Master Thread: Are measurements Everything or Nothing?

[John_Siau]

OK, but let's examine it properly, without masking by measuring system noise and without measurement misinterpretation. Below is the 335mW/4ohm. Can you see any hints of crossover distortion?? Audible crossover distortion?? All distortion components above H3 are below 0.0001%, thus below -120dB. Funny to speak about audibility.

View attachment 384016

Very nice. The worst-case H3 is -98 dB relative to 335 mW and this would be reproduced at an SPL of about -10 dB SPL with 90 dB speakers (completely inaudible, even if H3 was played by itself). The FFT provides the necessary information that was difficult to extract from the THD sweeps you provided earlier. It almost looks like the THD sweep was really a THD+N sweep, or it was limited by the THD+N of the measurement equipment. Nice amplifier.

 

[pma]

I think John would argue that you’re still 33x too high in power Though I doubt it will make a massive difference, it won’t suddenly have more distortion, mostly worse SNR.

It all goes down below amplifier noise, then, when you reduce power. No mystical xover distortion components. With this amplifier, transition from class A to AB is at about 0.32W, so 335mW was a good choice. However, you will see higher order harmonics rising from noise floor no sooner than above 1W. Still, very low in level. And, remember that uncontrolled crossover distortion usually looks like a "grass", "comb" of high order harmonics above the 7th, often higher than the low order harmonics.

Any simplification and any generalizing is usually wrong. What John said about "class AB" distortion might have been correct in case of the amplifier he tested, however, he brought no proof, no plots. What I am saying applies again to one specific amplifier and is design related. So please let us stop saying that class AB amplifiers will have audible crossover distortion at low power, it is in general nonsense and may be specific for certain design.

 

[John_Siau]

What I am saying applies again to one specific amplifier and is design related. So please let us stop saying that class AB amplifiers will have audible crossover distortion at low power, it is in general nonsense and may be specific for certain design.

This interesting discussion was started by someone asking if anyone had proof that amplifiers could sound different.

The ABX test I presented shows that two amplifiers can sound different, even when the specifications for both look quite good. In this specific case, class AB amplifiers were compared to an AHB2 which is a class AB amplifier with a parallel, passively-summed, class A feed-forward correction amplifier.

Both samples of the specific class AB amplifier under test showed output biasing problems (or crossover distortion problems in prior stages). No matter the exact cause of this crossover distortion, these traditional class AB amplifiers produced audible distortion at 0.01W playback levels, while fully meeting or exceeding the manufacturer's published specifications.

This implies that we need to put more emphasis on the 0.01W performance as well as on the performance measurements at the class A AB transition region. This is more of an issue than we would guess when looking at THD or THD+N numbers that are expressed in %.

It is important to think of THD in terms of dBW so that we can directly translate this to the playback SPL of the THD. It is unfortunate that we traditionally use % units for THD.

 

[ahofer]

The "lenient" threshold for Amp/DAC transparency here is -90db, so it seems like this AB amp wouldn't pass, right?

 

[pma]

One single number says nothing, regardless it is expressed in % or dB (any technically competent person is able to transfer % to dB and vice versa).

I cannot imagine a competent class AB amplifier with high crossover distortion products at 10mW power. So I would respectfully request for @John_Siau to post a proof in a form of a distortion spectrum graph of the ill behaving amplifier. It is the only way to lead a scientific discussion based on facts and not on the stories based on one’s authority, especially if there is a business interest.

 

[Newman]

Here is the evidence:

At Benchmark we ran an ABX test using a 1 kHz low-level tone. We compared a class AB amplifier to the Benchmark AHB2. The test level was 0.01 watt, producing an SPL of 67 dB at the listening position. The measured distortion produced by the class AB amplifier was reproduced at a level of 14 dB SPL. In contrast, the measured distortion produced by the AHB2 was well below 0 dBSPL (actual level was calculated at -21 dBSPL). Input voltages to the speaker terminals were set to 0.2828 Vrms using a precision volt meter.

Given these well-controlled conditions it was very easy to hear the difference between the two amplifiers in a fully-blind random ABX test. I scored 100% correct on every ABX test sequence I ran. Other listeners achieved similar results.

