Research Project: Infinity IL10 Speaker Review & Measurements

[amirm]

In any case distortion at such low frequencies are hardly a factor as our ears tolerate much much more there than higher in the spectrum.

The distortion is not at the indicated frequency. You have to multiply it by harmonic order to get the audible frequency. Once there and with high amplitude, these harmonic sprays easily reach into more audible bands.

 

[QMuse]

What is very rare for general public, can be very common for me.

I believe Toole was applying the term "very rare" in the context of speakers, not listeners.

So if you accept that, I suggest cutting back on the protests regarding distortion being a factor.

I am not "protesting" anything, I am simply not agreeing with what you're saying and providing arguments in a good hope that such discussion is desirable on a scientifically oriented forum.

 

[QMuse]

From what I can see distortion north of 100Hz of those 2 speakers is very similar and differences at LF distortion cannot come close to explain your listening impression of IL-10.

 

[amirm]

I am not "protesting" anything, I am simply not agreeing with what you're saying and providing arguments in a good hope that such discussion is desirable on a scientifically oriented forum.

Arguments are a dime a dozen. Repeating them over and over is even worse. You have done nothing constructive to advance the discussion. Go and take that Klippel distortion test and post the results. Conduct a blind test of speakers that have scores here. Do something other than demonstrating your unhappiness with me in countless posts.

 

[bobbooo]

You are saying woofer breakup doesn't generate distortion???

I think the point is that possible woofer breakup also manifests itself in the frequency response as @TimVG showed, and that in turn negatively affects the computed score. So even if what you hear turns out to be distortion, Olive's preference rating model takes that into account (albeit indirectly), leaving its efficacy intact.

 

[QMuse]

Arguments are a dime a dozen. Repeating them over and over is even worse. You have done nothing constructive to advance the discussion. Go and take that Klippel distortion test and post the results. Conduct a blind test of speakers that have scores here. Do something other than demonstrating your unhappiness with me in countless posts.

I am not "unhappy" with you, I don't even know you. I'm simply not accepting your arguments about distortion being problematic with IL-10 speaker and I have IMO provided sufficient number of counter arguments to support my point.
Going ad hominem means you ran out of arguments and replying in the same manner is certainly not something that I feel the need to do, so I'll leave it here as I anyhow said what I felt needed to be said.

 

[mjwin]

I'll add a thought experiment to this discussion that is hopefully productive:

I like thought experiments

But back to the specifics of loudspeaker anomalies, and your questions:

1. Are these nonlinear distortions provably impossible or negligible in the physical world?
2. How can we measure all the physically possible distortions that are audible, e.g. is THD + IMD enough, or do we need more than that (as is clearly the case with lossy compression algorithms, for example)?
3. Can these nonlinear distortions (whatever they are) explain the subjective audible differences Amir found here?

In good faith, these are my own thoughts regarding the above:

1. I'm not sure we want to go down that slippery route as it's not possible to prove the null hypothesis! We need to keep our focus on finding correlation between experimental data (listening tests) and measurable results.

2. Any sampled dataset has a finite information content. The examples which you gave of traditional THD/IMD test tones can be described very simply, and hence lossy compression algorithms can easily pack them into the available number of bits. Even a multitone test signal could probably pack down to say 320kbits/sec without any degradation. Compression algorithms primarily remove information deemed inaudible, and the better ones such as AAC, don't add (many) artifacts.

However, this is a slight digression since, in the case of speaker testing, I think we can agree that we're looking at trying to identify and analyse additive "distortion", the fact that what comes out of the speaker contains more than went in. I am quite keen on multitone tests as they're much closer to the types of signal which systems will encounter when playing music. The problem with these tests has always been that the result is qualitative. You look at the output FFT and visually judge the amount of hash between the tones.

So can we quantify the output of a multitone test in a way which might conceivably correlate to what we hear?

If we generate the original tones such that they each correspond to an FFT bin used in the analysis, we should be able to remove them completely. Then we're left with the residual hash. It's wideband, and occupies the whole bandwidth of the test and the first step would seem to be to filter it with a known audibility function. At its simplest this could be derived from the Fletcher-Munson curves. The next, and most complex process, would be to apply masking to the residual, the mask would be defined by a narrow bandwidth around each input tone, and follow known research (which I don't have to hand right now!). Ultimately, we would be left with a bunch of "noise". We then simply find its RMS value and arrrive at a figure. I'll call it a hash figure (since it keeps us away from preconceived notions...) The basis of empirical listening tests would then be to look for correlation between perceived quality and #. Maybe there will be none. Who knows?

