Reading about screen grid drive amps and checking at my tube stash, i found 6F5P which is essentially ECL85. The specsheet shows Ia vs Va characteristic with various G2 voltage, which is quite low (120V-170V). However, G2 requires much more voltage swing to drive, apart from the low impedance driver required to supply the grid current.
I then thought why not drive the G1 as well but at a fixed ratio to the G2 drive swing. Using 6N6P tube (ECC99 essentially), that ratio would be the tube's mu of 20. The idea is plate of the tube connects to G2 while the auto-bias resistor connects to G1. This should reduce the G2 drive swing requirement. You could of course replace the mu-ratio tube divider with resistive divider to set other ratio, AC coupled to G1.
Not sure if this will work.. Below/attached is the concept schematic. Obviously essential stopper resistors are not there yet. Don't take the part values seriously. What do you think? I'm concerned especially about the output tube's rp. Would it still be high? Do i need some kind of "schade" feeback to reduce it?
Preemptively answering the question of "why" = i'm not sure what to say other than "why not". It's just an idea. Maybe there's some hidden advantage that i can't yet see. At least this would increase the typically slim tube count to chassis area ratio on a SE amp 😀.. you can even improve it a la Aikido technique by utilising 6N6P's idle G1 grid.
I then thought why not drive the G1 as well but at a fixed ratio to the G2 drive swing. Using 6N6P tube (ECC99 essentially), that ratio would be the tube's mu of 20. The idea is plate of the tube connects to G2 while the auto-bias resistor connects to G1. This should reduce the G2 drive swing requirement. You could of course replace the mu-ratio tube divider with resistive divider to set other ratio, AC coupled to G1.
Not sure if this will work.. Below/attached is the concept schematic. Obviously essential stopper resistors are not there yet. Don't take the part values seriously. What do you think? I'm concerned especially about the output tube's rp. Would it still be high? Do i need some kind of "schade" feeback to reduce it?
An externally hosted image should be here but it was not working when we last tested it.
Preemptively answering the question of "why" = i'm not sure what to say other than "why not". It's just an idea. Maybe there's some hidden advantage that i can't yet see. At least this would increase the typically slim tube count to chassis area ratio on a SE amp 😀.. you can even improve it a la Aikido technique by utilising 6N6P's idle G1 grid.
Attachments
Last edited:
This idea has been looked at in a number of threads, starting with the DCPP amplifier thread (around half way I thinK). Later, I think, in the 6HB6 thread and then the 6GE5 thread.
DCPP:
http://www.diyaudio.com/forums/tubes-valves/151206-posted-new-p-p-power-amp-design.html
6GE5 thread:
http://www.diyaudio.com/forums/tubes-valves/278043-6ge5-triode.html
6HB6 thread:
http://www.diyaudio.com/forums/tubes-valves/276710-6hb6-finals-12.html#post4391057
Haven't seen using a driver tube in inverted/follower mode to do the Mu amplitude scaling before though. Driving a Mu scaled g1 along with the g2 allows using only half as much drive signal on g2, as g2 drive alone would require. Which can reduce the max positive excursion of g2 if g1 is driven from 0V. But does not help the max + g2 excursion if g1 is driven in its negative bias region.
Here are some curves and test circuit links:
http://www.diyaudio.com/forums/tubes-valves/271134-schade-cfb-exactly-equivalent-10.html#post4279487
http://www.diyaudio.com/forums/tubes-valves/271134-schade-cfb-exactly-equivalent-11.html#post4279626
b) 21HB5A in Mu scaled g2/g1 drive with g1 Schading
c) 21HB5A in g1 Schaded mode
d) 21HB5A in g2 drive only
a) test circuit
DCPP:
http://www.diyaudio.com/forums/tubes-valves/151206-posted-new-p-p-power-amp-design.html
6GE5 thread:
http://www.diyaudio.com/forums/tubes-valves/278043-6ge5-triode.html
6HB6 thread:
http://www.diyaudio.com/forums/tubes-valves/276710-6hb6-finals-12.html#post4391057
Haven't seen using a driver tube in inverted/follower mode to do the Mu amplitude scaling before though. Driving a Mu scaled g1 along with the g2 allows using only half as much drive signal on g2, as g2 drive alone would require. Which can reduce the max positive excursion of g2 if g1 is driven from 0V. But does not help the max + g2 excursion if g1 is driven in its negative bias region.
Here are some curves and test circuit links:
http://www.diyaudio.com/forums/tubes-valves/271134-schade-cfb-exactly-equivalent-10.html#post4279487
http://www.diyaudio.com/forums/tubes-valves/271134-schade-cfb-exactly-equivalent-11.html#post4279626
b) 21HB5A in Mu scaled g2/g1 drive with g1 Schading
c) 21HB5A in g1 Schaded mode
d) 21HB5A in g2 drive only
a) test circuit
Attachments
Last edited:
Here are some tests George (Tubelab) has done on this g2 and g1 drive:
http://www.diyaudio.com/forums/tube...your-screen-drive-circuits-9.html#post3600789
http://www.diyaudio.com/forums/tube...your-screen-drive-circuits-9.html#post3600789
I have several more dual drive circuits working, and some will be posted when I get a chance to catch up on everything since moving twice in the past 18 months. Most of my experiments have been based on sweep tubes, but I have had success with the usual audio tubes like the 6L6GC.
