ETA ESX-900 Measurement, Review, and Discussion

Discussion in 'Headphone Measurements' started by Vtory, May 6, 2021.

  1. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,847
    Trophy Points:
    113
    Location:
    East Coast
    ETA ESX-900 Review and Measurement

    IMG_7279.jpg


    Opening, Disclaimer, Setting Up


    First off, I’d thank Evan and ETA friends for adding me to the loaner tour of ESX900 (along with Genesis brothers as well). Was extremely excited about this opportunity largely because (1) I heard too many good things for Evan’s works since he messed with HFM but couldn’t have any chance so far (shame on me...); and (2) I am now planning to prepare post-covid life – more specifically, I must pick up a very serious closed-back pair for myself to alleviate hard work and tough schedules at work.

    Before moving forward, I want to make it hyper clear that no matter how harshly I will go, this is one awesome pair I have ever experienced. Good things were already well documented elsewhere, and to make my post as concise as possible, I took out all the shameless love-lettering statements, which could alter tone of this post greatly. Anything I don’t describe in this post would be no less than good/decent.

    Most of my listening was done with Soekris DAC 2541’s XLR headphone out with -49 to -50 decibel attenuations. Super high efficiency didn’t motivate me much about other amp options. The dac is configured with black filter with external 2x upconverting pre-process.

    ESX900 was very sensitive to how to position and seal. I had to pay an extreme caution -- enough stabilization time required in switching to and from different headphones. After a couple of days, I settled down with placing my ears slightly back and up. Tried to mimic similar positioning in my measurements as well.


    Aesthetics and Comfort

    The loaner unit weighted 505g with my scale. The OG was around 400g if I recall correctly.

    Except the ebony cups were super beautiful, ESX seemed similar with the og. Loose clamping helped long listening, but I didn’t feel it solidly seal or stick to my head.

    Additionally, I was not a big fan of TP front damping nor the cable came with esx (a little less soft than I prefer).


    Hype Verified


    Lots of previous reviewers/posters were deeply impressed by its bass. I was not an exception. Overall sound reminded me of Beats Solo 2 with the audiophile-grade technicality.

    Let me elaborate more. Good things first, ESX is by far the best performer when it comes to bass impact game. It could effortlessly handle lower bass down to 20hz or even below. Except actual physical vibration, the experience largely resembles that of gigantic subwoofers in decent living rooms (carefully configured of course). Literally brain-shaking. I’m not kidding.

    Two tracks impressed me particularly during the evaluation: Both from the film score for the dark knight (by Hans Zimmer and James Howard). In the first track (“Why so serious”), there are very challenging sections where brutally loud synth notes around 30hz mixed into heavy metal segments and metal chords at top volume. The next track “I’m not a hero” introduces a sustained and massive 20-25hz synth notes. My listening notes had this description: “bass is like a wall and there is constant pressure”. Maybe I should seriously consider re-watching this legendary film out of ESX.

    My references did these tracks also well, but Clear Mg (which I think the best bass-head performer among Focals) didn’t have this much loudness at this clarity level (I often found it somewhat SPL-restricted in bass unless sacrificing clarity). LCD-X does enough volume but lacks low-end weights by comparison.

    Downside is it seems to lose some delicacy and delineation. It may not be the most bass-resolving headphones. Frequency-wise details subdued on 40hz downward. Micro dynamics and finesse also start to show compromises in this region. The intro part in Beyonce’s “Partition” lack articulate and fine gradation I used to hear with Clear Mg or LCD-X. Not as withdrawn as HD650 though. This issue could be worsened by bass tonality as I found a little improvement by applying negative low shelf filter, but I felt like there was more fundamental omission.

    Note: Maybe I am just over-demanding. Dynamic drivers had inherent drawbacks below their resonance frequency. Planar / Ribbon / Electrostatic generally outperforms on average.

    I also noted bass transient per se was pretty good, but with its own recessed highs, attacks seem a little rounder and decays trail longer than I like. Particularly so when passages were busy.


