NC252MP (class D) vs. A250W4R (classAB) burst measurements into 4ohm//2.2uF load

[Hayabusa]

Guess I need to replace the NC502MP/NC252MP setup I've been using.

And NC400?

 

[pma]

After reading this thread, I would now choose a NCx or Nilai.
@pma clearly proved that there is a design issue with the NC design connected to a complex load.

Older line - Ncore - is definitely not bad, but I am sure it is good to be honest and to show not only good points, but also weaker points. Nilai and NCx seem to be better in handling a complex load (to SOTA parameters) and I am sure it is due to re-designed high-order feedback loop. It seems to me that Ncore loop was designed only with respect to resistive load and with resistive load it works well. But it is a class D, there is a switching frequency, FB is taken from output behind the LC filter (that reduces switching frequency content to acceptable level) and thus the FB action and transfer function is affected by the load, more affected by complex load than by a resistor. It is interesting to mention that the previous UcD180HG has almost no issue with the 2.2uF load, as I have already shown, and its feedback design took it probably into account. We have seen that in case of capacitive load component of 200nF - 4uF the load negatively affects resulting noise and distortion of the NC252MP in the way they become audible. Yes, very few speakers would bring such impedance to the amp output, however, the engineering design should count with such possibility.

When I do measurements and post a review here I have one big advantage/disadvantage at once. I do not rely on components sent by manufacturers or members, but I test my own components or I have to buy them in order to test them. That is why I bought AIYIMA A07, that is why I bought UcD180HG and NC252MP, that is why I bought Topping D10s and some other components. From those listed, I give my kudos only to the Topping D10s and in a way also to UcD180HG. I do not need to be polite in engineering technical terms, because I am not obliged to anyone, person or company, for providing me with the test object. I pay everything by myself so I feel free to post my findings. I do not make a living with audio, not even partially. I did it partially but finished it 11 years ago.

 

[sarieri]

But can't it just be resolved by using data from a speakers' impedance / phase diagram?

View attachment 274246

View attachment 274247

Let's focus on the circled area.
f = 110 Hz
Z = 8 ohm
phase = 50 ° cap.

Resolved vector =
6.13 Vert. // 5.14 Hor.
=> Xc = 5.14 ohm

C = (10^6/(2*3.14*110*5.14)) = 281.6 microFarad

Maybe my math is a bit rusty. It's been a while since I juggled vectors and this was in electrical grid and cabinet design so maybe some things can't directly be translated into audio and maybe I read the graph wrong.

Could someone explain why electrostat speakers are more likely to have issues when matching with NC252MP?

The calculation shown here takes a vertical capacitance into consideration comparing to the baseline model(pure resistive resistor). But to me, it seems that as long as the frequency plugged in is low, the calculated capacitance is gonna be quite large(hundreds of uF), even when Z is larger. For example, a woofer driver with impedance and electrical phase attached below has about 264uF calculated using this method. Following this logic, I don’t really understand why this makes electrostat special.

I’m asking this only out of intellectual curiosity.

 

[NTK]

Could someone explain why electrostat speakers are more likely to have issues when matching with NC252MP?

The calculation shown here takes a vertical capacitance into consideration comparing to the baseline model(pure resistive resistor). But to me, it seems that as long as the frequency plugged in is low, the calculated capacitance is gonna be quite large(hundreds of uF), even when Z is larger. For example, a woofer driver with impedance and electrical phase attached below has about 264uF calculated using this method. Following this logic, I don’t really understand why this makes electrostat special.

I’m asking this only out of intellectual curiosity.

What you showed is the impedance of a woofer, not an electrostatic speaker. Please see post #15 for the example impedance curves of some electrostatic speakers (the impedance curve of the Sound Lab A-1 is shown below). At 30 kHz, this electrostatic speaker is effectively a short (close to 0 Ω). (The impedance of this speaker is about the same as a 5 μF capacitor at ~10 - 20 kHz.) It is a very difficult load for a lot of amplifiers, not just the NC252MP.

Sound-Lab A-1 electrostatic loudspeaker Measurements

Sidebar 6: Measurements

www.stereophile.com

 

[sarieri]

What you showed is the impedance of a woofer, not an electrostatic speaker. Please see post #15 for the example impedance curves of some electrostatic speakers (the impedance curve of the Sound Lab A-1 is shown below). At 30 kHz, this electrostatic speaker is effectively a short (close to 0 Ω). (The impedance of this speaker is about the same as a 5 μF capacitor at ~10 - 20 kHz.) It is a very difficult load for a lot of amplifiers, not just the NC252MP.

