New Class A, Super-A, Non-Switching : need a revival ?

[bordodynov]

I solve the problem easier.
LoopGain = V (Out) / V (+, -), where V (+, -) there is a voltage difference at the inputs (+ and -) of the differential input amplifier.
See Blameless 3X3EF_new2AC.asc

 

[AndrewT]

post161

I finally got it going with the LTSpice. It was my fault.........I've been spending a lot of time on this the last few days, the simulation is so so slow. Just made a stupid mistake.

Anyway, I think I was right, the simple circuit is stable from my understanding on the simulation on the open and closed loop gain phase plot. I attached the new file with two identical circuits, one with closed loop gain of 21( the right). I change R33 on the left circuit from 1K to 0.1ohm ( in the feedback gain setting) to simulate openloop gain. I plot the gain and phase plot of both the open loop ( green) and closed loop (yellow/light grey).

You can see it's a single pole roll off from about 100Hz to about 5MHz before going to 2 poll roll off and the phase gone south. BUT we have a gain of 21, at the -3dB point of the amplifier ( vertical red line labeled -3dB), the phase shift is only -70deg. I have 180-70=110deg phase margin. The phase at 0dB is irrelevant as circled "Don't care".

The circuit should be stable. But then again, this is on paper/computer. Real circuit might be different.



Hi MiiB,

I try to look for Tien probe, I don't quite follow. But I think the way I make R33 to 0.1ohm should work close enough. When I search for Tien probe, I saw description using Spice derivitives which I don't understand. I am not very good with simulations.

Thanks

"Don't care" !
you should care.
Where the open loop crosses the 0dB gain line the slope of the roll off should be less than approx. -9dB/Octave. A bit steeper than single pole roll off. But this slope gives stability margins that are far too low. a tiny reactive load will push the amp into oscillation. Better to aim for -6dB/Octave slope at the 0dB gain crossing. But this gives a phase margin of ~45°. Again low levels of parasitic capacitance on the output pushes the amp into extreme overshoot or oscillation. You need to add a zero that operates near the 0dB cross over frequency to flatten the slope at crossover.
Your amplifier (if you have simulated it correctly) shows the gain plot steepening at the crossover frequency. This is exactly the opposite to what you require for good stability margins.

 

[Bob Cordell]

Nothing is wrong with the inductor, just it is not accurate enough. This method is similar with the one used in LoopGain.asc(also explained in the LTspice examples), where you get the loop gain instead open loop gain as with the inductor method. The Tian probe is explain in the LoopGain2.asc :

Here the open loop gain is determined from the closed loop system[1].
The open loop gain can be plotted by plotting the quantity:

-1/(1-1/(2*(I(Vi)@1*V(x)@2-V(x)@1*I(Vi)@2)+V(x)@1+I(Vi)@2))

Alternatively, you add the following line to your plot.defs file:
.func T.et.al() -1/(1-1/(2*(I(Vi)@1*V(x)@2-V(x)@1*I(Vi)@2)+V(x)@1+I(Vi)@2))
And then plot simply T.et.al()

This is an improvement over the technique shown in LoopGain.asc
because it (i) accounts for reverse feedback(it doesn't even
matter if you reverse the direction of the probe -- you still compute
the same open loop response) and (ii) the inserted probe elements
result in a smaller, sparser circuit matrix.

Depends on the word "enough". When just looking at the global loop gain and connecting the inductor method as I described earlier, and when the output impedance of the amplifier is very low compared to the value of the feedback resistor, the accuracy of the inductor method is more than enough, and it gives the loop gain (not the open loop gain). It fully accounts for the pole formed by the impedance of the feedback network and the input stage.

Cheers,
Bob

 

[dadod]

Depends on the word "enough". When just looking at the global loop gain and connecting the inductor method as I described earlier, and when the output impedance of the amplifier is very low compared to the value of the feedback resistor, the accuracy of the inductor method is more than enough, and it gives the loop gain (not the open loop gain). It fully accounts for the pole formed by the impedance of the feedback network and the input stage.

Cheers,
Bob

If you mean as you described in post #190, then to me it looks as LTspice LoopGain.asc(also explained in the LTspice examples). In this case you don't need an inductor. If I'm wrong could you put some picture how to do that.
Best wishes
Damir

 

[Bob Cordell]

If you mean as you described in post #190, then to me it looks as LTspice LoopGain.asc(also explained in the LTspice examples). In this case you don't need an inductor. If I'm wrong could you put some picture how to do that.
Best wishes
Damir

You always need an inductor to close the loop at DC.

In the arrangement I described, the loop is broken for AC by the inductor between the output node of the amplifier and the series feedback resistor. The node connecting the inductor and the feedback resistor is the node where the AC source signal is injected through a large coupling capacitor.

In this arrangement, the injected test signal must pass through the attenuation of the feedback network, so the gain to the output of the amplifier represents the loop gain.

Cheers,
Bob

 

[AndrewT]

Please attach your pics/info.

 

[dadod]

I tried here to simulate Loop Gain of my VFA with three different method, Tian, Middlebrook and inductor probe. Here are the results, no big differences.