3. I think that any explanation is more within the realms of psychoacoustics, and something which we don't need to consider too deeply until we have something tangible to explain! Correlation is what we're trying to show in the first instance. Of coure, that doesn't prove causality, but hey-ho... We already have the excellent work which draws correlation between spinorama data and listener preferences, as perhaps the primary measure of perceived speaker quality. Can # be a secondary measure?

 

[amirm]

I think the point is that possible woofer breakup also manifests itself in the frequency response as @TimVG showed, and that in turn negatively affects the computed score. So even if what you hear turns out to be distortion, Olive's preference rating model takes that into account (albeit indirectly), leaving its efficacy intact.

What would be the change in score if that hump is taken away?

 

[amirm]

I am not "unhappy" with you, I don't even know you.

I banned you once before for being obnoxious and rude especially towards me. You then made this new alias and here we are again. You seem to have allergic reaction to anything you have not read. Or if it relates to me knowing something you don't. You then grab on my pant legs and won't let go.

Note that there have been multiple calls to ban you and I have resisted it all. Don't go taking advantage of it by making fallacious statements like this. We are not dummies here.

I'm simply not accepting your arguments about distortion being problematic with IL-10 speaker and I have IMO provided sufficient number of counter arguments to support my point.

So what? What entitles you to keep repeating them? I am spending so much time answering them instead of testing more things.

Everything you have read, I have read. But I also have taken Harman tests and listened to this speaker and countless more. So sit back and stop being so aggressive. Allow for the possibility that there may be other factors in speaker sound. Don't take a hardline position beyond what even the researchers have taken.

Going ad hominem means you ran out of arguments and replying in the same manner is certainly not something that I feel the need to do, so I'll leave it here as I anyhow said what I felt needed to be said.

We should be so lucky.

 

[BYRTT]

...We are not dummies here...

Have a laugh i could feel like a big one now because when you tell that banned story thanks reminds me i always wondered what become of our good active member Krunok, the disguise thing in make a new alias reminds one of good old pink panthers...

 

[bobbooo]

What would be the change in score if that hump is taken away?

I suppose it depends on how exactly it's 'taken away' from the on-axis response and all the other responses that make up the PIR. What exact frequency range do you count as the 'hump'? Should you reduce it to be flat across that range or is that unrealistic and not representative of any real speaker?

Looking at @MZKM 's very useful 'Tonal Balance' plot for the IL10 that hump does seem to be the highest deviation from the mean (even if it's not really low-Q), as well as within the 2-5 kHz frequency range the human ear is most sensitive to, so it could be a fruitful avenue to investigate.

 

[GelbeMusik]

... whether it is theoretically possible for a speaker driver to have greater IMD than its THD chart would suggest.

A simple example: a bass has hardly any distortion deep down, no distortion further up. There is such a thing. But now, with a longer excursion, the shape of the suspension changes, possibly even resonantly. And thus the amplitude of a medium tone. This means that the excursion by the low tone modulates the amplitude of the higher tone: AM aka IM Without harmonic distortions in the two frequency ranges being relevant in themselves.

Whitebox testing would focus on the resonance of the suspension, and test the deformation by simultaneously reproducing bass. Sometimes the IM is acceptable, sometimes not. The popular nasty multitone story hardly reveals an effect, because the disturbance barely gets above the background noise. But with a thinned out signal the case becomes immediately noticeable. Also for the hearing.

The by now well known bass driver SB Acoustics 17NAC of Revel / Burchardt fame shows this behaviour of course, too, but very well damped. Another one at 6 times the price has chosen a way less damped "lowest loss" suspension, which keeps the HD part extremely low, but the generated IM spreads over a remarkably wide frequency range once the right tones are hit, and thus becomes disturbing.

As already mentioned, without any HD indications.

With very small, almost extremely small loudspeakers like here (IL-10), you can always expect occasional overload at HIFI-compatible volumes. Whether the spectrum of disturbances is limited or reaches far up depends on the construction. In general this is not tested at all, because you would have to take components up to H10 or so with you. IM components can also scatter very widely. I speculate that these, together with the strong resonances of hard diaphragms that are common today, contribute to an individual sound that eventually becomes noticeable. (You cannot remove this with an XO, because the generator is the motor/suspension, not the input!)) And from then on you always have it in your ears... just a speculation.

