I took my sweetest time on this, but here I finally am with my first article about E1DA dongles. This one is about the two models called 9038SG3 and 9038D. A subsequent article will cover PowerDAC V2.1.
9038SG3 is E1DA’s latest iteration of the 9038S model, which over time went through 3 generations – this being G3 in facts. The 9038S project, like PowerDAC, have always been designed around a balance-ended-only option. In conjunction with this third iteration of the 9038S project, howeverer, following quite a substantial flow of user requests E1DA decided to develop a single-ended (only) version, which is precisely what 9038D is.
9038SG3 can be purchased from E1DA’s AE shop, or directly from E1DA via paypal, for approx $105. 9038D has a regular price identical to 9038SG3 but it is currently not available: E1DA suspended production due to the excessive increase in chip costs – they rate that the higher price at which they would be forced to sell it would be unfair.
Table of contents
|Beyond spectacular cleanness and clarity||Not powerful enough to drive insensitive planar overears|
|Multiple sound optimisation capabilities offer nice experimentation possibilities||Some may not like “overly clean” sound tuning (can be mitigated)|
|Ridiculously inexpensive in light of their quality||9038D: some EMI sensitivity when paired to a phone|
|Very modest host power draw||9038D: lacks some power headroom for toughest planar IEMs|
|9038SG3: easily powers any IEM on the market including low impedance, low sensitivity planars, and most HPs too|
|9038SG3: very good EMI shielding|
|9038D: same sound quality as its balanced sibling|
|9038D: can be plugged into downstream amp|
A word on the manufacturer, and a few on this article
E1DA is a microscopical company. Indeed, a small family run business. The founder and key engineer in there is a Russia born guy now living in China called Ivan Khlyupin. He is an audio enthusiast, and an electrical engineer. Ivan is in charge of all hardware invention / designing, and his elder son takes care of relevant software development. The rest is chinese-cost and sadly chinese-quality contract manufacturing, which is why Ivan (literally) technically assesses and calibrates each sample one by one while or after assembly. Purest artisan’s pride DNA – which is probably why being Italian I feel a sort of natural empathy for the guy.
Their first project was the PowerDAC and it stemmed from a personal need: a low cost, powerful enough dac-amp to drive a pair of badass planar headphones. There wasn’t one on the market affordable and good enough at the same time, so Ivan DIY’d one. And then made it into a 50$ small-scale-industrialised marketable produt.
I’ve been following them for a while now and I own all their current “dongle” models (9038D, 9038SG3 and PowerDAC 2.1) which I of course purchased as a regular nameless customer.
As you will read here, and on a subsequent piece of mine dedicated to PowerDAC, E1DA’s products can easily be recognised as pure audio engineering competence concentrated into tiny, affordable little boxes. In a world overflooded by cheap and even not so cheap “meh-level” stuff, they feel relaxing like sea breeze.
Given their “vertical” technical nature, if there’s one thing E1DA lacks is vulgar-level communication. They do all they can to be super easily reachable via their Discord channel, and they are very responsive. Yet their documents, and their typical answers, are all very technical, with little to no concession to readability let alone accessibility for least competent laymen.
As a consequence high chances are their products get known, let alone “understood” and appreciated, only by already semi-skilled users at the very least. Take myself for one – I got onto them by pure chance and it took me quite some time to dig into some of the aspects of their stuff, and until now I even feel I only got part of that done.
So these are the main reasons why instead of a 4 – 5 pages small article this time I wrote an essay probably 10X longer. And I might even decide to update it in the future 🙂
Note: E1DA is to be pronounced “E one DAH”, following the sounds of the words “Ivan” and “Da” (“Yes” in Russian)
Features and description
The two devices are contained inside the same housing, a sandblasted anodized CNC-machined aluminium case, with marketing graphics (logo) and other data laser-marked in white on the black background.
Size is 48 x 22 x 9mm, weight is approx. 10-12g for either model. Simply put, either device is very small and lightweight which of course greatly facilitates it being used as a dac-amp external upgrade to any mobile device (phone, tablet, dap) which is capable of digital audio output through its USB port.
The housings’ sole apertures are the phone out on one of its small ends, and a USB-C port on the opposite end. Next to the USB-C port there’s also a tiny hole. No other elements, no buttons, no display.
Both 9038D and 9038SG3 are sold in a minimalistic small carton box, containing only the device itself (well protected). No cables / adapters are bundled. USB-C, Lightning and/or Y-USB-C cables can however be separately ordered from E1DA if needed. And they are needed. See more below in particular about appropriate Lightning cables.
9038SG3 and 9038D have very similar internal structures.
Both carry the same Comtrue USB bridge, featuring hw volume (actionable in .5dB increments) and furtherly tweaked with custom software.
Both devices’ heart is a ESS 9038Q2M DAC chip (spec sheet here) featuring current mode amplification, outstanding built-in jitter removal, and a host of user-configurable parameters including Master Clock frequency selection, FIR filter selection and customisable THD compensation coefficients (much more on this later).
Both devices have an amplification stage after the DAC but the two opamps are different on the two models. The internal power filtering also is somewhat different.
9038SG3 features an Analog Devices AD8397 (spec sheet here) opamp offering balanced end only output connectivity.
9038SG3’s internal power rail filtering structure is based on resistors and capacitors. Three different versions of 9038SG3 have been released over time. The earliest version adopted Yageo brand resistors + 2000µF capacitors. Such was a quite early version and according to E1DA shipped until a good 2 years ago.
Since then, 9038SG3 have been and still are equipped with Susumu brand (higher quality) resistors + either 2000µF or 3000µF capacitors. The higher the capacitance, the more efficient the power filtering.
9038SG3 SKU# did not change as internal equipment evolved over time. To tell which version a given unit is look at the product name engraved in white on 9038SG3’ housing: an underlined “3” (like this: “#9038SG3”) indicates a Susumu-equipped model. Looking at a Susumu 9038SG3’s housing, two or three white squares are engraved on the back side (opposed to the E1DA logo one): two squares mean 2000µF capacitance, three squares mean 3000µF.
For curiosity: the “squares” refer to the physical capacitors adopted inside. Two squares = 2 capacitors, three squares = 3 capacitors. By opening the enclosure (don’t! as you would have to reglue it later) they can easily be recognised as three lined-up orange “thingies” soldered on there.
9038D carries instead a TI OPA1622 opamp (spec sheet here) offering single ended only connectivity.
9038D also carries Susumu brand resistors, complemented with 4000µF total capacitance.
Both 9038SG3 and 9038D only offer a single digital input, via a USB-C port (fully USB-2 protocol compatible).
Both carrying the same internal USB bridge and DAC chips, both support the very same digital input specs which are:
- PCM up to 32bit / 384Khz (requires ASIO drivers on Windows, otherwise limited to 24bit / 192KHz)
- DSD up to 256 (again requires ASIO drivers on Windows, otherwise no direct DSD support available)
No drivers needed for full features availability on all other major supported OS. I know of very few other “dongles” supporting DSD256 on Android.
Next to the USB-C connector, on both 9038D and 9038SG3 there’s a tiny LED. Its lighting behaviour has the following meanings (identical on both models):
|Dongle is plugged-in||ON for 0.5 sec, then OFF|
|44.1 / 48KHz PCM stream||0.5Hz pulse (flashes ON every 2 sec)|
|88.2 / 96KHz PCM stream||1Hz pulse (flashes ON every 1 sec)|
|176.4 / 192KHz PCM stream||2Hz pulse (flashes ON every 0.5 sec)|
|352.8 / 384KHz PCM stream||4Hz pulse (flashes ON every 0.25 sec)|
|Direct-DSD stream||Steady ON|
When connected to a USB host 9038D will absorb approx. 435mW (87mA) even when not playing back. Consumption while playing will be even higher of course, from approx. 500 to 585 mW (100-117mA) when receiving a PCM stream (from 44.1 to 192Khz resolution respectively), and from approx. 500 to 670mW (100-134mA) when receiving a DSD-64 to DSD-256 direct stream.
