Jump to content
IGNORED

64bit multistage algorithm - DSD to PCM


mkrzych
 Share

Recommended Posts

Hello,

Anyone is aware of quite "eatable" explanation how this algo works in detail - just to understand better? In my case using A+ when I enable it what I am getting is 24/176.4kHz stream - so it means that is 4x redbook and downsample the DSD64 to the multiplication of the 44.1kHz up to the frequency supported by the dac as the result of it, am I right?

--

Krzysztof Maj

http://mkrzych.wordpress.com/

"Music is the highest form of art. It is also the most noble. It is human emotion, captured, crystallised, encased… and then passed on to others." - By Ken Ishiwata

Link to comment
Share on other sites

...additionally if the stream feeding the DAC is done from DSD64 to 176.4kHz PCM it will more likely include the ultrasonic noise above the real Nyquist frequency which may introduce some distortion to the amp and tweeter in the speakers - don't know if it's something I should worry about, shouldn't I? I know that's probably around -70dB down, but still could not be so nice.

--

Krzysztof Maj

http://mkrzych.wordpress.com/

"Music is the highest form of art. It is also the most noble. It is human emotion, captured, crystallised, encased… and then passed on to others." - By Ken Ishiwata

Link to comment
Share on other sites

...additionally if the stream feeding the DAC is done from DSD64 to 176.4kHz PCM it will more likely include the ultrasonic noise above the real Nyquist frequency which may introduce some distortion to the amp and tweeter in the speakers - don't know if it's something I should worry about, shouldn't I? I know that's probably around -70dB down, but still could not be so nice.

 

I had this situation for a year or more before I had a DSD capable DAC. It was never a problem.

One never knows, do one? - Fats Waller

The fairest thing we can experience is the mysterious. It is the fundamental emotion which stands at the cradle of true art and true science. - Einstein

Computer, Audirvana -> optical to EtherREGEN -> microRendu -> ISO Regen -> Pro-Ject Pre Box S2 DAC -> Spectral DMC-12 & DMA-150 -> Vandersteen 3A Signature.

Link to comment
Share on other sites

I had this situation for a year or more before I had a DSD capable DAC. It was never a problem.

 

Possibly Jud, but how do you know if it's not? Ok, not broken anything yet, but it could work on that.

--

Krzysztof Maj

http://mkrzych.wordpress.com/

"Music is the highest form of art. It is also the most noble. It is human emotion, captured, crystallised, encased… and then passed on to others." - By Ken Ishiwata

Link to comment
Share on other sites

...Continuing, some of the options already implemented in Foobar for instance give you explanation that when DSD->PCM conversion is used it uses 30kHz low pass filter in place to limit ultrasonic noise regardless of the other things in chain.

 

For A+, AFAIK we don't know. If I use 64bit precision and select "Forced Upsamplink No" for iZotope in the Audio Filters tab, A+ will send the stream to the DAC with all the ultrasonic noise there or limit it somehow (guessing that "Forced Upsamplink No" implicates no use of the Advanced parameters for the Audio Filters as well).

 

To convert DSD to PCM what I understood in case of 64bit multistage algo is First pass decimation to PCM 352,8 kHz from DSD (any FIR is used here?) and than second pass decimation to target frequency (possibly maximum of your DAC multiplication of 44.1kHz.

 

BTW, is there any option in Mac OSX to reroute the sound from A+ and capture it for instance for Audacity to see what A+ is actually sending to the DAC?

--

Krzysztof Maj

http://mkrzych.wordpress.com/

"Music is the highest form of art. It is also the most noble. It is human emotion, captured, crystallised, encased… and then passed on to others." - By Ken Ishiwata

Link to comment
Share on other sites

Possibly Jud, but how do you know if it's not? Ok, not broken anything yet, but it could work on that.

 

Hi Krsysztof. Yes, that's all I can say: I used A+ DSD-PCM conversion for over a year with 20+ year old speakers, and they worked perfectly during that period and afterward.

 

Are you seeing high levels of ultrasonics in your source material, or don't you know?

 

Interesting idea about capturing the output. I'm sure it's possible but don't know how.

One never knows, do one? - Fats Waller

The fairest thing we can experience is the mysterious. It is the fundamental emotion which stands at the cradle of true art and true science. - Einstein

Computer, Audirvana -> optical to EtherREGEN -> microRendu -> ISO Regen -> Pro-Ject Pre Box S2 DAC -> Spectral DMC-12 & DMA-150 -> Vandersteen 3A Signature.

Link to comment
Share on other sites

Hi Krsysztof. Yes, that's all I can say: I used A+ DSD-PCM conversion for over a year with 20+ year old speakers, and they worked perfectly during that period and afterward.

