What Causes This Output from Headphones?

[mike7877]

The circle covers 1kHz and 7kHz in the screenshot above. It can start a bit lower, like 800Hz, but not usually, and doesn't typically go much higher (it's actually usually diminished 1-2kHz sooner at ~5-6kHz).
See the really big dip after the circle? See how there's a move in the positive direction first? Literally every time a headphone has this circled behaviour, there is a rise right after it's done. It's also usually followed by a dip - and the dip is usually much narrower than the rise (like above), but not as severe as above... not nearly! More like equal to the rise, maybe 1.5x the rise.

I've seen this in many headphones - usually over head large cans (like HD 600) and more often open-back then sealed variants. Smaller headphones can have similar behaviour, but not the same (from what I've seen, they lack the aforementioned characteristic rise and dip).
This is probably present in over 60% of open back style headphones (like HD 600) and between 20-40% of closed back (I can't say how often as reliably for closed back headphones because I'm less interested in them so haven't looked at as many).

I believe this aberrant aspect of many 'phones' group delay plots is due to the same underlying cause:

I've tried considering driver membrane design, driver mounting, driver distances (to ear, from rear to inside of can), material, and every time I'm unable to isolate it because there's another model headphone done the exact same way which doesn't jump up and down from 2-4ms (p-p) for 1-3 octaves in the most critical bands (for detail, imaging, spatial queues)


I have two questions:

1.) The Mystery: Just what is it that causes this???

2.) Less Mysterious: How does this sound?



The mystery might not be so mysterious - things are a mystery until they're not

And the sound... Keep in mind - I do have a couple pair of headphones, but... they all have this disease!

I think I know the sound, but I could be wrong because I don't have a healthy control. Well, I do, but they're speakers.
The most obvious way it manifests to me is in peoples' singing - especially more airy singing. The "air" quality/aspect seems to be a result of really short, high frequency pulses at the tops and bottoms of the root note being sung. I believe the air-ier the sound of the singing, the louder these high frequency pulses are.
So, since these pulses happen at the peaks, they should be in time with the frequency. I believe that what I'm hearing through my headphones is these high frequency components not arriving in time (with the peaks and troughs of the root note). It's not off by a lot - the area I believe contains most of the energy of these high frequency components is 2-4kHz. Two to four kilohertz response of my HD 650s seems to arrive between +/- 1 to +/- 1.5ms. You may think 2-3ms variation isn't a lot, but it is! Example: When I use my PC to manipulate the output of my electric guitar like an amplifier, the 2.66ms round trip latency can be very distracting!
Lastly, yes, I did say I don't have a healthy headphone to compare, but I do have some good speakers and a decent room, and until I recently got back into headphones, I hadn't noticed this phenomenon on the vocals.

 

[KSTR]

Group delay is the derivative (slope of the curve) of phase vs. frequency (frequency on a linear scale), and derivatives are always more noisy than the source data. Good engineering practice is therefore to apply a resonable smoothing to the phase data before taking the derivative.

For HP measurement the phase is noisy/wiggly because of typical resonances seen in that frequency range. But those acoustic resonances are audibly benign unless in rare cases, also they come "by design" because they are unavoidable in over-the-ear headphones, sort of room resonances bascially.

These unsmoothed group delay plots exaggerate the actual impact big time and IMHO should best be ignored or at least mentally smoothed by the viewer.

 

[markanini]

Sound doesn't always move in a straight line or gets fully absorbed when it reaches a surface. So, it's likely related to internal reflections.

 

[swamps]

I've seen this in many headphones - usually over head large cans (like HD 600) and more often open-back then sealed variants.

For dynamic drivers this is most likely caused by internal reflections from the baffle, not an issue in of itself in this case. If the group delay is much higher around this region then it may be an issue with the driver itself and this driver issue very much audible as a smear.

Planar headphones (at least the good ones) have much less group delay at ~2kHz because they have no baffle, only the driver itself and whatever reflections caused by the earpad and the head/ear.

 

[TurtlePaul]

3 khz is around a 5” wavelength. I think this is from reflections inside the earcup reaching the ear at different times.

 

[kemmler3D]

You may think 2-3ms variation isn't a lot, but it is! Example: When I use my PC to manipulate the output of my electric guitar like an amplifier, the 2.66ms round trip latency can be very distracting!

Hold up, first of all, you have 2.66 total latency on a analog-digital-DSP-analog roundtrip? What the heck interface and PC are you using? And second, you can easily tell when it's in or out?

