Last effort: Tom insists on using total output noise. Telling us that 69uV (produced by a 3 times higher gain unit) is too much, perceivable for him. (with shorted input, ears 30cm from tweeter)
40uV from Mod86 (one with less gain) is OK.
But in the next paragraph Tom tells us that when using the MiniDSP setup, he receives 139uV total noise output, and he preferres that..
Again: 69uV from an idling FE modul, - disturbing.
139uV total noise, from an idling MiniDSP + Mod86 system.. (now This Is a real use case) - - - Is AOK.
Would like to repeat: I did not create this. I'm just reassuming here the amount of twisting needed, if we base our quality judgement on output total noise calculations..
Ciao, George
40uV from Mod86 (one with less gain) is OK.
But in the next paragraph Tom tells us that when using the MiniDSP setup, he receives 139uV total noise output, and he preferres that..
Again: 69uV from an idling FE modul, - disturbing.
139uV total noise, from an idling MiniDSP + Mod86 system.. (now This Is a real use case) - - - Is AOK.
Would like to repeat: I did not create this. I'm just reassuming here the amount of twisting needed, if we base our quality judgement on output total noise calculations..
Ciao, George
Last exercise: In Tom's house, with FE modules in the chain after the MiniDSP, one should obviously use attenuation, because of the module's higher gain.
Let's say a 5kohm pot adjusted for 3 times attenuation.
That is 1kohm output series resistance, adding 4nV/sqrtHZ extra noise to the FE input. So the original 15nV/sqrtHz gets to be 15,5nV sqrt/Hz, the original 69uV output noise gets to 71,4uV - - FE total output noise with attenuator.
Now the total output noise with MiniDSP will be:
SQRT((10*13,3uV)^2+(71,4uV)^2) = 151uV.
MiniDSP +Mod86 = 139uV
MiniDSP +myref FE = 151uV
This is a real use case calculation!
I will trust You with the conclusion.
Let's say a 5kohm pot adjusted for 3 times attenuation.
That is 1kohm output series resistance, adding 4nV/sqrtHZ extra noise to the FE input. So the original 15nV/sqrtHz gets to be 15,5nV sqrt/Hz, the original 69uV output noise gets to 71,4uV - - FE total output noise with attenuator.
Now the total output noise with MiniDSP will be:
SQRT((10*13,3uV)^2+(71,4uV)^2) = 151uV.
MiniDSP +Mod86 = 139uV
MiniDSP +myref FE = 151uV
This is a real use case calculation!
I will trust You with the conclusion.
Except it's not the real use case. Why on earth would I add an attenuator to the input of the MyREF? The MiniDSP 4x10HD already includes a volume control, so why add more attenuation? If you believe the gain of the MyRef FE is 3x too high, why not just redesign it for a lower gain?
I would also argue that 151 > 139, but maybe that's just me.
It just goes to show that attenuating before amplifying is best to be avoided. That's the concept of gain structure in a nutshell.
Your insistence on using input-referred numbers also don't represent the real use case. I am concerned about the noise at the output of the amplifier because I connect my speakers there. Do you connect yours at the input?
If you want some academic figure of merit for comparing two opamps, using the input-referred noise is a good place to start, though, as BrianL points out, there's more to it than just the wide-band noise number on page 1 of the data sheet.
There is a time and a place for the academic approach. The design phase of an amp is a good place for it, for example. But there is also a time for a broader perspective, including an analysis of the end-to-end performance of an audio system. That's what I have provided.
I think you may have forgotten to comment on my updated noise measurements Post #178. They show the Modulus-86 providing ~5 dB lower noise than the MyRef FE. 20*log10(69/40) = 4.74 dB. Imagine that! Do you still believe my measurements are biased or has issue that been resolved?
When I designed the Modulus-86, I used one of the best (if not the best) opamps available at the time. I designed for the best end-to-end performance of the completed audio system. That's how I ended up with an amp that offers balanced inputs, a relatively low gain, and low output noise. Naturally, I will always strive to find improvements where possible. As result, the Modulus-186, -286, and -686 have even lower noise (both in terms of output noise and input-referred noise).
I'm curious which design considerations led to the choice of 30 dB gain and the use of the LM318 in the MyRef FE.
Tom
I would also argue that 151 > 139, but maybe that's just me.
It just goes to show that attenuating before amplifying is best to be avoided. That's the concept of gain structure in a nutshell. Your insistence on using input-referred numbers also don't represent the real use case. I am concerned about the noise at the output of the amplifier because I connect my speakers there. Do you connect yours at the input?
If you want some academic figure of merit for comparing two opamps, using the input-referred noise is a good place to start, though, as BrianL points out, there's more to it than just the wide-band noise number on page 1 of the data sheet.
There is a time and a place for the academic approach. The design phase of an amp is a good place for it, for example. But there is also a time for a broader perspective, including an analysis of the end-to-end performance of an audio system. That's what I have provided.
I think you may have forgotten to comment on my updated noise measurements Post #178. They show the Modulus-86 providing ~5 dB lower noise than the MyRef FE. 20*log10(69/40) = 4.74 dB. Imagine that! Do you still believe my measurements are biased or has issue that been resolved?
When I designed the Modulus-86, I used one of the best (if not the best) opamps available at the time. I designed for the best end-to-end performance of the completed audio system. That's how I ended up with an amp that offers balanced inputs, a relatively low gain, and low output noise. Naturally, I will always strive to find improvements where possible. As result, the Modulus-186, -286, and -686 have even lower noise (both in terms of output noise and input-referred noise).
I'm curious which design considerations led to the choice of 30 dB gain and the use of the LM318 in the MyRef FE.
Tom
Tom,
This is a design which has history a bit deeper reaching that You seem to be unaware of.
Simple. You should ask Tim de Paravicini about it.
Penasa had his Musical Fidelity A370, he loved the sound of it, so he had recreated a more agile, quick, modern (and cool and green..) version of it. In the early years of the 2000. In 2005 he presented it to us, and I helped him put together the first english description.
The original design started it's existence in the 80-ies. P270, A370. (1986)
This is a design which has history a bit deeper reaching that You seem to be unaware of.
Simple. You should ask Tim de Paravicini about it.
Penasa had his Musical Fidelity A370, he loved the sound of it, so he had recreated a more agile, quick, modern (and cool and green..) version of it. In the early years of the 2000. In 2005 he presented it to us, and I helped him put together the first english description.
The original design started it's existence in the 80-ies. P270, A370. (1986)
Attachments
First off, let me say that both amps are by all accounts excellent designs and I'm impressed by the contributions made by both Tom and Dario to the DIY community.
Secondly, we really are picking at nits here with the noise floor being 130-140dB down.
But if we are going to start arguing over microvolts here then George does bring up a salient point - the difference in gain between the two designs does have to be taken into account in order to do an apples to apples comparison. As an example, what would you expect if you put a voltage divider on the output of the MyRef that would reduce the signal so that the effective gain was the same as the Mod68?
We can debate this ad nauseum or we can all just go on our merry way and acknowledge that the Mod86 is a nice low-noise design but the gain is at the lower end while the MyRef has slightly more output noise but is a higher gain amp.
Finally, I suspect that the ultimate determination of output quality is more influenced by the skills demonstrated by the constructor than anything else.
Take it easy, have fun and stay safe.
-bill
Secondly, we really are picking at nits here with the noise floor being 130-140dB down.
But if we are going to start arguing over microvolts here then George does bring up a salient point - the difference in gain between the two designs does have to be taken into account in order to do an apples to apples comparison. As an example, what would you expect if you put a voltage divider on the output of the MyRef that would reduce the signal so that the effective gain was the same as the Mod68?
We can debate this ad nauseum or we can all just go on our merry way and acknowledge that the Mod86 is a nice low-noise design but the gain is at the lower end while the MyRef has slightly more output noise but is a higher gain amp.
Finally, I suspect that the ultimate determination of output quality is more influenced by the skills demonstrated by the constructor than anything else.
Take it easy, have fun and stay safe.
-bill
Great. But the MyRef FE started around 2012ish. Why not update it? Why just reproduce an old circuit with the design considerations that prevailed then?
The history and folklore make for good marketing copy, but unfortunately neither improve performance.
