Ideas here and there on this forum made me wonder if anybody (other than myself) has used this topology in their own designs and with what results.
If I may be allowed a brief introduction: Since the distributed load output topology (lay name ultra linear or simply UL) came along about 1951, a number of investigations have shown that for the little extra complexity of providing 2 taps on an output transformer, significant advantages were yielded. Very briefly (since most know this), it gives about 80% of the advantages of pentodes and triodes combined, with very few of their disadvantages. Still almost pentode efficiency and output capability but with almost triode distortion and internal impedance (rp), it just about precludes use of either of those previous topologies from then on.
About 1948 P J Walker of the UK came with a further advantage in the Quad II, resulting from placing that part of the output transformer between G2 and B+ (thus carrying cathode current) on the cathode side of the configuration. This resulted in still further improvement of damping, independance of load characteristics and lowering of distortion. Bogen of the US were quick to adopt the same topology.
The penalty was that a higher G1 signal was required. E.g., in order to retain the UL characteristics a lower limit tap of 25% has to be maintained for 6L6s. Using the Quad modification, to fully drive a pair of 6L6s for some 60W output, now required about 360Vpp clean grid signal. This is no mean feat at low distortion. Walker used pentode drivers (EF86s), but he eventually settled for only a 10% equavalent screen "tap", thus easing his driver requirements at the expense of some increased output stage distortion.
For maximum output fixed bias conditions are required, where the first limit imposed is the relatively low max. value of 100K for Rg1. One way out would be a cathode follower driver, however with the possibility of added non-linearity because of the large signal amplitude required. A number of designers use this, also MOSFET drivers. I myself have tried a servo control feeding the bottom end of Rg1. The results were a promising reduction of some 5V of G1 drift down to about 25mV, with Rg1 as high as 470K for 2 tubes in parallel per pp side. (Some purists want no semiconductors in the signal path of a tube amplifier.)
Has anybody out there used a similar topology, and with what experience?
Regards
If I may be allowed a brief introduction: Since the distributed load output topology (lay name ultra linear or simply UL) came along about 1951, a number of investigations have shown that for the little extra complexity of providing 2 taps on an output transformer, significant advantages were yielded. Very briefly (since most know this), it gives about 80% of the advantages of pentodes and triodes combined, with very few of their disadvantages. Still almost pentode efficiency and output capability but with almost triode distortion and internal impedance (rp), it just about precludes use of either of those previous topologies from then on.
About 1948 P J Walker of the UK came with a further advantage in the Quad II, resulting from placing that part of the output transformer between G2 and B+ (thus carrying cathode current) on the cathode side of the configuration. This resulted in still further improvement of damping, independance of load characteristics and lowering of distortion. Bogen of the US were quick to adopt the same topology.
The penalty was that a higher G1 signal was required. E.g., in order to retain the UL characteristics a lower limit tap of 25% has to be maintained for 6L6s. Using the Quad modification, to fully drive a pair of 6L6s for some 60W output, now required about 360Vpp clean grid signal. This is no mean feat at low distortion. Walker used pentode drivers (EF86s), but he eventually settled for only a 10% equavalent screen "tap", thus easing his driver requirements at the expense of some increased output stage distortion.
For maximum output fixed bias conditions are required, where the first limit imposed is the relatively low max. value of 100K for Rg1. One way out would be a cathode follower driver, however with the possibility of added non-linearity because of the large signal amplitude required. A number of designers use this, also MOSFET drivers. I myself have tried a servo control feeding the bottom end of Rg1. The results were a promising reduction of some 5V of G1 drift down to about 25mV, with Rg1 as high as 470K for 2 tubes in parallel per pp side. (Some purists want no semiconductors in the signal path of a tube amplifier.)
Has anybody out there used a similar topology, and with what experience?
Regards
The QUAD is a cool output circuit.... i have wound these outputs and they are optimized very well....
I like the MacIntosh output stage.....they put equal share of load in plate and cathode, which makes for unity coupling and large drive signals at G1....
They used a cathode follower to drive G1....
The interesting thing about the follower is that the later revisions of the Mac circuit used "bootstraping" to linearize the follower...
For example early model uses 100K for cathode resistor...