Not a guitarist or even a real musician, but have played MIDI through DAWs with varying delay and I find that <10ms feels snappy, but under 5ms would be hard to detect for me.

Anyway, I'm not sure if these big spikes in GD are real or measurement artifacts.

You might try messing with some allpass filters in EQAPO to see if you can make the effect you're hearing better or worse. If adding phase distortion doesn't change what you're hearing here, then it must be because of something else.

 

[somebodyelse]

When I use my PC to manipulate the output of my electric guitar like an amplifier, the 2.66ms round trip latency can be very distracting!

Do you find the difference between standing 1m and 2m from your guitar amp distracting?

 

[mike7877]

Hold up, first of all, you have 2.66 total latency on a analog-digital-DSP-analog roundtrip? What the heck interface and PC are you using? And second, you can easily tell when it's in or out?

Not a guitarist or even a real musician, but have played MIDI through DAWs with varying delay and I find that <10ms feels snappy, but under 5ms would be hard to detect for me.

Anyway, I'm not sure if these big spikes in GD are real or measurement artifacts.

You might try messing with some allpass filters in EQAPO to see if you can make the effect you're hearing better or worse. If adding phase distortion doesn't change what you're hearing here, then it must be because of something else.

Lol yes - the interface isn't anything too special, just the Babyface Pro - the PC is quite fast, though, regarding latency, not necessarily the newesta 9600K @ 5200MHz (1.42V), 16GB DDR4 3900 CL 14 (14-14-14) (Samsung B-Die @ 1.63V). AIDA64's reported memory latency is 38ns (a typical 9600K with standard RAM (say DDR4 2666 CL18) running at stock 4.3GHz (all core boost) would report something in the 70s-80s.
Though the 9600K is getting older, DDR5 took a looooong time to be able to reach sub-40ns - it might still only be in the realm of overclocking, and before a year ago extreme overclocking (ie liquid nitrogen). I haven't paid as much attention to PC development since I built my last PC about a year after the 9600K. I didn't know if AMD was going to have another X3D variant (Intel did the same thing {strapping 128MB of ultra fast cache straight onto the CPU die in 2016/17 with their 5th gen and never did it again) so near the end of 3rd gen Ryzen I grabbed a Gigabyte x570s ACE Max motherboard (basically the best motherboard you could buy for overclocking the 5800X3D, though slightly overpriced at $500) and got my 5800X3D. It disappointed me a bit with its static clock not being able to be held above 3.6GHz (higher frequencies require turbo, and I always disable turbo: to minimize latency with audio you must disable turbo on Intel machines. Anyway, I built that thing, tweaked the heck out of it, moved my Gigabyte 3080 OC 10G (first edition with full hash rate allowed) into it (@1950MHz/21.0Gbps core/VRAM), and have been cruising with it since. I think 5800X3D will do me for 10 years, same with the 3080 (I waited in line for that thing on release - I was the second or third person in line, and they called me to pick it up a week later - shortages were HORRIBLE: except for the very first people, everyone basically had to wait 6-8 months or pay 2-3x MSRP for the card.
I haven't kept up with the most recent developments (Intel's 14/15th Gen or "Core Ultra" stuff, and AMD's 9000 series) because I've been busy and they don't concern me because my current machine does everything I need it to. The AMD which I use for everything + occasional FPS (first person shooter) games, its face is a 1920x1080 LCD with 280Hz refresh rate - the 3080 is perfectly sufficient, along with the 5800X3D, to drive it. Since they merged consoles with PC gaming, the CPU demands won't be rising too much into the future, and as GPU demands go up, I'm sticking at a lower resolution which will keep framerates up, easily above 120, most often over 200, for all games into probably 2030, some beyond. The GPU might start to feel a little stale once nVidia launches their 7000 series (for reference 3080 is 3000 series which came out in 2020, the 5000 series literally just came out). So yeah, since I don't keep up, maybe 38ns RAM performance is top end stock performance now. But my 2019PC matches it xD
If you want to configure your PC to support its lowest possible latency, you can message me (or others who read this who are interested - it's not too hard, though if you bought your PC from a company like Dell or HP and it's not a gaming PC with a normal BIOS, there's not too much on the hardware configuration end that you can do, except install the fastest RAM the CPU supports, and make sure you're running dual channel mode (have at least two sticks in)

 

[mike7877]

Sound doesn't always move in a straight line or gets fully absorbed when it reaches a surface. So, it's likely related to internal reflections.

