Understanding the Headphones.com IEM Measurements

Hey y’all! We’ve had a lot of people ask for an explanation of how we display IEM measurements, specifically questions like:

  • “What is ‘JM-1’ and why should I care about it?”
  • “Why wouldn’t you use the Diffuse Field HRTF of the rig itself—like the B&K 5128 or GRAS 45CA—when measuring IEMs?”

Well, that’s what I’m here to answer! Think about this as the “IEM-centric” companion post to the DF HRTFs used at Headphones.com thread elsewhere on this same forum.

Before continuing, I highly recommend readers take a look at my article about Diffuse Field to familiarize themselves with what Diffuse Field is and why we’re using it. It might also be good to read my article The Shape of IEMs to Come to familiarize yourself with the benefits of measuring using the 5128. @Resolve and I also worked on a video with a cheeky title that goes over a lot of what I’ll cover here.

To explain JM-1, I first need to explain the “population average” DF HRTF that you may have seen @Resolve talk about in our livestreams or video content, so let’s start there

The “population average” DF HRTF

The “population average” Diffuse Field HRTF we use comes from Determination of noise immission from sound sources close to the ears (Hammershøi & Møller, 2008), which serves as the basis for the ISO 11904 standard, which is the internationally-recognized standard for measuring sound devices close to the ear, consisting of ISO 11904:1 and ISO 11904:2.

For now lets focus on ISO 11904:1, which is the standard that contains our “full population average” DF HRTF.

When I say “full population average,” what I actually mean is that the ISO 11904:1 DF is a Diffuse Field HRTF measured at the eardrum of actual humans. The underlying data for ISO 11904:1 comes from two/three different sources, shown below.

Fig. 1

After aggregating the above data, Hammershøi and Møller provided a “literature average” of this data (along with a ±1 Std. Dev. from the mean), shown below highlighted in red.

Fig. 2

This is, as far as I’m aware, the best basis we have for a full human Diffuse Field HRTF in the literature. Thankfully, Hammershøi and Møller in their infinite wisdom and loveliness also provided us with a “literature average” for Diffuse Field HRTFs measured at the blocked canal, shown in highlighted blue above. This is important because it allows us to characterize which parts of the HRTF are due to which anatomical feature.

For example, if we have the full DF HRTF from the ear drum X, and the blocked canal DF HRTF of everything outside of the ear canal Y, we can do X - Y = Z where Z is the effect of the ear canal. Doing this results in the measurements below.

Fig. 3

Black is the full ISO 11904:1 human DF HRTF measured at the eardrum
Blue is the DF HRTF measured at the blocked canal
Red is the transfer function of the ear canal of the "average" human, derived from the difference between the above data points

Okay… but what does this have to do with the 5128 or IEMs?

Well, the issue seems to be that the DF HRTF of the measurement rig we use for IEMs (the Brüel and Kjær Type 5128) is meaningfully brighter than the population average DF HRTF, which means if we use this as our baseline, something that measures “well” to Mr. 5128 might actually be heard as brighter than what an average human may expect as “neutral” at their eardrum.

Fig. 4


Black is the ISO 11904:1 DF HRTF
Green is the B&K 5128 DF HRTF

We can see that above 3-4 kHz, the 5128’s DF HRTF is significantly brighter than the “average” human. But why?

Fig. 5

A sampling of 9 headphones measured at the blocked canal of the B&K 5128 (green) have significantly more treble than the same headphones measured at the same place on a group of 15 humans (blue), shown above. Credit to Sean Olive.

The same headphone and listener dataset as above, but measured on the GRAS 45CA with KB5001 pinnae in green and humans in blue

Because these are headphone measurements at the blocked canal, they do not include the effects of any anatomical factors except for the outer ear. Thus we are solely seeing the differences between the outer ears of the measurement fixtures we use and the outer ears of humans. And since it’s not just 5128 that has this issue, we’re using a version of this target for IEMs measured on 711 couplers as well (just with the 711 coupler’s error under 1 kHz accounted for)

This definitely poses a problem… and not just for headphones.

The Hypothesis

Okay, so we need a DF HRTF for IEMs, but which should we use?

The 5128’s is probably too bright, but any unique characteristics of the 5128’s DF HRTF that can be attributed to its outer ear rather than its ear canal are irrelevant for an IEM target, because IEMs do not interact with the outer ear.

We know we can separate the DF HRTF as measured at the eardrum into the contribution from the outer ear* and the contribution from the ear canal, as shown in Fig. 3 posted again below.

Fig. 3 (again)

We can see from the above that different frequency regions have differing levels of contribution from the outer ear and ear canal. For example: at 5 kHz, about 90% of the total eardrum DF response comes from the outer ear (blue trace above)

Fig. 6

@Joel Merrifield used this process to produce a weighted average DF where, in the regions dominated by the outer ear, he correspondingly shifted the weight towards a generic human average DF, while shifting the weight towards using the 5128’s DF in regions where the ear canal makes the more significant contribution. The result is the orange trace below, which has henceforth been dubbed the “JM-1” baseline.

Fig. 7

You can see in the area where the canal contributes more to the DF HRTF, the result looks more like the 5128’s DF HRTF in green. And vice versa, where the outer ear contributes more, the JM-1 baseline begins to look much more like the ISO 11904:1 DF HRTF.

This is the DF HRTF we are currently using for evaluation of IEM measurements at Headphones.com.

We calibrate IEM measurements using this baseline to display them against our preference bounds derived from the existing speaker and headphone literature from Harman.
Examples shown below—first calibrated, then raw.

Fig. 8

z
Fig. 9

What about 711?

