A few members and Abyss Headphones pinged me to let me know about this discussion, asking me if we had measured the ABYSS DIANA V2. At the time we had not. I looked at the measurements at AudioScienceReview that I was being asked about, and thought what I saw (the measurements, the conclusions from those measurements, and the subsequent discussion there) interesting enough that we should measure it, too.
Let's start by looking at AudioScienceReview's DIANA V2 frequency response measurement:
Fig.1 (above): AudioScienceReview's frequency response measurement of the ABYSS Headphones DIANA V2.
I agree with Amir at AudioScienceReview that this is an interesting, challenging headphone to measure. One look at the DIANA V2 (and its earpads) suggests that the way this headphone wears and couples is a little unconventional. We ran into another unconventional headphone recently (but unconventional in a different way than this one)
when we measured the Apple AirPods Max. As with the AirPods Max, when you're dealing with an unusual headphone, you sometimes have to make some special considerations about how you go about measuring it -- the DIANA V2 is certainly one of those headphones.
I think I should start by making clear that while I can assure you there were different methods at work here (which I'll explain shortly), there is also the matter of the difference in our primary headphone measurement fixtures. AudioScienceReview used a GRAS 45CA (with GRAS RA0402 ear simulators) and we used the Brüel & Kjær 5128. We also have the same GRAS 45CA setup as AudioScienceReview and have measured with it for six years, so I am very familiar with it.
Whereas 711 simulators (like in the 45CA) were established 40 years ago primarily to measure things like hearing aids and telecom devices, the 5128 is a much more recent fixture (coming from 12 years of research and development) that was intended to measure everything from hearing aids to earphones to headphones like this, and of course many other things. Crucially, the Brüel & Kjær 5128 simulates human hearing from 20 Hz to 20 kHz, versus the 711's simulation range of only 100 Hz to 10 kHz.
Also important to note is that there's a reference plane on a 711 simulator where the device under test is intended to be positioned to ensure the input/transfer impedance is accurate [1]. (I'm referring to the input/transfer impedance of average human ear drums and the input/transfer impedance of an average human ear canal.) With a 711 simulator, moving away from the reference plane can result in a narrower simulation bandwidth than its specified 100 Hz to 10 kHz range. With the 5128, the average human impedance is equally well represented at any plane within it (and, again, from 20 Hz to 20 kHz) -- that is, there is no such measurement reference plane with it.
NOTE: If you're interested in learning more about the Brüel & Kjær 5128, please watch
my presentation about the 5128 at the HBK (Hottinger Brüel & Kjær) Product Physics Conference. It's only 20 minutes long, but there is a lot of information in it that will help you understand some of the key developments and differences.
I'm unclear about some of the things I'm seeing in AudioScienceReview's measurement shown in Fig.1 above. The response > 5 kHz looks more like a rather random series of very narrow peaks/troughs that we might be able to eyeball an average through, but I think doing so might be a bit tortured.
In Fig.2 below, you can see the second of two of our measurement seatings of the DIANA V2 on the Brüel & Kjær 5128. While the level for the measurement in Fig.2 was set at 90 dBSPL @ 1 kHz and AudioScienceReview's was set at 94 dBSPL or 1 Pa at 425 Hz, for the sake of
this comparison that level difference does not effect the shape of the frequency response.
Fig.2 (above): ABYSS Headphones DIANA V2 frequency response as measured on the Brüel & Kjær 5128. (This is one of two seatings we had completed at the time of this post.)
Just as with AudioScienceReview's measurement in Fig.1, our measurement in Fig.2 is completely unsmoothed. You can see above 2 kHz there are substantial differences between our measurement and AudioScienceReview's measurement. I think it is very noteworthy in our measurement in Fig.2 that there is far less random hash > 5 kHz -- instead, there is a clear indication of an actual response. Again, the Brüel & Kjær 5128 simulates average adult human hearing from 20 Hz to 20 kHz, and the average human impedance is equally well represented at any measurement plane within the system.
In Fig.3 below, you can see the unsmoothed frequency responses from the two seatings we completed at the time of this post along with the smoothed average (1/12 octave) of these two seatings.
Fig.3 (above): ABYSS Headphones DIANA V2 frequency response as measured on the Brüel & Kjær 5128, showing two seatings unsmoothed, and the average of those two seatings smoothed (1/12 octave).
Even with two separate seatings (representing two different headband sizing settings and in different positions), averaging through even the highest frequencies -- all the way to 20 kHz -- does not require the creativity that would be required averaging through that same range in the measurement in Fig.1.
