Paper on UL Operation with Cathode Feedback
Gents,
The best modern paper I found on UL (with cathode feedback) is at Ayumi's Lab. You will need google translate to read it (even if it is not very good), fortunately, there are plenty of equations and graphs to go along with the write-up, so with some effort it is possible to get the gist of the paper... I have not read it since I found the link
I hope you experts will be able to parse and extract some good information from it to share with us all. 
Jaz
Gents,
The best modern paper I found on UL (with cathode feedback) is at Ayumi's Lab. You will need google translate to read it (even if it is not very good), fortunately, there are plenty of equations and graphs to go along with the write-up, so with some effort it is possible to get the gist of the paper... I have not read it since I found the link
I hope you experts will be able to parse and extract some good information from it to share with us all. Jaz
Johan - correct.
I think of it this way: The CFB windings are, as you point out, effectively in series with the plate and screen windings. So if one starts with a standard UL configuration and adds CFB to it, then by subtracting the number of cathode winding turns from both the plate and screen windings, the inter-electrode potentials within the tube would remain the same, and the effective loading on the tube also remains the same. Effectively, all we have done is rearrange the windings to shift the g1 drive signal AC reference potential from the output tube cathode to a different point in the output circuit - a point at a hypothetical tap on the primary winding at CFB%. So the basic open-loop performance of the output stage hasn't changed, but we have added local NFB at the cathodes, reducing both distortion and output impedance.
Based on this understanding, my approach is first to optimize the UL% so as to start from the best open-loop operating point, and then add the desired amount of CFB by shifting the winding percentages. If the open-loop performance is already optimized, there is no need to do extensive calculations on the CFB configuration - the problem becomes a simple matter of deciding how much g1 drive voltage can be provided vs. how much CFB is desired.
So for example, let's say we start with an ordinary UL transformer having N turns per side in the plate winding, and UL taps at 0.25*N. Now if we want to add 10% CFB while keeping the UL action the same (same Eg2-k conditions and the same effective loading on the tube), then the new transformer should have 0.1*N CFB windings, and 0.9*N plate windings with UL taps at 0.15*N turns.
Jaz - looks like a good write-up. I'll dig into it over the weekend.
I think of it this way: The CFB windings are, as you point out, effectively in series with the plate and screen windings. So if one starts with a standard UL configuration and adds CFB to it, then by subtracting the number of cathode winding turns from both the plate and screen windings, the inter-electrode potentials within the tube would remain the same, and the effective loading on the tube also remains the same. Effectively, all we have done is rearrange the windings to shift the g1 drive signal AC reference potential from the output tube cathode to a different point in the output circuit - a point at a hypothetical tap on the primary winding at CFB%. So the basic open-loop performance of the output stage hasn't changed, but we have added local NFB at the cathodes, reducing both distortion and output impedance.
Based on this understanding, my approach is first to optimize the UL% so as to start from the best open-loop operating point, and then add the desired amount of CFB by shifting the winding percentages. If the open-loop performance is already optimized, there is no need to do extensive calculations on the CFB configuration - the problem becomes a simple matter of deciding how much g1 drive voltage can be provided vs. how much CFB is desired.
So for example, let's say we start with an ordinary UL transformer having N turns per side in the plate winding, and UL taps at 0.25*N. Now if we want to add 10% CFB while keeping the UL action the same (same Eg2-k conditions and the same effective loading on the tube), then the new transformer should have 0.1*N CFB windings, and 0.9*N plate windings with UL taps at 0.15*N turns.
Jaz - looks like a good write-up. I'll dig into it over the weekend.
The easy answer: UL sounds like a linear amplifier should sound
Kindly forgive that one ...
I have seen Mr Moers' paper (have it here somewhere). I can recall one difficulty I had when first perusing it, and I am going to do a most wreckless thing here and that is state a difficulty I have before I have properly studied the paper again (it is rather lengthy). Possibly others sharper than myself can correct me.
