EARLY LOOK: E1DA Cosmos ADC - affordable high performance measurements for the audio hobbyist!

As you can see in the picture above, I have one of Ivan Khlyupin's (aka IVX) Cosmos ADCs. It's not available yet on the market, so watch for when it's released from E1DA in the days ahead. (I see that variants are available on Amazon now.) Thank you Ivan for reaching out and sending this unit for me to use!

[If you're wondering about the E1DA name, as explained by Ivan, it comes from the pronunciation of "Ivan" as sounding like "E1" in Russian as opposed to the Americanized "Eye-Ven". "DA" as in "dah" ("да", "yes") - so it means "Ivan Yes".]

While this unit is one of his builds for external testers, I suspect that it should be much the same as the final product when available hopefully later this month. Obviously there could still be some changes with the final production release.

I think the price is slated for around US$150. Also, in the days ahead, E1DA will be releasing the "APU" (Analog Processing Unit) that can complement an ADC for measurements with the ability to notch out the 1kHz fundamental among other features allowing even more accurate measurements of very high performance gear. And there's also the Cosmos DAC coming as well to complete the "trinity".

For today, let's have a peek at the Cosmos ADC. Let's explore how to get it going, some of its features, and although still early days, we can take a quick look at measurements with this tool, with some DACs I have here.

In the picture above, we can see that the main inputs for the ADC are the dual balanced XLR connectors. Notice that the markings indicate that they can have input levels up to 10Vrms (+22dBu). The input level can be adjusted to optimize the dynamic range when you're measuring devices of lower voltage. You adjust this with some DIP switches at the bottom of the unit (I have been told that these are high-quality gold plated switches, guaranteed for minimum 1000 cycles):

You can see that I received test sample "3".

As indicated on the enclosure, one can change the input level for each left and right channel. If you've measured DACs before, you've probably seen that many balanced outputs are 4Vrms so the 4.5Vrms configuration would be a good one to use that optimizes potential dynamic range for such devices.

The construction is robust, it feels like aluminum with good laser-etched lettering. Dimensions are 6cm wide x 10cm deep x 3.25cm tall, and the weight is 155g.

Notice that at the lower left on the front panel, there is also a 2.5mm TRRS "Aux" jack which can be fed up to "30V" as printed on the front. In our E-mails, Ivan clarified that the actual range for the "Aux" jack correlates with the dip switches for the XLR. So at XLR 1.7V setting, the "Aux" port will hit 0dBFS at 34.4Vrms. At 10V XLR setting, 0dBFS is at 43Vrms (variations between these values with the intermediate settings I presume). This means there's the opportunity to measure power amplifiers such as up to 462W into a 4Ω load when set to that 10V XLR configuration. Furthermore, Ivan mentioned the potential to use this TRRS input for calibration of his DACs (for example, you can see a video here that shows fine tuning for THD compensation on the E1DA 9038SG3 and 9038D using the "Tweak9038" Android app). We'll stick with the XLR input for the tests today.

Here's a look at the rear of the unit:

Some specifications etched back here.

Notice the single USB-C connector, there's a small amber LED that blinks when in use. The blinking rate is dependent on the sample rate: 48kHz = 0.5Hz (1 blinks/2s), 96kHz = 1Hz, 192kHz = 2Hz, 384kHz = 4Hz, and the 44.1kHz family blinks slightly slower by about 10% (ratio 44.1/48 x frequency).

One "controversial" characteristic to this design which I need to add here (as you can see in the comments, I had neglected to do so in the first release of this article), is that this device does not incorporate an input buffer which results in variable and relatively low input impedance. Here's how it looks like depending on the voltage dip switch settings:

1.7V (6.83dBu) = 640Ω         2.7V (10.85dBu) = 1kΩ
3.5V (13.1dBu) = 1.3kΩ        4.5V (15.28dBu) = 1.66kΩ
6.7V (18.74dBu) = 2.46kΩ    7.6V (19.83dBu) = 2.82kΩ
8.5V (20.8dBu) = 3.12kΩ        10V (22.22dBu) = 3.48kΩ

The reason for the lack of the input buffer in my discussions with Ivan are multifold including price target, complexity of circuitry, and also maintaining the best performance with the ADC chip (the ES9822 which we'll talk about below).