Later model splits this in half...56K and 56K then connect the cathode winding to middle of the 56K resistors with a HUGE cap....
This effectively makes the top 56K resistor have a huge AC impedance and looks like a current source...... I will add this to older Mac circuits.....It does make a audible difference for the better..... You can apply this technique in other applications..
Chris
I like the MacIntosh output stage.....they put equal share of load in plate and cathode, which makes for unity coupling and large drive signals at G1....
They used a cathode follower to drive G1....
The interesting thing about the follower is that the later revisions of the Mac circuit used "bootstraping" to linearize the follower...
For example early model uses 100K for cathode resistor...
Later model splits this in half...56K and 56K then connect the cathode winding to middle of the 56K resistors with a HUGE cap....
This effectively makes the top 56K resistor have a huge AC impedance and looks like a current source...... I will add this to older Mac circuits.....It does make a audible difference for the better..... You can apply this technique in other applications..
Chris
Hi Chris!!
I recall your previous contributions in this respect - was it almost a year ago? I actually wondered whether you will contribute again. I must go now, and will quickly look up my version of the Mac, but the one I have still uses an interstage transformer, so no problem there with getting sufficient Vg1 (other than that of having an extra transformer).
I presume the ubiquitous internet will have the new Mac, which I will look up. But thanks for popping in here again.
Regards!
I recall your previous contributions in this respect - was it almost a year ago? I actually wondered whether you will contribute again. I must go now, and will quickly look up my version of the Mac, but the one I have still uses an interstage transformer, so no problem there with getting sufficient Vg1 (other than that of having an extra transformer).
I presume the ubiquitous internet will have the new Mac, which I will look up. But thanks for popping in here again.
Regards!
Hi Johan,
Just flipping this discussion from my other thread where you mentioned that the trafo winding scheme could be given here maybe or via PM if you like - I would be interested for the future.
It's surprising that nobody else other than yourself or chris seem to have experimented along these lines - is it because of the trafo winding?
Another thing that Chris mentioned
reminded me of a bias scheme I referred back on my "all bias options" thread http://www.diyaudio.com/forums/showthread.php?s=&threadid=96987
where the cathode R is split into two Rs (one large, one small)with large one having a cap accross it. I wonder does this trick with the cathode R work as current source on upper R?
Still trying to learn
John
Just flipping this discussion from my other thread where you mentioned that the trafo winding scheme could be given here maybe or via PM if you like - I would be interested for the future.
It's surprising that nobody else other than yourself or chris seem to have experimented along these lines - is it because of the trafo winding?
Another thing that Chris mentioned
reminded me of a bias scheme I referred back on my "all bias options" thread http://www.diyaudio.com/forums/showthread.php?s=&threadid=96987
where the cathode R is split into two Rs (one large, one small)with large one having a cap accross it. I wonder does this trick with the cathode R work as current source on upper R?
Still trying to learn
John
Few seem to be aware that most implementations of cathode feedback are also a form of G2 feedback, i.e. 'ultralienar' - because although Vg2 is constant, Vg2-k is not. Pure cathode feedback on a pentode is only possible if the screen grid supply is referenced to the cathode and not ground.
IMHO it is a bit misleading referring to an ultralinear connection as particulairly triode-like because of it's equivalent Rp - it appears to me that if one still wants to retain some measure of pentode efficiency, the resulting Rp will be very high compared to a triode connection.
IMHO it is a bit misleading referring to an ultralinear connection as particulairly triode-like because of it's equivalent Rp - it appears to me that if one still wants to retain some measure of pentode efficiency, the resulting Rp will be very high compared to a triode connection.
Johan,
Have you read Walker's original Wireless World article describing his cathode-feedback circuit?
I searched for information on the construction of the OP transformer (Quad11) - there are some very incorrect statements out there!! I finally was given a drawing attributted to an "ancient guy" who actually wound them. The lamination used is not common and I have not been able to locate some down here.
I have wound a smaller transformer intended for use with EL84 or EL81 and the like, and wound the cathode windings bifilar in two sections. Very happy with measured responses, but have yet not got around to actually building something - having been recently inducted into SE Heaven..........
