Interested in design guidelines for minimizing H3 in LTP drivers. I know there has been a bit of work with tuning the k resistor in a simple grounded cathode stage, but I haven't seen any relating to LTP stages.
Has anyone done and published any such work? Tube selection, circuit design, anything?
Has anyone done and published any such work? Tube selection, circuit design, anything?
Valve choice, bias, anode load - basically the same issues as a grounded cathode stage. With an LTP there is the additional complication of re-entrant distortion from the second order at the cathode mixing with the incoming signal at the grid.
There is this paper by Kiebert:
http://www.clarisonus.com/Archives/...System Design Factors for Audio Ampifiers.pdf
Worth taking a look at the gm curves on the E55L datasheet, pages 5 and 9, which are plotted versus current (CCS tail use) on top page 9 and versus grid voltage on page 5 (grounded cathode stage).
These are fairly typical curves for most tubes, maybe with a more or less accentuated top or bottom curvature of the page 5 graph.
Page 7 compares pentode and triode configuration gm curves versus voltage.
I think this is shown for the grid1 function only however. A triode mode LTP needs to take into account any plate feedback swings, which lowers the -effective- V to I power law to just above 1.0+.
(gm is the 1st derivative of the V to I power law. So a square law V to I, I = kV^2, gives a linear ramping gm, gm = dI/dV = 2kV, like seen in the center of the typical gm versus Vgrid1 curves)
The triode plate feedback flattens out horizontally the gm curvature considerably, but the two gm's are still summing to a slightly humped (in the middle) gm curve. Only takes a little tail R to flatten it out then. Obviously, pinning the plate voltage of one side of an LTP circuit is a bad idea for linearity.
http://frank.pocnet.net/sheets/009/e/E55L.pdf
For an LTP stage, you are trying to get the sum of the page 9 gm curves versus current (one flipped around and overlapped) to come out to a constant. Putting a tail resistor under the LTP gives something in between the two curve sets.
For a grounded grid class A output stage you are trying to get the page 5 gm versus Vg1 curves to sum to a constant.
http://www.clarisonus.com/Archives/...System Design Factors for Audio Ampifiers.pdf
Worth taking a look at the gm curves on the E55L datasheet, pages 5 and 9, which are plotted versus current (CCS tail use) on top page 9 and versus grid voltage on page 5 (grounded cathode stage).
These are fairly typical curves for most tubes, maybe with a more or less accentuated top or bottom curvature of the page 5 graph.
Page 7 compares pentode and triode configuration gm curves versus voltage.
I think this is shown for the grid1 function only however. A triode mode LTP needs to take into account any plate feedback swings, which lowers the -effective- V to I power law to just above 1.0+.
(gm is the 1st derivative of the V to I power law. So a square law V to I, I = kV^2, gives a linear ramping gm, gm = dI/dV = 2kV, like seen in the center of the typical gm versus Vgrid1 curves)
The triode plate feedback flattens out horizontally the gm curvature considerably, but the two gm's are still summing to a slightly humped (in the middle) gm curve. Only takes a little tail R to flatten it out then. Obviously, pinning the plate voltage of one side of an LTP circuit is a bad idea for linearity.
http://frank.pocnet.net/sheets/009/e/E55L.pdf
For an LTP stage, you are trying to get the sum of the page 9 gm curves versus current (one flipped around and overlapped) to come out to a constant. Putting a tail resistor under the LTP gives something in between the two curve sets.
For a grounded grid class A output stage you are trying to get the page 5 gm versus Vg1 curves to sum to a constant.
Last edited:
Something that I noticed the hard way several years back.
If using a triode based LTP with a single ended input, one grid grounded or tied to an AC ground, care should be taken to provide roughly similar impedances to ground for each grid. If one grid is grounded through a low value stopper, while the other is tied to the wiper of a volume pot the driving impedances are not equal, and vary with the pot setting. This can cause different phase responses at HF because the Miller capacitance is working against a different impedance in each triode.
I noticed that the frequency response, and higher order harmonic distortion at the higher audio frequencies varied with volume pot setting.
If using a triode based LTP with a single ended input, one grid grounded or tied to an AC ground, care should be taken to provide roughly similar impedances to ground for each grid. If one grid is grounded through a low value stopper, while the other is tied to the wiper of a volume pot the driving impedances are not equal, and vary with the pot setting. This can cause different phase responses at HF because the Miller capacitance is working against a different impedance in each triode.
I noticed that the frequency response, and higher order harmonic distortion at the higher audio frequencies varied with volume pot setting.
While most tubes have a gm versus plate current curve that curves like a square root function, some tubes are better fitted to LTP (CCS tail) function.
The 12AT7 is notorious for distortion in SE mode (it is a non-linear mixer tube after all), but it has a gm curve versus current that is more amenable to CCS tailed LTP use.