You could of course replace the mu-ratio tube divider with resistive divider to set other ratio, AC coupled to G2.
I didn't AC couple, but I am using fixed bias, so I have a negative supply. G1 is tied to a resistive divider from G2 and the negative voltage supply is chosen to get the bias close the the desired value. I find that most tubes work best in the range of about 5 to 1 screen to control grid voltage.
Advantages, less drive voltage required than pure screen drive, less likely hood of blowing the screen grid due to lower peak drive voltage.
Disadvantage, higher output impedance than pure G1 drive. Some feedback will be needed somewhere.
You could of course replace the mu-ratio tube divider with resistive divider to set other ratio, AC coupled to G2.
I didn't AC couple, but I am using fixed bias, so I have a negative supply. G1 is tied to a resistive divider from G2 and the negative voltage supply is chosen to get the bias close the the desired value. I find that most tubes work best in the range of about 5 to 1 screen to control grid voltage.
Advantages, less drive voltage required than pure screen drive, less likely hood of blowing the screen grid due to lower peak drive voltage.
Disadvantage, higher output impedance than pure G1 drive. Some feedback will be needed somewhere.
Here are some assorted curve sets for the 21HB5A.
a) negative g1 drive only
b) g2 drive + 124V with g1 at -22V
c) g2 drive + 62V with g1 at -11V
d) positive g2 drive only
As you can see, negative g1 gives plate curve kinks. So 0V on g1 would be preferred for g2 drive alone.
For scaled g2 and g1 drive, the kinks will disappear as the plate current increases (g1 increasing toward 0), so negative g1 bias would be OK.
Scaled g2 and g1 drive also works fine, with no discontinuities that I can see, with both g2 and g1 going positive. This would require the least positive excursion of g2, so would be the safest for the tube. The g1 driver does then need to handle some small current, like the g2 driver always does.
Refuses to let me upload the files???
a) negative g1 drive only
b) g2 drive + 124V with g1 at -22V
c) g2 drive + 62V with g1 at -11V
d) positive g2 drive only
As you can see, negative g1 gives plate curve kinks. So 0V on g1 would be preferred for g2 drive alone.
For scaled g2 and g1 drive, the kinks will disappear as the plate current increases (g1 increasing toward 0), so negative g1 bias would be OK.
Scaled g2 and g1 drive also works fine, with no discontinuities that I can see, with both g2 and g1 going positive. This would require the least positive excursion of g2, so would be the safest for the tube. The g1 driver does then need to handle some small current, like the g2 driver always does.
Refuses to let me upload the files???
Last edited:
There have been persistent forum problems with file uploading this week. I could not upload a picture on Wednesday, Thursday it worked. Other members have had problems too.
http://www.diyaudio.com/forums/forum-problems/283433-cannot-upload-images-swap-meet.html
http://www.diyaudio.com/forums/forum-problems/283433-cannot-upload-images-swap-meet.html
Tried upload again.... and WORKED!!!
50V/div Horiz. 20mA/div. Vert. 21HB5A
a) negative g1 drive only
b) g2 drive + 124V with g1 at -22V
c) g2 drive + 62V with g1 at -11V
d) positive g2 drive only (g1 at 0V) (50 mA/div Horiz.)
I have now noticed some issues with using positive g1 along with positive g2.
Pos. g1 will start to draw noticeable grid current if g2 is not sufficiently higher, and puts a slanted flat into the plate curve then (at low plate V's). This probably is still not be an issue with scaled g2 and g1 as long as g2 is scaled well above g1.
The other effect I am seeing is positive g1 causes the diode curve (left side of plate curve knees) to increase (shift right, in plate voltage) some. +10V on g1 (for the 21HB5A) can make the higher current plate knees increase by +25V. This is NOT GOOD for efficiency. For a Mu 4.8 scaled tube (ie, 21HB5A), this could reach +20V on g1 and +50V delta on the plate curves.
50V/div Horiz. 20mA/div. Vert. 21HB5A
a) negative g1 drive only
b) g2 drive + 124V with g1 at -22V
c) g2 drive + 62V with g1 at -11V
d) positive g2 drive only (g1 at 0V) (50 mA/div Horiz.)
I have now noticed some issues with using positive g1 along with positive g2.
Pos. g1 will start to draw noticeable grid current if g2 is not sufficiently higher, and puts a slanted flat into the plate curve then (at low plate V's). This probably is still not be an issue with scaled g2 and g1 as long as g2 is scaled well above g1.
The other effect I am seeing is positive g1 causes the diode curve (left side of plate curve knees) to increase (shift right, in plate voltage) some. +10V on g1 (for the 21HB5A) can make the higher current plate knees increase by +25V. This is NOT GOOD for efficiency. For a Mu 4.8 scaled tube (ie, 21HB5A), this could reach +20V on g1 and +50V delta on the plate curves.
Attachments
Last edited:
minor correction to above pics:
plots a) and d) are at 50 mA/div VERT. (were some older pics on hand)
plots b) and c) were at 20 mA/div VERT. (can tell by the different slope on right side for limiting)
I have also noticed that positive g1 not only makes the diode line shift right (to higher plate V), but it also makes the g2 currents increase quite noticeably. I think what happens here is that + g1 focuses cathode current ONTO g2 (while normally NEG. g1 SHIELDS g2 from cathode current). This is NOT good!!