Also noteworthy in comparing these measurements is the symmetry between channels in both seatings in Fig.3. Again, placement with this headphone is temperamental, but we can still achieve reasonable symmetry with it using a placement technique that I presented in a video on these forums and at the 2018 ALMA annual conference (AISE) that I have since refined.
Even then, additional care was required with the DIANA V2 as tiny movements over the fixture with this headphone could break seal and radically alter the bass response in one or both channels. It's also very sensitive to movement above 1 kHz, but the bass was the tougher issue for us with placement. For the DIANA V2, we actually used the aforementioned placement technique to line the channels up above 1 kHz, then switched between 20 Hz and 30 Hz sine stimuli and used very small adjustments to get those as close as we could to being within 1 dB of each other. Once we were able to do that, we ran the sweeps for that seating.
NOTE: While not every headphone is matched enough to measure symmetrically (left-right), we have found that premium headphones are more likely to. We always give the benefit of the doubt at the outset of measuring a given set of headphones, assuming that symmetry is possible, and our placement method generally works very well in helping us to determine that quickly.
Again, though, looking at the AudioScienceReview measurement in Fig.1 -- especially the lack of channel symmetry and the relative hash above 5 kHz -- I was curious about why this was happening, and thought maybe the DIANA V2 somehow interacted with 711 simulators poorly at higher frequencies. Since we have the same fixture as that used for the measurement in Fig.1 we decided to have a go at a couple of quick DIANA V2 seatings on it. Again, using real-time frequency response monitoring whilst placing the headphone on the GRAS 45CA fixture (and then micro-adjusting at 20 Hz and 30 Hz), we got the two seatings you'll see below in Fig.4 and Fig.5:
Fig.4 (above): ABYSS Headphones DIANA V2 frequency response as measured on the GRAS 45CA (RA0401), seat 1.
Fig.5 (above): ABYSS Headphones DIANA V2 frequency response as measured on the GRAS 45CA (RA0401), seat 2.
As you can see in Fig.4, it appears we may not have had a good seating, but at least it was reasonably symmetrical left-right. In Fig.5 we had a better seating, and it was still reasonably symmetrical. Of particular interest to me, though, was the response in both of these measurements > 5 kHz compared to the measurement in Fig.1. As you can see, in both measurements (Fig.4 and Fig.5), the response above 5 kHz had more symmetry and was less hashy than AudioScienceReview's measurement in Fig.1. Also, even though we might reject the measurement in Fig.4 before averaging, it's still more symmetrical throughout and there's certainly less hodgepodge in the treble range when compared to Fig.1.
After testing on our GRAS 45CA (identical in configuration to AudioScienceReview's), I suspect that the results we are seeing in the measurement in Fig.1 were likely due to issues of placement of the headphone on the measurement fixture.
When it comes to AudioScienceReview's total harmonic distortion (THD) measurements, we also had different results. In Fig.6 and Fig.7 below are AudioScienceReview's DIANA V2 THD measurements:
Fig.6 (above): AudioScienceReview's total harmonic distortion (THD) measurement of the ABYSS Headphones DIANA V2 at 94 dBSPL, 104 dBSPL, and 114 dBSPL (425 Hz).
Our distortion measurements were lower at 94 dBSPL and 104 dBSPL, which I will show shortly. While choosing to show the Y-axis with a linear (versus logarithmic) scale is the judgment call of the measurer, I tend to prefer (and so use) a logarithmic Y-axis, as I find it easier to read and interpret. Again, that's a judgment call, and perhaps some prefer the linear Y-axis.
What puzzles me, though, is why AudioScienceReview is showing THD at 114 dBSPL. To put 114 decibels in perspective, I believe that's somewhere between the sound pressure level of a rock concert and an airplane taking off. 114 decibels is louder than some jackhammers. At 109 decibels, hearing impairment occurs after less than two minutes. I am unclear about why this level is being used to judge a headphone. In my opinion, looking to see how much distortion one gets from a headphone at 114 dBSPL is purely academic, and that's fine. However, I believe that should then be more of a sidebar item, not mixed in with more realistic, usable levels and then used to skew the appearance of a headphone's distortion characteristics.
So why was 114 decibels used? I think it was included for dramatic effect. After showing the measurement in Fig.6, AudioScienceReview's distortion assessment started with "Ouch!"
That THD measurement at 114 dBSPL was also used to create the conclusion that "Treble distortion shifts lower with increasing level" (see Fig.7 below). Is that a reasonable conclusion, though? I don't think so in this case. I'll get back to this shortly.
