EARLY LOOK (Part I): E1DA #9039S USB Balanced DAC/headphone amp - Super Hi-Res, Tiny Package! Also, let's tweak...

Hey everyone, it's time to have another look at a product from the engineering workshop of Ivan Khlyupin (IVX) at E1DA - the E1DA #9039S:

Notice balanced 2.5mm TRRS phono output.

As you can see, it's a prototype/preproduction (depending on whether the circuit could still change) unit with pen markings on the box for the number "9" to make sure the correct product was sent my way. 🙂

This is the latest iteration of E1DA's line of USB2.0 dongle DACs which includes the E1DA #9038D6K I had a look/listen to in late 2022. As you likely surmised, this update is based on the next generation of ESS Sabre DACs - the low-power, 2-channel, ES9039Q2M, using their sigma-delta Hyperstream IV modulator. The chip specifications list 130dB of dynamic range, -126dB THD and -120dB THD+N. We'll see in a little bit what Ivan has been able to "cook up" in his lab including some performance results. Back in 2022 with the #9038D6K DAC, he had already achieved -120dB THD+N (120dB SINAD) with single-ended output. How much better does it get!?*

As I sometimes do, due to time limitations, I'll split this write-up into 2 portions to present different roles/functions of the device. For today's Part I, let's just focus on the usual look-and-feel as well as using this device as a straight, unloaded balanced DAC. This will give us an idea of the performance capabilities of the ES9039Q2M in the hands of an experienced engineer. Next time we'll examine this device as a headphone amp.

The current anticipated price for this USB DAC/amp is presumably going to be less than US$150; I believe the exact number is still being crunched.

[* Asking how much better in terms of THD+N is of course a different question from "How much resolution does a person need for excellent high-fidelity reproduction?"!]

I. Physical product and a little bit about balanced headphone outputs.

Here's a look side-by-side with the #9038D6K (see previous review for more about build, weight, etc.):

As you can see, they're basically identical in form factor, both featuring a USB-C digital input. The only differences being the 3.5mm single ended port on the #9038D6K as opposed to the 2.5mm "balanced" port of the #9039S with markings and illustrations on the rear to identify the model. As usual, the rigid aluminum case feels sturdy and the laser-engraved lettering is sharp. The DAC is capable of handling up to 32/384 PCM and DSD256 (depends on the firmware, see below).

At this juncture, I think it's useful for us to talk about what "balanced" cabling means in the headphone world. First, let's look at the pin designations for a typical 3.5mm "single ended" vs. 2.5mm "balanced" headphone jack:

The 3.5mm single-ended TRS jack has Left and Right signals going to the respective headphone driver referenced to GND. The 2.5mm balanced TRRS jack instead has separate, active, complementary L+/L- and R+/R- connections per headphone. With the small 2.5mm connector, there's no GND pin in this system. Note that the larger 4.4mm Pentaconn balanced headphone connectors do have GND (see comparison image here).

In my mind, the term "balanced" is usually connected with line-level differential signaling such as from balanced DACs connected to balanced (pre)amps using XLR or 1/4" TRS/channel cables. With balanced differential line-level cables, signals are transmitted as equal-impedance hot/cold inverted pairs and the receiver extracts the difference between these pairs unlike the simpler single-ended cable referencing each channel to ground. Because it's the difference being extracted, noise like RF interference common to each line will be subtracted out. This is the "common mode noise rejection" term you might have seen mentioned.

Noise rejection would be particularly useful when running long lengths of cables (like >25' interconnects) between say your DAC to amplifier across a room or low-level signals that might be more easily corrupted by noise like microphone signals.

[Note: not all XLR analog signals need be differentially balanced; some could be "pseudo-balanced" for example - see here for interesting circuitry discussions. The Sabaj A20d-2022's 4.4mm headphone output is not differential balanced yet performs quite well.]

