Lumin U1 Mini best DSD -> PCM strategy


(paolo) #1

I gave up on DSD and picked a Metrum Acoustics Adagio DAC :slight_smile: :slight_smile: :slight_smile:

the Adagio does 384KHz and I only have (a few) DSD64 and 128 albums. Looks options are:

  • leave the U1 Mini do the conversion
  • set (appropriately) Roon’s DSP engine

advice is… ?

thanks


(Daniel Beyer) #2

I’d use Roon to do the conversion. It does a good job with DSD to PCM, imho.

However, if you only have a few there is also the option, if you are staying away from DSD, you could also get a program and convert the DSD to 384 PCM files and be done with it. Store the original DSD somewhere in case. I did this for the first couple of DSD downloads I purchased (PCM versions were not available) and before Roon had a good DSD converter and before I had a DAC capable of DSD. Now, I just use Roon to do the conversion at time of playback.


(paolo) #3

and… better set Roon to convert everything to 384 (not going to permanently convert externally, thanks: “few” actually means 50-60 :wink: ) or DSD64 -> 88.2 and DSD128 -> 176.4?


(Daniel Beyer) #4

LOL, for me a few was 6.

That depends on the Metrum. But if it likes incoming streams of 384 then no reason not to feed it everything, including DSD, at 384. Unless the DSD conversion, which can need some horsepower, causes your server stress.


(paolo) #5

:stuck_out_tongue:

thank you Daniel


(Peter Lie) #6

For USB: DSD -> 352.8kHz
For AES: DSD -> 176.4kHz

See which sounds better.

The reason I choose 352.8kHz not 384kHz is this:

DSD128: 5644800Hz / 384000 = 14.7 [Not integer!]
DSD128: 5644800Hz / 352800 = 16


#7

What I know many DACs in the market also do DSD in x48k samples. They have separate clock crystal that handles both x44.1k as well as x48k samples. In the case of x48k in DSD mode this translate to 6144000Hz, therefore there’s no truncation integer error when this converted to PCM.


(Peter Lie) #8

If we’re talking about upsampling from PCM to DSD, then yes 192kHz upsampling to 6144000Hz DSD128 (x48) is better than 5644800Hz DSD128 (if the DAC supports x48 DSD, some do not).

The subject of this thread is to convert existing DSD music files to PCM for playback by a PCM-only DAC. All normal DSD files that can be purchased belong to the 44.1kHz family (2822400Hz, 5644800Hz, 11289600Hz for DSD64, DSD128, DSD256 respectively). So 352.8kHz or 176.4kHz is better. (The exception is that when a DAC plays 48kHz family better than 44.1kHz family due to the way its clocking is done in hardware.)


(paolo) #9

set all DSDs to 352.8 PCM and sounding great :slight_smile:

wasn’t that happy with my previous settings of DSD64 -> 88.2, DSD128 -> 176.4 etc (only tried this with DSD64 though)


(Peter Lie) #10

This reminds me of:


(paolo) #11

may be!

but Metrum’s are no “conventional R2R” DACs :wink:
(here an explanation of how it works)


(Jussi Laako (Signalyst)) #12

Those plots I posted are from Metrum Musette…


(paolo) #13

… which only has one “DAC ONE” chip per channel whilst the Adagio has four “DAC TWO” (eight DAC ONE plus FPGA correction) chips per channel :wink:


(Jussi Laako (Signalyst)) #14

Paralleling extra converters only improves SNR, but doesn’t change the fundamental behavior. As long as the fundamental operation is similar the resulting output spectrum is similar. It doesn’t even need actual converter because the behavior is just pure math, one can simulate it in computer too.

As you can see from the plots there’s massive difference on output quality between running the converter at 44.1k vs running it at 352.8/384k. There’s also equally massive difference in amount of output jitter.

I’d be happy to see corresponding plots for Adagio or other models too.


(paolo) #15

Cees (the designer) says: “We also changed the DAC ONE module of the Pavane to the newer DAC TWO. As you’ll remember, we used an FPGA to split up the 24-bit domain into two streams, each fed to a single DAC One module. Afterwards we summed these streams to get back the full analog signal with improved low-level linearity”

“[…] doubling has a positive effect on noise, distortion and linearity”

not going into an argument (no will nor knowledge), Jussi: just trying to understand :wink:


(Jussi Laako (Signalyst)) #16

Yes, possibly improved THD as result of improved linearity which is always a challenge for R2R and one reason SDM DACs became majority.

But this doesn’t affect amount of image distortion which depends on combination of sampling rate and performance of the reconstruction filter. Those images also give rise to IMD in the audio band. In order to accurately reconstruct signal sampled at 44.1k you need to completely remove image frequency content above 22.05 kHz. This is where you need digital filters because for practical analog filters, the band between 20 kHz and 22.05 kHz is too narrow to reach -96 dB attenuation (needed for 16-bit RedBook). If you have 24-bit content sampled at 44.1k you need -144 dB attenuation in the same band. When you upsample to 352.8k, you only need analog filter that can do the same in 20 kHz to 332.8 kHz band which is much more realistic…

If you compare the two plots I posted, at 44.1k output rate first image is right above 22.05 kHz and only -16 dB down. While at 352.8k output rate first image is about -50 dB down and begins at 332.8k.

Still almost all DACs that do digital filters only to 352.8k cannot do it perfectly. So you have some left overs at multiples of 352.8 kHz. One reason is that their analog filters don’t usually start cutting already at 20 kHz, because that would cause huge phase shift in the top octave.

So for example 19 kHz sine at 44.1k sampling rate:

And the same, but upsampled to 384k this time: