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

[Sokel]

I know this continued discussion is frustrating for some. Especially given the brilliant and essentially "free to me" work from @amirm and others who have the equipment and patience to run tests on real world products without any reward from advertisers and manufactures. In my experience of HiFi I've never been so blessed with excellent information and guidance.

I have a mental model of a Class D (which is not the same as the classic block diagrams) of a black box that provides gain in the audio frequency bands superimposed with an RF device producing Radio Frequency signals which cannot be played by our speakers and which we cannot hear and which is mostly blocked from reaching our speakers by the output filter.

What this thread seems to be about is - perturbations to the RF amplifier - 1) whether a real world device can cause an interaction and 2) whether such an interaction perturbs the RF device, 3) how would we sensibly test and present this.

What is perhaps relevant is how many people are really satisfied with their Class D amps in the real world. Obviously, the vast majority of purchasers can not measure how well the amp works, but if there are real world perturbations to the RF device that folds artefacts such as noise and distortion back into the audio bands, we're not encountering complaints...

I hate to intervene to a highly technical debate but as is goes it's not technical any more (for some last posts at least with the name calling).

I only wish for all to be equally skeptical about any class or any damn little elecronic thing.
We want it to be our servant and praise our music,not the other way around.

If it fails to do so there are nice ways to set it on fire and dance around it,whatever that may be ,class D,class AB or class X.
I'm done reading this thread as it goes with brand and class lovers.

It was interesting,but...

 

[damonhill]

And voila, you inadvertently proved that indeed you don't understand. Last time I will suggest this before giving up that you have any actual curiosity: bone up on basic electricity, with particular attention to voltage dividers. If you can't or won't do that, you really have no business trying to interpret electrical data.

edit: I'd add in Thevenin. This is very basic stuff that will greatly enrich your understanding.

It's interesting that you keep trying to "play the man" as opposed to any cogent technical debate. Implying you know something, that you have an explanation, and yet it never comes.

So tell me, how would you have predicted the amp above might respond with a real speaker by just looking at the 4 and 8 ohm resistive plots?

 

[damonhill]

hate to intervene to a highly technical debate but as is goes it's not technical any more (for some last posts at least with the name calling).

Yes it is a shame some feel the need to do this.

 

[MaxwellsEq]

And voila, you inadvertently proved that indeed you don't understand. Last time I will suggest this before giving up that you have any actual curiosity: bone up on basic electricity, with particular attention to voltage dividers. If you can't or won't do that, you really have no business trying to interpret electrical data.

edit: I'd add in Thevenin. This is very basic stuff that will greatly enrich your understanding.

Sorry @SIY I don't get your point. I have a degree in Electronics and sat a few papers dedicated to complex circuit analysis using Thevenin and Norton equivalent circuits. I agree that, in theory, any circuit can be modelled in this way.

Here is what I think you are saying: the output impedance and the speaker load make (to a first order) a simple divider which you can do maths on. If the output impedance is zero, there will be no load dependency, for all non-zero (i.e. real-world) values, there is a load dependency. THEREFORE, if a frequency sweep into 4 and 8 Ohm loads demonstrates a wide variation, by simple deduction, we can assume a reactive (i.e. real-world) load will also have a dependency, so there is no need to measure with a complex load, because it's already clear there will be an impact. This I agree with. A complex load would just demonstrate the same or perhaps worse variation in frequency. The review just needs to flag this impact and advise people that the sound may be inconsistent, speaker to speaker.

But what happens if the device does not display any variation between 4 and 8 resistive loads. Can we safely deduce from this that a complex reactive load won't create frequency variations? Where I don't understand your point is that: in the real world, the output impedance is not purely resistive. Therefore I don't think it is safe to conclude that any real-world frequency impacts when driving reactive loads from a reactive output impedance are negligible. A Thevenin equivalent would demonstrate that there may be variations with reactive output impedances not demonstrated when one half of the circuit is purely resistive.

 

[damonhill]

Where I don't understand your point is that: in the real world, the output impedance is not purely resistive. Therefore I don't think it is safe to conclude that any real-world frequency impacts when driving reactive loads from a reactive output impedance are negligible.

Spot on. Amp output impedance isn't a constant WRT frequency, nor is it purely resistive.

The second part of my point is the clarity of information for the reader. Whilst the stereophile style of plots are only one specific complex load character, they do get the message across very simply how one amp differs from another in their response to real world loads. Something Amirs plots don't.

 

[SIY]

But what happens if the device does not display any variation between 4 and 8 resistive loads. Can we safely deduce from this that a complex reactive load won't create frequency variations? Where I don't understand your point is that: in the real world, the output impedance is not purely resistive.

You have it- if there's no difference in 8 and 4 traces, Zo = 0. If there is a difference that is frequency independent, the output Z is resistive. If there is a frequency dependent difference, then the output Z is reactive and again, that is the Thevenin equivalent source impedance from which interaction with loads is perfectly predictable.