Graeme
Have you read Walker's original Wireless World article describing his cathode-feedback circuit?
I searched for information on the construction of the OP transformer (Quad11) - there are some very incorrect statements out there!! I finally was given a drawing attributted to an "ancient guy" who actually wound them. The lamination used is not common and I have not been able to locate some down here.
I have wound a smaller transformer intended for use with EL84 or EL81 and the like, and wound the cathode windings bifilar in two sections. Very happy with measured responses, but have yet not got around to actually building something - having been recently inducted into SE Heaven..........
Graeme
Graeme,
Yes, I have that article - I think it was by Walker and Williamson, September 1952. I also corresponded with Dr Walker on that. He ends up with some "8-storey" maths (fractions of 8 expressions one on top of the other).
Wavebourn,
Exactly; that was what I was trying to mention, perhaps very briefly because one is sensitive about long posts. You will agree that the Quad topology is best understood by looking at it firstly as UL, then the move to shift the "common" part of the primary to the cathode. I saw some circuits with both a G2 tap on the primary as well as a cathode feedback winding. That made one wonder if the designer understood that he is now effectively getting an overly high G2 tap.
Ilimzn,
Correct as I have just stated. Again, not everybody appears to understand exactly what is going on (and as you do).
I can see your problem with regard to the UL effect, efficiency and rp. It is best realised by looking at a family of Ia-Va graphs for the UL mode. Although they are generally closer to triode than pentode, the efficiency is increased above triode operation in that towards low G1 voltage the graphs bend back to pentode style i.e. the desired lowest anode travel is almost as it was for the pentode. It must also be realised that G2 assists somewhat with output power. I have GE graphs for KT88 where in fact the maximum output power is somewhat higher in the UL mode than pure pentode. The graphs are scarce, but exist for certain tubes (KT66, KT88, 6L6 and EL34), showing the values of rp, output and efficiency that I stated above.
I will try to scan the KT88 graphs (the picture is rather small) and post them here; that should illustrate the general behaviour. They are similar for all power pentodes/beam tubes, only the values vary somewhat. (I could not find similar on the internet.)
Regards
Yes, I have that article - I think it was by Walker and Williamson, September 1952. I also corresponded with Dr Walker on that. He ends up with some "8-storey" maths (fractions of 8 expressions one on top of the other).
Wavebourn,
Exactly; that was what I was trying to mention, perhaps very briefly because one is sensitive about long posts. You will agree that the Quad topology is best understood by looking at it firstly as UL, then the move to shift the "common" part of the primary to the cathode. I saw some circuits with both a G2 tap on the primary as well as a cathode feedback winding. That made one wonder if the designer understood that he is now effectively getting an overly high G2 tap.
Ilimzn,
Correct as I have just stated. Again, not everybody appears to understand exactly what is going on (and as you do).
I can see your problem with regard to the UL effect, efficiency and rp. It is best realised by looking at a family of Ia-Va graphs for the UL mode. Although they are generally closer to triode than pentode, the efficiency is increased above triode operation in that towards low G1 voltage the graphs bend back to pentode style i.e. the desired lowest anode travel is almost as it was for the pentode. It must also be realised that G2 assists somewhat with output power. I have GE graphs for KT88 where in fact the maximum output power is somewhat higher in the UL mode than pure pentode. The graphs are scarce, but exist for certain tubes (KT66, KT88, 6L6 and EL34), showing the values of rp, output and efficiency that I stated above.
I will try to scan the KT88 graphs (the picture is rather small) and post them here; that should illustrate the general behaviour. They are similar for all power pentodes/beam tubes, only the values vary somewhat. (I could not find similar on the internet.)
Regards
Jkeny,
I am splitting this up as there are 2 topics. What Cerrem was referring to if I understood correctly is a solution to this business of maximum allowable value for the power tube grid resistor, especially in the case of fixed bias. (Too high a value can cause a runaway effect as a result of the grid beginning to conduct and going slightly positive - the full explanation is somewhat longer.)
Cerrem was referring to newer models of the McIntosh, where the power stage is driven by a cathode follower to provide a low enough impedance to drive the low grid resistor, then bootstrapping the driver stage (not the power stage) for improved linearity.
With the Quad topology there is signal available for bootstrapping, but to implement fixed bias at the same time matters become somewhat complex, although I will rethink that. (For starters using the cathode feedback will not cause more than a 3-4 fold increase in Rg. In the case of a cathode follower it is very different.) I mentioned earlier that at present I use a kind of servo circuit to buck the dc grid drifting effect with some success (an op-amp is required for that, but it is not in the signal circuit). This allows me to use a 10x higher Rg than specified. It also shows up the odd gassy power tube, where the servo correction goes to abnormal values - perhaps good to know. One can even install a LED indicator as warning that such a situation is developing.
Regards
I am splitting this up as there are 2 topics. What Cerrem was referring to if I understood correctly is a solution to this business of maximum allowable value for the power tube grid resistor, especially in the case of fixed bias. (Too high a value can cause a runaway effect as a result of the grid beginning to conduct and going slightly positive - the full explanation is somewhat longer.)
Cerrem was referring to newer models of the McIntosh, where the power stage is driven by a cathode follower to provide a low enough impedance to drive the low grid resistor, then bootstrapping the driver stage (not the power stage) for improved linearity.
With the Quad topology there is signal available for bootstrapping, but to implement fixed bias at the same time matters become somewhat complex, although I will rethink that. (For starters using the cathode feedback will not cause more than a 3-4 fold increase in Rg. In the case of a cathode follower it is very different.) I mentioned earlier that at present I use a kind of servo circuit to buck the dc grid drifting effect with some success (an op-amp is required for that, but it is not in the signal circuit). This allows me to use a 10x higher Rg than specified. It also shows up the odd gassy power tube, where the servo correction goes to abnormal values - perhaps good to know. One can even install a LED indicator as warning that such a situation is developing.
Regards
Not so mysterious
All of these configurations (UL, CFB) are just methods of programming a Mu factor for a pentode to get a triode result:
The current emitted by the cathode is a result of the total electric field seen there from the various electrodes:
I = k ( Gmg*Vg +Gms*Vs+Gmp*Vp)^1.5
For a Mu calculation, we set I equal to a constant and take derivatives:
Gmg*dVg+Gms*dVs+Gmp*dVp = 0
This equation is used to derive all the following results.
Mu is the ratio of output voltage change to input voltage change:
Mu = -dVout/dVinp
For the triode: Mu = -dVp/dVg = Gmg/Gmp
Gmp = 1/Rp so Mu = Gmg * Rp
For the normal pentode: (dVs = 0)
Mu = -dVout/dVinp = -dVp/dVg = Gmg/Gmp
Gmp is very small in the case of the pentode due to electrostatic screening by the screen with a fixed voltage on it, so Mu is very large.
For 6L6: Gmg = 6000/1000,000 Gmp = 1/Rp = 1/22500 so Mu is 135
For UL pentode: (dVs = U%*dVp)
Mu = -dVout/dVinp = -dVp/dVg = Gmg/(U%*Gms+Gmp)
Again, Gmp is very small, so we can simplify to:
Mu = Gmg/(U%*Gms) = MUtriode/U%
For a typical 43% UL tap and a 6L6 (MUtriode = 8 = Gmg/Gms)
we get Mu = 8/.43 = 18.6
For CFB pentode: (dVinp = dVg + C%*dVout, fixed voltage on screen grid with respect to ground)
Mu = -dVout/dVinp = Gmg/(C%*(Gmg-Gms) - Gmp)
again, Gmp is very small, so we can simplify to:
Mu = Gmg/(C%(Gmg-Gms)) = MUtriode/(C%(MUtriode-1))
which is approx = 1/C%
For a 6L6 (MUtriode = 8 = Gmg/Gms) with 15% CFB we get
Mu = 8/(.15(8-1)) = 7.62
For a 6L6 in Circlotron or McIntosh unity gain circuit (w/o bootstrapped screens) we get
Mu = 8/(.5(8-1)) = 2.286
With bootstrapped screen voltages, the Gms term drops out since dVs = 0
All circuits are triodes effectively, just different Mu's result. UL is good for getting a Mu greater than the tube's MUtriode
rating and CFB is good for getting a Mu less than the MUtriode rating. Consequently, Horizontal Output tubes (MUtriode range of 3 to 4 ) don't need CFB.