These figures are low, and even incredibly low when we consider the output this little kid is able to provide (see subsequent chapter).
Both models have an automatic Standby function, which can be enable/disabled by the user and is enabled by default out of the box. Thanks to such feature both 9038SG3 and 9038D will go into “Low Power Consumption” mode if they don’t receive any data from the host for 60 seconds – so when they are “on but doing nothing” so to say – and they will automatically wake back up when a new stream will start flowing again. Power absorption in such condition is circa 50% of the minimum required on quiesced playback status, so around 205-210mW.
I’ll provide more details much down below on how to enable/disable the Standby & Mute functions, and some caveats.
Considering the huge power especially 9038SG3 is able to deliver on mid, low and extralow (see below) the above figures are impressively low. Just a bit more than 600mW (only 120mA!) when decoding DSD256 and at high volume seems like a joke.
9038SG3 and 9038D are by far fully USB2 compliant, so no problem with any Android phone and if the phone has a relatively modern battery (3-4000mAh or more) 9038D / 9038SG3 will not meaningfully jeopardise battery duration.
They can be used on quite a few iPhone models too but choosing the Lightning cable that make it happen may be tricky. See much down below a dedicated chapter to this issue.
I thought about inserting a digression at this point regarding power demand on so-called “dongles” (like 9038SG3 and 9038D) and some considerations stemming from there, but my notes soon developed into something deserving a separate take. So please be patient, I’ll be issuing a standalone article on this ReallySoonNow™.
As mentioned above, 9038SG3 and 9038D carry a different amplifier module, which is the most significant difference between the two models.
9038D has one 3.5mm single-ended output port, which supports connection to earphone/headphones of any impedance and technology, and also connection to the single-ended input of an Amplifier device.
9038SG3 has one 2.5mm balanced-ended output port, which exclusively supports connection to earphone/headphones of any impedance provided they have a balanced-ended termination. Connecting Amplifier devices to 9038SG3 is not supported, not even via such Amplifiers’ balanced ended input ports – failure to comply to such exclusion will most surely physically damage the device.
9038SG3 features very significant output power:
- And 3.3Vrms @0dB vs high impedance loads
9038D output power is also quite interesting albeit definitely lower:
- And 2.75Vrms @0dB vs high impedance loads
Both offer extremely low output impedance, around 0,1Ω.
All figures come from E1DA, and correspond to measurements conducted at room temperature, and at 1% THD+N
It’s quite interesting to note here how 9038D’s output power decreases as impedance goes down, while 9038SG3 increases in the same condition. A nice opportunity to learn something. Let’s articulate.
As [ehm…] everyone [should] know[s], electricity laws say that when voltage stay the same, reducing load (impedance) makes current go stronger. That’s what apparently happens on 9038SG3, doesn’t it – while 9038D seems to break such rule.
More precisely, Ohm’s law says that slashing load impedance in half power will exactly double up. And if we notice, 9038SG3 doesn’t really cope with this. So 9038SG3, too, “breaks the rules” apparently ?
Neither does of course.
An amplifier can only provide up to a certain maximum amount of power, and in particular a certain maximum current intensity. Going beyond such limit would physically damage the device which is why there often (but not always) is some sort of soft or hard stop implemented to avoid that.
That’s however why by continuously slashing impedance by half we won’t (of course!) get indefinitely doubling power on any physically existing amplifier: eventually the power will start “growing slower”, then will start going down.
The sad part is that reading that a certain amp is able to deliver 4W at 32Ω, or 6V against 600Ω, does not give us any (any!!) information about how much power will that very device be able to deliver onto our 14Ω, 95dB/mW preferred IEM driver.
Such piece of information is most frequently missing, or unclear, on most amps’ spec sheets. Let’s use 9038D and 9038SG3 as examples now.
We know from above that (e.g.) 9038D provides circa 2.75V at 0dB (i.e. “at full volume” position) against very high impedances.
As we unplug high impedance headphones and start plugging headphones or earphones of lower and lower impedance, our 9038D will keep on providing 2.75V at 0dB “for a while”, i.e., until the earphone we plug will have a certain minimum impedance. From there on down, 9038D will start applying less than 2.75V on it, thus reducing the current intensity flow into the drivers, to keep it under its cap.
As a result, from that load impedance value on down we’ll see that 9038D’s power figures will not anymore “double up” as impedance halves down. They will initially start growing less than 2X, then will eventually go down.
Let’s do some math on the above numbers.
We know by the measures provided by E1DA that 9038D delivers 200mW on 40Ω. That corresponds to 71mA and 2,83V. Concede on some approximation error (actual ohms might have been like 40.2 or so, and rounded up for typographical rationales), and we found an impedance at which 9038D “can still afford” applying its max-V (circa 2.75V) at 0dB: that’s (circa) 40Ω.
Let’s look into 32Ω now. If it didn’t encounter its limits yet, OPA1622 (the op amp inside 9038D) should give us >230mW of power (2,75×2,75/32). Instead, we measure only 180mW. So not only the power has not gone up, but it even went down!
This tells us that on “some” load impedance value between 40 and 32Ω OPA1622 reaches its sweet point beyond which it starts slowing down on power to avoid exceeding its Current capabilities. In facts 180mW on 32Ω are 2,40V and 75mA. So the device went down in power compared to the 40Ω load case by reducing Voltage (2,40 down from 2,75), even if Current still went up a bit (75 up from 71mA).
At 16Ω 9038D delivers 120mW (so not at all twice the value at 32Ω, indeed 33% less!) corresponding to 1,39V and circa 87mA. See? Power went down in relation to a severe voltage reduction (1,39V down from 2,40V) while Current still went furtherly up.
It’s not written up above but let me add here that (circa) 87mA is OPA1622’s current cap. How do I know it? It comes from TI’s spec sheets.
Opamps’ spec sheets have to be taken with triple grain of salt as they offer cryptical data first of all, and even most importantly because they offer information about the broadest possible alternative implementations of that very chip. Simply put, it may well be that the figures “promised” by the chip manufacturer are not realistically reachable in the particular situation / implementation we are considering.
In this case, however, we find that the current 9038D delivers onto a 16Ω load matches quite nicely with the maximum current the chip’s manufacturer reports. So that’ll be it.
Which means that we now know even without measuring that onto furtherly lower impedances (14ohm, 12ohm, 8ohm…) 9038D will keep on delivering a maximum of 87mA, so it will be forced to apply lower and lower voltages to cope, and correspondingly its power figure will rapidly go down.
You can do the math yourself: Power in mW = Current in mA ^ 2 * Impedance in Ω. At 14Ω for example you can expect circa 106mW from a 9038D, give or take.
Let’s now look into 9038SG3.
We know its max V at 0dB is 3.3V. We also know it issues more than 340mW onto a 32Ω load, which corresponds to circa 103mA and 3,3V. So at 32Ω 9038SG3’s Voltage has not started to “go down” from max yet.
We also know it delivers circa 550mW@16Ohm, corresponding to 185mA and 2,87V. Here Current is higher than 103mA, Power is much higher (but not double!) than 340mW, and Voltage is lower than 3.3V. This tells us that “someplace” between 32Ω and 16Ω 9038SG3 starts to find the need to slow down, at least reducing its growth.
We finally know (always by measure, so within the measure’s error rate) that 9038SG3 delivers circa 600mW into 10.3Ω – corresponding to 243mA and 2,50V. Again: power goes up, but at en even lesser rate.
We do not have an official current cap value coming from E1DA about 9038SG3. AD8397 chip’s manufacturer talks about >300mA but that’s one of those cases where the information is of little use as AD8397 is a quite “professional” chip, designed with a lot of liberty (it does not have a proper current limiter, only thermal control) so reading on its specs that it can deliver up to 310mA is not fully indicative for us as the contour conditions for such performance may well not be those of a device like 9038SG3.