The most common case is a Zo that is slightly inductive since the loop gain rolls off causing Zo to rise at high frequency.

 

[tmtomh]

Well...I'm not exactly skeptical. Obviously the class D amps are working well for the vast majority. If anything I'd like to see more actual information on exotic speaker loads. I'm perfectly willing to believe that they may not be a problem.

Fair enough!

 

[fpitas]

Fair enough!

Once my J2 dies (if that ever happens. No First Watt amp has died yet) I'll replace it with class D. So I'm hardly turning up my nose.

 

[tmtomh]

This is not the conclusion drawn. You are stating that by testing an amp with two resistive loads, which show load variance, the amp is therefore fully characterized and let’s call it a day. This is a fundamental misunderstanding of how reactive elements affect an active circuit, in a different manner than resistive loads. For example, using @amp ‘s extreme load, by your logic I could replace the magnitude of its impedance at frequency X with an equivalent resistance of the same value, and get the same result. That’s not accurate. Reactive loads impart phase shift, and phase shift in any system with feedback alters behavior. It’s instructive to understand this load sensitivity, no different than with any other amplifier design.

Your comment represents yet another example of the issue that's driving this thread in circles. The question is not whether we all understand "how reactive elements affect an active circuit." Of course we get that. But that technical-sounding and ominous-sounding language simply boils down to something quite simple: an amp whose response is load-dependent will produce nonlinear measurements across the audible spectrum; when tested with a simple resistive load, the nonlinear measurements will plot a relatively smooth-looking graph, while with a complex load, the nonlinear measurements will plot a "wiggly" graph as seen in some examples above. But both graphs will clearly show load-dependent nonlinearities, and that's what we want and need to know.

What neither you nor @Holdt nor @damonhill nor anyone else arguing this position has shown, or provided any plausible theory or hypothesis for, is the claim that there are amps out there that show a total load invariance when tested with resistive loads, but suddenly reveal load variance when tested with complex loads. (And again, Amir has tested this - he just stopped after he failed to get any qualitatively different results with each different kind of load).

Now, all that said, you are of course correct that there are situations or applications - like when designing an amp, learning about circuits, or just engaging in exploration for curiosity's sake - when it can be interesting to look at the specific ways that a load-dependent amp's response might change when connected to various speakers (or complex loads meant to approximate certain speakers). But - and here's where the circular nuttiness of this thread comes in - that's not the purpose of an amplifier review. Load-dependence is a problem for an amp - once that problem has been discovered (or, in the case of load-invariant amps, ruled out), it doesn't matter for the purposes of the amp review exactly what the nonlinearities look like with this or that speaker. Once again, if you want a nonlinear response with your particular speakers, then use EQ or DSP, not a load-dependent amp. And if your response to that is, "Well of course I don't want my amp to have nonlinear response with my speakers!" - well, then, that's my point: in that case you don't want a nonlinear amp, and so you are best advised to avoid load-dependent amps. We don't need to know all the various different ways that load dependency can manifest itself in order to know that we want to avoid using that amp in our system.

 

[ahofer]

Load-dependence is a problem for an amp - once that problem had been discovered (or, in the case of load-invariant amps, ruled out), it doesn't matter for the purposes of the amp review exactly what the nonlinearities look like with this or that speaker.

For those of us who can't follow along as easily, and may be going down the drain in the 'circularity', is the NC 252 is load dependent by your definition?

 

[Deleted member 48726]

Still I seem to not have read an answer to how I should conclude how badly this amp is load dependent from looking at the resistive frequency response?
No one have answered this?

What about this one?

Will this behave completely flat with real speakers? No / yes / maybe?

Can I conclude that every amplifier with a slight rise in high frequencies are behaving as badly as the first example?
IMO it's not very telling what might or might not happen when they are connected to a real load from the resistive tests. I mean, if a rise of 0.5 dB at 20 kHz results in +2 dB at 50 and 1500 Hz and a drop of 2 at 5000 Hz then the test traces doesn't seem to transparent or useful. -What if it's flat with resistive load to 50 kHz and then rising..

The Crown 4-300 above has a +0.5 dB rise at 10 kHz. "Very modest load dependency" in the review. But it is the same kind of rise the amplifier from the first example exhibits (at 20 kHz) but yet the complex load response is all over the place.

 

[NTK]

Still I seem to not have read an answer to how I should conclude how badly this amp is load dependent from looking at the resistive frequency response?
No one have answered this?

@SIY has answered that question in post #486, quoted below.

You have it- if there's no difference in 8 and 4 traces, Zo = 0. If there is a difference that is frequency independent, the output Z is resistive. If there is a frequency dependent difference, then the output Z is reactive and again, that is the Thevenin equivalent source impedance from which interaction with loads is perfectly predictable.

