No-global-loop amplification

Hi All,

A Simpelstark amplifier has been prototyped, tested, measured, auditioned.
See the links below:

1) Version 1.1 schematic and PCB views;
2) Spectrum measurements;
3) Assembled board view.

I will most likely try to update the layout, designed by Alex MM, and test it later on.

Cheers,
Valery
_________________________________________________________
Original message - Einbahnstark, etc.

As many times discussed, feedback is inevitable, being present everywhere around us. We also use it every time we design an audio amplifier.
Important point – the way we use it.

No doubt, it’s possible to design a good (or not so good) amplifier with both global NFB loop in place or no global loop at all – in this case, several local NFBs are utilized for controlling important parameters.

After testing / auditioning different designs throughout the recent years, I come to conclusion I like the way the good no-global-loop amplifiers sound. Those amplifiers are well-known in the world of vacuum tubes for decades. However, my focus are solid-state amplifiers and hybrids, designed with series of local feedback (preferably covering a single stage) loops in mind. Damir and other members presented a number of excellent works in this area.

In this thread, I’d like to share my experience in designing an important part of any amplifier – the front-end, providing all the voltage gain – utilizing only local feedback loops.

Picture 1 illustrates possible arrangement in principle.
Input stage (transconductance, Gm = 0.5 mA/V) is a heavily degenerated CE circuit – that’s where the most of distortion is born, however at the volume of degeneration used, it’s low enough for the purpose.
Emitter degeneration provides a series-series feedback, providing the desired transconductance ratio and linearizing the transfer function.
2-nd stage is a high-precision cascaded current mirror with current gain of around 2x – highly linear mirroring with a little bit of gain.
Finally – passive transimpedance conversion – resistor load – also very linear, assuming both the output impedance of the current mirror and input impedance of the next stage are high enough.
To satisfy the latter assumption, we have to use a high-quality unity-gain buffer between this resistor and the OPS, as OPS’s input impedance, although it can be rather high, is normally more (especially in case of MOSFET drivers) or less capacitive and rather modulated (especially in case of BJT arrangement).
Small value capacitor provides minimal correction, eliminating possible frequency response artifacts above 1MHz.
It's easy to arrange a balanced input or a fully balanced architecture here - no global loops make it beautiful (just servo needs a bit of redesign).

Picture 2 illustrates practical implementation of the above ideas.
The circuit demonstrates pretty good performance – although, I’m presenting simulated results here, I have measured and listened a rather close (a bit less sophisticated) design, showing similar results and sounding just great.

Pictures 3, 4, 5 and 6 illustrate:
- Spectrum at 1 KHz (20V RMS0;
- Spectrum at 20 KHz (20V RMS);
- Square wave response at 20KHz (60Vpp);
- Frequency response.
Harmonics 2, 3 and very little bit of 5. Looks like a spectrum profile from the vacuum pentode 😊

There may be many different implementations of the stages, including jFETs at the input, folded cascode instead of current mirror, simpler unity gain buffer, etc.

In terms of OPS selection – it must be a low-distortion one. Class A / AB (well-designed CFP, properly biased EF, MOSFET outputs, some error-correction options, non-switching arrangements, etc.).
I used NS-OPS, showing THD < 0.05% open loop throughout the whole audio bandwidth.

What do you think?

Cheers,
Valery

P.S. A simpler no-global-loop headphone amp design is presented here:
Aureaux high-quality no-global-loop headphone amp
 

Attachments

  • @No-Global-Loop-Amplification.JPG
    @No-Global-Loop-Amplification.JPG
    108.3 KB · Views: 6,808
  • Einbahnstark_2.0-Sch.JPG
    Einbahnstark_2.0-Sch.JPG
    263.9 KB · Views: 7,175
  • Spectrum_1KHz_20V_RMS.JPG
    Spectrum_1KHz_20V_RMS.JPG
    109.1 KB · Views: 6,515
  • Spectrum_20KHz_20V_RMS.JPG
    Spectrum_20KHz_20V_RMS.JPG
    109.4 KB · Views: 5,880
  • Square-wave_20KHz_57Vpp.JPG
    Square-wave_20KHz_57Vpp.JPG
    89.7 KB · Views: 5,801
  • xFrequency-response.JPG
    xFrequency-response.JPG
    90.9 KB · Views: 1,089
Last edited:
Hi, Valery!

Nice approach and good linearity.

It also could be done slightly simpler, without highly degenerated diferential stages, which in turn adds noise.

Something like this:
IMG_6487.JPG

R4 define load of the input diamond buffer and also current consumed from rails.
Full Wilson mirrors reflects consumed current to I/V load.
 
Last edited:
Hi, Valery!

Nice approach and good linearity.

It also could be done slightly simpler, without highly degenerated diferential stages, which in case adds noise.

Something like this:
View attachment 637873

R4 define load of the input diamond buffer and also current consumed from rails.
Full Wilson mirrors reflects consumed current to I/V load.

Yep - diamond is a good one, I'm going to play with it :yes:
Using dfferential pairs makes it easy to arrange a balanced input.
On the other hand, we can try a dual-diamond configuration, for example.

Thanks for inspiration ;)
 
Using dfferential pairs makes it easy to arrange a balanced input.


Yup.
But the most valueable property of the balanced input is a good CMRR. It must reach at least 60 dB at 20 kHz.
So there are next thoughts.
1. Source output impedances must be aligned to 0,1% or better.
2. Balanced receiver must be placed before volume regulator and better to before input selector
3. Balanced receiver must be builded around 0,1% or better resistances, may be matched matrix.