When we look at how headphones are wired up, the main intent for "balanced" output is to be used as differential drive with each +/- wire connected to the respective pole of your headphone transducer. It's not about reducing common mode noise over long cables since it's unusual to have headphone cable lengths beyond a few feet anyways, especially for mobile devices. Rather, balanced headphone amps act as complementary "bridged" amplifiers, thus effectively doubling the maximum voltage swing that is delivered to your headphones. This ideal doubling of voltage (+6dB), and quadrupled power could obviously be beneficial for low-sensitivity headphones. Furthermore, because we don't use a common ground between the left and right channels, stereo crosstalk should be essentially eliminated (not that crosstalk is typically a problem). With precisely matching amplifiers, even-order harmonics can be suppressed (related to symmetry of the waveform). The down side is that each +/- amplifier as a pair needs to be very accurate in order to ensure low distortion and noise, output impedance is also doubled compared to single-ended and this could decrease damping factor.

This combination of pros and cons is why some manufacturers like Benchmark warn against the need for balanced headphone amps. In principle, low-noise, adequately powered single-ended amps should be all that's needed and should not be a problem especially with desktop headphone amplifier designs. As audiophiles, we should not hold on to an automatic assumption that balanced headphone amps are better. It depends on the product and as usual, I would look to the measured results to understand the device's capabilities rather than subjective testimonies only.

Have a look at this article for even more details on balanced headphone/amp functioning and wiring.

While I'm not specifically going to be measuring the #9039S as a "loaded" headphone amplifier in this Part I article, I figure it was good to mention the above as we look at the test results and you, dear audiophiles, can decide for yourself whether you need balanced headphone amplification.

II. Objective Performance

Over the years, while I've measured all kinds of devices, I have not actually tested 2.5mm balanced outputs from a little headphone DAC like this. As a result, to adapt the output as cleanly as possible to the very high-resolution E1DA "measurement rig" consisting of the Cosmos APU, ADC, and Scaler, I hacked together a 2.5mm balanced to dual 3.5mm balanced stereo phono cable.

Here's the pin recipe for those who want to build something similar:

The 2.5mm 4-pole balanced headphone connector I believed was initially used by Astell&Kern.

My cable ended up being 15", a convenient length but still relatively short so as to reduce interference pick-up. I took some inspiration from the official E1DA XLR adaptor to add the alligator clip for GND which can be connected to the USB-C shell (as you can see, there is some spacing between USB-C connector and the #9039S).

Since this device is meant for mobile applications, to achieve as much electrical isolation as possible from my MiniPC test rig, I used two Topping HS02 USB isolators both for feeding the ADC and the #9039S DAC to my computer (Ryzen MiniPC connected to the E1DA Cosmos components) like this:

MiniPC ↔ HS02 Isolator ↔ E1DA #9039S DAC →  custom cable → E1DA Cosmos "stack" (APU/Scaler/ADC) ↔ HS02 Isolator ↔ MiniPC

A peek at the DAC on the testbench wired up...

Since this is a prototype unit, Ivan supplied me with a number of pre-release firmware versions to try out. For the tests in this post, I'll use one called "zatoichi_v02.hex" which I found worked very well on the Windows PC with native UAC2 audio driver. [Zatoichi referring to the kick-ass blind swordsman.]

While the USB interface remains asynchronous, the "zatoichi" lineage internally synchronizes the DAC, and ComTrue CT7601 USB interface chip with a low-phase-noise NDK oscillator. While there are benefits with time-domain performance doing this (ie. better jitter), the CT7601 cannot handle sync mode and native DSD playback (only up to DSD128 DoP).

However, there is another firmware lineage with the most recent version I've seen being "9039S_v05.hex" which can be used for native DSD256 playback but has higher jitter. To be honest, I rarely use DSD these days (occasionally DSD64 for native SACD rips, DSD128 already great for PCM-to-DSD upsampling, and much less need for DSD256) so I'll stick with the "zatoichi" firmware here to demonstrate the best resolution achievable.

One last housekeeping item, I have set the DAC to "Linear phase fast" digital filter which is what I listen and do most of my measurements with. We'll talk about the digital filter in the tweaking app section at the end of this article which is how you adjust this.

