I really don't think the science of how subjectively beautiful a sound is is anywhere close to being advanced enough for a company to be very deliberate and precise. I think it's more of a try a bunch of stuff till it sounds nice type of thing. I think the makers of ladder dacs have inherently different design philosophies. But I have never designed a multibit DAC, or interviewed someone who has about their design philosophy, so I am really just making assumptions here.
I do know that the guys at Schiit have been very vocal about not designing the modi multibit around it's measurements. They claim they already did that with the regular modi. They prefer the modi multibit because it sounds better to them. As for how precise they were in crafting the details of their multi-bit sound, I do not know.
I think it is worth keeping in mind that resistor ladder DACs are valued not for how accurately they measure but for how pleasantly they distort. They are basically the DAC version of a tube amp. The value of a tube is not in how close it can sound to a solid state but how pleasantly it can twist the sound to the listener's preference.
For this reason, I am of the opinion that measurements are actually quite useless in judging these types of products that are engineered toward subjective tonal and timbral preference instead of objective performance.
Hi
@Noob - first off apologies if this is redundant; I scanned the thread but didn't notice someone making exactly the same point however my observational powers are sub obtimal.
Analog devices (ADI) makes the actual DAC inside the Yggdrasil, the AD5791.
The AD5791 datasheet is here. They market this as an industrial / measurement DAC so they don't publish FFTs, but there are 3 main measurements that strongly indicate the ENOB/SINAD/SNDR (all the same measurement) would be basically ideal (within 1 LSB or 1 "bit" of the total 20), assuming one reproduces the same test conditions that ADI did when characterizing the part (looks like it prefers a fairly high and clean voltage supply, and given Schiit's stance on power supplies and
@amirm finding AC mains spurs in the FFTs I wouldn't be surprised if the reference quality was the main problem).
Reason 1/3: On page 15/27, you'll see the output of the DAC fully settles to its final value under all measurement conditions within 1-1.5 microseconds. This means that it will work at any sampling rate below 500kHz - 1MHz. I can't find it now but I remember when reading about Schiit's proprietary filter that they do 8x oversampling (and their proprietary novelty is that every 8th sample is the original contained in the source), so if you're sending 44.1 - 96 kS/s data then the sampling rate is not the problem. Maybe there is settling-time related degradation with 192kS/s material.
Reasons 2 & 3: On page 9/27, you'll see INL and DNL measurements (Integrated Non Linearity and Differential Non Linearity;
you can learn about them here starting on page 5.5). TLDR is if the DNL and INL are less than 1 LSB across the entire input range then you have succeeded in designing an N-bit converter; if you made an N-1 bit converter it would be worse.
To my knowledge none of the 24-bit converters actually measure 24 bits of ENOB. The best DACs
@amirm has measured get between 20-21. The DAC chips (or ICs, integrated circuits) inside of them also only get about that, which means the engineers of DACs that measure 20-21 bits of ENOB like the topping D90 have succeeded in extracting all of the possible performance out of the DAC chip and have not ruined anything.
Finally, the AK4191EQ in the Topping D90
is a switched-resistor ("multibit") DAC, as they state in the first sentence of their product brief here, so if there was some fundamental flaw with switched resistor DACs preventing good measurements, this would be impossible.
Happy to address anything I've left out, and intent was just to shed some hopefully useful information about how data converters work and what all of these measurements mean.