Saturday 23 March 2024

EARLY LOOK (Part II): E1DA #9039S USB Balanced DAC dongle/headphone amp - DSD and performance under load. A few more words on the "need" for higher DAC fidelity!

Last week, I published Part I of the review/measurements on the upcoming E1DA #9039S USB dongle DAC/headphone amplifier. Already, we've seen that this ESS ES9039Q2M-based balanced DAC produces very clean, high-resolution sound.

As we continue, let's delve deeper into the performance of this little device. We'll have a look at some features like DSD performance and more importantly, let's examine the balanced amplifier output when subjected to headphone-like loads to see how well it performs.


First let's just get DSD over with. 😏

Honestly guys, I rarely use DSD these days because there's just not much native DSD content I listen to regularly (as discussed years ago, many SACDs are just upsampled PCM). Even if a recording is native DSD, it's more than likely just SACD/DSD64 which is rather noisy just above 20kHz whereas resolution at 24/192+ already would be of phenomenal quality with plenty of tools available in the studio for processing.

Upsampling PCM to DSD128+ can be done these days in realtime with software like HQPlayer but IMO the benefits are really not significant compared to DSP in the form of digital room correction or other processes like crosstalk cancellation for example. Since we cannot keep DSD in its native 1-bit format for DSP, multibit conversion is necessary, so we might as well use PCM! Furthermore, for many DACs including these ESS chips, internally we're looking at a multi-bit sigma-delta ("HyperStream IV") modulator which will modify your 1-bit input anyways; perhaps an issue for some DSD purists! (Some DAC chips like the AKM ones have a DSD Direct pathway.)

For testing DSD, let's use my Google Pixel 8 PRO (Android 14) phone with USB Audio Player PRO:

As I mentioned last time, the "zatoichi" firmware that I based my measurements on is able to handle up to DSD128 DoP and one would need to use the non-Zatoichi variant for native DSD playback up to DSD256.

To start, let me grab a reading of a clean PCM 24/96 1kHz 0dBFS off the phone just so we can compare with the DSD output.

We get THD+N of -122dB for this signal which is a little lower than the -125dB from my measurement MiniPC as discussed last week. This is because this 0dBFS 1kHz digital data is a dithered 24/96 FLAC file rather than using REW on my measurement computer sending 32-bit data with 25th-bit dithering. Typically with most DACs I don't notice a difference, but when we're down at this kind of resolution beyond THD of -140dB and THD+N of -120dB, the flat triangular dithering can have an impact numerically - not audibly of course!

As discussed a few years back, I base my DSD testing "by the book" - Sony's "Scarlet Book" that is - which described the standard for SACD. Specifically, for music, keeping peak signal at +3.1dBDSD (which is equivalent to -2.9dBFS PCM). Furthermore, I encoded my signals with freely available SOX incorporating DSD/PDM support. You'll find the "Arch Standard +3.1dBDSD 1kHz DSD Signal" in that previous post which is what I'll be using here.

"Zatoichi_v02.hex" Firmware:

So, starting with the synchronous Zatoichi firmware, let's look at DSD64 and DSD128 playback via DoP:

Looks great. A "perfect" DAC can extract around THD+N -116dB from that +3.1dBDSD DSD64 signal, so the -114dB we're seeing with the #9039S is excellent, almost ideal.

The fact that we're seeing -119dB with the +3.1dBDSD DSD128 signal implies that a full amplitude +6dBDSD signal would push beyond -120dB even though by Scarlet Book standards this would be defined as excessively "hot".

Notice that I'm showing the FFT up to 96kHz (192kHz ADC samplerate) which highlights the usual ultrasonic DSD noise, clearly worse with DSD64 with the noise rising from just beyond 20kHz while DSD128 will push the noise a bit further out and can be seen as a more gentle rise.

