A few days ago, I posted some questions about the Transnova amplifier design in the “acoustat-answer-man-is-here” discussion group in case he could share any insights from his time at Acoustat, where the design originated. I know he focuses on speakers and didn’t have much he could share. So, I mentioned I would bring up the topic on the solid-state group.
Over the years, I’ve read through a number of Jim Strickland's notes & schematics on the early Acoustat TNT Transnova amps (TNT-200, TNT-120) while comparing to the later Hafler Transnova versions (9500/9300, 9505/9303 – Hafler made other Transnova models; I just care about the Hafler Transnova designs that use Lateral MOSFETs). Some of Jim's earlier Transnova papers describe something called "complementary" feedback, which looks to be essentially positive feedback being applied to perform some error correction/cancellation. It looks like that aspect of the unique feedback arrangement was phased out of the Hafler designs. It is R26 in the TNT schematic snippet (blue) attached and Rc in the (attached) patent writeup. I don’t see a similarly located resistor is not present in the Hafler version (yellow attached schematic snippet).
Although both driver circuits are able to pump quite a bit of current into the MOSFETs (at least 40mA), I have read that Hafler front-end improved on the linearity vs. the Acoustat front-end. So, I’m curious if the complementary feedback was dropped because it was viewed as unnecessary with the “new & improved” driver circuit or were there other reasons?
I've heard the Hafler version model 9500 played through some nice electrostatics and my old Kef 104.2 speakers and liked what I heard but have never heard the Acoustat TNT models...
I know a few others around here have commented that they didn’t care for the sound of the Transnova design (anatech, djk, etc.); I couldn’t tell if they referred to the earlier Acoustat version or later Hafler version (or both).
While roaming around through the discussions, from time to time, I see some people around here post about experimental designs they have built that are somewhat similar (low voltage front-end driving grounded source Lateral MOSFET output stage that has voltage gain and floating, higher voltage rails), for example, Juma’s nice “Transnova-Schade OS/Amp. I’ve also seen the AES paper and Linear Audio article written by John Vanderkoy (and others) that uses a high performance op amp as the front-end driver for the same output stage instead (image from abstract attched). One of the key differences is Acoustat’s Tranova approach configure the discrete design front-end as a transconductor with local negative feedback around the output stage (Juma does also), while Vanderkooy’s approach uses the op amp as just a voltage gain block and the output stage has no local negative feedback.
Anybody else out there playing with this design? Any of you even happily cloned either the Acoustat or Hafler Transnova amps?
I attached some reading material that was available freely. Some of this seems harder to find now.
mlloyd1
Over the years, I’ve read through a number of Jim Strickland's notes & schematics on the early Acoustat TNT Transnova amps (TNT-200, TNT-120) while comparing to the later Hafler Transnova versions (9500/9300, 9505/9303 – Hafler made other Transnova models; I just care about the Hafler Transnova designs that use Lateral MOSFETs). Some of Jim's earlier Transnova papers describe something called "complementary" feedback, which looks to be essentially positive feedback being applied to perform some error correction/cancellation. It looks like that aspect of the unique feedback arrangement was phased out of the Hafler designs. It is R26 in the TNT schematic snippet (blue) attached and Rc in the (attached) patent writeup. I don’t see a similarly located resistor is not present in the Hafler version (yellow attached schematic snippet).
Although both driver circuits are able to pump quite a bit of current into the MOSFETs (at least 40mA), I have read that Hafler front-end improved on the linearity vs. the Acoustat front-end. So, I’m curious if the complementary feedback was dropped because it was viewed as unnecessary with the “new & improved” driver circuit or were there other reasons?
I've heard the Hafler version model 9500 played through some nice electrostatics and my old Kef 104.2 speakers and liked what I heard but have never heard the Acoustat TNT models...
I know a few others around here have commented that they didn’t care for the sound of the Transnova design (anatech, djk, etc.); I couldn’t tell if they referred to the earlier Acoustat version or later Hafler version (or both).
While roaming around through the discussions, from time to time, I see some people around here post about experimental designs they have built that are somewhat similar (low voltage front-end driving grounded source Lateral MOSFET output stage that has voltage gain and floating, higher voltage rails), for example, Juma’s nice “Transnova-Schade OS/Amp. I’ve also seen the AES paper and Linear Audio article written by John Vanderkoy (and others) that uses a high performance op amp as the front-end driver for the same output stage instead (image from abstract attched). One of the key differences is Acoustat’s Tranova approach configure the discrete design front-end as a transconductor with local negative feedback around the output stage (Juma does also), while Vanderkooy’s approach uses the op amp as just a voltage gain block and the output stage has no local negative feedback.
Anybody else out there playing with this design? Any of you even happily cloned either the Acoustat or Hafler Transnova amps?
I attached some reading material that was available freely. Some of this seems harder to find now.
mlloyd1
Attachments
-
transnova-g,h.pdf
-
the-transnova-topology.pdf
-
the-hafler-tradition.pdf
-
history-of-the-transnova.pdf
-
John Vanderkooy v9 jv abstract image.webp -
acoustat_trans-nova_twin-200_en.pdf
-
Hafler Trans Nova Amplifier 9300, 9500 Large.jpg -
US4467288 Distortion-Free Complemented Error Feedback Amplifier and Method.pdf
No expert, no ee, just building amps, preamps and speakers for 40 years...build and listen. I once put together this circuit, not knowing anything about it, but learning about transnova. It sounded ok, but was quite finicky to set.
Thread 'Amp out grounded, power supply as output, why?' https://www.diyaudio.com/community/threads/amp-out-grounded-power-supply-as-output-why.347506/
Thread 'Amp out grounded, power supply as output, why?' https://www.diyaudio.com/community/threads/amp-out-grounded-power-supply-as-output-why.347506/
It's negative feedback, otherwise it would blow up.
It even says negative feedback on the schematic.
The various schemes to gnd outputs or float supplies can have engineering advantages but as far as the audio is concerned it doesn't make a difference. The signal and control flows remain the same.
Jan
It even says negative feedback on the schematic.
The various schemes to gnd outputs or float supplies can have engineering advantages but as far as the audio is concerned it doesn't make a difference. The signal and control flows remain the same.
Jan
Go check out Broskies website TubeCad Journal. He wrote something up in the last 6 months I believe about the Transnova design and what made them different.
Jan:
I agree that straight-up positive feedback would be a big problem for an amplifier: latch-up or oscillations could make for flaming semiconductors and/or speakers!
I'm guessing it was called "Complement" negative feedback for a reason.
😉
Unusually, overall feedback goes from the MOSFET ops output to the non-inverting input of the driver stage, so the overall loop feedback is negative, as is provided by R27, the usual 24k to 33k resistor (with 1k to ground).
The MOSFET ops has it's own local negative feedback with the 680pF in parallel with the 4.7k or 56K (depending on which schematic you refer to) essentially from gate(s) to drain(s).
The weird one is the R26 feedback resistor in the TNT schematics, because it's connection across the driver stage links 2 consecutive stages that both invert (JFET Q2 and MOSFETs Q3 & Q4) for a net non-inversion, so R26 is not providing the usual direct negative feedback (it makes a funky delta network with the other feedback resistors).
The R26 value is pretty large compared to the other 2, so it cannot be contributing anything very bad since these amps have pretty decent specs.
But it is curious to me that having this resistor in place is claimed to provided improved performance, according to the white papers and patents, but is missing entirely in Hafler's later designs.
And I'm curious why.....
It is negative feedback from two different points - output and driver stage.
It makes sense to call it 'complementary' (just 'complement' is bad English) because two neg fb loops that complement each other.
Still negative fb of course.
A hunch: he played around because it oscillated with only output feedback, and with this scheme it didn't.
And presto - a new legend was born!
Jan
It makes sense to call it 'complementary' (just 'complement' is bad English) because two neg fb loops that complement each other.
