The E1DA Cosmos Stack: ADCiso and 2 APUs. An example of bad multichannel - Neil Young's 'Harvest' on DVD-A (2002), fixed in new Atmos mix.

E1DA Cosmos ADCiso Grade A center.
Cosmos ADC Prototype I've been using over the years to the left.

Readers here probably are aware that over the last few years, I've been using the E1DA line of devices, the Cosmos ADC, APU, and Scaler typically paired with PC-based software like Room EQ Wizard for most of my measurements, especially for the highest fidelity products like DACs. As a hobbyist exploring objective performance, there's obviously no need to spend thousands of dollars on devices like the Audio Precision (here's a peek inside the APx555B top-of-the-line model, around US$30k). Not enough return on investment unless one is doing professional product design and testing.

As usual, over time things evolve around here and since I needed an extra ADC for work-related purposes, I bought one of the newer E1DA Cosmos ADCiso Grade A to replace the prototype that Ivan sent me back in 2021 which I will repurpose elsewhere.

This post is mainly an update and comparison, refer to the original article on the Cosmos ADC for more details.

Here are some pictures for comparison. As you can see, externally, the differences are mainly just cosmetic. The exception is the rear where the ADCiso has two USB-C connectors:

Note red tape on the ADC Prototype from 2021.

ADCiso conveniently has the balanced input impedance listed (single-ended about 30% lower) and more explicit dip-switch graphics.

Very important "power first" comment for ADCiso.

As you can see, for the ADCiso, the data and +5VDC (<250mA) power USB-C connections have been separated to allow for better 'iso'lation between the digital data to/from your computer and noisy power and ground loops can be ameliorated with a separate power source like a battery. By doing this, in principle, we should be able to avoid needing USB isolator devices as I had previously demonstrated (such as the Topping HS02 for USB2 which is compatible with the E1DA ADC, or Intona USB3 isolator which unfortunately did not work with the E1DA ADC itself). Make sure to power up the ADCiso first before turning on your computer or connecting the data cable otherwise you won't get an input signal even if the computer sees the device.

If you already have a good isolator (eg. Topping HS02) and don't want to use a separate battery, then a USB-C splitter cable works fine and basically converts the ADCiso → ADC.

The ADC chip is the ESS ES9822Pro and in the ADCiso, the USB bridge is the Comtrue CT7601PR variant with 768kHz capability. When buying these units, note that there are different "grades" based on noise and SINAD/THD+N defined as such:

Grade    SNR (dB A-weighted)    SINAD (dB)

0            >129

A           128-129                            >123dB

B           126-128                            >122dB

C           >125

Note: Measured in MONO mode, SNR is A-weighted, and SINAD measured with 1kHz tone at -0.5dBFS, 48kHz, 20-20kHz.

As you would expect, the higher grade devices like 0 and A cost more than the B and C units. As with most things in life, there are diminishing returns. Assuming everything else is in order like flat frequency response, excellent linearity, miniscule jitter, would being able to measure an excellent DAC using a device with SNR >125dB(A) [Grade C] be any more audibly meaningful than accurately capturing SNR >129dB(A) [Grade 0]? I doubt it, but as a measurement device, it's always good to have better capabilities.

The ADC is calibrated from the company. I find it worthwhile grabbing the default parameters using Cosmos_Tweak just in case they get messed up or if I recalibrate. Here's what mine looked like:

My E1DA Cosmos ADCiso default values. As with DAC playback,
I typically run the filter at "Linear Phase Fast" although "Apo" is the default.

For very high-resolution devices, when it comes to 1kHz SINAD/THD+N and the dynamic range measurements, I'd be using the Cosmos APU in conjunction anyways, making the "grade" of the ADC less significant.

Let's look at the noise characteristics between the original prototype Cosmos ADC and the new ADCiso Grade A. The ADCs were set to MONO mode (volume control set to 100%), nothing connected to the inputs. Note that my settings are not the default since I typically measure at 96kHz (E1DA grades the device at 48kHz sample rate) and I have the input level at 6.7V instead of the default 4.5V out of the box:

Note the FFT parameters I typically use: 128k bins for 96kHz, B-H 7 windowing running at 32-bit/96kHz. I prefer 96kHz as a hi-res standard, allowing us to see if there's any kind of down-sampling happening and check for excess noise an octave above audible. Consider these my defaults used over the years unless specified otherwise.