So unless we actually measure that, we have no real way to devine if 9038SG3 will exceed 243mA current on even lower impedances but hey!, even this value is incredibly high – double so considering the device class we are talking about-
P.S. – for the most precise readers: all W, V and A values mentioned above are “rms”.
Sound and performance
I’m going to report about 9038SG3 first, then I will more easily cover 9038D in terms of differences from that.
In its stock calibration situation, so Out Of The Box as they say, 9038SG3 is easily one of the cleanest, most detailed and fast (short transients) dongle I ever auditioned. Indeed, from the sound structure standpoint it rivals much higher class devices.
Notes are exceptionally well separated and clean, while on the flip side they come accross somewhat lean, and this contributes to a general feeling of “scarce musicality” and “excess in detail” if that even makes sense of course.
Leveraging on its internal harmonics compensation generator via the Tweak9038 app, 9038SG3’s “presentation” can be altered to be made a bit “more musical”, “warmer”, even “tubey”. Sure, it takes some will on experimenting of course but possibilities are there. The effect however is not that of flipping the whole presentation into a dark/warm one. See much more on this below.
9038SG3 also has nice spatial rendering – “soundstage” as we call it is definitely OK. Not the level at which a Groove renders depth and height on highres (>96KHz) streams, but that’s related to Groove’s FR being uncommonly flat much beyond 20KHz more than anything else (see here for the full story). Barred that, 9038SG3 has nothing to envy to any other dongle I auditioned, at any price, on this respect.
9038SG3’s lack of sound coloration is obvious, but that’s possibly the second most outstanding feature I noticed right away – the first being the very high amount of power (current) this little box is able to deliver onto low impedance, low sensitivity drivers.
Give or take 9038SG3 delivers 550-600mW into 14 and even 10Ω loads: a sort of mini nuclear plant, perfectly capable to “move” deep insensitive drivers e.g. Final E5000 (14Ω 93dB), RHA CL2 (15Ω 89dB), Hifiman HE400S (22Ω 98dB) and pretty much any low impedance planar IEM you can think of, and with some plenty of room to spare.
As load impedance goes up 9038SG3 stays an uncommonly powerful thingie but starts to show its ropes of course (hey it’s an effing dongle…). For example SRH1840 (65Ω 96dB) are still kinda no problem, but Hifiman HE560 (45Ω 90dB) are no-no.
Lastly, 9038SG3 max voltage swing on high impedance drivers (3.3V) makes it more than decently fit to drive the likes of Sennheiser HD600 (300Ω 102dB/V = 96,7dB/mW) – on which the “tube emulation” is worth a spin, maybe even two… – see below how. 😉
Once all the above is clear, describing 9038D is relatively simple: it’s virtually identical in tonality, timbre, cleanness and technicalities, but delivers way less current on the low loads end, and also more modest voltage swing vs high impedances.
On the former part I guess I can call this yet another example of how a “balanced” scheme is not a requirement to the purpose of outstanding quality on sound output. This consideration apart, 9038D like 9038SG3 sounds magnificently well, and it can be tweaked and changed exactly like its sibling so it’s up to each one to leave it “more analythical” as in its OOB tuning, or a bit “more musical”.
The latter part reflects into a quite different applicative span for 9038D compared to 9038SG3.
While 9038D can still properly drive the likes of Tanchjim Oxygen (32Ω 110dB/Vrms=95,5dB/mW) or Final A3000 (18Ω 98dB) or the recently hyped 7Hz Timeless (14.8Ω 104dB), other drivers like Final E5000, SRH1840 or other more seriously harder to drive planars are at various degrees not ideal, or not viable altogether pairs.
Similarly although less seriously on the higher impedance end: 2.75V are OK to make HD600 sing, but there won’t be much room to compensate in case of low-level recorded tracks and/or level-punishing EQ schemes.
Very succintly put: 9038SG3 delivers incredibly clean sound and very good technicalities and so much power that it can act as a one-stop-source for all IEMs on the market bar none, and most Headphones too, bar high demand planars only. 9038D offers the same sound qualities, can drive “most” IEMs and a few Headphones from of a single ended connection so without requiring cable swapping.
Before I forget: 9038D is virtually immune from hissing when paired to oversensitive loads (Campfire Andromeda, Penon Volt…). 9038SG3 does hiss a bit on the same drivers.
And lastly: as quickly mentioned above and explained in better detail down below, 9038D can be used as a pure DAC connected to a downstream amplifier. Given its outstanding sound profile and its ease of integration on pretty much any host OS, such application might be something to seriously look into, in spite of its external “superpocketable dongle” format.
Hidizs S9 Pro ($119,00)
An educational case insofar as we are talking about almost identical-priced devices, and based on the very same DAC chip (ESS 9038Q2M).
First of all, both 9038D and 9038SG3 sound simply obviously better than S9 Pro. Their presentation is much more linear, clean and detailed compared to S9 Pro’s balanced output. S9 Pro’s high mids very easily tend to “overdo”, and the treble end lacks some air in comparison. S9 Pro also lacks any form of tweakability. S9 Pro’s single ended output is almost unaudible to me quality wise.
On the power delivery standpoint, 9038D’s output is marginally more powerful than S9 Pro’s balanced ended out, and more than twice its single ended one. 9038SG3 is roughly 50% more powerful than S9 Pro at 32Ω, and even most importantly S9 Pro runs into a serious current shortage from right around 16Ω on down, while 9038SG3 still provides something like 600mW vs 10Ω loads. In practice: 9038SG3 easily drives E5000 and planar IEMs, S9 Pro can’t even start trying doing that, or doing that at a comparable level.
Simply and perhaps a bit unforgivingly put: S9 Pro is a toy compared to 9038SG3, and less desireable (although by a smaller margin) even compared to 9038D.
Cayin RU6 ($250)
As you’ll read on my separate take on RU6 (due Soontm), in less than a million words RU6 has in its unique timbre its main if not sole reason to be. Its internal R2R technology implementation delivers in facts an audibly different nuance to notes, and that is likely the reason for the ticket price for the curious modest-budgeted audiophile.
The rest is unimpressive at best, often underwhelming. The R2R timbre is audible on NOS mode only – which sadly requires high-res (>= 96KHz) digital tracks to be fed from the outside, as its noise, distortion and FR rolloff on Redbook material is nearly comical – which is even a worse pity if we consider that amongst all that noise one can hear above decent imaging and note body. That’s probably why many say RU6 should exclusively be used on OS mode where reconstruction of 44.1KHz becomes decent-ish, with an at least reasonable sense of space, and much less audible noise. Sadly, the OS circuitry is deltasigma based which defeats most if not all the purpose in this case.
Be as it may, RU6 never comes even close to 9038SG3 / 9038D in terms of clarity, cleanness and detail retrieval.
Power wise the situation is similar to S9 Pro: RU6 is a quite modest-powered device, delivering “just” 213mW@32Ω on BE (similar to 9038D on its single ended, and much less than 9038SG3), slightly more than half on SE, and most importantly dropping quickly below 16Ω so in this case, again: nothing special on E5000, and forget cheecky planar IEMs, etc – unlike what is fully allowed by 9038SG3.
Unlike S9 Pro, RU6’s relatively modest output power at least comes with modest host power requirements – that’s a quite important note. Together with the fact that all the above is attached to a 2X price tag.
IFi GO Bar ($329 / €329)
Also Soontm you’ll read my full article on GO Bar. In the meanwhile…
The first big difference with 9038SG3 / 9038D is in the price of course: almost 3 times as much. Better be something serious in there doesn’t it.
Another thing is power. GO Bar is powerful on high and medium impedance loads. It swings a whopping 7.2V into 600Ω (more than twice 9038SG3), and 475mW into 32Ω (40% more than 9038SG3).
