smoking-amp said:
Guys in the Tube Forum time to time discuss so called Aikido amp, it is a vacuum triode loaded by identical vacuum triode, so some non-liniarities that depend on anode voltage are compensated. To compensate distortions that depend on anode current tranfrormer or CCS load is used. Audiofools prefer to fool own imagination obtaining needed gain from as less as possible of tubes that may be considered too powerful for given putposes.
If you look at Gioconda by Leonardo it is far not as good as a photography, it is made of horrible distortions, but people love it...
"Hilbert transformer?"
Well, there is a real circuit that does the 90 deg. transform, used to see it in the Amatuer Radio Handbook for early phasing SSB xmitters. One can also just integrate the signal to get 90 deg. for all components but then has a 1/omega term too.
But maybe this is not so bad. All higher harmonics automatically get attenuated by (1/omega) to some power. Reminds me of the J loop feedback patent with two cascaded stages, each with a capacitor feedback to act as an integrator stage. No wonder it kills distortion harmonics so well.
Here is some crazy thinking for a distortion cancellation scheme:
90 degrees gives cancellation for squared terms summed,
60 degrees gives cancellation for cubed terms summed,
45 degrees gives cancellation for quartic terms summed,
..... etc
Now we use two identical amplifiers and measure the level of harmonics in an FFT test.
Then we compose a second signal for the second amp using a "taylored" series of phase shifted components so that harmonics will all cancel when summed with the first amplifier. (not sure if this can be done using just two amplifiers though, but maybe).
Of course, the components in the amplifier FFT spectrum are not uniform with amplitude, so this may require something more like a real Taylor power series to adjust component amplitudes. We'll just tuck this complexity away into the digital preprocessing before the DACs.
Don
Well, there is a real circuit that does the 90 deg. transform, used to see it in the Amatuer Radio Handbook for early phasing SSB xmitters. One can also just integrate the signal to get 90 deg. for all components but then has a 1/omega term too.
But maybe this is not so bad. All higher harmonics automatically get attenuated by (1/omega) to some power. Reminds me of the J loop feedback patent with two cascaded stages, each with a capacitor feedback to act as an integrator stage. No wonder it kills distortion harmonics so well.
Here is some crazy thinking for a distortion cancellation scheme:
90 degrees gives cancellation for squared terms summed,
60 degrees gives cancellation for cubed terms summed,
45 degrees gives cancellation for quartic terms summed,
..... etc
Now we use two identical amplifiers and measure the level of harmonics in an FFT test.
Then we compose a second signal for the second amp using a "taylored" series of phase shifted components so that harmonics will all cancel when summed with the first amplifier. (not sure if this can be done using just two amplifiers though, but maybe).
Of course, the components in the amplifier FFT spectrum are not uniform with amplitude, so this may require something more like a real Taylor power series to adjust component amplitudes. We'll just tuck this complexity away into the digital preprocessing before the DACs.
Don
...than we use infrasonic oscillators to modulate frequency dependent phase shifts of several amps and hear a chorus instead of a solo.
speaking of digital preprocessing, I suspect Altec Lansing does something similar for their active speakers, JBL uses digital preprocessing to tune vertical dispersion in line arrays.
speaking of digital preprocessing, I suspect Altec Lansing does something similar for their active speakers, JBL uses digital preprocessing to tune vertical dispersion in line arrays.
"Aikido amp, it is a vacuum triode loaded by identical vacuum triode"
Yes, I am familiar with this design. It is based roughly on the "voltage mirror" idea.
See:
http://www.diyaudio.com/forums/showthread.php?postid=470523#post470523
I'm not sure the Aikido has it quite right yet though since it was derived with PSRR foremost in mind.
For a real voltage mirror and exact cancellation, one needs to match up the V/I curvature of the load with the V/I curvature of the amplifiying device. This does not necessarily, in fact not likely, lead to using the exact same device for the load as used in the amplifier stage. Particularly for using a triode as amplifier, since it is already nearly linear for a Hi-Z load. The Aikido is probably producing inverse distortion due to too much correction in the first stage, followed by another stage the "uses-up" some of the anti-distortion.
This non-linear load idea is workable with Mosfets - JFets, and I believe I have seen it mentioned in some IC design texts where device geometry can be tailored. It does however reduce all of the gain curve to the lowest gain on the curve. Not working for BJTs though, lowest gain is zero.
Don
Yes, I am familiar with this design. It is based roughly on the "voltage mirror" idea.