The gm curve versus Ip looks more like a linear ramp (which summed with its flipped around version, comes closer to constant gm)
http://frank.pocnet.net/sheets/093/1/12AT7.pdf
The 12AT7 is notorious for distortion in SE mode (it is a non-linear mixer tube after all), but it has a gm curve versus current that is more amenable to CCS tailed LTP use.
The gm curve versus Ip looks more like a linear ramp (which summed with its flipped around version, comes closer to constant gm)
http://frank.pocnet.net/sheets/093/1/12AT7.pdf
Woops. Just noticed the last line of my comments on post #3 should have been:
"For a grounded cathode, class A, P-P output stage, you are trying to get the (E55L page 5) gm versus Vg1 curves to sum to a constant.
---------------------------------------
On the tail resistance compensation scheme mentioned for an LTP, one is adjusting the curvature of the low current region (left side of the gm graphs) to get close to a linear ramp through the operating region. The two extremes shown in the graphs, are curvature upwards for the gm versus Vg1, and curvature downwards for the gm versus Ip. A tail impedance gives you the range in between, with some specific R tail getting close to a linear ramp throughout. And then opposing gm ramps sum to a constant gm through the region.
Another way of looking at it is, the tail R allows the individual tube gm's to increase at signal extremes by allowing more cathode current than the CCS tail alone would have allowed.
-----------------
Kieberts comment about a different 3rd harmonic term polarity between the triode and pentode case is somewhat confusing in his paper (not explained). (end of page # 26, ie. bottom of 2nd page)
Since the grid/cathode interface for either is identical. I can only take that to mean it is including the effect of plate feedback in the triode case. Which has the effect of greatly linearizing the triode case, bringing the effective gm curve closer to a constant with slightly humped curvature in the middle for the sum.
By increasing the current in the pentode case, one can reverse the 3rd harmonic sign, by operating in the upper gm region where the gm versus Vg1 curve is curving downward. So no need for a tail R to compensate, just pick the correct idle current. (or one could put some feedbacks on the screen grids to make it similar to a triode LTP case, even at lower currents)
One could also play off + 3rd harmonic sign (humped up gm in middle) against a subsequent circuit stage with
- 3rd harmonic sign (bowl shaped gm in middle). Much like 2nd harmonic cancellation is done sometimes. So triode drivers with pentode outputs might do it. Or visa-versa. Fun and games....
Need an FFT program to probe the stages.
"For a grounded cathode, class A, P-P output stage, you are trying to get the (E55L page 5) gm versus Vg1 curves to sum to a constant.
---------------------------------------
On the tail resistance compensation scheme mentioned for an LTP, one is adjusting the curvature of the low current region (left side of the gm graphs) to get close to a linear ramp through the operating region. The two extremes shown in the graphs, are curvature upwards for the gm versus Vg1, and curvature downwards for the gm versus Ip. A tail impedance gives you the range in between, with some specific R tail getting close to a linear ramp throughout. And then opposing gm ramps sum to a constant gm through the region.
Another way of looking at it is, the tail R allows the individual tube gm's to increase at signal extremes by allowing more cathode current than the CCS tail alone would have allowed.
-----------------
Kieberts comment about a different 3rd harmonic term polarity between the triode and pentode case is somewhat confusing in his paper (not explained). (end of page # 26, ie. bottom of 2nd page)
Since the grid/cathode interface for either is identical. I can only take that to mean it is including the effect of plate feedback in the triode case. Which has the effect of greatly linearizing the triode case, bringing the effective gm curve closer to a constant with slightly humped curvature in the middle for the sum.
By increasing the current in the pentode case, one can reverse the 3rd harmonic sign, by operating in the upper gm region where the gm versus Vg1 curve is curving downward. So no need for a tail R to compensate, just pick the correct idle current. (or one could put some feedbacks on the screen grids to make it similar to a triode LTP case, even at lower currents)
One could also play off + 3rd harmonic sign (humped up gm in middle) against a subsequent circuit stage with
- 3rd harmonic sign (bowl shaped gm in middle). Much like 2nd harmonic cancellation is done sometimes. So triode drivers with pentode outputs might do it. Or visa-versa. Fun and games....
Need an FFT program to probe the stages.
