Sunday 11 December 2022

REVIEW: E1DA #9038D6K Dongle DAC / Headphone Amp (Part II): Headphone Amp Power, DSD, Subjective Impressions, AMPT & Perfectionistic Tweaking!

A great sounding music listening station! (Samsung Galaxy Tablet, Drop+HiFiMan HE-4XX, classic Jacky Cheung cantopop.)

Okay friends, today we continue with Part II of the discussion on E1DA's #9038D6K portable DAC/headphone amp. Please refer to Part I last time for the general overview of DAC performance which already tells us about the PCM resolution of the device. The data from last time shows that for the price and size, this unassuming dongle is one heck of an accurate, high-fidelity, high-resolution digital-analogue converter.

This time, let's move forward with exploring other areas not yet discussed. Specifically, let's focus on resolution when under headphone-like loads, look at DSD performance characteristics, have a listen to the DAC, and given that E1DA is known to be tweaker/DIY-friendly (including making devices like the Cosmos line), let's also talk about what I would consider more "extreme" tweaking for the perfectionist audiophiles!

I. Headphone amplifier performance (and 5V USB current demand)

As a reminder from last time, this little DAC/headphone amp has excellent very low output impedance:

I've shown (Part I, Section I) how this low-impedance curve makes the frequency response highly "invariant" of different loads including headphone impedance fluctuations.

Let's now go further and examine the distortion characteristics and maximum power (to 1% distortion) we can expect from this little dongle. As usual, I've standardized on 20Ω, 75Ω, and 560Ω as representative resistive loads for low, medium, and high impedance headphone tiers.

For higher resistance/impedance loads, at least from 75Ω onward, the headphone amp is able to output up to 2.8Vrms without hitting 1% clipping. Into a low resistance 20Ω load, maximum power to 1% distortion is right at 200mW. This is consistent with the official E1DA spec of 180mW into 32Ω, also measured at 1kHz. Compared to others dongle DACs I've tested, for low impedance loads, you'll be able to get a bit more power from something like the iFi GO Bar at around 300mW into 20Ω (plus has balanced out, higher voltage ability good for higher impedance, less sensitive headphones). The Dragonfly Cobalt is less powerful at around 60mW <1% distortion into 20Ω.

Let's have a look at signal resolution like the 48/960/5472Hz Triple-Tone TD+N at 0.5V into 20Ω which has been a "standard" measurement I've use here for device comparisons:

A TD+N of -96.4dB (0.5V into 20Ω) is a good result and close to the -98dB of the Sabaj A20d-2022's headphone output. And even better than the Drop + THX AAA 789's -93dB; although obviously that amplifier can provide more power. Among mobile USB dongles, the #9038D6K is significantly better than the iFi GO Bar with a result of around -75dB.

And here's the 1/10-Decade Multitone 32, 0.5Vrms output into 20Ω, both channels driven:

As you can see, we've picked up a 60Hz hum + 120Hz harmonic through the resistive load/cabling. If we ignore those, there's more than 105dB of distortion-free range from the multitone peaks. In comparison the iFi GO Bar achieved an 85dB range with the same set-up.

In terms of the 5V USB current demand, with 20Ω load connected, idle current (as measured with USB tester) is at 0.11A, this increases to 0.18A when playing a 0.5Vrms 1kHz tone, 0.23A at 1Vrms, and 0.28A at 1.5Vrms 1kHz. If nothing is playing for 60 seconds, you'll hear a soft click as it enters power-saving mode down to 0.04A, nice. Over the years I haven't been consistent with these power measurements so will try to aim at doing this in a more standard fashion ahead. That 0.11A idle current is similar to the old SMSL iDEA and more recent iFi GO Bar, but higher than the 0.06A of the Dragonfly Cobalt.

II. DSD Playback

I think most new DACs these days are able to decode DSD. While it has become part of many company's default specs, I wonder how many listeners actually use this feature. For completeness, let's see some DSD performance numbers and graphs. The #9038D6K is able to play up to DSD256. For these tests, I used my Samsung Galaxy Tab S6 Lite (Android 12) running USB Audio Player PRO which worked well. For some reason I could not get DSD256 playing with the ComTrue Windows ASIO driver (using Foobar2000 or JRiver) nor my Raspberry Pi 4 "Touch" running current version of Volumio.