Don
All of these configurations (UL, CFB) are just methods of programming a Mu factor for a pentode to get a triode result:
The current emitted by the cathode is a result of the total electric field seen there from the various electrodes:
I = k ( Gmg*Vg +Gms*Vs+Gmp*Vp)^1.5
For a Mu calculation, we set I equal to a constant and take derivatives:
Gmg*dVg+Gms*dVs+Gmp*dVp = 0
This equation is used to derive all the following results.
Mu is the ratio of output voltage change to input voltage change:
Mu = -dVout/dVinp
For the triode: Mu = -dVp/dVg = Gmg/Gmp
Gmp = 1/Rp so Mu = Gmg * Rp
For the normal pentode: (dVs = 0)
Mu = -dVout/dVinp = -dVp/dVg = Gmg/Gmp
Gmp is very small in the case of the pentode due to electrostatic screening by the screen with a fixed voltage on it, so Mu is very large.
For 6L6: Gmg = 6000/1000,000 Gmp = 1/Rp = 1/22500 so Mu is 135
For UL pentode: (dVs = U%*dVp)
Mu = -dVout/dVinp = -dVp/dVg = Gmg/(U%*Gms+Gmp)
Again, Gmp is very small, so we can simplify to:
Mu = Gmg/(U%*Gms) = MUtriode/U%
For a typical 43% UL tap and a 6L6 (MUtriode = 8 = Gmg/Gms)
we get Mu = 8/.43 = 18.6
For CFB pentode: (dVinp = dVg + C%*dVout, fixed voltage on screen grid with respect to ground)
Mu = -dVout/dVinp = Gmg/(C%*(Gmg-Gms) - Gmp)
again, Gmp is very small, so we can simplify to:
Mu = Gmg/(C%(Gmg-Gms)) = MUtriode/(C%(MUtriode-1))
which is approx = 1/C%
For a 6L6 (MUtriode = 8 = Gmg/Gms) with 15% CFB we get
Mu = 8/(.15(8-1)) = 7.62
For a 6L6 in Circlotron or McIntosh unity gain circuit (w/o bootstrapped screens) we get
Mu = 8/(.5(8-1)) = 2.286
With bootstrapped screen voltages, the Gms term drops out since dVs = 0
All circuits are triodes effectively, just different Mu's result. UL is good for getting a Mu greater than the tube's MUtriode
rating and CFB is good for getting a Mu less than the MUtriode rating. Consequently, Horizontal Output tubes (MUtriode range of 3 to 4 ) don't need CFB.
Don
correction
Whups, was getting late last night.
For the CFB case,
dVinp = dVg - C%*dVout (I forgot the inversion in the CFB winding)
Putting this into the main equation:
Gmg*dVg+Gms*dVs+Gmp*dVp = 0
gives:
Gmg(dVin+C%*dVout)+Gms*C%*dVout+Gmp*dVout = 0
re-arranging:
Gmg*dVin +dVout(C%*Gmg+C%*Gms+Gmp) = 0
so Mu = -dVout/dVin = Gmg/(C%*Gmg+C%*Gms+Gmp)
simplifying by letting Gmp = 0
we get:
Mu ~= Gmg/(C%(Gmg+Gms))
then using the MU(triode rating) = MUt = Gmg/Gms
Mu ~= (Gmg/Gms)/(C%(Gmg/Gms + Gms/Gms)
reducing to
Mu ~= MUt/(C%(MUt + 1))
so for a 6L6 with 15% CFB we get
Mu ~= 8/(.15(8+1)) = 5.92
for a Circlotron or McIntosh (w/o screen bootstrapping)
Mu ~= 8/(.5(8+1) = 1.78
With screen bootstrapping, the Gms term drops out, so
Mu ~ = 8/(.5(8 + 0)) = 2
which is a little off since we left out the Gmp term. Gmp will reduce the Mu a little more due to the grid drive required to overcome the plate feedback. (from the pentode case Gmg/Gmp = 135 )
Don
Whups, was getting late last night.