 

Are you seeing high levels of ultrasonics in your source material, or don't you know?

 

 

I don't know, but technically speaking if I play let say DSD64 file from the old recording which was never recorded in the high resolution usually the musical content could be up to 20k (even less) and the rest in the noise when playing as 176.4kHz fed DAC. Now, how much of this kind of ultrasonic noise don't know, but it's there I guess...

--

Krzysztof Maj

http://mkrzych.wordpress.com/

"Music is the highest form of art. It is also the most noble. It is human emotion, captured, crystallised, encased… and then passed on to others." - By Ken Ishiwata

Link to comment
Share on other sites

...additionally if the stream feeding the DAC is done from DSD64 to 176.4kHz PCM it will more likely include the ultrasonic noise above the real Nyquist frequency which may introduce some distortion to the amp and tweeter in the speakers - don't know if it's something I should worry about, shouldn't I? I know that's probably around -70dB down, but still could not be so nice.

The noise floor of DSD64 usually rises rather sharply after 30 kHz. For this reason the SACD spec recommends a 50 kHz (analogue) lowpass filter.

 

Converting DSD to PCM is nothing other than a standard digital lowpass filter. For playback you get the best results by filtering at 50 kHz while keeping the sample rate at the highest your DAC accepts to avoid its upsampling filters adding anything unwanted. If you're paranoid about audio band effects of the lowpass filter you can use a gentle filter starting around 30-40 kHz and a sharper one to cut off any remaining noise above 88.2 kHz (if the target PCM rate is 176.4 kHz).

Link to comment
Share on other sites

This is an interesting thread as I'm converting DSD to PCM so I can use CanOpener Studio with my headphones. A+ is converting to 352.8 as my DAC supports it.

Roon Rock->Auralic Aria G2->Schiit Yggdrasil A2->McIntosh C47->McIntosh MC301 Monos->Wilson Audio Sabrinas

Link to comment
Share on other sites

The noise floor of DSD64 usually rises rather sharply after 30 kHz. For this reason the SACD spec recommends a 50 kHz (analogue) lowpass filter.

 

Converting DSD to PCM is nothing other than a standard digital lowpass filter. For playback you get the best results by filtering at 50 kHz while keeping the sample rate at the highest your DAC accepts to avoid its upsampling filters adding anything unwanted. If you're paranoid about audio band effects of the lowpass filter you can use a gentle filter starting around 30-40 kHz and a sharper one to cut off any remaining noise above 88.2 kHz (if the target PCM rate is 176.4 kHz).

 

So, meaning to use let say 12dB of steepness/octave and 1.0xNyquist in A+ settings right? Even if you use it, do we know if the first conversion from DSD to PCM 352.8kHz involved already any default low pass filtering?

 

Also in case of A+, maybe other software as well there is another step to downsample from 352.8kHz to the highest possible by the DAC sampling rate being that 44.1kHz multiplication - again, any default low pass filtering there (having Force Upsampler disabled).

--

Krzysztof Maj

http://mkrzych.wordpress.com/

"Music is the highest form of art. It is also the most noble. It is human emotion, captured, crystallised, encased… and then passed on to others." - By Ken Ishiwata

Link to comment
Share on other sites

Krsysztof, how would you set A+ for 12dB/octave cut? If I recall correctly, the filter order is about 4x the iZotope dB setting, so 1dB setting ~= 4th order = 24dB/octave.

One never knows, do one? - Fats Waller

The fairest thing we can experience is the mysterious. It is the fundamental emotion which stands at the cradle of true art and true science. - Einstein

Computer, Audirvana -> optical to EtherREGEN -> microRendu -> ISO Regen -> Pro-Ject Pre Box S2 DAC -> Spectral DMC-12 & DMA-150 -> Vandersteen 3A Signature.

Link to comment
Share on other sites

So, meaning to use let say 12dB of steepness/octave and 1.0xNyquist in A+ settings right? Even if you use it, do we know if the first conversion from DSD to PCM 352.8kHz involved already any default low pass filtering?

 

Also in case of A+, maybe other software as well there is another step to downsample from 352.8kHz to the highest possible by the DAC sampling rate being that 44.1kHz multiplication - again, any default low pass filtering there (having Force Upsampler disabled).

 

I'm not familiar with the settings available in A+, but let's look at some pictures to better see what we're dealing with.