With the RightMark PCM 32/96 test signal converted to DSD64/128/256 in SoX (as per my standard), let's compare with PCM 24/96 playback:

As you can see, the PCM results give us the best dynamic range, and lowest distortion numbers. This pattern is common with many DACs. Other devices including otherwise very high resolution ones like the Topping D90SE also perform very well in PCM but dip a bit with DSD. The first 3 measurements on the left are the default DSD64, 128, and 256 numbers. We can see that in terms of dynamic range and distortion values, DSD128 looks like it's the "sweet spot" of DSD performance for this DAC.

To the far right, I'm giving a little preview here of one of the "tweaks" you can do with the Tweak9038 App. Compared to the DSD64 measurement on the far left, I changed the MCLK (Master Clock) setting from 25MHz to 50MHz which improved the measured performance mainly by dropping the noise floor. We'll talk more of this later.

We can see this disparity between DSD and PCM performance in the overlay graphs:

For these measurements, the cables/connections were exactly the same. Interesting change in Stereo Crosstalk curves between the different modes and PCM vs. DSD.

We can see a rather striking difference in the noise floor of the Dynamic Range graph and also the lower IMD+N sweep when playing PCM content.

Okay, let's have a peek at the 1kHz +3.1dBDSD THD+N graphs at the various DSD rates using the test signals I posted here:

Note the ultrasonic noise shaped content out to 192kHz (384kHz sampling rate for the Cosmos ADC) as expected. For comparison, here's the -2.9dBFS 1kHz 24/96 PCM playback of equivalent amplitude (Cosmos APU notch not used), and as a bonus, since DSD512 is not supported natively, let's play it through USB Audio Player PRO converted from DSD512 to PCM in realtime:

As expected, we see the contrast between DSD and PCM - lower noise level in the audible frequencies and absent quantization error ultrasonic noise with PCM. I think it's impressive these days how there's enough processing power in even a mobile device like the inexpensive Samsung tablet to be able to do quite a good job with realtime DSD→PCM conversion.

I've suggested years ago that I believe accurate reproduction of higher samplerate DSD like DSD256 and DSD512 approaches the quality of good high-res PCM. I personally don't hear anything special about DSD to suggest that it sounds more "real" despite entertaining high-profile claimsDSD64 is a noisier data stream with the ultrasonic noise rising quite quickly slightly beyond 20kHz as seen in the measurements above and I suspect this may have a euphonic effect for some listeners (DSD lovers probably would reject such ideas as DSD256 sounding like good PCM 24/192 regardless ;-).

While the native DSD measured results are lower, note that this is still "hi-res" with better than 16-bit performance in the audible spectrum.

Finally, a look at the complex 1/10-Decade Multitone 32 through DSD:

Looks great. DSD64 and 128 play back cleanly although with DSD256, we're seeing a little bit of distortion creeping into the base of the 100Hz to 2kHz tones, but still below -95dB. Left and right channels perform equivalently and are well balanced throughout all these tests.

As suggested above, just as there are audiophiles and sales folks out there who swear by the sound of DSD, we know that there are designers who feel that the core of their products somehow must incorporate a 1-bit SDM phase to differentiate what they're doing. Examples of this would be Ed Meitner who converted all PCM into DSD128 for awhile back in the day with the MA-1 DAC but these days uses higher DSD rates. Another would be Ted Smith and PS Audio's love of DSD which now extends to Octave Records; for example, PS Audio's PerfectWave DirectStream DAC did some unique calisthenics where they upsampled everything to 30-bit/28.2MHz then downsampled it all into 1-bit DSD128(!!!). The PS Audio DAC did not measure particularly well with rather high noise, but I'm sure it can sound good just like this #9038D6K playing DSD despite being objectively noisier than PCM. There's also Playback Designs' FPGA, SDM-based DACs that upsample to presumably 1-bit/50MHz championed by Andreas Koch that I'm sure sounds great. I would love to see measurements of the noise floor in their modern MPD-8 DAC since the older MPS-5 SACD/CD player was also quite noisy.