Sadly, it hits against a wall of current limitation (circa 120mA) as impedance goes down. As a consequence GO Bar (balanced out) drives Final E5000 (14Ω 93dB) with good athleticism, although with less headroom compared to 9038SG3, but it won’t properly drive the likes of RHA CL2, which are instead perfectly managed by 9038SG3.
Probably due to its performances on higher impedances, or to lesser efficiency, or both, GO Bar, unlike 9038SG3 or 9038D, is a power w**re. It absorbs up to 4W while working, which is 800mA – so it is not USB2 compliant and by far so. Not all Android phones will drive it to its full power then: a laptop is required, or a battery in parallel with a phone. Oppositely, 9038SG3 and 9038D are very modest in terms of power needs vs their output power capabilities, and fully USB2 compliant.
One more thing is features. GO Bar misses the harmonic compensation and masterclock customisation infrastructure available on 9038SG3, and that’s not small stuff, and offers only 4 different FIR filters to choose from instead of 7. On the flip side GO Bar covers the user with features one nicer and/or sexyer than the other, all of which are totally missing on 9038D and 9038SG3: XBass and XSpace analog-domain effects, selectable low/high gain, integrated IEMatch, high quality integrated power filtering, and (for Tidal’s aficionados) MQA full decoding.
So in the end yes, GO Bar does give quite something more than 9038SG3 in return for that higher purchase price and a much higher host power need. I see 9038SG3 as a device delivering similar or better sound quality, and similar to much higher output power onto IEMs compared to GO Bar, in a nofrill package and for a fraction of GO Bar’s price.
Apogee Groove ($220)
As extensively reported on my piece about it, Apogee Groove is an oddball. A badass of an oddball if you wish, but still an uncommon device, with the pros and cons one may after all expect from oddity.
Groove’s output stage is based on proprietary technology and does not support crossover filters or similar circuitry, and all too often it also powers Balanced Architecture drivers (even single-driver models) very quirkily. To cut it short, Groove is mainly if not solely intended for Dynamic Drivers, which is of course an apriori fact to seriously consider when looking instead for a “universal application” DAC/AMP dongle.
That said, Groove swings 5V into 300Ω and 600Ω impedance cans, making it obviously more energetic compared to 9038SG3 and of course 9038D too, and to all other dongles on the market with the sole exception of iFi’s GO Bar.
On the opposite end Groove delivers less current than 9038SG3 which is why it can power Final E5000 (14Ω 93dB) well, but falls dramatically short when applied to RHA CL2, which 9038SG3 eats for breakfast instead. Always from the current delivery on mid/low impedance loads standpoint, Groove is OK-ish on SRH1840 (on the limit, let’s say), while 9038SG3 dances them around more “brilliantly”, and with much more headroom for sure.
Like 9038D, and unlike 9038SG3, Groove can be exploited as a DAC connected to a downstream amplifier.
Groove requires nearly 3 times the current 9038SG3 or 9038D do from their host, which is still USB2 compliant but a huge point to consider nevertheless.
Power profiles aside, Groove and 9038SG3 are very different in terms of sound presentation. Groove is way superior in terms of macro-dynamics (imaging) and even more so in terms of spatial drawing: I hardly can name a single mobile DAC device better than Groove on this.
On the other hand 9038SG3 is obviously less colored and has better subtlety on detail retrieval. Flipping the coin, Groove is gorgeously more “musical” than 9038SG3 and you won’t change the latter anywhere near the former via TCC tweakings.
In less than a million words: where applicable and therefore apriori comparable, Groove is more musical and sexyer, 9038SG3 is more technical, cleaner, sharper. Groove is more powerful on Sennheiser cans, 9038SG3 is way stronger on planar IEMs. I’m so happy I own both, and I would again buy both as these two together cover all possible needs south of a much higher end (and priced) DAP or battery-powered DAC-AMP.
E1DA developed a companion Android app for their 9038SG3 and 9038D dongles. It’s called Tweak9038.
The purpose of the app is giving the user access to most if not all customisable parameters offered by the ESS 9038Q2M chip, and it indeed succeeds in doing it reliably and quite easily too. The down side is that those parameters are quite technical stuff, and customising them to “make sound better” requires knowing what one’s doing – no worries though, you can’t damage anything if you do it wrong.
There are some limitations to the app, including:
- It only works on Android OS, and they are not planning to port it anywhere else. It anyhow technically might never work on iOS due to Apple limitations.
- It only communicates with 9038SG3 / 9038D via USB.
So the device must be plugged into the Android device where the app is via a USB cable, and only then the dongle configuration can be accessed, seen and changed. The modified configuration can be saved into the dongle’s own non volatile memory, and it will stay there even when the device is plugged onto a totally different device, even if not carrying the Tweak9038 app, and/or if not even Android-based.
Some may consider the app cost (10$) also a limitation. They are wrong. This app is totally brilliant, and adds a lot to 9038SG3 and 9038D’s value and to their uniqueness, as I’ll explain in a bit. It’s actually very cheap for all it offers and how accurately and reliably it works. If you have some strange problem with “expensive apps” (expensive? 10$? Well, ok…) just mentally add 10$ to the price you pay for the 9038SG3 or 9038D and you won’t fail noticing they will stay two incredibly inexpensive dongles in light of the over-amazing quality they offer.
So quit whyning already, and buy the bloody app to support its developers 🙂
As I mentioned above, Tweak9038 exclusively supports 9038D and 9038SG3. It does not support 9038SG2 or earlier. 9038SG3 is anyhow quite significantly better than 9038SG2, and still affordable enough that if I were a 9038SG2 owner I would not think twice about buying a 9038SG3 as an upgrade.
Enough foreword. In summary, Tweak9038 allows to:
- Customise the Minimum, Maximum and Default levels of the device’s hardware volume scale.
- Generate harmonic distortion (yes, you read well), and even do that diversely following playback volume.
- Customise the DAC’s clock frequency, and apply different values automatically based on the track sample rate.
- Select amongst different available reconstruction filters, and again choose different ones automatically depending on the track sample rate
- Enable/disable Standby and Mute options
- Save “sets” of all the above parameters under user-defined names, and recall + apply them to the device whenever liked.
- Lastly, scratch everything off just in case you need to return to the exact configuration and calibration that very device had when leaving the factory.
While some of these features may seem easy, others are quite obscure or at least they were to me. After some extensive use and a lot of curiosity applied, I must say this has been dual fun me: once because these tweaks resulted in amazing sound output, and twice because they gave me the occasion to study their rationales in deeper depth then I ever did in the past – which of course now helps me put other devices in a much more realistic and technically more correct perspective.
Here below I will go through most of the “stuff” I experimented and sometimes learnt. YMMV of course: if for you all this is already bread and butter well, just jump through 😊.
A special mention is deserved by E1DA’s support team, which is easily reachable via their Discord channel, and always available to provide competent and precise input.
Volume scale calibration
Out of the box, 9038SG3 / 9038D hardware volume is set to go from -127.5dB to 0dB, with the default value set to 0dB.
Hardware volume values are to be intended as “attenuation” values. So 0dB means “no attenuation”, that is, “leave the amp output fully undampened”, aka “go ahead, kill my ears!”. Oppositely, -127.5dB means “drop the output volume down by 127.5dB” which is a huge dampening. It equates to “shut the F up!”.
The Default volume value is the value the device will set the volume at whenever it is turned on. Given it’s a battery-less device, it will turn off every time it is disconnected from a host, and on when it is reconnected back. Default volume set to 0dB means: whenever you turn the dongle up set the volume to “full unbridled sound out”. Sounds scary doesn’t it. And yes, it is scary.
