Why is tuning IEMs flat after the resonance peak so difficult? Graph

[Fraxo]

So thanks to @ADU and his fantastic explanation on the cause of shallow insertion depth 7-9.5kHz resonance peak (and the other peaks) in IEMs, I'm lead to ask another question.

It became very clear that when using ear simulators - most IEMs struggle to have a linear frequency response before and after the resonance peak, but what stands out to me is the slope following the peak.
I was wondering why that is the case... Clearly a peak is a reason there will be depression around it, but it doesn't explain that deeper "dip", and some IEMs demonstrate a very aggressive slope which is audible in listening tests (not just visually or coupler related) and lacks as a result.
To clarify again - I hear those dips with a sine sweep as well and with pink noise, just as I hear the a resonance peak. Some IEMs slope over 12-14dB before peaking again somewhere post 12kHz

In most graphs I see and measure myself - that feels like the default tuning result of many models, and any other tuning in that range seems harder to achieve.

So to summarize - why is it so difficult to tune a measurable flat FR post the resonance peak? Is it driver/shell structure related? Etc...
Could there be a new driver design to battle the slope?
Would generally appreciate any input on this topic, even beyond my specific questions as I'm glad to learn

 

[someguyontheinternet]

High frequency measurements in general are not that reliable. Amir includes this paragraph in his reviews to make it clear:

Note: The measurements you are about to see are made using a standardized Gras 45C. Headphone measurements by definition are approximate and variable so don't be surprised if other measurements even if performed with the same fixtures as mine, differ in end results. Protocols vary such as headband pressure and averaging (which I don't do). As you will see, I confirm the approximate accuracy of the measurements using Equalization and listening tests. Ultimately headphone measurements are less exact than speakers mostly in bass and above a few kilohertz so keep that in mind as you read these tests. If you think you have an exact idea of a headphone performance, you are likely wrong!

The important part being ,,Ultimately headphone measurements are less exact than speakers mostly in bass and above a few kilohertz so keep that in mind as you read these tests."

There is also a video where he roughly outlines the process you have to use when looking at high-frequency measurements. I'd have to look for the video, but in summary you need to create sort of an "envelope" around the peaks and valleys to get an estimate of the response.

To get a good EQ that will also work on your ears you should work with a sweep or noise and manually tune filters against a reference headphone or speaker.

 

[Jimbob54]

Do you want them to extend roughly flat after the peak towards 20khz or to hit a target roll off ?

 

[Fraxo]

High frequency measurements in general are not that reliable.
To get a good EQ that will also work on your ears you should work with a sweep or noise and manually tune filters against a reference headphone or speaker.

As I've explained - I don't just base it on visual graphs but on listening. It's very noticeable with sine sweeps and pink noise + EQs...
I've made numerous tests with my ears, confirmed with couplers and compared to other measurements in addition real life experience. It's audible.

I hear a significant depression post the resonance peak, beyond the level it peaked at so graphs match what I hear quite well.

 

[Fraxo]

Do you want them to extend roughly flat after the peak towards 20khz or to hit a target roll off ?

If you read my previous post you'd see @ADU made a great point that there's an expected post 11kHz resonance peak due to the ear canal's structure, so I wouldn't expect it to go flat or extend that extremely (def not 20kHz lol).

I would prefer that the peak was mitigated as much as possible and that the 7-11kHz range would be more linear and not so reduced after the peak.
There's this unavoidable "valley" that seems to appear on most IEMs in variable levels.

 

[Jimbob54]

If you read my previous post you'd see @ADU made a great point that there's an expected post 11kHz resonance peak due to the ear canal's structure, so I wouldn't expect it to go flat or extend that extremely (def not 20kHz lol).

I would prefer that the peak was mitigated as much as possible and that the 7-11kHz range would be more linear and not so reduced after the peak.
There's this unavoidable "valley" that seems to appear on most IEMs in variable levels.

But if you ignore the post 11khz "second" peak, there is always going to be a pretty steep post 8k roll off. Are you asking why this is?

 

[Curvature]

Beyond 10kHz most current ear simulators do not model the impedance of the ear correctly. Tuning is still possible, especially broadly, but imprecise.

At these frequencies you're into individual anatomical differences. Subjective EQ is probably the most useful tool for compensating. I don't see a way around it. We can't expect IEMs to work excellently in the HF as is other than by luck. It's a technological limit.

 

[AdamG]

Wild ass guess. Because most people can’t hear much beyond 12k anyway. Demographically speaking.

 

[Fraxo]

Beyond 10kHz most current ear simulators do not model the impedance of the ear correctly. Tuning is still possible, especially broadly, but imprecise.

Once again - I was talking and pointed out the post "8k" resonance peak area even if it doesn't extend much further. And once again - I'm talking about real life experience with my ears that confirms it.
It's all in the actual post and my further comments my dude... Seems like people just comment without reading on this one haha.

 

[Fraxo]

Wild ass guess. Because most people can’t hear much beyond 12k anyway. Demographically speaking.

Well the slope starts to happen from 7-9kHz usually and is prevalent within 99% of people's hearing range...

 

[Curvature]

Once again - I was talking and pointed out the post "8k" resonance peak area even if it doesn't extend much further. And once again - I'm talking about real life experience with my ears that confirms it.
It's all in the actual post and my further comments my dude... Seems like people just comment without reading on this one haha.

You misunderstood my post.

You hear what you hear in the HF because of the combo of the IEM response, insertion depth, seal and your ear canal geometry and dimensions. Manufacturers are currently limited about what they can do in the HF region to address all of those features and account for individual listener variation, with one of the main reasons being that most industry standard measurement gear is inaccurate beyond a certain point.

 

[majingotan]

If you read my previous post you'd see @ADU made a great point that there's an expected post 11kHz resonance peak due to the ear canal's structure, so I wouldn't expect it to go flat or extend that extremely (def not 20kHz lol).

I would prefer that the peak was mitigated as much as possible and that the 7-11kHz range would be more linear and not so reduced after the peak.
There's this unavoidable "valley" that seems to appear on most IEMs in variable levels.

Did you PEQ the 11-13 KHz range by a factor of 1-5 dB and how did cymbal percussions sounded with them?

I have heard the 3 IEMs on the graph and personally own one of them (the most muted response from 11-13 KHz) and tried adding 5 dB. The cymbals timbre sounded a bit strident, as if I'm listening next to it rather than a couple of feet away from it

 

[Fraxo]

You misunderstood my post.

You hear what you hear in the HF because of the combo of the IEM response, insertion depth, seal and your ear canal geometry and dimensions. Manufacturers are currently limited about what they can do in the HF region to address all of those features and account for individual listener variation, with one of the main reasons being that most industry standard measurement gear is inaccurate beyond a certain point.

I understand there are many variables, but my question was - what causes this specific dip for so many companies to struggle with it, since it's audible and does match real life listening. For example - if you told me that a specific angle in the insertion method is causing... or if drivers are factory tuned as such due to.... that would help. I do enough testing to personally know (u of course are welcome to disagree if you tested otherwise irl) that it's a prevalent, common issue with a lot of "safely" tuned IEMs. A majorly hyped one (at the time) to demonstrate how bad this effect is - is the Moondrop Aria. Aggressive dip and demonstrates all that's bad about it imo.

 

[Fraxo]

Did you PEQ the 11-13 KHz range by a factor of 1-5 dB and how did cymbal percussions sounded with them?

I have heard the 3 IEMs on the graph and personally own one of them (the most muted response from 11-13 KHz) and tried adding 5 dB. The cymbals timbre sounded a bit strident, as if I'm listening next to it rather than a couple of feet away from it

I'll humble myself as I start to wonder if perhaps I didn't explain it well enough. To remain consistent I'll quote myself from previous comments, as I did specify several times:

"I would prefer that the peak was mitigated as much as possible and that the 7-11kHz range would be more linear and not so reduced after the peak.
There's this unavoidable "valley" that seems to appear..."

So to answer your question @majingotan - 11-13kHz range is extremely more difficult to tune, varies a lot for people and would need insane Q factor. The only way to tune very high treble properly (which I have) is to increase the dB\Octave ratio of the Q factor, since normal filters would simply not suffice and either create way too narrow and sharp increase, or be too wide and affect other areas (that's of course after finding your personal resonance peak for the model).
I've demonstrated it here:

In order to tune extremely high freq there's many issues and those online tools won't account for so it's irrelevant.

Back to this:
"I would prefer that the peak was mitigated as much as possible and that the 7-11kHz range would be more linear and not so reduced after the peak.
There's this unavoidable "valley" that seems to appear..."

I clearly refer to the 7-11kHz, so for example this is the "valley of death" in the Aria haha:

Might even go with 7-12k range or around that, but the evil slope is at it's lowest before reaching the 12kHz mark so boosting would be lower, and I would'nt quantify the dB boost according to graphs. So if found the correct spot and dB\oct etc to not affect other freq around it (and confirming with measurments) - I'd go by ear and not just assume 5-6dB there is what's necessary, in many cases it would be less.

Your models don't demonstrate much of that, although the Andromeda might be a suspect if normalized to 8k, but it seems to fit deeper or tuned to present the peak further so it shouldn't lose detail as much. I'd need to measure myself to know for sure, I'm just saying you prbly didn't necessarily correct what you intended to.

View attachment 236080

A few known ones with the dip some of HBB's lower budget tuning sets:

Which if someone likes then by all means have fun with it, but there's simply way too much information missing after the resonance peak for me personally. The Olina fakes it out by recovering more post 11k but now we're in peaksville. Again - all tried and tested with my ears, not a noob graph master here.
Not many people understand that a sharp dip doesn't just recess, it also creates "sharpness" around it so that's why the aria sounds "metallic" to some although it's quite a warm set otherwise with it's treble quantity.

Back to high treble:
EQing dips and peaks might sound extremely unnatural or easily be done incorrectly, as it requires a lot of figuring out first. You'd have to find your own resonance peak with every IEM (as graphs often normalize to 8k but yours might be higher or lower with each model), then measure it to that peak and only then adjust that EQ band with a capable EQ that has adjustable dB\oct features and perhaps even linear phase to test, as extreme sharp eq changes change the phase drastically etc....

I hope I was able to explain a bit better and perhaps help further regardless, but I was wondering if there's a specific known limitation with IEMs that makes it much harder to avoid dips in the treble area. Peaks are hard to combat due to natural Resonance Peaks of the ear canal, I get that. But with all the R&D done, I wonder and want to learn more if someone has some experience.

Also would love to expand on anything I've mentioned up here and gladly debate disagreements if you've reached different conclusions

 

[Cars-N-Cans]

Problem with that region is the wavelengths are really short, and there are standing waves and cancellations in the ear canal, as you probably know. Your physical anatomy will also impact the IEMs response. Any time you have something that's relatively large compared to the acoustic waves that are travelling through it in some fashion, life gets more complicated, and the measuring equipment less reliable on small scales. There is also the possible issue of perception as well. I can still hear to 16 kHz, and above about 8-9 kHz the pitch becomes ever more compressed with each linear increment in frequency. Its hard to tell how people perceive that region, assuming they can still hear it, since there is so little energy there in music and sound reproduction. Really there is no easy answer due to all the variables that are at play. At best they could try to just make it roll off in an orderly fashion, but the odds of getting that same response once you stuff them in your ears is slim to none unless you happen to have the same ears as a GRAS fixture or whatever you test with, and you can get them to go to the same depth each and every time. And then there is the coupler itself and how accurately, or inaccurately it represents the ear canal, etc. etc. etc. I suppose you could try some form of in-ear measurement if you have the equipment, but even then its not likely to be exact. Personally I just try to split the difference by listening and leave it at that. Its a shortcoming that's not likely to go away any time soon since unlike speakers and rooms that are macro-sized objects that have easily measured characteristics, we are living things that don't easily accommodate instrumentation.

 

[Cars-N-Cans]

View attachment 236166
Which if someone likes then by all means have fun with it, but there's simply way too much information missing after the resonance peak for me personally. The Olina fakes it out by recovering more post 11k but now we're in peaksville. Again - all tried and tested with my ears, not a noob graph master here.
Not many people understand that a sharp dip doesn't just recess, it also creates "sharpness" around it so that's why the aria sounds "metallic" to some although it's quite a warm set otherwise with it's treble quantity.

One thing that is missing from all of that is the group delay. Without that, there is no way to determine if the peak is from the IEM itself having some internal resonance, or if it originates inside the measurement fixture from standing waves. Any standing wave will be seen as the group delay suddenly exhibiting a large, narrow increase in duration around the area where the measurement shows a peak in the response. If its flat and only on the order of microseconds in that region, then its more likely from the IEM itself or just a resonance in the fixture. I'm sure resident acousticians here at ASR can elaborate more on the finer details. Edit: Although I think this is mainly more of an issue with headphones, but still it shows the uncertainty of drawing conclusions above 8 kHz in the measurements.

 

[Fraxo]

One thing that is missing from all of that is the group delay. Without that, there is no way to determine if the peak is from the IEM itself having some internal resonance, or if it originates inside the measurement fixture from standing waves. Any standing wave will be seen as the group delay suddenly exhibiting a large, narrow increase in duration around the area where the measurement shows a peak in the response. If its flat and only on the order of microseconds in that region, then its more likely from the IEM itself or just a resonance in the fixture. I'm sure resident acousticians here at ASR can elaborate more on the finer details. Edit: Although I think this is mainly more of an issue with headphones, but still it shows the uncertainty of drawing conclusions above 8 kHz in the measurements.

This is the group delay for a peaky measurement at the resonance peak.

Does that answer what you meant or not?

 

[Cars-N-Cans]

This is the group delay for a peaky measurement at the resonance peak.

Does that answer what you meant or not?

Thanks. Yes, Id presume this probably is a resonance in the IEM (edit: The peak between 8-9k) since they correlate with each other and its only a modest increase. Basically, put most simply there is a moderate resonance which delays the signal a bit since it takes time to excite it. It would be interesting to examine some that don't have any peaks. I suspect it would continue to be minimum phase (i.e. around 0 msecs) through that region. This probably implies its the IEMs, but there is still the likely possibility of these being influenced by the surrounding ear canal or cavity in the measurement fixture, but I don't know enough about IEMs to say for sure what the potential sources of the resonances could be, or their solution for that matter.