I asked Ivan (IVX) about the reason for the higher DSD noise in this DAC, here's a bit of our conversation:
"Higher DSD has more and more THD+N and MCLK needs to be higher and higher. If you switch MCLK to 25MHz, only DSD64 will work. Hence, higher DSD needs a max MCLK 100MHz to work better. Also, not a bad idea to keep physical opamps BW 50kHz due to lots of HF noise of DSD mode.

I had no chance to use 100MHz in a compact DAC because I used the internal ES9038Q2M oscillator + good quality crystal 52MHz. That solution provided the best jitter performance regardless of cost. Yeah, ES9038 contains a wonderful low-phase-noise oscillator circuit on board. 100Mhz crystal (i.e. 33.3(3)MHz at 3rd harmonic) would work there only in overtone mode i.e. more complex schematic + not sure regarding the jitter. So, I have a perfect PCM performance for reasonable money and tiny-sized housing ;)

BTW, I remember discussion, when I asked folk why nobody tried ES9038 internal OSC, and why we believe if ESS designers created a good DAC but not a low-noise OSC? "No way", you should use gold-plated Crystek for $50 to be happy, they said!
Finally, I used a HK company's crystal for $0.15 and the jitter is negligible."
So basically, we're talking about compromises here between small size, cost, and the fact that DSD ultrasonic noise can affect performance of the board.

That last paragraph is great! Often, we see in audiophile discussions and in magazines all kinds of statements about the "need" to spend lots of money on expensive parts seemingly based on sentimental pet theories around certain brand names like Crystek oscillators or beliefs that even an audio streamer computer benefits from OCXO clocks (previously discussed here). As we saw last time in Part I, the #9038D6K achieves fantastic low-jitter performance without anything fancy or expensive. Remember audiophiles, when you hear of flowery claims, make sure you "stay grounded" and ask about the evidence especially with "High End" expensive products and the attribution theories a company might propose for why their "special" designs are better than others. I suspect most of the time, companies will have nothing to offer you except to ask you to "just listen". This is fine, but these days, since so much of modern equipment already sounds great, that's not evidence that they've done any unique research to justify claims that often are just used to push MSRP up.

Apart from no-nonsense audio hardware designers like Ivan, you're not going to get this kind of "inside info" from the usual company sales reps nor mainstream audio magazines, nor YouTube/website promoters creating infomercials that typically will tout some kind of "Higher MSRP is better sound!" ideology.

III. Subjective

I've been listening with this little DAC using a few of my wired headphones over the last 2 weeks - Sennheiser HD800AKG Q701Drop+HiFiMan HE-4XX planar (top image), even the modded Dekoni Blue. While I did use it with my IEMs (like the CCA C12 and Etymotic ER-4B), these ultrasensitive low-power transducers were simply not particularly "challenging" for the #9038D6K to drive! Suffice it to say that this mobile DAC proved to be very low noise with these sensitive IEMs using my few-years-old Huawei P30 Pro phone. I noticed the "Warning: 9038D contains a high-sensitive I/V stage that may cause audible noise when your phone uses LTE  4G, 5g." on the E1DA website. I did not hear any noise issues on my end.

With full-sized headphones, what's clear is that this device provides a decent amount of power for such a small package connected to a phone/tablet. While I would not be able to turn the volume up with the low-sensitivity Dekoni Blue (based on Fostex T50RP Mk3) to the same amount as my desktop Drop+THX AAA 789 headphone amp, this dongle has enough oomph to drive these headphones to louder than my usual listening levels at least.

Even before looking at the objective data, I could tell the little DAC sounded very "precise" (unlike something like the Cayin RU6 which was audibly non-transparent). Frequency reproduction across the audible range sounded excellent and the quality of reproduction basically tracked the limitations of the headphones paired with the DAC. For example, the Sennheiser HD800 can be a bit harsh with treble accentuation on some music such as the very nice synth score of Cyberpunk 2077 (2020, DR7) - have a listen to "V". But on other headphones like the Drop+HiFiMan HE-4XX, the mellower sound signature was more pleasant with longer listening sessions.

In contrast, I enjoy the "airy" presentation of the Sennheiser HD800 more when listening to acoustic music like say Trio Mediæval's SOLACIUM (2021, DR13) than a thicker, "meatier" sound of the Dekoni Blue.