Until now I talked about the dongle’s “internal” (“hardware”) volume values, but we don’t normally “directly see” those values. What we most often see is a volume slider on our music player software, or even on the Operating System of the machine the dongle is connected to. Such slider is usually labelled as going from “zero” (meaning “zero volume”, 0%, or “silent”) to “100” (or “100%”, “full volume”, “full loudness”).
If our music player app’s volume slider is “linked” to 9038’s internal hardware volume (and it usually is – either automatically or by manually switching it on as you can do on most sw player’s settings) then out of the box the music player’s “zero” volume maps into 9038’s -127.5dB attenuation, and on the opposite end the music player’s “100” volume maps into 0dB attenuation. And the “default volume position” at power-up will be “full volume” position, or, the music player’s latest used volume position (depends on the situation).
Twea9038 app allows for customising all 3 of these default hardware volume settings. But why should you?
Well first of all: the default volume level. It is much, much safer if we set it low, instead of high, let alone “full up” (i.e. “0dB”). This is simply because sooner or later we risk to forget to bring the volume down before hitting “play”, and doing this with the default volume level set too high blasts so loud sound into our ears that we can (seriously!) be permanently harmed.
Secondly: the lower end value. -127.5dB is such a huge attenuation value that unless we plug an extremely oversensitive driver in, starting from “volume full down” will require moving the volume slider a lot before getting some decent sound pressure (“loudness”) out.
First impression in such case might be that 9038S has very weak amplification (“hell… I need to bring it to 80% to get some loudness even on these simple IEMs…!”). But that’s not the case. Moving the volume slider further up the sound pressure will raise very strongly, up to deafening levels.
The real problem is that the “Min” (starting) value is way too low for practical purposes. Setting the minimum hw volume value to a more convenient setting “fixes” this. Which setting is exactly recommendable depends on the impedance and the sensitivity of the actual drivers we “usually” pair 9038 with. For my sets I found that a value of -80dB is OK.
Lastly, let’s consider the upper end value of the scale.
9038SG3, especially, is powerful. Quite seriously powerful I mean, as it can swing 3.3Vrms into a high impedance driver, or – on the flip side – push up to half watt (!) into a 16Ω load.
The downside of all such power is that (again) a wrong move with the music player’s volume control can deafen the user for good, especially if this happens while using IEMs or Headphones which are not “impossible” to drive for 9038SG3, which is like… 95% of the existing ones (and 100% of those in my possession).
Setting the upper end of the hw volume scale to a value lower than 0dB is a safeguard in such sense. Once set at (say) -10dB, this means that when (willingly or by mistake) music player’s volume is slammed to “full up” 9038SG3 won’t release all its possible power onto the drivers, bit quite a bit less.
Similarly to the bottom end value case, the “right” (“most practical”) value to choose for the upper volume end largely depends on which earphones / headphones are part of one’s rotation. If most drivers are very sensitive stuff like Campfire Andromeda, Penon Volt or the like, a pretty low value is recommended! Oppositely, if drivers at hand are hard to drive planars, or insensitive and/or high impedance DD’s, it may be best to leave the value near 0dB, or just below that.
Besides writing values into Tweak9038’s GUI, there’s also another “hidden way” to adjust the Maximum hardware volume boundary by “fiddling” in special ways with the host OS volume slider. This works on multiple different hosts (MacOS, Windows or Linux, and Android – in such last case UAPP is required).
Here’s the scoop:
Bring the host volume slider to 0% (so “all the way down”), and then quickly raise volume + bring it back down to zero% + raise volume again. The “gesture” is like “pulling the volume slider up from zero and quickly slamming it back down, then bouncing back a bit”. The “bounce up” should not exceed 50% of the totale volume slider run space. Do this “bounce” trick 3 times in a row and this will result in a -30dB Max Volume value being instantly set. A sort of quick way to impose a hard “volume limiter”, if you wish.
Bring the host volume slider to 100% (so “all the way up”), and then quickly lower volume + bring it back to 100% + lower volume again. The gesture is specular to the previous one, it’s like “pulling the volume slider down from 100% and quickly slamming it back up, then bouncing back down a bit”. Same caveat as before: the “bounce down” should be less than 50% of the total volume slider run space. Do this 3 times, and Max Volume value will increase by 3dB. Do this another 3 times and it will increase another 3dB, so 6dB in total. And so on.
I don’t know if you agree but I find this so brilliant… 😊
I have two minor negative points to mention, too.
One: even if a front end music player app directly controls 9038SG3 / 9038D’s hardware volume, it has however no way to know its absolute Min and Max values as they are set inside the dongles by the Tweak9038 app. So for example it did happen to me to spend some sweet time wondering (while swearing) why the heck my 9038SG3 could not make a certain IEM sound really loud even at “full volume” (on the player), only to remember much later that I had set the Max hw volume value to a low level myself.
Two: someone may feel more comfortable if the GUI mentioned volume values in Vrms units in addition to dB units. This would be easy to implement as there is a precise formula linking the two, and someone did already put this in the wishes box to E1DA, so I trust it will eventually happen. Until then, we can calculate them manually as follows :
Output voltage [Vrms] = FullScale voltage [Vrms] * 10 ^ (Attenuation [dB] / 20)
So for example an attenuation of -10dB results in:
- (For 9038D) 2.75 Vrms * 10^ (-10dB/20) = 0,87 Vrms full scale voltage
- (For 9038SG3) 3.33 Vrms * 10^ (-10dB/20) = 1,04 Vrms full scale voltage
A “distortion generator” – to help reduce distortion
Sounds like a paradox doesn’t it.
Ideally, a DAC/AMP should be a “purely transparent” device, reconstructing and then powering the exact analogue wave described by the digital samples it is fed with.
The term “distortion” often widely generically indicates “any” deviation from such ideal. Overdoing with the volume yielding into clipping is called distortion. Noise floor is sometimes also called distortion. Etc.
More properly speaking, “distortion” has to do with “harmonics”.
For a somewhat technical intro at what harmonics are, start here. But let me vulgarise as always.
An “harmonic” is a replica of a certain sound (called the “fundamental tone”), featuring a frequency which is 2X, 3X, 4X … nX compared to (i.e. an integer multiple of) that of the fundamental tone.
Harmonics corresponding to 2X, 4X, 6X […] their fundamental tone frequency are called “even order harmonics”. Those corresponding to 3X, 5X, 7X […] the fundamental tone frequency are called – guess what – odd order harmonics.
Harmonics can be both good or bad, in a sense.
When playing a musical instrument (say: a guitar) one may develop techniques to produce certain harmonics together with, or even instead of, a certain “pure note” – of course aiming at a special sound effect. These are “good” harmonics, we do want those to be there. Beyond their name, if you want, such harmonics acquire the same dignity as any other note played by that original instrument.
On the flip side, unwanted harmonics are generated in parallel to their fundamental notes by many sorts of disturbances involving the sound source (a musical instrument, or an analogue and/or digital sound reproduction device).
Long story short: pretty much every time a note is “played”, “some harmonics” happen too, which are in general of the “unwanted” kind.
Harmonics typically come with a lower amplitude (they are less loud) compared with their fundamental tone, and also often fall outside the audible frequency range.
Audible harmonics can be perceived as a change in music’s timbre, or as some odd notes or accents audible here and there which are not supposed to be part of the original music. All these effects are often referred to as “sound coloration”.
Harmonics falling outside the directly audible range (so above approx. 16 – 18 Khz) will still alter the sound purity, as they impact e.g. on sound timing such as echoes, reverberations, etc. They modify the “sense of space” which that specific music would generate when played “more cleanly”. Additionally, harmonics fundamental tones around the same frequencies will interact producing Intermodulation Distortion (IMD), a further type of distortion.
Total Harmonic Distortion is the ratio between the “force” of all these unwanted sounds (the harmonics) taken together, divided by the “force” of the “originally intended music”. The lower such ratio the best of course, as it means those little bastards (the harmonics) are so “weak” they don’t effectively affect the purity of the ideal sound (significantly).
