Easy to use transformer model with examples:
https://www.diyaudio.com/community/threads/ltspice-hierarchical-ultralinear-opt.379431/post-6853094
https://www.diyaudio.com/community/threads/ltspice-hierarchical-ultralinear-opt.379431/post-6853094
This should get you started. It is for a Stancor A8056 6.6k:8 ohm transformer with 10hz-50kHz bandwidth. Inductances etc worked out from the spreadsheet in this link:
https://www.diyaudio.com/community/threads/spice-transformer-model-spreadsheet.181578/
https://www.diyaudio.com/community/threads/spice-transformer-model-spreadsheet.181578/
I measure the secundary windings at less than 0.1 ohm. My DMM had the resolution but not the accuracy, so I pulled out an older Simpson analogue meter with a low R scale. It measured 0.05 across both windings. In other words, the meter saw it as almost a short. The Simpson requires a calbration cycle to measure resistance and I trust it. Does it make sense to enter something like 0.025 in the spreadsheet for each of the two secundary windings?
Last edited:
taps at 0-4-8 if that's what you mean. I "measure" 0.05 between 0 and 8.
The meter unfortunately does not have a mirror in the scale, so 0.05 means somewhere between 0 and 0.1
The meter unfortunately does not have a mirror in the scale, so 0.05 means somewhere between 0 and 0.1
The 4 ohm tap will be about 70% of the secondary winding turns (impedance is related to the square of the number of turns) if you want to split the resistance accurately, although it will make no significant difference to the result.
Added an OPT using measured values.. Does not give a nice sinewave at the output (looks like its has a AM modulation on it), and a .ac analysis gives me a "Matrix is singular error" back. So I probably did something wrong.
Most DIYers can't really measure less than an Ohm reliably, but fortunately it doesn't matter until you get down to speaker level impedances. Now that we're there, we need to consider the relative importance of the numbers. This will seem heretical to modern audiences but software level modeling is generating more rather than less understanding of design choices. We tend to believe our models with less scrutiny than we Otter.
Another way of saying this is that models, like shotguns and revolvers (to use a firearms analogy) are experts' tools, and only useful after a deeper understanding of their limitations. We cannot observe reality directly*, we can only observe our models, and must understand their limitations. We must have models (it's inherently human, and probably true for other processing machines) but we can't conflate our models with the world.
If you can hear something as being "wrong" you will be able to see it on a 'scope, if you're looking at the right spot. Humans are amazingly subtle in hearing lots of stuff and amazingly oblivious to lots of other stuff. The (linear) things we're most sensitive to are things like amplitude variations across critical bands, resonances, anything unexpected, etc. Our predecessors' priorities are all visible on an amplitude vs. time basis.
"HiFi" adds an additional level of what Casteneda called "intention" to the chain. We're seldom called on to focus our attention in our daily lives - we try to numb ourselves from it if possible - but we restore ourselves, at least our attention, when actively listening to music. We're not crazy; we're just intention deprived.
(*) We operate on a model of reality that runs about 200mS behind clock time. This is measurable with highest trained athlete sprinters leaving a starting gate at random-time impulse ("starter's gun"). Our visual processing time is amazingly powerful to be able to generate our reality model including conversion from the spherical retina data input to our "imagined" rectilinear world view, our "imagined" color vision (there are no "colors" in the external world, only varying wavelengths of about an octave of EM radiation), and our Darwinian weighted data input weighting for danger, etc. Recursive programming ("Darwinian evolution" is equally valid) got us here and left us with some artifacts. Could be worse. But we live in an "imagined" model of the world and operate on that model and not on the world directly. Extra processing syncs our hearing with our slower visual, yada yada, obviously without "our" intentional help. Been working fine since I was born - nothing to fix here.
Totally off topic, but possibly intellectually interesting: a well practiced human can draw and fire a handgun, striking an opponent who is pointing a firearm at him/her/them, before the armed opponent can fire; IOW, less than 200mS. Not recommended of course, unless options have narrowed drastically. But that scales our human reality, which is not very "real-time".
All good fortune,
Chris
Another way of saying this is that models, like shotguns and revolvers (to use a firearms analogy) are experts' tools, and only useful after a deeper understanding of their limitations. We cannot observe reality directly*, we can only observe our models, and must understand their limitations. We must have models (it's inherently human, and probably true for other processing machines) but we can't conflate our models with the world.
If you can hear something as being "wrong" you will be able to see it on a 'scope, if you're looking at the right spot. Humans are amazingly subtle in hearing lots of stuff and amazingly oblivious to lots of other stuff. The (linear) things we're most sensitive to are things like amplitude variations across critical bands, resonances, anything unexpected, etc. Our predecessors' priorities are all visible on an amplitude vs. time basis.
"HiFi" adds an additional level of what Casteneda called "intention" to the chain. We're seldom called on to focus our attention in our daily lives - we try to numb ourselves from it if possible - but we restore ourselves, at least our attention, when actively listening to music. We're not crazy; we're just intention deprived.