This made a lot of sense to me when I started thinking about possible causes, but after a while I saw that there are so many similar headphone designs - some of them have this problem, and others just... don't!

 

[mike7877]

For dynamic drivers this is most likely caused by internal reflections from the baffle, not an issue in of itself in this case. If the group delay is much higher around this region then it may be an issue with the driver itself and this driver issue very much audible as a smear.

Planar headphones (at least the good ones) have much less group delay at ~2kHz because they have no baffle, only the driver itself and whatever reflections caused by the earpad and the head/ear.

Is a smear like what I described in the OP?

With more time thinking + process of elimination, it really seems like the driver has to be at fault for this.
The chart I used for headphone group delay is from here, Amir's HD 650 review. I've looked at the drivers of the HD 650s... instead of typing a long description, I'll paste a picture (so much easier lol)

The orange circle is the copper voice coil. The driver is kind of like a ring radiator tweeter with around middle instead of a phase plug.
What Sennheiser does is, they have about 60% of the driver (the part outside the voice coil) covered with a very thick material - a circle in the middle is about the same size as the dome part of the driver in the center. I believe this is to optimize frequency response to the listener, somewhat focusing the beam at the center. Being so close to a relatively large driver, frequency response is sure to vary from position to position, so kind of collecting them all and gathering them to point at the ear probably leads to a more balanced frequency response.

At first I was thinking, maybe there's some weird interaction because of the textured outer part of the driver (woofer surrounds can cause some issues when part of it is moving out of phase with the cone). But just now I got to thinking - the probable, above, frequency response smoothing might come at this cost - redirecting the sounds to the middle like is done, causes internal diffraction in the rear chamber of the chassis, and since the path is longer, these frequencies arrive later (or sooner, or whatever they do).

I wonder if this could have been fixed with more distance to the ear and bigger cans!

I honestly wouldn't mind if each can was the size of my head!
As long as it isn't heavy their size could be anything!

Of course the common materials we have available to use to make headphones are only so stiff per their weight, and compensating for that increases cost,


I just want to add, this isn't necessarily the reason for so many of these headphones' messy 1-5kHz group delay plots. It might be the HD 650's and some others, but I'm pretty sure some of the headphones I've seen with this problem have different internal configurations... I guess in those cases it would have to be the driver... seeing how the surround is hump-shaped, there are many angles at which these frequencies will be being emitted at - constructive and destructive interference would be expected then.
But does this affect group delay, or only frequency response? The distances we're talking inside of headphone cans aren't any further than 120mm.

Actually... that's about 2.8kHz - centered pretty much perfectly in the problematic range I gave which is worst in the middle!

But then the remaining question - why do some headphones NOT have this problem?! They're about the same size and shape, -also, I just scrolled through a thread of a bunch of different headphone drivers;they all look pretty similar...

 

[mike7877]

Do you find the difference between standing 1m and 2m from your guitar amp distracting?

Good point - my speakers are about 2m from me, so it's more like 9ms when the round trip is 2.66ms. I had a feeling it was longer than the stated 2.66ms!

I just looked this up - apparently a "good" drummer will keep time to within ~5ms.

Anyone who can tell a "good" drummer from a mediocre drummer - if they were to learn an instrument and then try to play it with its sound coming 9ms late, they would definitely notice a problem. Unless, maybe, if they were a worse than mediocre player. Usually people can hear better than they sound, though, so even if they couldn't play to within 9ms, they'd probably still be able to hear. Unless they just weren't blessed with rhythm at all (hey it happens).

My name is actually mike7877 FS - friends call me mike7877steadyclock_vanishingly-low-jitter. Once a year, when the clock flips from 23:59 to 0:00 on Dec 31, I go into my garage and clap into a microphone - it's connected to the world atomic clock. Yes, it is me, keeping all the satellites from crashing into each other - you're welcome!

But seriously... just like there are people with perfect pitch and perfect relative pitch, there are also people who can keep time exceptionally well. I am one of them.

 

[somebodyelse]

Anyone who can tell a "good" drummer from a mediocre drummer - if they were to learn an instrument and then try to play it with its sound coming 9ms late, they would definitely notice a problem.

You would HATE playing a real organ - the delay from key press to sound reaching you can be an order of magnitude more.

Unless, maybe, if they were a worse than mediocre player. Usually people can hear better than they sound, though, so even if they couldn't play to within 9ms, they'd probably still be able to hear. Unless they just weren't blessed with rhythm at all (hey it happens).