So I would say the optimum setup for Mu scaled g2 g1 drives would bias g1 at half the usual neg. bias (compared to g1 drive only). Then the scaled g2 g1 drives would bring g1 up to 0V and g2 up to +Vg2max (approx. Mu times g1 Vbias) for maximum + signal. (Each grid providing equal effective current output drive with the Mu scaled drives.) Drive(s) could still exceed that on signal overdrive, but efficiency would start to decrease some beyond there. The kinks in the plate curves will be full size near zero plate current (g1 at -Vbias), but shrink to nothing at max current (g1 at 0 V). So the load line is never bothered by them. At least overdrive will only be pushing Vg2 up further at half the rate (in g2 g1 scaled mode) of a pure g2 drive setup.
Schade feedback shouldn't really effect the DC setup points (for AC coupled Fdbk anyway). But it will effect the R (or tube) dividers to get the proper g2 g1 drive scaling. Schade to g1 will be Mu times more effective than to g2 for the same feedback, as far as reducing output Z and distortion. But for two similarly (Mu) scaled Schade feedbacks, they should both contribute the same I think.
plots a) and d) are at 50 mA/div VERT. (were some older pics on hand)
plots b) and c) were at 20 mA/div VERT. (can tell by the different slope on right side for limiting)
I have also noticed that positive g1 not only makes the diode line shift right (to higher plate V), but it also makes the g2 currents increase quite noticeably. I think what happens here is that + g1 focuses cathode current ONTO g2 (while normally NEG. g1 SHIELDS g2 from cathode current). This is NOT good!!
So I would say the optimum setup for Mu scaled g2 g1 drives would bias g1 at half the usual neg. bias (compared to g1 drive only). Then the scaled g2 g1 drives would bring g1 up to 0V and g2 up to +Vg2max (approx. Mu times g1 Vbias) for maximum + signal. (Each grid providing equal effective current output drive with the Mu scaled drives.) Drive(s) could still exceed that on signal overdrive, but efficiency would start to decrease some beyond there. The kinks in the plate curves will be full size near zero plate current (g1 at -Vbias), but shrink to nothing at max current (g1 at 0 V). So the load line is never bothered by them. At least overdrive will only be pushing Vg2 up further at half the rate (in g2 g1 scaled mode) of a pure g2 drive setup.
Schade feedback shouldn't really effect the DC setup points (for AC coupled Fdbk anyway). But it will effect the R (or tube) dividers to get the proper g2 g1 drive scaling. Schade to g1 will be Mu times more effective than to g2 for the same feedback, as far as reducing output Z and distortion. But for two similarly (Mu) scaled Schade feedbacks, they should both contribute the same I think.
Last edited:
Some further thoughts on g2 g1 combined drives.
As you can see from the earlier plate curves, g2 drive gives the best raw linearity (equal curve spacing versus drive voltage steps). However, when Schade (or any) local feedback is considered, the g1 drive gives more loop gain to linearize better.
So the net result is probably a near tie between them.
Combined g2 g1 drives, with Schading, gets complex (the Schade Fdbk affecting the R dividers, and drive current is needed for the g2 drive at least). The result of g2 g1 combined drives gives raw linearity 1/2 way between g2 or g1 only drive setups (and gain 1/2 way between). With Schade local Fdbk set up though, the net result will be near the same as g1 or g2 only drive, with Schade Fdbk.
So combined g2 and g1 drives is interesting...., but complex...., for nearly the same net results. So I think I will be staying with the easiest g1 drive for now. Now a pre-designed and debugged PC board design for g2 g1 drive (a'la Tubelab) could even out that playing field.
As you can see from the earlier plate curves, g2 drive gives the best raw linearity (equal curve spacing versus drive voltage steps). However, when Schade (or any) local feedback is considered, the g1 drive gives more loop gain to linearize better.
So the net result is probably a near tie between them.
Combined g2 g1 drives, with Schading, gets complex (the Schade Fdbk affecting the R dividers, and drive current is needed for the g2 drive at least). The result of g2 g1 combined drives gives raw linearity 1/2 way between g2 or g1 only drive setups (and gain 1/2 way between). With Schade local Fdbk set up though, the net result will be near the same as g1 or g2 only drive, with Schade Fdbk.
So combined g2 and g1 drives is interesting...., but complex...., for nearly the same net results. So I think I will be staying with the easiest g1 drive for now. Now a pre-designed and debugged PC board design for g2 g1 drive (a'la Tubelab) could even out that playing field.
Last edited:
Lately, I have been more interested in a scheme similar to that shown below. Essentially a P-P cascode stage, with a low voltage, moderate current (Mosfet) class A drive stage below, and a tube cascode stage above. Then Gyrators or CCS's remove the class A portion of the DC currents in between, so the top tube stage operates in class aB mode for efficiency.
This eliminates crossover distortion and gm variation, while providing the efficiency of class aB. In other words, it's class A linear without the HEAT. No need to run all that DC idle current through the tubes, when it just cancels out in the OT anyway.