Okay, time for some results...

0dBFS 1kHz Sine Waveform:

Let's just start with a look at the 1kHz 0dBFS sine waveform as captured in the ADC:

As you can see, at the macroscopic level of the full waveform, there's literally nothing to see except clean sine waves that overlap. Left channel at 3.49Vrms and the right at 3.48Vrms (basically 3.5Vrms). This 0.01Vrms (0.02dB) difference is insignificant and simply indicates very precise channel balance.

RightMark Comparison (24/96):

To start, here's a look at the results from the old RightMark test performed at 24/96. (These days, 16/44.1 is easily reproduced accurately and presents no challenge for any decent DAC so let's not bother with that here - I can confirm that the #9039S produced exemplary 16/44.1 results.)

The summary table above includes comparison data for the #9039S with #9038D6K (ES9038Q2M chip), the Sabaj A20d-2022 (ES9038PRO), SMSL DO100 (dual ES9038Q2M), RME ADI-2 Pro FS R Black Edition (AKM AK4493, 24dBu = 12.28Vrms peak output), and Chord Mojo (proprietary FPGA DAC).

As you can see, the #9039S produced excellent performance from such a little USB-powered dongle DAC. Although I didn't have the data easily accessible to load up, these results are clearly well beyond dongle DACs like the AudioQuest Dragonflies of yesteryear and much more in line with high-fidelity balanced desktop devices these days.

Of all the DACs in the comparison table, the E1DA #9038D6K and Chord Mojo are the only single-ended devices. I intentionally did not bother comparing this with really poor quality recent dongle DACs like the Cayin RU6.

Summary graphs:

Very good frequency response (basically flat with only -0.14dB at 20kHz), low noise, very low crosstalk and impressive IMD+N sweep. Notice how the ES9039Q2M in this little DAC is surpassing both the dual-ES9038Q2M's of the SMSL DO100 and even the ES9038PRO in the Sabaj A20d-2022 in many ways. Of course, those are full-sized desktop DACs so they provide advantages and various desirable features (like XLR output, multiple inputs, remote control, physical volume knob, on-screen setting customizations, etc.).

Dynamic Range:

Using the Cosmos Scaler's +26.7dB gain, we can get a good look at the dynamic range this DAC is capable of with a -60dBFS 1kHz tone:

Unweighted, we're seeing a DAC capable of about 128dB dynamic range.

Often in specs sheets we see dynamic range figures quoted as A-weighted, let's have a peek:

Great. With A-weighting, we get 130dB(A) dynamic range consistent with the ESS specs sheet.

Single-Tone Tests - THD(+N)/SINAD:

Using the E1DA APU (1kHz notch filter), we can get a very detailed look at the 1kHz THD+N/SINAD:

Beautifully clean THD+N of -125dB (SINAD 125dB) with the harmonics well suppressed below -140dB relative to the 1kHz tone; THD itself down at -140dB. 60Hz mains noise absent. Inter-channel output level difference only 0.02dB, consistent with other evidence of excellent channel balance.

Distortion across audible frequencies:

Let's check to make sure the low distortion isn't just at 1kHz! Here's a -0.5dBFS sweep across the audible spectrum:

Well-controlled THD+N throughout. There are some fluctuations in the 3rd harmonic across the frequencies but otherwise the 2nd harmonic level looks stable and generally the relative harmonic structure doesn't change much. Again, these are very low levels of distortion.

THD vs. Level:

How about the harmonic distortion across output levels?

Very nice. All the way up to 0dBFS (3.5Vrms) the THD+N follows the output level more-or-less linearly. Note that I'm measuring this without the aid of the APU so at 0dBFS I'm pushing it to the edge of the Cosmos ADC resolution. No 0dBFS clipping which we saw with the #9038D6K previously. The 2nd harmonic (red) generally remains low along with good suppression of other even harmonics. There's more 3rd but the levels are low and intermingles with other odd-order harmonics like the 5th and 7th.