The most common case is a Zo that is slightly inductive since the loop gain rolls off causing Zo to rise at high frequency.

For those of us who can't follow along as easily, and may be going down the drain in the 'circularity', is the NC 252 is load dependent by your definition?

I couldn't find a graph of the NC252MP frequency response measurements with both 4 & 8 ohm loads, so I'll use the NC502MP graph. You can see that the 4 & 8 ohm measurements almost lie on top of each other (in this case, within about 0.1 dB). That indicates there is very little load dependency.

Using @Holdt's first example from Stereophile's measurement. There are significant separations (by several dB's in the vertical positions) between the FR curves, and that indicates load dependence -- the amp gives a lower output when load impedance is lower. The separations between the curves, however, do not change much with frequency (frequency independent using SIY's terminology), and that indicates that the output impedance is largely resistive.

The curve from the simulated load is wiggy because the impedance of the simulated load varies with frequency. When the impedance of the simulated load is high at a certain frequency, the measured response will be higher (use the voltage divider equation to understand it mathematically); conversely when simulated load impedance is low. Since the impedance of the Stereophile simulated load goes up and down and up and down across the frequency range, the frequency response went up and down with it.

The Crown and Fosi graphs show frequency dependent variations, and that means the output impedance is reactive. That is due to the class-D output filters, and either not implementing post filter feedback, or a not completely effective implementation.

 

[Matias]

I couldn't find a graph of the NC252MP frequency response measurements with both 4 & 8 ohm loads,

Here

 

[jimk1963]

I'll skip copying any one commenter's replies here, as all of you are arguing about Thevenin and passband warble. The issue found by the OP has nothing to do with simple voltage divider math at the output of the amp - nothing at all. And I doubt any EE's on here appreciate all the condescending language about basic circuit theory. But I digress. The "wiggly line" arguments you guys are having, while also tossing in Thevenin, are based on linear systems theory. When a circuit is modeled as a cascade of linear elements, of course a passband's variation can be easily predicted.

While interesting (especially extracting a PA's output impedance - a task in itself), it is missing the larger point of the OP's findings, and also a critical reason for load-pull testing in general. What we look for with load-pull tests isn't just passband warble, but rather nonlinearities that otherwise don't manifest themselves with purely resistive loads. Nonlinearities that, for the hundredth time, may manifest themselves differently in a Class D due to its architecture (heavy output feedback, modulation-based design, ...). The OP found some weird noise aliased to baseband when he tortured the amp's output with a highly capacitive load. Thevenin has zero to do with this finding. It happens because somewhere in the amp, likely in the output feedback circuit (my best guess), the reactive load causes the amp's self-oscillation frequency to change/modulate and/or some other non-Thevenin-at-all nonlinear action is happening. This is what I found most interesting in the first place, and still do.

It's especially interesting (to me) that not all Class D's behave the same under stressed loads. Doesn't mean the amp is "bad", especially if the load required to induce passband noise is beyond what the amp will ever see. But it does highlight the differences in how Class D architectures are implemented. And again, admittedly mostly from a designer's viewpoint rather than a consumer purchase decision viewpoint - I find these differences intriguing and worth consideration of ways to better understand this. Perhaps a topic for another thread, to avoid confusing consumers on here trying to make a purchasing decision. I certainly haven't concluded the NC25x or NC50x amps are "bad", in fact I'm exploring purchase of these for my HT surrounds. They're cheaper than the new-kid NCx series and still fantastic amps, as Amir has kindly shown.

 

[SIY]

I'll skip copying any one commenter's replies here, as all of you are arguing about Thevenin and passband warble. The issue found by the OP has nothing to do with simple voltage divider math at the output of the amp - nothing at all. And I doubt any EE's on here appreciate all the condescending language about basic circuit theory. But I digress. The "wiggly line" arguments you guys are having, while also tossing in Thevenin, are based on linear systems theory. When a circuit is modeled as a cascade of linear elements, of course a passband's variation can be easily predicted.

While interesting (especially extracting a PA's output impedance - a task in itself), it is missing the larger point of the OP's findings, and also a critical reason for load-pull testing in general. What we look for with load-pull tests isn't just passband warble, but rather nonlinearities that otherwise don't manifest themselves with purely resistive loads. Nonlinearities that, for the hundredth time, may manifest themselves differently in a Class D due to its architecture (heavy output feedback, modulation-based design, ...). The OP found some weird noise aliased to baseband when he tortured the amp's output with a highly capacitive load. Thevenin has zero to do with this finding. It happens because somewhere in the amp, likely in the output feedback circuit (my best guess), the reactive load causes the amp's self-oscillation frequency to change/modulate and/or some other non-Thevenin-at-all nonlinear action is happening. This is what I found most interesting in the first place, and still do.