Last edited:
Since one has limited options for flattening the gm sum of an efficient class AB P-P pentode output stage (no significant class A overlap), other than maybe UL Fdbks, it could make good sense to use a differential driver stage with the opposite 3rd harmonic polarity (or gm sum curvature). The pentode outputs will have gm curving upwards through a large central region in P-P class AB. Hard to compensate the whole operating region though, since the gm curvature changes to downward (-) curvature at high current. (gm versus Vg1, page 5 E55L, top right side) Class AB will generally have a full range gm sum that is like an almost flat bottomed V, the bottom overlap part being curved upward like a shallow bowl. And the top of the V edges curving over outwards like wings. (again, gm versus Vg1, page 5 E55L, top right side)
I think the only way to really fix all that mess is massive amounts of Neg Fdbk for class AB. (put most of it in local N Fdbk form so as to not diss the OT phase shifts)
Or try some clever class A to class B conversion scheme:
http://www.diyaudio.com/forums/tube...1-g2-driven-amp-concept-worth-pursuing-2.html
I think the only way to really fix all that mess is massive amounts of Neg Fdbk for class AB. (put most of it in local N Fdbk form so as to not diss the OT phase shifts)
Or try some clever class A to class B conversion scheme:
http://www.diyaudio.com/forums/tube...1-g2-driven-amp-concept-worth-pursuing-2.html
Last edited:
Basics for beginners:
In an LTP then perfect balance of the 2 sides will result in minimum 2H distortion.
What distortion is left is then mainly 3H distortion from the non-linearity of the tubes themselves.
So choose a tube with good linearity for minimum 3H.
Cheers,
Ian
In an LTP then perfect balance of the 2 sides will result in minimum 2H distortion.
What distortion is left is then mainly 3H distortion from the non-linearity of the tubes themselves.
So choose a tube with good linearity for minimum 3H.
Cheers,
Ian
I started this thread because I noticed h3 seemed high in most simple LTPs I designed. Clearly in a diff amp h2 is low so h3 stands out more. But still...
So I did a few sims to get an idea of relative distortion of LTP vs other topologies. I took the LTP in the Citation V, a good simple 6CG7 LTP with ample supply and long tail. At max signal (around 35Vp each out), I get h2=-50dB and h3=-41dB, THD=0.91. And a gain of x5.2. If I put a CCS in the tail, h2 goes down by 8dB but h3 stays identical.
I don't know how to improve h3 in that particular LTP (or any similar simple LTP), but I do know that I can design a simple VA+Concertina that does much better. I ran a 6SL7 into 6SN7 Concertina, same output level, and I get h2=-54dB, h3=-63dB, THD=0.22. Gain of x65. I'd say much better.
Where am I wrong here? Are my spice models faulty? Comments?
So I did a few sims to get an idea of relative distortion of LTP vs other topologies. I took the LTP in the Citation V, a good simple 6CG7 LTP with ample supply and long tail. At max signal (around 35Vp each out), I get h2=-50dB and h3=-41dB, THD=0.91. And a gain of x5.2. If I put a CCS in the tail, h2 goes down by 8dB but h3 stays identical.
I don't know how to improve h3 in that particular LTP (or any similar simple LTP), but I do know that I can design a simple VA+Concertina that does much better. I ran a 6SL7 into 6SN7 Concertina, same output level, and I get h2=-54dB, h3=-63dB, THD=0.22. Gain of x65. I'd say much better.
Where am I wrong here? Are my spice models faulty? Comments?
Yes, and that is the way it's supposed to work.
For your cathodyne (concertina) comparison, was that also putting out 35Vpk each side?
Was the 6CG7 LTP being fed a very clean signal when it was putting out 35Vpk per side?
For the 6CG7 LTP, what were the operating points, and what were the plate loads?
Make sure the 6CG7 is running at a linear spot for the voltage swing required. In some cases this will work out better with less than 4mA plate current per triode and relatively large (47k) plate load resistors. 3HD should be lower if you run the 6CG7 with higher current, but then you need smaller value plate load resistors and a higher B+ voltage.
Which 6CG7 model are you using? Have you tried a 6SN7 model to compare? (They should be very similar.)
--
Last edited:
Are you primarily trying to implement a phase splitter? In that case, putting a 3rd harmonic canceling R (approx 1/gm range) in the LTP tail will not be so helpful, since it will take unbalanced loads to get equal outputs then. I would suggest arranging a drive from an attenuated (1st inverter) plate, to drive the grid of the 2nd side of the LTP. Then put the comp. R tail in to get rid of 3rd harmonic. Some re-adjusments likely required to get the grid drive level right for equal outputs. (paraphase, see-saw, or the floating paraphase splitter, tail R adjusted to eliminate 3rd harmonic)
Another alternative are the beam deflection tubes like 6JH8. It's like a CCS'd tail LTP in a bottle, except without the 3rd harmonic problem. (linear deflection mode from the deflectors gives low distortion outputs) One could also arrange an attenuated (1st plate to 2nd) deflector drive to get twice the output. (or floating paraphase arrangement) Put cascode Fets above the plates if you want lots of gain. (typically a max gain of 6 using CCS plate loads without the cascodes, less with R loads).
http://frank.pocnet.net/sheets/093/6/6JH8.pdf
Another alternative are the beam deflection tubes like 6JH8. It's like a CCS'd tail LTP in a bottle, except without the 3rd harmonic problem. (linear deflection mode from the deflectors gives low distortion outputs) One could also arrange an attenuated (1st plate to 2nd) deflector drive to get twice the output. (or floating paraphase arrangement) Put cascode Fets above the plates if you want lots of gain. (typically a max gain of 6 using CCS plate loads without the cascodes, less with R loads).
http://frank.pocnet.net/sheets/093/6/6JH8.pdf
Last edited:
Putting a CCS in the tail will reduce 2nd because it forces balance and balance is what eliminates 2nd.
