If this has been covered before I cant find it all in one place in the archives.
What are everyones thoughts on parallel outputs? I intend to parallel some 807/6L6 type valves in AB2 triode. Although discussion on all types in general would be appreciated.
Apart from the advantages of possibly better output transformer and averaging of characteristics. What are the other possible advantages and most importantly disadvantages?
Cheers Matt.
What are everyones thoughts on parallel outputs? I intend to parallel some 807/6L6 type valves in AB2 triode. Although discussion on all types in general would be appreciated.
Apart from the advantages of possibly better output transformer and averaging of characteristics. What are the other possible advantages and most importantly disadvantages?
Cheers Matt.
Disadvantages: more tubes = more sockets. Advantages: many.
Including, if you put enough tubes in parallel you don't need an output transformer.
Jim
Including, if you put enough tubes in parallel you don't need an output transformer.
Jim
Another disadvantage is the risk of current hogging but this can be minimized to some extent by adding a resistor of 100 ohms in each plate circuit.
Yup. But not as bad as SS. Anyway, I wish people would say "This is for a home stereo or a guitar anp or a pirate radio transmitter modulator".
Matt B.H. ?
Jim
Matt B.H. ?
Jim
Sorry, yes its for home use. I know 80-100W is lots but I already have the PS transformers and valves/sockets etc. I just need to get some output transformers.
I have read that some people think parallel outputs degrade the sound. I can only see advantages though.
Cheers Matt.
I have read that some people think parallel outputs degrade the sound. I can only see advantages though.
Cheers Matt.
Input capacitance of the stage increases when you parallel the tubes. So, You need to make sure that driver stage can provide enough current to drive output stage.
Well, as Don Smoking Amp once explained it to me:
A triode is any number of parallel triodes internally.
Any one path by itself would have perfect Mu and
parallel curves that never "lean over to the right".
Its the average of those many slightly different
triode paths, each with slightly different cutoff
thats makes for the leaning curve set you would
typically see. And more so, the less truly parallel
and well controlled the internal pathing.
Now your beam power tube dumbed down to a
triode probably has really good internal parallel
sameness, like a high quality triode. But start
pluggin in multiple tubes, you can expect more
and more the leaning over curves of the triode
with sloppier internal construction.
I would look into Triodlington, where a single
high quality triode provides 1/(Mu-1) plate to
cathode feedback path for a power transistor's
collector to base.
Thus loaning the sand: high input impedance,
low output impedance, and proper triode rule
of Mu. Without having to parallel a bunch of
real tubes to get required transconductance.
And avoiding degradation of the triode curve
in the blender.
Like magically duplicating single malt Scotch in
a current mirror. Would it be better than a huge
but blended barrel of the real thing? I don't know
when sandy mirror magic might be the better or
worse than a blended reality?
If you are going into AB2, you can only use
the PNP (Sziklai CFP) version of Triodlington.
Else forward grid current gets multipled too.
But transconductance of such a pairing is so
high, no need or reason to operate into AB2...
A triode is any number of parallel triodes internally.
Any one path by itself would have perfect Mu and
parallel curves that never "lean over to the right".
Its the average of those many slightly different
triode paths, each with slightly different cutoff
thats makes for the leaning curve set you would
typically see. And more so, the less truly parallel
and well controlled the internal pathing.
Now your beam power tube dumbed down to a
triode probably has really good internal parallel
sameness, like a high quality triode. But start
pluggin in multiple tubes, you can expect more
and more the leaning over curves of the triode
with sloppier internal construction.
I would look into Triodlington, where a single
high quality triode provides 1/(Mu-1) plate to
cathode feedback path for a power transistor's
collector to base.
Thus loaning the sand: high input impedance,
low output impedance, and proper triode rule
of Mu. Without having to parallel a bunch of
real tubes to get required transconductance.
And avoiding degradation of the triode curve
in the blender.
Like magically duplicating single malt Scotch in
a current mirror. Would it be better than a huge
but blended barrel of the real thing? I don't know
when sandy mirror magic might be the better or
worse than a blended reality?
If you are going into AB2, you can only use
the PNP (Sziklai CFP) version of Triodlington.
Else forward grid current gets multipled too.
But transconductance of such a pairing is so
high, no need or reason to operate into AB2...
Last edited:
Oh yeah, that too. AB2 and complex waveforms (music) don't go together.
Go class A (it's been done forever) or class D (it's almost never been done with tubes).
Jim
As Irakli said... you will need one hell of a driver to drive two parallelled grids in an AB2 configuration. Even in AB1, many PPP amps sound bad because the driver is inadequate (i.e. the designer took a PP design and simply added output tubes, disregarding the driver stage. Don't do that.).
Kenneth
Kenneth
I was not suggesting any problem with AB2, except that NPN Darlington arrangement
would multiply grid forward currents. PNP Sziklai would have no such troubles.
Was the only point about AB2 I was trying to make earlier... but I can expand on it...
I'm not against AB2 with music, except it usually takes sand to drive it properly. And
then you might just as well have thrown your sand into a Triodlington, and have more
transconductance than you could ever dream. You can do both of course, but its sort
of pointless. Once you have nearly unlimited current of sand under the firm voltage
control of one really good triode, what reason did we need the added hassle of tryin
to drive AB2 on both sides of the forward current knee?
If your talkin' one of those transmitter triodes thats in AB2 all the time, the knee is
maybe not a problem. But you still need big current to drive it. I think you still end
up throwing sand at the problem of driving a triode act like sand with Mu, so why?
Triodlington operated below the grid current knee is a better deal all ways round...
would multiply grid forward currents. PNP Sziklai would have no such troubles.
Was the only point about AB2 I was trying to make earlier... but I can expand on it...
I'm not against AB2 with music, except it usually takes sand to drive it properly. And
then you might just as well have thrown your sand into a Triodlington, and have more
transconductance than you could ever dream. You can do both of course, but its sort
of pointless. Once you have nearly unlimited current of sand under the firm voltage
control of one really good triode, what reason did we need the added hassle of tryin
to drive AB2 on both sides of the forward current knee?
If your talkin' one of those transmitter triodes thats in AB2 all the time, the knee is
maybe not a problem. But you still need big current to drive it. I think you still end
up throwing sand at the problem of driving a triode act like sand with Mu, so why?
Triodlington operated below the grid current knee is a better deal all ways round...
Last edited:
What I do (and would advise everybody) is I never, ever build a circuit I don't understand 200%. Because after building it, it will need to be troubleshooted and then what...?
There are tons of exotic VT or VT/SS circuits around which all have their strong points but they all have a weak point, i.e. the above.
There's beauty in simplicity too. And BTW, adding a pair of output tubes only generates +3dB which is barely audible.
Just my 2c
Kenneth
There are tons of exotic VT or VT/SS circuits around which all have their strong points but they all have a weak point, i.e. the above.
There's beauty in simplicity too. And BTW, adding a pair of output tubes only generates +3dB which is barely audible.
Just my 2c
Kenneth
Hi Ray_Moth,
You said "Another disadvantage is the risk of current hogging but this can be minimized to some extent by adding a resistor of 100 ohms in each plate circuit. "
I've seen the use of plate resistors, eg Ampeg SVT, but never really understood the "Why?" of it.
Can you explain?
Thanks
You said "Another disadvantage is the risk of current hogging but this can be minimized to some extent by adding a resistor of 100 ohms in each plate circuit. "
I've seen the use of plate resistors, eg Ampeg SVT, but never really understood the "Why?" of it.
Can you explain?
Thanks
This should catch you up to the 200% understanding threshold. Or maybe not...
Nothing wrong with real parallel Triodes, except how averaging "cheapens" the
curve exactly like less accurate internal construction would... On the other hand,
the average has lower noise. I doubt that merit matters much for output stage...
This duty screams for current multiplication...
Attachments
Last edited:
Well thanks for the replies so far🙂
I know I will only get 3dB more but its the headroom that counts. My speakers are 90dB at 1W 1M 200W RMS and they can take this, kids fingers I have learnt are far more damaging😡. So I will probably only ever use an average of maybe 6W whilst playing fairly loud but the excursions will be up there in the 80W range occasionaly.
I will be using fet followers as in the 6l6 AB2 thread. So grid input capacitance is no prob.
Sorry for rambling, So to sum up with advantages first.
More power, although twice the power is only a max of 3dB in the real world.
A possibly better output transformer due to lower leakage inductance and parrasitic capacitance from the lower load impedance and hence less turns etc. etc.
Averaging of characteristics, although I can only see this being beneficial with vast amounts of output valves ie an OTL or some crazy multi KW regae rig. This leads to a good point by Kenpeter that the average may be a poor representation of the ideal. I can see it being an advantage for paralleling thousands of fets for a phono input stage to abtain a far lower noise, its of no use in an output stage. So on to disadvantages:
More heater current, this is no problem for me with my power transformers🙂 but it may cause hum problems due to increased electromagnetic radiation from the exposed higher current ac wiring.
More sockets, oh well🙂
The main problem I can envisage is finding reasonably matched sets of output valves or arranging a suitable bias arangement and the direct coupling of A/AB2. Maybe a seperate follower per valve with provision for adjustment of each valve which would require capacitor coupling to the fet and the additon of another time constant. Bias servo? even more sand, its turning into a house of valves on stilts of sand😉
So I guess the advantages/disadvantages sort of weigh up.