Really low-resonance loudspeakers are boring - all experts like to say that - but they seem to me to be more pleasant in the long run.

Finally an anecdote (again): a highly respected PA midrange driver, low HD, flat amplitude response, only humble ondulations up high, that famous one with the flat suspension failed, according to my criteria, miserably. The bandwith was already limited digitally to a quite narrow 400Hz to 1,5kHz or so, but the - aurally quite disruptive - IM went up to 10kHz. Even higher than the raw frequency response of the driver! What a beast of a kinky sound signature! But some people still swear on it.

 

[NTK]

I think the point is that possible woofer breakup also manifests itself in the frequency response as @TimVG showed, and that in turn negatively affects the computed score. So even if what you hear turns out to be distortion, Olive's preference rating model takes that into account (albeit indirectly), leaving its efficacy intact.

Frequency response is not the right test nor indicator for nonlinear distortions.

A broken up diaphragm can lead to a very bad case of IM distortions. Once the diaphragm is broken up (i.e. gone into buckling), it destroys its performance for all other frequencies because the diaphragm stiffness is gone. The change in the stiffness can be understood using a structural column as analogy. For a slender vertical column, it is initially stiff against a compressive load. But as you increase the compressive load and once the buckling limit is passed, the column buckles and its stiffness immediately goes almost to nil.

Therefore, if the diaphragm is broken up by a, say, 3 kHz tone, the 100 Hz tone (or tones at any other frequencies) it produces won't be good anymore.

 

[KaiserSoze]

Oh gosh, you just reminded me that one of their blind tests of distortion is online and many of us took it: https://www.audiosciencereview.com/forum/index.php?threads/distortion-listening-test.8152/

Despite some quirks in the test protocol which makes it hard to take, many of us had no trouble hearing distortions specific to speakers. These were my results for example:

As it happens, Tracy Chapman is also used by Harmah for their listening tests. As you see, I was able to get to -27 dB and could have done more if the test had not stopped me.

Indeed a number of people finished it to the end (-45 dB):

This is why I draw the -50 dB line in my distortion measurement graphs:

And for this speaker:

Those peaks are below 32 dB around 1 to 2 kHz.

As I said, the notion that speaker distortion is never audible is just preposterous. In comparison to other speakers it may not stand out. But by itself compared to original content, it definitely changes the sound and derates it.

This test on Klippel befuddles me because the sample that is supposed to be clean sounds so awful that it is amazing that anyone gets further than -20 dB. I've never heard such an awful music sample before. It sounds like it is severely compressed for one thing. This kind of test is extremely sensitive to the sound samples that are used. Distortion is much more apparent with some sounds than with others. Klippel really made a mess here, and it befuddles me. It would have been a very useful thing if he used had samples of very clean synthesizer music, for example, or maybe xylophone music, or maybe some classical guitar. Or some mellow woodwinds. A nice clean recording of the first 35 seconds of George Gershwin's "Rhapsody in Blue" would make an excellent test for audibility of distortion. The only test tone of any sort was an overlay of two tones, and for Klippel to have used this in a distortion audibility test is downright bizarre. Talk about the rigging the deck before the game begins.

 

[TimVG]

I think the point is that possible woofer breakup also manifests itself in the frequency response as @TimVG showed, and that in turn negatively affects the computed score. So even if what you hear turns out to be distortion, Olive's preference rating model takes that into account (albeit indirectly), leaving its efficacy intact.

Actually I wasn't even paying attention to the score. I was suggesting possible reasons why Amir's listening impressions were 'decent' instead of 'great' and that instead of looking at specific (potential) issues, it's important to address the more obvious matters first. Now, this can only be solved by means of deduction and assessment - and for now the only person able to do so is Amir, but understandably given the amount of reviews he's pumping out (not just speakers) this cannot happen overnight. In fact we should be very grateful for the amount of data we're getting for free here. I'm tempted in buying the pair off of eBay to do some testing myself but the shipping and import taxes will set me back more than the actual speakers.