Yup, the Grade A results are better. The ADCiso Grade A is quieter with more of the noise level in red below the purplish ADC Prototype. As you can see, the ADC Prototype I've been using these last few years would be an "upper Grade B" model (-127.6dB A-weighted) and the difference isn't big compared to the new ADCiso at -128.2dB(A).

Let's now compare the Cosmos ADC and ADCiso (no APU) with this stack of DACs - the Topping D10s (single-ended), Topping D10 Balanced, and the Linn Klimax DS/2 EXAKT variant (XLR-out):

The most challenging DAC is the Topping D10s because it's unbalanced and completely USB-powered. That loopback set-up from Ryzen 7 MiniPC → D10s DAC → Cosmos ADC → MiniPC can be prone to ground loops and other electrical noise.

To demonstrate issues, let's look at a 1kHz 0dBFS THD+N with the E1DA ADC connected directly to the computer (noisy) vs. in conjunction with the Topping HS02 USB isolator vs. the new ADCiso Grade A with a standard 5V lithium battery pack without USB isolator:

Notice left channel noise floor very slightly higher in the Prototype
though still down around -145dB.

As you can see, without a USB isolator, the E1DA ADC (prototype) was prone to quite obvious noise that was nicely ameliorated with the Topping HS02 USB isolator. The ADCiso is able to perform well without using the isolator. Yes, I could still plug the ADCiso into the Topping HS02 + separate battery but there would be no significant benefit to doing this.

Notice that the ADC Prototype + USB isolator and ADCiso both gave an equivalent THD+N result averaged to -109dB both channels. 2nd and 3rd harmonics look similar with good even harmonic (4, 6, 8) suppression. Some variation in the higher frequency harmonics and low-level noise but we're looking at stuff below -130dB and this is single-ended output which is prone to picking up interference.

When we move over to balanced DACs, even when USB-powered off the computer, the situation changes markedly. Here's the Topping D10 Balanced:

Even without a USB isolator, the balanced signal remains free from excess noise. We've seen this over the years in USB isolator device testing (for example the Topping HS01 test).

Both the E1DA ADC and ADCiso showed a slightly higher right channel noise floor at lower frequencies and slightly higher left channel noise at higher frequencies with this particular Topping D10 Balanced. Not that we would ever hear the difference!

So you must be wondering about that Linn Klimax DS/2 'EXAKT' - here's the XLR balanced out. I'll just show the ADC Prototype + Topping isolator vs. ADCiso Grade A:

While the Linn streamer is an older model (circa 2015-2016, prior to the Katalyst architecture like in the DSM/2), it is much more expensive than the Topping DACs above which are just US$100-150 DACs. As you can see, the 1kHz 0dBFS measurement did not pose much of a challenge to the measurement E1DA Cosmos ADCs.

I also bought another E1DA Cosmos APU for work-related purposes:

Note the red soft plastic piece used as protection against accidentally switching on the +48V phantom power.

With two of these, it'll allow me to run 2-channel THD+N/SINAD and SNR measurements simultaneously. Other than mild calibration difference in the compensation curve between the original unit and this one (discussed in my 2022 article, essentially just a 0.05dB peak difference), there's no other noticeable change between the early APU and this new one.

So let's use the APU's notch filter and grab some results from the Sabaj A20d 2022 DAC which is based on the ESS ES9038PRO chip.

On the right are a couple of lithium batteries to power the 2 APUs, Scaler, and ADCiso!

A closer look at the 2xAPU + Scaler (200kΩ balanced / 100kΩ unbalanced input impedance) + ADCiso - "The E1DA Cosmos Stack":

Short <6" Phono-Phono cables from APU out to Scaler in to reduce risk of noise.

Here are the results from "The E1DA Cosmos Stack", THD+N and SNR of the Sabaj A20d 2022:

Beautiful. Now that's a hi-res DAC.

This result of -124dB THD+N and almost 130dB of dynamic range is pretty well state-of-the-art among consumer DACs with 4Vrms balanced output. Of course these days we have even higher THD+N/SINAD devices based on newer ES9039 chips (with other beneficial characteristics like this) that can achieve THD+N -125dB and lower (like even the little #9039S USB dongle!).