Fig.7 (above): AudioScienceReview's total harmonic distortion (THD) measurement of the ABYSS Headphones DIANA V2 at 94 dBSPL, 104 dBSPL, and 114 dBSPL (425 Hz).
Below are our THD measurements of the DIANA V2:
Fig.8 (above): ABYSS Headphones DIANA V2 total harmonic distortion (THD) at 90 dBSPL (1 kHz) as measured on the Brüel & Kjær 5128.
Fig.9 (above): ABYSS Headphones DIANA V2 total harmonic distortion (THD) at 94 dBSPL (425 Hz) as measured on the Brüel & Kjær 5128.
Fig.10 (above): ABYSS Headphones DIANA V2 total harmonic distortion (THD) at 104 dBSPL (425 Hz) as measured on the Brüel & Kjær 5128.
NOTE: We did not test at 114 dBSPL, as, in my opinion, there is simply no good reason to. The DIANA V2 is actually used by us at the Head-Fi office regularly, so I decided to keep the tests reasonable -- and useful...and fair.
At 94 dBSPL (425 Hz) and 104 dBSPL (425 Hz) -- which are two of the three levels measured by AudioScienceReview -- our measured THD was lower. At 94 dBSL, AudioScienceReview's measured THD at 20 Hz was < 3% (Fig.7), and we measured < 1% at 20 Hz (Fig.9). At 104 dBSPL, AudioScienceReview's measured THD at 20 Hz (Fig.7) was around 8%, and we measured > 3% (Fig.10).
NOTE: On the GRAS 45CA (RA0401) we measured at both 90 dBSPL (1 kHz) and 94 dBSPL (425 Hz), and the measured THD levels at 20 Hz were nearly identical to our Brüel & Kjær 5128 measurements (but there were slight differences through the midrange and treble).
I suspect our measured THD may be lower than AudioScienceReview's for a couple of reasons: First, we use a Herzan acoustic and vibration isolation enclosure. Given the physics involved, our Herzan enclosure is not entirely effective at lower frequencies, but it does help. Second, our measured frequency response had quite a bit more energy approaching 20 Hz. At a level of 94 dBSPL (425 Hz), both the left and right channels were at around 87.5 dBSPL at 20 Hz in our measurements, which is around 10 dBSPL higher than one of the channels in AudioScienceReview's measurement in Fig.1 and around 5 dBSPL higher than the other. Since both AudioScienceReview and we are expressing THD as a ratio, this should lead to a lower measured percentage in our measurement given our higher measured output there.
As for AudioScienceReview's conclusion that "Treble distortion shifts lower with increasing level," we have come to a different conclusion. We measured 104 dBSPL on only one of the two seatings, so isolating the THD for that seating at 90 dBSPL (1 kHz), 94 dBSPL (425 Hz), and 104 dBSPL (425 Hz), we measured the THD in Fig.11 below:
Fig.11 (above): ABYSS Headphones DIANA V2 total harmonic distortion (THD) at 90 dBSPL (1 kHz), 94 dBSPL (425 Hz), and 104 dBSPL (425 Hz) as measured on the Brüel & Kjær 5128.
As you can see in Fig.11, from 90 dBSPL to 104 dBSPL, treble distortion does not shift lower with increasing level.
At some point in the future we will likely add two or three more measurement seatings to better round out our DIANA V2 measurements.
We've been measuring headphones for six years at Head-Fi HQ, having performed
thousands of them over that time. Just today, we were at
Head Acoustics' US headquarters for the third social-distanced day of measuring ANC headphones in the past couple of months. We discuss headphone measurements with headphone engineers and product teams almost daily, routinely sharing and discussing our results. At any given time at Head-Fi HQ, we have a number of prototypes and pre-production headphones that we are asked to both measure and evaluate subjectively. And still, six years later, we still feel like headphone measurement newbies here every time a headphone like the Apple AirPods Max or ABYSS DIANA V2 challenges our measurement routines.
I have not looked at all of AudioScienceReview's headphone measurements, but certainly with this one I think AudioScienceReview should consider re-measuring it and working on better and more careful placement of the headphone on their measurement fixture. Again, we would like to improve our DIANA V2 measurements by adding a two or three more seatings ourselves. I do strongly agree, though, that the ABYSS Headphones DIANA V2 is more challenging than most headphones to measure.
Unless otherwise noted, the measurements in this post were made with:
References:
1. Johansen B, Jønsson S, inventors; Brüel & Kjær Sound & Vibration Measurement A/S, assignee.
Human like ear simulator. United States Patent Application US 2015/0124978 A1. United States Patent and Trademark Office. 7 May 2015.