As the name of the group implies, this is mostly traditional choral music sung in reverberant rooms/halls with delicate overlap of vocals. Plenty of detailed, complex harmonics and nuances of the human voice on display in the recording if you're interested in this genre. Good rendering of the soundstage and the impression of the space in which the recording was done. Nice rendering of foreground lead voices separated from the background harmonies as on the track "Abba, hjärtans Fader god".

I have heard many good things over the last few months about the recent Pink Floyd Animals remix (James Guthrie's 2018 mix, DR12). While I don't consider this album as enjoyable as Dark Side... or The Wall, the new stereo mix is pretty good with cleaned up instrumentation (especially the percussion brought forward with better-defined bass) and clearer vocals. Sound effects like the barking of the dogs around 9:30 of "Dogs" sound clearer, a bit brighter, and a greater sense of 3D spatial depth compared to the original CD release.

As a multichannel music lover though, I would still very much encourage audiophiles to experience the 5.1 version on the Bluray!

Harry Connick Jr's recent album Alone With My Faith (2021, DR9) is an interesting melting pot of poppy, jazz-tinged, traditional hymns and new material, recorded during the pandemic (as you might have expected of the title). Check out "Amazing Grace" with its foundation of piano and solo voice built up over the verses with choir, percussion, and brass accompaniment. Great use of the Hammond B-3 organ in "The Old Rugged Cross".

I don't think I ever imagined I would hear an album with Harry singing "Panis Angelicus". I usually think of recordings like Charlotte Church back in the day...

Finally, let's turn the volume up and play some rock - Lenny Kravitz's Greatest Hits (2000, DR8). Great adrenalized guitars on hits like "Are You Gonna Go My Way", and "Fly Away". The #9038D6K effortlessly reproduced the soundstage with guitars panned wide, percussion and vocals layered nicely around the center image on "Always On The Run", handling the dynamic compression with grace. Likewise on the slower tunes like "Believe", there's good layering and clarity of the studio multitrack mix as it highlights the intentionally distorted lead vocals. Great bass depth on the techno-dance-inspired "Black Velveteen".

Other than through headphones either on my smartphone or the Samsung tablet in the top picture, I was also listening to some of the albums through the Raspberry Pi 4 "Touch" streamer running RoPieee in my main sound system:

E1DA #9038D6K hanging off Raspberry Pi 4 streamer at the back going into main system.

While the DAC doesn't quite look "right" integrated into a large audio system like this, don't be biased by appearances as if that determines audio performance! It sounds just as good as many of the more expensive DACs and a simple Raspberry Pi streamer + high quality DAC like this would be a great "starter" system until you buy something that might look more fitting for full-sized hi-fi. ;-)

IV. Your turn... Archimago's Musical Performance Track (AMPT)

It's your turn should you wish to accept the challenge ;-).

Have a listen to a hi-res 24/96 capture direct from the E1DA #9038D6K:

E1DA #9038D6K DAC/Headphone Amp (ESS ES9038Q2M, RCA) AMPT

The AMPT recording above was created based on the standard procedure posted here. As usual, copyrighted short segments have been utilized on the basis of "fair use" for the purpose of education, research, and public testing.

V. Summary

Overall then, what I hear is basically what I've come to expect from E1DA; the #9038D6K is a USB-powered dongle headphone DAC built upon well-engineered high-resolution performance. Subjectively, I have nothing to complain of and would happily suggest that in the majority of situations, this little device would sound as good as much larger, more expensive DACs. It has enough power for most typical headphones with up to 2.8Vrms into high impedance loads, and up to ~200mW into 20Ω for those low-impedance devices like modern IEMs.

Having said this, detailed measurements tell us to be mindful of higher distortion when played loud above -0.5dBFS as discussed in Part I - for headphone users, be mindful of hearing loss if one is routinely listening so loud as to push a device like this up that far! And as discussed above, DSD playback is significantly noisier than its PCM performance, so I would not suggest this dongle if highest resolution DSD playback is essential for your needs. Despite the clear objective difference, I would argue that the DSD performance is still more than "good enough" for mobile listening if you happen to have some DSD material on your device.