For the CFB case,
dVinp = dVg - C%*dVout (I forgot the inversion in the CFB winding)
Putting this into the main equation:
Gmg*dVg+Gms*dVs+Gmp*dVp = 0
gives:
Gmg(dVin+C%*dVout)+Gms*C%*dVout+Gmp*dVout = 0
re-arranging:
Gmg*dVin +dVout(C%*Gmg+C%*Gms+Gmp) = 0
so Mu = -dVout/dVin = Gmg/(C%*Gmg+C%*Gms+Gmp)
simplifying by letting Gmp = 0
we get:
Mu ~= Gmg/(C%(Gmg+Gms))
then using the MU(triode rating) = MUt = Gmg/Gms
Mu ~= (Gmg/Gms)/(C%(Gmg/Gms + Gms/Gms)
reducing to
Mu ~= MUt/(C%(MUt + 1))
so for a 6L6 with 15% CFB we get
Mu ~= 8/(.15(8+1)) = 5.92
for a Circlotron or McIntosh (w/o screen bootstrapping)
Mu ~= 8/(.5(8+1) = 1.78
With screen bootstrapping, the Gms term drops out, so
Mu ~ = 8/(.5(8 + 0)) = 2
which is a little off since we left out the Gmp term. Gmp will reduce the Mu a little more due to the grid drive required to overcome the plate feedback. (from the pentode case Gmg/Gmp = 135 )
Don
further explanation
In case you are all wondering how pentode characteristic curves (with their knee in the curves on the left side) could possibly be considered as high Mu triode curves:
The knee is due to the screen grid capturing of plate current when the plate voltage isn't about 50 or 100 volts above the screen. The nearly flat part of the curves is the high Mu triode portion remaining, with a very high triode Rp (hence slight tilt).
If we could put a grounded, grid wire aligned, shadow grid immediately behind the screen grid wires to prevent screen grid return capture of plate current, we would get a plate curve set without knees.
The plate curves would contintue their slightly tilted trajectory to the left into the negative plate voltage territory. As long as +Vs*Gms were greater than (-Vg*Gmg) + (- Vp*Gmp), normal (but small) plate current would flow.
Since we operate pentodes in the flat (slightly tilted actually) plate current region anyway, this abstraction is not so unreal.
Pentodes are just high Mu, high Rp triodes.
CFB and UL are just ways to program lower triode Mu's.
Don
In case you are all wondering how pentode characteristic curves (with their knee in the curves on the left side) could possibly be considered as high Mu triode curves:
The knee is due to the screen grid capturing of plate current when the plate voltage isn't about 50 or 100 volts above the screen. The nearly flat part of the curves is the high Mu triode portion remaining, with a very high triode Rp (hence slight tilt).
If we could put a grounded, grid wire aligned, shadow grid immediately behind the screen grid wires to prevent screen grid return capture of plate current, we would get a plate curve set without knees.
The plate curves would contintue their slightly tilted trajectory to the left into the negative plate voltage territory. As long as +Vs*Gms were greater than (-Vg*Gmg) + (- Vp*Gmp), normal (but small) plate current would flow.
Since we operate pentodes in the flat (slightly tilted actually) plate current region anyway, this abstraction is not so unreal.
Pentodes are just high Mu, high Rp triodes.
CFB and UL are just ways to program lower triode Mu's.
Don
Smokin'
Ditto from me - I'm wading thru' your calcs now - as an independent review if you like.
The calcs attached as appendix to the Williamson and Waker paper are reasonably useless. Wrong assumptions in number of places uag1 ug2g1 used interchangably, wrong values plugged into expressions, letting RL -> infinity etc. Its surprising that I haven't seen an article taking them to task about it - unless it went over most folks heads and they were afraid to take on the "authority".
Cheers,
Ian
Ditto from me - I'm wading thru' your calcs now - as an independent review if you like.
The calcs attached as appendix to the Williamson and Waker paper are reasonably useless. Wrong assumptions in number of places uag1 ug2g1 used interchangably, wrong values plugged into expressions, letting RL -> infinity etc. Its surprising that I haven't seen an article taking them to task about it - unless it went over most folks heads and they were afraid to take on the "authority".
Cheers,
Ian
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.