Cathode resistance raises plate resistance.
It will linearize GM, but in doing so, fights
against internal voltage feedback of Mu.
Which do you prefer to be linear? Voltage
or Current? Sorry, no free lunches here.
I do agree the transformer requirements are
eased. Both for parallel and for Triodlington.
Miller adds up for Parallel. But the Miller of
Triodlington's transistor is hidden by the
local feedback. You only have to drive the
miller of one small triode's plate.
Last edited:
I still don't get it. How can the cutoff become more remote by parallelling tubes? The cutoff cannot become more remote than that of the "most remote" tube. Where else would the current flow?
I don't see how averaging slightly different curves can change the shape of the curve. Unless the errors would be in fact correlated, that is.
Again, statistics learns us the exact opposite, i.e. that deviations from the mean curves will decrease when parallelling tubes with independent errors.
I don't question your observations in any way, but I'm trying to reconcile them with first principles. I'd really like to understand why you saw what you saw.
Kenneth
I don't see how averaging slightly different curves can change the shape of the curve. Unless the errors would be in fact correlated, that is.
Again, statistics learns us the exact opposite, i.e. that deviations from the mean curves will decrease when parallelling tubes with independent errors.
I don't question your observations in any way, but I'm trying to reconcile them with first principles. I'd really like to understand why you saw what you saw.
Kenneth
A real triode is any number of parallel triodes inside itself.
The more you got, the better chance some of em will be
really remote ... And the average won't cutoff till the very
last path is pinched away.
The bigger the sample set, the more remote the average.
A sharp cutoff indicates either a very small sample set, or
a tightly controlled variance.
The more you got, the better chance some of em will be
really remote ... And the average won't cutoff till the very
last path is pinched away.
The bigger the sample set, the more remote the average.
A sharp cutoff indicates either a very small sample set, or
a tightly controlled variance.
Okay I follow you there, but we'll need to parallel a lot of devices if we are to be 99% sure one of them is really remote, so practical it ain't.
No! By definition, the average cannot undergo a systematic change in function of the number of samples... 😕
I still can't see how the shape of the curve can change by parallelling samples with statistically independent deviations from a mean.
The only explanation I can see is that the cutoff would appear to be sharper, when the compound curves are scaled up N-fold along the vertical (current) axis and then compared side-by-side to the single device curves. But that is just a scale effect and is trivial, surely this is not where you are getting at?
I say it again I'm really trying to see how it works but some things just don't add up 😕
Kenneth
Remote: The slope of the curve where it meets 0mA?
Or maybe he difference between that slope and the
slope of RP? Don't get me to lying, its over my head...
The bigger set, the more ROUNDED the bell's ends.
And in this case, round on the ends don't mean linear...
I don't know this holds true for really big sample sets?
Probably at some point new statistics get swamped. As
its only those that add new extremes that affect shape.
Its a question that probably requires the experiement to
be run differently. The time I did: My goals were to find
the minimum number of ideal Y^1.5 <= X < Y^2 curves
that could faithfully emulate a 300B. The best answer I
could come to empirically seemed to be four. And curves
closer to square law were working better for me than
curves closer to x=y^1.5 "triode law"...
That answer don't tell us squat about how many sets
before average finally settles down and stops changing?
Nor does it tell the number of actual samples or variance
around a single cathode assembly, nor spread between
triodes of different cathodes? I'm no math whiz, so go
figure...
Or maybe he difference between that slope and the
slope of RP? Don't get me to lying, its over my head...
The bigger set, the more ROUNDED the bell's ends.
And in this case, round on the ends don't mean linear...
I don't know this holds true for really big sample sets?
Probably at some point new statistics get swamped. As
its only those that add new extremes that affect shape.
Its a question that probably requires the experiement to
be run differently. The time I did: My goals were to find
the minimum number of ideal Y^1.5 <= X < Y^2 curves
that could faithfully emulate a 300B. The best answer I
could come to empirically seemed to be four. And curves
closer to square law were working better for me than
curves closer to x=y^1.5 "triode law"...
That answer don't tell us squat about how many sets
before average finally settles down and stops changing?
Nor does it tell the number of actual samples or variance
around a single cathode assembly, nor spread between
triodes of different cathodes? I'm no math whiz, so go
figure...
Last edited:
Any slightly wider space between two grid wires, hotspot on the cathode,
bump or crease on the plate, minor misalignments... I don't think you need
a LOT of samples to smear the cutoff.
Two triodes can look near ideal (but different) by themselves. Once you
average them, differences then take the appearance of a remote cutoff.
