DSD -> PCM conversion workflow

[gordinho]

I've got a wiim mini and my DAC does not even understand DSD so I'm looking at converting the few SACDs to PCM so I can easily stream them. My rationale brain says I should just convert them all the way down to 16/44.1 but seems like most people decimate to 24/88.1.. and well, more is better right lol ? (no!)

I'm curious for those of you that are converting DSD layer to PCM what you are converting to. My main interest in some of these SACDs are the different mastering or transfer from analog sources not really the DSD format.

 

[Scytales]

I'm not doing what you plan to do, but, from a rationale perspective, I think it would be better to find a tool that would permit you to decimate (that is : low pass filter the DSD modulation in the digital domain to obtain a PCM modulation) at least at 24 bits accuracy and a sampling frequency that is a multiple of 44.1 kHz.

Actually, if I remember correctly some digital analysis I saw from a Italian PhD Prof. with a special software tool he had developped, in the audio bandwidth DSD is capable of something between 18.5 and 25 equivalent bits of resolution in the digital domain depending on the pro-audio delta-sigma modulator used to produce the DSD distributed on SA-CD. Hence, it makes no sense to convert DSD to PCM in only 16 bits, unless your equipement does not accept more than CD-like bitdepth. The larger bitdepth would also help if you intent to do some digital signal processing on the resulting PCM files.

As for sampling frequency, dowsample to a multiple of the original 44.1 kHz base frequency (FS, DSD being 64FS) should probably be better to avoid digital resampling artifacts.

Edition :

Tascam has a free software tool for that kind of work, but it is only two channels, I think : https://tascam.com/int/product/hi-res_editor/top

 

[DVDdoug]

As for sampling frequency, dowsample to a multiple of the original 44.1 kHz base frequency should probably be better to avoid digital resampling artifacts.

In the real world of DSP it doesn't matter if the sample rate divides evenly or is a factor. The process/algorithm is the same and anti-alias filtering is always applied when down-sampling. So for example, you can't simply throw-away every-other sample when cutting the sample rate in half.

16/44.1 is probably "good enough" since it's generally better than human hearing. (In a proper blind ABX test, you usually can't hear the difference in a high-resolution original and a copy downsampled to CD quality.)

On the other hand... I don't want to speak for Amir, but I believe his philosophy is "why not" high resolution (or higher resolution), because there might be a difference...

And the "pro studio standard" is 24/96.

There is no "bit perfect" (mathematically reversible) conversion between PCM and DSD.

 

[gordinho]

In the real world of DSP it doesn't matter if the sample rate divides evenly or is a factor. The process/algorithm is the same and anti-alias filtering is always applied when down-sampling. So for example, you can't simply throw-away every-other sample when cutting the sample rate in half.

16/44.1 is probably "good enough" since it's generally better than human hearing. (In a proper blind ABX test, you usually can't hear the difference in a high-resolution original and a copy downsampled to CD quality.)

On the other hand... I don't want to speak for Amir, but I believe his philosophy is "why not" high resolution (or higher resolution), because there might be a difference...

And the "pro studio standard" is 24/96.

There is no "bit perfect" (mathematically reversible) conversion between PCM and DSD.

Right, I'm thinking I will probably just go with the "why not" approach and just do 24/88 (1:32 decimation).. Then I just need to worry about the gain adjustment..

But I still don't quite understand why XLD it asks me to choose a SRC algorithm if I put 88.2 sample rate but it doesn't if I do 1:32 decimation. I guess I could compare the files...

 

[Ricardus]

As for sampling frequency, dowsample to a multiple of the original 44.1 kHz base frequency (FS, DSD being 64FS) should probably be better to avoid digital resampling artifacts.

I can't believe people still believe this and are still repeating this. Please stop. It's not true.

 

[Blumlein 88]

Doesn't matter if you convert to 24/88.2 or 24/96. Go 88.2 if you like, won't hurt anything.

 

[Vincent Kars]

The problem with DSD is its high quantization noise.
In case of DSD64, this starts at 22 kHz. Limiting the sample rate to 48 kHz, blocks out this noise. You must filter out the high frequencies anyway.
24 bit is a good choice as DSD has a substantial bandwidth.

 

[Blumlein 88]

The problem with DSD is its high quantization noise.
In case of DSD64, this starts at 22 kHz. Limiting the sample rate to 48 kHz, blocks out this noise. You must filter out the high frequencies anyway.
24 bit is a good choice as DSD has a substantial bandwidth.

Yes, this is true. So one can go with 48 khz or one can say they are trying to keep the original close to the original and go with something higher. I'd be fine with 48 khz myself. Also lots of DACs have various (too slow roll off) filters now so maybe going with a higher rate takes that effect out.