Research Project: Infinity IL10 Speaker Review & Measurements

[Thomas_A]

Is it possible to show the simulated in-room response of the IL10, Focal Aria and M106 in the same graph?

 

[tuga]

At least we have Dr. Olive's attention.

View attachment 71133

The Spin looks good yet there is apparently something wrong with the sound (we haven't yet ruled out a problem with the speakers).
Is this a case of missing the forest for the trees?

 

[tuga]

Is it possible to show the simulated in-room response of the IL10, Focal Aria and M106 in the same graph?

Yes. Download REW, import the "Estimated In-Room Response.txt" file of each speaker (available for download as a Zip archive at the end of the review) and select "All SPL".

 

[Thomas_A]

Yes. Download REW, import the "Estimated In-Room Response.txt" file of each speaker and select All SPL.

Thanks!

 

[MZKM]

Thanks!

That won’t be 100% correct. The graph Klippel exports is slightly incorrect for the in-room response (the standard had confusing directions and they took it the wrong way). I have to calculate the graphs correctly for my data. I can make a graph of all 3, give me a few minutes.

 

[MZKM]

Thanks!

Zoomed in to see differences more easily.

Here is comparison their "Tonal Balance", which is a normalized weighted average I do based on their on-axis and PIR:

 

[GelbeMusik]

Thanks for the hint on the "aperture effect". It is something I wanted to convey to some folks discussing (again) time alignment (again) and again, always forever.

I don't know if you've read Toole's book?...

... for loudspeakers of all sizes and prices, and 0.996 for similar bandwidth bookshelf speakers (p < 0.0001). This is not accidental. ...
The big missing ingredient in the analysis of loudspeaker performance is a measure of non-linear distortion that reliably correlates with perception. The common harmonic and intermodulation metrics are almost useless. That is where perceptual (simultaneous and temporal masking) factors are absolutely required in the metrics.

Almost useless for the industry. Toole serves the industry with consultancy to some success, which isn't bad at all. But, "useless" is just not true from the perspective of the customer. One might decide to set a safe limit, as the host here already did with -50dB of THD.

This is a metric. It is engineering where it starts: will the pyramid stay upright, will the bridge hold, will the dam not break, will the car break == decelerate under every circumstance etc. Setting a safety limit for starters, actively decide for a big margin, refinement comes later. Fault is not an acceptable option. Basic engineering. Btw, this crude metric is even well accepted in decision making. It discriminates the doable versus the impossible.

Devoid of any further knowledge one has to go for the best available--in that particular regard. Here it is distortion. It costs a dollar more, or even two. But the feature which I'm addressing, namely lowest possible distortion, is realized by many drivers today. To chose less capable drivers is cutting corners. There is no other compromise to distortion-free (virtually) drivers than cost. And the raise in cost is very limited, indeed. A few dollars max, way below the cost of the shiny enclosure for instance.

So, I offer my metric not only to Toole but to the whole industry. (If they want to donate, my bank account still has some appetite.) Set a limit from which we already know, that it is safe, regarding human hearing. Done--for the moment. We have a metric that works and can eventually start engineering (WTF, sorry)

 

[TimVG]

The Spin looks good yet there is apparently something wrong with the sound (we haven't yet ruled out a problem with the speakers).
Is this a case of missing the forest for the trees?

Too early to draw meaningful conclusions. EQing the speaker to match a different measured one (such as the M16) within +/- 0,5dB anechoically, and checking for differences below the transition frequency in-room, would be a good start. Another valid test would be (since bi-amping is an option) to correct the behaviour of the individual drive units past their passband. In one of my DIY builds (the one in my avatar) I could hear issues I wasn't seeing clearly in the summated result. The 2213nd woofer was however showing breakup issues that were being masked, and only showed themselves when measuring the drive units seperately, the difference was barely visible afterwards but it made a (to my ears) big improvement in terms of overall sound. I would encourage @amirm to give it a try when he's feeling bored

 

[GelbeMusik]

View attachment 71135

Maybe of less interest. I revived a pair of Canton GLE 100, which were found on the sidewalk as abandoned bulk junk. Only the tweeter fuses were burned. The XO showed some burning marks too. Most probably an accident with karaoke / feedback.

Ehm, this is how they look like as a pair, squeezed into some corners in-room @3meters, the dip at 20kHz comes from the mike not being centered perfectly.

Distortion, pair about 96dB at 3 meters:

These little guys are 40 (fourty) years old--each. So, what do You discuss for 18 edit: 19 pages now, and counting?

 

[edechamps]

Is it possible to show the simulated in-room response of the IL10, Focal Aria and M106 in the same graph?

You can use Loudspeaker Explorer which is specifically designed to do that kind of thing. However it does use the Klippel PIR which as @MZKM explained is slightly wrong (fixing that is still on my todo list).

 

[mjwin]

I don't post much on this forum, but I've been watching this thread develop & am a little perturbed by the tide of hostility which seems to directed at a genuine attempt to advance the state of knowledge. Of course, there's also a great deal of constructive discussion here, and that is to be applauded.

Personally, I think that Amir is doing a fine job of teasing out some of the second order effects which impact on perceived loudspeaker sound quality. Research is an iterative process where we're constantly trying to build on what we have, and question and improve the accuracy of our models based on observational data.

A lot of people here seem to be getting very upset that their dream single quality metric (spinorama) is somehow invalidated by this latest observation. This is really not the case. Ok, so the "one absolute metric" aspect might be called into doubt, but, as has been amply shown here by numerous quotations from Toole, etal, it was never intended to be the ultimate indicator of sound quality, simply a practical metric to use in the specification and evaluation of loudspeakers.

I think everyone would agree that frequency response has always been the primary quality metric for judging sound reproduction. Harman's body of research which, for the first time showed meaningful correlation between off-axis frequency response and listener preference, has been a huge step forward in this regard. Amir and colleagues are simply attempting to build on this body of knowledge to give better differentiation amongst high quality speakers which already show excellent frequency response and directional characteristics.

It seems that the term "THD", too is causing some upset. And, yes, I do (just) remember the THD wars of the 1970s! "Total Harmonic Distortion". The term makes harmonics seem bad and its use is, I feel, rather counter-productive. It's not the harmonics in themseselvs which are the problem, as musical instruments themselves generate large numbers of harmonics, often in excess of their fundamental tone; it's called "timbre!". The issue with measuring harmonics of a steady tone is simply that their presence in a test signal indicates nonlinearities in the system. The result, due to intermodulation, will be additional sum & difference products of all of the signals passing through it. And that is not good.

I think that it's important to understand what we are actually measuring when we look at "THD" figures. THD is actually a very good indicator of the linearity of a system at a given frequency. We're looking at the way in which the transfer function (input versus output) deviates from a straight line, and THD can give us an indication of the severity of this. By separating out the individual harmonics, as is routine with FFT based calculations, we can also get an idea of the type of non-linearity, be it square-law or cubic, etc. Although completely automated by using swept tones and mathematical analysis, for a given frequency the measurement consists of putting a pure (sine wave) tone into the speaker, and then taking the output (via a linear microphone), subtracting the original frequency, and measuring what is left. This residual consists of harmonics of the original signal (at 2, 3, 4.. times the frequency) plus noise, etc. Although this latter aspect is less relevant for speaker testing.

In the case of the IL10 speaker, we get a sharpish peak in our "THD" plot. This means that, at 1.7KHz other higher frequencies are present in the output. The sharp peak would suggest a resonance of some kind, and is indicative of a resonance of some sort. As has been suggested, it could be cabinet resonance, or a break-up mode in the speaker cone. In any case, the sound output due to distortion at this point is only 1.5% higher, which is about 0.13dB or so depending on how you add the signal and distortion product. Although hardly noticeable on the frequency response plot, this level of non-linearity is hardly insignificant and with a complex (music) signal can give rise to considerable unwanted distortion products due to intermodulation. Of course the actual audibility of this is subject to listening tests, and is sure to vary from listener to listener and type of music, etc. As Earl Geddes has pointed out, no direct correlation between THD/IMD figures and "sound quality" has been found, but I do think that Amir's comments of a "grungy" sound bear some consideration.

I'll just add that intermodulation distortion is generally linked to a thickening "dirty" quality to the sound, which is exactly what Amir is reporting. Ok, this is an anecdotal observation, but anyone who has listened to electronic music from the 1960s will be aware of the use of "ring modulators" and also know of this sound. Personally, I'm not sure of the attraction, as the resultant noise is quite horrid!

