On Group Delay and Spatial Effects

[Presently42]

Er, actually, this thread should probably be in the psychoacoustics area. Can a mod move it?

Introduction and observation:
Following the fairly well received post on the relationship between group delay and spatial effects in headphones, I thought I'd start a new thread on the matter, gathering further evidence, eliciting further thoughts and opinions, and acting eventually as a repository for the results of armchair research, that I hope to conduct in an effort to prove or disprove the hypothesis, that

Hypothesis:
Group delay (as well as frequency response; particularly around 8 kHz and possibly around 2 kHz - aka, the Blauert bands) is responsible for much of the out-of-head soundstage and instrument separation phenomena.

Definition:
Let's define soundstage to be sound emanating over the sum of the two angles formed between the nose and the ears; depth, the apparent distance from the centre of the head to the location of the sound; and imaging, the location of various sounds in the soundstage.

Examples:

1. two sounds coming only from a single point in the middle of the head indicates a soundstage, depth and image of zero
2. sounds going from ear to ear with each source of sound a pinpoint in the space close to the head indicates a soundstage of 180 degrees (I refuse to use radians), a small depth and high imaging.
3. Sounds surrounding the head coming from afar and seemingly diffusely indicates a soundstage of 180 degrees, large depth and low imaging

Data:
Let's begin with some data and comments gathered by our host, Amir. Note, that the list is as yet incomplete: I'll be updating it more tomorrow and in the days to come. For now, there's no especially clear image - except this: headphones with good soundstage seem to have clean group delay graphs, with some messiness around the 2 kHz region, and a spike around 9 kHz. One pattern that is emerging is, that too messy a graph indicates too many confused reflections, lessening the spatial effect - whilst too clean a graph indicates too few reflections and a non-existent spatial effect. I hope to further refine these findings as I gather data and am able to form better analyses on them.

Maker - Model - Group Delay Graph - Amir's related comments
Beyerdynamic DT990 rather good spatial qualities
Beyerdynamic T1 v2 [No comment given.]
Bower & Wilkins P5 [No comment given.]
Focal Celestee soundstage is mostly whacked in the middle of your head
Focal Clear [No comment given.]
Focal (Drop) Elex [No comment given.]
Focal Utopia a halo of sound about 25 to 30% outside of your head with very nice instrument separation
HEDD Headdphone maybe a bit better than a typical headphone with its center of your head response
Hifiman Ananda speakers away from your ear [...] It nicely pushes the sound out some and leaves room for instruments to image between it and the ear.
Hifiman HE-6 Spatial effects were reasonable
Philips Fidelio X2HR Shaving off those resonances also collapses the soundstage by some 20 to 30%. [...] increasing the strength of the [high] shelving filter [...] cost even more in soundstage. [... It has] good spatial effects
PSB M4U 1 massive spikes in high frequencies indicates reflections inside the cup [...] The [eq] boost between 1 and 5 kHz opened the soundstage and detail/instrument layering.
Sennheiser HD650 Where is not so good is the spatial qualities. There is a nice open circle that the sound sources emanate from. It starts from your ears and circles forward. There is no sensation of space beyond that. Nor that nice layering [...] I always get a feeling of the sound closing
Sennheiser HD800S What was remarkable and uncanny was separation of instruments. It was as if this headphone would take every element in the music, pull it apart, and then position it in different spatial locations in a 6 inch space around each ear. I wouldn't call it "soundstage" as much was it was this fun and captivating effect.
Sony MDR-7506 little soundstage [...] dull, middle of the head image [...] boring spatial effect.
Superlux HD668B The large drivers provide good bit of spatial information (separation of instruments), creating a soundstage that dangles from your ears down and to the back with the mono content behind your forehead.

 

[Sunrise]

I wonder why no one has any comments. I think the relationship between listening experience and group delay is significant.

 

[Presently42]

As do I - but I no longer am of the opinion, that this is a productive way of predicting the spatial effects of headphones: quite simply, there are too many headphones with decent spatial effects, which cannot be accounted for by the group delay graph. Pity! But there seems to be something more going on.

Whilst it's certainly possibly, that group delay in specific regions increases certain types of spatial effects, more likely is, that a number of causes are extant1. My own small experiments have had rather poor outcomes, and I confess, that I've moved on from trying to figure out exactly what's going on. Have you any thoughts on the topic?

1: I'd guess at the angle of the driver; the distance between the driver and the ear; and the make and material of the earcup. Whilst the former two can (and maybe should!) be documented, that a relationship be found between their values and the combined perceived spatial effects values, the third is still very opaque, and requires more research: for example, do materials of different sonic reflective values produce different spatial effects? Quite honestly, there's more research to be done here, than I'm able to do - especially given, that I've no budget to acquire and test many pairs of headphones! I was hoping the group delay graph would suffice - but it seems to not, sadly.

Edit: There might be something to be gleaned from @Floyd Toole 's book. I should have a look....