Let's have a look at the 1/10 Decade Multitone 32 played back as DSD128 for a more complex signal:

Yup, not bad at all. The signal was pushed as loud as possible without clipping and I've included the peak noise FFT (light blue) over a number of minutes of playback showing freedom from distortion for at least 110dB below the multitone peaks (similar to the PCM version of this test shown last time).

"9039s_v05.hex" Firmware:

Now let's look at the alternative non-Zatoichi firmware which runs the ES9039Q2M and USB interface chips asynchronously but provides the option of native DSD playback to DSD256:

FFT up to 96kHz.
As usual, notice the reduction in ultrasonic noise with higher DSD speeds.

Excellent still. Honestly folks, I would have no issues with using this firmware if I needed to play DSD256. For the perfectionists, notice that the THD values are a bit better with the Zatoichi firmware for both DSD64 and DSD128 even though the THD+N are about equivalent.

Tip: This might not be relevant once the final #9039S hardware & firmwares are released. For a couple of nights I was wondering why the non-Zatoichi firmware measured unusually poorly on my test bench. I found out that it was because my "THD DAC" calibration settings and output level got messed up when I flashed the firmware. Easily fixed with the "E1DA_9039_Tweak" app I mentioned last time in Part I. Might want to double check each time firmware changed just in case.


II. Performance as headphone amp

As with any headphone amplifier, the #9039S will need to manage the typical impedances from headsets out there. While many headphones these days have a relatively flat impedance profile, some do vary a bit and there's a whole range of impedance levels so it would be nice for the headphone amp to handle everything from say 20Ω to 600Ω without excess distortion and with adequate power.

As discussed a few years ago, I typically measure loads at 20/75/560Ω levels representative of low/medium/high-impedance headphone loads.

For convenience, I'm going to deviate and start using the E1DA Accessory Load Board. I made a 3.5mm balanced phono-to-test probe/GND cable as seen here:

The Load Board has both 2.5mm balanced and 3.5mm single-ended connectors to plug into your headphone jack. The centrally located DIP switches allow setting the load amount from 10-300Ω or open/no-load for each channel. Make sure to set the last two DIP switches to the proper balanced/single-ended configuration to make sure not to damage the amp.

Here's what it looks like when testing:

As you can see, I'm tapping the balanced L+/L- test points with the 2.5mm phono connector plugged into the #9039S output.

Since frequency response is the most important thing to get right, the first thing we must examine is the headphone amplifier's output impedance; the lower the impedance, the better "load invariance" including better damping factor. This can be measured based on voltage drop between the unloaded state and with a low 20Ω load, calculating the output impedance:

With a flat output impedance curve averaging <0.2Ω across 20Hz to 20kHz, I suspect you'll have no trouble driving any sane headphone using the #9039S. While I don't have any balanced headphones with wide impedance swings, we can have a look at the frequency response under various loads and estimate what we might see when a headphone of fluctuating impedance is plugged in:

Notice only 0.5dB range on Y-axis.

As you can see, I put the cursor at 1kHz so that we can read the dB for each load. Notice that between "No Load" (which would be "infinite" but in actuality 200kΩ for the E1DA Scaler used) and 16Ω is only about 0.1dB. This implies that even if you have a headphone-from-hell that had impedance fluctuations from say 1kΩ to 16Ω across the audible spectrum, all you might hear is an insignificant fluctuation of 0.1dB across the frequency response. This is what "load invariance" looks like.

Beyond frequency response, how does this headphone amplifier perform in terms of distortion under load? Check out these results with a low 16.2Ω load (1kHz tone, -80dBFS stepped sine up to 0dBFS):

Cursor placed at 1Vrms, and -120dB for reference.

As you can see, there's clipping at 0dBFS/3.5Vrms. But so long as you avoid pushing the listening volume to 100% and music not frequently clipping or hitting peak level (!), the headphone output is able to achieve phenomenal resolution even at -0.5dBFS (3.3Vrms):

Wow! THD+N of -124dB into 16Ω at 3.3Vrms. That's just a mere 1dB reduction from THD+N -125dB unloaded at 0dBFS. At 3.3Vrms, the amplifier is pushing 0.7W into the 16Ω load using this USB dongle.

Let's examine the harmonic distortion performance across the audible frequencies (not just 1kHz) at various output levels:

Very well behaved. As expected with a high-quality balanced device, the even harmonics are well suppressed. Odd harmonics tend to be higher. Note that I'm not able to benefit from the E1DA Cosmos APU (1kHz notch filter) for these loop-back stepped sine THD vs. frequency measurements, hence the THD+N of the 1kHz measurement sits at a "mere" -118dB @ -1dBFS into 16Ω.

Here's my Triple-Tone into 16Ω so we can have a look at harmonic and intermodulation products along with a TD+N resolution score:

-111dB on the Triple-Tone TD+N into 16Ω is excellent. Within the audible frequencies, the highest distortions are below -114dB, nothing of concern. Unloaded, the TD+N measured -117dB so there is a drop as expected, but this is not much given the punishing low-impedance conditions and more complex signal.

As we increase the load, the results are predictably excellent with no clipping to the full 3.5Vrms level. Here are the stepped sine 1kHz graphs into 20.5Ω, 80Ω, and 300Ω loads:

3.5Vrms into 20.5, 80, and 300Ω loads correspond to peak power of 0.6W, 150mW, and 40mW respectively. Often, we see headphone amp specs documenting power into 32Ω - for the #9039S, this would be 380mW which should be more than adequate for the majority of mobile listeners! Obviously, check your headphone sensitivity and make sure to protect hearing acuity if you're in need of high-power headphone amplification. 😐

Since I like to try for apples-to-apples comparisons, let's have a look at the Triple-Tone TD+N result into 20.5Ω at 0.5Vrms:

Distortion products below -120dB.

TD+N of -108dB in this test is excellent, besting the Sabaj A20d 2022's headphone out which achieved about -99dB and the Drop+THX AAA 789 at -93dB - both of which are full-sized desktop devices that can provide more power even though lower resolution score.

#9039S, world class balanced DAC & portable headphone amp performance in the palm of your hand... Literally.

III. Final Summary

Taking Part I and today's Part II results together, objectively and subjectively, the E1DA #9039S is a phenomenal performing little USB-powered balanced DAC/headphone amplifier with 2.5mm phono output. The measured performance is in line with some of the highest fidelity desktop DACs. While the final MSRP is yet to be announced, I'm quite sure the price will represent seriously good value for an objectively "super-high-fidelity" audio converter than can handle up to 32/384 PCM and DSD256.

As a little headphone amplifier, the output impedance of <0.2Ω flat across the audible frequencies will mate well with basically any non-exotic headphones without significant frequency variation. The fact that this device is able to maintain low noise and distortion at up to around 3.3Vrms into a very low 16Ω load (0.7W) with better than -120dB THD+N is certainly a remarkable feat!

If you get one of these, digging deeper into the performance, the Zatoichi firmware lineage is best for highest fidelity playback so long as you don't need native DSD256 support (ie. up to DoP DSD128 is probably adequate for the vast majority of listeners). That's really just super nit-picky when we're talking about a DAC that's easily capable of better-than THD+N of -120dB!

As with Ivan's other E1DA line of dongle DACs (like the #9038D6K), there are also opportunities for tweakers to play with the registers as discussed last time. I'm not sure if there will be an Android "Tweak9039" like the Tweak9038 app discussed previously.

Despite the abuses delivered over a number of evenings of testing with different computers, phones, and various loads, this little dongle DAC handled it with grace. Even playing at full volume at extended periods of time into low impedance loads, the metal case only got mildly warm.

Considering that Ivan has been able to squeeze beyond -125dB unweighted THD+N out of this little ES9039Q2M chip that's only specified for -120dB already, I wonder what he can do with the ES9039PRO with specs beyond 130dB dynamic range and even better THD+N or maybe even dual-9039Q2M's?! Will see I guess... I assume he still has his eye on making the larger Cosmos DAC one day. 🙂


A few more words on finding balance in audiophilia...

