So we now have a speaker with >5% distortion which isn't audible subjectively. Is a SINAD of 96dB poor?

[Ron Texas]

The point of this site is 42.

 

[amirm]

Is it clear how loud the listening test was done at? We know how loud the distortion measurement was made at, and we know that lower playback levels would give lower distortion.
Some more info from @amirm before we go to far help. But it is a very interesting area to explore.

As shown, it was 90 dB SPL at one meter. I had to wear ear protection for the test.

Normal standard I have now (same as soundstage) is 96 dB at 1 meter. But I feared that may damage these little speakers.

 

[amirm]

My point was that Amir didn't find the distortion audible when he listened loud - yes, not a peer reviewed study ))

I have not said this. I have no way of turning the distortion on and off to determine if it is audible. I *can* do this with two amps or DACs, one of which distortions while both have the same frequency response. This is key: frequency response is by far the dominant factor in evaluating speaker sound. My brain won't focus on distortion as long as there are frequency response errors.

 

[Frank Dernie]

I have not said this. I have no way of turning the distortion on and off to determine if it is audible. I *can* do this with two amps or DACs, one of which distortions while both have the same frequency response. This is key: frequency response is by far the dominant factor in evaluating speaker sound. My brain won't focus on distortion as long as there are frequency response errors.

Sorry, I extrapolated what you wrote which was:-
"I was pleased that even at extreme, there is no sudden distortion."

 

[amirm]

So have I and many others, that is why I thought it was worth having it all in one thread, and particularly since nobody has satisfactorily (to me at least) explained why one rule for speakers and a different one for electronics is OK from a SQ pov, I understand making low distortion even frequency response electronics is relatively trivial and getting anywhere near with speakers impossible but that is a practical rather than SQ matter.

Three things:

1. If I serve you salty ice cream, you are not going to care how good the flavor of vanilla is in it! Frequency response is so dominant as far as "speaker distortion" goes that it is by far the dominant factor in determining likeability of a speaker. Your brain instantly goes for evaluating that as I just post.

2. With electronics such as DACs, getting distortion and noise to incredibly low levels is essentially "free." A $99 DAC gets there. There is no reason to reward electronics that are not well engineered because in most cases, you are not given a discount for that lowered performance, either in purchase price or cost to manufacture.

3. There is a huge gulf between electronics distortion measurements and acoustic. I can modify simple parameters and drastically change the nature of the distortion shown. We are seeing the effects of the room, microphone distortion, fixture rattling, measurement technique (sweep type, window function), etc. None of these factors come into play with electronics measurements.

 

[amirm]

Sorry, I extrapolated what you wrote which was:-
"I was pleased that even at extreme, there is no sudden distortion."

That was meant to say there was no sudden break up of either driver bottoming out, or the amp clipping, or going into compression.

 

[Tom C]

https://www.audiosciencereview.com/...rformance-power-amps.10026/page-3#post-366387
Fred Jacquot’s explanation makes intuitive sense to me. Any error existing in the first component in the chain is going to be increased by errors made by components positioned later in the chain. When a DAC is the first component, its errors will be the most obvious because those errors are amplified and added to by every component following in the chain. Speakers will have the least impact since they are the final component.

http://www.sengpielaudio.com/calculator-thd.htm
When I use this calculator, I get the same answer no matter what order I put different distortion numbers, i.e., if I use different numbers for Device 1 and Device 2, then swap the numbers entered for Device 1 and Device 2, the distortion sum comes out the same. Shouldn’t the order matter? Or is that the difference in using dB instead of %?

When an amplifier reproduces the input signal perfectly, but adds harmonic tones, we call that harmonic distortion. When an amplifier clips, it no longer reproduces the input signal accurately, and the output is much more objectionable. Same is true when frequency response is nonlinear, because the relative levels at output don’t match the input. So, some types of distortion are more objectionable than others. Which would imply that you can’t lump them all together and come up with a single number to use as the threshold of audibility. The tolerance level will differ according to type.

 

[Blumlein 88]

I feel that DACs above 0.1% start becoming irritating. ("TDA1543" from China). Time for a double blind.

Or we can make it even quicker if somebody knows of some distortion-injecting filter in Audacity, then we can play until it becomes detectable. All while on a speaker with a distortion higher than that.

Pkane's Distort software. It was made for this. Free download. Best such thing I've seen for trying distortion profiles.

https://distortaudio.org/

 

[Rick Sykora]

Three things:

1. If I serve you salty ice cream, you are not going to care how good the flavor of vanilla is in it! Frequency response is so dominant as far as "speaker distortion" goes that it is by far the dominant factor in determining likeability of a speaker. Your brain instantly goes for evaluating that as I just post.