 

[Sunrise]

그룹딜레이(GD)의 중요성과 FR의 완결성 - 영디비

완성된 글은 아니지만 회원님과 대화가 오간다면 괜찮지 않을까 해서 올려봅니다. 본문은 저의 청취 경험에 전적으로 기반한 내용이라 틀린 내용이 있을 수 있습니다. ASR은 헤드폰에서 그룹딜레이(이하 GD) 측정치를...

www.0db.co.kr

Based on my personal experience, I wrote this article. You can use google translation.

 

[Presently42]

Hmm, I'm afraid, that googly translate doesn't handle Korean terribly well; and what Korean I did learn is certainly insufficient for these purposes! With that in mind, I didn't see any correlation made between group delay and spatial effects, other than perhaps what I've mentioned above. I also am not sure, that your article distinguishes between causality and correlation: whilst you nicely noted certain kinks present in both the group delay and frequency response of a given headphone, perhaps due to the translation, I was unsure, if the one caused the other - or were merely present in both. Let's recall, that resonances are visible in many graphs: they appear in frequency response graphs as peaks; in spinoramas as peaks in all curves; in phase and impedance plots as kinks; and in group delay graphs - though here, their traits are a bit harder to detect, seemingly.

In any event, I don't think, that spatial effects have much of anything to do with resonances. Rather, I suspect, that spatial effects have to do with reflections at particular frequencies; and that these frequencies determine the type of spatial effect. Unfortunately, I've heretofore been unable to create good experiments to prove or disprove this: my attempts at doing so via the group delay graphs has mostly failed, as there are too many outliers. Perhaps in your article, you discovered something on this? As I said, the googly translate really wasn't up to the task.

In another thread, @Floyd Toole discusses a technique for mimicking various venues in a single room with a few speakers. The technique, seemingly, uses modifications in both the frequency and time domains: certain frequencies are produced more than once (aka, a reflection). If we assume, that headphones are speakers (other than their positioning), then we can also assume, that this mimicry can be achieved through headphones - something we already know is true, from the Smyth virtualiser and other such techniques.

What I'd hoped to achieve (and still hope to!!!), was a way to predict the spatial effects of headphones (and maybe speakers - though that seems to be mostly directivity). I'd noticed a correlation between spatial effects and group delay. I'd hypothesised, that group delay might indirectly indicate reflections: if a set of frequencies has a large group delay, the these frequencies might also be being reflected many times in the volume of space between the head and the earcup - and that these reflections (time differences) were perceived as spatial effects. I also hypothesised, that the greater the amount of energy at certain frequencies, the greater the effect. I'm not sure I'm wrong - but I'm fairly sure the group delay graph isn't indicative of these reflections.

Given the importance of spatial effects in speakers (envelopment et al.), I'm rather surprised, that they're not more discussed with headphones: we already know, that the stereo effect is so highly addictive as to mask the flaws of an otherwise mediocre speaker. I'd propose, that one reason the HD800 are so widely praised, is due to their superb spatial effects. Indeed, I'd go so far as to say, that spatial effects are nigh as important in headphones as frequency response. Perhaps @Sean Olive has done some research on this topic?

Anyway, I'm glad the post is finally getting a bit of attention: the more mind think on the topic, the more likely, a good result will be found!

 

[Dentin]

The 9 kHz group delay spike is caused by some sounds bouncing off the pinna before entering the ear canal, which causes phase cancellation and creates a pinna notch in the upper treble response. Rtings already measured how pinna interaction changes with speaker positioning and found that positioning them at 30° also led to the formation of such a notch, correlating with staging depth. I'd guess the rest of the messiness also corresponds to outer ear interaction at those frequencies, which may or may not enhance imaging based on how accurately it emulates outer ear activation caused by speakers.

FWIW, I figured that since most headphones are effectively minimum-phase transducers, I'd be able to emulate open-back stereo imaging purely by changing frequency response. EQ'ing by ear, I PEQ'ed down the 7.5-10 kHz region, though I needed to use multiple peak filters and be really precise with the treble to avoid imaging incoherency. I find it does the job extremely well if I'm careful enough with filter placement. I also EQ'd down 2 kHz for staging width. I tried centering my peak filters at 3 kHz at first but found that it made the timbre unnatural.

e: So this is what the EQ preset I'm using right now looks like (dots mark 100/1,000/10,000 Hz):

e2: I'm not sure why, but the notch also sounds like it's responsible for what audiophiles call sub-bass slam. If it's not deep enough or the sides aren't elevated enough, I get textureless or plasticky bass. I notice that many headphones that audiophiles claim are better able to take bass boost "without distorting" have a deep notch in GRAS measurements.

 

[Jens]

I am convinced that the group delay mess from 1 - 2kHz to around 5 kHz
band that looks solid on the group delay diagram is caused by the thin
foam (and the nature of it) “protection” on the acoustic driver that is inside the headphones.

Therefore, it should be simple to listen to the headphones with and
without this band of group delay by removing the foam and listening.

I believe that this group delay band indicates dramatic changes in
the phases of the sound. These phase changes move some of the sounds in
space. If I assume that our ability to observe/hear phase change is the
same in free listening as it is with headphones, then it is only the
phase mess below 1500 Hz that will change our sound in space perception.

Hypothesis: the more messy the group delay is below 1500 Hz, the more
“space sound” is perceived.

Jens