Let me state at the outset that it is the best analytical paper I have encountered on the subject - no problem there. But from the outset the argument goes about the "middle" between the convex pentode and concave triode curves relating Ia and Va. That is fine - but amplifier linearity goes about the Vg1 vs Ia relationship. That has little to do with Ia -Va relationship per se; it has to do with where different such curves for different Vg1s disect the load line. There is still a non-linear relationship between Vg1 and Ia for the different screen taps.
Here I must stop for now; toward the end of the paper something referring to this appears, and my best simplistic conclusions come from the previously mentioned KT88 GEC graphs for distortion vs. output and the given graphs for the EL34 in this Moers paper, also available from Mullard data sheets. Only as said, I was confused as to why the initial emphasis on the linearity of the Ia-Va relationship, without involving Vg1 from the start.
Over to other worthy members to broaden my horizons.
Kindly forgive that one ...I have seen Mr Moers' paper (have it here somewhere). I can recall one difficulty I had when first perusing it, and I am going to do a most wreckless thing here and that is state a difficulty I have before I have properly studied the paper again (it is rather lengthy). Possibly others sharper than myself can correct me.
Let me state at the outset that it is the best analytical paper I have encountered on the subject - no problem there. But from the outset the argument goes about the "middle" between the convex pentode and concave triode curves relating Ia and Va. That is fine - but amplifier linearity goes about the Vg1 vs Ia relationship. That has little to do with Ia -Va relationship per se; it has to do with where different such curves for different Vg1s disect the load line. There is still a non-linear relationship between Vg1 and Ia for the different screen taps.
Here I must stop for now; toward the end of the paper something referring to this appears, and my best simplistic conclusions come from the previously mentioned KT88 GEC graphs for distortion vs. output and the given graphs for the EL34 in this Moers paper, also available from Mullard data sheets. Only as said, I was confused as to why the initial emphasis on the linearity of the Ia-Va relationship, without involving Vg1 from the start.
Over to other worthy members to broaden my horizons.
Last edited:
Modification of a non-linear function by means of non-linear feedback, if you like.
Johan - I have the same disagreement with Mr. Moers' analytic approach, and Ayumi's similar approach (although I do think both are insightful articles). The load-line linearity is indeed the primary function of interest, with Vg1 vs. Ia relationship dominating, and Va vs. Ia becoming more significant at lower rp. What a straightened-out Va vs. Ia curve helps with is linearity into reactive loads (elliptical load line).
However, my main difficulty is that these analytic approaches don't predict distortion accurately - only the basic performance parameters. In particular, what I really want to see in a push-pull amplifier is the composite curves for the two tubes together at the selected bias point. The poorly modeled low- Ia region is quite important, especially at lower bias currents. The best tool I've found is the TCJ push-pull calculator, but it only models triodes. I suppose one could coax LTSPICE into plotting such curves, but I am somewhat dubious of model accuracy at low current, and where g2 current is concerned (and g1 current too, for that matter). Norman Koren's models would be my starting point if attempting this. Then the problem becomes one of getting good input data and parameter fitting to accurately model Ia characteristic vs. both g1 and g2, plus the grid currents, which starts to smell like it might involve a bunch of measurement work. I'm not sure the datasheet curves have quite enough info.
So at this point in my exploration of the literature, I am coming to the conclusion that there are too many variables and non-linearities involved to simulate accurately without a good deal of effort to create better models. For this reason, I am heading the empirical route, to directly measure the push-pull stage. This is still going to be quite a lot more work than I was initially planning when I started the project. Hopefully it will be worthwhile.
However, my main difficulty is that these analytic approaches don't predict distortion accurately - only the basic performance parameters. In particular, what I really want to see in a push-pull amplifier is the composite curves for the two tubes together at the selected bias point. The poorly modeled low- Ia region is quite important, especially at lower bias currents. The best tool I've found is the TCJ push-pull calculator, but it only models triodes. I suppose one could coax LTSPICE into plotting such curves, but I am somewhat dubious of model accuracy at low current, and where g2 current is concerned (and g1 current too, for that matter). Norman Koren's models would be my starting point if attempting this. Then the problem becomes one of getting good input data and parameter fitting to accurately model Ia characteristic vs. both g1 and g2, plus the grid currents, which starts to smell like it might involve a bunch of measurement work. I'm not sure the datasheet curves have quite enough info.