As there ain’t such thing as a “perfect” device, of course there is no such thing as a “totally non-distorting” DAC/AMP. Alternatively said, our audio gear’s THD will always be >0.
Audio equipment engineering and manufacturing of course includes keeping distortion as low as possible, for as low industrial cost as possible of course.
There are structural (I like to call them “static”) causes for harmonic distortion: the quality of the electronics, the cleanness of its implementation, etc. A badly engineered device based on crappy components will produce distortion under any operative conditions; under the very same operative conditions, a better engineered device based on higher quality components will produce lower distortion.
Then, there are those which I call the “dynamic” causes. Operative temperature, for example, is a factor: electronics do change their behaviour with temperature, and that makes a difference in their audio behaviour. EMI/RFI interference even more so. And load: depending on the impedance of the connected headphones the source device will “behave” a bit differently and will generate different distortion patterns – in very general terms, distortion goes up as impedance goes down. Even volume: the same amp will distort less when operated at mid-volume, more when pushed at the top of its capabilities. Etc.
So not only the most skilled manufacturer in the world will be unable to deliver a zero-THD device, but even their best device will have an always somewhat variable THD, as some of that THD depends on how that device is being “actually used”.
Is there a way to cope with such distortion depending on usage conditions? Well… in part, yes.
If we know which parasite harmonics is the device generating under certain usage conditions (e.g., when a given headphone is connected), then we can create some harmonics ourselves which are “equal-but-opposite” compared to the unwanted ones, thus effectively “cancel them out”.
The ESS 9038Q2M chip does have a sort of built-in harmonics (compensation) generator, and indeed that’s what E1DA exploits to first of all calibrate each and every 9038D or 9038SG3 unit prior to shipping.
By design 9038SG3 and 9038D aim at a THD of -125dB as a target value. A value of THD = -125dB is considered ideal, -124dB is considered “standard”, and -123dB is the threshold below which that very unit will be sold as b-stock.
Given what we just noted, we wonder: under which effective usage conditions are such THD values verified?
E1DA reports: they calibrate and measure their 9038SG3/9038D devices with a 32Ω load, and at an operative temperature of 25°C. Then they observe the (unwanted) harmonics during playback of a reference signal while going from min to max volume, and they set compensation values into the the ESS 9038Q2M chip harmonics generator to cancel them out.
Such compensation values are finally burnt into the device’s firmware before shipping. No matter how hard we subsequently mess with the harmonics generator for experimenting etc (see below), we can always go back to E1DA’s original “factory values” by tapping on Restore Factory Settings on the Tweak9038 app.
During real life use we will of course plug all sorts of different impedance drivers into our 9038SG3 or 9038D dongles. When their impedance will be significantly different from 32ohm those factory-pre-set compensation values will be less effective to the purpose. What can we do then?
One: experiment “by ear”.
Simply reach out for the Tweak9038 app and modify the values on the “THD panel”. Doing this while playing music, the result is hearable in real time. So anyone can judge by oneself if the change is adding or removing distortion, and by how much.
As distortion often also depends on the volume at which the device is being made to work, the Tweak9038 app allows to define 3 sub-ranges of the entire volume range. Range 1 goes from -127.5dB to Low Threshold (blue-green), Range 2 from Low Threshold to High Threshold (orange), and Range 3 from High Threshold to 0dB.
Threshold values can be freely modified. To set them either drag the bullets atop their vertical bars, or tap their values (the blue-green and orange figures atop the bar) and directly key the new number in.
Once Ranges are defined, tap on each of the 3 “THD Edit” buttons, at the top, and enter harmonic correction value for each of those Ranges.
It is possible to generate even (2nd) and odd (3rd) harmonic values. Each value must be entered in dB, and by ticking the Invert flag we flip the harmonic’s phase.
Similarly to how it works for the Thresholds, harmonic values can be input either by dragging un/down the 2 orange bullet atop the animated “graph spikes” at the centre of the screen, or by tapping on the number values within the frame on the upper-right.
Suppose we don’t want to take volume variations too much into account: how do we define a “flat” correction, all equal for the entire volume range?
There’s 2 ways to do that: either input the exact same values into “THD Edit” for all three ranges, or define Threshold values such that… only Range 3 is effectively ever active (i.e.: set both Low and High Threshold to -127.5dB), and input correction values only under Range 3’s THD Edit space.
Two: go the engineeristic way
As I very briefly mentioned above, THD is typically inversely proportional to load impedance. Which means that E1DA’s factory calibration, centered upon a 32Ω load, will deliver ideal results for 32Ω but results will still be much more than decent at higher impedances; viceversa it will “need some help” – so to say – when pairing 9038SG3 or 9038D with sub-16Ω IEMs.
To find out as accurately as possible which new values optimise a 9038SG3 (or 9038D) when paired with a specific earphone / headphone, some equipment, and following a similar procedure, to what E1DA uses and does in-house to pre-calibrate 9038S3G and 9038D will be required.
And guess what: E1DA develops and sells such equipment. It’s called Cosmos, and does exactly that (and much more). Here is the link to the description – I will not go in more detail here, this article is already long enough isn’t it.
A “distortion generator” – to actually add some distortion
We do all we can to get distortion-free DACs. We even calibrate them in respects to our headphones, one by one, to compensate for load-dependent distortion… why the hell would we want to “add” distortion???
As I mentioned above, 9038D and 9038SG3 come with whopping -125dB (or so) THD which is a monumentally good value for such a device and especially price class.
And in facts they do sound… clean. Holy cow they really do!
Tell you what: maybe a tad too much ?
In an ideal situation, when we listen to our preferred digital tracks we want to hear the hell of the detail, and layering, and separation and all that exactly as it is “contained” in the track file.
We typically assume that the digital information inside our CDs, or FLAC, WAV or DSF files, is “the” thing, it’s a “given perfection”, and our task is finding the best gear we can possibly afford to convert that into wonderful music reaching to our ears with the “highest fidelity” possible towards such allegedly “perfect” starting point.
Is such assumption correct ? No it’s not.
Bad recordings are of course a thing, to begin with. But there’s much more and much worse.
Music publishers do mess with music “purity” inside their masters to compensate for most of their paying customers very likely going to play back that track on supercheap, not at all hifi-grade gear.
Such “mastered/remastered music for cheap gear” will “sound better” (or “less worse”) on cheap gear, but will reveal all sorts of unwanted sonic features (compressed dynamic range, lack of definition etc etc) when played back on higher level, low-THD equipment.
The opposite may also be a problem, sometimes.
Suppose we have a very good digitally mastered edition, with no or minimal compression, no artificial panning, etc etc – a good audiophile level job. But, we are accustomed to listening to it with some “coloured” gear. So much so, and for such a long time, that our brain got biased: for us that song’s “home” sound is that colored.
Then one day we listen to that same digital track with a much lower-THD device. While we’ll certainly appreciate the higher definition, better technicalities, etc, chances are our brain may decode such newly conquered “transparency”, or “neutrality”, into “lifelessness”, “lack of musicality”.
E1DA 9038SG3 and 9038D’s harmonics generator can (incredibly in a sense, but really) help also for this case. By fiddling with the THD compensation values we can “add some colour” to sound, making it “more musical” in a sense.
There’s even a way to simulate the voicing of a tube amp – that’s mainly about playing with the 2nd harmonic. The Tweak9038 app even offers “tube emulation” presets, those are labelled “SE” – there’s one for 9038D one for 9038SG3 – all it takes is loading them, and they can be furtherly tweaked of course.
When on your quest for more colored sound you may also want to remember that
- the lower the MCLK, the lower the DR (more on this below)
- linear filters often tend to make notes less sharp (more on this below, too)
Playing with 9038SG3 / 9038D’s sound tweaking gauges is just amazing 😊 This video (by E1DA) shows a live demo of the game.