Sound-Lab A-1 electrostatic loudspeaker Measurements

Sidebar 6: Measurements

www.stereophile.com

But 5uF is low comparing to the calculation I did. I thought this thread is showing that capacitance of speaker driver is a hard load for amp and therefore pure resistor load is not an ideal load to test speaker. But now the impedance/phase graph here shows that at 20k, even though the impedance is very low(lower than 2ohm), but since the frequency is very high, the capacitance is still low(less than 5uF) comparing to the calculation I did on that dynamic woofer driver.

If the point is just low impedance implies hard load, then aren’t we back to the original spot that more resistor value of load need to be chosen to test a amp?

Or maybe the model including only a vertical capacitor is still an oversimplification and thus doesn’t imply much about the situation?

 

[MaxwellsEq]

But 5uF is low comparing to the calculation I did. I thought this thread is showing that capacitance of speaker driver is a hard load for amp and therefore pure resistor load is not an ideal load to test speaker. But now the impedance/phase graph here shows that at 20k, even though the impedance is very low(lower than 2ohm), but since the frequency is very high, the capacitance is still low(less than 5uF) comparing to the calculation I did on that dynamic woofer driver.

If the point is just low impedance implies hard load, then aren’t we back to the original spot that more resistor value of load need to be chosen to test a amp?

Or maybe the model including only a vertical capacitor is still an oversimplification and thus doesn’t imply much about the situation?

The phase difference is also tough. The current leads the voltage, which it doesn't do with a pure resistive load.

 

[WaterFlame]

The same I can say about the Nilai .

BTW, 4R7//2.2uF to 4R7//33uF is no especially difficult load at 1kHz as a test tone. Any amp should be able to pass it and any good class AB amp passes it. What it in fact tests is a FB stability, loopgain design.

View attachment 292374

Hi, I can send you my Nilai stereo for testing that you can torture test for about a month. Drop me a PM. I an curious to see how 19-20khz imd or higher than 1khz fft would look like under different loads.

 

[pogo]

Looked at the NAD M33 and M10 and none had complex load traces?

M33 = M23: Link

 

[sarieri]

M33 = M23: Link

Correct me if I’m wrong but to me that test only shows that the power capability does not change with capacitive resistor, it’s not showing anything about SINAD getting worse or staying at the same level right?

 

[pogo]

Correct.
But not all Purifi implementations will behave the same way. See, for example, the PSU section of the T+A M200.

 

[sarieri]

Correct.
But not all Purifi implementations will behave the same way. See, for example, the PSU section of the T+A M200.

Are you saying that different implement(with different PSU) will result in different capability of handling capacitive loads? So does this mean that capacitive load will put more pressure on psu, so “bigger” psu will in some sense alleviate the problem?

 

[Nyattnyatt]

I don’t know much, but I am curious. If you were to add a second filter after the feedback loop, could you lesson or stop the oscillations from the add’l capacitance of the revised load (4 ohm//2.2 uF)?

 

[dguillor]

I don’t know much, but I am curious. If you were to add a second filter after the feedback loop, could you lesson or stop the oscillations from the add’l capacitance of the revised load (4 ohm//2.2 uF)?

If you did, it would likely change the frequency response of the speakers.

 

[Nyattnyatt]

Even if the filter is set high? Similar to what’s already used in the amp? Say 12 dB/ octave at 50kHz?

 

[dguillor]

Even if the filter is set high? Similar to what’s already used in the amp? Say 12 dB/ octave at 50kHz?

The required compensation will be different for every model of loudspeaker, and I don’t think that it will correspond with the amp output filter.

 

[Head_Unit]

An unplanned oscillation may damage tweeters if they actually are able to attempt to play the frequency. I think it's a long shot if it's way beyond their normal operating range.

The point was that class D amps are always oscillating in ultrasonic range. We don't hold that against them because the frequency is so high. And level very low relative to max power.