Thank you.
That's a nice academic exercise. It does the same as comparing based on the input-referred noise.
Which is basically what I wrote in my review.
Finally, I suspect that the ultimate determination of output quality is more influenced by the skills demonstrated by the constructor than anything else.
A good amp starts with a good design. A good implementation follows. We agree on that.
Cheers.
Tom
Thanks
Absolutely.
Yes but at the same time both I and (IMHO) Tom, with different approaches, aimed to give consistent results for builders.Finally, I suspect that the ultimate determination of output quality is more influenced by the skills demonstrated by the constructor than anything else.
I've did extensive research and listening tests on parts producing a strict BOM (or better two, one industrial and the other using audiophile parts) while Tom focused on solid engineering practices and also with a standard BOM.
Tom, an important part of the Paravicini's amplifiers sound that enchanted Mauro derived by the use of the LM318 which have a distinct sound signature (or distortion profile if you wish), sweet and reminiscent of tube sound.
So, designing the original My_Ref, he changed the output FETs with the LM3886 (tried first with TDAs without success) but the sound signature has been preserved, how? Thanks to the LM318.
Musical Fidelity improving their design during time changed the LM318 power supply regulators and this approach gave results also to some original My_Ref builders.
So, after some time Mauro decided to explore that path and improved the My_Ref with better power supply, a DC servo and a double current pump to have more power, this was the My_Evo (no more a public design).
The My_Evo had much lower distortion and much better sound.
Sadly Mauro made a single limited batch of boards and I wanted to have one.
It's for this reason that the My_Ref FE has born, I've made my evolution of the My_Ref following same Mauro's goals for the My_Evo (but omitting the DC servo and the second pump) and obvioulsy a PCB.
See: https://www.diyaudio.com/forums/chi...-edition-help-pcb-evaluation.html#post2720683
The LM318 has never been in discussion since I loved (and still love) the My_Ref sound signature.
JosephK has experimented with other opamps and the necessary compensation adjustements leading to even less distortion and improved sound quality BUT with a different sound signature.
In any way the design and performance of the FE has been vastly improved over the original My_Ref, equaling the My_Evo one.
This despite preserving an obsolete and noisy part like the LM318, which however gives the My_Ref his peculiar flavour.
The next step will be the noisy LM318 input stage replacement with discrete parts (like Musical Fidelity did in the A370 MK2) leading to further performance improvement.
Attachments
If "must contain an LM318" is one of your design constraints, then you did the best you could with it. That said, the LM318 is the bottleneck in terms of the performance of the MyRef FE, so if improving performance becomes a priority you know where to start.
The MyRef FE is different than above schematic in that the LM3886 in the MyRef FE is configured as a Howland current pump. So you're colouring outside the lines already...
Tom
The MyRef FE is different than above schematic in that the LM3886 in the MyRef FE is configured as a Howland current pump. So you're colouring outside the lines already...
Tom
I know, and I've also an upgrade path as depicted above
I'm not so expert and neat... Mauro did it.The MyRef FE is different than above schematic in that the LM3886 in the MyRef FE is configured as a Howland current pump. So you're colouring outside the lines already...
Tom
If You observe better, You will notice that the Musical Fidelity output stage is a common source complementary push-pull design. That is, the load sees directly the drain.
What You have in front is a discrete voltage controlled current source. It is the load itself which does the final I/V conversion, and the output voltage that has developed on the speaker itself is taken as input to the feedback loop.
The output stage is current output, high impedance.
It does not absorb the return EMF signal from the speaker, to compensate for it is left totally for the feedback controlled loop.
The speaker is active part of the feedback loop.
This is the spark of genius of de Paravicini, and this is the most characteristic streak of the design.
Which had been correctly identified by Mauro Penasa, who had been no less genius, and faithfully reproduced it in his twist on the original design, by the application of a chip power opamp based bipolar voltage controlled current source. He had chosen the Howland current pump as the most appropriate topology.
So, it's on the contrary: not colouring outside the lines, it is the utmost fidelity to the core of the design principle.
If You observe better, You will notice that the Musical Fidelity output stage is a common source complementary push-pull design. That is, the load sees directly the drain.
What You have in front is a discrete voltage controlled current source. It is the load itself which does the final I/V conversion, and the output voltage that has developed on the speaker itself is taken as input to the feedback loop.
The output stage is current output, high impedance.
It does not absorb the return EMF signal from the speaker, to compensate for it is left totally for the feedback controlled loop.
The speaker is active part of the feedback loop.
This is the spark of genius of de Paravicini, and this is the most characteristic streak of the design.
Which had been correctly identified by Mauro Penasa, who had been no less genius, and faithfully reproduced it in his twist on the original design, by the application of a chip power opamp based bipolar voltage controlled current source. He had chosen the Howland current pump as the most appropriate topology.
So, it's on the contrary: not colouring outside the lines, it is the utmost fidelity to the core of the design principle.
Nah. The feedback is on the inside of the output resistor. It's a opamp driving a grounded base amplifier. That turns the output of the opamp into a current which drives the driver stage and output stage. That's an old trick. National Semiconductor featured the circuit in an app note many moons ago.
The only "special sauce" I see is that the input stage of the LM318 has been replaced by a pair of discrete devices.
Tom
The only "special sauce" I see is that the input stage of the LM318 has been replaced by a pair of discrete devices.
Tom
I've attached the National Semiconductor circuit. It's not an exact match for the Musical Fidelity amp, obviously, but the overall topology is very similar. You can find the full app note here: https://www.ti.com/lit/an/snoa600b/snoa600b.pdf
It's from 1981...
Tom
It's from 1981...
Tom
Attachments
Last edited:
Furthermore: As Dario has mentioned, the original Penasa design had been presented here in 2005:
My "audiophile" LM3886 approach
It got grabbed up quickly by young Russ White, who had developed his own pcb layout version of a single channel. (the original design by Penasa had been a complete stereo amplifier) (many of us had built that)
Effectively that pcb became the first 'product' of Twisted Pear.
Somehow not totally happy, Mauro later on had developed a further version, the Myref Evolution, which design had introduced the parallel current pump structure, a different power supply dedicated to the control opamp in the loop, the clipping control, and a secondary DC servo control loop.
That design had never been released to the public domain.
And here enters Dario, who had convinced Mauro to let it be released to the public a simplified (but quite faithful) variation of the Evolution design - and it become the FE edition.
My "audiophile" LM3886 approach
It got grabbed up quickly by young Russ White, who had developed his own pcb layout version of a single channel. (the original design by Penasa had been a complete stereo amplifier) (many of us had built that)
Effectively that pcb became the first 'product' of Twisted Pear.
Somehow not totally happy, Mauro later on had developed a further version, the Myref Evolution, which design had introduced the parallel current pump structure, a different power supply dedicated to the control opamp in the loop, the clipping control, and a secondary DC servo control loop.
That design had never been released to the public domain.
And here enters Dario, who had convinced Mauro to let it be released to the public a simplified (but quite faithful) variation of the Evolution design - and it become the FE edition.
Myself, having a view of the design a bit differing from that of Dario, had started in 2015 to experiment with more modern, 'up to date' substitutes in the control opamp position.
That is, I appreciate very much Dario's faithful conservative method. He is the 'keeper of the flame' for the original sound represented by the original choices.
Me, just out of curiosity, wished to try how this same principle could work with more recent units in the control opamp position.
I have a faint impression that originally the choice of de Paravicini had been the fastest, high slew rate but still precision, most evoluted opamp in it's age. (LM318)
If we think of it like that, it could be changed to the present equivalents today..
Doing this, I think to have discovered little further tweaks like the elimination of the servo loop and further tweaking the compensation scheme.
So that upgrading of the design had happened already some years ago and free to be tried by the interested.
As well as the original design concept so faithfully maintained by Dario.
That is, I appreciate very much Dario's faithful conservative method. He is the 'keeper of the flame' for the original sound represented by the original choices.
Me, just out of curiosity, wished to try how this same principle could work with more recent units in the control opamp position.
I have a faint impression that originally the choice of de Paravicini had been the fastest, high slew rate but still precision, most evoluted opamp in it's age. (LM318)
If we think of it like that, it could be changed to the present equivalents today..