On the other hand, we can try a dual-diamond configuration, for example.


Yup. And load them to each other.
With proper layout technique you could achieve low parasitic capacitance to ground and improve CMRR accordingly.
 
Agree - "external" balanced receiver will give better result in terms of CMRR.
We use THAT-1206 ones in our preamp - those are fantastic.

Exactly, dual-diamond configuration with the diamonds, loaded to each other. Rather simple and efficient. Also, a good starting point for a fully balanced setup (including two OPS modules).
 
What do you think?

Cheers,
Valery

Hi Valery, I think it's interesting to read your philosophical ideas, not many lay out their philosophies regarding audio amplifier designing as you have done here, but rather plant out tiny tidbits "here and there" so one have to spend quite some time here on DiyA in order to get some idea what each ones philosophy is, even if I don't have so much to add I can only say that I concur with many of your sound philosophies. :)

Regards Michael
 
Valery, as always you have very good ideas how to design different amplifiers. You use diamond OPS here, how you do thermal compensation here, I never built any diamond power OPS yet (only for preamp)? How you get THD 0.002% if OPS alone has 0.05%?
I designed no GNFB but never built yet, to many project, have you built that one, if yes, how you like the sound?
This is my NGNFB amp and it uses current mirror with current gain of 2.12 GainWire-NGNFB-classB-PowerAmp
Those are my preamps with no GNFB, also uses current mirror. http://www.diyaudio.com/forums/anal...onveyor-voltage-amplifier-24.html#post3447084 and http://www.diyaudio.com/forums/solid-state/235695-nfb-line-amp-gainwire-mk2-97.html#post4629715
Maybe you can get some ideas from it, like to use supper pairs.

P.S. pity you don't use LTspice ;)
 
Exactly, dual-diamond configuration with the diamonds, loaded to each other. Rather simple and efficient.


Valery, are you ever tried even simpler four-quadrant self-biased cascoded IPS like this?
IMG_6491.JPG

Or matched arrays like THAT340/ALD1103? Last one have dedicated substrate connection.

THAT1206 seem to have 7 kOhm at input in-series and could degrade noise performance in low-level applications.
 
Last edited:
Valery, as always you have very good ideas how to design different amplifiers. You use diamond OPS here, how you do thermal compensation here, I never built any diamond power OPS yet (only for preamp)? How you get THD 0.002% if OPS alone has 0.05%?
I designed no GNFB but never built yet, to many project, have you built that one, if yes, how you like the sound?
This is my NGNFB amp and it uses current mirror with current gain of 2.12 GainWire-NGNFB-classB-PowerAmp
Those are my preamps with no GNFB, also uses current mirror. http://www.diyaudio.com/forums/anal...onveyor-voltage-amplifier-24.html#post3447084 and http://www.diyaudio.com/forums/solid-state/235695-nfb-line-amp-gainwire-mk2-97.html#post4629715
Maybe you can get some ideas from it, like to use supper pairs.

P.S. pity you don't use LTspice ;)

Hi Damir,

Good to see you here :cool:
Part of the circuit, shown on the schematic, does not include OPS - diamond on the right is the unity gain buffer. Connecting the OPS right to the transimpedance stage results in significantly higher level of distortion because of OPS's input impedance modulation. Buffer does the job perfectly.
So, distortion figures shown on the pictures are true for the front-end only.
NS-OPS dictates overall level of distortion at around 0.05%, running open loop.

Cheers,
Valery
 
Valery, are you ever tried even simpler four-quadrant self-biased cascoded IPS like this?
View attachment 637998

Or matched arrays like THAT340/ALD1103? Last one have dedicated substrate connection.

THAT1206 seem to have 7 kOhm at input in-series and could degrade noise performance in low-level applications.

Hi Pavel,

Yes, I have used this kind of self-biased IPS arrangement, based on 2sk246/j103 pairs in one of X4 front-end options.

Noise - at low levels yes, but at line levels - no problem ;)
 
Diamondstark

Following Pavel's post, I have tested the concept with diamond IPS.

Initial setup did not include common base cascodes U8, U9 - overall distortion level was roughly 1/2 from what we've seen with differential pairs in place, still copying that initial profile.
The cascodes noticeably reduce 2-nd harmonic, improving overall performance, with 3-rd harmonic dominance. This configuration also shows rather good result in terms of IMD.

It's also cool to be able to use +/-15V regulated rails for IPS, allowing low-voltage complementary transistors.

Live performance is going to be slightly lower, strongly depending on components selection, as a number of important parts are either out of any NFB loop, or form the local loop itself.
 

Attachments

  • Diamondstark_1.0-Sch.JPG
    Diamondstark_1.0-Sch.JPG
    293.6 KB · Views: 1,794
  • Spectrum_1KHz_20V_RMS.JPG
    Spectrum_1KHz_20V_RMS.JPG
    85.3 KB · Views: 1,789
  • Spectrum_20KHz_20V_RMS.JPG
    Spectrum_20KHz_20V_RMS.JPG
    87.2 KB · Views: 1,503
  • Square-wave_20KHz_57Vpp.JPG
    Square-wave_20KHz_57Vpp.JPG
    50.7 KB · Views: 1,495
  • xIM_14-15KHz_20V_RMS.JPG
    xIM_14-15KHz_20V_RMS.JPG
    86.7 KB · Views: 1,632