(*) We operate on a model of reality that runs about 200mS behind clock time. This is measurable with highest trained athlete sprinters leaving a starting gate at random-time impulse ("starter's gun"). Our visual processing time is amazingly powerful to be able to generate our reality model including conversion from the spherical retina data input to our "imagined" rectilinear world view, our "imagined" color vision (there are no "colors" in the external world, only varying wavelengths of about an octave of EM radiation), and our Darwinian weighted data input weighting for danger, etc. Recursive programming ("Darwinian evolution" is equally valid) got us here and left us with some artifacts. Could be worse. But we live in an "imagined" model of the world and operate on that model and not on the world directly. Extra processing syncs our hearing with our slower visual, yada yada, obviously without "our" intentional help. Been working fine since I was born - nothing to fix here.
Totally off topic, but possibly intellectually interesting: a well practiced human can draw and fire a handgun, striking an opponent who is pointing a firearm at him/her/them, before the armed opponent can fire; IOW, less than 200mS. Not recommended of course, unless options have narrowed drastically. But that scales our human reality, which is not very "real-time".
All good fortune,
Chris
Last edited:
Thanks Ite, indeed that was the problem. The ac simulation now works, however, apart from getting the gain about correct the curves are totally different from the real world measurements.
My ultimate goal was not to simulate a working amplifier, but to improve the actual amplifier. It seems to miss the ability to drive real speakers at low frequencies, so I'll start adding some NFB to see how that changes the behaviour.
My dummy speaker at the moment is an 8 Ohm resistor, which I might replace by a complex impedance as suggested before. And do a few more spot checks at lower frequencies.
My ultimate goal was not to simulate a working amplifier, but to improve the actual amplifier. It seems to miss the ability to drive real speakers at low frequencies, so I'll start adding some NFB to see how that changes the behaviour.
My dummy speaker at the moment is an 8 Ohm resistor, which I might replace by a complex impedance as suggested before. And do a few more spot checks at lower frequencies.
What do you mean with totally different curves ?
Transformers in spice are over simplified models, and only accurate over a narrow frequency range and operating conditions. With primary inductance and K-factor you can set the 3dB points to match your measurements. Outside the 3dB points is unknown territory. Did you have a look at my model (post 21)? It includes parasitic capacitance you can play with, but don't expect to much.
Is the DC-balance ok in your amplifier?
Transformers in spice are over simplified models, and only accurate over a narrow frequency range and operating conditions. With primary inductance and K-factor you can set the 3dB points to match your measurements. Outside the 3dB points is unknown territory. Did you have a look at my model (post 21)? It includes parasitic capacitance you can play with, but don't expect to much.
Is the DC-balance ok in your amplifier?
Ite, I posted the measured bode plots in posts 16 and 18. They look a lot better than what LTspice produces with a simulated OPT. That being said, I have no details about the OPT so the OPT was simulated using Ohms measurements. And I would agree that that is an estimation rather than a simulation.
Is the DC balance ok? Yes, but not perfect. The phaseshifter needs some tweaking, and I need to check the 6550A's. They run at 95 mA. Which is a bit cold I guess. But since I wanted to switch between triode and tetrode operation I needed to make sure it does not exceed the maximum power dissipation in both modes.
The poor performance in the low area may be solved by getting some better speakers. I do not see any drop-off in the low, only hear it.
Next is to look at the distortion again. In '93/94 I used an HP334a, and I believe it was satisfactory. But maybe an o'scope with FFT will show different stuff.
Is the DC balance ok? Yes, but not perfect. The phaseshifter needs some tweaking, and I need to check the 6550A's. They run at 95 mA. Which is a bit cold I guess. But since I wanted to switch between triode and tetrode operation I needed to make sure it does not exceed the maximum power dissipation in both modes.
The poor performance in the low area may be solved by getting some better speakers. I do not see any drop-off in the low, only hear it.
Next is to look at the distortion again. In '93/94 I used an HP334a, and I believe it was satisfactory. But maybe an o'scope with FFT will show different stuff.
Changed speakers for good ones (Thiel CS0.5), made a significant difference. Changed setup of first tube to 390 ohm kathode resistor and 68k anode, that put the phaseshifter in the 1/4 kathode - 3/4 anode supply voltage split. Changed second stage anode resistors to 33k to allow/produce a larger voltage swing. Sound is now a lot better. Last thing I did was to "unbalance" the phaseshifter a bit, which resulted in slightly more overall distortion, but the forced unbalance causes 2nd order harmonics rather than 3rd. It makes the voices a bit mellower. We're not looking at large amounts of distortion (< 0.2% @ 16Vpp output in 8 ohm (4Wrms) in LTSpice. (I added Ite's OPT and tweaked the parameters a bit to make it look like the real OPT). No feedback used, operating in Triode mode. Quite happy now!