Doing this, I think to have discovered little further tweaks like the elimination of the servo loop and further tweaking the compensation scheme.
So that upgrading of the design had happened already some years ago and free to be tried by the interested.
As well as the original design concept so faithfully maintained by Dario.
I read Penasa's original document when I was writing my review of the MyRef FE. You will find a link to it in the introduction section.
The folklore and history is nice, but ultimately what matters – at least to me – is the end-to-end system performance. I appreciate the backstory, though.
I'm still not convinced that the Musical Fidelity design is a trans-conductance amp (schematic in Post #189). You're controlling the voltage at the 'inside' of the output resistor (4x0.47 Ω in parallel) with no feedback on the 'outside' of the resistor. Compare that with attached from Penasa's MyRef design doc.
I think you'll find that if you simulate (or measure) the output impedance of the Musical Fidelity amp, you'll find that it is low.
The output impedance of the MyRef FE is relatively low as well (see attached).
Tom
The folklore and history is nice, but ultimately what matters – at least to me – is the end-to-end system performance. I appreciate the backstory, though.
I'm still not convinced that the Musical Fidelity design is a trans-conductance amp (schematic in Post #189). You're controlling the voltage at the 'inside' of the output resistor (4x0.47 Ω in parallel) with no feedback on the 'outside' of the resistor. Compare that with attached from Penasa's MyRef design doc.
I think you'll find that if you simulate (or measure) the output impedance of the Musical Fidelity amp, you'll find that it is low.
The output impedance of the MyRef FE is relatively low as well (see attached).
Tom
Attachments
Tom,
Can we agree in that the MF amp output stage consists of 3 or 5, parallel couple of push-pull, complementary (mos) FETs with output taken from the drain?
That is, those FETs are operating in 'common source' topology.
""
In electronics, a common-source amplifier is one of three basic single-stage field-effect transistor amplifier topologies, typically used as a voltage or transconductance amplifier."" wikipedia.
So they operate as a voltage controlled current source.
Before feedback, that is the output signal. Current.
This also means, that for external signal drive, such as motional EMF from the speaker, the output stage does not absorb anything, shows high impedance.
this current output then feeds the load (speaker).
It gets converted (a voltage drop develops on the speaker impedance)
This becomes the feedback signal to the system.
After this feedback loop is closed, the complete amplifier is a classic, low output impedance driver.
External observer will not quickly notice a difference.
But the internal workings are very different. The mechanisms are different, and behaviour are different.
For example, output impedance is low, but much higher than in case of a classic design with EF output drive.
Compare the mod86 and myref FE 'damping factor' values.
Can we agree in that the MF amp output stage consists of 3 or 5, parallel couple of push-pull, complementary (mos) FETs with output taken from the drain?
That is, those FETs are operating in 'common source' topology.
""
In electronics, a common-source amplifier is one of three basic single-stage field-effect transistor amplifier topologies, typically used as a voltage or transconductance amplifier."" wikipedia.
So they operate as a voltage controlled current source.
Before feedback, that is the output signal. Current.
This also means, that for external signal drive, such as motional EMF from the speaker, the output stage does not absorb anything, shows high impedance.
this current output then feeds the load (speaker).
It gets converted (a voltage drop develops on the speaker impedance)
This becomes the feedback signal to the system.
After this feedback loop is closed, the complete amplifier is a classic, low output impedance driver.
External observer will not quickly notice a difference.
But the internal workings are very different. The mechanisms are different, and behaviour are different.
For example, output impedance is low, but much higher than in case of a classic design with EF output drive.
Compare the mod86 and myref FE 'damping factor' values.
One of Mauro's explicit design targets was to 'linearize' the output impedance of his design.
That is, avoid that classic upward curve with frequency.
Obviously this could be done inside limits of physics: he could not bent downwards the high frequency part, but could equalize the low frequency behaviour.
See below the comparison between Mod86, Myref FE output impedances.
That is, avoid that classic upward curve with frequency.
Obviously this could be done inside limits of physics: he could not bent downwards the high frequency part, but could equalize the low frequency behaviour.
See below the comparison between Mod86, Myref FE output impedances.