With that THD vs. Output Level data, we can calculate the linearity of this DAC:

Deviation of a mere 0.2dB down at -125dBFS. Down at the expected limits of the ADC and I'm pretty sure no human being would complain. 😉

Multitone Tests:

Next, let's go beyond simple single-tone THD(+N) tests and examine some dual and multitone results. Since it's commonly done and you can compare this with graphs published for example in Stereophile, here's the CCIF intermodulation test consisting of 19 & 20kHz tones, both channels driven:

All distortion products in the audible frequencies are below -130dB. Very low intermodulation distortion even for such a strenuous test.

1/10 Decade Multitone 32:

Notice that I've plotted the 1/10 Decade MultiTone 32 with peak values (light purple) and the parameters I used (24/96, 1M FFT, 50% overlap). Even after a few minutes of audio capture (about 3 minutes) averaging over 32 samples, peak noise/anomalies at most reached around -110dB.

There's a little ultrasonic noise out around 28kHz which is insignificant down at -115dB. Presumably just picked up some interference from the computer or power supply. Again, no 60Hz hum or more significant anomaly in the audible frequencies.

Triple-Tone TD+N (48/960/5472Hz):

With both left and right channels playing, I'm seeing an average Triple-Tone TD+N of a mere -117dB. This is one of the best results I have seen over the years. While we can see a little "skirt" at the base of the 48Hz tone below -120dB, distortions are otherwise down below -130dB of the primary tones. The only other DACs I've come across with higher results are the ES9038PRO-based machines like the Topping D90SE (-120dB) or Sabaj A20d-2022 (-118dB) with balanced XLR and higher output levels (≥4Vrms).

Jitter:

A couple of tiny inconsequential sideband pairs down at -140dB in the 16-bit test well below the LSB jitter modulation signal, and one pair below -150dB in the 24-bit test.

As you can see, there is absolutely nothing here of concern using the Zatoichi synchronous firmware. Unless shown otherwise, I find that jitter is simply an inaudible non-issue with modern USB DACs like this. We're looking at basically ideal performance.

Google Pixel 8 Pro connected to E1DA #9039S, Drop+HiFiMan HE-4XX. If you look closely, I'm playing the SACD rip of Friday Night In San Francisco in DSD64, compressed using WavPack which USB Audio Player PRO can decode.

III. Subjective:

As usual, I listened to the DAC for a few days before I subjected it to the measurements. In the picture above, I have the HiFiMan HE-4XX with balanced wiring connected to the #9039S. As discussed in Section I, balanced cabling should not be complicated (it's really just plugging the +/- headphone drivers to the proper balanced connector), but yet aftermarket cables can be quite expensive! So while I have headphones like the Sennheiser HD800 that would be great for balanced use, I'm still on the lookout for reasonably priced cables. (Heck, might even end up making my own or just get inexpensive adaptors like these MMCX/3.5mm to HD800.)

Anyhow, with the HiFiMan HE-4XX, I had a listen to the recent remaster from Analogue Productions of their Steely Dan Can't Buy A Thrill (1972, 2023 SACD, DR11 for both SACD and regular CD). The new AP transfer sounds good and it's clear through high quality DACs like the #9039S that the remaster has boosted the bass. To confirm this impression, I loaded up the tracks to compare; here's the first track on the album, "Do It Again":

I normalized the mid-range frequencies between 1.5-3kHz in the overlay. It's not unusual to see remasters like this one done by Bernie Grundman adding a bit of EQ to modernize the sound, especially extra bass as in this example below 500Hz by about +2dB. The regular CD sounds really good already as is typical of the Steely Dan discography so feel free to just add a couple of dB's with your own EQ to emulate the remaster! 😉

IMO, there comes a time for these old analog transfers and remasters of music from the '70s and earlier to end (once we hit the '80s, most recordings transitioned to digital already). Yeah, labels can keep doing the "remaster" thing to sell a few more albums to collectors I suppose. However, over time, even though on the digital side (ADC, mastering tools) we can achieve even higher resolutions, the analog master tapes are getting older (Can't Buy A Thrill is 52 years old now) and as we've discussed, they deteriorate over time. Even the best master tapes achieve <80dB dynamic range which is easily digitized within the limits of 16-bits for decades already.

