Power amplifier THD+n tests

[fpitas]

Audio devices are for making audio. Of course you can investigate past what's audible.

Tell me why one wouldn't measure with 100MHz bandwidth

What kind of tweeters do you own?

 

[Trell]

Audio devices are for making audio. Of course you can investigate past what's audible.

Tell me why one wouldn't measure with 100MHz bandwidth

Because this is audio for human hearing?

 

[voodooless]

Tell me why one wouldn't measure with 100MHz bandwidth

That’s a massive straw man and you know it. The graphs show distortion of the audible frequencies. For that you need more than audible bandwidth.. 100 MHz does not contain any measurable distortion products of audio from 1 Hz to 20 KHz.

 

[fpitas]

That’s a massive straw man and you know it. The graphs show distortion of the audible frequencies. For that you need more than audible bandwidth.. 100 MHz does not contain any measurable distortion products of audio from 1 Hz to 20 KHz.

To be fair, we don't know how high he can hear.

 

[voodooless]

To be fair, we don't know how high he can hear.

That is irrelevant. Air cannot move at 100 MHz.

 

[fpitas]

That is irrelevant. Air cannot move at 100 MHz.

Sometimes a joke is just a joke

 

[IAtaman]

If it did not make sense to me that the two graphs showing THD+N calculations had different bandwidths, my first reaction would be to ask. This is a forum after all. People ask questions. But not you. You don't need to ask. You already know better. You are here to make a recommendation.

How fascinating.

 

[mike7877]

Because this is audio for human hearing?

Exactly

 

[mike7877]

That is irrelevant. Air cannot move at 100 MHz.

Wrong.
Because these are audio devices, meant to make sound for people to hear...
Which is why 22 makes sense, like the hierarchy measurement.
And for consistency

 

[mike7877]

If it did not make sense to me that the two graphs showing THD+N calculations had different bandwidths, my first reaction would be to ask. This is a forum after all. People ask questions. But not you. You don't need to ask. You already know better. You are here to make a recommendation.

How fascinating.

Argue for why they should be different. Wait... they shouldn't be!

It's not me in the wrong. And if you read what's transpired in this thread you'd agree these folks are trying to push my buttons. Maybe you haven't read the last page in the half. But if you have, I'd have to think you might be.

I am the most laid back person, and nearly never on the offensive. A few people gaslighting me on the internet isn't going to change the truth or that I know what it is.


To be clear, it's not that I don't think 45kHz bandwidth measurements are useless, just if there is only one chart and the choice is 22 or 45, it should be 22 (like all Amir's hierarchy charts - which amp and which dac are best/the order from best to worst, left to right)

Edit: and yes, I came into this thread with a take on the situation. The same take as you. You know how reasonable advocating for both to be 22kHz is. Nobody's said anything to oppose this view except that measurements aren't done here to determine how things might sound. I'd wager most people are looking at specs for a hint (strong or weak) at how something might sound. Having 45kHz showing 13dB more noise at 1kHz/5w 4 ohms vs 22kHz 1kHz 5w 4 ohms, is a weird inconsistency. Something I think should be solved, using 22kHz so the numbers reflect what will be projected into the room or at one's ears.

 

[mike7877]

That’s a massive straw man and you know it. The graphs show distortion of the audible frequencies. For that you need more than audible bandwidth.. 100 MHz does not contain any measurable distortion products of audio from 1 Hz to 20 KHz.

And there is no audible distortion between 22 and 45kHz (my entire point...)

All I did was exaggerate opposing claims (by saying 100Mhz) so it became obvious how ridiculous others' assertions were

 

[restorer-john]

A 22kHz bandwidth is way too narrow. I use 200kHz, as does just about every single audio analyzer ever made. This is because THD is typically calculated to the 10th harmonic. Of course you can narrow a bandwidth, but if you don't have it in the first place, numbers cannot be compared.

But then came along amplifiers with high levels of HF THD and spewing garbage above 20kHz. Times change and now we have reviewers making up their own bandwidths to keep manufacturers happy and manufacturers narrowing the bandwidth to even 20kHz so their silly numbers look better to the uninitiated.
But I'm not testing class D amplifiers.

 

[mike7877]

What kind of tweeters do you own?

ATC SH25-76s.
-6dB @ 25kHz
-2 @ 20kHz
I can hear up to 18.4kHz

The reason I said 100Mhz is in post 31

 

[mike7877]

A 22kHz bandwidth is way too narrow. I use 200kHz, as does just about every single audio analyzer ever made. This is because THD is typically calculated to the 10th harmonic. Of course you can narrow a bandwidth, but if you don't have it in the first place, numbers cannot be compared.

But then came along amplifiers with high levels of HF THD and spewing garbage above 20kHz. Times change and now we have reviewers making up their own bandwidths to keep manufacturers happy and manufacturers narrowing the bandwidth to even 20kHz so their silly numbers look better to the uninitiated.
But I'm not testing class D amplifiers.

On this site (what we're talking about), the DAC and amplifier charts use 22kHz. I say the frequency sweep should as well, unless the hierarchy charts should be switched to 45.

I think if all went to 45, all devices with no multi tone/power (like img 2 in OP) would need to be retested or removed.

