CCS IC's are the worst example. DigiKey either has 1000+ in stock, or they have none, with a 10+ week lead time (the current situation).
Thanks for the Fritz papers
thevoice,
Thanks for the Langford Smith stuff, I'd never seen that before. As you said, its not covered in RDH 4 although I see he quoted pg 570 and pg 564 in his references.
It covers some of the stuff Don was concerned about - in particular optimum loading.
I've printed it off for some "light" reading.
Cheers,
Ian
thevoice,
Thanks for the Langford Smith stuff, I'd never seen that before. As you said, its not covered in RDH 4 although I see he quoted pg 570 and pg 564 in his references.
It covers some of the stuff Don was concerned about - in particular optimum loading.
I've printed it off for some "light" reading.
Cheers,
Ian
Hello thevoice,
Thank you very much for this article series! Very informative. I do however have to eat humble pie now, but only half of it.
The article shows that there is indeed a strong constraint on the primary impedance of the xfmr. It generally has to be 10% to 25%
above the optimum Z for pentode operation. (the increase due to the effectively reduced screen voltage from pentode operation requiring increased plate voltage to get up to max power again) The low end 10% is for best distortion, the high end 25% is for best power.
The UL tap % appears to then be mandated by obtaining a match between the effective triode Zout and the xformer Zload. This combination (preferred primary Z and effective load match) essentially determines a single preferred tap % for that tube. Although they don't state this, it just comes out as agreeing with their measurement.
Incidentally, they refer to a voltage feedback model used by Williamson and Walker in their "Amplifiers and Superlatives" article as being confirmed for Zout and gain measurements by their data. This model is similar to my effective triode model except they calculated everything with respect to the pentode rather than with respect to the triode. Their measured graphs are in accordance.
The second article on UL by Langford-Smith & Chesterman uses the 6V6 and comes up with a different optimum %UL tap. So these optimum taps are different from tube to tube. The Kt66 came out at 43% winding tap (18.5% impedance) and the 6V6 came out at 22.4% winding tap (5% impedance). It may be that the range of optimum tap % is relatively limited for available tubes however.
L-S & C also show that distortion does NOT shoot up at higher %taps as H&K had in their graph. In fact it stays constant and low. Power drops down with higher %UL tap though. (Obviously H&K grossly overdrove the g1 for their chart)
L-S & C also mention an upcoming article to compare UL with CFB. You wouldn't happen to have a copy of that?
Don
Thank you very much for this article series! Very informative. I do however have to eat humble pie now, but only half of it.
The article shows that there is indeed a strong constraint on the primary impedance of the xfmr. It generally has to be 10% to 25%
above the optimum Z for pentode operation. (the increase due to the effectively reduced screen voltage from pentode operation requiring increased plate voltage to get up to max power again) The low end 10% is for best distortion, the high end 25% is for best power.
The UL tap % appears to then be mandated by obtaining a match between the effective triode Zout and the xformer Zload. This combination (preferred primary Z and effective load match) essentially determines a single preferred tap % for that tube. Although they don't state this, it just comes out as agreeing with their measurement.
Incidentally, they refer to a voltage feedback model used by Williamson and Walker in their "Amplifiers and Superlatives" article as being confirmed for Zout and gain measurements by their data. This model is similar to my effective triode model except they calculated everything with respect to the pentode rather than with respect to the triode. Their measured graphs are in accordance.
The second article on UL by Langford-Smith & Chesterman uses the 6V6 and comes up with a different optimum %UL tap. So these optimum taps are different from tube to tube. The Kt66 came out at 43% winding tap (18.5% impedance) and the 6V6 came out at 22.4% winding tap (5% impedance). It may be that the range of optimum tap % is relatively limited for available tubes however.
L-S & C also show that distortion does NOT shoot up at higher %taps as H&K had in their graph. In fact it stays constant and low. Power drops down with higher %UL tap though. (Obviously H&K grossly overdrove the g1 for their chart)
L-S & C also mention an upcoming article to compare UL with CFB. You wouldn't happen to have a copy of that?