This eliminates crossover distortion and gm variation, while providing the efficiency of class aB. In other words, it's class A linear without the HEAT. No need to run all that DC idle current through the tubes, when it just cancels out in the OT anyway.
Last edited:
Reading your valuable experiment explanations, i do have to agree with you. Since we still need to introduce some kind of local feedback to reduce rp, there isn't any significant advantage to driving g2 or dual drive other than novelty. Unnecessary complexity. I'll put this G2 experiment on the back of the line.
Still working on it, but also trying to rebuild my lab, and my life at the same time. Progress stalls a lot, sometimes for weeks at a time. Last week I flipped on power to my bench (still using extension cords) only to see a wisp of smoke come from my HP8903A audio analyzer.
Two blown fuses inside on the power supply board, replacing them allows the instrument to power up, but it is brain dead except for a few random flashes on the readouts. Yes, it is a 35 year old machine, but reliability has been terrible. This will be the fifth time I have had to fix it, and there were obvious repairs before me. It came from the Motorola Paging plant before it was bulldozed and made into a condo / shopping complex. The plant where I worked is becoming a medical park.
For a mild mannered HiFi amp, maybe not. Pete's big red board can get you 125 WPC in G1 drive mode without going far into AB2.
Pure G2 drive will get you better plate efficiency, and lower idle current for the same or better distortion level as an equivalent G1 driven amp. This allows more output power for a given set of output tubes, or cooler operation at a given power level. A well designed G1 driven P-P amp may have a plate efficiency in the 60% range. I have seen over 70% on a G2 driven amp. The plate saturation voltage, at least on sweep tubes, is lower when G2 is driven.
There is however a rather nasty failure mode on a G2 driven amp that would probably never occur in a HiFi amp even if you cranked some disco music well into the clipping region. I however have managed to create some rather spectacular vacuum tube fireworks complete with exploding mosfets by playing my guitar into a screen driven amp that was squeezing 120 watts from a pair of tiny 6BQ6GT's.
I want all of my amps to be capable of this kind of abuse, because someone somewhere WILL try it. It's usually me (like who else plays their guitar through a 300B amp?), but there are other postings on the Tubelab forum where a select few have taken my boards where even I haven't been. So I have looked at other drive circuits to extract maximum efficiency without blowing up any parts. G1 - G2 "dual drive", and cathode drive both show potential, but I'm not there yet.
Note, Yeah, I know about these, but they are new comers. I have been playing my guitar through my TSE since I made it. I test all my amps with a guitar, and liked the way it sounded. I did find, and fix, a failure mechanism in the SSE in this manner before the boards ever went live.
https://www.eurotubes.com/store/pc/the jj one guitar amp.htm
Siegmund Diamond | 300B tube guitar amplifier handmade
Well, I was going to mention that the cathode driven, Class A to Class aB conversion scheme (above in post 11) could be converted to P-P screen drives (which it can), but I discovered last night that it still suffers from gm doubling when there is a little "a" left in the Class "aB". Some kind of servo is required for the Gyrators to get exactly half the class A current nulled.
But the idea can still be used for a screen drive scheme. I would call this "bipolar" screen drive. Essentially, the screen grids pick up (or intercept) a near constant fraction of the cathode current (like 10%) until the plate voltage drops down near or below the g2 voltage. So if one drives the screen grid with a current drive, it provides a current gain "Beta" like a bipolar transistor, which droops off at high current (low plate V actually), just like a transistor.
The current drives would allow a Class A driver current output stage (pentodes) to drive the screen grids (using CCS loads, or constant Beta PNP bipolars, as "mirrors").
And Gyrators or CCS level could then again null out half the Class A idle currents before reaching the screen grids, to get class aB. (But that must be EXACTLY 1/2, or perfect class B, unfortunately to avoid gm doubling). More work is needed on this Class conversion concept in either form.
Conventional P-P voltage drive (either g1 or g2 drive) does not suffer, typically, from the gm doubling problem (with conduction overlap), since the gm's ramp up from zero in that case. (while current drive forces instantaneous full programmed current out) One is faced with summing two opposing gm step functions in the current drives case, while the voltage drives case requires summing two opposing ramps. Which works out much better for class aB, since the ramps sum to a constant gm in the overlap region. Once again, the old tried and tested approach of V drive works best. (for either g1 or g2 drive)
But the idea can still be used for a screen drive scheme. I would call this "bipolar" screen drive. Essentially, the screen grids pick up (or intercept) a near constant fraction of the cathode current (like 10%) until the plate voltage drops down near or below the g2 voltage. So if one drives the screen grid with a current drive, it provides a current gain "Beta" like a bipolar transistor, which droops off at high current (low plate V actually), just like a transistor.
The current drives would allow a Class A driver current output stage (pentodes) to drive the screen grids (using CCS loads, or constant Beta PNP bipolars, as "mirrors").
And Gyrators or CCS level could then again null out half the Class A idle currents before reaching the screen grids, to get class aB. (But that must be EXACTLY 1/2, or perfect class B, unfortunately to avoid gm doubling). More work is needed on this Class conversion concept in either form.