For 711 measurements of IEMs, we are using a version of the same baseline above that incorporates the average difference between 711 and 5128 measurements of IEMs. This is because the 711 coupler is a less accurate system, and using this difference to alter the preference bounds for 711 can improve the consistency and accuracy of our analysis of these measurements.

Below is the difference (red) between the Hisenior Mega5EST measured on the 711 coupler (teal) and on the B&K 5128 (purple).

Fig. 10

When you measure many IEMs between the two systems, the average difference looks something like the dotted blue curve below.

Fig. 11

The sections < 100 Hz and > 4 kHz are greyed out because we cannot have confidence in the “average” difference across IEMs being representative of what you will see with any given IEM.

For example, the average in blue incorporates measurements of IEMs that use both dynamic drivers as well as balanced armatures for bass, and the latter tend to “lose” more bass when moved from 711 to 5128.

That’s why the Mega5EST (red), which uses dynamic drivers in the bass, loses less bass than the average (blue) when moved to the 5128.

The differences in the treble are due to a few factors, including length of the air volume between eardrum and IEM. This factor isn’t going to be consistent between IEMs or even between individual seatings of the same IEM, and it’s also one area where we know the 5128’s ear canal is longer than the 711 coupler. That’s why the treble peaks in the 711 measurement in Fig. 10 (teal) shift downward in frequency when the IEM is measured on 5128 (purple).

Fig. 12

Fig. 12 shows the 4620 + Human Diffuse Field HRTF baseline used for Headphones.com’s IEM measurements in pink, and the 711 version of the same HRTF in light blue.

The future

While this is a baseline that has a fair smidge of theoretical underpinning, it is still just a mathematical derivation based on other available data. In short, it’s not The Real Deal, and we would definitely want The Real Deal in order to judge IEMs as fairly as possible.

(We’d also like to make sure of just how close our $50,000 ear simulator is to the actual humans it purports to simulate)

So, we’ve recently been working on measuring the canal transfer function of the 5128 in order to see just how close Joel’s initial calculations were, and how viable the idea of using the 5128’s canal with the “population average” blocked canal data to get an idealized HRTF for IEM evaluation is.

The measurements of the 5128 canal are an ongoing project and preliminary results have confirmed that the JM-1 hypothesis is more or less correct. Once we have a result we feel 100% confident in, this will be shared with the community. But we’ll be continuing to use JM-1, as it has become a widely used reference in the community.

For now, this post just serves as an explanation of What The Hell JM-1 Is And Why We Use It, as well as a homebase for discussion surrounding that.

We at Headphones.com just wanted to make sure there was an easy resource for this information out there so that its easier for people to get hip to what we’re doing here.

Thanks for reading! I’ll keep this post updated as changes are made.

7 Likes

Hi all, this is my first post in the forum here, but I’m the “JM” in JM-1. I’ll keep an eye on this thread in case people have questions, but @listen_r is more than qualified to answer them as well!

8 Likes

My sources have told me that this man is not a trusted user.

/s
Glad to finally have you here, Joel!

3 Likes

Thanks Joel and Griffin ! First post here as well.

6 Likes

Ahhhh so glad to have you here, Maya! Truly a huge get for this forum :grin:

3 Likes

Thank you very much!

I’m ashamed to admit it, but while I’m amazed with the work done by Joel Merrifield with this composite compensation target, I still don’t quite get it why just using the ISO 11904:1 as the reference for IEM evaluation would be inadequate (apart from the modest sample size it’s based upon)?

Or rather what issues with that approach JM-1 is aimed towards alleviating.

It feels like this should be rather obvious, yet I fail to grasp it due to a lack of knowledge in the field and somewhat compromised cognitive capacity.

Fwiw, once we’ve finished verifying our testing of the canal transfer function of the 5128, we very well might just use ISO 11904:1

Our aim in doing this is mostly to characterize the contributions of the ear itself to the HRTF, but its entirely possible that the contribution of the ear to the HRTF and the contribution of the ear to IEM measurements aren’t correlated, in which case using ISO 11904:1 makes a good amount of sense. Hope that answers your question :slightly_smiling_face:

1 Like

You do mean the outer ear, right? Intuitively it would seem that they wouldn’t be correlated and personally I’d be quite surprised if turns out that they are. Due to the character of sound source localization in case of IEMs and everything… If you mean something analogous to HpTF for circumaural headphones, where head geometry influences the propagation of soundwaves by impacting acoustic coupling/leakage, inner reflectivity of the headphone, angle of incidence for the transducer etc; but not in the same exact way it would do in case of different sound source localization like with speakers in the room for example. Due to the distance at least and as such the amount of expected reflections, air volume/acoustic impedance among other things(?)… Outer ear geometry may contribute to most, if not all(?) the same factors as well, but in somewhat different way and magnitude/degree with IEMs. Ugh, Im agraid I’m struggling to express what little I seem to understand even, so I’ll cut this short :sweat_smile:

That’s a very interesting issue. I wonder what would be the way to find out.

Please, excuse me this little rant, it may be painfully uneducated, exacerbated by the language skills that are sadly inadequate for the task/topic at hand.

I’m looking forward to your future findings eagerly.

Huge thank you all for your incredible work, and thank you Listener for your patience and eagerness to help!

Actually no, I was referring to the ear canal/eardrum!

We are close to knowing how exactly the ear canal of the 5128 contributes to the full DF HRTF, but we’re not quite certain that the 5128’s ear canal would be contributing the same transfer function to the measurements of IEMs themselves.

In short, we don’t yet know if this transfer function when using 5128 to measure speakers is equally applicable/useful for how IEMs behave in the canal itself. Initial measurements indicate poor agreement between (the difference between 5128 and 711’s canal transfer functions) and (the difference between IEMs measured on 5128 and 711), but more testing is definitely needed.