Don
You've stirred my memory and I found a reference to UL in one of my books "High Fidelity Sound Reproduction" 2nd Ed. E Molloy (ed.) Newnes (1958).
"It will be noted from Fig. 5 that the optimum point for operation is approximately 0.2 or 20 per cent. This is applicable to KT66 valves, the optimum ratio for Mullard EL34s being 43 per cent.
An even more efficient variation of this circuit is used by the Quad II (Fig. 6). Here the common portion of the winding is inserted in the cathode, resulting in the voltage being effectively applied to the grids as negative feedback.
Fig. 5: (below) Ratio of common to total windings for an ultra-linear circuit. (Showing reduction of harmonic distortion obtained by choice of optimum tapping point for tetrode. Main feedback loop disconnected."
"It will be noted from Fig. 5 that the optimum point for operation is approximately 0.2 or 20 per cent. This is applicable to KT66 valves, the optimum ratio for Mullard EL34s being 43 per cent.
An even more efficient variation of this circuit is used by the Quad II (Fig. 6). Here the common portion of the winding is inserted in the cathode, resulting in the voltage being effectively applied to the grids as negative feedback.
Fig. 5: (below) Ratio of common to total windings for an ultra-linear circuit. (Showing reduction of harmonic distortion obtained by choice of optimum tapping point for tetrode. Main feedback loop disconnected."
Attachments
I've just been reading the Radiotronics article and a thought has occurred to me...
To me, the obvious meaning of the term "tapping point" is as a proportion of the number of primary turns, but Langford-Smith muddies the waters by insisting on it being the proportion of primary impedance. If the originator of the KT66 graph used impedance as their metric but Mullard used turns ratio, then the results for KT66 and EL34 would have been in broad agreement. And frankly, I would expect them to be in agreement because it would have made good commercial sense for GEC and Mullard to make valves that could be substituted for one another in a commercial amplifier with only a change of bias resistor - not a change of output transformer.
To me, the obvious meaning of the term "tapping point" is as a proportion of the number of primary turns, but Langford-Smith muddies the waters by insisting on it being the proportion of primary impedance. If the originator of the KT66 graph used impedance as their metric but Mullard used turns ratio, then the results for KT66 and EL34 would have been in broad agreement. And frankly, I would expect them to be in agreement because it would have made good commercial sense for GEC and Mullard to make valves that could be substituted for one another in a commercial amplifier with only a change of bias resistor - not a change of output transformer.
I may, but if I do I don't seem to have labelled it properly, I'll have a look in the next day or two when I'm not at work. We may get lucky.Originally posted by smoking-amp L-S & C also mention an upcoming article to compare UL with CFB. You wouldn't happen to have a copy of that?
Don
I'm intending to put quite a few of these articles up on the net for both historical and technical interest (some more the former than the latter), time is the issue, as always.
Friday morning fog
Thanks for checking. I will try the library to see if any univiersity around has them. I assume this journal is long out of print. Being from Australia, not many will have it here. Sounds like it had a lot of good articles, maybe someone should try to do an anthology reprint!
Optimum Transformer Zload thoughts:
I have been reviewing the data and have come to some conclusions on why an optimum transformer impedance exists (and hence an optimum %UL too) for UL (and pentode too).
If xfmr. Zload impedance is too low, then B+ and screen voltages are low, which leads to low bias voltage on grid1. Since more current is needed for the same power at low Zload, g1 headroom dissappears. So a miminum Zload exists, where g1 headroom is just sufficient to stay out of g1 grid current and distortion.
On the high xfmr. Zload end, the plate voltage swings around hugely, and plummets far below screen voltage, causing screen current distortion. See L-S & C page 70-71 figs 1A & 2A for 10K and 20K xfmrs.
Another factor affecting screen current is the %UL tapping as seen in L-S & C page 57, fig. 4. Low %UL taps prevent the screen from following the plate to minimise screen current draw when the plate voltage bottoms out. This effectively rules out %UL taps below 18.5% (impedance) or 43% turns (at least for the tube used for the graph).