Anyhow, the #9039S sounded excellent whether playing the CD layer or as a DSD64 rip. It's clean and hi-fi sounding which is as it should. Transients sound tight, loads of detail. Very easy to hear the relative boost in the bass frequencies compared to the regular CD even with the inexpensive HiFiMan HE-4XX headphones.

For those interested in contemporary classical, I recommend a listen to Mason Bates' work Anthology of Fantastic Zoology (Live) (2016). Fantastic sounding capture of the performance. Beautifully transparent through the #9039S. I really like the repeating "sprite" motif (track 2) being played by the various instruments interspersed within the orchestra presenting a beautifully dynamic piece which on a good system should be highly titillating 🙂. I would also recommend trying this with crosstalk cancellation using speakers. You should be able to hear a very wide soundstage and precise placement of those "sprites" among the orchestra without "bunching up" of the instruments. As a live recording, there's an excellent amount of "realism" to be exploited so try not to miss out on that!

Your turn... Here's the AMPT recording for your listening pleasure; as usual, feel free to compare with recordings from other DACs:

E1DA #9039S AMPT Recording (24/96)

Sure... It might look odd, but slap on some short analog 2.5mm-to-XLR connectors to a balanced amp, paired with longer lossless USB digital cables. Note: discussed a little more in Part II - noticed some hum/grounding issues when tried so not recommended, just use a proper desktop DAC. :-) 

IV. Summary for Part I:

As I mentioned at the start, this article is an examination of the E1DA #9039S as a DAC, without a lower-impedance (headphone) load connected. What we're seeing is a very high-resolution balanced DAC, based on the ESS ES9039Q2M chip, with 1kHz 0dBFS/3.5Vrms output capable of 130dB(A) dynamic range, clean THD -140dB and THD+N -125dB! In concert with the excellent linearity, resistance to jitter, and concomitant high-resolution multitone performance, this DAC places among some of the objectively highest fidelity devices out there. Considering the tiny form factor, this might come as a surprise to some.

At likely less than US$150 (price currently not formalized), we're already seeing levels of performance equivalent and at times even beyond the previous generation ES9038PRO products. Another beautiful example of technological progress pushing the envelope of performance at lower price points, with each step becoming "even more perfect".

As usual, it's a pleasure to test out stuff from Ivan/IVX's workshop - yet another fascinating, affordable, high-fidelity product from E1DA!

Next time in Part II, let's continue to look at this little amp/DAC and see how well it performs when paired with headphone-like loads including characteristics like output impedance, and a quick peek at DSD playback.

--------------------

Fun with "E1DA_9039_Tweak"!

Before we end, like with other E1DA DACs, for those who want to get deeper into some of the hardware register-level fine-tuning, there is the "E1DA_Tweak_9039" program:

My default settings used for tests above.

CLK_GEAR changes the master clock speed (0 = 49.152MHz, 1 = 49.152MHz/2, 2 = 49.152MHz/4, 3 = 49.152MHz/8). Higher clock (ie. 0) results in better performance with lower noise but higher current draw which may or may not be significant on low-power mobile devices. For my test results above, I left this and the THD DAC values as default. The manufacturer default digital filter (under DAC1) is "Linear phase apo", but notice that I selected the "Linear phase fast" filter which is my usual preferred listening and testing default. Other than this, everything else was left untouched for the results presented above in the text.

Honestly guys and gals, these days, I don't care much about filter settings because it makes little difference unless the filter creates significant frequency effects (like the Pono Player from close to a decade back). Extreme filtering like say PGGB makes no sense to me unless one has a DAC with very poor built-in filter options and you desperately needed an alternative.

