This is the regular thing that repeats itself every 6 - 12 months. I’ve been answering to these kind of chains for the past over 20 years, since I’ve been designing delta-sigma modulators for some 25 years now.
3 Likes
Yes, of course.
Yes it does, when some format that is actually not really used by anybody for actual converters is being advocated as some kind of solution. At the moment, it is at best some intermediate abstract format, apart from the very few R2R DACs (that don’t really technically perform so great).
No it is not.
By the way, you can’t edit 24-bit signal.
I guess a minimum would be 32-bit fixed- or floating-point, but it would still be PCM. At least according to my definition of PCM. You seem to have another, but I won’t go into semantics again.
BTW, the Cosmos ADC, which is one of the best ADCs, uses ES9822 PRO, which is not “native PCM”.
And it is likely 1-bit modulator stream just like ESS DAC chips.
The chip can also output raw data using 7 GPIO data pins, so I doubt it’s a 1-bit modulator. Also, the signal path from the datasheet seems to suggest it’s actually 6 bits:
Regardless, decimating to PCM doesn’t seem to do much damage.
This is all fascinating, but for the real-world music consumer, how does it matter? Does any of this confer real-world audible advantages of one format over the other? Does DSD consumption consume less power than PCM consumption?
What is the point?
That doesn’t tell how many bits the modulator is, just that the data is transfered in chunks of 6 bits. Remember that it has 4 ADCs inside. Out of 7 raw output pins one is clock and six are used for raw data for all the 4 ADC channels.
It can also output max DSD512. It has 5 pins for doing DSD output, one for clock and 4 for data for all the 4 ADC channels.
Like my DSC1 design transfers 32 bits to the conversion elements with 1-bit SDM input stream.
It is just unnecessary, and you can see from the datasheet how much they cut corners. It would be likely so much better if you would just take it’s raw data or DSD data instead of PCM output.
It has also configurable DSD dither level (5 bit value)…
Don’t try to use Cosmos ADC for measuring phase responses though, since those IIR decimators will introduce phase shifts. And the last stage FIR is nothing to call home about, with max 115 dB stop-band attenuation. Reason seems to be that the filter coefficient precision is max 24-bit. But afterall, it has been designed for recording music, not for making measurements.
For example RME ADI-2 Pro, both it’s 705.6/768k 32-bit PCM output and DSD256 output have exactly same noise-shaper noise slope. There, the DSD format is not the limiting factor.
For the real-world music consumer, it doesn’t really matter. The quality is high enough so that these discussions are now only theoretical - and a source of marketing buzzwords.
1 Like
Are you suggesting that the folks here are the typical “real-world music consumer”? 
3 Likes
Well, I would strongly disagree with that. That’s like saying MP3 is a good as Red Book.
I much prefer DACs that are process DSD natively and I really prefer DACs, like the Holo Audio May, that don’t do much if anything with the signal they receive. I have that I get the best sound quality using HQPlayer to up sample both PCM and DSD to DSD256.
Maybe I am not a “real world consumer” whatever that might be…
I interpreted @Bill_Janssen’s term as “a healthy member of the homo sapiens species who listens to music”. That should also answer @Fernando_Pereira’s question.
2 Likes
You got to be kidding. Humans are highly variable in their perception and discrimination abilities, experience, taste, and many other dimensions. I’m pretty healthy, and I think Mikaela Shiffrin is too, we both ski, but you’d not even try to compare her ability to read and run a course with mine 
Discrimination abilities, experiences etc. don’t make anyone superhuman. You can’t train your ear to hear well over 20kHz or to resolve pitch differences well under one cent or to hear sounds well under 0dBSPL etc. But if it makes you happy, I can change the definition to “a highly discriminating, highly capable, extremely experienced, healthy member of the homo sapiens species who listens to music”. I excluded “taste” on purpose.
Given that most people, under controlled conditions, actually can not tell MP3 (at 320 Kbps at least) from Red Book, let alone Red Book from high resolution signal, there’s quite a bit of doubt whether we’re talking audible differences or confirmation bias here…
4 Likes
I know the people at Bell Labs who ran those evals back when. Some musically trained subjects could make MP3-lossless distinctions reliably for certain kinds of material.
MP3 can be encoded in 32, 44.1 and 48kHz and between 32 and 320kbps. You need to be specific about the rate used in tests.
OK, let me add some definition. By “real-world music consumer” I mean someone with a good ear who likes to listen to music, but is not a recording engineer, or in any audio-related business, and is not particularly interested in all the bizarre paraphernalia of audiophile equipment rigs. Someone with a spouse and a mortgage and three kids to put through college, so has to be careful with spending, but is willing to pay for Roon. Someone who just wants a good transparent system to listen to good recordings through.
What’s the point of DSD for that person?
And yes, I do realize that everyone on this forum is an outlier, as Fernando pointed out.
1 Like
Well, that reduces the range quite a bit 
If you also add age, all bets are off. I stand by my definition tough, to make sure the bar is as high as possible.
That’s a good point. In theory, you can make an IIR linear phase in the passband, but I don’t know if that’s the case. I guess I’ll just give it a try. I’ll play a sine sweep through my DAC with a linear phase filter and capture with Cosmos. You could argue that the phase shift in the IIR used in the DAC plus the one in the IIR used in the ADC cancel themselves out, but what are the chances?