This is as "transparent" a mobile DAC (playing PCM) as I have ever heard or measured to this point. The portable USB dongle-DAC market has grown substantially with all kinds of devices over the last few years. At least in the "audiophile" segment, clearly we have proceeded far beyond early models like any of the Audioquest Dragonfly devices. These days, USB dongles can sound significantly cleaner, quieter with more power to drive headphones. At the current price of around US$130 at a time when inflation is still high and pandemic-related supply chain issues still unresolved, the E1DA #9038D6K IMO is good value considering that something like the Dragonfly Cobalt is still almost twice the price.

For those who like to see what the innards look like as a sign of quality, here are a couple of pictures I received from Ivan of the production PCBs you might find interesting:

Four Kyocera AVX conductive polymer capacitors (not cheap!) on the underside. Each 1500µF.

And yet, we're still not done with this article! ;-)

VI. Bonus: Tweak9038 Android App - deeper perfectionistic tweaking!

While measurements were collected for the most part in its default state, a complete review of the #9038D6K demanded that I discuss the fact that settings can be further controlled with the Tweak9038 application for Android devices. This app costs US$10 in the store, is not necessary for music playback, but if you have a geeky, perfectionistic curiosity, it's very useful!

If there's one thing I've learned over the last couple years connecting with Ivan of E1DA, it's that this guy is incessantly "hardcore" when it comes to squeezing out performance at a budget! ;-)

Here's a look at the main Tweak9038 screens. Most of the features are self-evident. You can also see the default settings for the early production #9038D6K being reviewed:

Main screen, the default "Master Clock" and "Filter" settings on the early production unit I'm using.

From the main screen we can see that there are buttons to adjust harmonic distortion compensation parameters ("Thd") for the DAC, select the master clock setting for each samplerate - both PCM and DSD, and adjust which digital filter you want the DAC to use for each PCM samplerate. We saw already in Part I that at 44.1kHz, a minimum phase setting was used, but changes to linear phase at 96kHz; now you see the actual settings under the hood.

There's a volume controller up at the top of the main screen along with adjustable "Max" and "Min" limits. For example, in Part I when we were looking at maximum resolution, we know that there's a bit of saturation above -0.5dBFS, so we can click on the "0.0dB" and pull it down to "-0.5dB" as a limit if we want to use this as a pure DAC and want to avoid the risk of that extra distortion up there. By the way, pulling this down to "-3.0dB" would limit the max output to 2Vrms which would be a typical "line level" voltage like many DACs. Likewise, -127.5dB is a very large amount of attenuation; feel free to change this to something like -100dB or -75dB if you want a convenient and practical number.

When you change these settings, they are automatically saved on the little DAC's non-volatile memory so you can disconnect it from the Android device and use it elsewhere. Settings can be saved and there are also "official" presets you can default to and the ability to import settings as well so don't worry about screwing things up. In the "Settings" menu (not shown), there's also a factory reset option to get you back to a default state.

For filter setting, I'm generally a "Linear Phase, Fast" guy (unless I have the option to play with nuances of course) and prefer steep "brickwall" type Chord-like filters - maybe not needing to be *that* steep. I'd like to think that in 2022,  rational audiophiles aren't afraid of impulse response "pre-ringing" and appreciate the benefit of linear phase filters for DAC playback by not causing group delay in the time domain, unlike say MQA which still touts "temporal blurring" benefits despite the lack of evidence.

As for the MCLK settings, this will affect noise floor and harmonics and you can tweak for each samplerate. As an illustration, here's what changing MCLK does to the +3.1dBDSD 1kHz DSD64 THD+N measurement:

As you can see, running the MCLK higher up to 50MHz with DSD64 drops the noise level, but low-level discrete non-harmonic noise peaks are accentuated so there's a balance one should consider. I assume this is why E1DA defaulted to the 25MHz setting despite 50MHz achieving a better measured THD+N.

Compared to master clock speed and digital filter settings which are fairly straight forward to try, what takes a bit more effort are the "Thd" and "THD Edit" screens:

Like the tweaking I showed with the Cosmos ADC earlier this year, these settings change the coefficients fed to the ES9038Q2M chip for 2nd and 3rd harmonic distortion compensation. E1DA tests and optimizes these values for each DAC to perform well with a generic 32Ω load so by default the performance is already excellent. The Tweak9038 app simply gives the end user the power to optimize further for your specific DAC + headphone pairing.