Conventional P-P voltage drive (either g1 or g2 drive) does not suffer, typically, from the gm doubling problem (with conduction overlap), since the gm's ramp up from zero in that case. (while current drive forces instantaneous full programmed current out) One is faced with summing two opposing gm step functions in the current drives case, while the voltage drives case requires summing two opposing ramps. Which works out much better for class aB, since the ramps sum to a constant gm in the overlap region. Once again, the old tried and tested approach of V drive works best. (for either g1 or g2 drive)
Last edited:
Well, on SECOND thought, after further analyzing the Class A to Class aB conversion scheme above, I think it CAN be made to work (either the cascode or screen drive versions).
The trick is to set the Gyrators (which suck off the class A DC currents) so they have an idle voltage just below turn-on of the output power tubes. The Gyrators will automatically adjust their current to consume the average current sent by the Class A driver stage devices (approx. 50% each side from a CCS tail under the class A driver).
As soon as the driver stage upsets the idle condition (ie, signal input), the excess current (and complementary side decreased current) will instantly (limited by driven output tube input capacitance and driver idle current, so like uSecs) turn on one output device (and turn further off the complementary device). There is NO overlapped conduction to cause gm doubling, as long as the Gyrator V set points are just below output stage conduction thresholds. So there is no "a" in the class "aB" after all this way. It just makes PERFECT class B!
This needs to be simulated of course to verify this. But I think I am onto something NEW UNDER the SUN after all. This may be a really interesting scheme after all. It will work for the P-P screen driven arrangement too, since that can be current driven (with Beta gain no less, like 10 or 20), similar to the cascode arrangement (except Beta = 1 for cascode).
So the screen driven version does not need SS Mosfets for the class A driver, due to the screen current Beta gain feature. A plain pentode differential stage with a CCS tail can perform driver duty then. So an all tube version is possible this way. I would probably just use a SS CCS tail for the differential driver stage and for the two un-loading Gyrators.
Happy Day!!!
The Gyrator set points could be set automatically by a servo loop that monitors the minimum output stage OT center tap current (which would be just zero or microAmps). That should bring turn-on times down to sub uSecs.
The trick is to set the Gyrators (which suck off the class A DC currents) so they have an idle voltage just below turn-on of the output power tubes. The Gyrators will automatically adjust their current to consume the average current sent by the Class A driver stage devices (approx. 50% each side from a CCS tail under the class A driver).
As soon as the driver stage upsets the idle condition (ie, signal input), the excess current (and complementary side decreased current) will instantly (limited by driven output tube input capacitance and driver idle current, so like uSecs) turn on one output device (and turn further off the complementary device). There is NO overlapped conduction to cause gm doubling, as long as the Gyrator V set points are just below output stage conduction thresholds. So there is no "a" in the class "aB" after all this way. It just makes PERFECT class B!
This needs to be simulated of course to verify this. But I think I am onto something NEW UNDER the SUN after all. This may be a really interesting scheme after all. It will work for the P-P screen driven arrangement too, since that can be current driven (with Beta gain no less, like 10 or 20), similar to the cascode arrangement (except Beta = 1 for cascode).
So the screen driven version does not need SS Mosfets for the class A driver, due to the screen current Beta gain feature. A plain pentode differential stage with a CCS tail can perform driver duty then. So an all tube version is possible this way. I would probably just use a SS CCS tail for the differential driver stage and for the two un-loading Gyrators.
Happy Day!!!
The Gyrator set points could be set automatically by a servo loop that monitors the minimum output stage OT center tap current (which would be just zero or microAmps). That should bring turn-on times down to sub uSecs.
Last edited:
Here is a rough circuit for the Current Driven Screens - Perfect Class B (for the output tubes).
Fairly simple to implement. DC coupled too.
The driver tube screens are used for local voltage feedback (from UL taps) to lower the final tube output Z and minimize voltage distortion. The driver plate and screen grid voltages are in phase for each driver tube, and could be made to proportionally track by adjusting the R2, R1 ratio. That would keep the driver screen grid current linear with voltage, so as to not disturb the Fdbk attenuation. 12HL7 tubes for the driver stage have remarkably linear-constant g2/g1 Mu versus current, so would work well here for low distortion (any residual dist. would be class A triode P-P signature). A 10 Watt or so driver tube like 12HL7, 12GN7, 6HB6, 6HZ8, 6JZ8, 6GF5 .... can easily handle the screen drive currents directly.
Output tubes could be most any TV Sweep tube with low g2/g1 internal Mu. (ie, suitable for screen grid drive)
The Gyrators get adjusted for 0 Volt driver output at idle, and each absorbs 1/2 of the bottom CCS tail current at idle. Any input signal unbalances the driver currents, leading to one side screen driving the output tube into proportional conduction (x Beta current gain). The driver stage needs to be fairly linear so as to not shift the AVERAGE plate voltage output at large signals. (ie, no 2nd H, or 3rd H in differential, bias shift) That would mess up the Gyrator set points if the average voltage shifted. Matched drivers and matched output tubes likely desirable.
Some droop in the screen current gain, Beta, at high current will tend to smoothly limit clipping, however the local UL voltage feedbacks will keep things linear over most of the range.
Since the voltage feedbacks to the driver stage are B+ referenced off the OT, the B+ should be regulated.
Some global N Fdbk could be added (with an input gain/splitter stage added) to compensate for OT resistance and leakage L if wanted.