Like when mismatching impedance of amps with speakers/headphones, you will see a voltage "droop" if you connect high impedance devices into this ADC relative to the impedance table above. Yeah, this will have an effect on the results; the pragmatic question is how big of an impact this will have and that I suspect will be an interplay between the ADC and your test device and its level of performance. I'll certainly monitor this in the days ahead as I try out different gear and obviously I will still have my RME ADC to compare.

When plugged in, the device is recognized automatically as USB Audio Class 2 (UAC2) audio input so it can be used by Windows without special drivers:

As you can see, Windows will list up to stereo 32/384 PCM capabilities. I didn't test this with Mac or Linux which I believe should work fine.

You will need to install the ComTrue 7601 USB Bridge driver for ASIO; I used the current version 6.0.1.20210429. I had no problems with the ASIO driver in WaveSpectra. But ran into various software complaints in ARTA, and Room EQ Wizard. I eventually settled on ASIO4All in REW which worked fine up to 192kHz.

Software utilizing MME (RightMark Audio Analyzer, and SpectraPLUS-SC), WDM (ARTA) worked fine as well. And I don't anticipate issues with WASAPI (no problems in WaveSpectra).

Here's how REW looks with ASIO4All and the Cosmos ADC listed as input device:

Using ASIO4All - note Cosmos ADC as input. For the record, I'm using REW 5.20 RC9 currently. I haven't updated in a few months as this version has worked well for me thus far. Ivan warned that some builds might have accuracy issues when calculating THD(+N) so be mindful of any unusual results.

The heart of the unit is based on ESS Tech's recent converter chip, the ES9822 PRO (see brief datasheet here, more detailed info requires an NDA). This chip was first made available in late 2020 and is described by the company as "the world's highest performance 32-bit analog-to-digital converter". It's a 2-channel device, with the datasheet claiming -117dB THD+N per channel in stereo mode, and -118dB THD+N in mono mode where the two channels are combined. Based on what I'll show below, these numbers appear conservative!

I mentioned above that E1DA's DACs can be fine tuned with THD compensation, so too this ADC using the "E1DA_Cosmos_Tweak" software. Here's what it looks like when I fired it up and clicked "Connect&Read" to get the default parameters:

The default numbers for 2nd and 3rd harmonics were calibrated prior to the unit being sent to me to correspond to results on the Audio Precision SYS27XX with a -0.5dBFS signal. You can also change which filter setting is applied to the ADC channels; "Linear phase apo" and "Linear phase fast" both worked well in my testing.

Notice the slider below the "2ch MODE" and "1ch MODE" labels. The switch between mono and stereo modes is linked to the UAC2 volume level. If you set the slider here to <50% towards that -77.5dB side, it will switch to "stereo" mode, towards 100% is "mono" mode. You can also change the volume setting in the Windows "Device properties" section for the input level which can be saved as default.

The "E1DA_Cosmos_Tweak" software will adjust the values in real-time so you can see the effect they have on the FFT. You can save the values (click "Connect&Read"). The only thing that's not saved here is that slider between stereo/mono modes.

What good would an "EARLY LOOK" at the ADC be without some measurements?! ;-)

Before we look at an actual DAC output, let's have a peek at the noise floor for this ES9822-based ADC compared to the RME ADI-2 Pro FS which I have been using with its internal AKM AK5574. To compare similar settings, let's put the RME ADI-2's input level to +19dBu and the Cosmos ADC dip switches to 6.7V making the peak input level about equivalent - here is the FFT for "silence":

Because ASIO4All with REW is limited to 192kHz, the comparison above goes up to 96kHz. Notice how flat the Cosmos is compared to the increasing ultrasonic noise seen with the RME's AK5574. In practice, notice that the AK5574's noise floor is still below -120dBFS by 100kHz so it's not like this level of noise would be at all problematic when doing music recordings. However, when we're measuring gear, it's nice to maintain that flat noise level so we can see potential distortions that could be obscured. Despite the fact that I'm running the Cosmos ADC off my computer (Intel NUC 6i5SYH) with no special USB isolation, this remains clean.