2. With electronics such as DACs, getting distortion and noise to incredibly low levels is essentially "free." A $99 DAC gets there. There is no reason to reward electronics that are not well engineered because in most cases, you are not given a discount for that lowered performance, either in purchase price or cost to manufacture.

3. There is a huge gulf between electronics distortion measurements and acoustic. I can modify simple parameters and drastically change the nature of the distortion shown. We are seeing the effects of the room, microphone distortion, fixture rattling, measurement technique (sweep type, window function), etc. None of these factors come into play with electronics measurements.

In 2 of the last 3 passive speaker tests, there was roughly 2% THD at about 2kHz.
It is easy to design a speaker that does not have this degree of distortion.
So, audible or not, we are striving to determine why (and if the speaker, what it might indicate)?

 

[Soniclife]

As shown, it was 90 dB SPL at one meter. I had to wear ear protection for the test.

I was asking how loud the listening test was, not the distortion measurement.

 

[scott wurcer]

https://www.audiosciencereview.com/...rformance-power-amps.10026/page-3#post-366387
Fred Jacquot’s explanation makes intuitive sense to me. Any error existing in the first component in the chain is going to be increased by errors made by components positioned later in the chain. When a DAC is the first component, its errors will be the most obvious because those errors are amplified and added to by every component following in the chain. Speakers will have the least impact since they are the final component.

This makes little sense on several levels. First of all most of the gain in an audio signal path is power gain not voltage gain, in fact every time there is a passive level control it is an amplitude reduction followed by a fixed gain. Secondly, SINAD is a proportional measure of desired signal to impairment, barring pathologically bad electronics in between DAC and speakers what is presented to the speakers maintains that. That's why the -60dB speaker at the end masks the contribution of the rest.

 

[tuga]

Often in electronics when the levels of harmonic distortion are high, so are the levels of intermodulation distortion.
Is that also the case with loudspeakers?

Distortion values versus frequency at constant voltage for the three comparison models. In addition to the lowest distortion values, the Neumann dome (Sennheiser) also delivers a significantly higher sound pressure

Intermodulation distortion over frequency at constant level for the three comparison models. The Neumann dome (Sennheiser) is on average 10 dB lower in distortion than the other two already very good models

source: https://www.fidelity-online.de/neumann-kh-420-messungen/

 

[tuga]

KEF Reference 1 maximum level - Fig. 10 Maximum level with a maximum of 3% distortion (red) and a maximum of 10% distortion (blue). The highest power was 400 W at 8 Ω. The measurement is carried out with 185 ms long sine bursts. The 10% measurement is only carried out for the low frequency range. The green curves show the sensitivity for 1W / 1m and the curve calculated for 400 W (+26 dB).

KEF Reference 1 intermodulation distortion 2m - Fig. 11/12 Intermodulation distortion at 85 dBA averaging level at 2 m and 4 m distance under free field conditions. Excitation signal: multisine with the spectral distribution of a medium music signal and 12 dB crest factor.

KEF Reference 1 intermodulation distortion 4m - Fig. 11/12 Intermodulation distortion at 85 dBA averaging level at 2 m and 4 m distance under free field conditions. Excitation signal: multisine with the spectral distribution of a medium music signal and 12 dB crest factor.

Another measurement regarding intermodulation distortion was carried out with an averaging level of 85 dBA at typical listening distances of 2 m and 4 m under free field conditions. The peak level in this measurement was 101 dB, also at a distance of 2 m and 4 m. A multisine with 60 excitation frequencies and a weighting according to EIA-426B for a medium music signal is used as the test signal for this measurement. The signal has a crest factor of 12 dB and is therefore close to a music signal. For the evaluation, all spectral components added by the loudspeaker are evaluated, which contain both harmonic distortions and intermodulation distortions. The two graphics from Fig. 11 show the spectrum of the excitation signal (green curve), the measured spectrum of the signal emitted by the loudspeaker (red) and the distortion components extracted from it (blue). Both are displayed with individual spectral lines and summed up in 1/6 octave bandwidth. The overall distortion component (harmonics and intermodulation THD + IMD) is for the listening level of 85 dBA Leq at 4 m distance at a low –31.3 dB corresponding to 2.7% and for 85 dBA at 2 m distance it is only 1%.
At first glance, the 85 dBA averaging level sounds like little, but this quickly becomes clear on closer inspection. 85 dBA corresponds to 89 dBZ (unweighted), which is measured in 4 m in the free field. In relation to 1 m, this results in 101 dBZ averaging levels. The peak values are again approx. 12 dB higher, where you have already reached a remarkable 113 dBpk. Assuming a sensitivity of 82 dB 1W / 1m, this requires a peak value in the power of 1.26 kW. Converted to a sinusoidal signal, this is 630 W. A powerful amplifier cannot do any harm. It should also cope well with the impedance minimum of 3.2 Ω.

source: https://www.fidelity-online.de/kef-reference-1-messungen/

 

[tmtomh]

Ive been complaining about this double standard for a while now. I gave up trying to prove my point because it’s lost in the sea of noise that is constant bashing on poorly measuring dacs.