Screen current distortion is obviously the main criteria limiting lower %UL and higher Zout xfmrs. Lower screen current tubes and better grid alignment might allow this balance to shift toward lower %UL. (Lower power output toward the triode mode end being the other major constraint in the optimum selection balance, due to DC screen voltage boost being lost.)
[Conclusion: Using a MOSFET screen driver with its drain current returned to the plate tap (using a floating + boost supply) should allow elimination of this screen current distortion and improve overall UL distortion noticeably, for pentodes too.]
I now suspect that matching the effective triode Zout to the xfmr Zload via %UL selection is a second order affect that just happens to be doable within the main constraint window (window from screen dist. versus power out). This matching allows an additional dip in distortion through what appears a novel mechanism. My reasoning follows:
1) The match calculated between effective triode Zout and xfmr Zload always seems to occur (for the cases analyzed so far) for the Zout occuring at the maximum signal peak when optimum %UL is being used, so is signal level dependant. (Zout varies with signal amplitude due to the Ik^(-1/3) term.)
2) Looking at the fig. 5 graphs, page 59 in L-S & C, you will notice a minimum notch in the THD curve versus input signal that can be moved over to near the max. signal (zero grid condition) by selection of higher %UL tap. This is following exactly the effective Z matching expectation versus %UL.
3) L-S & C mention on page 57-58, "overload characteristics & dist." that the output signal peaks look different in optimized UL versus pentodes, with a flat top instead of a rounded saturation.
My guess is that the Z matching is straightening out the usual curved saturation plate curve by providing better matching at the peak output versus low signal level.
If so, then this scheme should work for P-P class A triode output also. Zload = Rp instead of the more usual Zload = 5*Rp
I say P-P class A output because setting Zload = Rp makes for bad 3/2 power law current distortion in SE triode. But this is likely mainly 2nd harmonic, so it can be cancelled safely in triode class A P-P. (and it cancels out in UL mode as well) The Z match also maximises power output for triode mode.
So MY design criteria for UL would now be the following:
1) Use an XFMR Zload that is toward the lowish side, but above the minimum Zload that avoids grid1 current for the maximum pentode rated power.
As L-S & C suggest, this should occur about 10% to 25% above the optimum Zload for pentode, due to the effective drop in screen voltage in UL versus pentode (because the screen is not staying put at the DC level, needs a higher operating V, hence higher Z xfmr)
2) next calculate the %UL to get the effective triode Zout to match the xfmr Zload. [One may have to reverse this order due to the limited availibity of anything other than 43% taps. (ie. figure the optimum xfmr Zload for a 43% tap) But see below on MOSFETs.]
3) use MOSFET followers for driving the screens (primary taps used for the gate drives, or resistive dividers if no taps are available or at a suitable %) and connect the MOSFET drains to the plate taps with floating +boost supplies (to allow the MOSFETs to operate when peak plate V is below screen V). This way, screen current is returned to the plate winding and does not cause distortion.
Note: it is probably possible to come up with some bootstrap scheme using diodes and caps to make the + boost drain supplies from just plate V swing. To be continued....
Don
Thanks for checking. I will try the library to see if any univiersity around has them. I assume this journal is long out of print. Being from Australia, not many will have it here. Sounds like it had a lot of good articles, maybe someone should try to do an anthology reprint!
Optimum Transformer Zload thoughts:
I have been reviewing the data and have come to some conclusions on why an optimum transformer impedance exists (and hence an optimum %UL too) for UL (and pentode too).
If xfmr. Zload impedance is too low, then B+ and screen voltages are low, which leads to low bias voltage on grid1. Since more current is needed for the same power at low Zload, g1 headroom dissappears. So a miminum Zload exists, where g1 headroom is just sufficient to stay out of g1 grid current and distortion.
On the high xfmr. Zload end, the plate voltage swings around hugely, and plummets far below screen voltage, causing screen current distortion. See L-S & C page 70-71 figs 1A & 2A for 10K and 20K xfmrs.