Fairly simple to implement. DC coupled too.
The driver tube screens are used for local voltage feedback (from UL taps) to lower the final tube output Z and minimize voltage distortion. The driver plate and screen grid voltages are in phase for each driver tube, and could be made to proportionally track by adjusting the R2, R1 ratio. That would keep the driver screen grid current linear with voltage, so as to not disturb the Fdbk attenuation. 12HL7 tubes for the driver stage have remarkably linear-constant g2/g1 Mu versus current, so would work well here for low distortion (any residual dist. would be class A triode P-P signature). A 10 Watt or so driver tube like 12HL7, 12GN7, 6HB6, 6HZ8, 6JZ8, 6GF5 .... can easily handle the screen drive currents directly.
Output tubes could be most any TV Sweep tube with low g2/g1 internal Mu. (ie, suitable for screen grid drive)
The Gyrators get adjusted for 0 Volt driver output at idle, and each absorbs 1/2 of the bottom CCS tail current at idle. Any input signal unbalances the driver currents, leading to one side screen driving the output tube into proportional conduction (x Beta current gain). The driver stage needs to be fairly linear so as to not shift the AVERAGE plate voltage output at large signals. (ie, no 2nd H, or 3rd H in differential, bias shift) That would mess up the Gyrator set points if the average voltage shifted. Matched drivers and matched output tubes likely desirable.
Some droop in the screen current gain, Beta, at high current will tend to smoothly limit clipping, however the local UL voltage feedbacks will keep things linear over most of the range.
Since the voltage feedbacks to the driver stage are B+ referenced off the OT, the B+ should be regulated.
Some global N Fdbk could be added (with an input gain/splitter stage added) to compensate for OT resistance and leakage L if wanted.
Last edited:
One problem found (with the perfect class B idea).
The differential driver stage outputs must swing identical amounts for + and - signal, so that the Gyrator's steady current will not drift versus signal amplitude. Ie, the average driver output voltage must stay constant versus signal amplitude. The loading on the driver outputs must be symmetrical to do that. But the actual output tube(s) screen grid only loads each driver for one polarity.
The earlier cascode version above included thermionic catch diodes for negative driver swings to mimic a symmetrical load. A cascode input pretty much looks like a thermionic diode due to the fixed grid voltage. Easy to emulate.
However, a screen grid input has a very different characteristic, determined by how much cathode current it intercepts, which depends on the plate voltage (and hence the load even). Not going to be easy to emulate that.
So for the screen drive version, one may want to give up on the self adjusting Gyrator approach, and just use a manually tuned, fixed, CCS up top for each driver pull-up. No CCS drift then with signal amplitude. But manual tune-ups may be needed to compensate for various tube aging effects. Guess that's nothing new. But not really welcome. The CCS is a little easier circuit to construct than the Gyrator. Will need some resistance across the CCS (Rstabilize) to stabilise it, since the input Z of the screen grid is only finite while conducting, and one is balancing the top two CCS against the bottom CCS, when the screen grid is not conducting (0 V). (Ie, adjust top CCS current set pot to hold the driver output at 0 V) Or else use a Gyrator with an hour long time constant.
New diagram attached:
The differential driver stage outputs must swing identical amounts for + and - signal, so that the Gyrator's steady current will not drift versus signal amplitude. Ie, the average driver output voltage must stay constant versus signal amplitude. The loading on the driver outputs must be symmetrical to do that. But the actual output tube(s) screen grid only loads each driver for one polarity.
The earlier cascode version above included thermionic catch diodes for negative driver swings to mimic a symmetrical load. A cascode input pretty much looks like a thermionic diode due to the fixed grid voltage. Easy to emulate.
However, a screen grid input has a very different characteristic, determined by how much cathode current it intercepts, which depends on the plate voltage (and hence the load even). Not going to be easy to emulate that.
So for the screen drive version, one may want to give up on the self adjusting Gyrator approach, and just use a manually tuned, fixed, CCS up top for each driver pull-up. No CCS drift then with signal amplitude. But manual tune-ups may be needed to compensate for various tube aging effects. Guess that's nothing new. But not really welcome. The CCS is a little easier circuit to construct than the Gyrator. Will need some resistance across the CCS (Rstabilize) to stabilise it, since the input Z of the screen grid is only finite while conducting, and one is balancing the top two CCS against the bottom CCS, when the screen grid is not conducting (0 V). (Ie, adjust top CCS current set pot to hold the driver output at 0 V) Or else use a Gyrator with an hour long time constant.
New diagram attached:
Last edited:
Next Iteration of the Current Driven Screen Drive - Perfect Class B
Using OT bootstraps now, instead of the top CCS or Gyrators, for the driver pull-ups. The bootstrap voltages (standard UL taps if possible) would be chosen to provide the nominal voltage swing that would normally drive the screen grids with a nominal load. Then the Rboot resistors act like current sources, with constant voltage across them. Rboot value is chosen to drop B+ to 0V with the bottom CCS current/2 through it, setting the turn-on threshold at idle to 0 V. The CCS tail is adjustable to fine tune the idle Vscreens.