So at this point in my exploration of the literature, I am coming to the conclusion that there are too many variables and non-linearities involved to simulate accurately without a good deal of effort to create better models. For this reason, I am heading the empirical route, to directly measure the push-pull stage. This is still going to be quite a lot more work than I was initially planning when I started the project. Hopefully it will be worthwhile.
Howdy, All: Been away on work pressures and fixing to do it again but wanted to take a few minutes to join in. I must say that this is one of, if not the most interesting, threads for me in the site!
Merlinb, I guess I'm letting myself be too rigorous and exact. After dealing for years with all kinds of servo systems where a sample of the output is compared to the signal and the difference applied as feedback to correct imprecisions, that being true negative feedback, calling the application of a signal to one element where that is not being compared to the input or used to generate an error signal, negative feedback, is kind of hard on me. That modification of the operating point via G2 connection is a mechanism for linearizing the stage to which it is applied, no doubt about it!
I think the path Chad is taking and some of the insights to which Johan points make up the right attitude and illuminate the path. The empirical approach, once theory gets us to the vicinity, can be quite practical and very valid as far as I am concerned. Plenty of projects over the years have been successfully concluded with some empiricism applied judiciously, following theory
OK, gotta get ready for my day trip tomorrow, looking forward to more!
See Y'all later!
Rene
Merlinb, I guess I'm letting myself be too rigorous and exact. After dealing for years with all kinds of servo systems where a sample of the output is compared to the signal and the difference applied as feedback to correct imprecisions, that being true negative feedback, calling the application of a signal to one element where that is not being compared to the input or used to generate an error signal, negative feedback, is kind of hard on me. That modification of the operating point via G2 connection is a mechanism for linearizing the stage to which it is applied, no doubt about it!
I think the path Chad is taking and some of the insights to which Johan points make up the right attitude and illuminate the path. The empirical approach, once theory gets us to the vicinity, can be quite practical and very valid as far as I am concerned. Plenty of projects over the years have been successfully concluded with some empiricism applied judiciously, following theory
OK, gotta get ready for my day trip tomorrow, looking forward to more!
See Y'all later!
Rene
Could you please clarify the issue a bit more? Is there something we can do to improve the pentode model and/or equivalent circuit?
That will be great to see more measurement data, looking forward to your results. Ayumi's simulation results were matched up pretty closely to his measurement results even though they were for SE operation only. So perhaps those models are still of some use for predicting distortion, or for parameter optimization, etc.
Anyway, I used Ayumi's pctube program to plot the UL composite curves and came up with an estimate for the THD, which looks reasonably close to the available data (but do correct me if I am wrong...)
EL34 UL Composite Characteristic with UL Tap=43%, Eg1=-25V
The grid curves indicate the input voltage swing, the blue line is the composite load line, and the green lines are the individual tubes' load line:
An externally hosted image should be here but it was not working when we last tested it.
Estimated THD:
An externally hosted image should be here but it was not working when we last tested it.
Jaz
Last edited:
Hi Jaz:
A straight line Va - Ia characteristic (constant Vg1) has no direct bearing on distortion. What is needed is a straight line Vg1 - Ia characteristic. So, although the plate curves may look impressive for UL compared to the convex pentode or concave triode curves, this means relatively little for distortion.
