Yes, this is a great thread. Hope my bane will benefit others heading in a similar direction. Everyone's statements about the ST70 are no doubt true, but just to clarify, I really don't have anything like the original:
I certainly don't claim this 5-20 amp with ST70 iron to be the end-all, but I do know my issues are a little past personal bias. There is not simply a roll-off of the highs (as stated earlier, the response actually begins to peak over 3K, not fall) and the sound is not something my ear needs to be trained to hear, it's downright disappointing. I am not new to transformer coupling, for that matter. My DAC/preamp is transformer coupled (looks similar to a Raven, actually), as is the DCX2496 feeding the subs. Using Gary Pimm's mods for that hardware, xfmr coupled in and out. Nice mod to get highest S/N out of the DCX2496.
Adding the transformer coupled output of the preamp really increased the detail and instrument presentation. Part of that could be from the 4.5:1 step-down ratio, lowering output impedance, but the amorphous Lundahl really came alive after 25 hours. So up to this point, my experience with transformer coupling has been very positive; there just seems to be something right about it. Possibly too much of a good thing, IDK.
Not complaining (though it probably sounds like it), just stating that I know something isn't right, beyond simple frequency response. If it really turns out to be the 12B4, I may just burn those cheap tubes in effigy. 1K Rp and they sound that bad? Have had some great results with the 6n6p also; I may compare those to the 7119.
Got a bunch of 10 ohm Mills for the cathodes; thanks for the suggestion. Wasn't sure if this was to assist in triode balance, or to keep oscillation to a minimum. I'll take the former with a WW.
Grid stoppers are always 1/2W carbon comps for me; I never deviate from that.
zigzagflux said:
I certainly don't claim this 5-20 amp with ST70 iron to be the end-all, but I do know my issues are a little past personal bias. There is not simply a roll-off of the highs (as stated earlier, the response actually begins to peak over 3K, not fall) and the sound is not something my ear needs to be trained to hear, it's downright disappointing. I am not new to transformer coupling, for that matter. My DAC/preamp is transformer coupled (looks similar to a Raven, actually), as is the DCX2496 feeding the subs. Using Gary Pimm's mods for that hardware, xfmr coupled in and out. Nice mod to get highest S/N out of the DCX2496.
Adding the transformer coupled output of the preamp really increased the detail and instrument presentation. Part of that could be from the 4.5:1 step-down ratio, lowering output impedance, but the amorphous Lundahl really came alive after 25 hours. So up to this point, my experience with transformer coupling has been very positive; there just seems to be something right about it. Possibly too much of a good thing, IDK.
Not complaining (though it probably sounds like it), just stating that I know something isn't right, beyond simple frequency response. If it really turns out to be the 12B4, I may just burn those cheap tubes in effigy. 1K Rp and they sound that bad? Have had some great results with the 6n6p also; I may compare those to the 7119.
Got a bunch of 10 ohm Mills for the cathodes; thanks for the suggestion. Wasn't sure if this was to assist in triode balance, or to keep oscillation to a minimum. I'll take the former with a WW.
Grid stoppers are always 1/2W carbon comps for me; I never deviate from that.
Okay, now we've got some good progress.
Swapped the 12B4 out with the 7119. I know it took me way too long to do this, but had too much to do around the holidays.
Finally obtained the sound quality I was expecting !! Soundstage is correct, actually deeper than the ST70. It's like everything has aligned itself correctly, and is coherent. Detail that was previously washed out is now displayed with pinpoint accuracy. Decay of voices and cymbals is clear; background is black.
Interestingly enough, the peaking over 10kHz is still there, and for as much as I can measure, the amp behaves identically by the numbers. But to the ears there is no comparison. Now, since the peaking is there, the highs are a little on the bright side, which contributes a little fatigue, but I will proceed to investigate zobels on the IT secondaries to try and smooth things out.
The phase shift is also still present, which doesn't excite me. But as far as I can tell at this point, it isn't ruining the presentation. Maybe I will try out a Lundahl IT at some point, but the mechanics of swapping these out in the chassis isn't trivial.
Gain is over 3 times as high, which I also need to address with the preamp span, but that can be fixed easy enough.
And all this with alligator clips and flying leads. Amazing what a tube swap can do. Fine recommendations, Bud and Lynn. You weren't kidding about that first stage.
I'll continue to keep the thread updated with progress on the remaining tweaks.
Swapped the 12B4 out with the 7119. I know it took me way too long to do this, but had too much to do around the holidays.
Finally obtained the sound quality I was expecting !! Soundstage is correct, actually deeper than the ST70. It's like everything has aligned itself correctly, and is coherent. Detail that was previously washed out is now displayed with pinpoint accuracy. Decay of voices and cymbals is clear; background is black.
Interestingly enough, the peaking over 10kHz is still there, and for as much as I can measure, the amp behaves identically by the numbers. But to the ears there is no comparison. Now, since the peaking is there, the highs are a little on the bright side, which contributes a little fatigue, but I will proceed to investigate zobels on the IT secondaries to try and smooth things out.
The phase shift is also still present, which doesn't excite me. But as far as I can tell at this point, it isn't ruining the presentation. Maybe I will try out a Lundahl IT at some point, but the mechanics of swapping these out in the chassis isn't trivial.
Gain is over 3 times as high, which I also need to address with the preamp span, but that can be fixed easy enough.
And all this with alligator clips and flying leads. Amazing what a tube swap can do. Fine recommendations, Bud and Lynn. You weren't kidding about that first stage.
I'll continue to keep the thread updated with progress on the remaining tweaks.
Very interesting, but very strange as well. It would seem that a 12B4 would be more than capable of driving an interstage, especially if the 7119 could do it, albeit with lower gain. Maybe, the 12B4's were matched well dynamically, or they didn't have the same distortion profiles.
Just guessing...
Just guessing...
My thoughts exactly; I'm very surprised the 12B4 had problems.
Messed around with zobels; only appears to be necessary on the first IT. So far using a 470 pF in series with a 20K resistor. Measured performance is quite flat, but I question the audible results. A little on the dry side maybe, but certainly less stringent and fatiging. I'll continue to play around with less compensation and adjust by ear. At least the test equipmement gives me some ballpark values to play around with.
Messed around with zobels; only appears to be necessary on the first IT. So far using a 470 pF in series with a 20K resistor. Measured performance is quite flat, but I question the audible results. A little on the dry side maybe, but certainly less stringent and fatiging. I'll continue to play around with less compensation and adjust by ear. At least the test equipmement gives me some ballpark values to play around with.
By now you probably have a healthy respect for just how transparent the driver and output stages really are, and how much coloration other amplifiers have in these same stages. If this topology has any real downsides, it's the ruthlessly unforgiving nature of the driver and output stages - there's nowhere for colorations to hide, and the demands on the input and preceding components are very severe. I wish it wasn't that way, but that's how it seems to work out.
As for RC-compensating the secondaries, don't forget to use Silver Mica small-value caps. These sound really good, and are ideal for RIAA equalization applications.
And for something really different, don't forget you can HF equalize the cathode resistors of the input tubes with a bypass shunt capacitor across the 10-ohm resistor. This "speeds up" the tube by dropping Rp at the plates. You can play with different values of R (increase the R up to 100 ohms if desired) and C on the cathode side and see what works best for you. Tune by adjusting square waves on the secondary of the interstage transformer.
As for RC-compensating the secondaries, don't forget to use Silver Mica small-value caps. These sound really good, and are ideal for RIAA equalization applications.
And for something really different, don't forget you can HF equalize the cathode resistors of the input tubes with a bypass shunt capacitor across the 10-ohm resistor. This "speeds up" the tube by dropping Rp at the plates. You can play with different values of R (increase the R up to 100 ohms if desired) and C on the cathode side and see what works best for you. Tune by adjusting square waves on the secondary of the interstage transformer.
Lynn:
Have played around with various mods, trying to obtain what I am looking for. Adding the zobel to the first IT secondary did smooth out the response, but I could never get an improvement in the sound, only in measurements. This was my experience with my transformer-output linestage; I could make square waves look better with zobels, but they never improved the sound.
Also tried my favorite triode, the 6n6p. I have branded this guy "the little tube that could." It does very well in a differential circuit driving a transformer (that's what my linestage consists of). Rp around 1.6K, I believe it was the Russian's copy of the 5687, from what I read. Anyway, at 150V and 20mA, I like what this tube can do for the first stage. My only aversion to its use is the fact my linestage also uses it, and identical tubes in successive stages, while not unheard of, isn't my preference. I have a suspicion whatever sonic signature (distortion) the first stage might add will only be made worse (multiplied ?) by its use in the next stage. Just conjecture. It has, no doubt, removed some of the stridency of the 7119, but the peaking response is still there nonetheless, and it can be heard clearly.
As a result, the subjective sound quality is still not what it could be. Side by side, I still prefer the Mullard 5-20 with ST70 iron, and this is simply unacceptable for me.
If I may post a paragraph from your Karna page that I think has some pertinence:
Being that I've done mostly everything I can without major mechanical modifications or reverting to 45's, I've focused again on the IT's. When taxes are completed and paid for (OUCH) I will buy some Lundahl iron like you suggested, though will probably start with the amorphous LL1692A. It was specifically requested by Kevin Carter for tubes in the Rp range we are dealing with.
So, I performed your Lissajous tests above on the first three transformers. I would like to mention that the test results are essentially identical regardless if the succeeding triodes are installed or not, so at least for these results, the Rp of the driver is not a big player (my opinion). Driving an open secondary was the same as driving an active grid. Sorry I don't have any cute color pictures to show; I can make screen captures with my Tek 2024B. If you want to see anything in particular, let me know and I'll get it.
First, the Tribute input splitter. Swept from 20Hz to 30kHz. Extremely straight line, no opening of the display at all, which indicates good phase angle match. Very much at a 45 degree angle, which would indicate good magnitude match. No real problems that I can see with the first stage. Again, tubes installed or not, same result.
Next, the first IT. Nice closed figure at 45 degrees until about 6kHz, where things just barely open up, and we start to deviate from 45 degrees. This is clearly seen at 10kHz, and is somewhat drastic at 20kHz (elliptical with large major axis and small minor axis). Looked at the behavior in the time domain to see what was going on. Beginning at 6k, one of the phases (call it W1) begins to have a magnitude less than the opposite phase (call it W2), which gets more obvious as frequency is increased. When looking at the sine waves, the magnitude differences are much easier to see than the phase differences, until about 22kHz, where W1 actually starts to head in a leading direction, moving from 180 degrees toward almost 90 degrees. At this point, the magnitude has risen again, so that at 24 kHz, the magnitudes of W1 and W2 are equal, but the phase shift is not 180, but 90 ! Further increases in frequency above 24 kHz cause W1 to increase in magnitude greater than W2 (just the opposite behavior) and the phase shift returns to 180 degrees, somewhere around 27 kHz. Curiously enough, further increases in frequency produce pretty decent phase match, but the magnitude match is quite poor. Maybe at this point I'm no longer in iron, but air territory.
Whew! That's a lot of words. I have no problem providing captures, but this is a pretty good description of what happens.
Then, the 2nd IT. Essentially similar behavior, but I would say it's somewhat improved in that the changes occur at extended frequencies, like the 2nd IT has greater bandwidth. We begin to open up the Lissajous pattern around 15kHz instead of 6kHz, and really start to open at 23kHz. The phase shifting of W1 occurs around 25 kHz, and we are widely elliptical at 30kHz.
I will be most interested to know if anyone can provide their experiences of how this type of behavior might sound. From what I understand, earlier vintage transformers had similar measurement, with various opinions of their sound. I'll also be interested to compare the measurements and sound of a Lundahl in its place. I know the LL1689AM/PP I use for the linestage output transformer had very tight and 45 degree patterns through 30kHz. The un-zobelled secondary is pretty darn flat, and so far, whatever limitations the equipment has is unknown, since it is limited by the other equipment in the signal path.
The search continues.
Have played around with various mods, trying to obtain what I am looking for. Adding the zobel to the first IT secondary did smooth out the response, but I could never get an improvement in the sound, only in measurements. This was my experience with my transformer-output linestage; I could make square waves look better with zobels, but they never improved the sound.
Also tried my favorite triode, the 6n6p. I have branded this guy "the little tube that could." It does very well in a differential circuit driving a transformer (that's what my linestage consists of). Rp around 1.6K, I believe it was the Russian's copy of the 5687, from what I read. Anyway, at 150V and 20mA, I like what this tube can do for the first stage. My only aversion to its use is the fact my linestage also uses it, and identical tubes in successive stages, while not unheard of, isn't my preference. I have a suspicion whatever sonic signature (distortion) the first stage might add will only be made worse (multiplied ?) by its use in the next stage. Just conjecture. It has, no doubt, removed some of the stridency of the 7119, but the peaking response is still there nonetheless, and it can be heard clearly.
As a result, the subjective sound quality is still not what it could be. Side by side, I still prefer the Mullard 5-20 with ST70 iron, and this is simply unacceptable for me.
If I may post a paragraph from your Karna page that I think has some pertinence:
Being that I've done mostly everything I can without major mechanical modifications or reverting to 45's, I've focused again on the IT's. When taxes are completed and paid for (OUCH) I will buy some Lundahl iron like you suggested, though will probably start with the amorphous LL1692A. It was specifically requested by Kevin Carter for tubes in the Rp range we are dealing with.
So, I performed your Lissajous tests above on the first three transformers. I would like to mention that the test results are essentially identical regardless if the succeeding triodes are installed or not, so at least for these results, the Rp of the driver is not a big player (my opinion). Driving an open secondary was the same as driving an active grid. Sorry I don't have any cute color pictures to show; I can make screen captures with my Tek 2024B. If you want to see anything in particular, let me know and I'll get it.
First, the Tribute input splitter. Swept from 20Hz to 30kHz. Extremely straight line, no opening of the display at all, which indicates good phase angle match. Very much at a 45 degree angle, which would indicate good magnitude match. No real problems that I can see with the first stage. Again, tubes installed or not, same result.
Next, the first IT. Nice closed figure at 45 degrees until about 6kHz, where things just barely open up, and we start to deviate from 45 degrees. This is clearly seen at 10kHz, and is somewhat drastic at 20kHz (elliptical with large major axis and small minor axis). Looked at the behavior in the time domain to see what was going on. Beginning at 6k, one of the phases (call it W1) begins to have a magnitude less than the opposite phase (call it W2), which gets more obvious as frequency is increased. When looking at the sine waves, the magnitude differences are much easier to see than the phase differences, until about 22kHz, where W1 actually starts to head in a leading direction, moving from 180 degrees toward almost 90 degrees. At this point, the magnitude has risen again, so that at 24 kHz, the magnitudes of W1 and W2 are equal, but the phase shift is not 180, but 90 ! Further increases in frequency above 24 kHz cause W1 to increase in magnitude greater than W2 (just the opposite behavior) and the phase shift returns to 180 degrees, somewhere around 27 kHz. Curiously enough, further increases in frequency produce pretty decent phase match, but the magnitude match is quite poor. Maybe at this point I'm no longer in iron, but air territory.
Whew! That's a lot of words. I have no problem providing captures, but this is a pretty good description of what happens.
Then, the 2nd IT. Essentially similar behavior, but I would say it's somewhat improved in that the changes occur at extended frequencies, like the 2nd IT has greater bandwidth. We begin to open up the Lissajous pattern around 15kHz instead of 6kHz, and really start to open at 23kHz. The phase shifting of W1 occurs around 25 kHz, and we are widely elliptical at 30kHz.
I will be most interested to know if anyone can provide their experiences of how this type of behavior might sound. From what I understand, earlier vintage transformers had similar measurement, with various opinions of their sound. I'll also be interested to compare the measurements and sound of a Lundahl in its place. I know the LL1689AM/PP I use for the linestage output transformer had very tight and 45 degree patterns through 30kHz. The un-zobelled secondary is pretty darn flat, and so far, whatever limitations the equipment has is unknown, since it is limited by the other equipment in the signal path.
The search continues.
zigzagflux said:
The distortion (of the following stage) is critically dependent on phase match. For example, a 3-degree spread in phase between the 2 secondaries can degrade even-order distortion by 10 dB - and a 3-degree phase shift is just barely visible on a scope. 90 degrees of phase shift between the two half-sections means that all PP distortion reduction is lost, along with a distortion profile that is almost certainly worse than a simple SE amplifier.
HF phase shift between the two halves of a PP amplifier is very undesirable, since the spectral content of the distortion becomes extremely frequency-sensitive - not to mention bizarre things like rapid phase rotation of the individual harmonics. (Remember, when the phases are re-summed, it's the inphase voltage vector that gives the magnitude of the audio signal, and the antiphase voltage vector that controls the phase and magnitude of the even-order harmonics. Small changes in these vectors result in large changes to the cancellation behavior and distortion residue.)
For this test, you want to drive the primary of the IT's with a symmetric drive - I assume you have set up your sound card so the L and R channels are in antiphase, and each channel is driving each half of the primary. Otherwise, you are unrealistically testing the IT as a SE -> PP phase-splitter, which is a much more difficult test. Input transformers can act as decent phase-splitters, but this is pretty difficult to impossible for IT's, which operate at higher levels and impedances.
Assuming your test protocol is valid (symmetric drive on the primaries), it would be worthwhile playing around with independent RC loading on the half-secondaries to see how closely they can be matched. The data you have for the first IT is not very promising, to be honest, and the second IT doesn't seem much better. The requirement - it is not an option - for the Karna and Amity amplifiers is a 3-degree phase match at 20 kHz (on the secondaries) with a matched, symmetric drive on the primaries.
I have measured the Lundahl LL1635 and saw tight (1 degree) phase match up to 50 kHz, with 90 degrees of phase shift occurring at 110~120 kHz. When used a phase splitter, though, even the LL1635 starts coming apart around 20~25 kHz. Matching the capacitances in an IT is very, very difficult, which is why I suggest trying different RC's on each half of the secondary as a stopgap measure.
The fact that the new amplifier is sounding worse than a Mullard 5-20 with Stereo 70 output transformers means that none of the benefits of the Amity/Karna architecture are being realized - it's like a BMW M3 accelerating slower and handling worse than a delivery truck. Something is very wrong with this picture.
The inherent distortion of the Karna is more than 10X lower than equivalent Mullard/Dynaco amplifiers, and although we have no distortion measurements, your description sounds like both the input and driver section are not operating the way they should, with high distortion, poor bandwidth, and limited headroom. With no distortion data and an unknown tube complement, I can only make rough guesses at where the problem lies.
Taking it piecemeal, which is always a good approach when debugging, feel free to hybridize the amplifier with the respective portions of the Mullard 5-20. What does it sound like with, for example, with a Mullard front end and driver and Karna output section? Or a Mullard front end and Karna driver and output section?
The most basic test of all would be a mildly revised Mullard with zero feedback, 300B outputs, and the existing ST70 output transformers (which match 300B's just fine). If that sounds worse, you have bad 300B's or maybe just aren't into the sound of triodes. If it sounds better - as it most certainly should - then the problem does indeed lie in the driver and input section.
you might also look into swapping which secondary is connected to which position. Also, are the IT's on the same side of the metal chassis plate as the tube bulbs and how close are they to those tube bulbs, if they are mounted on the same chassis face? These things will couple with everything that has an EM field and the tube innards appear to be the worst offender. You might also remove the metal endbells for a look into how they are reacting with your amps EM environment.
The problems you are addressing are antenna events in the coils. The core is out of the picture above 1 kHz, as far as participating in the transformation process. It does set minimum rise time limits above this frequency, but everything else is direct coupling and that is aided or not by the capacitance's it sees on both sides of the transformer.
Bud
The problems you are addressing are antenna events in the coils. The core is out of the picture above 1 kHz, as far as participating in the transformation process. It does set minimum rise time limits above this frequency, but everything else is direct coupling and that is aided or not by the capacitance's it sees on both sides of the transformer.
Bud
Thanks, Bud, for reposting GPimm's info about the IT's. That's a good point about the miswiring - it took several tries with the Lundahl's to figure out which set of wires went where. Fortunately, they don't need to be in-circuit to do the test with the signal source and the scope. Just attach the clip-leads and go for it.
The comment about the independent trimming of the RC loading was meant seriously; use the same R's, but independently trim the C (which only need to be a few Pf anyway) for each half of the secondaries. If I recall right, Peerless even disguised small balancing caps inside one of their famous PP output transformers just so they could be capacitively balanced at HF - so it's not like it hasn't been done before, and by famous Western Electric engineers at that.
Hmm - speaking of tricks with capacitors, anyone here tried the Lundahl capacitive bypass between primary and secondaries? That's what they recommend for one of their specialized ultrawideband interstages.
The comment about the independent trimming of the RC loading was meant seriously; use the same R's, but independently trim the C (which only need to be a few Pf anyway) for each half of the secondaries. If I recall right, Peerless even disguised small balancing caps inside one of their famous PP output transformers just so they could be capacitively balanced at HF - so it's not like it hasn't been done before, and by famous Western Electric engineers at that.
Hmm - speaking of tricks with capacitors, anyone here tried the Lundahl capacitive bypass between primary and secondaries? That's what they recommend for one of their specialized ultrawideband interstages.
A bit of information to digest and reply to; thanks for the suggestions. I will try to be thorough as possible in my responses to the critical points.
Hmmm. Well, I'm getting closer. I don't know if you mean on a YT scope trace, which phase shift I cannot see, or on a XY scope trace, which I can see easily, as I'll show in a bit.
Busted. How about this, in lieu of a sound card? I verified the input splitter again, both for phase shift and magnitude match. It was visually perfect past 80 kHz. So, assuming no issues there, I simply placed the 7119 back in circuit and drove the first IT actively with a function generator into the splitter. I suppose I could have some mu differences between the triodes, but any tube I place in the socket would have those differences anyway? Would this be a valid enough test for compensating the IT?
If so, some pictures of xy plots:
6 kHz reference:
10kHz below:
14kHz below:
18kHz below:
22kHz below:
Also attached is a pdf showing the relative voltage outputs of each half-secondary. I'm getting pretty close in phase and magnitude match, though still having some frequency response rise. As shown in the pdf file, one secondary is loaded with 470pF, 13.7K and the other with 390pF, 27.36K. I have a large assortment of micas, and for testing purposes used 50K pots for the loading resistors. I suspect I will need ANOTHER zobel across the entire secondary to smooth out the rising response, which will require re-tweaking of the current values.
So, does this mean I'm going about this wrong, or does the scope tell the story? Being that I used pots, I could perfect the selection, and it didn't look like identical R's permitted magnitude matching in the 20kHz region. The values above are what they are, and it wasn't all that simple to find the magic spot; either the phase gets all messed up, or the magnitudes are off. You mention a few pF (I envision 33pF is a 'few') and I have 470pF. Reasonable, or is something else going on?
Is it normal to use 3 zobels on each IT used? I rarely see this in published designs, but that's what I believe I will end up with. My assumption was other Karna designs out there don't have all these zobels being used, yet my response and balance (with the 7119) is disappointing without them. I just don't get it.
Thanks for your continued support, Bud. I don't understand this approach. What would change when feeding the opposite grid? Wiring is essentially suspended in air, so I would think circuit capacitances are equal lead to lead.
Per your recommendations of months back, I mounted the input splitter and IT's under the aluminum top plate. All tubes are above the plate, along with the output transformer. Filament transformer is on the opposite side of the amp, under the plate. When determining which xfmr goes where, I energized the filament xfmr in its predetermined corner. Then all other xfmrs were mounted and oriented for minimum hum pickup from this magnetic field generator. As fortune would have it, these locations were also providing negligible hum pickup from each other. That is, I would energize the second IT and verify low hum pickup in the other xfmrs. This was a very successful procedure for me, one of the few I can claim to work for me.
Comments on the data? Am I heading in the right direction?
Lynn Olson said:
Hmmm. Well, I'm getting closer. I don't know if you mean on a YT scope trace, which phase shift I cannot see, or on a XY scope trace, which I can see easily, as I'll show in a bit.
Lynn Olson said:
Busted. How about this, in lieu of a sound card? I verified the input splitter again, both for phase shift and magnitude match. It was visually perfect past 80 kHz. So, assuming no issues there, I simply placed the 7119 back in circuit and drove the first IT actively with a function generator into the splitter. I suppose I could have some mu differences between the triodes, but any tube I place in the socket would have those differences anyway? Would this be a valid enough test for compensating the IT?
If so, some pictures of xy plots:
6 kHz reference:
An externally hosted image should be here but it was not working when we last tested it.
10kHz below:
An externally hosted image should be here but it was not working when we last tested it.
14kHz below:
An externally hosted image should be here but it was not working when we last tested it.
18kHz below:
An externally hosted image should be here but it was not working when we last tested it.
22kHz below:
An externally hosted image should be here but it was not working when we last tested it.
Also attached is a pdf showing the relative voltage outputs of each half-secondary. I'm getting pretty close in phase and magnitude match, though still having some frequency response rise. As shown in the pdf file, one secondary is loaded with 470pF, 13.7K and the other with 390pF, 27.36K. I have a large assortment of micas, and for testing purposes used 50K pots for the loading resistors. I suspect I will need ANOTHER zobel across the entire secondary to smooth out the rising response, which will require re-tweaking of the current values.
Lynn Olson said:
So, does this mean I'm going about this wrong, or does the scope tell the story? Being that I used pots, I could perfect the selection, and it didn't look like identical R's permitted magnitude matching in the 20kHz region. The values above are what they are, and it wasn't all that simple to find the magic spot; either the phase gets all messed up, or the magnitudes are off. You mention a few pF (I envision 33pF is a 'few') and I have 470pF. Reasonable, or is something else going on?
Is it normal to use 3 zobels on each IT used? I rarely see this in published designs, but that's what I believe I will end up with. My assumption was other Karna designs out there don't have all these zobels being used, yet my response and balance (with the 7119) is disappointing without them. I just don't get it.
BudP said:
Thanks for your continued support, Bud. I don't understand this approach. What would change when feeding the opposite grid? Wiring is essentially suspended in air, so I would think circuit capacitances are equal lead to lead.
BudP said:
Per your recommendations of months back, I mounted the input splitter and IT's under the aluminum top plate. All tubes are above the plate, along with the output transformer. Filament transformer is on the opposite side of the amp, under the plate. When determining which xfmr goes where, I energized the filament xfmr in its predetermined corner. Then all other xfmrs were mounted and oriented for minimum hum pickup from this magnetic field generator. As fortune would have it, these locations were also providing negligible hum pickup from each other. That is, I would energize the second IT and verify low hum pickup in the other xfmrs. This was a very successful procedure for me, one of the few I can claim to work for me.
Comments on the data? Am I heading in the right direction?
Mounting as you have done does remove one significant potential source of problems, excellently done by the way.
The wiring comments were to make sure you explored the capacitance to ground issues that can also create problems with these devices. G Pimm found that there was a best way to hook these items up and while I did implement that in the layout drawings there is no guarantee that it is correct in your situation.
As an added interesting, item here is a gif from my main tutors last book. The book was written by Nat Grossner and is "Transformers for Electronic Circuits" 2nd edition ISBN # 0-07-024979-2 and this diagram is from page337 of that book. It shows the sort of layout found in your IT's, though yours are push pull and drive both sides of the secondary. Nat's comments on IT's and other transformers that drive essentially capacitive loads, is that they were a PITA to get right and he failed to understand why capacitors were not good enough for this job. Nat was, incidentally, the man behind the invention and commercial development of Met Glass, typically now referred to as amorphous core material.
As for continuing to support your efforts, that comes as a given for anyone who purchases my audio transformers and you are certainly welcome to what advice I can bring to the discussion.
Bud
The wiring comments were to make sure you explored the capacitance to ground issues that can also create problems with these devices. G Pimm found that there was a best way to hook these items up and while I did implement that in the layout drawings there is no guarantee that it is correct in your situation.
As an added interesting, item here is a gif from my main tutors last book. The book was written by Nat Grossner and is "Transformers for Electronic Circuits" 2nd edition ISBN # 0-07-024979-2 and this diagram is from page337 of that book. It shows the sort of layout found in your IT's, though yours are push pull and drive both sides of the secondary. Nat's comments on IT's and other transformers that drive essentially capacitive loads, is that they were a PITA to get right and he failed to understand why capacitors were not good enough for this job. Nat was, incidentally, the man behind the invention and commercial development of Met Glass, typically now referred to as amorphous core material.
As for continuing to support your efforts, that comes as a given for anyone who purchases my audio transformers and you are certainly welcome to what advice I can bring to the discussion.
Bud
Attachments
Thanks again, Bud. Not only you, but a number of people have submitted welcome and pertinent suggestions for improvement and resolution. All of them I take seriously, though I must admit I place you and Lynn at the top, since you both have the most direct experience with this specific design (obviously). I hope others will benefit from the wealth of information you provide, and should the 'easy' route not work out for me, I will pursue some of the other topologies. Kevinkr has always been available and ready to provide solid advice.
I must admit, my knowledge of transformers is limited to power, where size is measured in MVA (megavoltamperes), and winding configurations and capacitance are designed to fulfill BIL (Basic Impulse Level), for the purpose of surviving lightning strikes. Most of what I've learned about transformers came from my short stint at Waukesha Electric Systems, the largest American manufacturer of medium power transformers. A little different, and in my opinion, the audio transformer is more difficult to understand, design, build, and implement. In power systems, inductance is easily compensated for, faults are (mostly) easily detected and cleared, and harmonics are well understood. Measurement equipment never lies, and I am aware of no common issues that do not have a test designed to validate. In the audio realm, I am not so sure; how does one measure imaging, and is it correct or coloration? Why does one amp sound so different from another, when they measure so similarly?
I digress. Below is the final product of what I will implement on the secondary of the first interstage. When this is installed with proper mounting (no more alligator clips) I will pursue a similar approach with the second IT. Then, listening tests will begin. I must admit, the compensation network looks a little ridiculous, but the response is happily flat. Phase shift, viewed on the yt plot, is essentially impossible to notice up to 22k. Magnitude balance between the two phases is also beyond eyesight; having the RMS measurements available in the Tek scope are very handy for checking balance. (yes, the absolute accuracy is maybe 3%, but I have checked relative and repeatable accuracy, and that appears to be better than 1%)
Only by expanding the xy plot can the phase shift be seen. Very slight opening at 18k, very obvious at 22k, but the 45 degree angle is maintained, owing to magnitude balance. Crossing my fingers this is adequate for superior sound.
I must admit, my knowledge of transformers is limited to power, where size is measured in MVA (megavoltamperes), and winding configurations and capacitance are designed to fulfill BIL (Basic Impulse Level), for the purpose of surviving lightning strikes. Most of what I've learned about transformers came from my short stint at Waukesha Electric Systems, the largest American manufacturer of medium power transformers. A little different, and in my opinion, the audio transformer is more difficult to understand, design, build, and implement. In power systems, inductance is easily compensated for, faults are (mostly) easily detected and cleared, and harmonics are well understood. Measurement equipment never lies, and I am aware of no common issues that do not have a test designed to validate. In the audio realm, I am not so sure; how does one measure imaging, and is it correct or coloration? Why does one amp sound so different from another, when they measure so similarly?
I digress. Below is the final product of what I will implement on the secondary of the first interstage. When this is installed with proper mounting (no more alligator clips) I will pursue a similar approach with the second IT. Then, listening tests will begin. I must admit, the compensation network looks a little ridiculous, but the response is happily flat. Phase shift, viewed on the yt plot, is essentially impossible to notice up to 22k. Magnitude balance between the two phases is also beyond eyesight; having the RMS measurements available in the Tek scope are very handy for checking balance. (yes, the absolute accuracy is maybe 3%, but I have checked relative and repeatable accuracy, and that appears to be better than 1%)
Only by expanding the xy plot can the phase shift be seen. Very slight opening at 18k, very obvious at 22k, but the 45 degree angle is maintained, owing to magnitude balance. Crossing my fingers this is adequate for superior sound.
An externally hosted image should be here but it was not working when we last tested it.
Balance is way more important than overall frequency response, which I put fairly far down on the list unless the 20 kHz region is screwed up. But phase balance is super-important to distortion performance for any PP design - it's those vectors, y'know.
Audio transformers are cool devices and a great way to break ground loops, but stray capacitances are everywhere. And not usually symmetric. A simple example is power transformers. The amount of hum you get in a system depends on the phasing of the primary - AC power is delivered as hot and neutral (not balanced) and the winding of the primary is NOT symmetric with respect to capacitance to ground. Thus, one side of the primary has high capacitance to ground (and this should be connected to the neutral), and the other has much lower capacitance to ground (and this should be connected to hot). But, big joke, most power transformers have no markings which primary lead is low-capacitance and which is high-capacitance, so it's purely a matter of chance how the power transformers in several different bits of hifi equipment get connected in production. Start to see the problem?
Something as seemingly innocent as a small amount of stray C in a power transformer results in HF hash from the power line getting coupled to the chassis, and the odd phenomenon of an isolated chassis floating up to surprising voltages (when measured with a high-impedance DVM). When the multiple chassis are interconnected via the RCA cabling, the voltage differences almost disappears, but now currents flow in the ground circuits and you get hard-to-track-down hum and buzzing sounds. All of this trouble thanks to rather small values of stray C from primary to core to chassis.
So ... getting around to audio transformers, do NOT assume capacitances to core and ground are symmetric. It takes extraordinary measures to achieve this. This is why the markings on the primary and secondary leads are important. The leads with the most capacitance belong on the ground side, and the ones with the least on the high-voltage side. Woe betide anyone that gets them mixed up. Capacitance meters are very, very useful for finding out which leads do what on transformers.
The worst case scenario is getting one half-secondary right, and the other half-secondary wrong. With isolated secondaries you can do this, you know, and it screws up performance unbelievably bad. I've done this myself on the very symmetric Lundahl's - they measure really weird when this happens. Lots of opportunities for getting it wrong when you have ten leads coming out of a little mystery box - primaries, secondaries, electrostatic screen, and case and core ground. Cores in particular should never be allowed to float - for safety reasons alone, as well as having the transformer perform as designed.
Once you start seeing the world in terms of stray capacitance - a few pF here, a few pF there, everywhere a pF pF - it starts looking like a RF-centric world. Unlike the RF guys, though, we care (a lot) about distortion and noise, so there are additional criteria we have to meet. What comes in handy are accounting for all of the stray C's in the tubes and transformers, and having symmetric point-to-point wiring to minimize capacitance where we don't want it.
Audio transformers are cool devices and a great way to break ground loops, but stray capacitances are everywhere. And not usually symmetric. A simple example is power transformers. The amount of hum you get in a system depends on the phasing of the primary - AC power is delivered as hot and neutral (not balanced) and the winding of the primary is NOT symmetric with respect to capacitance to ground. Thus, one side of the primary has high capacitance to ground (and this should be connected to the neutral), and the other has much lower capacitance to ground (and this should be connected to hot). But, big joke, most power transformers have no markings which primary lead is low-capacitance and which is high-capacitance, so it's purely a matter of chance how the power transformers in several different bits of hifi equipment get connected in production. Start to see the problem?
Something as seemingly innocent as a small amount of stray C in a power transformer results in HF hash from the power line getting coupled to the chassis, and the odd phenomenon of an isolated chassis floating up to surprising voltages (when measured with a high-impedance DVM). When the multiple chassis are interconnected via the RCA cabling, the voltage differences almost disappears, but now currents flow in the ground circuits and you get hard-to-track-down hum and buzzing sounds. All of this trouble thanks to rather small values of stray C from primary to core to chassis.
So ... getting around to audio transformers, do NOT assume capacitances to core and ground are symmetric. It takes extraordinary measures to achieve this. This is why the markings on the primary and secondary leads are important. The leads with the most capacitance belong on the ground side, and the ones with the least on the high-voltage side. Woe betide anyone that gets them mixed up. Capacitance meters are very, very useful for finding out which leads do what on transformers.
The worst case scenario is getting one half-secondary right, and the other half-secondary wrong. With isolated secondaries you can do this, you know, and it screws up performance unbelievably bad. I've done this myself on the very symmetric Lundahl's - they measure really weird when this happens. Lots of opportunities for getting it wrong when you have ten leads coming out of a little mystery box - primaries, secondaries, electrostatic screen, and case and core ground. Cores in particular should never be allowed to float - for safety reasons alone, as well as having the transformer perform as designed.
Once you start seeing the world in terms of stray capacitance - a few pF here, a few pF there, everywhere a pF pF - it starts looking like a RF-centric world. Unlike the RF guys, though, we care (a lot) about distortion and noise, so there are additional criteria we have to meet. What comes in handy are accounting for all of the stray C's in the tubes and transformers, and having symmetric point-to-point wiring to minimize capacitance where we don't want it.
Okay, that makes sense. I thought the suggestion was to swap the leads feeding the grids, which would also reverse the polarity. Wasn't following that idea. But I follow what you are saying about which end of the winding gets 'grounded'. I will investigate.
Bud, I followed your drawing, which seemed to indicate the 5 and 6 (red, blue) leads would be the grounded ends. Have you the ability to predict which one would be more likely based on construction?
Bud, I followed your drawing, which seemed to indicate the 5 and 6 (red, blue) leads would be the grounded ends. Have you the ability to predict which one would be more likely based on construction?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.