REVIEW: E1DA #9038D6K Dongle DAC / Headphone Amp (Part I) - PCM DAC performance. And (potentially) higher dynamic range in Spatial Audio 2-channel fold-downs.

Little E1DA DAC to Sennheiser HD800 headphones via 1/4" adaptor... Looks ridiculous if walking around in public with this! But sounds great. ;-)

A few weeks back, I received a package from Ivan Khlyupin (IVX) of E1DA containing both the Cosmos Scaler and the device we'll be looking at today - the E1DA #9038D6K DAC / headphone amplifier.

This is the latest iteration of the #9038D models which feature 3.5mm single-ended output, to be distinguished from the #9038S models which are balanced devices with 2.5mm output jacks. And the "6K" refers to 6000μF (!) of power rail filtering (original #9038D has 4000μF).

As the name implies, internally, these DACs are based on the ESS ES9038Q2M low-power chips. They are capable of playing back PCM to 32/384 as well as up to DSD256. Over the last year or two, we've seen a proliferation of DACs based on this component and I would say that overall, the performance on these devices have in general been very good. Thanks to the low power consumption, we can now have mobile converters that perform at least to the resolution of many desktop DACs while being fully USB-powered.

To generate the amplified headphone output, op-amps used include the OPA1622 and OPA1602 based on the E1DA specs. To achieve very high resolution, internal components are advertised to be quality Susumu resistors, Murata C0G caps, Kyocera AVX tantalum caps, and connectors from Taiwan. In my discussions with Ivan, I know in the last while it has been slow for him to source some of these components and at times difficult to avoid fake parts out there!

Let's have a look at the little DAC:

We can see both the USB-C and 3.5mm phono sides. Notice on the USB-C side there's a small LED that turns on when plugged in.

The enclosure has the typical look of other E1DA devices. Sandblasted black aluminum with laser-engraved lettering. Dimensions are a minuscule 48 L x 22 W x 9mm D as shown on the box, weighing around 12gm (<0.5oz). The device connections are simple, just a USB-C female connector on one end, and a standard 3.5mm stereo headphone jack the other. Other than that small LED on the USB-C side, there is otherwise no physical buttons or other indicators.

It came in a simple cardboard box as in the picture with no accessories. You might be able to purchase packages with included USB cable. (A very short 6" USB-C to USB-C would be great for cell phone use, USB-C to USB-A might be needed if plugging into a streamer or computer.

E1DA #9038D6K with CCA C12 IEM; this would be a more realistic DAC/headphone pairing when casually walking around town as opposed to the HD800 above! Note the flashing orange LED when connected and playing audio.

As usual, I put the little DAC through its paces by listening to the device for a week, then transfer it to my test bench to get some measured results before having a listen again afterwards. I will reserve comments on subjective matters until later on (Part II) in this review series.

I. Oscilloscope, Digital Filter, Impulse Response, Output Impedance...

While there is an Android app called TWEAK9038 (US$10) which can adjust a number of settings we'll look at in Part II, given that there are no adjustment buttons or other selectors on the unit, let's measure the device in its default state and see how this performs for a new owner.

Starting then with the usual 0dBFS 1kHz sine and -3dB bandlimited and unlimited square waves:

At 0dBFS, notice output level hits 2.84Vrms. Channel balance looks great with basically a perfect overlay of left and right channels (sine waveform).

Minimum phase filtering is used at 44.1kHz without pre-ringing on the non-bandlimited square wave. Here's the square waveform at 96kHz:

Interesting. At 96kHz, the filter has changed to a linear phase one (presence of pre-ringing). This tells us that the DAC has the ability to adjust the filter setting depending on the samplerate - one of the tweaks one can play with to be discussed later.

As you can see, with the increased bandwidth at 96kHz and linear phase filter (no group delay), the transient edge of the square wave is significantly steeper.