On the matter of simulation, I have found that there is some difficulty in creating a good SPICE model which is faithful to the real tube behaviour. Years ago, I spent some time with a program (I forget the name) which could overlay a graph of simulated curves onto an image from a datasheet. You could adjust some model parameters and try to make a SPICE model with curves that accurately match the datasheet curves. But there are two problems:
1) the datasheet curves are not always very accurate, having been drawn by hand from tabulated data, long before computers were available, and
2) even assuming the original graph was accurate, it can be rather difficult to match the curves - one parameter adjustment might improve the match at low Ia, but cause some unwanted curvature at high Ia, or vice-versa. I found the models had insufficient parameters to produce an exact match, and that pentode curves were harder to match than triode curves.
So it is difficult for me to believe the models are reliable enough. Maybe at 10% distortion they are good enough, but around 1% distortion and less, I don't think so. When it comes to distortion, we care a lot about the subtle shape of the curves ... the second- and third- order accuracy, not just the coarse features which are adequate to determine basic operating point and performance parameters. We also care about the grid current, which is non-linear and will affect the drive voltage and therefore distortion, if the source has non-zero impedance. Modeling additionally the g2 characteristic as required for UL operation, only adds yet another layer of complexity to this, and more opportunity to introduce errors into the model's prediction of reality.
Norman Koren's models are a step in the right direction, but I think one still needs better quality input data for this task. Garbage in = garbage out, as they say. So, this is why I think it smells like a bunch of measurement effort. If one doesn't trust the original hand-drawn graphs to be accurate enough, then it's time to build a computerized curve tracer. After measuring curves, then it's time to make the accurate models (with enough parameters), and then build a simulation and analyze the results... I think the effort to do all of this would be more productively put towards directly measuring the distortion of a test circuit. Perhaps I do not give enough credit to the available SPICE models. Anyway, it's hard to tell if the models correlate well with reality unless you have good empirical data to compare the simulation against.
So, that's why I say, time to get some more measurement data!
A straight line Va - Ia characteristic (constant Vg1) has no direct bearing on distortion. What is needed is a straight line Vg1 - Ia characteristic. So, although the plate curves may look impressive for UL compared to the convex pentode or concave triode curves, this means relatively little for distortion.
On the matter of simulation, I have found that there is some difficulty in creating a good SPICE model which is faithful to the real tube behaviour. Years ago, I spent some time with a program (I forget the name) which could overlay a graph of simulated curves onto an image from a datasheet. You could adjust some model parameters and try to make a SPICE model with curves that accurately match the datasheet curves. But there are two problems:
1) the datasheet curves are not always very accurate, having been drawn by hand from tabulated data, long before computers were available, and
2) even assuming the original graph was accurate, it can be rather difficult to match the curves - one parameter adjustment might improve the match at low Ia, but cause some unwanted curvature at high Ia, or vice-versa. I found the models had insufficient parameters to produce an exact match, and that pentode curves were harder to match than triode curves.
So it is difficult for me to believe the models are reliable enough. Maybe at 10% distortion they are good enough, but around 1% distortion and less, I don't think so. When it comes to distortion, we care a lot about the subtle shape of the curves ... the second- and third- order accuracy, not just the coarse features which are adequate to determine basic operating point and performance parameters. We also care about the grid current, which is non-linear and will affect the drive voltage and therefore distortion, if the source has non-zero impedance. Modeling additionally the g2 characteristic as required for UL operation, only adds yet another layer of complexity to this, and more opportunity to introduce errors into the model's prediction of reality.
Norman Koren's models are a step in the right direction, but I think one still needs better quality input data for this task. Garbage in = garbage out, as they say. So, this is why I think it smells like a bunch of measurement effort. If one doesn't trust the original hand-drawn graphs to be accurate enough, then it's time to build a computerized curve tracer. After measuring curves, then it's time to make the accurate models (with enough parameters), and then build a simulation and analyze the results... I think the effort to do all of this would be more productively put towards directly measuring the distortion of a test circuit. Perhaps I do not give enough credit to the available SPICE models. Anyway, it's hard to tell if the models correlate well with reality unless you have good empirical data to compare the simulation against.
So, that's why I say, time to get some more measurement data!
Thanks for the explanation, I can appreciate the "garbage in = garbage out" comment, many of SPICE models unfortunately fall into that category, even the Koren pentode models do not model the control grid currents particularly well.