    Not Bad… But

    In the midrange and upward, ESX presented what bio cellulose drivers are capable of. Great tone, unique timbre, and Evan’s secret sauce on highs combinedly led me to very pleasing and forgiving session with virtually any tracks. Listening fatigue is the least relevant word in describing ESX. Its lower midrange decay seemed a little slow to me (reminded me of ZMF Eikon) and slight cup resonance was noticeable, too. Not necessarily bad. They made a lot of vocal and some instrumental tracks more emotionally touching and “speaker-like” (in spectral balance).

    One drawback I noted was that in dense tracks its midrange performance doesn’t fit well. In such a context, ESX’s relative tonal balance between lower mid and upper bass is a bit too much right skewed. I occasionally heard muddiness and over-bloom which were distracting a bit.

    I have also been struggling with ear gain parts around 3khz. ESX often sounded too subdued and resulted in loss in every kind of presence, bite, and sizzle to my liking. In less carefully recorded tracks, this character also harmed separation and placement of music components.

    Listening to Kubota Osamu’s “Presto” with ESX, which I hadn’t played in years, I found the sound was a little off in various aspects than I recall. This rendition should show some magical moments in haunting melodies, sensual harmonies, and massive dynamic contrasts all created by Kubota’s rhythmic and creative uses of electronic piano and synth. But attacks were too soft and congested to achieve so. I first blamed its recording quality which is subpar at best. But switching to LCD-X resolved almost every issue by better handling lower treble, eventually enabled me to hear an occasional touch of wow.

    “Colors” composed/played by Takayuki Isikawa and sung by Sunny presented similar issues. Both vocal and piano parts sounded too distant and less present than I favor. ESX conveyed a good bass experience in this track which was good. But the total trance experience was dragged by too much attenuated lower and mid (maybe some upper as well) trebles and did not align with the ways how I wanted to enjoy this music.


    Wrap-Up

    Even with some minor shortcomings I nitpickingly elaborated above, there’s no doubt that ESX is a great achievement. More so considering my own traumatic auditions of TH900 OG. I really, really, liked how ESX sounded.

    Next question is, as a potential buyer in the market (like I said in the opening), do I have to or want to buy it as a work headphone?

    Well, maybe not. There are several elements that prevent me from being fully into ESX.

    First, ESX’s limitations in my perception bug me obviously in many lower tracks I listen day to day.

    Second, ESX’s over-sensitivity in positioning may not be the greatest experience I would require for my own pairs. Every time I wear ESX, I had to start from weird feeling (temporary channel imbalance included). After some adjustments and stabilization, everything became ok. This could improve if I spend more time as an owner, but hard to justify myself given my other headphones do not present this.

    Third, practically speaking, ESX isn’t cheap as buyers must prepare the own pair of TH900 (I doubt if sub-TH900 alternatives such as X00 do the equivalent technical justice). Factoring in TH900’s price, the total cost may approach to a used pair of Verite. I spoke to one of the ESX owners before, and his hypothetical price (assuming he might be unlucky) was between $1.5k to $2k.

    Last but not least, I just started my research. Want to hear other alternatives (from both major and boutique players) before making a firm conclusion. And in this respect, ESX900 gave me a lot of insights and food for thoughts. Also, during my evaluation, I didn’t particularly bother typical compromises of closed backs, which may be telling something.

    Of course, looking forward to getting my hands on ETA’s other works down the road.

    That’s pretty much about my subjective evaluation. In the subsequent post(s), I will share my measurements (to be posted tomorrow or so). Will also provide my comments or thoughts on something interesting I found during the evaluation.

    Spoiler: ESX is rather a decent measuring headphone, but it did show several interesting behaviors I haven’t seen in my other measurements.


    upload_2021-5-6_23-4-8.png

    Relevant Thread(s):

     
    • Like Like x 11
    • Epic Epic x 4
    • List
    Last edited: May 8, 2021
  2. E_Schaaf

    E_Schaaf MOT: E.T.A Headphones

    Pyrate BWC
    Joined:
    Mar 28, 2017
    Likes Received:
    9,611
    Trophy Points:
    113
    Location:
    Staten Island, NY
    Home Page:
  3. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,847
    Trophy Points:
    113
    Location:
    East Coast
    First things first. Here are my usual set of measurements: FR, distortion, and waterfall plots for both channels.

    eta_esx_900_fr_lr.jpg

    Slight imbalance between left and right channels observed. However, as I posted above, this headphone is very sensitive to positioning. Above is the most successful results averaged over 3 different measurement sets (to reduce random human error). Subjectively, after some adjustments, channel imbalance would not bother eventually (center images were not exactly at the center but everything to an acceptable extent).