 

[Thomas_A]

View attachment 71135

Zoomed in to see differences more easily.

Here is comparison their "Tonal Balance", which is a normalized weighted average I do based on their on-axis and PIR:
View attachment 71138

Thanks, I got essentially the same in REW. The tonal response from the Revel and the Infinity is quite similar while the Focal is voiced a bit darker.

 

[GelbeMusik]

As Earl Geddes has pointed out, no direct correlation between THD/IMD figures and "sound quality" has been found, but I do think that Amir's comments of a "grungy" sound bear some consideration.

I'm absolutely on the same page regarding THD and instruments, even electronica. But one should really not take "the Geddes" for granted. The test would have, if at all addressed amps and A/D/A/DCs the like, not the very much more complex loudspeaker.

For example Doppler aka frequency modulation is not easily heard over headphones. But with speakers in a room there is not so much difference to amplitude modulation anymore. Resonances together with distortion products, never investigated maybe, could become objectionable, while each wouldn't be too bad alone. Etc.

Again, don't quote Geddes. For me his case isn't settled at all.

 

[ctrl]

If by algorithm you mean the Preference Rating then I don't agree that it is taking into account small yet audible problems, only tonal balance on- and off-axis.

Thought I made myself clear:

The Spinorama and the score give us a measure of the probability of how the loudspeaker in question will sound to most consumers and how flexibly one can influence the sound of the loudspeaker.

The "small yet audible problems" do not play a decisive role in achieving a meaningful score.
As stated in the previous post, the other minor influencing factors should be discussed in the review itself and, if they have a decisive influence, indicate an outlier.

Can highly recommend the ingenious post by @edechamps, which makes it clear that the score is about trends and probabilities and not about figure skating scores where the decimal point decides.

The score gives the trend, the blue line shows that at ninety percent reliability the first twenty ranks can appear in any order in the individual scores, but the mean trend is expressed by the score (if I interpret @edechamps data correctly).

It will never be possible to make a 100% reliable prediction about the tonal evaluation of loudspeakers, because if all possible sound-relevant factors are taken into account, it becomes too complex for a simple algorithm and of course there will always be individual preferences of the listeners.

In order to be able to make even more reliable predictions about the characteristics of loudspeakers, machine learning is a must.

If @amirm ever has a few thousand dollars too much because he no longer has to buy dog food after a pack of hungry, escaped lab rats ate his dog, he could buy an dummy head for binaural recordings and use it to record sound clips of the measured loudspeakers.

The forum would then rate these sound clips against always the same reference and thus provide the database for a self-learning algorithm that relates the measured data (frequency responses, distortions, CSD, ...) to the ratings.

 

[mjwin]

[...] For example Doppler aka frequency modulation is not easily heard over headphones. But with speakers in a room there is not so much difference to amplitude modulation anymore. Resonances together with distortion products, never investigated maybe, could become objectionable, while each wouldn't be too bad alone. Etc.

The Doppler modulation is indeed another issue which needs to be investigated thoroughly. It impacts on the theoretical advantages of multi-drive-unit (3+) speakers. As with all engineering solutions there's a trade off. Poorer vertical directivity vs lower doppler & intermodulation distortion... There's a lot of work still to be done.

Again, don't quote Geddes. For me his case isn't settled at all.

My apologies. I guess he's a "can of worms" as we say!

My point was that I've not seen any study where an existing distortion metric correlates directly with listener preference. I should do my own original research, but for now I thought I'd blame someone else lol!

[edit] I'll just add that in terms of formulating a more advanced test, a multitone test signal would be interesting. It's certainly the most brutal test & covers the full bandwidth of the speaker. The plots look "pretty" but the problem is how to evaluate the result numerically. If we generate the multitones to each exactly correspond to an FFT bin, then we should be able to remove them entirely. The resultant mess of intermodulation products could then be summed to give a single figure. But then we would have no differentiation between problems across the frequency range. We woud need to weight the frequency spectra according to some (yet) unknown criteria of human sensitivity.

 

[ctrl]

Almost useless for the industry. Toole serves the industry with consultancy to some success, which isn't bad at all. But, "useless" is just not true from the perspective of the customer. One might decide to set a safe limit, as the host here already did with -50dB of THD.