Is this relative or absolute synchronization, and if absolute, from whose point of view? Path differences for an orchestra can easily be more than 9ft. It's trickier for amplified bands with monitors I've often seen performances where I thought timing was off, but never thought about it at this level.

 

[swamps]

Is a smear like what I described in the OP?

Maybe but probably not. My best example of smear is the AKG K612 compared to the K701 since I had both. They both have the exact same internal structure, same materials and all except the driver.

Here is the CSD of the K612 (courtesy of Solderdude):

Now compare the 2kHz to the K701:

I've normalized the FR with EQ (they aren't that different to begin with aside from the massive roll-off after 10kHz) and same pads, the K612 sounds like a photo slightly out of focus. It's like soggy crackers, it doesn't taste bad but you can tell something's wrong.

What Sennheiser does is, they have about 60% of the driver (the part outside the voice coil) covered with a very thick material - a circle in the middle is about the same size as the dome part of the driver in the center.

I only tried out the HD650 a few times and from the images online I thought this material was thin like the AKGs, but if it's thick that explains why the soundstage is so squished.

From my experiments any thick material covering the outer edge of the driver and directly in front of it massively reduces the soundstage.
Of course this is a tradeoff and I definitely wouldn't remove this part covering the driver as it'll ruin the FR with all sorts of fun resonances.

Based on absolutely nothing I'll say this is one of the main reasons why the HD6xx series is so well behaved overall.

I wonder if this could have been fixed with more distance to the ear and bigger cans!

I've tested this as well and it doesn't fix it. This only adds "depth" but not "width". You could make a bigger headphone out of the HD650 but as long as there is this thick material in front of the driver the soundstage width won't improve.

*Not an ideal apples to apples comparison because I experimented on different drivers but they all behave the same.

 

[Heinrich]

For dynamic drivers this is most likely caused by internal reflections from the baffle, not an issue in of itself in this case. If the group delay is much higher around this region then it may be an issue with the driver itself and this driver issue very much audible as a smear.

Planar headphones (at least the good ones) have much less group delay at ~2kHz because they have no baffle, only the driver itself and whatever reflections caused by the earpad and the head/ear.

Is 'smear' a word in context of sound? You address a topic that I'm especially after. The headphone shall, more or less depending on 'preference', replicate an outer diffuse soundfield around the listener in the listener's ears. Sounds more complicated than it is. What would be measured at the eardrum of the listener sitting in a diffuse soundfield, compared to what the headphone brings to the eardrum.

With headphones, other than with IEM, the ear's pinna is involved in the process. It has to do all the reflections that you may see in the groupdelay. In parts the spikes are evidence that the concept works. To remove them altogether is not a valuable target.

But the question, if all those spikes are good in that sense is another and quite valid one. Sure you know about the head related transfer function. How it really looks like is not easy to understand, because the literature mostly shows smoothed curves. In order to evaluate the referring spikes the real, detailed HRTF of the dummy head or HATS respectively as used in the measurement is needed.

Btw: planars are not as simple as the industry conveys. In contrast to what one might expect, the relatively chaotic movement of the diaphragm at higher frequencies may help with the 'diffuse' paradigm.

 

[Heinrich]

What would be measured at the eardrum of the listener sitting in a diffuse soundfield, compared to what the headphone brings to the eardrum.

Before you tell me, rightfully, that I'm a bit wrong on this. The equivalence between the diffuse soundfield and the headphones response, both measured at the eardrum, is meant as amplitude frequency response. It is not so that the headphone itself shall make the sound diffuse. That's my personal speculation. The original article by Günther Theile remains a bit vague in this regard. I cannot refer to any in-depth literature here.

We are not bats with a that fantastic directional hearing due to nose and ear shape. But I would expect that the pinna plays a role in forming the actual pattern of soundwaves coming down to the eardrum. Non-minimum phase reflections and a quite irregular, personal amplitude response is the result. If a headphone is to mimic listening to speakers, then the pinna is cleary in the game. The saying is, that angled cups are benefitial for instance. I can only wonder, why IEMs became that successful--me included ;-)

 

[swamps]

Is 'smear' a word in context of sound?

Not really that used but at least it usually means the same thing. You probably won't get it but for me it's like the old FXAA antialiasing on Crysis, it looks like "vaseline on screen" but the screen is now blurry sound.

The equivalence between the diffuse soundfield and the headphones response, both measured at the eardrum, is meant as amplitude frequency response.