Setting a custom clock frequency
Tweak9038 allows to set the ESS 9038Q2M chip’s Master Clock frequency at 3 different values: 12.5Mhz, 25MHz or 50MHz.
A higher clock frequency lowers the noise floor, produces better note definition, sharper attack, better space reconstruction, but generates more high order harmonics (higher distortion).
Lower clock frequency is the opposite: less distortion, a bit higher noise floor, softer note contours, more “intimate” stage.
To give an idea, 12Mhz has lower THD vs 25MHz but is 1-1.5dB(A) worse in terms of SNR/DR.
The effect is indeed quite apparent especially if you have a trained audio ear already. If you are not particularly ahead in your critical listening experience, try “extremising” the values: slam all THD compensation to zero, set clock to 50MHz, and chose a minimum phase filter – you should hear all notes definitely more “sculpted”.
By the way, the fact that Tweak9038’s “Tube emulation preset” profiles are only defined at 12Mhz frequency is indeed consistent with the above: music comes across softer (not fuzzier of course, definition is still there), less “carved in stone”. Like tubes do.
Similarly to the other areas of intervention, for MCLK selection too Tweak9038 app allows to pre-set which clock value to use depending on the track’s sampling rate, and save the full association table under a custom named preset file, which one can load and apply at leisure.
For my taste, lower clock speeds are a better compromise on lower sample rates – and higher clock speeds “fit” (my tastes) better on higher sample rates.
Reconstruction filters: why we need them, and why Tweak9038 is cool
Takes as it should be, deeply understanding reconstruction filters would require a treaty on signal processing. If you are technically inclined an elementary starting point might actually be this Wikipedia page.
I wrote and rewrote this chapter a few times, was never happy of its contents as when reading it back I felt like I wuold not understand myself if I were to do it from what I had just written.
So in the end I spun it into a separate article. It’s here.
The ESS 9038Q2M chip adopted inside 9038SG3 and 9038D offer 7 different reconstruction filters (Linear Phase Slow and Fast, Minimum Phase Slow and Fast, Apodizing, Brick Wall and Corrected Minimum Phase) + 1 “ESS-Reserved” filter.
Assuming you read the above, or you know from before and even better than me, Linear Phase Slow and Fast, and Minimum Phase Slow and Fast filters don’t need much presentation I guess.
Brick Wall and Apodizing are variations of a Fast Linear filter. Corrected Minimum Phase is a not-very-slow Min Phase filter. The R (ESS-reserved) is similar to the Apodizing, but with less ripple.
Linear Phase Fast Filter
Linear Phase Slow Filter
Minimum Phase Fast Filter
Minimum Phase Slow Filter
Apodizing Fast Filter
Hybrid Fast Filter
Brick Wall (fast) Filter
9038SG3 and 9038D allow the user to freely select their preferred filter – which happens via the Tweak9038 app of course. Not so many other dongles offer the same possibilities (at any price, by the way).
Veeery widely said, I personally tend to apply Min Phase Slow to >=88.2KHz tracks, and Min Phase Fast to Redbook tracks.
What I find absolutely brilliant, and unique, here is that the Tweak9038 app makes it possible to map which filter is to be used based on the sampling rate of the incoming digital file.
So I can set things up such that (e.g.) on all tracks < 88.2Khz a Min Phase Fast is automatically used, while the DAC automatically switches onto a Min Phase Slow filter when resolution goes above 96Khz. Or whatever other pairing you might instead find best for your ears 😉
Standby and Mute
9038SG3 and 9038D have built-in automatic Standby and Mute functions.
Standby will set the device in Power Saving mode if it does not receive data from the host for 1 minute. This is of course very nice to reduce power consumption when you are not actively listening to music. Power Saving mode reduces consumption by 50%.
Mute will also turn DAC output off while the device is in Power Saving mode.
There is a single drawback in leaving Standby and Mute working their automatic way: when 9038 starts back receiving again data from the host while in Power Saving mode it may induce annoying and sometimes quite loud “pops” on the drivers. The more annoying and lower the higher the drivers’ sensitivity.
Standby and Mute can be turned “off” by accessing Tweak9038 Settings panel, and just tap to remove the flags on the two options.
Settings persistence and “Presets” management
This part may seem a bit confusing at first, at least it was for me. Let me try to make it simple and straight.
The Tweak9038 app makes all of the above illustrated 9038SG3 and 9038D parameters visible to the user, and allows the user to change them.
As soon as you plug either 9038D or 9038SG3 into the Android device running Tweak9038, its currently active parameters are read-in, and shown by the Tweak9038 app.
Whenever any change is made to one of the parameters shown by the Tweak9038 app while the dongle is connected, such new value is instantly saved onto the connected 9038SG3 / 9038D non-volatile internal memory.
Due to such non-volatility, all values will persist even after unplugging the dongle from the Android device running Tweak9038, and after plugging it onto a totally different device, regardless of such device’s OS.
So again: Tweak9038 shows the values which are written “inside the dongle”, and allows the user to “prepare”, “tune” his 9038SG3 / 9038D how he prefers, save the values back into the dongle, and use it, so tuned, wherever he wants, without ever needing the Tweak9038 app once again.
Clear till now? Good.
In addition to the above, Tweak9038 allows for saving “full sets” of such parameters. Such sets are however saved onto the Android device hosting Tweak9038. You do that by tapping on the Save frame, on the main app screen, and then giving the preset a name.
Again: Presets are not saved onto 9038SG3 / 9038D. They are kept local to the Tweak9038 app.
Existing Presets can be accessed by tapping on the Preset frame, on the main app screen. Once there, Presets will be found under 2 different categories – accessible by tapping on the first 2 buttons atop: User are those previously saved by yourself, and Official are those supplied by E1DA.
After tapping on a User Preset, it’s possible to SET, EXPORT or DELETE it.
By tapping on “SET”, the Preset’s values are written all together and all at once onto the dongle’s non-volatile memory, and saved. Exactly the same as if they were input one by one by hand.
User Presets can be Exported. Which is meaningful as Presets can also be later Imported – for instance in case a friend wants to pass us one of his sets, or to acquire a special Preset developed by E1DA which is not included inside the standard Tweak9038 app distribution set.
Official Presets cannot be Exported (no need to) nor Deleted (so you can’t mess up). Therefore, when tapping on an Official Preset the system just asks for confirmation before applying its values, and that’s it.
One last note: whenever a Preset is saved into Tweak9038’s workspace, all of the configuration values of the currently connected dongle are automatically saved onto the named file, including Min, Current and Max Volume figures. But, when a Preset is recalled from storage, and “SET” (applied) onto the currently connected dongle, the system asks wether Volume parameters need to be also Set, or those need to be left at their current “live” values.
Using 9038D with an external Amp
It is possible to exploit 9038D’s 3.5mm single ended phone out as a preamp out, and connect it to a downstream Amplifier. It’s quite logical to assume that, very likely if not always, 9038D will be connected to a desktop transport of sort for this application – like a Windows or Linux machine.
To get best results it is recommended to use the Tweak9038 app (see above) to:
- Apply appropriate THD compensation values
- Disable Mute and Standby
- Set Max Volume to -3dB (this way Max Vout will be 2 Vrms – this is only required if this is the max allowed input value for our Amp)
It is optionally possible to save a Preset for this, especially in case one plans to dedicate 9038D to this application only occasionally, and needs therefore an easy way to switch back and forth between these settings and others more appropriate for mobile use paired with headphone or IEMs.
Sole doubt is: how do I devine the “appropriate THD compensation values” to apply when 9038D’s load is represented by the amp? No worries.
The 9038D has a nice matte-finished black metal housing. I would call the front side the one with the E1DA white logo engraved on, of course. Let’s flip it to the back side. Near the end corresponding to the USB-C plug there are some other minuscule-font-size engravings. On my unit I read:
Calibrated unit: DR 124.7 / 125.4 TCC 2 / -70
I already mentioned far above that E1DA calibrates each and every unit upon manufacturing. These figures actually regard my own unit (so will in general be different from anybody else’s unit).
