I’ve searched the forum, and haven't found an adequate answer to this question.
What are the downsides of running parallel output tubes in a SE amp?
I have ran across some comments here, and elsewhere that suggest it isn’t a good idea, but I haven’t seen a good explanation as to why. Is it the complexity of getting two tubes to operate at the same point (matching), or is there more to it than that?
Thanx,
Casey
What are the downsides of running parallel output tubes in a SE amp?
I have ran across some comments here, and elsewhere that suggest it isn’t a good idea, but I haven’t seen a good explanation as to why. Is it the complexity of getting two tubes to operate at the same point (matching), or is there more to it than that?
Thanx,
Casey
Who knows? I have tried matching to within a percent or so but only at the operating point. Didn't help. The sound loses focus and clarity. Sounds even worse than paralleling input tubes. I don't think a convincing explanation exists.
I also fail to see how the logic most people give holds. When you look at tubes like a 45, you have two plates and two grids on either side of a VV filament.
Since the tube characteristics have to do with the various element spacings a slight shift of any one of these structures will give you two different tubes in parallel within the same envelope, one could argue they could never be identical yet they never seem to blur
if you look at the way the current is drawn form the plate, you will actually see a "shadow" of the filament with diffuse edges on the plate structure. The fact that the plate pattern is not a line that mirrors the filament suggests that different distances are traveled by various electrons which again nets you varying tube characteristics (ie multiple tubes in parallel) within the same envelope.
Whether you have multiple tubes in parallel in a single envelope, or in multiple envelops, you still end up with a single set of average chatacteristics, but seeing the distince multiple structures sure might give the brain a nudge that blurring of the sound is happening 🙂
dave
dave
Since the tube characteristics have to do with the various element spacings a slight shift of any one of these structures will give you two different tubes in parallel within the same envelope, one could argue they could never be identical yet they never seem to blur
if you look at the way the current is drawn form the plate, you will actually see a "shadow" of the filament with diffuse edges on the plate structure. The fact that the plate pattern is not a line that mirrors the filament suggests that different distances are traveled by various electrons which again nets you varying tube characteristics (ie multiple tubes in parallel) within the same envelope.
Whether you have multiple tubes in parallel in a single envelope, or in multiple envelops, you still end up with a single set of average chatacteristics, but seeing the distince multiple structures sure might give the brain a nudge that blurring of the sound is happening 🙂
dave
dave
That old chestnut.....
Yes, there are many different views and not many facts.
I have, and will, argue that if parallel SE causes loss of "definition" then so should class A push-pull - or any other class of push pull (within the region where both valves are conducting).
Yes, there are many different views and not many facts.
I have, and will, argue that if parallel SE causes loss of "definition" then so should class A push-pull - or any other class of push pull (within the region where both valves are conducting).
Re: That old chestnut.....
I'll go one step more and suggest that if parallel tubes do indeed blur due to their differences, the PP should be the worst offender, since you are not only dealing with mismatched devices, but you have to add the mismatched driving circuits in parallel, and the mismatched iron that "adds" the parallel signals together 🙂
dave
dhaen said:
I'll go one step more and suggest that if parallel tubes do indeed blur due to their differences, the PP should be the worst offender, since you are not only dealing with mismatched devices, but you have to add the mismatched driving circuits in parallel, and the mismatched iron that "adds" the parallel signals together 🙂
dave
Konnichiwa,
I am stomped for any specific explanation, but here is some experience.
I once has a 300B PSE Amplifer. It used seperate cathode resistors and capacitors for biasing (maybe a "wrong" approach) and had also seperate grid coupling capacitors and grid leak & grid stopper resistors (actually, the whole thing was switchable PP & PSE, but that as they sayis another story, I preferred PSE over PP).
When I replaced my 90ish db/W/m speakers with some 104db/W/m speakers I decided to see what would happen if I pulled one output valve and merely re-connected the 8 Ohm Speaker to the 4 Ohm Tap, the AMp was cruising at a few 100mW into that load.
This pretty much keeps all variables equal to a resonable degree.
What I found was that with some music the difference between SE and PSE was small enough to be counted under "uncertain if there was any". BUT with some music and some recordings SE had a much greater degree of resolution, a much greater degree of "see-troughness" and "thereness".
That said, I did NOT find the same effect to anywhere near the same degree with a Kondo (ANJ) Kegon, which in comparison to my own Amp uses fixed bias common in value to both valves with individual cathode followers per valve, but we did not test as much. I still personally felt however that overall the Single SE Shinri sounded a little purer.
Equally, the Arthur Loesch preamp circuit that circulated on the Joe-List. This offered two options of line-stage, one with a PSE 6922/6DJ8 and the other with a single section of 5687. Here the variables become very divergent, but having build both versions more than once again I felt the PSE version to sound a little more opaque in terms of resolution, compared to the "single SE".
I cannot offer any reasons why this SHOULD be so, but that is what I have observed (BTW, paralleling power devices, including IC Poweramp's also seems to cause audible problems, but they are more various).
Draw your own conclusions and best do your own testing and listen to what happens in PSE vs SE (and perhaps also PP), it is IMHO a quite instructive and illuminating thing to do.
Sayonara
valveitude said:
I am stomped for any specific explanation, but here is some experience.
I once has a 300B PSE Amplifer. It used seperate cathode resistors and capacitors for biasing (maybe a "wrong" approach) and had also seperate grid coupling capacitors and grid leak & grid stopper resistors (actually, the whole thing was switchable PP & PSE, but that as they sayis another story, I preferred PSE over PP).
When I replaced my 90ish db/W/m speakers with some 104db/W/m speakers I decided to see what would happen if I pulled one output valve and merely re-connected the 8 Ohm Speaker to the 4 Ohm Tap, the AMp was cruising at a few 100mW into that load.
This pretty much keeps all variables equal to a resonable degree.
What I found was that with some music the difference between SE and PSE was small enough to be counted under "uncertain if there was any". BUT with some music and some recordings SE had a much greater degree of resolution, a much greater degree of "see-troughness" and "thereness".
That said, I did NOT find the same effect to anywhere near the same degree with a Kondo (ANJ) Kegon, which in comparison to my own Amp uses fixed bias common in value to both valves with individual cathode followers per valve, but we did not test as much. I still personally felt however that overall the Single SE Shinri sounded a little purer.
Equally, the Arthur Loesch preamp circuit that circulated on the Joe-List. This offered two options of line-stage, one with a PSE 6922/6DJ8 and the other with a single section of 5687. Here the variables become very divergent, but having build both versions more than once again I felt the PSE version to sound a little more opaque in terms of resolution, compared to the "single SE".
I cannot offer any reasons why this SHOULD be so, but that is what I have observed (BTW, paralleling power devices, including IC Poweramp's also seems to cause audible problems, but they are more various).
Draw your own conclusions and best do your own testing and listen to what happens in PSE vs SE (and perhaps also PP), it is IMHO a quite instructive and illuminating thing to do.
Sayonara
Yhats what I was beginning to suspect
Differing views can be a very good thing, but after a while , if no body of evidence shows up, I have to wonder about there validity.
The SE/PP debate that still rages on is a good example. At first the SE crowd had nothing to go on but their ears, and were being told that it was nothing more than the "euphonic" effect of adding distortion. At first blush, this SEEMED to hold water, then the "hard" evidence started to mount. The best treatment I have read to date as to why our ears seems to contradict what our test equipment tells us regarding SE is….
http://www.next-power.net/next-tube/articles/Cheever/cheever.pdf
…it’s a real eye (ear) opener. I HIGHLY recommend it.
Since the parallel/ single debate has been going on for about as long, and nothing comparable in the way of evidence has shown up, I tend to believe it is an article of faith in the “Simpler is ALWAYS Better” doctrine.
Casey
Differing views can be a very good thing, but after a while , if no body of evidence shows up, I have to wonder about there validity.
The SE/PP debate that still rages on is a good example. At first the SE crowd had nothing to go on but their ears, and were being told that it was nothing more than the "euphonic" effect of adding distortion. At first blush, this SEEMED to hold water, then the "hard" evidence started to mount. The best treatment I have read to date as to why our ears seems to contradict what our test equipment tells us regarding SE is….
http://www.next-power.net/next-tube/articles/Cheever/cheever.pdf
…it’s a real eye (ear) opener. I HIGHLY recommend it.
Since the parallel/ single debate has been going on for about as long, and nothing comparable in the way of evidence has shown up, I tend to believe it is an article of faith in the “Simpler is ALWAYS Better” doctrine.
Casey
i was just visiting a friend and he had a 4 pillar 45 sitting right there. Upon close inspection it was clear that the grid/filament structure was not even close to being centered inside the plate stucture.
talk about two dissimilar tubes in parallel.
dave
talk about two dissimilar tubes in parallel.
dave
That's why I prefer planar triodes and also single plate tubes such as EC88 and EC8010 at low levels...
Honestly, though, if you parallel two tubes, you end up with a significantly different animal, so of course it will not sound the same.
On the other hand, I've read stuff where the author dislikes parallel tube sections but recommends parallelling plate resistors for audible improvement!
John
jlsem said:
the argument can also be made for the circular plate/grid/filament structures like many of the eimac tubes. Rather than two distinct tubes, there becomes an infinite number of gradually varying tubes.
but the fact that the "new tube" which has 2X the Gm always has the problems seems a bit fishy. plus, why does addign another plate to the EC 8010 more than quadruple its value 🙂
dave
Yo Dave,
Here's my take - connecting 2 electrode systems together forces the elements to be at the same potential, which gives you one composite device. No blurring.
Where this model breaks down is that there is inductance/capacitance associated with those connections, which enables 1 device to fight the other til they eventually settle into steady state [do we ever get there with music?].
In a single envelope, those parasitics are minimised. 2 separate tubes, more lead length.
The way to test this theory is to take a dual triode and parallel the halves up with short and very long leads? Or even just insert an RF choke.
Just my theory anyway
Here's my take - connecting 2 electrode systems together forces the elements to be at the same potential, which gives you one composite device. No blurring.
Where this model breaks down is that there is inductance/capacitance associated with those connections, which enables 1 device to fight the other til they eventually settle into steady state [do we ever get there with music?].
In a single envelope, those parasitics are minimised. 2 separate tubes, more lead length.
The way to test this theory is to take a dual triode and parallel the halves up with short and very long leads? Or even just insert an RF choke.
Just my theory anyway
Hi all
There is more, Casey. And there is a precise physical reason.
It is not so. Paralleling tubes does not give us tubes with averaged
characteristics but a different animal. I'll try to explain why.
Suppose to have two tubes (triodes) which perfectly adhere to the Child's law
but with different gain, for instance.
Now, immagine the tubes being completely paralleled. Suppose one tube
has a gain of 3 and the other a gain of 4, for example, and suppose
that the currents of the two triodes are the same for vg= 0 Volt.
Then, for a given value of grid voltage, say -20 Volt, the tubes
will have different currents and, if we change the anode to cathode
voltage, different anodic curves.
If we make an orizzontal average of the curves we end with a
curve with average characteristic, that is average gain in
this case.
But the action of paralleling tubes make a vertical average, since
the total current is the sum of the current of each tube.
so we end with a curve with the following equation
i=k[V+m1*vg]^1.5+ k[V+m2*vg]^1.5
and this is not the Child law.
But..., maybe it is a better one..not so.
It will result clear from the following graph.
You can see the curve corresponding to
m=3 and m=4 ( in blue)
the orizzontal average (green) that has m=3.5
and the the vertical average (red)
which appears very distorted in the small current region.
in this region, in fact, the mean is done betweeen only
one curve and the zero line (because one tube is OFF).
So, paralleled tubes have anodic curves different from
single tube. In particular, the curves result more
distorted in the region of small currents.
Note that it is a geometrical effect, it is not important
that the law is the Child's one. Only in case of
a linear device the orizzontal and vertical average will coincide,
and only at high currents.
I don't know if it is the cause of sound difference nor if this is a good explanation. surely, it is a possible explanation.
bye
Federico
There is more, Casey. And there is a precise physical reason.
It is not so. Paralleling tubes does not give us tubes with averaged
characteristics but a different animal. I'll try to explain why.
Suppose to have two tubes (triodes) which perfectly adhere to the Child's law
but with different gain, for instance.
Now, immagine the tubes being completely paralleled. Suppose one tube
has a gain of 3 and the other a gain of 4, for example, and suppose
that the currents of the two triodes are the same for vg= 0 Volt.
Then, for a given value of grid voltage, say -20 Volt, the tubes
will have different currents and, if we change the anode to cathode
voltage, different anodic curves.
If we make an orizzontal average of the curves we end with a
curve with average characteristic, that is average gain in
this case.
But the action of paralleling tubes make a vertical average, since
the total current is the sum of the current of each tube.
so we end with a curve with the following equation
i=k[V+m1*vg]^1.5+ k[V+m2*vg]^1.5
and this is not the Child law.
But..., maybe it is a better one..not so.
It will result clear from the following graph.
You can see the curve corresponding to
m=3 and m=4 ( in blue)
the orizzontal average (green) that has m=3.5
and the the vertical average (red)
which appears very distorted in the small current region.
in this region, in fact, the mean is done betweeen only
one curve and the zero line (because one tube is OFF).
So, paralleled tubes have anodic curves different from
single tube. In particular, the curves result more
distorted in the region of small currents.
Note that it is a geometrical effect, it is not important
that the law is the Child's one. Only in case of
a linear device the orizzontal and vertical average will coincide,
and only at high currents.
I don't know if it is the cause of sound difference nor if this is a good explanation. surely, it is a possible explanation.
bye
Federico
Attachments
CV said:
seems like a good distinction between the two.
fscarpa58 said:
I agree with your assessment, but think you miss the idea that this same thing happens with in a single tube also.
Lets stick with the planar model of a tube and consider the Single plate 2A3. I chose this tube because the harp shaped filament most closely resembles a planar emitting surface. This gives us a single plane cathode with two individual grid and plate planes on either side which can be considered two tubes in parallel. Even if we assume the elements are parallel, its a pretty fair guess that the distances will have some difference between them which will translate to different gains for each plate-grid structure.
I fail to see how this conceptually is any different than two discrete tubes in parallel and if on the surface the concept is not consistent, then any conclusions drawn from the concept seem suspect to me.
If we look at the "single" tube construction and consider what happens under a number of situations the "large number of tubes in parallel" view becomes more clear.
Lets stick with the single plate 2A3 and see what happens if we remove one half of the plate structure. Since the gain is determined by the spacing and all we did was cut the plate area in half, the gain will remain unchanged. The reduction of the plate area does change the Gm (cuts it in half) and if we hold the mu constant and halve the Gm we get double the Rp. This remains consistent with what one would expect with the parallel connection of tubes.
If we keep the parallel plane model with the same grid and plate structure and spacings and switch from a harp filament to a single V filament, we end up with a tube with the same gain, but a smaller area of the plate is used so Gm will go down. Everyone seems ot think that the harp filament of the 2A3 was done to attempt to mimmic the planar cathode surface, but I wonder if it was done to make full use of the plate area allowing the 700 ohm Rp from a tube that would have a 1500 ohm Rp with a VV filament.
If we move to a tube like a 45 with a VV filament and pretend one of the V's does not light again we get a different tube. Since the Gm is related to the area of the plate and its distance from the cathode, when you remove 1/2 the cathode you essentially remove the plate area adjacent to it from the picture and again end up with 1/2 the Gm. Now imagine what happens if we take this tube with 1/2 of the filament lit and remove the plate adjacent to it leaving the plate by the unused filament connected. The plate area remains the same, but the distance from cathode to plate has increased so again the Gm goes down AND the cathode-grid-plate spacings have also changed so the mu will be different.
I have a PDF of a 40 page article by Kusunose on triode design, and while the math gets a bit deep, when looking at it it quickly becomes clear that the design of a triode can be looked at as the averaging of a number of different triodes in parallel. To then enclose all of those averages in a single envelope and assume what is inside is 100% the mean rather than the average of a bunch of different dimensions (tubes) is conceptually flawed imo.
If anybody wants to look at the kusunose, I'll put it up. (its an 8 meg pdf)
dave
Dave said:
How does that stand up?
That was my immediate thought too. Then I realised that those single tubes produced where the linearity was bad at the bottom end, would have been rejected at test.
How does that stand up?
It may have also been an (successful) attempt at reducing distortion. In directly-heated triodes, distortion decreases as the diameter of the filament wire decreases. A row of many very thin filaments results in lower distortion while supplying copious emission necessary for a power tube.
John
Hi Dave, All
You are completely right, Dave. And no, I do not miss
that idea. I left it for this post...
Have you never asked to yourself why, in the small current
region, the anodic curves of the triode diverge from the
child law? The cause is, for the main part, the
phenomenon described by you.
Many are the reasons, for the most technological, that produce
differences between predicted and actual behavior of a triode:
the not uniform distribution of temperature of the cathode, chiefly in
direct heated tubes, the less than perfectly uniform step of the grid
spiral, in signal triodes, and in general, all those differences
between ideal and real geometry that cause a one-dimensional
idealized structure, like the Child’s one, to diverge from the complex
three-dimensional reality of a triode.
Now, think at the electronic flow (or the current) existing
between anode and cathode as the sum of elementary flows
standing in zones characterized by slightly different values of
the properties (say gain, for instance). The total current will
result from the integral of these contributions.
The effect is of the same kind, but more smoothed, of that
described in my previous post (one can see it in the attached
picture).
The models proposed during the years by some researchers
represent an attempt to mathematically describe the phenomenon.
Unfortunately they (Koren's one, etc.) are not based on physical considerations so often resulting in a great number of parameters.
Bye
Federico
Fitting of Child’s low (6SN7 curves from
http://www.audiomatica.com/tubes/6sn7.htm )
You are completely right, Dave. And no, I do not miss
that idea. I left it for this post...
Have you never asked to yourself why, in the small current
region, the anodic curves of the triode diverge from the
child law? The cause is, for the main part, the
phenomenon described by you.
Many are the reasons, for the most technological, that produce
differences between predicted and actual behavior of a triode:
the not uniform distribution of temperature of the cathode, chiefly in
direct heated tubes, the less than perfectly uniform step of the grid
spiral, in signal triodes, and in general, all those differences
between ideal and real geometry that cause a one-dimensional
idealized structure, like the Child’s one, to diverge from the complex
three-dimensional reality of a triode.
Now, think at the electronic flow (or the current) existing
between anode and cathode as the sum of elementary flows
standing in zones characterized by slightly different values of
the properties (say gain, for instance). The total current will
result from the integral of these contributions.
The effect is of the same kind, but more smoothed, of that
described in my previous post (one can see it in the attached
picture).
The models proposed during the years by some researchers
represent an attempt to mathematically describe the phenomenon.
Unfortunately they (Koren's one, etc.) are not based on physical considerations so often resulting in a great number of parameters.
Bye
Federico
Fitting of Child’s low (6SN7 curves from
http://www.audiomatica.com/tubes/6sn7.htm )
Attachments
So trace curves in parallel...
I have been following this post with attention to try to understand a bit more about PSE vs SE.
If Federico is right, the decrement in sound performance of PSE vs similarly designed SE versions using the same tube is a matter of erroneous calculation of the parameters for operating thre tubes.
Following the same rationale, using a curve tracer to obtain the combined behavior of paralleled tubes, will produce a different Vp vs Ip graph for every Vg. Using that graph to obtain the operating conditions of the paralleled tubes instead of the graph from the datasheet (for one tube), would allow a better estimation of the circuit behavior.
Am I right? Do those expensive commercial PSE amps are designed in such fasion?
Cheers,
Rada.
I have been following this post with attention to try to understand a bit more about PSE vs SE.
If Federico is right, the decrement in sound performance of PSE vs similarly designed SE versions using the same tube is a matter of erroneous calculation of the parameters for operating thre tubes.
Following the same rationale, using a curve tracer to obtain the combined behavior of paralleled tubes, will produce a different Vp vs Ip graph for every Vg. Using that graph to obtain the operating conditions of the paralleled tubes instead of the graph from the datasheet (for one tube), would allow a better estimation of the circuit behavior.
Am I right? Do those expensive commercial PSE amps are designed in such fasion?
Cheers,
Rada.
I have seen this explanation posted before, can't recall where or by whom, but sounds good to me.
For a single tube, the design is sufficiently "coherent" ie. same geometry thru-out, that the sound is OK. For a small number of tubes in parallel, the chances are that manufacturing alignment differences between the tubes will produce a distorted summation (the different Mu effect, etc), but for a large # of tubes in parallel, the variations average out unless one has a truly "bad" tube in the lot. So on this hypothesis, one should be able to parallel a small number of tubes successfully if one takes considerable care in matching their characteristics on a curve tracer.
Even the difference between well liked tube types and poor sounding tube types may be explainable by similar reasoning. Good sounding tubes tend to have a nice flat Mu curve versus plate current and plate voltage. Tubes with badly curving Mu curves are composed of non consistent geometry sections acting like different tube types mixed together.
I think to get a nice flat Mu curve, the geometry needs to be such that the grid falls on an equipotential between the plate and cathode. The grid wires need to be sufficiently close together to not cause shadows or "inselbildung" or "island effect", on the cathode. The cathode also needs to be of constant geometry/ temperature etc.
Hi Dave,
Is that "kusunose" article on your web site? I would like to read it. There are also some tube design articles here:
http://frank.pocnet.net/other/RCA/VTD_RCA/index.html
http://frank.pocnet.net/other/RCA/ETD62/index.html
Don
For a single tube, the design is sufficiently "coherent" ie. same geometry thru-out, that the sound is OK. For a small number of tubes in parallel, the chances are that manufacturing alignment differences between the tubes will produce a distorted summation (the different Mu effect, etc), but for a large # of tubes in parallel, the variations average out unless one has a truly "bad" tube in the lot. So on this hypothesis, one should be able to parallel a small number of tubes successfully if one takes considerable care in matching their characteristics on a curve tracer.
Even the difference between well liked tube types and poor sounding tube types may be explainable by similar reasoning. Good sounding tubes tend to have a nice flat Mu curve versus plate current and plate voltage. Tubes with badly curving Mu curves are composed of non consistent geometry sections acting like different tube types mixed together.
I think to get a nice flat Mu curve, the geometry needs to be such that the grid falls on an equipotential between the plate and cathode. The grid wires need to be sufficiently close together to not cause shadows or "inselbildung" or "island effect", on the cathode. The cathode also needs to be of constant geometry/ temperature etc.
Hi Dave,
Is that "kusunose" article on your web site? I would like to read it. There are also some tube design articles here:
http://frank.pocnet.net/other/RCA/VTD_RCA/index.html
http://frank.pocnet.net/other/RCA/ETD62/index.html
Don
- Status
- Not open for further replies.