Remember that by itself, a single 1kHz 0dBFS THD+N measurement below say -100dB among otherwise good quality DACs is unlikely to be meaningfully audible (see THD listening blind test). There's no need to chase the best THD+N/SINAD number; it's just one of a number of parameters we can look at as evidence of low-noise, low-distortion, and high quality engineering.

So there you have it, an update on my measurement kit as we go forward using the E1DA Cosmos ADCiso Grade A. The results obtained with this device would not be much different compared to the ADC Prototype I've been using already (<1dB variation). It's simply refinement of the hobbyist audio test kit that doesn't cost too much and will yield results well beyond the limits of human hearing. Hmmm... in some areas, a simple test kit like this might exceed APx555B capabilities; if only scripting capabilities could be added to REW to help automate testing 😉.

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

To end this post, I thought it would be interesting to discuss an obvious example of bad-sounding multichannel!

As you probably know, I'm an advocate for multichannel audio (Atmos being a modern variant) due to the ability to better convey 'immersion' and 'realism' in recordings compared to standard 2-channel high-fidelity.

While there are many excellent-sounding multichannel recordings, of course nothing's perfect and there are some pretty atrocious ones out there as well. A good example of a bad-sounding 5.1 mix is Neil Young's Harvest (1972) that came out way back in 2002 on the DVD-A format as a 30th Anniversary edition.

I was quite disappointed when I got this DVD-A years ago so hadn't listened to it for awhile and was reminded more recently to give it another try. Yup, my original memories were correct - the sound sucked. A disservice to multichannel audio and ultimately a disservice to an excellent album. I had been wanting to have a better look at why this is for awhile, so let's rip it with DVD Audio Extractor and have a peek.

That's obviously very weird. Notice how strong the left and right surround channels are! With most of the audio coming from behind, it comes across as highly diffused and unfocused. In comparison, the front left/right, especially the right channel is downright anemic - why!? Did Neil Young listen to this himself before it was released? 🤨

I suspect whoever did the mix or DVD-A disc mastering back in 2002 had possibly flipped the front and surround channels around, resulting in an extremely "rear heavy" sound. If we just moved channels 5/6 ↔ 1/2, this could probably sound quite reasonable (haven't tried).

Interestingly, the track "The Needle and the Damage Done" (Live) on the same DVD-A turned out alright with a more normal front presentation as it should be if attending a concert:

How strange. Does this mean that whoever remixed the other non-live tracks intended to blast the rear channels!?

Thankfully, issues and errors get fixed. Over time, remixing happens and we end up with better versions as the tools and techniques also evolve. Here's the current Atmos mix (down-converted to 5.1) from Apple Music, I believe this has been available since 2023:

Definitely more reasonable with normal front left/right content, low-amplitude center channel is not unusual although I personally like music to use the center more for anchored vocals, minimal LFE (subwoofer unnecessary for this track), and generally lower-level left/right surrounds mainly for ambiance although the artist/mixer might elect to put instruments/vocals back there.

The Neil Young example I think is interesting because of the obvious channel anomaly and not necessarily the recording itself ("lo-fi" like say Weezer's [Blue Album] available in Atmos). If you like Neil Young's material, many of his albums have been released in multichannel/Atmos already. One standout for me is Harvest Moon (1992, DR13 original 2.0, DR14 multichannel/Atmos down-folded to 5.1) with noticeably cleaner vocals and harmonica such as on the title track. The less heavy-handed application of reverb is very noticeable and much appreciated! Also, as expected, the "spatial" surround had more natural width and depth, less "crammed" feeling and allows for a more nuanced appreciation of the instruments compared to the 2-channel mix.

[I know Neil Young likes analogue to the point of making senseless claims against digital audio. He might be a good musician, but historically has been questionable when it comes to understanding audio technology. Nonetheless, his multichannel/Atmos versions sound pretty good and generally, IMO, are significant upgrades from 2-channel digital and vinyl mixes for fidelity and immersion.]

Hope you're all enjoying the music and having a wonderful June as we approach the solstice this year!