See:
http://www.diyaudio.com/forums/showthread.php?postid=470523#post470523
I'm not sure the Aikido has it quite right yet though since it was derived with PSRR foremost in mind.
For a real voltage mirror and exact cancellation, one needs to match up the V/I curvature of the load with the V/I curvature of the amplifiying device. This does not necessarily, in fact not likely, lead to using the exact same device for the load as used in the amplifier stage. Particularly for using a triode as amplifier, since it is already nearly linear for a Hi-Z load. The Aikido is probably producing inverse distortion due to too much correction in the first stage, followed by another stage the "uses-up" some of the anti-distortion.
This non-linear load idea is workable with Mosfets - JFets, and I believe I have seen it mentioned in some IC design texts where device geometry can be tailored. It does however reduce all of the gain curve to the lowest gain on the curve. Not working for BJTs though, lowest gain is zero.
Don
Reading this thread, and coming back to variable capacitances - it gets worse. You get rectification effects from anywhere where the charging and discharging characteristics are different - like a follower terminated by a resistor, for instance.
I have recently been experimenting with a simple amp, where one stage had this problem, a MOSFET gate nonlinear capacitance changed the gain of the previous stage so much it had latch up phenomena peculiar to speciffic signal amplitude, which manifested itself as a 'deadband' with isteresis at a certain amplitude.
Perhaps not entirely on topic, but it shows that nonlinear componets can have some rather insidious effects, which would be very difficult to analyse, modeling them by the usual distorsion mechanisms.
I have recently been experimenting with a simple amp, where one stage had this problem, a MOSFET gate nonlinear capacitance changed the gain of the previous stage so much it had latch up phenomena peculiar to speciffic signal amplitude, which manifested itself as a 'deadband' with isteresis at a certain amplitude.
Perhaps not entirely on topic, but it shows that nonlinear componets can have some rather insidious effects, which would be very difficult to analyse, modeling them by the usual distorsion mechanisms.
re: Wavebourn ---- Aikido
A triode gives a theoretical constant gain of Mu when loaded with a Hi-Z load. So it does not need a non-linear load. In practice, the Mu is not so constant for most tubes versus Vp and Ip, and gets worse with real resistive loading. Trending back towards 3/2 power I versus Vin as the plate voltage gets immobilzed by a low Z load. So some partial diode (but generally quite small) correction could be used with many triodes to compensate for resistive loading curvature.
Interstage tube plates usually only have the next grid to drive so resistor loading is usually set at 5x plate Rp and is quite linear. Output stages are heavily loaded, and could use some diode correction, but the diode will eat up most of the output power in these cases.
Notice the Aikido uses a resistor in series with the top triode for its load. (and also the same value for the bottom tube's cathode resistor) The Aikido seems to approach this solution by using identical top and bottom setups in order to null power supply noise at the center. I believe one should really use an FFT analyzer to determine how much diode versus resistive loading gives linear results for a given tube. The bottom triode's cathode resistor greatly increases the triode's output Z, so it is not really seeing an identical AC load imposed on it.
There is likely some theoretical formula that can be worked out for a first approximation for linear response, but no one has worked it out yet. Needs to take into account the resistive load as well as the geometry factors (gm and rp) of both tubes. I should work this out, someday...
The physical geometry of the loading tube must be such as to cause it to operate over the same part of the curved response as the amplifier tube. So not just any tube will work for the load. I suspect something more like a HV rectifier tube may be a better match. For example, putting a high current rectifier or triode as the load would cause it to operate AC wise over a tiny portion of its curvature, so it would just look like a low value resistor. A HV low current diode will operate AC wise over a much larger part of its curvature. One needs to get this optimised to match the gain tube's curve.
re: ilimzn
I can see this happening with a JFet for sure, surprised to see a Mosfet can do it too. Some earlier ideas on using a varicap diode might be of some use. One really needs an anti-varicap for nulling capacitance change. But using a backward (with respect to the offending junction diode) diode with floating reverse bias can give some cancellation similar to that in a diffl. ampl. stage. The curvatures don't match up, so odd order variation remains. Some opto-isolators are around that have photovoltaic diodes to generate a floating bias voltage.
--------
On my earlier idea of limiting the error feedback to the audio band only, so as to minimize re-entrant distortion, it just occured to me that that is exactly what most tube amps already do, due to their limited freq. response output transformer. By having a high power filter after the final stage, ultrasonic feedback is prevented and ultrasonic distortion output to the speaker is prevented too. So no need for the feedforward ultrasonic corrector to aid the speaker. This idea is in use all the time and works fine!!!