Another factor affecting screen current is the %UL tapping as seen in L-S & C page 57, fig. 4. Low %UL taps prevent the screen from following the plate to minimise screen current draw when the plate voltage bottoms out. This effectively rules out %UL taps below 18.5% (impedance) or 43% turns (at least for the tube used for the graph).
Screen current distortion is obviously the main criteria limiting lower %UL and higher Zout xfmrs. Lower screen current tubes and better grid alignment might allow this balance to shift toward lower %UL. (Lower power output toward the triode mode end being the other major constraint in the optimum selection balance, due to DC screen voltage boost being lost.)
[Conclusion: Using a MOSFET screen driver with its drain current returned to the plate tap (using a floating + boost supply) should allow elimination of this screen current distortion and improve overall UL distortion noticeably, for pentodes too.]
I now suspect that matching the effective triode Zout to the xfmr Zload via %UL selection is a second order affect that just happens to be doable within the main constraint window (window from screen dist. versus power out). This matching allows an additional dip in distortion through what appears a novel mechanism. My reasoning follows:
1) The match calculated between effective triode Zout and xfmr Zload always seems to occur (for the cases analyzed so far) for the Zout occuring at the maximum signal peak when optimum %UL is being used, so is signal level dependant. (Zout varies with signal amplitude due to the Ik^(-1/3) term.)
2) Looking at the fig. 5 graphs, page 59 in L-S & C, you will notice a minimum notch in the THD curve versus input signal that can be moved over to near the max. signal (zero grid condition) by selection of higher %UL tap. This is following exactly the effective Z matching expectation versus %UL.
3) L-S & C mention on page 57-58, "overload characteristics & dist." that the output signal peaks look different in optimized UL versus pentodes, with a flat top instead of a rounded saturation.
My guess is that the Z matching is straightening out the usual curved saturation plate curve by providing better matching at the peak output versus low signal level.
If so, then this scheme should work for P-P class A triode output also. Zload = Rp instead of the more usual Zload = 5*Rp
I say P-P class A output because setting Zload = Rp makes for bad 3/2 power law current distortion in SE triode. But this is likely mainly 2nd harmonic, so it can be cancelled safely in triode class A P-P. (and it cancels out in UL mode as well) The Z match also maximises power output for triode mode.
So MY design criteria for UL would now be the following:
1) Use an XFMR Zload that is toward the lowish side, but above the minimum Zload that avoids grid1 current for the maximum pentode rated power.
As L-S & C suggest, this should occur about 10% to 25% above the optimum Zload for pentode, due to the effective drop in screen voltage in UL versus pentode (because the screen is not staying put at the DC level, needs a higher operating V, hence higher Z xfmr)
2) next calculate the %UL to get the effective triode Zout to match the xfmr Zload. [One may have to reverse this order due to the limited availibity of anything other than 43% taps. (ie. figure the optimum xfmr Zload for a 43% tap) But see below on MOSFETs.]
3) use MOSFET followers for driving the screens (primary taps used for the gate drives, or resistive dividers if no taps are available or at a suitable %) and connect the MOSFET drains to the plate taps with floating +boost supplies (to allow the MOSFETs to operate when peak plate V is below screen V). This way, screen current is returned to the plate winding and does not cause distortion.
Note: it is probably possible to come up with some bootstrap scheme using diodes and caps to make the + boost drain supplies from just plate V swing. To be continued....
Don
Hardware test progress
I am actually making some progress on the test hardware here. I have the SE triode config working for 6L6. I still have to do the MOSFET screen follower yet. I'm looking at a bootstrap scheme for the MOSFET drains to connect to the plate taps. Maybe by Sunday or Monday I will start taking spectrographs. I will want to play with the setup a bit first, to narrow it down to what graphs show interesting stuff.
Don
I am actually making some progress on the test hardware here. I have the SE triode config working for 6L6. I still have to do the MOSFET screen follower yet. I'm looking at a bootstrap scheme for the MOSFET drains to connect to the plate taps. Maybe by Sunday or Monday I will start taking spectrographs. I will want to play with the setup a bit first, to narrow it down to what graphs show interesting stuff.