Forgive if this was answered in a much later page, but
- The frequency was not so high. 450k --> 66 k, still above human hearing range but not unthinkable to perhaps intermodulate? Plus what is the impedance of electrostatics up there? And...
- 1.7 volts!?! That might not fry all tweeters instantly however it's not insignificant.
On the one hand I can't offhand agree that in actual use the results shown here are a total kiss of death as @pma seems to feel, especially since most people listen not so carefully at not so loud levels. BUT here at ASR I consider the difference versus some other forums is to hold a beady eye towards well-done engineering that does NOT have weird problems. In this case it seems the NC is not really suitable for electrostats compared to a (beefy!) AB.
- I think it's fine to be bothered by problems like this even if the real world relevance is murky. Hey this is the same forum worshipping noise and distortion far below audible limits, isn't it?

 

[amirm]

- 1.7 volts!?! That might not fry all tweeters instantly however it's not insignificant.

Huge number of class d amps are sold in both consumer and professional spaces. No reports have come of any such problems. There is natural roll off in speakers at these frequencies.

 

[MaxwellsEq]

There is natural roll off in speakers at these frequencies.

And almost all (normal) content has a strong downwards slope from the middle frequencies to the highest frequencies, so there is a lot less musical energy being put into the system.

Sure, if there is a self-oscillating resonance, content is irrelevant.

But as Amir and everyone else keeps pointing out, there's simply no recorded evidence that Class D is damaging anything.

 

[atmasphere]

I don’t know much, but I am curious. If you were to add a second filter after the feedback loop, could you lesson or stop the oscillations from the add’l capacitance of the revised load (4 ohm//2.2 uF)?

Again, what you are seeing is how the amp damps the oscillation starting up. It has to do this with every iteration so it looks like an oscillation.

This amp is of a variety called 'self oscillating' which is to say the amp is already oscillating- the oscillation is its switching frequency.

The way this is done is by adding so much feedback that the phase margin of the design is exceeded by the feedback (IOW, the feedback, instead of being negative at audio frequencies, is phase shifted so far that its positive at the oscillation frequency). So the amp goes into oscillation and begins switching as soon as its turned on.

An advantage of doing this is the enormous amount of feedback that can be employed.

The feedback network is a special design. The formula for it is known as the 'oscillation criteria' and has 7 or 8 variables at least. So its not a trivial matter to change it; part of the issue is overload of the amp (at which point it stops switching). When the amp starts up again, it must find the same solution (oscillation frequency) or Bad Things happen. The oscillation criteria is set up so there is only one oscillation solution.

But as I explained in my first post on this thread, the issue seen with the ESL isn't actually an issue. ESLs are a special challenge for class D amps because the output filter of the amp has an inductor in it and the ESL is highly capacitive. The two form a resonant frequency. If the amp does not have enough feedback or has none at all, it will oscillate at that frequency. We can see in this case it isn't doing that, since that amp is no-where near full power. This is simply a signal generated every time the inductor interacts with the load's capacitance, and then the feedback of the amp shutting down that resonance (which it can't do at zero Volts output). So you see a little signal but its not an oscillation in the traditional sense.
********************************

42 pages and this conversation is still going on about a thing that's of no concern??

Tell you what. Find a zero feedback class D amp (they exist), put it on that load and then see what happens. You might want to have your safety glasses on.

 

[Head_Unit]

Huge number of class d amps are sold in both consumer and professional spaces. No reports have come of any such problems. There is natural roll off in speakers at these frequencies.

there's simply no recorded evidence that Class D is damaging anything.

42 pages and this conversation is still going on about a thing that's of no concern??

I find it amazing that in a forum where surely many would feel an Apple dongle is inadequate and worship some Topping DAC or AHB2 amp etc with S/N of 120 dB, that so many say 1.7V is irrelevant. Please note I am NOT saying it is a problem...but it seems hypocritical to diss a dongle whose performance is most likely 99% transparent to 99% of the people with 99% of the music-and then dismiss the 1.7V etc as negligible.
- Even if it's not a real world problem, wouldn't you prefer the amp NOT do this?
- The bigger question, what happens with other class D amps? Is this inherent to the class? (based on @atmasphere's post I'd think so, unless the design extends the feedback loop out past the filter as a colleague of mine was one of the first to do back in the day. Then I'd *think* this doesn't happen?)

Anyway again a big kudos to @pma for doing these investigations, it's always interesting to find odd behaviors, which sometimes lead to insight which leads to improvement.