LTP has 3rd from two sources: 3rd order curvature in the valves themselves, plus re-entrant 2nd from the cathodes mixing with the input signal via the 2nd order curvature of the valve. The latter effect is missing from a grounded cathode stage. One thing to try for an LTP is a remote-cutoff triode, as that should have intrinsic 3rd of the opposite sign to a normal triode so the re-entrant 3rd will partly cancel instead of add to it.
LTP has 3rd from two sources: 3rd order curvature in the valves themselves, plus re-entrant 2nd from the cathodes mixing with the input signal via the 2nd order curvature of the valve. The latter effect is missing from a grounded cathode stage. One thing to try for an LTP is a remote-cutoff triode, as that should have intrinsic 3rd of the opposite sign to a normal triode so the re-entrant 3rd will partly cancel instead of add to it.
Yes
These are LTSpice sims, not actual measured circuits, so yes. The Citation V LTP (6CG7) is fed single sided, exactly per the HK schematic.
As noted, the operating point is exactly as found in the Citation V. I put the exact circuit in the sim. Va=375, Vk=150, plate loads 22k, Ia about 3mA per tube.
Ayumi as found in the sim thread and yes
Yes and I mistakenly assumed that eliminating h2 would lessen the h3 due to IM between h1 and h2.
That's an interesting idea. Are you suggesting an LTP made of one regular and one remote cutoff triode?
I don't use the Ayumi 6CG7 or 6SN7 model because it seems to disagree with the other models (I think it's an outlier). Try this one instead, just to compare (courtesy of Wayne Clay):
Code:
* 6SN7_GE LTSpice model
* Modified Koren model (8 parameters): mean fit error 0.0771164mA
* Traced by Wayne Clay on 10/11/2013 using Curve Captor v0.9.1
* from General Elctric data sheet
.subckt 6SN7_GE P G K
Bp P K I=
+ (0.01996138472m)*uramp(V(P,K)*ln(1.0+(-0.06003194624)+exp((4.721146098)+
+ (4.721146098)*((20.78010878)+(-111.9910206m)*V(G,K))*V(G,K)/sqrt((29.06231184)**2+
+ (V(P,K)-(6.821890538))**2)))/(4.721146098))**(1.378722311)
Cgk G K 2.8p ; 0.2p added
Cpk P K 2.2p ; 0.5p added
Cgp G P 4.5p ; 0.5p added
Rpk P K 1G ; to avoid floating nodes
d3 G K dx1
.model dx1 d(is=1n rs=2k cjo=1pf N=1.5 tt=1n)
.endsDoes that one give you very different results?
Also, as I've been told many times now, real life results can be very different from spice results.
--
No. The two triodes must be identical if you want 2nd to cancel.dgta said:
The snag with the LTP is that 2nd cancels in differential mode but adds in common mode, hence you get 2nd at the cathodes.
Thanks for posting that model. It gives me identical results on the couple circuits I tried, but I'll keep it for future use. You can never have enough models 🙂
Interesting, it's another one of those things that just goes by as the way to do things until somebody takes another look. I imagine it taken care of easily with a ganged pot wired as a variable resistor on the passive side. Any trouble with that?
I think the take-away message here is that both grids in a triode LTP should see consistently low impedance to ground if the stage has much gain. You would need a four-gang input level control to balance miller effect in a stereo amp with high-gain triode LTP input stages. It's better to simply ditch the input level control and rely on the preamp's low output impedance.
I'd say there's a message here. We humans are pack animals and although we all like to declare our rugged individualism we can't seem to help ourselves from asking the other guy what he's doing. I say go for the Concertina and make life simpler for yourself. It's staring you in the face.
There are complications to having a concertina over a LTP such as needing generally a high voltage for it. In some amp designs it is not a disadvantage ,but with designs needing a low voltage for the output tubes that can play havoc. But even this can be overcome over come with careful design.
Right now I'm uses designing an amp that uses the two advantages that a concertina has. Those two advantages are very low higher order distortion and double the mu of the same two tubes (grounded cathode, concertina) over an LTP pair. It uses both 6SN7s for all of the voltage gain and parallel high sensitivity output tubes for the current gain in the push-pull. "No PPP" may be another sacred cow that is also a good candidate for a barbecue.🙂
Last edited: