I have extensively used power toroidals as output and input transformers. The choice of which way to wire the primaries has proven absolutely critical. Wired in one direction and bass is rolled off and treble peaks badly at 60-70khz, wired the otherway both these issues fade away.
A scope and decent signal generator are absolutely essential to setting up an amp using large amounts of iron.
Shoog
A scope and decent signal generator are absolutely essential to setting up an amp using large amounts of iron.
Shoog
Any competent transformer designer will always put the high-capacitance side of the winding on the low-voltage side of the circuit. By low-voltage I mean low AC voltage swing, not the DC potential. Symmetric capacitances are very, very hard to do, and only found in the highest-performance fully balanced devices.
Where misunderstanding arises is between the transformer engineer and the circuit engineer; the best old-school guys were quite aware of the shortcomings of the parts they were using, and would cleverly design the circuit around them. Tek scopes, for example, were always designed around the actual physics of the tubes, caps, transformers, and electron-beams of the CRT's, and not the claims of the vendors. You see the same approach with Bell Labs - who actually invented negative feedback and were very aware of its shortcomings in audio applications.
Many modern engineers, though, have fallen in love with oversimplified computer models and an opamp-cookbook approach to design. This is where a lot of glib assumptions about the behavior of parts arises - it's just simple ignorance, really, masked by gee-whizzery of fancy computer displays. Very much akin to the mathematical architects of the credit-derivatives disaster - some of the econometric models (designed by Ph.D's from the best schools) had no allowable input variables for house prices ever decreasing!
The real world is our friend; we have to talk to the guys that design the parts we use and know how to measure what's going on. As the world has found to its cost, with tens of trillions of dollars up in smoke and credit markets in disarray, models can be very misleading. Always measure and try to understand the physics of the device.
Where misunderstanding arises is between the transformer engineer and the circuit engineer; the best old-school guys were quite aware of the shortcomings of the parts they were using, and would cleverly design the circuit around them. Tek scopes, for example, were always designed around the actual physics of the tubes, caps, transformers, and electron-beams of the CRT's, and not the claims of the vendors. You see the same approach with Bell Labs - who actually invented negative feedback and were very aware of its shortcomings in audio applications.
Many modern engineers, though, have fallen in love with oversimplified computer models and an opamp-cookbook approach to design. This is where a lot of glib assumptions about the behavior of parts arises - it's just simple ignorance, really, masked by gee-whizzery of fancy computer displays. Very much akin to the mathematical architects of the credit-derivatives disaster - some of the econometric models (designed by Ph.D's from the best schools) had no allowable input variables for house prices ever decreasing!
The real world is our friend; we have to talk to the guys that design the parts we use and know how to measure what's going on. As the world has found to its cost, with tens of trillions of dollars up in smoke and credit markets in disarray, models can be very misleading. Always measure and try to understand the physics of the device.
Been over 5 weeks since I have accumulated enough valid information to post. I have made some good progress, developed additional issues/questions, but all part of the learning experience. First, I will try to address most of the previous 4 pages of posts.
Each stage has been converted over to essentially an identical topology of what is posted on Lynn's web site, with the 40uF motor run + teflon bypass from B+ to cathode. Didn't see any huge difference in measured performance, things still looked relatively poor.
Input stage: tried a number of tubes, 12B4 (lowest Rp), 7119, and 6n6p. When looking at Lissajous patterns, there was for the most part no difference in performance between the tubes as viewed on the scope. Still had that annoying phase shift at frequencies less than 20k, and the phase shift wasn't symmetrical either. Extensive experimentation with zobels to try and improve this was successful to a degree (see my prior posts), but heavy loading was necessary, which didn't give me much confidence.
Purchased a Lundahl LL1692A, which was designed for Rp ranging from 1.5K to 5K. This transformer was specified as a 1.75:2 stepup, but I wanted to use it in a stepdown configuration for better drive. In accordance with others' experience, this transformer absolutely performed differently depending on how the windings were arranged. Tried 16 different configurations to find the best performer. Measurements indicated phase/magnitude balance from excellent to ridiculous. Moral of the story; experiment and measure prior to installing. FWIW, I was not seeing any difference in performance with the ONetics when I changed winding configurations. Installed the LL1692A in place of the first IT, feeding off the 6n6p. Measured phase shift and balance were significantly improved, without any compensation whatsoever. This was most encouraging! Based on prior experience with this tube, I decreased the cathode resistor to pull more current through the triode. This effectively lowers the Rp even more, and as as long as I have the headroom available at the grid, typically makes this tube sound better. Phase shift was quite tight up to about 21kHz, opening up only slightly after that, and holding good phase/magnitude symmetry up past 40kHz. I had real problems keeping the ONetics magnitudes matching throughout the sweep range; absolutely no problem with the Lundahl. Okay, now that's some good progress.
Also tried using in step-up configuration, which interestingly enough required a different winding arrangement than stepdown to optimize its performance. In the end, both stepup and stepdown arrangements exhibited successful results. I kept the IT as a stepdown since I didn't require any additional gain, and minimizing the next stage's reflected capacitance is a good thing.
Once the first IT was replaced with the LL1692A, listening tests (mono at this point) confirmed the importance of phase balance, as things noticeably improved, though clearly I was not done. Better, but not completely resolved. There was, to be fair, a bit of linear peaking with the Lundahl from 14K up, but by the time we reach 20K, we are talking maybe 1 or 2 dB. It was possible to flatten this out with simple equivalent zobels from each hot to common, but for now, the zobels were left out. Even though there was a rise in response, the phase and magnitude balance was still very good. Now I could look into the use of the 7119 to improve the sound quality, but I am still optimizing with the scope. There is no obvious measurable difference between the 7119 and 6n6p as far as phase balance goes. Yet there was a significant difference in measurements between the Lundahl and ONetics.
Now that I had reasonable performance with the input stage, I could pursue the driver stage. The original concern I had regarding the grid problems was easily solved; I was simply overdriving the driver stage way beyond what was necessary to drive the output stage to clipping. Just a big oversight on my part, with the root cause to be revealed later.
I found the 46 to be a very good driver once the grids were properly driven. Phase balance measured at the 300B grids were looking pretty good, but not as good as what I had obtained with the input stage. What if I used another LL1692A for the driver stage? Purchased another one (part of the reason this process has taken 5 weeks to report back) and inserted as the 2nd IT. Again, phase balance and magnitude match was improved, though less improvement was required over the input stage. From what I can determine, the 46 is a strong enough driver to overcome the stray capacitances that the input stage is so limited by. I was able to drive the 300B grids positive by maybe 5 or 10 volts before visible clipping was seen, but this ability to drive the grid positive was dependent on frequency. Contrary to my assumption, I could drive the grids more positive at high frequencies than at low. This could be in part to the rising response of the LL1692A matching the grid capacitance of the 300B; just a guess. But just the fact I was able to drive the grids positive at all with low visible distortion and good phase balance was reason enough for me to celebrate.
Next post to discuss my findings with the output stage.
Each stage has been converted over to essentially an identical topology of what is posted on Lynn's web site, with the 40uF motor run + teflon bypass from B+ to cathode. Didn't see any huge difference in measured performance, things still looked relatively poor.
Input stage: tried a number of tubes, 12B4 (lowest Rp), 7119, and 6n6p. When looking at Lissajous patterns, there was for the most part no difference in performance between the tubes as viewed on the scope. Still had that annoying phase shift at frequencies less than 20k, and the phase shift wasn't symmetrical either. Extensive experimentation with zobels to try and improve this was successful to a degree (see my prior posts), but heavy loading was necessary, which didn't give me much confidence.
Purchased a Lundahl LL1692A, which was designed for Rp ranging from 1.5K to 5K. This transformer was specified as a 1.75:2 stepup, but I wanted to use it in a stepdown configuration for better drive. In accordance with others' experience, this transformer absolutely performed differently depending on how the windings were arranged. Tried 16 different configurations to find the best performer. Measurements indicated phase/magnitude balance from excellent to ridiculous. Moral of the story; experiment and measure prior to installing. FWIW, I was not seeing any difference in performance with the ONetics when I changed winding configurations. Installed the LL1692A in place of the first IT, feeding off the 6n6p. Measured phase shift and balance were significantly improved, without any compensation whatsoever. This was most encouraging! Based on prior experience with this tube, I decreased the cathode resistor to pull more current through the triode. This effectively lowers the Rp even more, and as as long as I have the headroom available at the grid, typically makes this tube sound better. Phase shift was quite tight up to about 21kHz, opening up only slightly after that, and holding good phase/magnitude symmetry up past 40kHz. I had real problems keeping the ONetics magnitudes matching throughout the sweep range; absolutely no problem with the Lundahl. Okay, now that's some good progress.
Also tried using in step-up configuration, which interestingly enough required a different winding arrangement than stepdown to optimize its performance. In the end, both stepup and stepdown arrangements exhibited successful results. I kept the IT as a stepdown since I didn't require any additional gain, and minimizing the next stage's reflected capacitance is a good thing.
Lynn Olson said:
Once the first IT was replaced with the LL1692A, listening tests (mono at this point) confirmed the importance of phase balance, as things noticeably improved, though clearly I was not done. Better, but not completely resolved. There was, to be fair, a bit of linear peaking with the Lundahl from 14K up, but by the time we reach 20K, we are talking maybe 1 or 2 dB. It was possible to flatten this out with simple equivalent zobels from each hot to common, but for now, the zobels were left out. Even though there was a rise in response, the phase and magnitude balance was still very good. Now I could look into the use of the 7119 to improve the sound quality, but I am still optimizing with the scope. There is no obvious measurable difference between the 7119 and 6n6p as far as phase balance goes. Yet there was a significant difference in measurements between the Lundahl and ONetics.
Now that I had reasonable performance with the input stage, I could pursue the driver stage. The original concern I had regarding the grid problems was easily solved; I was simply overdriving the driver stage way beyond what was necessary to drive the output stage to clipping. Just a big oversight on my part, with the root cause to be revealed later.
I found the 46 to be a very good driver once the grids were properly driven. Phase balance measured at the 300B grids were looking pretty good, but not as good as what I had obtained with the input stage. What if I used another LL1692A for the driver stage? Purchased another one (part of the reason this process has taken 5 weeks to report back) and inserted as the 2nd IT. Again, phase balance and magnitude match was improved, though less improvement was required over the input stage. From what I can determine, the 46 is a strong enough driver to overcome the stray capacitances that the input stage is so limited by. I was able to drive the 300B grids positive by maybe 5 or 10 volts before visible clipping was seen, but this ability to drive the grid positive was dependent on frequency. Contrary to my assumption, I could drive the grids more positive at high frequencies than at low. This could be in part to the rising response of the LL1692A matching the grid capacitance of the 300B; just a guess. But just the fact I was able to drive the grids positive at all with low visible distortion and good phase balance was reason enough for me to celebrate.
Next post to discuss my findings with the output stage.
Okay, input stage has largely been corrected by the LL1692A, and the 300B grids appear to be driven well by the driver stage, again using LL1692A. Both stages exhibited some degree of linear peaking, so I performed an overall frequency response test to see how things were.
Absolutely ruler flat. Hmmmm. This is unexpected. There is a clear, visible rising response when looking at the 300B grids, yet at the speaker terminals, ruler flat response. Does the frequency response of the output transformer fall off that well to perfectly match the rising response of the front end? I had my doubts.
Don't know how valid of a test this is, but it sounded good to me. AC coupled a 10X probe at the grids, checking the Lissajous pattern of the output stage. At 18 kHz:
Okay, that looks darn nice. Tight phase balance. Magnitude is a little skewed off 45 degrees, and correct me if I'm wrong, but that will be expected with tubes whose mu does not perfectly match. I also expect this will correct itself at the single ended 4 ohm output winding ? Looking on a time scale, the sine waves looked very clean, low distortion and good matching. So then I attached the probes, ac coupled, to the 300B plates:
Now that doesn't look very promising, does it? I understand there has been a lot of Rp discussion about my tube selection on the first two stages, but if the 300B cannot drive a resistive 4 ohm load at 2W and low distortion, I have some serious concerns. Something is preventing the 300B from accurately driving the output transformer.
As a double check, I compared a grid signal to its corresponding plate signal, and inverted one channel. Mismatch between plate and grid was obvious, and it looked to me like slewing of the output stage. I can acquire an image of this if necessary, but fundamentally, input and output are not matching.
This was the root cause of the "falling response" in the output stage. Not frequency-dependent gain differences, per se, but distortion of the output stage that added together to produce a net effective attenuation of the output when measured on an rms basis. Also was seeing an impedance mismatch with the output transformer when looking at input/output levels, as if the transformer was wound with too many primary turns. Per Bud's suggestion, I measured the turns ratio with a 100V ac input, measuring the output. Numbers right on the money, with the correct turns ratio. Yet, when connected in the circuit with varying frequencies, the voltage ratios are different, owing I believe to the distortion phenomenon above.
So I need to resolve the distortion in my output stage. If I were to venture a guess, the behavior of the output iron is very similar to the performance of the former IT's ??? I can't imagine what else it could be. I have 6 different 300B's, and they all show the same performance in circuit. Grids of the 300B's look great, better than they ever have. Good balance, low distortion.
I expect this is the final significant issue I am dealing with on this amp. There are other nuisance issues, such as the hum cancellation circuit which is showing curious behavior, and my future intentions as to circuit topology, but that can be addressed at a later date. For now, what could be going on with my output stage?
Thanks for the assistance.
Absolutely ruler flat. Hmmmm. This is unexpected. There is a clear, visible rising response when looking at the 300B grids, yet at the speaker terminals, ruler flat response. Does the frequency response of the output transformer fall off that well to perfectly match the rising response of the front end? I had my doubts.
Don't know how valid of a test this is, but it sounded good to me. AC coupled a 10X probe at the grids, checking the Lissajous pattern of the output stage. At 18 kHz:
An externally hosted image should be here but it was not working when we last tested it.
Okay, that looks darn nice. Tight phase balance. Magnitude is a little skewed off 45 degrees, and correct me if I'm wrong, but that will be expected with tubes whose mu does not perfectly match. I also expect this will correct itself at the single ended 4 ohm output winding ? Looking on a time scale, the sine waves looked very clean, low distortion and good matching. So then I attached the probes, ac coupled, to the 300B plates:
An externally hosted image should be here but it was not working when we last tested it.
Now that doesn't look very promising, does it? I understand there has been a lot of Rp discussion about my tube selection on the first two stages, but if the 300B cannot drive a resistive 4 ohm load at 2W and low distortion, I have some serious concerns. Something is preventing the 300B from accurately driving the output transformer.
As a double check, I compared a grid signal to its corresponding plate signal, and inverted one channel. Mismatch between plate and grid was obvious, and it looked to me like slewing of the output stage. I can acquire an image of this if necessary, but fundamentally, input and output are not matching.
This was the root cause of the "falling response" in the output stage. Not frequency-dependent gain differences, per se, but distortion of the output stage that added together to produce a net effective attenuation of the output when measured on an rms basis. Also was seeing an impedance mismatch with the output transformer when looking at input/output levels, as if the transformer was wound with too many primary turns. Per Bud's suggestion, I measured the turns ratio with a 100V ac input, measuring the output. Numbers right on the money, with the correct turns ratio. Yet, when connected in the circuit with varying frequencies, the voltage ratios are different, owing I believe to the distortion phenomenon above.
So I need to resolve the distortion in my output stage. If I were to venture a guess, the behavior of the output iron is very similar to the performance of the former IT's ??? I can't imagine what else it could be. I have 6 different 300B's, and they all show the same performance in circuit. Grids of the 300B's look great, better than they ever have. Good balance, low distortion.
I expect this is the final significant issue I am dealing with on this amp. There are other nuisance issues, such as the hum cancellation circuit which is showing curious behavior, and my future intentions as to circuit topology, but that can be addressed at a later date. For now, what could be going on with my output stage?
Thanks for the assistance.
Sure. If used in step up, the datasheet recommendation Alt.M is correct.
If used in step down, follow the datasheet Alt.M for the 'C' windings, but the A and B configuration is different. Instead, connect 8-10, 7-2, and 5 to 3. B+ ties to 7-2, and 9 and 4 goes to the plates.
Of course, you will want to test first to make sure it's right for your application. Amazing how crazy the balance would get when you had a different configuration.
If used in step down, follow the datasheet Alt.M for the 'C' windings, but the A and B configuration is different. Instead, connect 8-10, 7-2, and 5 to 3. B+ ties to 7-2, and 9 and 4 goes to the plates.
Of course, you will want to test first to make sure it's right for your application. Amazing how crazy the balance would get when you had a different configuration.
The 300B's look sick - or more particularly, one of them looks sick, based on the one-sided twist in the waveform. Perhaps one of them is gassy, or some fault in the output transformer is putting excessive capacitance in the HF load of one tube but not the other.
Looking at the second X/Y plot, with the phase opening up only on the lower half of the display, I'm curious what would happen if the pair of tubes were reversed. If the opening up in the display then moves to the top of the display, well, then you have your culprit: a bad tube. If it stays the same (on the bottom of the display) you have a bad socket, bad wiring, or an asymmetric output transformer. It's either one or the other; swapping the tubes will answer the question in a decisive way.
If you can see a visible difference on the scope between grid and plate, that is gross distortion - anything that is visible to the eye represents several per cent distortion, far in excess of acceptable limits.
I am suspicious one or more of the tubes has poor emission, due to gas, faulty processing, or the filament not operating at the correct temperature. I've seen DHT's that had less filament current draw than factory spec; when that happens, the plate curves are all wonky, particularly on the high-current side, thanks to inadequate emission. It sounds ridiculous, but modern 300B's have been assembled with the wrong set of filaments, or half of the filaments non-functional.
If one of the tubes is completely dead (no emission at all), remember, with conventional PP and transformer coupling, there is the illusion of it still working, thanks to the transformer acting as a see-saw and swinging the plate voltage of the inactive side up and down.
Looking at the second X/Y plot, with the phase opening up only on the lower half of the display, I'm curious what would happen if the pair of tubes were reversed. If the opening up in the display then moves to the top of the display, well, then you have your culprit: a bad tube. If it stays the same (on the bottom of the display) you have a bad socket, bad wiring, or an asymmetric output transformer. It's either one or the other; swapping the tubes will answer the question in a decisive way.
If you can see a visible difference on the scope between grid and plate, that is gross distortion - anything that is visible to the eye represents several per cent distortion, far in excess of acceptable limits.
I am suspicious one or more of the tubes has poor emission, due to gas, faulty processing, or the filament not operating at the correct temperature. I've seen DHT's that had less filament current draw than factory spec; when that happens, the plate curves are all wonky, particularly on the high-current side, thanks to inadequate emission. It sounds ridiculous, but modern 300B's have been assembled with the wrong set of filaments, or half of the filaments non-functional.
If one of the tubes is completely dead (no emission at all), remember, with conventional PP and transformer coupling, there is the illusion of it still working, thanks to the transformer acting as a see-saw and swinging the plate voltage of the inactive side up and down.
Thank you, Lynn. Your recommendations confirm for me what I have already checked, and I believe I may have asymmetry in the output transformers.
The behavior does not follow the tubes as they are swapped; it stays with the leads of the transformer. I have 4 JJ's and 2 Valve Arts, and identical performance regardless of how they are installed.
Further, I checked left and right channels, and performance is identical from amp to amp; bottom side always distorted.
For some eye candy (I like playing with my scope), I've attached some further pictures. This one of yellow-grid and blue-plate, obvious distortion on the plate:
Here is an FFT of the grid. This taken on the unmodified amp with the ONetics, hence the 2nd and 3rd harmonics- the Lundahl shows very little distortion. Doesn't matter, since I'm only comparing grid to plate.
Then as a comparision, FFT of the plate:
Again, same results regardless of tube installed.
THANKS TO ALL for their help, and whatever other assistance I can be to those attempting a similar build, I will provide whatever advice and pitfall warnings I can. I look forward to swapping out some parts and getting a listen to what this amp is really supposed to sound like. Other than incorrectly modding the design to use fully differential input and driver stages (which I have corrected), at least my build was correct. I comfort myself in the fact that I'm not a complete failure (only partly a failure, as my friends are currently mocking me as spending all the time and money only to kind of 'start over'). C'est la vie.
When we stop learning, we stop growing. What have I learned ? Breadboard all designs, even those that are tried, true, and tested, since even a single component can throw you into a tailspin.
The behavior does not follow the tubes as they are swapped; it stays with the leads of the transformer. I have 4 JJ's and 2 Valve Arts, and identical performance regardless of how they are installed.
Further, I checked left and right channels, and performance is identical from amp to amp; bottom side always distorted.
For some eye candy (I like playing with my scope), I've attached some further pictures. This one of yellow-grid and blue-plate, obvious distortion on the plate:
An externally hosted image should be here but it was not working when we last tested it.
Here is an FFT of the grid. This taken on the unmodified amp with the ONetics, hence the 2nd and 3rd harmonics- the Lundahl shows very little distortion. Doesn't matter, since I'm only comparing grid to plate.
An externally hosted image should be here but it was not working when we last tested it.
Then as a comparision, FFT of the plate:
An externally hosted image should be here but it was not working when we last tested it.
Again, same results regardless of tube installed.
THANKS TO ALL for their help, and whatever other assistance I can be to those attempting a similar build, I will provide whatever advice and pitfall warnings I can. I look forward to swapping out some parts and getting a listen to what this amp is really supposed to sound like. Other than incorrectly modding the design to use fully differential input and driver stages (which I have corrected), at least my build was correct. I comfort myself in the fact that I'm not a complete failure (only partly a failure, as my friends are currently mocking me as spending all the time and money only to kind of 'start over'). C'est la vie.
When we stop learning, we stop growing. What have I learned ? Breadboard all designs, even those that are tried, true, and tested, since even a single component can throw you into a tailspin.
Well, one more check. Since this is a zero-feedback amplifier, reverse the primary leads of the OPT and see what happens. If your suspicions are correct, the distortion will follow the leads, and the diagnosis confirmed.
A minor way to explore the symmetry of the OPT's to compare balanced vs unbalanced drive for the speakers - you can do this by connecting the 0-ohm tap to ground (unbalanced drive), vs connecting the 4-ohm tap to ground (balanced drive). Different speaker connections may affect the balance of the overall transformer. (Never operate the OPT without having one of the speaker taps connected to ground - this is a safety requirement.)
More dynamic range on the FFT would be nice - will the scope do an average of say, 16 sweeps, and show more resolution at the bottom of the range? It would be nice to see the harmonic structure in more detail.
A minor way to explore the symmetry of the OPT's to compare balanced vs unbalanced drive for the speakers - you can do this by connecting the 0-ohm tap to ground (unbalanced drive), vs connecting the 4-ohm tap to ground (balanced drive). Different speaker connections may affect the balance of the overall transformer. (Never operate the OPT without having one of the speaker taps connected to ground - this is a safety requirement.)
More dynamic range on the FFT would be nice - will the scope do an average of say, 16 sweeps, and show more resolution at the bottom of the range? It would be nice to see the harmonic structure in more detail.
I'm getting the hang of this. I predicted you would recommend that, and have already tried it 
Swapping primary leads and/or the grounded lead of the secondary produced no change.
The FFT I'll have to look into. I do have Arta, but have been procrastinating the learning curve. In the interim, the Tek FFT was a tool for me to get a quick overview of where my distortion has been entering with minimal effort. I venture to say its resolution won't get much better.
Swapping primary leads and/or the grounded lead of the secondary produced no change.
The FFT I'll have to look into. I do have Arta, but have been procrastinating the learning curve. In the interim, the Tek FFT was a tool for me to get a quick overview of where my distortion has been entering with minimal effort. I venture to say its resolution won't get much better.
While you are anticipating, please check to see if the white to blue and white brown DC resistance is within 2 ohms. If it is any greater than that an internal wire has snapped due to external temperature extremes or rough handling during shipping, as they were within that range through final test or they would not have shipped to you.
The normal no load to full load AC power regulation is 3.5% for these transformers when full load equates to 30 watts AC RMS. So I am quite concerned with your reported voltage drop and only 13 watts of power, plus high distortion and asymmetric phase balance.
If the phase problem does not track the winding connection arrangements it becomes less likely that the transformer is the culprit here.
I am also interested in the swing voltages out of the 46 tube and how closely they are tracked in amplitude by the interstage. I have yet to find a listing of the various capacitance's for that tube but that IT was designed to match from 800 to 1200 ohms of load impedance and the 46 has about 1700. This is not a gross mismatch but will cause odd things to creep in at the frequency extremes. The reason for looking for the 46 tube internal capacitance's is they will affect the IT just as will those of the 300 b on the other side.
How does the Lundahl perform in the first IT position when hooked up to service 1 to 1, rather than in the much easier to handle step down conditions? Given a bit of time and a small amount of extra business I could be persuaded to provide step down IT's for your investigations. I am not looking for a comparison here, our companies have two widely divergent design philosophies, and I am always willing to learn from those I respect.
Bud
The normal no load to full load AC power regulation is 3.5% for these transformers when full load equates to 30 watts AC RMS. So I am quite concerned with your reported voltage drop and only 13 watts of power, plus high distortion and asymmetric phase balance.
If the phase problem does not track the winding connection arrangements it becomes less likely that the transformer is the culprit here.
I am also interested in the swing voltages out of the 46 tube and how closely they are tracked in amplitude by the interstage. I have yet to find a listing of the various capacitance's for that tube but that IT was designed to match from 800 to 1200 ohms of load impedance and the 46 has about 1700. This is not a gross mismatch but will cause odd things to creep in at the frequency extremes. The reason for looking for the 46 tube internal capacitance's is they will affect the IT just as will those of the 300 b on the other side.
How does the Lundahl perform in the first IT position when hooked up to service 1 to 1, rather than in the much easier to handle step down conditions? Given a bit of time and a small amount of extra business I could be persuaded to provide step down IT's for your investigations. I am not looking for a comparison here, our companies have two widely divergent design philosophies, and I am always willing to learn from those I respect.
Bud
zigzagflux said:I'm getting the hang of this. I predicted you would recommend that, and have already tried it
Swapping primary leads and/or the grounded lead of the secondary produced no change.
What!? If swapping the primary leads produced no change, that excludes OPT primary imbalance. I am referring to the X/Y phase shift measurement, by the way, not the scope stuff. Well, that's really weird.
Something is very, very wrong here. The critical measurement is the X/Y phase balance measurement, as shown above. The opening of the trace on the minus phase indicates something bad and nonlinear is happening. It should not be there at all, in any amplifier.
This is as if one of the scope channels has a bad amplifier - they are phase-matched, aren't they? Are the attenuators in the scope or probes working OK when you measure the plate voltages? Are the caps exactly matched? Is one of the caps starting to break down at the very high plate voltages? (Swap stuff to find out what's going on.)
If swapping the tubes and swapping the primary leads produced no change, then this is starting to point away from the amplifier, unless there is some kind of truly wacko unidirectional saturation fault in the OPT. But that makes no sense. The magnetic core is out of the picture above 1 kHz anyway - it's all direct air coupling in there, so no saturation issues. I would go ahead and swap OPT's. but something is very strange here. I was expecting the fault to follow the primary leads.
Another test: swap the grounding on the secondaries from the 0-ohm lead to the 16-ohm (or whatever the highest-impedance is) leadout wire. You have to track down the source of the asymmetry and distortion. Also double-check to see that the core really is grounded, instead of floating off into the ether.
One last thought: maybe this is DC imbalance partially saturating the core. Very long shot. Put matched sampling resistors with values something like 10 to 100 ohms in series with the plates. Be very careful measuring the voltage drops across the resistors. The OPT doesn't want to see more than 5~10 mA of static DC imbalance, and less is better.
My thoughts would have to completely echo Lynn's, something very fishy going on here. I would focus strongly on assuring that the scope is not the culprit here, and check both probes for phase and amplitude response across the whole audio bandwidth. Tell us that you are using well matched scope probes, preferably same brand and type or that at least you have checked one against the other to assure they are virtually identical.
Whoa, here guys, I didn't make myself clear when I answered above. It definitely appears to be the output transformer.
When I said I swapped primary leads and it "produced no change" I meant to imply that it followed the leads. That is to say, there was no improvement in performance when leads were swapped. Very sorry for the poor statement. The distortion follows the primary wire, is that the best way to state it?
And yes, when I grounded the 4 or 8 ohm tap instead of the labeled common, there was still the same distortion on the same primary lead, so no change with that.
Brand new factory calibrated Tek 2024B scope, about 4 months old.
Writing from work right now, but I did do some basic resistance checks before and after installation. No problems; very good DC resistance balance. I'll check again, but both amps are doing the same thing; I'd be surprised if both failed identically.
Bud, the LL1692A worked essentially the same with regards to phase balance/frequency response in both step up (1.75:2) as well as step down (2:1.75). Granted, the winding configurations had to be changed depending on how I wanted to use it, but once the right configuration was found, operation was as expected. I chose step down just for that added security, and I really don't need the gain.
Post 84 shows the signals at the 300B grids. Looks pretty good to me, so my assumption is that the 46's are doing a good job. Maybe not as well as a 45, but the distortion after all the other issues were resolved, as far as I can tell, is now at the output stage.
I'll get some more thorough information tonite. Thanks.
When I said I swapped primary leads and it "produced no change" I meant to imply that it followed the leads. That is to say, there was no improvement in performance when leads were swapped. Very sorry for the poor statement. The distortion follows the primary wire, is that the best way to state it?
And yes, when I grounded the 4 or 8 ohm tap instead of the labeled common, there was still the same distortion on the same primary lead, so no change with that.
Brand new factory calibrated Tek 2024B scope, about 4 months old.
Writing from work right now, but I did do some basic resistance checks before and after installation. No problems; very good DC resistance balance. I'll check again, but both amps are doing the same thing; I'd be surprised if both failed identically.
Bud, the LL1692A worked essentially the same with regards to phase balance/frequency response in both step up (1.75:2) as well as step down (2:1.75). Granted, the winding configurations had to be changed depending on how I wanted to use it, but once the right configuration was found, operation was as expected. I chose step down just for that added security, and I really don't need the gain.
Post 84 shows the signals at the 300B grids. Looks pretty good to me, so my assumption is that the 46's are doing a good job. Maybe not as well as a 45, but the distortion after all the other issues were resolved, as far as I can tell, is now at the output stage.
I'll get some more thorough information tonite. Thanks.
Okay, some of this might be repetition, but hopefully I will get confirmation what I see happening. Let me know if you need anything else. If I could do a better job with FFT on short notice, I would. Set 64 averages only for FFT measurements; otherwise straight sampling.
Equipment setup: 4 ohm 100W resistor used as dummy load. Because it shows the least amount of distortion, I used the amp with LL1692A in each of the IT locations. All transformer cores are grounded. Output was 18 kHz at about 4W. Common wire on the secondary of OPT grounded. All probes were compensated prior to measurements, set to 10X.
Reference input from preamp; 3rd harmonic (the first obvious one) is about 58dB down from fund:
Captures of the 300B grids:
The 2nd harmonic above is about 50dB down from the fund.
No changes made, captures of the 300B plates:
The 2nd harmonic above is 29dB down from the fund.
Obviously not very good results. Just to repeat the tests again, I lifted the secondary bond, then proceeded to ground the 4 ohm tap, then the 8 ohm tap. No visible change in results. Returned to the common lead grounded. Switched the two ends of the OPT primary leads; the wide end of the Lissajous followed the lead, in other words, the scope pattern was inverted, fat end now on the upper left. I called this 'no change' above, which was way too ambiguous, as in my mind the transformer was still the problem; nothing else changed. Last step was to swap 300B's; no change in the scope trace when tubes were swapped.
Just for giggles, upped the output to 8W. View of the grids:
Below is a view of the plates:
And then a comparison of the grid (scope channel inverted) to the corresponding plate:
Any other measurements you would like to see with my admittedly limited test equipment?
Equipment setup: 4 ohm 100W resistor used as dummy load. Because it shows the least amount of distortion, I used the amp with LL1692A in each of the IT locations. All transformer cores are grounded. Output was 18 kHz at about 4W. Common wire on the secondary of OPT grounded. All probes were compensated prior to measurements, set to 10X.
Reference input from preamp; 3rd harmonic (the first obvious one) is about 58dB down from fund:
An externally hosted image should be here but it was not working when we last tested it.
Captures of the 300B grids:
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
The 2nd harmonic above is about 50dB down from the fund.
No changes made, captures of the 300B plates:
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
The 2nd harmonic above is 29dB down from the fund.
Obviously not very good results. Just to repeat the tests again, I lifted the secondary bond, then proceeded to ground the 4 ohm tap, then the 8 ohm tap. No visible change in results. Returned to the common lead grounded. Switched the two ends of the OPT primary leads; the wide end of the Lissajous followed the lead, in other words, the scope pattern was inverted, fat end now on the upper left. I called this 'no change' above, which was way too ambiguous, as in my mind the transformer was still the problem; nothing else changed. Last step was to swap 300B's; no change in the scope trace when tubes were swapped.
Just for giggles, upped the output to 8W. View of the grids:
An externally hosted image should be here but it was not working when we last tested it.
Below is a view of the plates:
An externally hosted image should be here but it was not working when we last tested it.
And then a comparison of the grid (scope channel inverted) to the corresponding plate:
An externally hosted image should be here but it was not working when we last tested it.
Any other measurements you would like to see with my admittedly limited test equipment?
Hi Zigzagflux,
I will admit to running in a bit of a vacuum here, but I am now wondering whether or not your expectations of the opt performance at 18kHz are realistic or not. I think the only way to know this for sure would be to do the same experiment on another reputedly "unimpeachable" opt. I suspect that leakage inductance and stray capacitances are probably not balanced between the primary halves of the winding, (nor reasonably can they be without multiple prototype iterations of that EI transformer) and I am not sure that you can necessarily do a lot better without a huge expenditure. (A C-core type where you can configure the windings for best behavior may be your best option - a Lundahl or similar.) The distortion numbers you quote sound pretty decent for the power levels involved and the fact that you are running open loop.
The lissajous test regimen you have designed is a very demanding test for an opt to meet, and I think you might want to compare actual known phase shifts at various frequencies to see how bad this actually is. Do you know the actual number of degrees of phase shift represented by these patterns?
My designs do not perform any better and I suspect in a lot of instances may perform significantly worse.
It is not too uncommon in my experience for EI PP opts in the 30W and higher wattage classes to fall quite short of ideal performance in the HF region.. You would think that driving a transformer primary from a low impedance source would naturally result in the best possible performance, but unfortunately that is not the case - add in a little leakage inductance and you can quickly see that a much higher source impedance (rp) may in fact result in better performance. The trade offs are always between adequate primary inductance, sufficient core area for good linearity at low frequencies and keeping leakage inductance and stray capacitance in check - these conflict and by necessity these choices are a compromise.
Seems (to me) like a C-core based output transformer would be the only solution likely to result in a significant improvement in performance.
How does this amplifier actually sound?
You are unlikely to ever need more than a watt or two at these frequencies..
And..
As always jmtcw...
I will admit to running in a bit of a vacuum here, but I am now wondering whether or not your expectations of the opt performance at 18kHz are realistic or not. I think the only way to know this for sure would be to do the same experiment on another reputedly "unimpeachable" opt. I suspect that leakage inductance and stray capacitances are probably not balanced between the primary halves of the winding, (nor reasonably can they be without multiple prototype iterations of that EI transformer) and I am not sure that you can necessarily do a lot better without a huge expenditure. (A C-core type where you can configure the windings for best behavior may be your best option - a Lundahl or similar.) The distortion numbers you quote sound pretty decent for the power levels involved and the fact that you are running open loop.
The lissajous test regimen you have designed is a very demanding test for an opt to meet, and I think you might want to compare actual known phase shifts at various frequencies to see how bad this actually is. Do you know the actual number of degrees of phase shift represented by these patterns?
My designs do not perform any better and I suspect in a lot of instances may perform significantly worse.
It is not too uncommon in my experience for EI PP opts in the 30W and higher wattage classes to fall quite short of ideal performance in the HF region.. You would think that driving a transformer primary from a low impedance source would naturally result in the best possible performance, but unfortunately that is not the case - add in a little leakage inductance and you can quickly see that a much higher source impedance (rp) may in fact result in better performance. The trade offs are always between adequate primary inductance, sufficient core area for good linearity at low frequencies and keeping leakage inductance and stray capacitance in check - these conflict and by necessity these choices are a compromise.
Seems (to me) like a C-core based output transformer would be the only solution likely to result in a significant improvement in performance.
How does this amplifier actually sound?
You are unlikely to ever need more than a watt or two at these frequencies..
And..
As always jmtcw...
kevinkr said:
Agreed. Actually, I ordered a test LL1620 yesterday. I could also compare to some spare A470 iron I have, which if I recall are 4.3K p-p.
Not at present; it's a little more difficult since the zero crossing is on the money. I will have to correlate time to quadrature axis separation, then then convert to degrees ???
Well, don't forget, I have built one of your designs (feedback, yes) and it is currently serving as my reference. Nice little workhorse amp.
In mono, it doesn't sound too bad, to be honest. Equal to or just barely better than your ST70 mod since getting the IT issues resolved. However, as Lynn has previously stated, the driver and output stage are essentially flawless, it's the input stage that is the weak link. These measurements do not seem to show a flawless output stage.
Agreed. But does it make sense for the IT to operate better than the OPT? I have no idea; this is the first time I've delved into an open loop project. That feedback, with all the troubles it creates, can certainly mask a lot of weaknesses in the rest of the system. Whether you think it's a good thing or not, can you blame the manufacturers in the 50's and 60's for pursuing GNFB? They thought they stumbled on the grail or something.
Zigzagflux,
I agree with Kevin's last post regarding the OPT's.
If THD is fine up to 10KHz, in your place, I would check the IMD using the SMPTE method.
I would also explore a fixed bias by grounding the cathodes of the 300B's. This is a rather simple and effective supply for class A2:
http://www.icl.co.jp/audio/english/2a3/2a3amp.htm
Maybe the sound will change a lot in your system.
Enjoy your holidays,
45
P.S.
Regarding the better behavior of IT's in comparison to OPT's, it makes a lot of sense to me. Actually this is what you need, in principle. Otherwise, better going for other solutions.
I agree with Kevin's last post regarding the OPT's.
If THD is fine up to 10KHz, in your place, I would check the IMD using the SMPTE method.
I would also explore a fixed bias by grounding the cathodes of the 300B's. This is a rather simple and effective supply for class A2:
http://www.icl.co.jp/audio/english/2a3/2a3amp.htm
Maybe the sound will change a lot in your system.
Enjoy your holidays,
45
P.S.
Regarding the better behavior of IT's in comparison to OPT's, it makes a lot of sense to me. Actually this is what you need, in principle. Otherwise, better going for other solutions.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.