WaveSpectra allows me to use the ASIO driver at 384kHz, here's what the noise floor looks like extended to 192kHz (131k FFT):

Beautiful.

As I noted last week, I have the Topping D90SE DAC here for review, but at this point, I have not run detailed measurements yet. Instead, let's start by looking at something like the Topping D10 Balanced which was discussed and measured a month back.

We know that the D10 Balanced DAC output goes up to around 4.25Vrms. So I would start with setting the Cosmos ADC to the appropriate input level. As such, let's flip a few dip switches and set it to the 4.5V configuration like this:

The dip switches are pretty small so make sure you have your reading glasses on if needed ;-). I used a wooden toothpick which worked well and won't scratch like a metal pin.

Notice in the picture above the small Cosmos ADC LEDs are green when there's a signal (it goes dark when no signal), indicating that it's functioning in stereo mode when I took that picture.

While in stereo mode, we can easily flip between the right and left channels as input captured above. If we compare this to previous measurements using the RME, indeed we can see the Cosmos ADC is able to achieve cleaner results and appreciate the relative differences between the two channels.

To get even cleaner results, we can turn on "mono" mode for the ADC, play the 1kHz 0dBFS tone through both XLR outputs, and achieve even lower noise level - at the expense of averaging the 2 channels in this configuration:

-118dB THD+N is what was found using the Audio Precision APx555 for this DAC. An impressive start and it would not be unreasonable to think that the Cosmos can be pushed even further when paired with an even higher fidelity DAC.

As per the comments below, one could also look at a single channel at a time in "mono" mode by using a cable splitter and presenting the same output to each ADC channel; this will give you a better look at individual channel performance of course.

Even though 1kHz THD+N provides a convenient benchmark of sorts for DACs, obviously we should look at other (likely audibly more important) characteristics. As an audio device on the computer, the Cosmos ADC can also be used with RightMark Audio Analyzer. Here's what I got for the Topping D10 Balanced at 24/96 (stereo mode, otherwise cannot measure crosstalk):

Again, nice and clean measurements. The apodizing filter does roll-off earlier just above 40kHz, so in the future, I might set the ADC channels to "Linear phase fast".

Next, we can have a look at the J-Test and note how well the Topping D10 Balance performs on jitter!

The very low noise floor accentuates the jitter-modulation least-significant-bit square wave in the 16-bit J-Test. Seriously folks, no need to worry about jitter these days even with inexpensive DACs!

If we compare the 24-bit result above with the RME ADI-2 Pro FS, we can see that the RME's graph "skirts" a little bit at the base of the 12kHz signal, plus the noise floor is higher:

Obviously, it would be ridiculous to be concerned about that tiny "skirt" in the RME measurement down at -140dB when used for audio recording! Nonetheless, its absence is indicative of the Cosmos' accurate temporal performance. An impressive demonstration of the precision of this tool.

For fun, let's rescale my 24-bit J-Test to compare with the ASR graph, 16 average, 256k FFT. I prefer a wider aspect ratio for esthetic reasons. Cosmos ADC set to 2-channel mode looking at the R channel only (arbitrary, L channel looks about the same):

We can easily see the 24th-bit 250Hz square pulse. Very low noise level below -160dB. A pair of sidebands around 12kHz and also a pair +/-6kHz below -150dB. Nice and flat noise floor.

While the Cosmos ADC is a high-resolution ADC intended for balanced input, you can indeed put some XLR-to-RCA adaptors on and measure. For example, here's the Topping D10s plugged into the Cosmos ADC:

Notice the blue Cosmos ADC front LED color - the device is in MONO mode when I took the picture. Since I know the DAC output is ~2V, I set the Cosmos dip switches to the 2.7V level.

Right and left channels THD+N:

The device measures better than -111dB THD+N, again cleaner than the RME ADI-2 Pro FS result previously mainly due to better noise floor. And if you compare the harmonics, we can see that my measurements show a similar structure of relative peaks. If we compare this with ASR's measurements, we see that the 3rd order harmonic predominated there as well. Likewise, measurements show the highest distortion peak to be less than -120dB.