I think this is the biggest limitation of headphone measurements, they cannot show why an on-ear like the Sony ZX110 won't sound as good as a HD800S after EQ.
An obvious start is the HD800S makes a lot more literally air volume (not SPL).

I'd like to test someday with impulcifier how much the headphone's sound propagation really matters when normalized to my HRTF and then measure whatever I can afterwards but alas, I lack good speakers, a good room, and a decent measuring rig.

@Heinrich this thread may interest you, there are all sorts of fun stuff like adding delay to match reflections and much more tbh.

Replicate imagined speaker sound to headphones. (With Impulcifer)

Hi. I live in South Korea, and I enjoy exploring and studying various audio information from all over the world. I use a translator. I'm sorry the sentence wasn't smooth. I used the speaker set up in the in-room and was quite satisfied. But I had no choice but to dispose of the speaker because...

www.audiosciencereview.com

 

[Lion♡]

'd like to test someday with impulcifier how much the headphone's sound propagation really matters when normalized to my HRTF and then measure

The thread link below is mine, and I appreciate your interest in it.
If I may share a personal opinion, I believe that each headphone/IEM has its own unique characteristics (or flaws), and in most cases, they work reasonably well.
But somehow... headphones feel somewhat unstable. Since the pinna is included and real-ear measurements are possible, externalization is quite easy. However, if we assume an in-room environment, it feels like a very weak reflection is applied overall in the BRIR setting.
On the other hand, IEMs can sometimes be difficult to externalize, depending on the situation. However, most IEMs perform well in their role as "playback devices."
So now, I only use Apple EarPods and Custom IEMs.
However, even if certain characteristics (or flaws) are present, whether they have a significant impact depends on the listener and the specific headphones. Some may notice them clearly, while others may not.

 

[Heinrich]

... "vaseline on screen" but the screen is now blurry sound.

So now, I only use Apple EarPods and Custom IEMs.

Ja, the term is clear, and is not. It directs the imagination to a cause, which I think is not appropriate. There is a deeper problem with headphones, especially with IEMs.

Above mentioned Günther Theile argues that the amplitude, when listening to headphones, at the eardrum should resemble (closely?) a diffuse field. The reasoning behind this is pretty simple.

The HRTF depends strongly on the incident angle of the sound. See the net for "HRTF HATS". Therefore would the hearing conclude from any specific sonic signature the direction of the sound. Conversely, to shape the amplitude response of the headphone to HRTF in diffuse field would take away directional cues, as "diffuse" is inherently free of "direction".
To avoid directional cues when chosing the tuning of a headphone is essential, as with stereo it should not only represent sound coming from one direction, but many in between left and right.

This argument is flawed in some aspects, me thinks. Only for an example, to conclude direction from a specific HRTF needs the sound to be known, how it sounds differently from various directions. This is rarely the case, especially with stereo, because the sound source doesn't--move.

On the original question, where the 'grass' in the group delay comes from when measiring on a G.R.A.S or other HATS, I said it may be, but only in parts, the interaction of the sound field with the pinna. It is natural so far, and should be appreciated. The success of IEMs rounds out the mystery.

ps: the 'preference' approach of Harman, equalize until you like it, is grand indeed.

 

[Lion♡]

The success of IEMs rounds out the mystery.

I'm not sure if you were replying to me or to Swamps, but my IEM experience is based on the fact that I use personalized HRIR/BRIR.
As you know, HRIR and BRIR naturally include not only direct sounds but also countless reflections from various angles, each incorporating my personal HRTFs. Additionally, HPCF is applied to properly equalize the IEMs/headphones so that these elements can be heard correctly.
However, since this isn't directly related to the thread's main topic, I won’t go further into discussions about diffuse fields or directionality. I just happened to come across this thread, and when I saw my thread link attached, I was happy to join in. But I don’t want to hijack the discussion, so I’ll refrain from further participation.

 

[Heinrich]

But I don’t want to hijack the discussion, so I’ll refrain from further participation.

One other example: in speaker stereo a source may be paned to full left. Its spectrum is kept as is, let's assume that the HRTFs of 0° front and 30° from the side do not difffer that much. Such a position of 30° left in speaker stereo would translate over headphones to 180° left.
But
- the sound from full left has the wrong HRTF signature
- the sound on the right ear is missing entirely
- if there was sound on the other ear it should have a significant signature from the HRTF

I'm going to look into your approach, @Lion

@mike7877, me to, that's all not about the group delay and reflections etc., sorry, but reflections are expected, as explicated above But how useful the interaction with the pinna, or better to say the HRTF is, I do not know ...