A Ferrite core balancing inductor is added in series with the bootstraps, to allow isolation for the non-linearity of the screen voltage drives versus the output voltages. So that would be a fairly low turns dual winding xfmr, just to handle screen non-linearity. Both the Rboot and the Ferrite balancer compliance together allow for variation of the final load R (changing screen drive V level versus output V level). So some extra turns on the balancer xfmr could be useful to isolate load variation from affecting the constant current pull-ups, but the Rboot could handle that alone with some AC current variation then (distortion influence). The Ferrite core balancer is chosen for HF response, since the driver stage is acting as a current source, and will require fast voltage change to satisfy that. It also allows both sides of the driver to contribute some current drive to the active screen.
Looking fairly practical now. The driver tubes do have to handle full + to - max screen voltage swing (ie, 2x the max + screen V), which is pushing the voltage specs for the 12HL7 or 12GN7 drivers. But they're cheap. I can burn a few up before resorting to a 6GF5 or something similar. The B+ will definitely have to be regulated now since it affects both the screen turn on thresholds as well as the local V loop feedbacks (to the driver screens).
Using OT bootstraps now, instead of the top CCS or Gyrators, for the driver pull-ups. The bootstrap voltages (standard UL taps if possible) would be chosen to provide the nominal voltage swing that would normally drive the screen grids with a nominal load. Then the Rboot resistors act like current sources, with constant voltage across them. Rboot value is chosen to drop B+ to 0V with the bottom CCS current/2 through it, setting the turn-on threshold at idle to 0 V. The CCS tail is adjustable to fine tune the idle Vscreens.
A Ferrite core balancing inductor is added in series with the bootstraps, to allow isolation for the non-linearity of the screen voltage drives versus the output voltages. So that would be a fairly low turns dual winding xfmr, just to handle screen non-linearity. Both the Rboot and the Ferrite balancer compliance together allow for variation of the final load R (changing screen drive V level versus output V level). So some extra turns on the balancer xfmr could be useful to isolate load variation from affecting the constant current pull-ups, but the Rboot could handle that alone with some AC current variation then (distortion influence). The Ferrite core balancer is chosen for HF response, since the driver stage is acting as a current source, and will require fast voltage change to satisfy that. It also allows both sides of the driver to contribute some current drive to the active screen.
Looking fairly practical now. The driver tubes do have to handle full + to - max screen voltage swing (ie, 2x the max + screen V), which is pushing the voltage specs for the 12HL7 or 12GN7 drivers. But they're cheap. I can burn a few up before resorting to a 6GF5 or something similar. The B+ will definitely have to be regulated now since it affects both the screen turn on thresholds as well as the local V loop feedbacks (to the driver screens).
Last edited:
Screen Drive with Power Steering
I was thinking over what happens (in the above schematic) when the voltage pickoff taps (UL taps shown) have a different voltage from what would give constant current through the Rboot resistors. Any difference from the ideal voltage tap value would just produce an extra voltage across the Ferrite xfmr up until its voltage range is exceeded. If the Ferrite xfmr were removed from the path however, it would cause variation in the "constant" current loading of the drivers.
So, removing the Ferrite xfmr and varying the pickoff taps from current neutrality, allows one to provide feedback current that is either assisting the driver, or bucking the driver at the output tube screen grids. The bucking arrangement would just be the usual Schade local N Fdbk. The assisting range however would provide positive feedback. Now the screen grids have near constant Beta, (current gain) in the small to moderate signal range, so they require a drive current that is proportional to output current. If one were to select pickoff taps that provided the nominal screen drive currents to just match the nominal load requirements, then the driver circuit would only need to supply miniscule "steering" currents to control the output.
Essentially the positive feedbacks would then be achieving "unity" feedback, giving the inner "inverted Schade" loop infinite gain in that case. The drivers still can control all this easily, since they have more "real" gain enclosed in the negative feedback loop, due to the driver gains. Essentially a classical positive feedback loop enclosed within a negative feedback loop. Unity P Fdbk producing infinite gain within the N. Fdbk loop.
Normally one cannot push the positive feedback loop to infinite gain at low signal level, since a grid V drive gm's increase with current would turn all into an oscillator (practically speaking; theoretically it's possible to exceed infinte gain with sufficient neg. Fdbk around it, like in Hawksford Error Correction). But anyway, the Beta factor in current drive -decreases- with higher output, so everything becomes more stable. No run-away condition here.
The main issue would be final load variation, a heavier speaker loading (like 2 Ohms) would decrease the P. Fdbks, so effectively increasing the output Z since more driver current would be needed. A non-existant load would increase the P. Fdbks beyond unity, so possibly turning all into an oscillator. Sufficient drive capability from the drivers to over-ride any feedback variation would guarantee stability still, so I would keep the 10 Watt driver tubes. The high gm frame grid drivers will also produce output Z limited only by the OT winding resistance (or an outer global N Fdbk loop), regardless of speaker loading.
With infinite gain available to the N. Feedback loop under small to moderate signal levels, distortion should be reduced to vanishingly low. And the parts construction complexity is reduced without the Ferrite feedback isolation xfmr.
(besides no need for a screen grid drive follower)
This positive feedback bootstrapping arrangement for the driver loads can also be used with any of the more typical Voltage drive screen drive setups, as long as the drivers can contain any P Feedback runaway. (Available driver currents sufficient to overcome any excessive P Fdbk.)