Don
A triode gives a theoretical constant gain of Mu when loaded with a Hi-Z load. So it does not need a non-linear load. In practice, the Mu is not so constant for most tubes versus Vp and Ip, and gets worse with real resistive loading. Trending back towards 3/2 power I versus Vin as the plate voltage gets immobilzed by a low Z load. So some partial diode (but generally quite small) correction could be used with many triodes to compensate for resistive loading curvature.
Interstage tube plates usually only have the next grid to drive so resistor loading is usually set at 5x plate Rp and is quite linear. Output stages are heavily loaded, and could use some diode correction, but the diode will eat up most of the output power in these cases.
Notice the Aikido uses a resistor in series with the top triode for its load. (and also the same value for the bottom tube's cathode resistor) The Aikido seems to approach this solution by using identical top and bottom setups in order to null power supply noise at the center. I believe one should really use an FFT analyzer to determine how much diode versus resistive loading gives linear results for a given tube. The bottom triode's cathode resistor greatly increases the triode's output Z, so it is not really seeing an identical AC load imposed on it.
There is likely some theoretical formula that can be worked out for a first approximation for linear response, but no one has worked it out yet. Needs to take into account the resistive load as well as the geometry factors (gm and rp) of both tubes. I should work this out, someday...
The physical geometry of the loading tube must be such as to cause it to operate over the same part of the curved response as the amplifier tube. So not just any tube will work for the load. I suspect something more like a HV rectifier tube may be a better match. For example, putting a high current rectifier or triode as the load would cause it to operate AC wise over a tiny portion of its curvature, so it would just look like a low value resistor. A HV low current diode will operate AC wise over a much larger part of its curvature. One needs to get this optimised to match the gain tube's curve.
re: ilimzn
I can see this happening with a JFet for sure, surprised to see a Mosfet can do it too. Some earlier ideas on using a varicap diode might be of some use. One really needs an anti-varicap for nulling capacitance change. But using a backward (with respect to the offending junction diode) diode with floating reverse bias can give some cancellation similar to that in a diffl. ampl. stage. The curvatures don't match up, so odd order variation remains. Some opto-isolators are around that have photovoltaic diodes to generate a floating bias voltage.
--------
On my earlier idea of limiting the error feedback to the audio band only, so as to minimize re-entrant distortion, it just occured to me that that is exactly what most tube amps already do, due to their limited freq. response output transformer. By having a high power filter after the final stage, ultrasonic feedback is prevented and ultrasonic distortion output to the speaker is prevented too. So no need for the feedforward ultrasonic corrector to aid the speaker. This idea is in use all the time and works fine!!!
Don
or just put output inductor in the amp's output?
re: lumanauw "or just put output inductor in the amp's output?"
Yes, seems like that should do it. The NFB would come off after the filter, ie directly from the speaker port. Since this will be causing phase problems for stability big time, maybe it becomes the miller stability compensation pole too? Trouble is that speaker Z will be affecting the filter frequency now.
Tube amp xfmrs have multiple taps for different speaker Z to compensate.
Also, tube output xfmrs have a distributed C and series leakage L, so they have a faster rolloff than just an RL filter. Must be some way to do it. Combine with the Zobel network maybe? The phase/stability problem becomes more severe with an LC. But this might also fix the RF incursion thru the output problem too.
Don
Yes, seems like that should do it. The NFB would come off after the filter, ie directly from the speaker port. Since this will be causing phase problems for stability big time, maybe it becomes the miller stability compensation pole too? Trouble is that speaker Z will be affecting the filter frequency now.
Tube amp xfmrs have multiple taps for different speaker Z to compensate.
Also, tube output xfmrs have a distributed C and series leakage L, so they have a faster rolloff than just an RL filter. Must be some way to do it. Combine with the Zobel network maybe? The phase/stability problem becomes more severe with an LC. But this might also fix the RF incursion thru the output problem too.
Don
smoking-amp said:
On my earlier idea of limiting the error feedback to the audio band only, so as to minimize re-entrant distortion, it just occured to me that that is exactly what most tube amps already do, due to their limited freq. response output transformer. By having a high power filter after the final stage, ultrasonic feedback is prevented and ultrasonic distortion output to the speaker is prevented too. So no need for the feedforward ultrasonic corrector to aid the speaker. This idea is in use all the time and works fine!!!
Don
...also, they have the shorter feedback path on high frequencies via some cpaitor from anode of some tube...
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.