Still negative fb of course.
A hunch: he played around because it oscillated with only output feedback, and with this scheme it didn't.
And presto - a new legend was born!
Jan
EdGr said: ...This amplifier has one strength (ability to drive a capacitor) and multiple weaknesses."
Ed:
I'm aware this design was conceived/motivated by a desire to create an amp to drive electrostatics. I'm curious if you'd be willing to say more about what you see as the multiple weaknesses? And I'm assuming you mean weaknesses relative to other topologies.
Jan said: ... A hunch: he played around because it oscillated with only output feedback, and with this scheme it didn't."
Jan:
This hunch seems reasonable to me. 🙂 I would think the optimum resistor value for the complement feedback might be a little bit touchy, so am also curious about why some sort of trimmable adjustment wasn't used. Something else to play around with once I start messing about with this ....
Ed:
I'm aware this design was conceived/motivated by a desire to create an amp to drive electrostatics. I'm curious if you'd be willing to say more about what you see as the multiple weaknesses? And I'm assuming you mean weaknesses relative to other topologies.
Jan said: ... A hunch: he played around because it oscillated with only output feedback, and with this scheme it didn't."
Jan:
This hunch seems reasonable to me. 🙂 I would think the optimum resistor value for the complement feedback might be a little bit touchy, so am also curious about why some sort of trimmable adjustment wasn't used. Something else to play around with once I start messing about with this ....
The common-source output stage has high distortion, output impedance, and Miller capacitance. The output stage relies heavily on feedback to produce acceptable performance.
(In contrast, a source or emitter follower output stage has 100% feedback built-in to its transistors).
The floating power supply is a noise source.
Ed
(In contrast, a source or emitter follower output stage has 100% feedback built-in to its transistors).
The floating power supply is a noise source.
Ed
I haven't read all the documents yet, but at first sight, I agree with the first sentence of Ed's post #9. It looks like you have local negative feedback around the output stage (A), overall negative feedback (B) and positive feedback around the input stage (C), presumably to boost the overall loop gain. I think it also matches the description in the patent:
It's nice to see something else than the zillionth variant of the boring old Lin configuration for a change!
It's nice to see something else than the zillionth variant of the boring old Lin configuration for a change!
Last edited:
Unless Strickland was hired by Hafler company.
I feel the peak of Hafler amp was the one before the Transnova, like Hafler 9290. I don’t like the Transnova topology.
At the end of the previous century, when I was working on my error correction patent, I spend many, many months analyzing some of the wildest feedback schemes people have come up with. Without exception, I always was able to start with the actual setup and simplify it down to a single global feedback loop. No matter how many local, global, positive or negative feedback loops were present.
In the end, all these 'smart' sytems can be simplified to a forward gain block with global feedback. In the end, it all boils down to the difference between open loop gain and feedback return gain to determine the distortion reduction. Certainly never 'distortion-free'. I see no reason why in this case it would be different.
The fact that he named it a 'a distortion free amplifier' shows he really doesn't grasp feedback and control systems theory.
Jan
In the end, all these 'smart' sytems can be simplified to a forward gain block with global feedback. In the end, it all boils down to the difference between open loop gain and feedback return gain to determine the distortion reduction. Certainly never 'distortion-free'. I see no reason why in this case it would be different.
The fact that he named it a 'a distortion free amplifier' shows he really doesn't grasp feedback and control systems theory.
Jan
While I am here, does anyone per chance have this paper:
S. Holmes, G. J. Sussman, J. Boyk, On the Characteristic Spectra of 28 Varieties of Feedback, Cambridge: Beekeepers Press, 1899.
(and I think the year of the latter should be 1998, not 1899 ...).
Jan
S. Holmes, G. J. Sussman, J. Boyk, On the Characteristic Spectra of 28 Varieties of Feedback, Cambridge: Beekeepers Press, 1899.
(and I think the year of the latter should be 1998, not 1899 ...).
Jan