An up-dated schematic is attached. I've shown the Rboot and V Neg. Fdbk taps moved to the plate ends of the OT, since Rboot now will nominally require higher AC voltage swing. These OT end points may not give the required voltage swing for unity P. Fdbk, likely too much. A practical implementation might use TWO Rboot resistors for each side, each roughly twice the R value, one going to the UL tap and the other to the OT plate end, then vary the ratio between them to get the proper Bootstrap V (screen drive current ultimately).
The V Neg. Fdbk resistors (R2) might be better connected to the UL taps for closer coupling to the secondary winding (on most OTs). Their R2/R1 V attenuators can be adjusted for most any tapping.
I think the design has become simple enough now that a physical construction would be useful to check this out. This is really simpler than most conventional screen drive setups even, since no current boosting screen drive followers are required.
With the idle screen voltages adjusted to 0V (trimmer on the bottom tail CCS), and the infinite N. Fdbk gain available for small to moderate signals (one could use less than "infinite" gain by reducing the bootstrapping) this should have distortion well below what even a full class A P-P amp could deliver. And it runs with class B efficiency no less!
So you will soon be able to throw out all those obsolete classical amps out there.
(I'll provide an address for disposal) Hehe.....
I was thinking over what happens (in the above schematic) when the voltage pickoff taps (UL taps shown) have a different voltage from what would give constant current through the Rboot resistors. Any difference from the ideal voltage tap value would just produce an extra voltage across the Ferrite xfmr up until its voltage range is exceeded. If the Ferrite xfmr were removed from the path however, it would cause variation in the "constant" current loading of the drivers.
So, removing the Ferrite xfmr and varying the pickoff taps from current neutrality, allows one to provide feedback current that is either assisting the driver, or bucking the driver at the output tube screen grids. The bucking arrangement would just be the usual Schade local N Fdbk. The assisting range however would provide positive feedback. Now the screen grids have near constant Beta, (current gain) in the small to moderate signal range, so they require a drive current that is proportional to output current. If one were to select pickoff taps that provided the nominal screen drive currents to just match the nominal load requirements, then the driver circuit would only need to supply miniscule "steering" currents to control the output.
Essentially the positive feedbacks would then be achieving "unity" feedback, giving the inner "inverted Schade" loop infinite gain in that case. The drivers still can control all this easily, since they have more "real" gain enclosed in the negative feedback loop, due to the driver gains. Essentially a classical positive feedback loop enclosed within a negative feedback loop. Unity P Fdbk producing infinite gain within the N. Fdbk loop.
Normally one cannot push the positive feedback loop to infinite gain at low signal level, since a grid V drive gm's increase with current would turn all into an oscillator (practically speaking; theoretically it's possible to exceed infinte gain with sufficient neg. Fdbk around it, like in Hawksford Error Correction). But anyway, the Beta factor in current drive -decreases- with higher output, so everything becomes more stable. No run-away condition here.
The main issue would be final load variation, a heavier speaker loading (like 2 Ohms) would decrease the P. Fdbks, so effectively increasing the output Z since more driver current would be needed. A non-existant load would increase the P. Fdbks beyond unity, so possibly turning all into an oscillator. Sufficient drive capability from the drivers to over-ride any feedback variation would guarantee stability still, so I would keep the 10 Watt driver tubes. The high gm frame grid drivers will also produce output Z limited only by the OT winding resistance (or an outer global N Fdbk loop), regardless of speaker loading.
With infinite gain available to the N. Feedback loop under small to moderate signal levels, distortion should be reduced to vanishingly low. And the parts construction complexity is reduced without the Ferrite feedback isolation xfmr.
(besides no need for a screen grid drive follower)
This positive feedback bootstrapping arrangement for the driver loads can also be used with any of the more typical Voltage drive screen drive setups, as long as the drivers can contain any P Feedback runaway. (Available driver currents sufficient to overcome any excessive P Fdbk.)
An up-dated schematic is attached. I've shown the Rboot and V Neg. Fdbk taps moved to the plate ends of the OT, since Rboot now will nominally require higher AC voltage swing. These OT end points may not give the required voltage swing for unity P. Fdbk, likely too much. A practical implementation might use TWO Rboot resistors for each side, each roughly twice the R value, one going to the UL tap and the other to the OT plate end, then vary the ratio between them to get the proper Bootstrap V (screen drive current ultimately).
The V Neg. Fdbk resistors (R2) might be better connected to the UL taps for closer coupling to the secondary winding (on most OTs). Their R2/R1 V attenuators can be adjusted for most any tapping.
I think the design has become simple enough now that a physical construction would be useful to check this out. This is really simpler than most conventional screen drive setups even, since no current boosting screen drive followers are required.
With the idle screen voltages adjusted to 0V (trimmer on the bottom tail CCS), and the infinite N. Fdbk gain available for small to moderate signals (one could use less than "infinite" gain by reducing the bootstrapping) this should have distortion well below what even a full class A P-P amp could deliver. And it runs with class B efficiency no less!
So you will soon be able to throw out all those obsolete classical amps out there.
(I'll provide an address for disposal) Hehe.....
Last edited:
- Status
- Not open for further replies.