You can read the entire writeup here: "Power Amplifiers - A "First Watt" ABX Test"

Given the results of this test, there is no question that power amplifiers can sound different even when they have respectable specifications at high output levels. This specific test shows that we should pay more attention to the 0.01 watt THD.

Thanks for this John.

IMHO the article needs to include measured frequency responses of the specific sample amplifiers used for this test, when connected to the speaker used for the test, measured at the speaker terminals. This helps to eliminate the risk that there is an audible difference in the frequency responses when connected to the specific speakers and cables as a system. (Of course if your test, as apparent from your article, was only of a 1 kHz test tone, then the FR measurements I described are not needed. But to only test with a single test tone is not good enough - see below.)

The article also needs to include 'baseline' ABX tests with no signal present, demonstrating that the amplifiers cannot be detected from their idle noise output alone. Because you haven't eliminated the possibility that they can be told apart based on idle noise.

Finally and most importantly, I don't see any evidence that you passed an ABX test when playing music at 0.01 watts approximating 67 dB SPL. Audio research is littered with examples of artefacts that are audible with specific test tones, but inaudible when playing music. You have not conducted the 'test of truth' when the amp is being used for its intended purpose. This is especially disappointing because your article repeatedly describes in theory the effect when playing music, but neglects to 'put it to the test'. "Musical details", "between the transient peaks", "change the harmonic character of musical voices", "add a fatiguing harshness". Many ambit claims, none of them tested.

Even your tone test, to be consistent with your article's references to music-playing between peaks, should have the 0.01 watt test tone interspersed by kick drum samples at 25-65 watts (per your article) and say, 110 bpm. Let's see if the ABX is "easy to pass 25/25" now. In contradiction to your ambit claims, I suspect that the ear 'turns the gain down' in these conditions and is less discriminating of the 'musical detail' level between the peaks.

Even then, it still is not good enough, if you want to claim that the effects are audible (not only audible, but preferable...try that test), then you need to demonstrate it with typical music recordings....and backed up with measured frequency responses as per my first paragraph.

You need to re-test and re-write, if you want to do more than mislead (by drawing musical conclusions from test signal results).

cheers

 

[pma]

This implies that we need to put more emphasis on the 0.01W performance as well as on the performance measurements at the class A AB transition region. This is more of an issue than we would guess when looking at THD or THD+N numbers that are expressed in %.

It is important to think of THD in terms of dBW so that we can directly translate this to the playback SPL of the THD. It is unfortunate that we traditionally use % units for THD.

10mW power, dBW scale, class AB and ancient class D amplifier. I do not see any valuable hints to explain audible differences, if they would exist.

------------

Adding Purifi EVAL2 high gain setting to the list. As one can see the difference between the 3 amps at 10mW is only in noise level. In my opinion, the 10mW test is pointless unless the amplifier is totally incompetent design.

 

[Nuni]

It's laughable that you would think that's what I'm saying.


So you are saying you're clueless about science?

As you have no idea whatsoever about my academic qualifications I would say I laugh at your comment and treat it with the scorn it deserves. (You seriously have no clue.)

 

[pma]

Here is another challenge, prove that everything that can be heard is measured. do not just say it can.

I would rise the challenge if that "everything that can be heard" would be supported by properly made double blind test protocol. If not and if based on a mere subjective claim, then there is no reason to rise the challenge.

 

[voodooless]

While you're at it, please prove that there is no teapot floating between Earth and Mars!

 

[Jimbob54]

As you have no idea whatsoever about my academic qualifications I would say I laugh at your comment and treat it with the scorn it deserves. (You seriously have no clue.)

Probably time to park the dare/ double dare dialogue from the last few days please people. The far more interesting discussion in this thread currently is regarding actual measurements of amps vs blind test results- suggest we all see where that leads. Thanks

 

[ahofer]

As you have no idea whatsoever about my academic qualifications I would say I laugh at your comment and treat it with the scorn it deserves. (You seriously have no clue.)

Nobody cares about your credentials or lack thereof. You certainly aren’t doing them any credit with that sort of puffery, or by attempting to shift the burden of proof.

 

[krabapple]

As you have no idea whatsoever about my academic qualifications I would say I laugh at your comment and treat it with the scorn it deserves. (You seriously have no clue.)

I don't know what kind of hack you are, it's true, but also irrelevant and I don't care.

My question "So you are saying you're clueless about science?" was prompted by this gem of yours:

Here is another challenge, prove that everything that can be heard is measured. do not just say it can.

You think that's a scientific approach?

 

[krabapple]

I just don't want people coming here and coming to the conclusion that a Marantz AV 10 sounds better than an Anthem AVM 90 and many people here seem to think they essentially sound the same, that's simply not true. Anthem sound is far superior and anyone that has owned both of these units would understand that in the real world vs peaking at numbers and casual conversations with you experts here. I trust the numbers from Amir and his advice but let's not stretch the truth on Marantz and Denon as being right in line with the sound everyone hears out of these units not just me.

Ok, so after a detour to speakers , we're back again to AVRs. Or is it DACs in AVRs? The original subject was DACs. Then on that thread you started garbling about Marantz, Denon, and Anthem.

I'm thinking your serial vagueness is intentional.

 

[krabapple]

Here is the evidence:

At Benchmark we ran an ABX test using a 1 kHz low-level tone. We compared a class AB amplifier to the Benchmark AHB2. The test level was 0.01 watt, producing an SPL of 67 dB at the listening position. The measured distortion produced by the class AB amplifier was reproduced at a level of 14 dB SPL. In contrast, the measured distortion produced by the AHB2 was well below 0 dBSPL (actual level was calculated at -21 dBSPL). Input voltages to the speaker terminals were set to 0.2828 Vrms using a precision volt meter.

Given these well-controlled conditions it was very easy to hear the difference between the two amplifiers in a fully-blind random ABX test. I scored 100% correct on every ABX test sequence I ran. Other listeners achieved similar results.

You can read the entire writeup here: "Power Amplifiers - A "First Watt" ABX Test"

Given the results of this test, there is no question that power amplifiers can sound different even when they have respectable specifications at high output levels. This specific test shows that we should pay more attention to the 0.01 watt THD.

The question was whether we can predict difference from measurements. You found a distinct measurable difference in distortion at very low wattage, and you heard it using a test tone in a blind test. Your result suggests we need to measure more, specifically low wattage THD.

(It does not, it must be emphasized, suggest that 'we can't measure everything we hear' from amps...and I know you have not suggested that, though others have)

Maybe we need to measure more... if that difference proves relevant to normal listening.
Did you compare the two amps blind using music, spoken word or something else more typical of the consumer experience? Did hearing the test-tone difference prime you to perform at p<0.05 with more typical content?

 

[SIY]

Your result suggests we need to measure more, specifically low wattage THD.

I was reminded of the cheat to show that 24 bit can be distinguished from 16 bit encoding by the expedient of recording a tiny signal, then turning the volume way up to hear the noise floor.

This, of course, is worse since it uses a broken unit (in this case, clearly misadjusted bias).

 

[pma]

I measured just one amplifier with terrible crossover distortion, which had specific low power issues. It was based on OPA549

https://www.ti.com/lit/ds/symlink/opa549.pdf

and the results were like this:

I believe it would be a candidate for a positive ABX test result.

 

[John_Siau]

One single number says nothing, regardless it is expressed in % or dB (any technically competent person is able to transfer % to dB and vice versa).

I cannot imagine a competent class AB amplifier with high crossover distortion products at 10mW power. So I would respectfully request for @John_Siau to post a proof in a form of a distortion spectrum graph of the ill behaving amplifier. It is the only way to lead a scientific discussion based on facts and not on the stories based on one’s authority, especially if there is a business interest.

Here are the plots for the offending amplifier (shown in green on all graphs). The second amplifier used in the ABX test was the AHB2 (shown in red on all graphs). These tests plots were made in 2016. The offending amplifier was an RA-500 "studio reference" amplifier. We had two samples that were both purchased new at the time of the test. Both behaved identically. This amplifier was passively cooled and was designed to be rack mounted. I suspect the amplifiers may have been under biased because of thermal issues related to the passive cooling and the need to run in a studio equipment rack. It is also possible that there was some problem at the factory and these units got shipped with incorrect bias settings. I searched our facility and we no longer have these two samples. In any even, the RA-500 is no longer available.

The first plot is an FFT of the 1 kHz, 1Watt output spectrum of both amplifiers immediately after a cold startup.