It's especially interesting (to me) that not all Class D's behave the same under stressed loads. Doesn't mean the amp is "bad", especially if the load required to induce passband noise is beyond what the amp will ever see. But it does highlight the differences in how Class D architectures are implemented. And again, admittedly mostly from a designer's viewpoint rather than a consumer purchase decision viewpoint - I find these differences intriguing and worth consideration of ways to better understand this. Perhaps a topic for another thread, to avoid confusing consumers on here trying to make a purchasing decision. I certainly haven't concluded the NC25x or NC50x amps are "bad", in fact I'm exploring purchase of these for my HT surrounds. They're cheaper than the new-kid NCx series and still fantastic amps, as Amir has kindly shown.

I'm certain your knowledge of RF technology is excellent. But unfortunately, this does not directly translate to audio. And to see past some of the parlor tricks unfortunately takes some direct experience with audio measurement, especially of the particular technologies being discussed. The load-dependent frequency response variations waved around by some of the less knowledgeable folks are EXACTLY due to Thevenin, but have little to do with the original questions of driving amplifiers outside of their design limits that you're talking about.

 

[tmtomh]

For those of us who can't follow along as easily, and may be going down the drain in the 'circularity', is the NC 252 is load dependent by your definition?

Good question! I don't know for sure, but @Matias posted a measurement graph above that suggests the 252MP is load-invariant: Post #493

In that graph, it appears the 4 ohm and 8 ohm response is identical until you get out to about 55-60kHz (and even there the 4 vs 8 ohm difference appears to be not much more than the L-R channel differences).

 

[jimk1963]

I'm certain your knowledge of RF technology is excellent. But unfortunately, this does not directly translate to audio. And to see past some of the parlor tricks unfortunately takes some direct experience with audio measurement, especially of the particular technologies being discussed. The load-dependent frequency response variations waved around by some of the less knowledgeable folks are EXACTLY due to Thevenin, but have little to do with the original questions of driving amplifiers outside of their design limits that you're talking about.

Stability testing of an amplifier is hardly a parlor trick, if that's what you're implying. And yes, these concepts apply to amps at any frequency, so please, enough with the condescension. It really isn't necessary here. I'm not interested in passband warble arguments, where Mr. Thevenin has a role, at least at lower powers where the amp's output impedance is mostly invariant at a given frequency. I'm interested in unexpected nonlinear behavior of Class D's, whose circuitry is vastly different than Class AB / A amp topologies. My hunch is Class D designers are already well aware of these concerns, as evidenced by their outstanding performance under a variety of loads. But there's always more to learn.

 

[pma]

It is the opposite than @SIY is saying. RF susceptibility, input stage rectification, demodulation very often directly translate to audio. 20Hz - 20kHz view is short-eyed. Unfortunately, this is not a forum with many knowledgeable and experienced circuit designers.

 

[Deleted member 48726]

@SIY has answered that question in post #486, quoted below.



I couldn't find a graph of the NC252MP frequency response measurements with both 4 & 8 ohm loads, so I'll use the NC502MP graph. You can see that the 4 & 8 ohm measurements almost lie on top of each other (in this case, within about 0.1 dB). That indicates there is very little load dependency.

Using @Holdt's first example from Stereophile's measurement. There are significant separations (by several dB's in the vertical positions) between the FR curves, and that indicates load dependence -- the amp gives a lower output when load impedance is lower. The separations between the curves, however, do not change much with frequency (frequency independent using SIY's terminology), and that indicates that the output impedance is largely resistive.

The curve from the simulated load is wiggy because the impedance of the simulated load varies with frequency. When the impedance of the simulated load is high at a certain frequency, the measured response will be higher (use the voltage divider equation to understand it mathematically); conversely when simulated load impedance is low. Since the impedance of the Stereophile simulated load goes up and down and up and down across the frequency range, the frequency response went up and down with it.

The Crown and Fosi graphs show frequency dependent variations, and that means the output impedance is reactive. That is due to the class-D output filters, and either not implementing post filter feedback, or a not completely effective implementation.

An answer I learnt something from. Kudos! -It took a while for someone to explain it proper. At least for me to understand. Now I get much more from the 4 / 8R measurements than before.

Still I think it's not telling much of what can happen with real loads though. I mean with this knowledge I can now tell if
A) The amp is somewhat load dependent or
B) The amp appears not to be load dependent.

And it appears you have to look for very small deviations on the resistive tests because it looks like 0.5 dB resistive variations could result in over 2 dB variations with complex loads. -And what about higher loads?

As I see it the resistive tests merely imply than reveal much.

 

[pma]

Still I think it's not telling much of what can happen with real loads though

It is not for at least 2 reasons:
- says nothing about possible stability issues
- says nothing about SOA limitations

Again, views of real circuit designers are missing here. Opinions swim on the surface.