Since 22 reflects the audible spectrum and these are audio devices, 22 makes the most sense


Edit: maybe class D amps should be done at 200kHz as a separate test to see if they dump a lot of energy into tweeters (or just waste energy)

 

[restorer-john]

On this site (what we're talking about), the DAC and amplifier charts use 22kHz. I say the frequency sweep should as well, unless the hierarchy charts should be switched to 45.

I think if all went to 45, all devices with no multi tone/power (like img 2 in OP) would need to be retested or removed

You appear confused.

A frequency response plot does not measure or take into account any noise or THD. It measures the product's flatness to a sweep/chirp/stepped sine. It can go out from 0Hz to 500kHz and be very meaningful.

Noise (the +N part) of THD+N must be specified over a bandwidth, as noise is spread across the entire range you measure. But, you have to allow for the harmonics you want to capture. At 22kHz BW, you cannot get a meaningful THD for anything above a few kHz. Some argue 20KHz and 6.67kHz as being the upper limit of test frequencies, but that only allow the 3rd of 6.67kHz.

Noise was traditionally pretty much a fixed number spec and it made sense to specify the noise figure as a standalone number in uV (with or without weighting/bandwidth) and THD was really THD+N. But now we have gear where the noise increases with level, sometimes outstripping the THD sometimes not. So the all encompassing THD+N (SINAD) throws them in the same bucket and ascribes a single number that varies across the spectrum with harmonics falling off as the bandwidth limit is hit. Hence the sometimes very deceptive charts.

 

[dlaloum]

You need 45 kHz bandwidth to measure distortion at 15 KHz and higher. Otherwise you’ll need to limit to a max of 10 kHz or so.

That’s why I rather look at the multitone and brought it up. It’s a much more realistic view of how it handles real music in the audible range.

If there is a big difference between the 22 kHz and 45 kHz bandwidth measurements, it means there is quite a lot of things going on above 20 kHz. That should not be audible indeed, but that’s not really what that graph is about. You’ll probably not hear any distortion above 6~7 kHz at all…

We also do don’t use A-weighting, which would make sense for audibility.

Yes there is a whole discussion about the how THD should be measured with regards to bandwidth, and which measurement is relevant to our hearing and psycho-acoustics...

Strictly speaking THD to 20kHz requires measuring out to at least 40khz to include 2nd harmonic, and 60kHz or more if we want to include 3rd...

So for strict THD definitions we probably should do the measurements out to 192kHz (a convenient and readily available standard) to allow for additional harmonics to be included.

But, if we are filtering anywhere - and that would include speaker tweeter capabilities, crossovers, protection circuits etc... Then our true frequency response is band limited and we possibly should be discussing the "effective" THD - the frequency cut-off of which would be determined by the combination of both amp and speakers. - Plenty of speakers roll off substantially after 20kHz - others have extended response out to 35kHz and more (I chose 35kHz because that is the spec for mine!)

In the case of my Crown amp, with onboard DSP, it's output is by design band limited to 20khz.... (or supposed to be!)

So with my Crown amp, and a more typical high frequency limited tweeter - the true frequency response out of the speakers are likely to be circa 20kHz ... 22kHz seems a fair additional margin - so THD band limited to 22kHz is a decent representation of what we actually get to listen to in the real world, whereas the theoretical wideband THD has much more limited theoretical relevance.

Plenty of amps out there are wideband designs - with bandwidths extending to 50Khz or more... but measuring them out to 50kHz may be an academic engineering exercise, rather than a useful specification in evaluating audible performance.

 

[Rja4000]

the trend of so many of the threads I start, the first replies are not replies, rather statements that are only somewhat relevant or even entirely irrelevant.

Well, I've read your OP twice and, although I (think I) understood what you were saying, I didn't see any question but a statement that both measurements should use the same BW.

Then, how to be surprised that answers contain statements as well ?

If you allow me, without making a personal criticism, if that kind of shift happens often in your threads, maybe you should try to be more clear on your expectations in your posts.

And, to come back to (what I think is) your point:
You may whish that same BW is used for both, because it will bring kind of peace of mind by ending with the same value, but, as explained by others, there are even more reasons to use a different BW, as we see here.

And, anyway, Amir decided to use different BW, so the question is not really open.
It's much more important, from my point of view, to keep coherency in measurements from review to review than to try to please everyone (which won't happen anyway).

 

[Mnyb]

@mike7877

We already had a tread about this amp and already established that it has other problems that are more audible than the THD . It’s load dependency ie it’s frequency response varies with the speaker impedance.

 

[Mnyb]

Here is the old tread .

How do these two charts come from amps with [essentially] the same THD+n measurement (1kHz @ 5W / 4 ohm)

How can this: And this: Both come from amplifiers with: PA3s: -87 & -85 dB THD+n 3700H: -89 & -87 dB THD+n Edit: literally every PA3s line is above -70 and all but maybe 1% of the 3700H's are below -70! I have the Denon X3700h, and for sure the charts above are a much better...

www.audiosciencereview.com

 

[MaxwellsEq]

- the THD+n measurement at the very top is done @ 22kHz, multi tone: 45.

The two images below show a 13dB diference

I needed to read your OP twice to understand what you are saying. You are correct - the single SINAD number is 13dB better than the line in the graph showing the THD+N at 1kHz across the band.

I think you are making the case that a quick glance at the number, and the amp's location in the comparison table which are based on 22kHz may mislead. Whereas other charts show poorer results because they are based on 44kHz. You are arguing that this is confusing and the measurements should stick to one or the other frequency. Is this what you are saying?