I find the whole obsession about dacs bizarre...

Don't forget, though, that @amirm has pointed out that distortion and noise are cumulative. According to him, it's not a matter of a 1% THD speaker masking the distortion in a 0.01% THD DAC. It's that the DAC creates its distortion and the speaker distorts that distorted signal further.

I don't have any special expertise on this question; just pointing out what Amir has said.

 

[tuga]

Left - Piano, Right - Oboe, Bottom - Voice, Source - Here

A small % of THD from reproducing lower frequency components is not going to change the shape by much. A change in frequency response will.

In many orchestral pieces of the late Romantic period you have some 120 (one hundred and twenty) instruments, sometimes playing simultaneoulsy different parts of the music, sometimes accompanied by a large choir.

Add very wide dynamic and frequency range, and massive assortment of timbres, a mix of legato and stacatto and ambience cues and you get an explosive mix that is very demanding for the playback system, particularly the speakers.

 

[scott wurcer]

Don't forget, though, that @amirm has pointed out that distortion and noise are cumulative. According to him, it's not a matter of a 1% THD speaker masking the distortion in a 0.01% THD DAC. It's that the DAC creates its distortion and the speaker distorts that distorted signal further.

You will find a 0.01% electronic signal chain (this is -80dB which should now be commonly as bad as you get) followed by a 2nd dominated speaker at -40dB will have -40dB +- a few thousandths of a dB. You can run a math simulation trivially.

Noise is also not linearly additive see; https://en.wikipedia.org/wiki/Friis_formulas_for_noise

 

[Blumlein 88]

Long and short is the speaker will always be the determining distortion level of your system unless something else in the chain is broken. The electronics are going to set the noise level.

Which is one big reason I don't like SINAD for electronics. Give me distortion, and it will almost always be inaudible levels. And give me noise levels separate. They too should be inaudible and if not knowing only SINAD confuses the issue. You can have poorly gain-staged systems that will have audible noise without any distortion audible beyond what the speaker provides.

 

[andreasmaaan]

I can not find corroboration that .1% is audible with music at any frequency.

Gaskell reviewed the literature on distortion audibility in 2011 and produced this helpful table:

I also can not find much that supports that 3rd order distortion is easy to spot. (@.1% levels) Most of the opions I find suggest 2nd, 3rd are particulaly benign.

2nd is extremely benign, sure. Third is more benign than 4th or 5th, but far less benign than 2nd.

Having said that, the third will tend to be least benign when it falls within our most sensitive range of hearing (about 2-5kHz). This is obviously the case for fundamental frequencies between about 600Hz and 2000Hz.

Below is another graphical representation of the research from Zwicker & Fastl, this time summarising results from studies of audibility thresholds of masking of a secondary tone by a fundamental tone. Note that these studies took place in quiet (which is actually quite important FWIW, because we don't normally listen to audio in our own homes under such conditions), and using only single tones--this is therefore a highly contrived situation.

Imagine a fundamental tone is produced at 1kHz/60dB. The third, fourth, and possibly fifth harmonics will be audible in the section of the graph I've coloured in red:

Is there any robust speaker distortion testing that has been done that is published and available? Nearly all of what I find is very small studies or dialog on the chat forums. I find some theoretical articles and some very strong guesses but not any real robust stuff (such as Harman's frequency response testing)

It doesn't make sense IMHO to study speaker distortion in isolation, as it is not possible to control all relevant variables when switching speakers.

There is no reason why 1% 2nd harmonic produced by a speaker sounds any different from 1% 2nd harmonic produced by an amplifier (so long as the relative phases are identical in each case).

 

[wwenze]

Often in electronics when the levels of harmonic distortion are high, so are the levels of intermodulation distortion.
Is that also the case with loudspeakers?

Definitely.

I just thought of an idea.

Throw in the 32-tone IMD that has been done for DAC/AMP/etc.

It shows THD, IMD, even frequency response in one shot.

 

[SIY]

Definitely.

I just thought of an idea.

Throw in the 32-tone IMD that has been done for DAC/AMP/etc.

It shows THD, IMD, even frequency response in one shot.

I've done that in a few of my published reviews.