As a note when comparing measurements between different DAC units, we need to anticipate some variability especially when we're looking at such low distortion levels! I think it's important to remember the provenance of the device as well. The ASR measurements were done on a DAC supplied by the company so it's worth being mindful of potential selection bias. If I, as a manufacturer, send a DAC early on around the time of release to be measured, I would probably "cherry pick" a better performing one. In comparison, the unit I have here was purchased from retail channels a bit later.

Switching the ADC over to "mono" mode to average the two channels now:

We see averaged harmonic peaks, again the highest distortion component comes from the 3rd harmonic. All harmonic distortion peaks below -120dB of the fundamental. Again, consistent with my previous measurements of the D10s with the RME ADC and generally what was seen by ASR. Nice to not see any 60Hz hum despite the RCA connection/cable.

Early impressions...

As an audiophile who's targeting my writings in this blog to those who desire high-fidelity audio reproduction, the E1DA Cosmos ADC represents truly a "new world" of opportunity to go beyond the usual subjective evaluation of audio gear at a minimal cost of admission. This is as good an invitation to democratize and empower high-quality measurements by hobbyists as I have ever seen.

Already, this device has opened up for my testing an opportunity to peer even deeper into lower noise levels and examine distortion amounts with greater resolution. As suggested in the title of this post, I consider this an "early look" into the potential of the Cosmos ADC. In the days ahead, I will be incorporating this tool into my reviews and other measurements.

In terms of opportunities to improve the product, on the hardware side, while I really like the different voltage levels available, the tiny dip switches can take a little time to set. Physically larger switches would be easier, but ultimately some kind of 8-step hardware button or robust software switching would be most convenient. Obviously this would add cost, potentially quite significantly in order not to degrade performance.

I've had no problems with the native Windows UAC2 driver. Software that use MME, DirectSound, WASAPI seem to be fine. Working with the ASIO driver is however a bit of a hit and miss depending on what software is being used. ASIO seems to work well with WaveSpectra. The ASIO driver doesn't load with SpectraPLUS complaining of no output ports. REW doesn't like it when asked to use the ComTrue ASIO driver and will perform a "soft crash" that allows me to still save out stuff I'm doing but requires a restart. However, using ASIO4All, with input pointed at the Cosmos ADC works fine and this allows different in/out devices in REW. Unfortunately, doing this also limits me to 192kHz in REW. ARTA works well using the WDM driver, but I find the UI, measurement mechanism, and overall display of the FFT/data clearly less refined than REW.

Ideally, multi-device support in REW with WASAPI and/or ASIO natively would be fantastic. I must say that despite the age of WaveSpectra, it's quite impressive how well it handles the hodgepodge of audio driver variants in Windows!

Do not forget that the opportunities as an audiophile to explore with tools like this is not without personal cost. What I mean is that you'll need to take time, learn and challenge yourself with the audio technology. Learn how to read the measurements, even better, perform them yourself. Along the way, you will learn how to best set up a computer measurement system, appreciate what does or does not add to noise levels (for example, you'll easily see if a cable makes a difference, expensive power supplies, fancy streamers, ethernet switches and DAC output, etc.). Most importantly, you will be able to identify "snake oil" by virtue of some things obviously making no significant impact at all which you can then verify with listening. As with other things in life, taking up a challenge, obtaining knowledge (not just hanging on to opinions, or mere impressions), can yield great dividends. If we are open to changing our perspective to wherever the experience and data leads us, I believe this will make us better audiophiles, and people in general.

Beyond home audiophiles, it would be remiss of me to not suggest that this ADC would also be a great tool for all those audio DIY'ers looking to verify their circuit designs. Heck, this level of performance might even be more than good enough for professionals who want an inexpensive measurement tool as a preliminary "screen" for low noise and low distortion before sticking their designs on a calibrated Audio Precision.