To end, on the subjective side, an audiophile/reviewer can easily make all kinds of comments, claims and beliefs that are fanciful and false if accepted without critical thinking and unconfirmed by others. By nature, one of the outcomes of capitalism is to create a psychological desire in the consumer to spend money for "upgrades" whether he/she needs it or not. One consistent theme in this blog is the importance of recognition that as humans, we do have thresholds to our perceptual abilities and cognitive limits. Let's be honest, when (especially elderly) audiophiles insist that they can hear all kinds of differences between excellent devices like DACs, we have to be at least a bit suspicious. Especially when we measure extremely low noise levels, low distortion, minuscule jitter, excellent frequency response, etc.

So too on the objective side, we need to make sure we do our own "reality checks". For example, assuming we believe there is a correlation between harmonic distortion as measured by something like THD+N / SINAD and fidelity, how ideal of a number do we need? Even taking a "round number" of say 100dB SINAD at full amplitude, I suspect this would easily satisfy the desires of the vast majority of audiophiles. Obviously it's not hard to find a good DAC these days surpassing this level of low distortion. (As discussed in the listening tests a few years back, I actually don't even think the vast majority of listeners would complain with THD -75dB.)

Over the years, with progressive improvements in DAC performance as we're seeing here, obviously there comes a time when THD+N and SINAD numbers become badges of honor for our engineering friends as they achieve milestones in signal purity pushing further into the limits of Johnson-Nyquist noise! Their job is to push the frontiers into these higher and higher levels of performance whether as consumers we want or need this. I know that Ivan takes great joy in that challenge with his designs. 😉

So, is the THD+N/SINAD War over for DACs? Yeah, as an audiophile, I think fidelity of DACs these days is well beyond the task of satisfying the needs of human hearing even with small, inexpensive devices. High-fidelity audio lovers have "won" not with any single device but rather bountiful options at all kinds of price points and esthetic variations. We are very blessed to be living in such a time of progress and affordability.

As expressed before, I think even the Topping D10s was "perceptibly perfect" already with no significant flaws at a very low price back in 2021. I know that the "high-end", high-priced audio industry might not appreciate such comments, and it goes against the beliefs/intents/desires of certain reviewers/writers/YouTubers... Oh well, too bad. Some things are just factual.


Finally, I wanted to mention the albums above by Chinese singer Liu Liang Lu (劉亮鷺) who might already be familiar to Asian audiophiles. I recently had a listen to 舊夢不需記 ("No Need To Remember Old Dreams") (Cantonese, 2020, DR12) and 江湖情 ("Love In Jianghu") (Mandarin, 2021, DR12), two fantastically recorded and produced albums in the "audiophile male vocal" genre.

For years, the sound quality achieved by many albums I've heard from overseas IMO has put much of the North American "audiophile" material to shame. Notice the excellent dynamic range (DR12) of these two albums. Excellent resolution, deep bass, nice capture of vocal and instrumental timbre/color, fantastic soundstage width, precision of instrumental and vocal placement.

Alright folks, time for Spring Break. Hope you're enjoying your music - whether you're mobile or otherwise!

Can't speak, got a beat in my headphones...

Addendum (March 24, 2024): I received E-mails about the availability of this DAC/headphone amp. The last note I received from Ivan was that he couldn't find the "right solid-polymer flat caps 1000-1500uF in China... So, currently doing re-layout of the PCM to fit alternative caps". Otherwise, "pretty much ready to start"; so I assume coming soon.

Highly recommend checking out the E1DA Discord for discussions.

Addendum (March 27, 2024): I had a listen to the little DAC/amp in the main system tonight. A bit contorted cablewise with using balanced 2.5mm → dual XLR (male) then XLR (female) → TRS Balanced cables to Topping PA5 MkII+ amplifier → Paradigm Signature S8 v.3 front speakers.