 

[dadod]

Now here are the same simulations but on my CFA. Maybe I am doing wrongly this simulations here, but I can't see what.

 

[dadod]

Now here are the same simulations but on my CFA. Maybe I am doing wrongly this simulations here, but I can't see what.

I expected differences for CFA between Tian probe and other simpler method, but did not expected so big differences, so I suspect I did something wrongly here.

 

[padamiecki]

but is nonswitching OPS useless outside of GNFB loop?
Is it possible to use it as just OPS?
How?

Hi
My findings in general are that most of NS ops are performing very well inside gnfb loop making the hf harmonics profile quickly descending but they have ugly behavior as unity gain buffers.
There is only one exception from this "rule": Krill ops.
Do you know other power NS unity gain buffers for ngnfb operation?

 

[wahab]

Now here are the same simulations but on my CFA. Maybe I am doing wrongly this simulations here, but I can't see what.

What must be plotted is the relative voltage and phase at the generator outputs, the amp act as a counter force in respect of the generator, so what should be computed is (output voltage/input voltage) , the output is taken at the amp output and the input is taken at the hot plug of the generator, wich is also the inverting input (before the feedback network) of the amp..

FI here a plot of one of your CFA design :

 

[vzaichenko]

Hi Pawel,

I have tested this configuration with NS-OPS outside the global loop:
http://www.diyaudio.com/forums/soli...e-old-ideas-1970s-ips-ops-66.html#post4648520

It worked well, demonstrating distortion low enough and pleasant sound, although a slight difference in bass reproduction was noticed, comparing to the same Front-end + OPS combination covered by the normal GNFB loop.

OPS board on the photo is an early prototype. The current version is modular - the front-end is a snap-on board, placed on to of the "carrier" OPS module.
See the first post of the thread linked above - there's a reference table with many things you may find interesting there

Cheers,
Valery

 

[LKA]

Hi
My findings in general are that most of NS ops are performing very well inside gnfb loop making the hf harmonics profile quickly descending but they have ugly behavior as unity gain buffers.
There is only one exception from this "rule": Krill ops.
Do you know other power NS unity gain buffers for ngnfb operation?

not NS, class AB mosfet unity gain buffer and HEC

 

[padamiecki]

Hi Pawel,

I have tested this configuration with NS-OPS outside the global loop:
http://www.diyaudio.com/forums/soli...e-old-ideas-1970s-ips-ops-66.html#post4648520

It worked well, demonstrating distortion low enough and pleasant sound, although a slight difference in bass reproduction was noticed, comparing to the same Front-end + OPS combination covered by the normal GNFB loop.

OPS board on the photo is an early prototype. The current version is modular - the front-end is a snap-on board, placed on to of the "carrier" OPS module.
See the first post of the thread linked above - there's a reference table with many things you may find interesting there

Cheers,
Valery

Thank you Valery for reminding,
indeed your version of NSops is not that bad, I tried to translate it into LTspice and the results are not that bad, would you check if there was no errors during writing the sch?
Attached asc for ltspicers

 

[vzaichenko]

Thank you Valery for reminding,
indeed your version of NSops is not that bad, I tried to translate it into LTspice and the results are not that bad, would you check if there was no errors during writing the sch?
Attached asc for ltspicers

Schematic looks fine, however, performance strongly depends on both static bias (R20) and dynamic bias clamping (R14) settings.

Attached is the real 20KHz spectrum, measured on the live prototype (50W @ 8 ohm). The dominant 2-nd harmonic stays below -80db (0.001%).
Most likely, R20 and R14 are not set to the optimal values in your sim

 

[AndrewT]

Thank you Valery for reminding,
indeed your version of NSops is not that bad, I tried to translate it into LTspice and the results are not that bad, would you check if there was no errors during writing the sch?
Attached asc for ltspicers

Can I assume these are the output stage distortion figures for a stage that does not have feedback wrapped around it?

 

[vzaichenko]

Andrew, that's right - global loop does not cover the OPS.

Attached are simulation results of the well-tuned NS-OPS (spectrum and table form). It shows slightly higher distortion, than the live prototype
BTW, there's a typo mistake in my previous post: -80db = 0.01%

 

[vzaichenko]

Pawel, I see now - in your LTSpice model the low-impedance sources virtually connect both PD+ and ND- inputs (top and bottom sides of the bias spreader) to Ground - the OPS is zero-biased in this case, regardless of the trimmers position.

In my simulation I drive the OPS from the low-distortion IPS, so OPS is working in the normal impedance conditions, being driven as usual.
I can publish the model, showing these results, but is will be a Multisim one.

 

[padamiecki]

I can publish the model, showing these results, but is will be a Multisim one.

yes, please,
but anyway I have the impression that, the sims are worse than real life, again

 

[Havenwood]

Hi, what about Super Stability Non-Feedback amplifier by Denon?
The amplifier that have this technology is POA-8000.

It´s look like feedfoward amplifier. I never hear before about Denon´s non-switching technologies.