Speaking of Audio Precision, obviously the AP gear provides many more features in hardware and software (see the previous post on the innards of the APx555). From what I have seen thus far, under typical circumstances, it might not be unreasonable to suggest that the Cosmos ADC can perform low-noise measurements down to a level between the AP SYS27XX and APx555 models. Obviously there are many factors to consider and as usual, the "devil's in the details". As a guy who appreciates good value, to speak of this less-than-US$200 ADC in the same breath as measurement devices orders of magnitude higher in price is I think simply a reflection of what technological progress can achieve! Smaller, less expensive, potentially better in some ways.

Although I could wrap up the article at this point, I simply could not resist showing one final set of results. No fooling around, let's have a look at this:

Updated September 14.

Let's use an inexpensive XLR splitter to present the same channel to each input for mono. I put a little red tape on one end just for consistency in case there are very tiny differences between the lengths.

Updated September 14.

That's my retail Topping D90SE DAC at 4V (not even the higher 5V setting, or approaching the Cosmos ADC's XLR 10V upper limit), -1.9dBFS level (not even ideal -0.5dBFS), no 1kHz analogue notch filter for more precise examination of the harmonic residual (ie. no APU), captured unweighted comparing "stereo" and "mono" modes. In stereo mode, this is already showing THD+N of better than -119dB! In mono mode, both channels have THD+N better than -121dB, minimum "skirting" only below -140dB, harmonic distortion peaks less than -130dB.

Ladies and gents, fellow hi-fi audiophiles - when it comes to achieving "Good Enough" quality for human hearing, I hope everyone appreciates that we're assuredly beyond that at this point in the history of audiophile DACs at very reasonable price points!

More about the Topping D90SE in the days ahead.

While I have not tried it, I don't see why one could not use this ADC to record music. The issue is that you would still need to plug microphones into a pre-amp, and a standard USB audio interface with features like phantom power will be much more convenient and do the job well. There's a wealth of options out there from M-Audio, Presonus, Focusrite, RME, etc. I see this ADC as a specialized high-resolution tool which is why in future reviews/measurements, sometimes I'll use the Cosmos ADC, other times, the RME ADI-2 Pro FS could be more convenient. I see this as no different than selecting the appropriate software depending on what I'm testing.

Thanks again Ivan!

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

That's the "Cosmos Load" board. I'll leave it to you to figure out what that's about. ;-)

Head over to the Discord channel for further discussions/info about E1DA products...

Article updated September 12, 2021:

- Added information on impedance.

- Added both "stereo" as well as individual channel "mono" measurements for Topping D90SE for comparison.

Article updated September 14, 2021:

- Busy at work but found some time and updated the Topping D90SE stereo and mono measurements tonight. Cleaned up some anomalies I found in the last couple evenings when experimenting with cable and software configurations. Discovered a couple things about the D90SE and measurement with Cosmos ADC which I'll talk more about when I write about the D90SE in the weeks ahead.

Addendum September 15, 2021:

Hey guys, as per RJA4000's suggestion, here's some data from my RME ADI-2 Pro FS - stereo mode and mono mode with XLR splitter measuring the Topping D90SE. Same cables used as the results above with the Cosmos ADC:

The mono processing is a bit better, slightly more than 1dB of THD+N improvement with lower noise level using the Topping D90SE.

Okay, now as for using a low 1Vrms signal (I turned the D90SE volume down to -12.5dB), and the lowest input level for both ADCs - "+4dBu" on the RME ADI-2 Pro FS and "1.7V" (+6.83dBu) on the Cosmos ADC.

For best results, let's just look a the mono processing mode:

This is where the ability of the RME to set a lower input level and higher impedance achieves an overall better noise level. However, if we look at the FFT itself, notice how on the Cosmos the distortion level is significantly cleaner with THD better than -129dB (similar to the 4V graph above), and there's no "skirting" at the base of the 1kHz tone.

Finally, I wanted to check at the 1V level, despite differences in ADC input levels and impedance (for the record, the Topping D90SE output impedance is 100Ω), let's just make sure frequency response remains flat as one would expect using both devices:

Yeah, no problem and I would say equivalently flat from 20Hz to 20kHz (24/96 sampling, hence up to 48kHz). "Linear phase fast" roll-off filter used with both RME and E1DA ADCs.