It sounded great on the Android phone as in the image above, but noticed that there was some noise pick-up using the fanless MiniPC and less noise but not pristine with the Raspberry Pi 4 "Touch".

As such, I would stick with using the little DAC/amp for the intended purpose of being a balanced headphone amp! Wait for the Cosmos DAC...


  1. That's very impressive, especially considering that it's a USB bus powered device. Out of curiosity—did you also try measuring using your RME as an ADC? I hope, there is no "equipment bias" due to the fact the entire measurement chain is from the same manufacturer :)

    1. Hi Mikhail,
      Yeah, that's amazing performance! Even if not USB powered, I don't know which other headphone outputs would be able to achieve better than -120dB THD+N into 16-ohms.

      Yup, one evening I did use the RME to do a quick measure to make sure the set-up was good; this was when I was fooling around with the tweaking (discussed in Part I). Clearly the resolution of the DAC was beyond what the RME could measure without using a notch filter!

  2. Archimago,

    You hit upon this subject in regards to measuring vanishingly small differences in DAC performance but..

    To me seeing all these measurements, and the amount of time folks over on ASR spend measuring DACs, it's almost like the audiophile version of Phantom Limb syndrome. We used to have analog sources - turntables, cartridges etc - which not only measured quite different but were often way in to the audible zone, so this stuff really mattered. But once we moved to digital, well, it got mostly academic. But audiophiles still have to scratch that phantom limb itch of measuring stuff, whether there are audible differences or not.

    The same goes for the "subjectivist" side as well: once we got essentially transparent/accurate digital sources did all the tweak stuff stop? Of course not. Subjectivist/golden eared audiophiles have found every which way to "improve" the sound of digital with all manner of whacky implausible tweaks.

    For me digital is a done deal. I have a Benchmark DAC2L. I never have to think about digital again.

    1. Hey there Vaal,
      Certainly can appreciate your position and I agree as well. After testing a number of DACs and running blind listening myself - volume controlled, removed as much other variable as possible - the differences between DACs are clearly at best subtle so long as the measurements indicate that we're listening to bona fide "high-fidelity" DACs without unusual distortions primarily in the frequency domain (time domain accuracy of course beneficial, but nowhere as important unless severely anomalous!).

      Yup, the Benchmark DAC2L has all the characteristics of an excellent DAC capable of hi-res performance so I would have no concerns soundwise and could happily be a "forever DAC"! No... There is no subjective "magic" in the technology itself; as I've suggested for years, the "magic" lies with the artists and how they convey their art to use as we enjoy the music, not what some "iconic" audio designers believe they delivered to the consumer.

      Love the "Phantom Limb" analogy :-).

      This is why over the last year I haven't been measuring as much DACs and spending more time with the inexpensive TI chip amps or talking about "immersion". Unless there's something likely new I can experience and demonstrate (like this E1DA headphone amp pushing the boundaries of resolution into low impedance loads!), I've actually declined a few offers from companies wanting to send me products for review.

      I think it's good that the ASR guys continuing to publish the data; this puts useful pressure on companies that are not actually putting out good technology but still advertising as "high-fidelity" or "high-resolution" since these terms have an objective definition we should as an audiophile hobby be vigilant about (rather than that silly "high end audio" term which is mostly about just spending disposable $$$$ and bragging rights).

  3. Hi Archi,
    Are you going to try to listen to music with this tiny DAC on your main audio system with speakers? I would be interested to know your impressions to see how it sounds with big speakers.

    1. Hey there fgk,
      Posted a little addendum on this. Sounded really good with the Google Pixel Pro 8 phone and USB Audio Player PRO; clean sound connected to the Topping PA5 Mk II+ amplifier.

      However some issues with crackling noise using that set-up with other devices so I would not recommended at least with that balanced Topping amplifier combo. Might be better with other amplifiers or maybe using USB isolator with the computer (didn't check). I suspect it's the grounding and will need to double check.

      I would stick with using this for headphones as intended...