For the record, here's my "Digital Filter Composite" (DFC) graph for the tested "Linear phase fast" setting:

Notice the overloading with the wideband white noise (0dBFS peaks) suggesting limited digital filter headroom. This is typical for most chip DAC filters. Only devices with the best digital filtering that gracefully manages "intersample overs" will not show this behaviour (like the Chord Mojo).

And here's the DFC for the "Linear phase apo" setting which was the factory default:

Apodizing filters roll off the frequency response a bit earlier so you can see that drop in level between the green 19kHz and 20kHz peaks. Linear phase filters do show pre-ringing in the impulse response; but that's generally nothing to be concerned about as discussed years ago. Linear phase filters do not introduce phase shifts and steep/"fast" filters provide some of the tightest transients.

See the ES9039Q2M data sheet for more about the other filter options and impulse responses from each setting.

In the Tweak app, notice that the left and right 2nd and 3rd harmonic THD DAC levels can be fine-tuned to your liking, including increasing them substantially like this:

As you can see, adding big numbers (16-bit signed integer) to these parameters resulted in elevated harmonics. Feel free to experiment! Maybe you'll enjoy some potentially "euphonic" distortion from these elevated low-order harmonics, as one might find with tube amplifiers, although the higher harmonics above the 3rd are still much cleaner than an actual tube device!

Addendum: For those who might not be aware, the addition of these 2nd and 3rd harmonics intentionally introduces non-linearities into the waveform. So while we might use a single tone like the 1kHz sine to look for harmonic distortions, and think of dual-tones/multi-tones as intermodulation tests, they're just different ways of evaluating the underlying non-linear distortions. For example, those 2nd = 15000, 3rd = 10000 THD DAC settings will affect the 19+20kHz CCIF intermodulation test as well:

Notice the markedly worse IMD-power measurement which went from a minuscule -127dB (graph above in Section II) to this very poor -71dB.

An audiophile therefore cannot say: "I don't mind the sound of added low-order harmonics thanks to the tube amp I recently bought, but I hate it when a component adds intermodulation distortion!" THD and IMD are not separate phenomena. 

Again, presumably some audiophiles might subjectively experience euphonic pleasure listening to music with non-linearities that will result in intermodulation anomalies (which inherently increases with THD)?

Instead of purposely adding harmonics, obviously you can use these for fine-tuning and I was able to drop the 2nd harmonic of the left channel even further but this did not change the overall THD(+N) and the default values are already highly optimal.

There is another parameter called "IDAC_NUM" which is an option in the "CLK_GEAR" dropdown that seems to adjust the digital filter function. For example, one can set this from default 0 to IDAC_NUM 11 which will result in this interesting waveform:

Yup, a NOS-like 1kHz sine wave. An option to try if you like NOS (non-oversampling) distortions!

If we look at the frequency-domain FFT with this setting, we see that it only allows imaging distortions out to 2fs; so for example, here's a 18/19/20kHz 16/48 triple-tone and we can see the unfiltered mirror image at 28/29/30kHz causing the "squaring" of the waveform:

Wide bandwidth FFT sampled at 768kHz using RME ADI-2 Pro FS ADC (high frequency noise floor elevation is typical of the ADC, not from the DAC).

This is why I call it "NOS-like", or maybe more appropriately "NOS-lite" for those who just want a taste of NOS without too much ultrasonic high frequency content that could make the amplifier work too hard! 🙂

For the "full-spectrum-imaging-distortion" NOS afficionados, I suggest you try IDAC_NUM 10 for an even-more-NOS-like experience: 😍

Distortions are now splattered all the way out to >300kHz, baby! In listening sessions with this setting, it sounded good with maybe a subjective accentuation of the bass presence and sense of fullness. Interesting. As expressed in the past, distortions can be pleasant and I have no issues with some preferring added distortions ("euphonophilia") over strict technical accuracy.

You might even be able to combine this with different filters like DAC1 "Min phase slow" for subtle variations (I didn't personally try). Ivan tells me that parameters like these IDAC_NUM values are undocumented. Have fun.

I hope you're enjoying the music, dear audiophiles! The spring season is almost here; how time flies...