But as far as the measurement data are concerned, Langford-Smith, Mullard and GEC have published quite a bit of them back in the day, are those not useful for your application? If not, what's missing from them?
But as far as the measurement data are concerned, Langford-Smith, Mullard and GEC have published quite a bit of them back in the day, are those not useful for your application? If not, what's missing from them?
My vote is to let Chad off the posting hook so he can produce the experimentally derived data and models! Can't wait to see those! Chad, I hope you do wind up including the MOSFET drive G2 version of UL in your experiment matrix. If all is as my primitive analyses have shown, the only real drawback to that implementation will be lower power compared to the tapped transformer since the Screen power will not be added to the output power. But that should, on the average, be no more than a few percent and at least for my goal, unimportant. I won't care if my amplifier is 40W instead of 50W, for example.
Thanks to all! Now, back to work, did not get back from my day trip yesterday until late.
Y'all take care!
Rene
Can't wait to see those! Chad, I hope you do wind up including the MOSFET drive G2 version of UL in your experiment matrix. If all is as my primitive analyses have shown, the only real drawback to that implementation will be lower power compared to the tapped transformer since the Screen power will not be added to the output power. But that should, on the average, be no more than a few percent and at least for my goal, unimportant. I won't care if my amplifier is 40W instead of 50W, for example.Thanks to all! Now, back to work, did not get back from my day trip yesterday until late.
Y'all take care!
Rene
Alright, well I've not been getting warm fuzzy feedback from the transformer winders on the tapped teriary idea - it's apparently just too many taps to be realistic, and would cost and arm and a leg to have it built. To bring the cost down, the number of taps would have to be cut down, significantly limiting it's usefulness. It did not go unnoticed in my reading, that Langford-Smith & Chesterman used a tapped autoformer for their experiments, so I already had an awareness that the complicated transformer idea may not be feasible.
Plan B was similar to the historical test setups - a tapped autoformer, but with a standard OPT in parallel. Perhaps not ideal from a distortion perspective, as the two cores will draw different excitation currents and won't be tightly coupled, but perhaps good enough to be usable within a limited bandwidth. What would really be ideal though, is something more like a variac - continuously variable. Hmmm...
Now, this may or may not be a good idea, but we'll soon find out. A 3-phase variac is on it's way. The idea is to put two of the variacs in series, connected such that the tie point becomes a "center tap", and the wipers will track symmetrically outward, becoming the screen grid "taps". The third variac could be used for position measurement - could even enable servo drive for complete automation (not sure I'm that ambitious, but we'll see first how the initial testing goes). The "CT" and both ends of this arrangement will be AC coupled to the primary of a standard OPT. I went with a 400Hz model on the (possibly incorrect) presumption that it might have better coupling and bandwidth than a 50 or 60Hz version. The trade-off is that the core size and inductance will be limited, so I'll have to avoid low frequency operation.
Any glaring (or subtle) errors with this logic?
Plan B was similar to the historical test setups - a tapped autoformer, but with a standard OPT in parallel. Perhaps not ideal from a distortion perspective, as the two cores will draw different excitation currents and won't be tightly coupled, but perhaps good enough to be usable within a limited bandwidth. What would really be ideal though, is something more like a variac - continuously variable. Hmmm...
Now, this may or may not be a good idea, but we'll soon find out. A 3-phase variac is on it's way. The idea is to put two of the variacs in series, connected such that the tie point becomes a "center tap", and the wipers will track symmetrically outward, becoming the screen grid "taps". The third variac could be used for position measurement - could even enable servo drive for complete automation (not sure I'm that ambitious, but we'll see first how the initial testing goes). The "CT" and both ends of this arrangement will be AC coupled to the primary of a standard OPT. I went with a 400Hz model on the (possibly incorrect) presumption that it might have better coupling and bandwidth than a 50 or 60Hz version. The trade-off is that the core size and inductance will be limited, so I'll have to avoid low frequency operation.
Any glaring (or subtle) errors with this logic?
Attachments
![Screen Shot 2013-06-03 at Mon, 6-3 [ww23] 11.44.30 AM.jpg](/community/data/attachments/323/323510-b97c56f890fbd22d9793dc0124a936dd.jpg)
I'm afraid the variacs will have less inductance than your OPT, thus ruining the low frequencies. And if you connect the center of the variacs to B+ use a cap. For this to succeed I think you need the smallest type of variac, or a high-volage type which may not exist.
Success.
Steven
Maybe use a high-resistance pot with a follower, but you need a higher supply voltage.
Success.
Steven
Maybe use a high-resistance pot with a follower, but you need a higher supply voltage.
Last edited:
Anyone tried this Sower OPT? While not continuously variable, it does provide 3 possible screen tap configurations.
From the post, Chad is already aware of the low frequency limit, presumably that means that those low frequencies will be avoided at high power levels. What is the voltage rating of the variac?
If one assumes an induction level at rated voltage at 400Hz of, say, 16000 Gauss (keeping in mind that these are designed for power, not audio, and tend to be pushed harder for that reason) then the actual induction level at actual voltages/frequencies can be easily scaled, with the added proviso that this level is additionally predicated on what voltage is applied to what percentage of the windings.
Back to the dual transformer idea, the fact that less than perfect results will be obtained for all the reasons given (though I'm not that concerned about excitation power, it should be relatively low in a well designed transformer and goes down linearly as frequency increases, but not so core power loss which is resistive in nature, it is non linear with frequency though it should still go down with frequency all other things being equal). Perhaps with the dual transformer (or variac/transformer) you can still obtain the right shape curves and expect better actual numbers with a purpose built transformer.
Not sure what AC coupling is going to buy, but I must be forgetting something. Perhaps earlier in the thread you must have said that you wanted a different level DC supply for the screens?
The real potential problem I see with the variac idea is that there is no intentional coupling between variac sections at all so the OPT is going to provide that function. The side being driven toward cutoff will couple its above B+ voltage to the one end of the corresponding variac and that in turn couple to the screen, and vice versa. Is this the way you see it? If the volt-second product thus applied to the variac is appreciable then its excitation current might indeed be high. There will actually be some cancellation of flux on the side swinging high (picture the DC flowing in one direction, the AC in the opposite) but the currents are additive on the low side. Running at 1KHz will let you about double the AC end to end rating of the variac, though the core losses will then go up.
I say just fire it up and see what happens!
Good luck!
Rene
If one assumes an induction level at rated voltage at 400Hz of, say, 16000 Gauss (keeping in mind that these are designed for power, not audio, and tend to be pushed harder for that reason) then the actual induction level at actual voltages/frequencies can be easily scaled, with the added proviso that this level is additionally predicated on what voltage is applied to what percentage of the windings.
Back to the dual transformer idea, the fact that less than perfect results will be obtained for all the reasons given (though I'm not that concerned about excitation power, it should be relatively low in a well designed transformer and goes down linearly as frequency increases, but not so core power loss which is resistive in nature, it is non linear with frequency though it should still go down with frequency all other things being equal). Perhaps with the dual transformer (or variac/transformer) you can still obtain the right shape curves and expect better actual numbers with a purpose built transformer.
Not sure what AC coupling is going to buy, but I must be forgetting something. Perhaps earlier in the thread you must have said that you wanted a different level DC supply for the screens?
The real potential problem I see with the variac idea is that there is no intentional coupling between variac sections at all so the OPT is going to provide that function. The side being driven toward cutoff will couple its above B+ voltage to the one end of the corresponding variac and that in turn couple to the screen, and vice versa. Is this the way you see it? If the volt-second product thus applied to the variac is appreciable then its excitation current might indeed be high. There will actually be some cancellation of flux on the side swinging high (picture the DC flowing in one direction, the AC in the opposite) but the currents are additive on the low side. Running at 1KHz will let you about double the AC end to end rating of the variac, though the core losses will then go up.
I say just fire it up and see what happens!
Good luck!
Rene
These particular variacs are rated 120V @ 350Hz, but the 117% tap at the end of the winding is brought out to it's own terminal. So it's easier to think of them as rated for 140V. A series pair will thus handle 280Vrms, which is pretty reasonable.
AC coupling indeed merely buys a lower screen bias, which is one of the main things I want to investigate.
Interestingly, the manufacturer states the 400Hz version can also be run at 60V & 60Hz, and the specs show that the 60Hz and 400Hz versions also have the same number of turns. Considering the core sizes look to be roughly 2:1, the inductance of the 400Hz version should actually be about half that of the 60Hz version, which is better than one might expect just from the nameplate specs. However, I'm really not sure how much inductance these will have - I am guessing around 1 or 2H. As long as it's adequate and the core losses aren't too bad, I'm optimistic the setup will do what's needed.
So, the first thing I'll be doing when these arrive is measure their inductance and look at their BH characteristic to see what they can handle. I may have to live with the fact that this setup won't be able to measure more than modest power levels - so be it. The first couple of watts are the ones that matter most from a psycho-acoustic standpoint, while distortion figures at power levels closer to clipping are considerably less meaningful. In any case, the historical literature clearly establishes that UL configurations tend to maintain good distortion performance right up until they run out of swing. So, I'm prepared to settle for distortion optimization at low to moderate power levels, and let the chips fall where they may near clipping.
The lack of coupling between the two variacs is one of my major concerns. I'm actually more worried about stability issues when / if the tubes approach class B territory - another reason to keep the power levels low. I think I'll just have to try it and see.
I've got some setup work to do before this is going to be up and running. Actually, I've got to do some cleanup in the lab and make enough space for this first!
AC coupling indeed merely buys a lower screen bias, which is one of the main things I want to investigate.
Interestingly, the manufacturer states the 400Hz version can also be run at 60V & 60Hz, and the specs show that the 60Hz and 400Hz versions also have the same number of turns. Considering the core sizes look to be roughly 2:1, the inductance of the 400Hz version should actually be about half that of the 60Hz version, which is better than one might expect just from the nameplate specs. However, I'm really not sure how much inductance these will have - I am guessing around 1 or 2H. As long as it's adequate and the core losses aren't too bad, I'm optimistic the setup will do what's needed.
So, the first thing I'll be doing when these arrive is measure their inductance and look at their BH characteristic to see what they can handle. I may have to live with the fact that this setup won't be able to measure more than modest power levels - so be it. The first couple of watts are the ones that matter most from a psycho-acoustic standpoint, while distortion figures at power levels closer to clipping are considerably less meaningful. In any case, the historical literature clearly establishes that UL configurations tend to maintain good distortion performance right up until they run out of swing. So, I'm prepared to settle for distortion optimization at low to moderate power levels, and let the chips fall where they may near clipping.
The lack of coupling between the two variacs is one of my major concerns. I'm actually more worried about stability issues when / if the tubes approach class B territory - another reason to keep the power levels low. I think I'll just have to try it and see.
I've got some setup work to do before this is going to be up and running. Actually, I've got to do some cleanup in the lab and make enough space for this first!
Well, keep in mind that while inductance is proportional, all else being equal, to the area of the core, it is inversely proportional to the length of the magnetic path, again all else being the same. How are the cores different, that is between the true 60Hz version and the one you are getting? Also, the fact that you can run your units at a disproportionately high voltage when reducing the frequency by a factor of 6 tells me that the core may not be made up of thin laminations appropriate for 400Hz work (I always specified .004" lams for 400Hz) so the limit is more for core losses and the margin before saturation is then higher for your units than if they were true 400Hz cores. Don't know which way that will work for you but at least core losses are more resistive than reactive.
Can't wait until you do fire up things!
Rene
Can't wait until you do fire up things!
Rene
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.