As triodes cutoff, they take themselves out of the average. Your logic
based upon large sets may not accurately represent cutoff behavior.
The slope of a single triode path may not be remote at all. But Gm is quite
small compared to the whole parallel set, so the slope of that last cutoff
looks much more remote in context of the bigger sum...
Last edited:
Maybe I'm missing something about the justification of the triodlington. I'm not that familiar with that circuit and don't understand it completely. But... wouldn't paralleling the transistors above the tube just move the problems your talking about (remote cutoff) to similar problems (but not neccessarily identical) that occur in paralleling transistors. I'm not sure I understand the justification, especially if it in anyway changes the "tube" sound.
How does this circuit sound? That's the bottom line. And related to that, I've found a not that close correlation to paralleling unidentical tubes to poor sound. One does need to match current levels on both sides of an opt if going for PPP. Related to this: it really doesn't seem like it would be neccessary to have individual source followers for each tube. I think the cost/complexity/benefit analysis comes out strongly in favor of one driver circuit and a good method to balance "total" current in PP.
Actually, it should be easier to balance total current in a PPP amp. Using statistical analysis the deviation from the norm becomes less and less on each side of the transformer as you parallel more tubes. I think one just needs to make sure there's a 1 ohm resistor in each individual cathode circuit so you can "rough balance" by measurement and then moving tubes around. The fine balance would be the usual circuit adjustment.
How does this circuit sound? That's the bottom line. And related to that, I've found a not that close correlation to paralleling unidentical tubes to poor sound. One does need to match current levels on both sides of an opt if going for PPP. Related to this: it really doesn't seem like it would be neccessary to have individual source followers for each tube. I think the cost/complexity/benefit analysis comes out strongly in favor of one driver circuit and a good method to balance "total" current in PP.
Actually, it should be easier to balance total current in a PPP amp. Using statistical analysis the deviation from the norm becomes less and less on each side of the transformer as you parallel more tubes. I think one just needs to make sure there's a 1 ohm resistor in each individual cathode circuit so you can "rough balance" by measurement and then moving tubes around. The fine balance would be the usual circuit adjustment.
There is another nice feature to accepting the idea of paralleled output tubes. It opens up whole vistas of new possible interesting circuit topologies. In addition to the usual OTL variations there are interesting ones using OPTs.
All of these topologies utilize either either cathode follower or source follower drivers for the paralleled output tubes.
What you end up doing is trading front end complexity and multiple amplification stages for fewer amplification stages and high transconductance. This is especially meaningful if you don't have to have phono amplification. You can use any fairly low power tube that can work with the amplification you have. Often the best sounding tubes don't have a lot of power.
You could obviously get away pretty easily with one single LTP input stage using a 6sn7 into a source follower and follow up with EL84s paralleled for however much power you need. It would probably sound pretty good if the Baby Huey is any indication.
Another possibility, (if anyone was radical enough to try it), is simply using about 10 each 6sn7 in a single final amplification stage with no front end amplification at all, input going straight into the source follower. In PP you use an input transformer to do the phase splitting duties and if you don't have enough amplification make it a 1/6 input transformer. Of course there's all sorts of varying combinations where you use at most 1 stage of tube "voltage" amplification. Depending on the circumstances you could get away with no tube voltage amplification whatsoever, provided you have a standard 2 volt rms source.
All of these topologies utilize either either cathode follower or source follower drivers for the paralleled output tubes.
What you end up doing is trading front end complexity and multiple amplification stages for fewer amplification stages and high transconductance. This is especially meaningful if you don't have to have phono amplification. You can use any fairly low power tube that can work with the amplification you have. Often the best sounding tubes don't have a lot of power.
You could obviously get away pretty easily with one single LTP input stage using a 6sn7 into a source follower and follow up with EL84s paralleled for however much power you need. It would probably sound pretty good if the Baby Huey is any indication.
Another possibility, (if anyone was radical enough to try it), is simply using about 10 each 6sn7 in a single final amplification stage with no front end amplification at all, input going straight into the source follower. In PP you use an input transformer to do the phase splitting duties and if you don't have enough amplification make it a 1/6 input transformer. Of course there's all sorts of varying combinations where you use at most 1 stage of tube "voltage" amplification. Depending on the circumstances you could get away with no tube voltage amplification whatsoever, provided you have a standard 2 volt rms source.
I think what Ken was getting at about the composite curve changing shape with more tubes has to do with tube cutoffs effectively removing tubes from the sample, so averaging does not fully apply (no negative effects to average against positive). For example, if nine tubes cut off at -20V on Vg, but one odd tube cuts off at -30V, then it alone will determine the composite curve shape in the -20 to -30 V range. But I guess one could still argue that it is only a 10% effect now, compared to the higher current levels of the composite bunch. Sorta similar to adding more NFdbk to an amplifier, less total deviation but new higher order artifacts.