 

[GelbeMusik]

A broken up diaphragm can lead to a very bad case of IM distortions. Once the diaphragm is broken up (i.e. gone into buckling), it destroys its performance for all other frequencies because the diaphragm stiffness is gone. The change in the stiffness can be understood using a structural column as analogy. .

The analogy doesn't fit, as is often the case. The 'break up' doesn't indicate a broken (whatever). All the consequence You mention is actually not present.

 

[tecnogadget]

Luckily this thread started to stabilize in the last pages, I find very annoying the undeserved hits that @amirm had to take from several posters that entered in a toxic loop...

But still not much progress has been done, so is there any proposed test in the 30 pages that we could rescue (especially distortion related) ?
What about Amir re-listening using the EQ filter made by @QMuse in order to rule out Tonal Balance mismatch vs REVEL ??

 

[bobbooo]

Frequency response is not the right test nor indicator for nonlinear distortions.

Sure frequency response does not alone fully characterise the distortion, but a relatively large change in frequency such as that seen at 4-5 kHz for the IL10 can have a common cause with nonlinear distortion (that cause here hypothesised to be woofer breakup). That correlation between frequency response change and nonlinear distortion due to a common cause could be all it takes to modify the computed score to indirectly take that distortion into account in terms of its affect on preference, especially considering research has shown distortion to have a much lower influence than frequency response on preference.

 

[AnalogSteph]

I'll be echoing a post from page 10:

IME- grunginess and lack of clarity heard by @amirm comes from untamed midbass driver resonance at 4.7kHz. I made few loudspeakers with midbasses that had such resonance (metal/ceramic membrane) and one can easily hear it if not tamed by the crossover.

Reading this thread from the start, I was thinking "I betcha this model has a metal cone woofer". They are quite notorious for out-of-band breakup resonance peaks. Those may look suitably innocuous in FR measurements even when not sufficiently taken care of yet people may still very much hate them. The spike in 3rd harmonic at ~1.6 kHz (along with it still being at ~1.5% 3rd at 86 dB, indicating that it may remain present for a good long while below that) is the best indicator here. That would be my bet on why Amir didn't seem to like this speaker. He certainly would have been more likely to pick up on this distortion peak than most of us.

Proving this (conjecture from informed circles) would require amending the crossover to include a ~4.7 kHz notch filter for the woofer, implementing it for one speaker and comparing it to the second, unmodified sample (the performance of which should be verified beforehand). A quick back-of-the-envelope calculation indicates that paralleling L1 with ~560 nF + a bit of R to not wreck ultimate rejection altogether (I'd start out with somewhere around an ohm maybe) seems worth a shot; I should probably get out LTspice instead.
Either way asking a plain ~2.8 kHz 2nd order lowpass to deal with a nasty resonance peak at ~4.7 kHz seems a bit much. This may not yet have been common knowledge when the IL10 was designed 20 years ago, but I do think speaker designers have been aware of this potential snag for quite a while now.

This is one of those cases where interpretation of the measurements is just as critical as the measurements themselves.

Speaking of which, we don't have any multitone tests because that feature wasn't included in the license, right? Bummer. We could really use some IMD testing.

 

[GelbeMusik]

... such as that seen at 4-5 kHz for the IL10 can have a common cause with nonlinear distortion (that cause here hypothesised to be woofer breakup). That correlation ...

Nope. Break-up doesn't even make distorsion. It does nothing, except a change in linear response aka amplitude frequency response. Only if the AFR is exaggerated, by the diaphragm breaking-up in resonance, that amplifies distortion products, which are generated elsewhere (!) in the frequency range.

Many drivers happily take advantage of break-up, to everybodies pleasing, while keeping HD at very low level.

The said effect of HD/IM amplification wasn't recognized, or communicated the least, when the stiff cone tech came up, like hexacone, metal and so on.
The advertizing claimed that in particular soft cones where susceptible to HD/IM due to lacking rigidity. The other way round it was. The common resonance(s) of the stiff cones were excited by distortion products and generated a specific signature, which, sure enough, was taken as supernatural clearity. It is not so, that the HD was experienced as THE second or THE third harmonic. Due to the resonance it would be perceived as an up- and down-swelling side note of fixed frequency. Of course such an effect would be pronounced if at exactly that frequency the resonance appears as a peak in the AFR too. It would render the affected frequency band a focus of attention, which was not set by the music, but by the speaker.