The vertical axis (dBrA) is dB relative to 1 W.

From this FFT, you can see that the offending amplifier was working properly when stone cold. It did display a fair amount of power supply noise, but probably low enough to be inaudible with the 87 dB speakers used in the test. The 120 Hz and 180 Hz power supply noise may have been slightly above the threshold of hearing, but I specifically do not recall hearing AC line noise.

Also please note that H3 and H5 are too low to reach 0 dB SPL with 87 dB speakers.

Additional notes:

The 1kHz fundamental was removed with the AP2722 notch filter to improve the resolution of the FFT. This was a technique that we had to use with the older Audio Precision test stations. As many of you may know, the APx555 uses this technique internally, but then splices the 1 kHz fundamental back in.

I am trying to locate the FFT that shows the crossover distortion after the amplifier was warm but not overheated. These measurements were made in 2016 and we have since upgraded to APx555b test stations, so these plots may be lost. I will post them if we locate them.

I do recall that the perfect scores on the ABX tests were achieved by listening for harmonics of the 1 kHz tone.

The THD+N vs. Power curves always displayed the effects of crossover distortion. The minimum THD+N of the offending amplifier (red curve) is -74 dBW (at 0.01W), and this is what was measured and documented when the ABX test was run.

Please note that this "malfunctioning" amplifier still met the manufacturer's specifications for THD+N.

Here is the THD vs. Output Power sweep. The THD+N level is -72 dBW which is just 2 dB lower than the THD+N test above. This confirms that the THD+N reading was mostly distortion. Note that the -72 dBW THD would have produced about 15 dB SPL at the listening position. This agrees with our written notes regarding the ABX test. The harmonics caused by the biasing problem were high enough to be audible.

So what we have here is a very specific case of one brand and model of class AB amplifier that easily drifted into an under biased condition. Once it warmed up and entered this under biased condition, THD could be heard on a 1 kHz tone at very low power levels (0.01 Watt) and a listening level of 67 dB SPL (amplitude of 1 kHz tone). The sum total of the harmonics would have reached about 15 dB SPL at the listening location.

Conclusions:

Harmonic distortion at 15 dB SPL was heard when playing a 67 dB SPL 1 kHz tone. This was confirmed with perfect ABX scores from multiple listeners. The high scores indicate that the distortion was easy to hear when using a pure 1 kHz test tone.

The distortion was caused by crossover distortion in the offending amplifier (see green curves). We had two RA-500 amplifier samples and both had the same problem on they each had it on both channels. They were purchased at the same time, so it is hard to say if it was a production problem or a design problem.

The audibility of the distortion was only confirmed on a 1 kHz test tone. This may imply that it could be heard on some music samples, but the answer to this question is well outside of the scope of this simple test. A much more involved test would be required to determine audibility when playing music.

This crossover distortion defect was present in both channels of two band new RA-500 power amplifiers. This problem could have been solved with thermally stable bias circuits, or by proper factory adjustments, or a combination of both. Clearly these two test samples were not thermally stable. For those of us who design power amplifiers, it seems incredible that these amplifiers shipped with this biasing problem. It is hard to know if this is unusual, especially in lower cost class AB power amplifiers. The RA-500 was a low-cost "studio reference" class AB power amplifier.

This is why we need people like Amir and John Atkinson, who have provided us with a large database of tests.

Thanks to everyone who has made comments regarding this ABX test. I appreciate the need to keep the conclusions specific to the test that was conducted.

 

[JP]

Harmonic distortion at 15 dB SPL was heard when playing a 67 dB SPL 1 kHz tone. This was confirmed with perfect ABX scores from multiple listeners. The high scores indicate that the distortion was easy to hear when using a pure 1 kHz test tone

The audio was a pure tone for all the listening tests? Was this disclosed in the marketing blog? Any attempts made with music?

 

[pma]

So this is OPA549 at 10mW/4ohm, but it is an example of the ill-designed chip amplifier unsuitable for audio. The 10mW specific test is not needed, crossover distortion is seen at any signal level. It is a generator of crossover distortion that is visible even with the oscilloscope.

(Edit: THD crossover distortion hump shown in higher X-axis resolution, at 1kHz test frequency)

In case that any commercial audio amplifier shows such crossover distortion, the designer should be punished .