DR refers to my unit’s Dynamic Range, and the two figures refer to the left and right channel respectively.and mean:
Dynamic Range : 124.7 dB LEFT / 125.4 RIGHT
THD Compensation Coefficients: 2 / -70
TCC stands for THD Compensation Coefficients, and that’s what we are looking for now, as those figures are what’s needed on my very unit to minimise distortion on “No Load Condition”.
As I explained above, THD changes based on various dynamic situations, one of which is the impedance of what gets connected to 9038D’s output (the “load”).
So here E1DA is telling me what’s the compensation to apply when I connect my 9038D to… nothing ?!? Well let’s dig better into this technical wordage.
Connecting “nothing” to the output can be said in a more electrical-engineering-friendly way as “connecting to an open circuit”.
When voltage is applied to an “open circuit”, no current will pass through. “Of course… there is not even the effing wire!!”. Well… (again) an electrical engineer would rather say that there is a “wire”… with an infinite resistance (!).
And finally, amplifiers have very high Input Impedances. Not “infinitely” high, but “very” high nonetheless.
Now let’s connect the dots: those TCC values reported on the back of 9038D’s enclosure are the settings needed to minimise distortion when nothing is connected to 9038D’s output, i.e., when something with infinite resistance is connected there. So, they are a good approximation of the figures needed to minimise distortion when something with a very high – albeit finite – resistance is connected. Like an Amplifier, for example 🙂
How do I do that? Of course via the Tweak9038 app. Reach for the THD panel. Set both Low and High Threshold to -127.5dB. This way only Range 3 will ever be “active”. Open Range 3 panel and (in my specific case) set 2nd harmonic to 2dB, and 3rd harmonic to 70dB, while also ticking (only) 3rd harmonics Invert checkbox to reflect the “-” sign.
Should I plan to use 9038D as a “fixed volume” DAC, and only regulate volume on the downstream connected Amp, I would also want to set its Min Volume = Current Volume = Max Volume to -3dB, this way effectively forcing the device to always output 2Vrms flat.
MasterClock and Filter selection panels have nothing to do with the output connection, so I will leave my usual settings map in there.
And finally, I will save the whole thing under a User Preset called e.g. “9038D Dac”.
Heck! This way I overwrote my 9038D’s factory-imposed THD settings, those offering the least distortion when using a 32Ω IEM. And only now I realise I did not save the previous configuration into a User Preset before modifying it ☹
No worries. Original factory-recorded THD compensation data are hard coded into the firmware. By accessing Settings / Restore Factory Settings on Tweak9038 app I can swing back to those values in no time.
Using 9038SG3 with an external Amp
Some Amplifiers offer balanced input, and their owners would prefer exploiting that route, especially when the Amp also offers a full balanced internal structure.
Too bad that, simply put, such connectivity is not supported by 9038SG3. I found E1DA’s tech support explanation to why is this the case so efficient that I can’t find a better way then quoting that directly here:
DAC and Amp need to be grounded to each other to ensure safe operation. The headphone output of the 9038S has four pins: Hot/Cold Left and Hot/Cold Right. There’s no GND pin. Therefore, it is not possible to ground it to your Amp. People have ignored our warnings before, connected their 9038S to external Amps via a 2.5mm to 2x XLR adapter, and have fried their 9038S as a result. The output really is for headphones only. If you want something that you can use both with headphones/IEMs and with external DACs, get a 9038D.https://discord.com/channels/483873307251310592/608625162115612693/1001215598203768883
If you really don’t want to use the Tweak9038 app
If you are really really really unwilling to pay the 10 bucks for the app, then – even if you shouldn’t deserve it 😀 – there’s a B-plan for you: carefully flashing some pre-cooked firmware made available by E1DA themselves.
The firmware flashing package exclusively exists for Windows OS. Additionally, it requires Comtrue ASIO drivers to be pre-installed to operate correctly.
E1DA makes all the required sw packages available as free downloads of course, and they come with a collection of various alternative firmware versions ready to be flashed on either 9038D or 9038SG3.
It’s of course needed to select the files inside the folder called like the device model which is supposed to be updated of course.
Inside each “model” folder there’s a single file named “Tweak […]”. That is the firmware required for the Tweak9038 app to work. So basically, it’s the firmware that comes preinstalled from factory. Once one of the other firmware files gets flashed in, the Tweak9038 app will not be able to work on that device anymore, and to restore its functionality it’s required to re-flash the Tweak[…] firmware first.
The various files contain non-app-tweakable firmware configurations, quite clearly described by their names.
It’s worth nothing that:
- Firmware names containing “noSTBY” disable the automatic Standby feature
- Firmware names containing “SE” simulate a Tube amp sound signature (set 2nd harmonic to -50dB)
Special notes about iPhones (with a final Android hint)
Don’t be misguided [too much] by the above-mentioned power drains imposed by 9030D or 9038SG3 on the host, going well beyond 100mA which is known as a hard apriori limitation on Apple’s design.
In spite of that, both 9038SG3 and 9038D can be connected to iPhones via a Lightning connector cable but… whether it will work or not depends on the cable and on the iPhone specific model (!). Some models won’t ever work, other models won’t work with some cables (no power to turn 9038SG3 or 9038D up at all), and finally shall work with some other cables.
E1DA maintains an incredibly helpful if rustic worksheet – available here – collecting internal and users’ experiences: which cable works (or not) with which iPhone model. Check both the Yes/No and Voltage tabs to have the full information you want. The sheet also includes leads on where those cables can be purchased.
Once said it is possible, of course it stays legitimate to wonder whether it’s convenient to indeed connect 9038SG3 / 9038D alone to an iPhone, thereby heavily shortening its daily battery durability.
I myself use 9038SG3 (and Groove for that matter) on the go with “transport pack” made of a tiny DAP + a small lightweight powerbank, kept together by some Bluetack and connected to the dongles by means of a special Y-USB cable.
But that’s another story, and applies to quite a few Android phones too. This specific topic will be covered by a separate article of mine which will be published, well, you know already when.
Considerations & conclusions
You really got through all this article reading it all till here? Heck! I owe you a coffee at the very least. You deserved it.
As I hope I suceeded in saying and explaining why, E1DA 9038SG3 is a very good battery-less DAC-AMP, and an even more brilliant product overall. Unbelievably powerful, drives all nastiest IEMs including no matter which planar IEM, and most HPs as nothing – surrenders only against the most demanding planar overears. This, while offering outstanding sound clarity and superb detail retrieval.
9038D is also extremely good, indeed as good as 9038SG3 sound wise with the sole caveat of output power capabilities which are lower than 9038SG3 but still higher than 90% of the other mobile sources around. Can be the right choice for fixed-cable IEMs and anyhow most other IEMs around bar (some) planars only.
The Tweak9038 app further allows to “play” with the device timbre.
For how much ya’ll know I love my two Grooves, if I were to recommend “one” dongle only for true audiophile use at the lowest possible cost I would name 9038SG3 without the shadow of a doubt. Excellent, indeed.
As already mentioned at the beginning, the 9038SG3 and 9038D I am talking about have been personally purchased.
Very much lastly: I’ve had as always loads of fun going through all this with my audiophile friend Simone Fil, also an enthusiast E1DA user, and much deeper than me in technical audio competence – which of course I ruthlessly exploit. Quite some of the above content is the direct descent from our late evening chats and common findings.
P.S. – last minute news
On their Discord channel Ivan just recently announced he found a good successor for 9038D’s opamp. Same power, even a bit better THD and DR. It’s code-named 9038D6K as it will also have 6000µF power filtering capacitance, up from 4000 on og 9038D.
So it seems 9038D will be available again… Soontm.