Notice the ability to change the parameters for 3 amplitude "Ranges" based on the lower and higher threshold settings.

Ivan/E1DA has a video showing how this is done using the Load Board (US$20, detailed pictures here):

As you can see in the video, the harmonic distortion tweaking is not hard to do but one must have the prerequisite load and ADC measurement system hooked up. Admittedly, I don't expect most users would have the time or inclination to do this - only the most perfectionistic and obsessive. ;-)

Although I have the Load Board here, I don't have the necessary cables to connect the board output to the ADC for the time being. Let me instead give it a try with my homemade headphone load jig. Here's a picture of what I did on my test table with the AKG Q701 headphone:

Before getting started with just changing values in Tweak9038, we need a baseline. Here's what the THD vs. step level graph looks like in this #9038D6K + AKG Q701 "system" with default factory settings:

Notice that I selected the three Ranges to partition the graph a bit so I can customize parameters in each output region. Notice that the 2nd and 3rd harmonics are very low already below -16dBFS so there's not much tweaking to be done down there; best to just leave well enough alone!

As shown in the video above, the user can adjust the 2nd and 3rd harmonic compensation settings to achieve the lowest distortion. In my case, I'm running a clean 1kHz 0dBFS (24/96) tone in USB Audio Player PRO on the tablet through the #9038D6K DAC into the Cosmos ADC, monitoring REW RTA in realtime, using Tweak9038's volume control to target the output level to tweak. Then play with the harmonic compensation sliders to make sure the best result is achieved; it's often easy to over/under-shoot the optimal setting. This probably sounds more complicated than it is in practice.

With a bit of time, trial and error, here are the final tweak settings I settled on:

Set ranges to <-16dBFS, -16 to -8dBFS, and >-8dBFS.

Here are the before-and-after optimization 1kHz THD vs. level step graphs:

Notice that the differences are subtle below -16dBFS but you can see it if you compare the black THD line in relation to the dotted noise floor. Much more significant differences can be seen at higher output levels above -16dBFS. We can better appreciate the change if we overlay the THD graphs. Here it is for both right and left channels:

The left channel measured a little noisier than the right, likely the result of headphone driver idiosyncrasies. As noted above, differences are minimal below -16dBFS. However, above -16dBFS, the tweaked 2nd and 3rd harmonic compensation settings have resulted in a -10dB improvement in THD by around -6dBFS, benefiting both channels. That's a pretty neat trick! Notice the clipping above -1dBFS as my AKG Q701 has an average impedance around 60Ω. Obviously, different headphone loads can result in variable patterns.

This kind of "power" for end-users to implement their own tweaks and optimizations, catered to the users' hardware, is I think an opportunity to go forward with achieving "personalized high-fidelity playback". While we can already use EQ to change frequency response, and DSP to correct both frequency and time-domain issues, this is an interesting glimpse at a way to minimize distortion amounts. Having said this, and while having the ability to show the effect objectively, don't ask me to blind-test the difference. ;-) I'll leave that to the headphone "Golden Ears" to give this a try and discuss potential benefits!

As usual, visit the E1DA Discord channel for more discussions on their products.

Have fun tweaking away friends. Just make sure to enjoy the music along the way... ;-)

Wishing you and yours a wonderful Advent Season as well!


  1. Hi Archimago. As always thanks for the measurements and also for some music recommendations ;)
    I wanted to ask if you encounter any problems with cellular noise (or other) when using e1da with phone?

    1. Hi Milan,
      I see there is a warning but I have not noticed a problem with my phone. I'm still on a 4G network and use a short USB cable (something like 3" I think) between the phone and #9038D6K. Even with IEMs have not had an issue.

    2. Thanks, I'm also still on 4G. Don't need anything faster as it's more than enough for a phone usage ;)

    3. Yup, I agree Milan,
      I got my wife her new iPhone 14 this year which is 5G but can't say I noticed anything special for her or my needs. Maybe one day there'll be some "killer app" for the technology I might notice in daily use.

      Happy holidays!

  2. I request DA-AD loop test (AES-118 aka Df metric).