Tarzan, I think the 100R resistors are to stop oscillation or give the outputs an easier time when driven well into clipping.
Cheers Matt.
I know I will only get 3dB more but its the headroom that counts. My speakers are 90dB at 1W 1M 200W RMS and they can take this, kids fingers I have learnt are far more damaging😡. So I will probably only ever use an average of maybe 6W whilst playing fairly loud but the excursions will be up there in the 80W range occasionaly.
I will be using fet followers as in the 6l6 AB2 thread. So grid input capacitance is no prob.
Sorry for rambling, So to sum up with advantages first.
More power, although twice the power is only a max of 3dB in the real world.
A possibly better output transformer due to lower leakage inductance and parrasitic capacitance from the lower load impedance and hence less turns etc. etc.
Averaging of characteristics, although I can only see this being beneficial with vast amounts of output valves ie an OTL or some crazy multi KW regae rig. This leads to a good point by Kenpeter that the average may be a poor representation of the ideal. I can see it being an advantage for paralleling thousands of fets for a phono input stage to abtain a far lower noise, its of no use in an output stage. So on to disadvantages:
More heater current, this is no problem for me with my power transformers🙂 but it may cause hum problems due to increased electromagnetic radiation from the exposed higher current ac wiring.
More sockets, oh well🙂
The main problem I can envisage is finding reasonably matched sets of output valves or arranging a suitable bias arangement and the direct coupling of A/AB2. Maybe a seperate follower per valve with provision for adjustment of each valve which would require capacitor coupling to the fet and the additon of another time constant. Bias servo? even more sand, its turning into a house of valves on stilts of sand😉
So I guess the advantages/disadvantages sort of weigh up.
Tarzan, I think the 100R resistors are to stop oscillation or give the outputs an easier time when driven well into clipping.
Cheers Matt.
Sorry, an R in the PLATE circuit will only lose power, certainly NOT eliminate current hogging.
To solve that, you need a decent sized R in each CATHODE circuit, but again 100R will lose a lot of power.
Regards, Allen
Hi kenpeter,
the circuit is interesting! But can you please repeat how the worsening effect works when parallelling triodes, I don't completely understand the reasoning behind that. It confuses me because, in statistics, we can actually prove that the error (variance) on the transfer function decreases when parallelling devices. For example, one could make a precision resistor by parallelling many lower-precision parts. In the same vein, parallelling triodes should minimize inter-device deviations.
Kenneth
the circuit is interesting! But can you please repeat how the worsening effect works when parallelling triodes, I don't completely understand the reasoning behind that. It confuses me because, in statistics, we can actually prove that the error (variance) on the transfer function decreases when parallelling devices. For example, one could make a precision resistor by parallelling many lower-precision parts. In the same vein, parallelling triodes should minimize inter-device deviations.
Kenneth
I have lost track the GIFs of those functions I originally plotted.
But here's basically how I proved to myself that Smoking-Amp's
theory was functionally exact truth....
It started with a triode emulator that was basically a Schaded
IGBT stacked on top of a vacuum diode. It simply duplicated a
real diode curve, parallel, and with forced perfect Mu spacing.
This curve set looks way too ideal, not like a real triode at all.
Curves that do not "lean over to the right". And cut off much
too sharply. Don gave his insight why that might be (multiple
internal paths with slightly different character, especially the
cutoff) in the real thing.
So I plot two sets of ideal (but slightly different) curves, and
average as-if they were operating in parallel. Now it begins to
lean and cutoff more remotely! But took no less than four sets
of curves to average convincing approximation of a real 300B.
Adding more and more curves to this average begin to make
the resulting emulation less like the expensive planar triode,
and more the typical curve set of a cheaper construction.
This proved not to be a practical circuit for a triode emulator,
and I abandoned it in faver of Schaded FETs. But did leave me
with the strong feeling that Don was exactly right about what
makes real triode curves bend away from the ideal parallel Mu
as they do.
The more triodes in parallel that are not exact duplicates, the
more remote the averaged cutoff will become. Mu will bunch
together at the bottom. And curves at the right side of the
graph will be tilted at a more severe angle those at the left.
The noise killing might be useful if all vales were oriented a
bit differently? I think the Mu of all the other triodes fights
against an external noise when it is picked up by only one...
Again, thats maybe not relevant benefit to an output stage.
But here's basically how I proved to myself that Smoking-Amp's
theory was functionally exact truth....
It started with a triode emulator that was basically a Schaded
IGBT stacked on top of a vacuum diode. It simply duplicated a
real diode curve, parallel, and with forced perfect Mu spacing.
This curve set looks way too ideal, not like a real triode at all.