    Gradual down-slope up to 3khz and more recessed 5-9khz could be associated with its Solo2-ish sounding. Technicality is hugely different, but both headphones share similar characters in positive and negative aspects imho.

    The plot above presents very good low end extension, but I'd emphasize it is heavily subject to air gap and leak. Without pushing pads ahead of measurements, there were often few decibel loss in 60hz downward -- both objectively and subjectively.

    eta_esx_900_dist_l.jpg
    eta_esx_900_dist_r.jpg
    I remember the OG TH900 had pretty good distortion profile (unfortunately I've not measured the og with this rig). ESX seems an extension. Some measurement artifacts here and there, but nothing to worry too much.

    Nevertheless, I am a little bit curious if the specific tuning decision affected higher order harmonics. Note that ESX900 -- and likely Geneses too -- uses internal backwave diffusion, which may create non-linearity by nature.

    eta_esx_900_waterfall_l.jpg
    eta_esx_900_waterfall_r.jpg
    Midrange energy seems to stay at low level after initial decay. Closed backs with more proactive absorption (or perforated and/or suede pads as well) may behave better in lower to middle midrange, but such always come with its own cost.

    In highs, fairly clean pattern observed as expected for BC drivers.

    Overall, as I said in the op, this is very decently measuring headphone.

    Phase and group delay results are a little more interesting to me. I will post them later with comments.
     
    • Epic Epic x 2
    • Like Like x 1
    • List
  4. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,847
    Trophy Points:
    113
    Location:
    East Coast
    Tried iphone, topping, and laptop out (dell xps). All very enjoyable, more so than my references. However, I somehow felt like that's resulted by lacking transparency. 2541's staging and transient supremacy were less clear to me with ESX900, which relatively made the other sources less annoying.

    I couldn't test Erish and Jotunheim 2 as both are currently in the factory for service (what a misfortune..). I tried Magni 3+ but found no benefit over 2541 hp out. 2541 hp out was rather more transparent.

    And thanks for reminding me of the perforations. I should have described that, but forgot somehow in editing my review. lol. Having perforations had clear benefit in making it more open up in highs. Not too much difference in bass tho. And very interestingly, I failed to get objective supporting for that. FR more or less similar between before and after (no significance over run to run variance). I will post FR delta graphs later.
     
  5. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,847
    Trophy Points:
    113
    Location:
    East Coast
    eta_esx_900_fr_varience.jpg

    Yup. This is easily by far the most difficult headphone to measure (in terms of stable and consistent results) among I experienced. Please keep this in mind in looking at ESX-900 measurements.
     
  6. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,847
    Trophy Points:
    113
    Location:
    East Coast
    As promised, here are phase measurements for ESX-900.

    Note that I don't do this analysis very often as the results are either very experimental (a lot of unverified interpretation, confounders, etc) or mostly useless (more or less the same across products).

    I am not 100% advocate of Dr Toole, but my measurement experience so far kinda supported his assertion that headphones are almost minimal-phase device. But rarely phase information can be interesting -- particularly when it comes to the 3rd party modifications. I noted such importance in measuring Aiwa's vintage ortho.

    Let me first make clear that my position is NEVER like responses should be minimally phased. No. Nah. Never! As I said above, all my phase investigation is experimental at best, doesn't have any authority, and most importantly purely done out of my own curiosity.

    For those who have zero/vague idea about what the hell phase means. Theoretically, it's one of two major components of all kind of waves in frequency domain. It indicates that unit slice of wave arrives to the reference point earlier or later (but I don't recommend to use this interpretation as it's rather confusing). Those who are comfortable with high school AP math or above, phase corresponds to imaginary parts of complex numbers and also equivalent to angular part of trigonometirc functions.

    For quick and practical understanding, REW provides fantastic non-technical explanations in its online manual. Interested party has better checking them out: Link1, Link2

    Lastly, I specifically looked at ESX-900's phase as I wasn't successful to use EQ to tailor this headphone. Phase sometimes tells me some story that aren't obvious in my usual measurement set.

    Ok, let's see the results now.

    eta_esx_900_phase_l.jpg
    eta_esx_900_phase_r.jpg

    One thing we can quickly notice is the phase discontinuously filliped at 8k-ish hz in both sides. That was better shown in group delay plots.

    eta_esx_900_gd_l.jpg eta_esx_900_gd_r.jpg

    I don't think it meaningful to assign too strong interpretation to raw values in phase/group delay. Instead, I am rather interested how distant actual phase is from the estimated minimum phase. This translates to the concept of excess group delay (i.e., actual group delay at f - minimum group delay at f). In ideal minimum phase system, excess delay must be zero across entire spectrum.

    In the plots above, you will see that's not the case from the following points.

    • Peaks at 8.7khz (focus on white curves)
    • Dips at 4-5khz
    Please ignore the latter for now. That's not useful information -- compensation artifact. I will explain later in a separate post.

    For the reference, here is what it would look like in usual cases (focus on the dotted line):

    hd650_group_delay.jpg

    And this is not only for HD650. All my unmodded headphones behaved identically.

    If you closely compare group delay and phase plots side by side, it's straightforward to notice peaks at the group delay well matched to phase flipping point.

    I am suspecting this measured behavior is strongly associated with Evan's unique tuning method. If I correctly understood, he proactively used internal wave diffusion, reflection, and interference to get the right tonality in highs he wanted. I don't know the exact recipe, thus can't say anything firmly though.

    In the context of speaker's reproduction in the real room, minimum phase is easily violated when multiple different room modes and reflections interact (very typical textbook example). It's not good-or-bad stuff. Things just happen that way. It's interesting to see ESX-900 presents similar behaviors.

    While I don't know if this is associated with good or bad experience, there's one important takeaway. EQ may not be the good tool for this headphone. I will omit the technical detail (By all means, I am far from the very qualified person to educate it to the mass), but both theoretically and empirically EQ cannot apply in a linear way to the mixed phase system. Equalized results are unpredictable at best or even introduce unintended bad things at worst. We can still EQ anyhow with a lot of trial and error and iterative adjustments though (or maybe complicated math can help).

    I will investigate this point again with geneses when the loaner units here later this month or so. Personally very curious if this happens again to those open-backs. Will follow up and report back for sure.

    PS. Minimum phase does not guarantee EQ-friendliness. EQ may not work as intended for various reasons (sometimes what we can measure differs from what we want to measure..). Minimum phase is just a necessarily condition (never a sufficient one).
     
    Last edited: May 8, 2021
  7. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,847
    Trophy Points:
    113
    Location:
    East Coast
    Long story short, that's because SBAF compensation did not or did insufficiently address phase components in calibration.

    sbaf_comp_calcurve.jpg

    Above is the sbaf compensation calibration data I am using. As shown in the plot, it did calibrate "SPL" (aka wave amplitude) on highs.

    If boosting or cut naturally happens in acoustic domain, for each frequency, there must be a phase change. Naively speaking, all peaks and dips cannot exist without nonzero phase.

    In digital domain, and specifically the way how REW calibrates require users to input BOTH spl and phase calibration. The latter part is largely ignored probably because not that useful in 99% of headphone cases. Unlike SPL info, phase is also tricky to benefit from averaging multiple data points.

    I can estimate minimum phase calibration either empirically or computationally, but I was procrastinating to apply that to the compensation curve because I thought the expected benefit wasn't very worthy. Working with different data resolutions isn't very fun to me.

    sbaf_comp_egd_dip.jpg

    Anyway, based on the results I got so far, I am >80% confident that 4.38khz dip in the excess group delay is attributable to the lack of phase calibration in sbaf compensation.
     

Share This Page