This type of limit is too simple. One should already consider the order of the harmonic distortion, as one would do when evaluating the HD2-HD5 diagram.

Therefore it would be "more correct" to use at least the weighting depending on the order according to Shorter/GedLee, because this takes the psychoacoustic factor of the masking into account.

The generated, weighted THD curve would then correspond more to the acoustic perceptions of distortion.

Since @amirm measures harmonic distortion at 86dB and 96dB, two different "safe limits" are needed, since the human ear becomes less sensitive to distortion at higher sound pressure levels.

 

[edechamps]

The score gives the trend, the blue line shows that at ninety percent reliability the first twenty ranks can appear in any order in the individual scores, but the mean trend is expressed by the score (if I interpret @edechamps data correctly).

Almost. The whiskers are 99% ranking confidence, not 90%. That is, there is a 99% chance a Genelec 8341A will be preferred over a Kef Q100. I'm not entirely sure I understand what you mean by "reliability the first twenty ranks can appear in any order". (Probability is hard…)

 

[Thomas_A]

View attachment 71135

Zoomed in to see differences more easily.

Here is comparison their "Tonal Balance", which is a normalized weighted average I do based on their on-axis and PIR:
View attachment 71138

Is that PIR data available? I see that there are some differences compared to that I downloaded. I made a comparison with my real room data for my DIY speaker (which in this one also contains room effects below 300 Hz, now fixed with Audyssey).

 

[pozz]

In the case of the IL10 speaker, we get a sharpish peak in our "THD" plot. This means that, at 1.7KHz other higher frequencies are present in the output. The sharp peak would suggest a resonance of some kind, and is indicative of a resonance of some sort. As has been suggested, it could be cabinet resonance, or a break-up mode in the speaker cone. In any case, the sound output due to distortion at this point is only 1.5% higher, which is about 0.13dB or so depending on how you add the signal and distortion product. Although hardly noticeable on the frequency response plot, this level of non-linearity is hardly insignificant and with a complex (music) signal can give rise to considerable unwanted distortion products due to intermodulation. Of course the actual audibility of this is subject to listening tests, and is sure to vary from listener to listener and type of music, etc. As Earl Geddes has pointed out, no direct correlation between THD/IMD figures and "sound quality" has been found, but I do think that Amir's comments of a "grungy" sound bear some consideration.

This is mentioned over and over by Toole in his book and elsewhere: problematic resonances are a major detractor from quality. So distortion figures as such are not the target other than their indications of a poor or inappropriate driver, crossover design or cabinet construction.

There have been other speakers measured by Amir with distortion in that area. The main reason this has gotten so much attention here is because the IL10 is otherwise an excellent speaker.

What science are you referring to, preference polls? That's not science perhaps scientifically gathered data.
And no, reflections are not universally preferred.

If you listen on-axis to a single speaker in the middle of the room what impact does the room/directivity have in the sound at the listening spot?

@tuga Below is a page from Toole's book. The data is based on the following study led by Ben Shirley: “The Effect of Stereo Crosstalk on Intelligibility: Comparison of a Phantom Stereo Image and a Central Loudspeaker Source” http://www.aes.org/e-lib/online/browse.cfm?elib=14174



This is another paper, this time by Earl Vickers, that uses the above study as well: https://www.sfxmachine.com/docs/FixingThePhantomCenter.pdf He states:

Breaking this down in terms of the FR data above:

• Anechoically, measuring stereo using a dummy head produces the crosstalk null around 2kHz, which corresponds to head width.
• Anechoically, measuring a single center channel produces elevated response around the same area due to HRTF.
• However, in a reflective room, that difference between a stereo and single center speaker is made much smaller: this is due to the effect of reflections, which fill in the missing spectrum, though not completely. All this is with measurements and without touching preference tests. [Edit: This means that applying too much absorption treatment will cause crosstalk nulls to reappear.]
• You may cite studies about standing waves and their destructive/constructive interference and ask if they contradict the above. They do not. The link between interference and room response is perceptually accurate only when averaged above the transition region, since hearing itself follows an averaging function. This does not mean that all rooms are created equal: they are not. There are distinct advantages to having specialized acoustical construction, although the construction standards are very general, especially for things like reverb time vs. frequency and noise curves. Not a lot of psychoacoustic research has been on on room construction, especially for small rooms. Nothing like comparative blind testing. But putting together what is available shows that the average living room is ok. Not great, but ok.
• The specific Harman/Revel conclusion was that the compensatory effects of reflections and direct response should be accounted for and addressed in speaker design as well. The idea here is to characterize the role of the loudspeaker as much as possible given that rooms themselves are not the subject of design. Thus the full spin measurement. Again, this barely touches on preference studies. This is just good analysis of loudspeaker/room interactions. Where preference enters is when the goal shifts from characterization to design and the attempt to define optimal targets for loudspeaker FR response and the resulting room curve. At this point you can say: the Harman approach to speaker design is just one among many, but that statement just has little to do with the analytic procedure for characterizing speaker behaviour.
• You could add that speaker distortion should be measured on top of the spins to get the real picture. With the IL10 it is perhaps the first time that this kind of measurement matters given how good that speaker measured otherwise. I'm of the opinion that those two resonances on its FR are just that: mechanical issues that should have been addressed before the speaker was released for sale. If they were to be suppressed the loudspeaker would likely lose the negative qualities, although I still feel that the 4kHz roughness off-axis is very important given the ear's sensitivity in that range. I don't think we can make too many claims about distortion beyond that. For loudspeakers with no obvious tonal or mechanical issues the role of distortion is unclear.

Thanks for the hint on the "aperture effect". It is something I wanted to convey to some folks discussing (again) time alignment (again) and again, always forever.



Almost useless for the industry. Toole serves the industry with consultancy to some success, which isn't bad at all. But, "useless" is just not true from the perspective of the customer. One might decide to set a safe limit, as the host here already did with -50dB of THD.

This is a metric. It is engineering where it starts: will the pyramid stay upright, will the bridge hold, will the dam not break, will the car break == decelerate under every circumstance etc. Setting a safety limit for starters, actively decide for a big margin, refinement comes later. Fault is not an acceptable option. Basic engineering. Btw, this crude metric is even well accepted in decision making. It discriminates the doable versus the impossible.

Devoid of any further knowledge one has to go for the best available--in that particular regard. Here it is distortion. It costs a dollar more, or even two. But the feature which I'm addressing, namely lowest possible distortion, is realized by many drivers today. To chose less capable drivers is cutting corners. There is no other compromise to distortion-free (virtually) drivers than cost. And the raise in cost is very limited, indeed. A few dollars max, way below the cost of the shiny enclosure for instance.

So, I offer my metric not only to Toole but to the whole industry. (If they want to donate, my bank account still has some appetite.) Set a limit from which we already know, that it is safe, regarding human hearing. Done--for the moment. We have a metric that works and can eventually start engineering (WTF, sorry)

I don't completely understand your perspective. You still insist on the importance of distortion metrics but have quoted no metrics or data other than Amir's. The issue as I see it is the complete lack of data other than the anecdotal. There is no comprehensive perspective regarding distortion offered by anyone. Even when taking into account the limitations of Geddes' approach, there's a spreadsheet available with all the calculations built-in. We just have to punch in the numbers to see what comes out. The comparison with the preference score would be very instructive and worthy of additional analysis from a statistical perspective.

Despite doing a lot of research I haven't found any firm design targets for distortion. My main conclusion is that loudspeaker designers use distortion measurements as diagnostic tools primarily. There's nothing like an accepted target distortion curve in the industry, only a set of vague expectations. Psi Audio for example, stated passingly in one interview that they consider the limit to be -40dB (to be precise, the statement was more like: it's really good that our speakers show distortion at -40dB or less).


The final thing I wanted to add is that despite all the comments about Toole no one has yet pulled out his book and addressed the Olive test and the IL10 result specifically. The result is here:


It seems like his conclusion, based on the anechoic data, was that the IL10 (Loudspeaker 1) was a really good apart from minor issues. Maybe the Klippel measurement on ASR has showed that this was overstating the case, somewhat. This perhaps isn't that surprising when compared to what they were used to as the typical result:

 

[GXAlan]

@amirm , I know you're busy with the Denon project -- but any thoughts on IMD as an explanation for why the IL10 didn't sound as good as the M16 to you?