Curves that do not "lean over to the right". And cut off much
too sharply. Don gave his insight why that might be (multiple
internal paths with slightly different character, especially the
cutoff) in the real thing.
So I plot two sets of ideal (but slightly different) curves, and
average as-if they were operating in parallel. Now it begins to
lean and cutoff more remotely! But took no less than four sets
of curves to average convincing approximation of a real 300B.
Adding more and more curves to this average begin to make
the resulting emulation less like the expensive planar triode,
and more the typical curve set of a cheaper construction.
This proved not to be a practical circuit for a triode emulator,
and I abandoned it in faver of Schaded FETs. But did leave me
with the strong feeling that Don was exactly right about what
makes real triode curves bend away from the ideal parallel Mu
as they do.
The more triodes in parallel that are not exact duplicates, the
more remote the averaged cutoff will become. Mu will bunch
together at the bottom. And curves at the right side of the
graph will be tilted at a more severe angle those at the left.
The noise killing might be useful if all vales were oriented a
bit differently? I think the Mu of all the other triodes fights
against an external noise when it is picked up by only one...
Again, thats maybe not relevant benefit to an output stage.
Last edited:
Hey Kenpeter,
That is an interesting circuit! But personally I would use something like a 6DR7 or 6EM7 tube. What if you replace RLoad with a CCS or RCath with LES's? Ouch! Now my brain hurts.
Jim
That is an interesting circuit! But personally I would use something like a 6DR7 or 6EM7 tube. What if you replace RLoad with a CCS or RCath with LES's? Ouch! Now my brain hurts.
Jim
By paralleling output tubes (say 2 on each of the push pull sides) what you get is:
mu - stays the same
rp is halved
gm is doubled
Halving rp is VERY worthwhile in that it shifts the low frequency corner presented by the Output transformer Primary Inductance and tube rp lower. It also shifts the two (2) high frequency corner frequencies higher, 1 from rp and leakage inductance and 1 from rp and winding capacitance.
Doubling gm MAY be very useful also - many of the feedback schemes (those that degenerate gm) work by trading output tube gm for reduced rp. This helps as noted above BUT also helps to linearize the tubes. Those current equalization resistors mentioned in a post above should therefore be in the cathodes and NOT the anodes (such that they also degenerate the gm).
Make Sense?
Cheers,
Ian
mu - stays the same
rp is halved
gm is doubled
Halving rp is VERY worthwhile in that it shifts the low frequency corner presented by the Output transformer Primary Inductance and tube rp lower. It also shifts the two (2) high frequency corner frequencies higher, 1 from rp and leakage inductance and 1 from rp and winding capacitance.
Doubling gm MAY be very useful also - many of the feedback schemes (those that degenerate gm) work by trading output tube gm for reduced rp. This helps as noted above BUT also helps to linearize the tubes. Those current equalization resistors mentioned in a post above should therefore be in the cathodes and NOT the anodes (such that they also degenerate the gm).
Make Sense?
Cheers,
Ian
RLoad can be CCS or Choke no prob. RCath could be LED as long
as RLoad or CCSLoad sets a sensible quiescent current instead.
I think LED on bottom + choke on top would be begging for sand
runaway. Choice of triode boils down to quality, don't have to be
a high current monster.
In the NPN Darlington configuration, the emitter drop could be
a useful cathode bias. Though the emitter voltage drop is not a
constant, you can expect it to color the result somewhat...
In N-CH Darlington configuration, VGS-on gives more practical
Volts with which to bias the cathode. But now you some gate
capacitance phase shift in the Mu feedback loop. I think that
makes the N-CH MOSFET circuit a lot less desirable than use
an LED at the Cathode and PNP at the plate...
You need a bleeder resistor across VBE or VGS in either case...
Fixed resistance across a fixed voltage gives constant current.
There's not really enough working volts here for a complicated
CCS, the resistor alone does just fine.
as RLoad or CCSLoad sets a sensible quiescent current instead.
I think LED on bottom + choke on top would be begging for sand
runaway. Choice of triode boils down to quality, don't have to be
a high current monster.
In the NPN Darlington configuration, the emitter drop could be
a useful cathode bias. Though the emitter voltage drop is not a
constant, you can expect it to color the result somewhat...
In N-CH Darlington configuration, VGS-on gives more practical
Volts with which to bias the cathode. But now you some gate
capacitance phase shift in the Mu feedback loop. I think that
makes the N-CH MOSFET circuit a lot less desirable than use
an LED at the Cathode and PNP at the plate...
You need a bleeder resistor across VBE or VGS in either case...
Fixed resistance across a fixed voltage gives constant current.
There's not really enough working volts here for a complicated
CCS, the resistor alone does just fine.
Last edited:
- Status
- Not open for further replies.