Here's the "Digital Filter Composite" graph and the impulse response:

I've updated the look of the DFC graphs, using 1/48-octave smoothing for the white noise and noise floor graphs, but left the 19 + 20kHz graph unsmoothed for obvious reason. Note that the DAC output was reduced to -0.7dB to better capture filter performance as we will discussed below.

Indeed, at 44.1kHz, we see a minimum phase, steep filter which does allow a little bit of "aliasing"/imaging beyond the ~22.05kHz cursor. Intersample overloading happens with high amplitude signals as seen in the white noise tracing with 0dBFS peaks which as expected improves when taking the signal down by -4dB.

Since this little DAC is intended to be primarily used as a mobile headphone amp, let's have a look at the output impedance:

A slight increase in impedance at the 20Hz and 20kHz extremes.

Excellent, flat, low impedance across the audible spectrum. The E1DA specs rates the output impedance at 0.07Ω @1kHz. On my test system, I can confirm that it's below 0.09Ω average and indeed 0.071Ω at 1kHz.

With such low output impedance, it should be highly "load independent" and we should see this in the frequency response:

That's exactly what we see. Whether it's one of the resistive loads like my standard 20/75/560Ω, or the Polk Ultrafit 2000 headphone with highly variable impedance, the frequency response does not change (I had to separate the traces above in order to not end up with a single line). Compare the above to the Chord Mojo for example - as you can see, I did not need to separate the traces given the clear variation in frequency responses:

II. Unloaded DAC PCM performance

How good is this little DAC resolution-wise compared to others?

To conduct most of these tests, I'll be using my Beelink SER4 as measurement computer → Topping HS01 (USB isolator) → E1DA #9038D6K DAC → phono cable → E1DA Cosmos Scaler → Cosmos ADC → Beelink computer. You'll see use of RightMark (6.4.5 Pro) as well as Room EQ Wizard software. A quick testing note, I found compatibility issues with using both the Cosmos ADC and #9038D6K DAC's ComTrue drivers. On my testing machine, I uninstalled the software and just used Microsoft's default UAC2 drivers and WASAPI which worked fine.

Let's start with some comparison of devices using the RightMark battery...

RightMark 16-bit/44.1kHz:

As usual, we start with the most common frequency rate and bit-depth - 16/44.1, CD specifications. This is "standard" lossless resolution:

The results and comparisons are not surprising. Almost any decent DAC these days will be able to perform 16/44.1 conversion with minimum issues and typically DACs that can perform at "hi-res" levels will show similar results. This is simply because of a "ceiling effect" for 16/44.1 not being much of a challenge for good high-fidelity devices these days. IMO, in a blinded, volume controlled listening test, I believe most reasonable DACs will sound the same with 16/44.1 content as suggested a few years back.

In the summary and graphs above, the most "interesting" device was the Cayin multibit R2R dongle DAC which clearly did not perform as well as the others. As for the E1DA #9038D6K, it clearly performs as a high-resolution DAC should, basically as good as others, including very high performance desktop DACs like the Topping D90SE and Sabaj A20d-2022.

Notice the rippling frequency response of the Topping D10s using its default "apodizing" filter. Although it looks odd, the resolution of the measurements are such that the less than +/-0.1dB fluctuation looks significant, but in use, the ears would likely not be able to detect this especially at the higher frequencies.

RightMark 24/96:

Time to get serious with some hi-res measurements:

We continue to see the pattern above with the Cayin dongle R2R DAC performing significantly worse than the others. As you can see, the E1DA #9038D6K is a very impressive little DAC, capable of fidelity essentially equivalent to top-notch desktop devices like the Topping D90SE and Sabaj A20d-2022. I had to zoom into the noise floor graph to show that the #9038D6K measured equivalent to these.

RightMark 24/192 & 24/384:

While modern DACs can readily do 192kHz samplerate these days, I personally rarely keep 192kHz content anymore. Nonetheless, a well-engineered device should continue to perform well and it's good to check.

The data continues to look very nice. Although the RightMark software is a bit buggy so I can't show the crosstalk and IMD+N sweep nicely, we can see that frequency response extends beyond 50kHz (about -1.75dB @50kHz) using the default filter setting, and the noise floor is very quiet out to around 100kHz. Again I'm zooming into the noise floor which makes the Cayin RU6 R2R look horrible in comparison even though the noise level within the audible frequencies are below -100dBFS!

Just because I can ;-), at the far right, there's the E1DA DAC's 24/384 result showing excellent performance has been maintained with this even higher samplerate. This is in practice only of interest for folks who use software like HQPlayer to upsample.

THD+N & Stepped Sine:

Let's move on and take a reading of the highest THD+N / SINAD this DAC is able to achieve. This was found around -0.7dBFS which correlates to 2.6Vrms:

Indeed as advertised, it's hitting THD+N of -120dB or thereabouts, and unweighted dynamic range of close to 124dB.

Be mindful that at 0dBFS there's a bit of overload, with increased distortion as seen here:

An increase up to THD+N of -96dB (0.0016%) above -0.5dB output level is not a big deal and I highly doubt anyone would practically notice beyond running measurements like these. Nonetheless, feel free to drop the max output level a little bit (like down to 97/100 in Windows) if you want to avoid it.

By the way, I see that this distortion at 0dBFS is less severe than the 88dB SINAD reported in the ASR measurement of the previous version of the #9038D.

Next, let's have a look at the harmonic distortion components across the audible frequency from 20Hz to 20kHz at various output levels, let's use -0.5, -3, -6 and -12dBFS:

Looks good. Across the frequencies, we're not seeing any surprising spurious irregularities in harmonic distortion and the results here typical of ESS DACs. Notice the 3rd order harmonic tends to predominate on the -0.5, -3dBFS and -6dBFS levels although the 2nd harmonic likes to poke up ~400Hz. By -12dBFS, harmonics are all low with the 2nd minimally predominant.

Here's the 1kHz output level stepped sine graph:

Generally clean except for that saturation in the highest 0.5dB or so. We see a slight rise in THD around 100mV, but I wouldn't worry about such a small potential ESS "hump" up to -100dB THD. As discussed before, ESS DACs, at least the current HyperStream II models, tend to show fluctuations especially in the odd harmonics across the amplitudes.

Using the detailed step data above here's the calculated output level linearity:

Linearity is equivalent in both channels (left one shown). Less than 0.5dB deviation down to -125dBFS and less then 0.1dB deviance down to -115dBFS. Excellent, especially considering that this has a single-ended phono jack. I trust "Golden Ear" audiophiles would not complain.

Triple-Tone TD+N and 1/10-Decade Multitone 32:

Okay, let's end off this section with a couple of multi-tone tests to see how cleanly this DAC performs with more complex signals.

The Triple-Tone TD+N is quite clean with harmonic and intermodulation products down at -120dB below the primary tones. TD+N score of -114dB is simply excellent. Compare this to the Chord Mojo scoring around -104.5dB, or the remarkably poor Cayin RU6 at only -56dB! Again, this little DAC is closer to a desktop device like the Sabaj A20d-2022 which scored -111dB with the RCA output (improving to -118dB however with XLR out).

Likewise the 1/10-Decade Multitone 32 complex signal (running at 24/96, 1M points FFT) looks very good. No noise or distortion down to -110dB below the tonal peaks. Better than other mobile dongle DACs I've come across.

III. Jitter

Looks beautiful guys and gals. These are essentially ideal Dunn Jitter Tests FFTs. We can easily see the normal jitter-modulation signal in the 16-bit test.

I see that more recently, ASR measurements have looked at the 24-bit J-Test FFT down to 250Hz where the LSB "square" wave modulation tone is, let's do the same:

There you go. Notice how low the noise level is as we pick up even that tiny undithered 24th-bit 250Hz toggle down at -140dB (with lower level odd-harmonics).