I don't see any problem with paralleling more tubes if you can match them up well enough (a curve tracer, and take into consideration the additional paralleing effects mentioned earlier). I have heard comments about a few tubes in parallel sounding "off", and comments that many tubes in parallel were sounding fine. Probably just need a curve tracer matchup more so for fewer tubes.
I don't see any problem with paralleling more tubes if you can match them up well enough (a curve tracer, and take into consideration the additional paralleing effects mentioned earlier). I have heard comments about a few tubes in parallel sounding "off", and comments that many tubes in parallel were sounding fine. Probably just need a curve tracer matchup more so for fewer tubes.
Last edited:
When I said one might get away with no tube voltage amplification I really meant "except" for the final amplification stage, of course.
The idea of the only voltage amplification being the last stage was because I recently got 10 6sn7GTAs. I don't know how well matched they are and have no setup to test them but it is an interesting idea. Max plate dissipation is 5 watts for each of the two plates when run parallel within tube. If one ran it at 4 watts X 2 that's 8 watts per tube. At 25 % efficiency that's 2 watts per tube minimum. At 5 watts at 40% (assuming AB2) thats 4 watts per tube maximum. 6 tubes per channel would give 12 watts/channel minimum, 24 watts maximum.
I don't know if a 6sn7 can be operated in ab2. I know it wasn't meant to be. I also don't know if a single source follower could handle the equivalent of 6 individual input capacitance ( each 3 tubes being driven would by a single follower would drive the equivalent of 6 grids.) An amp like that would
definitely not need a separate input voltage amplification stage with the right input transformer. I didn't mean to hijack the thread but it is an interesting back of the napkin idea that sort of got away from me. Sorry
I just realized a big, big, big plus to this idea. No coupling capacitors required at all. Normally the coupling capacitor is in front of the source follower. But in this case the input transformer would isolate that requirement. And you wouldn't need complicated circuitry to provide that direct coupling which is a big plus.
Again, sorry for going off on a tangent.
The idea of the only voltage amplification being the last stage was because I recently got 10 6sn7GTAs. I don't know how well matched they are and have no setup to test them but it is an interesting idea. Max plate dissipation is 5 watts for each of the two plates when run parallel within tube. If one ran it at 4 watts X 2 that's 8 watts per tube. At 25 % efficiency that's 2 watts per tube minimum. At 5 watts at 40% (assuming AB2) thats 4 watts per tube maximum. 6 tubes per channel would give 12 watts/channel minimum, 24 watts maximum.
I don't know if a 6sn7 can be operated in ab2. I know it wasn't meant to be. I also don't know if a single source follower could handle the equivalent of 6 individual input capacitance ( each 3 tubes being driven would by a single follower would drive the equivalent of 6 grids.) An amp like that would
definitely not need a separate input voltage amplification stage with the right input transformer. I didn't mean to hijack the thread but it is an interesting back of the napkin idea that sort of got away from me. Sorry
I just realized a big, big, big plus to this idea. No coupling capacitors required at all. Normally the coupling capacitor is in front of the source follower. But in this case the input transformer would isolate that requirement. And you wouldn't need complicated circuitry to provide that direct coupling which is a big plus.
Again, sorry for going off on a tangent.
Last edited:
Member
Joined 2009
Paid Member
I like how you think !
http://www.diyaudio.com/forums/tube...g-triode-fet-complementary-feedback-pair.html
The combination I looked at was of Triode and MOSFET to reduce the biassing requirements but I never built it. I like the name Triodlington but with the CFP arrangement it begs for something like Sziklode 😀
Did I not mention it (#20)? MOSFET gate cap phase shift inside the plate feedback loop?
Well I suppose even a BJT has Miller to shift feedback phase. Its just a matter of degrees.
Sure MOSFET probably works, but is phase shifted Mu still Mu enough to be considered
a parallel current mirror to the reference triode? There is DC coupling at the gate for sure,
but I'm not sure the gate capacitance and stopper can be considered "direct" coupling?
PNP is still the current multiplier of choice. Hard to beat good PNP for dead flat Beta.
Well I suppose even a BJT has Miller to shift feedback phase. Its just a matter of degrees.
Sure MOSFET probably works, but is phase shifted Mu still Mu enough to be considered
a parallel current mirror to the reference triode? There is DC coupling at the gate for sure,
but I'm not sure the gate capacitance and stopper can be considered "direct" coupling?
PNP is still the current multiplier of choice. Hard to beat good PNP for dead flat Beta.
Last edited:
I'm looking at 6DR7 (actually 13DR7). Section #1 is very close to half of a 6SL7 (old school?). Section #2 isn't a "high current monster". I found a schematic after I picked that tube but it does not have any component valuses listed. I figered I need a 62K 1 watt and 2 3.6K 2 watt resistors. Now I.m looking for those. Why are there more shoe stores than tube stores?
Jim
The 6SN7GTA/B certainly can be operated in Class *2, since it was intended for vertical deflection duty. Even includes control grid radiator "wings".
That's why this forum is a wonderful resource. Thanks Miles for correcting my bad assumption. Is the 6SN7GTA/B fully equivalent to the 6SN7 besides the stepped up power? Is it as linear as the low power version, and as a point of interest, did they both originate for vertical deflection duty?
I may have to look closer at this idea for an amp. Now the only real question remaining is if one source follower mosfet can drive 6 grid/plate capacitances of 4pf each. With an in-circuit average mu of 10 that should work out to 240 pf of miller capacitance. Can one mosfet drive that much capacitance? I know there must be a simple formula but I don't know it. I'm just glad there's people here who are more knowledgable than I am about this stuff.
I have been contemplating paralleled 6SN7GT's for quite some time, after reading a tubecad journal article about paralleled triode connected EL84 output stages. I figured why use that many pretend triodes when you could just use a pile of real triodes, and why not the best small signal triode while you're at it? I figure a triode-loaded grounded-cathode input stage cascaded into a interstage transformer in order to split the phase, and then eight 6SN7GT (four bottles) per OPT side, individual cathode resistors or LED arrays per tube. Eighteen 6SN7/12SN7 Would make for a ridiculous Super Amp™. Toroidal Mains transformers for output most likely, as I've spoken to several who have had great results with them. It would even have XLR connectors between the Transformer and the output stage for future experiments with balanced preamps.
This amp, should it ever get built, would physically be an absolute beast! Dual monoblock style would most likely be the way it happens. If I can come up on a sack of 6SN7 big enough they are definitely getting stowed away for this design.
This amp, should it ever get built, would physically be an absolute beast! Dual monoblock style would most likely be the way it happens. If I can come up on a sack of 6SN7 big enough they are definitely getting stowed away for this design.
Why not use the Russian equivalent 6H8C then? They are affordable in large numbers and well-built. I like them 🙂
Kenneth
Kenneth
Because unless your pile of real triodes are impossibly well matched,
almost any pretend triode performs better than the pile of real ones.
Averaging a pile of small perfect triodes that don't match makes for
a big imperfect one.
And a beam power tube's construction almost assures it will behave
extremely linear as a triode (all internal paths the same) . Only the
most thoughtfully constructed (expensive) of real triodes could hope
to compete.
Now if you wanted to use a bank of beam power tubes (strapped
as Pentodes) slaved to a single beam power tube strapped as the
feedback Triode. You can avoid averaging any mismatched cutoff.
Not significantly different than N-CH Triodlington, except that you
don't get a free 4V drop to abuse for cathode bias....
Without sand, its a little more complicated, but it can be done.
http://www.tubecad.com/2009/11/blog0176.htm
Last edited:
Ken
I think one could say, using your argument, that one should never use tube PP because no two tubes ever have consistent characteristics between the two sides. This is a hobby for most of us and hobbys are enjoyable because we can try things. Its not like work where we have a boss telling us we how we have to do things. At least I would hope that hobbys are like that.🙄
I think one could say, using your argument, that one should never use tube PP because no two tubes ever have consistent characteristics between the two sides. This is a hobby for most of us and hobbys are enjoyable because we can try things. Its not like work where we have a boss telling us we how we have to do things. At least I would hope that hobbys are like that.🙄
The 6SN7GTA/B is equivalent to the older, lower plate dissipation version. The main differences are an enhanced plate dissipation rating, and the extension of the plate characteristic into positive Vgk territory. Most of the 6SN7s that you see today are the GTA version anyway. So far as I can tell, the 6SN7 wasn't originally a vertical deflection type, but was quickly adapted to digital and quasi-digital operations (during WW II, and essssss-loads of 6SN7s found their way into radar sets) and was the choice for vertical deflection duty after the war when TV started coming into its own during the late 1940s.
A power MOSFET can easily drive 240pF of load capacitance (22K1 @ 30KHz). The r(don) of power MOSFETs is extremely low, so this is NBD.
- Status
- Not open for further replies.