Don
I have been intently reading this thread every day. Many of Don's conclusions are based on the onset of grid (G1) current as a limiting factor. Tubecad simulations show that a 6L6GC or a KT88 can be operated well into A2 and the distortion will remain low. I have used A2 on several DHT's (including the 300B) with good results, but I haven't tried it on indirectly heated tubes. These two tubes have been successfully used in AB2 push pull circuits, so grid current may not be a limiting factor with a good driver design (yet another mosfet). Here again testing is needed to prove (or disprove) the simulations.
I started assembling a test setup last night. I am using one of my TubelabSE boards with an octal socket wired into it. This is capable of more grid current than a 6L6 or KT88 can eat. I am using a 0 to 400 volt regulated supply for the plate voltages and a 0 to 555 volt supply for the screen mosfet. This will allow total flexibility without worrying about bootstrapping the fet voltage.
The mosfets and other components should arrive today, but another family emergency may postpone my testing.
I started assembling a test setup last night. I am using one of my TubelabSE boards with an octal socket wired into it. This is capable of more grid current than a 6L6 or KT88 can eat. I am using a 0 to 400 volt regulated supply for the plate voltages and a 0 to 555 volt supply for the screen mosfet. This will allow total flexibility without worrying about bootstrapping the fet voltage.
The mosfets and other components should arrive today, but another family emergency may postpone my testing.
"straightening of a transfer function (i.e. increasing it's order) is by-product of a NFB."
Yes, I agree, and the NFB IS increasing with greater %UL (for a fixed Zload anyway). So this could be the entire explanation for the flat-topping peak observed. (with no NFB in the pentode case)
But notice in Fig. 5 page 59 of L-S & C that the distortion minima is sensitive to voltage drive level versus %UL. My guess is the voltage sensitivity of the minima, THD dip, position could be ascribed to the changing (lowering) of Mu effective versus %UL (via the NFB). But the fact that there is a minima/THD dip at all versus drive level suggests some non-linear mechanism with drive level that is compensating too.
So I'm guessing that the Ik^(-1/3) term in Rp eff. is the non-linear effect. Do you have a better explanation? I'm all ears.
Don
Yes, I agree, and the NFB IS increasing with greater %UL (for a fixed Zload anyway). So this could be the entire explanation for the flat-topping peak observed. (with no NFB in the pentode case)
But notice in Fig. 5 page 59 of L-S & C that the distortion minima is sensitive to voltage drive level versus %UL. My guess is the voltage sensitivity of the minima, THD dip, position could be ascribed to the changing (lowering) of Mu effective versus %UL (via the NFB). But the fact that there is a minima/THD dip at all versus drive level suggests some non-linear mechanism with drive level that is compensating too.
So I'm guessing that the Ik^(-1/3) term in Rp eff. is the non-linear effect. Do you have a better explanation? I'm all ears.
Don
"dependances of anode current on grid voltages and on anode voltage are non-linear! "
Yes, true. But generally these non-linearies cause increasing distortion with greater drive level. Getting a distortion drop with greater drive level is unusual. By deliberately slightly mis-matching Z at low level, the Ik^(-1/3) term MIGHT be able to compensate by bringing Z back into match at higher levels.
Screen current taking off at high level could be another effect, but it would make output Z higher I think. But maybe it (screen current dist.) could be compensating for the 3/2 power law distortion in current from heavy loading fortuitously.
Well, it will be interesting to see whether fixing the screen current distortion with the MOSFET drain connection improves or worsens UL distortion.
Don
Yes, true. But generally these non-linearies cause increasing distortion with greater drive level. Getting a distortion drop with greater drive level is unusual. By deliberately slightly mis-matching Z at low level, the Ik^(-1/3) term MIGHT be able to compensate by bringing Z back into match at higher levels.
Screen current taking off at high level could be another effect, but it would make output Z higher I think. But maybe it (screen current dist.) could be compensating for the 3/2 power law distortion in current from heavy loading fortuitously.
Well, it will be interesting to see whether fixing the screen current distortion with the MOSFET drain connection improves or worsens UL distortion.
Don
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.