And to push the testing even further, let's accelerate the J-Test all the way to "quad" 192kHz and see how well the little DAC manages to remain "sideband free":

As indicated, this is an insane "accelerated" J-Test which nobody does ;-). The primary signal is up at 48kHz, and although we see a pair of sidebands, they're totally irrelevant at -130dB, not to mention ultrasonic. A torture test not just of the DAC, but also the measurement system.

Seriously folks, with good modern DACs in the last few years, temporal "jitter" anomalies are not an issue at all... As if jitter was ever all that audible unless very severe.

An evening at the "test bench"... ;-)

IV. Summary of Part I...

So far, what I've presented of the E1DA #9038D6K DAC shows that this is a highly accurate DAC. In the default state, it uses a typical ESS minimum phase steep filter for 44.1/48kHz playback, and can output up to 2.84Vrms unloaded at 0dBFS. As a DAC, the RightMark results show a level of resolution on par with some of the best desktop converters. This is confirmed with tests like the 1kHz THD+N showing that even a tiny dongle DAC can now achieve a superb -120dB! The only caveat being the increased distortions above -0.5dBFS. Even so, this latest #9038D6K (2022) model seems to be better than the previous version with less distortion at 0dBFS.

This excellent performance extends into the frequency and level stepped sine tests. Jitter is essentially non-existent. Multitone tests like my triple-tone TD+N (with a superb result of -114dB) continue to confirm the findings. While we will delve into the device more as a headphone amplifier next time, already we can see that output impedance is supremely low with an average across seven measurements within the audible spectrum of <0.1Ω.

In most measured analogue output characteristics, this device surpasses something much more expensive like the Chord Mojo 1 or the IMO overpriced Audioquest Dragonfly dongles (I'm especially disappointed in the Dragonfly Cobalt despite the hype over the years). It's also head and shoulders better as a hi-res converter compared to something like the Cayin RU6! As usual, while I can say this as a statement of objective reality, it doesn't mean subjective preference has to correlate with the higher fidelity findings for any one individual.

Alright friends, that's quite a bit of data to process! Let's end here for Part I. We'll discuss more next time with both objective data (headphone amp performance, DSD) and subjective impressions in Part II.

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

Recently, I was contacted by a reader Jeremy about his look at Apple Music Spatial Audio fold-down of playback for 2-channel headphone listening (the "virtualized" surround sound). He made a video about it here using his Mac and BlackHole to route the digital output for capturing:

It's encouraging to actually see the waveforms of these Atmos tracks showing improved dynamic range, potential euphonic changes in tonality (deeper bass in some tracks), and also an interesting look under the hood at the files downloaded from Apple Music when streaming.

To me, Atmos/Spatial Audio vs. "hi-res" is analogous to the HDR feature with 4K. While the extra resolution at 24/96 and 4K might be appreciable with some of the best albums and movies, often the improved resolution is not significant unless one has fantastic equipment (like a larger screen TV, very resolving speakers/headphones, top notch room) and of course excellent eyes and ears. The moment you add HDR to the 4K image, or multichannel/Atmos to the audio signal, anyone can see and hear the difference clearly even without an A/B test to accentuate the nuances.

Let's hope that going forward the music production folks continue to build on the momentum of Spatial Audio/Atmos and create even more compelling content. This is how we "make music dynamic again" ;-). Maybe add a "Loudness" button in software for folks who want dynamic compression when listening and walking around town or on the bus so nobody complains that the Spatial Audio downmix is too soft!

Thanks Jeremy for the tip about Vera Blue and her albums Perennial (2017, DR5) and Mercurial (2022, DR5)!

I'm curious if anyone has also done a capture with multichannel output using BlackHole beyond the 2-channel headphone processing on the Mac?

Happy December everyone! Have fun as we go forward into the Christmas and New Years holiday. The older I get, how seemingly quicker the years go by...