 

This is a spectrogram of some music recorded at DSD64 (downsampled to 96 kHz):

 

dsd64-spgram.png

 

We can see occasional peaks extending up to 30 kHz, above which the DSD noise rises rapidly. A plot of the peak spectral density in the same clip after plain downsampling to 384 kHz looks like this:

 

dsd64-pcm384.png

 

Here we can clearly see the music content dropping off and the DSD noise rising after 30 kHz. Above 100 kHz the noise is at the same level as much of the music.

 

If we apply a 12 dB/octave lowpass filter at 32 kHz we instead get this:

 

dsd64-lp-pcm384.png

 

The DSD noise now peaks at -90 dB/Hz around 90 kHz. If the high-frequency noise was causing any problems before, this should take of that.

 

We could of course have applied a sharp filter directly at 32 kHz, but some people are afraid this might have unpleasant effects (ripple etc.) in the passband. This solution offers a compromise.

Link to comment
Share on other sites

Apples built in and Ozone both don't go higher than 20kHz for a low pass filter. Other options?

 

iZotope 64-bit SRC bundled with A+ allows setting what I believe is the sq rt of 2 over 2 point of the filter (roll off to .707 of the original response) by adjusting the multiplier of "Nyquist."

One never knows, do one? - Fats Waller

The fairest thing we can experience is the mysterious. It is the fundamental emotion which stands at the cradle of true art and true science. - Einstein

Computer, Audirvana -> optical to EtherREGEN -> microRendu -> ISO Regen -> Pro-Ject Pre Box S2 DAC -> Spectral DMC-12 & DMA-150 -> Vandersteen 3A Signature.

Link to comment
Share on other sites

Interesting idea about capturing the output.

 

As a side note, with OS X, you can do this with Audio Hijack. As I just determined with some difficulty, if you want to use AH to capture output from A+, you need to turn off Direct Mode in A+ first. (Doh!)

 

--David

Listening Room: Mac mini (Roon Core) > iMac (HQP) > exaSound PlayPoint (as NAA) > exaSound e32 > W4S STP-SE > Benchmark AHB2 > Wilson Sophia Series 2 (Details)

Office: Mac Pro >  AudioQuest DragonFly Red > JBL LSR305

Mobile: iPhone 6S > AudioQuest DragonFly Black > JH Audio JH5

Link to comment
Share on other sites

Krsysztof, how would you set A+ for 12dB/octave cut? If I recall correctly, the filter order is about 4x the iZotope dB setting, so 1dB setting ~= 4th order = 24dB/octave.

 

I initially thought that setting steepness to 12dB is enough, but it seems that indeed it to cover the octave it needs to be at least 24dB right?

 

Anyway, for me using A+ I am considering that if Force Upsampler is disabled, meaning set to None all the advance parameters are not taken into account, right? If so, still looking for an answer what is used when playing DSD file converted to PCM using two block path: DSD2PCM=358.4kHz -> PCMtoPCM (in my case to 176.4kHz)?

 

Edited: Well, it's technical divagation, but maybe I should forget about the software settings and just use my DAC built in filters like Rega explained:

 

Filter settings 1–3

(low sample rates 32/44.1/48K)

1. Linear phase half-band filter

2. Minimum phase half-band filter

3. Minimum phase apodising filter

 

Filter settings 1–3

(medium & high sample rates 88.2/96 & 176.4/192K)

1. Linear phase soft-knee filter

2. Minimum phase brickwall filter

3. Linear phase apodising filter

The filters have a greater effect on higher sample rates because of the higher Nyquist frequency which is 1⁄2 of the sample rate. This means the filter has a wider frequency range in which to be active.

 

Using for instance filter 1. for DSD2PCM material done for old 50/60 jazz gives to my ears the best sound.

--

Krzysztof Maj

http://mkrzych.wordpress.com/

"Music is the highest form of art. It is also the most noble. It is human emotion, captured, crystallised, encased… and then passed on to others." - By Ken Ishiwata

Link to comment
Share on other sites

Interesting thing to me is that all those FIRs are quite similar in the passband if not the same - what they're doing is suppressing higher frequencies above 20kHz - maybe except brick wall filters where they have sharp cut off. According to my Rega manual it's saying that I should stick to the Linear phase half-band filter for sampling rates up to 48kHz and Linear phase soft-knee filter for the higher sampling rates - filter 1. What is the difference between half band and brickwall in general? Half band meaning that it will start cutting off the frequency from 22.05kHz for redbook?

--

Krzysztof Maj

http://mkrzych.wordpress.com/

"Music is the highest form of art. It is also the most noble. It is human emotion, captured, crystallised, encased… and then passed on to others." - By Ken Ishiwata

Link to comment
Share on other sites

FYI, A+ when playing ISO image the decimation is done directly to a frequency power of two of 44.1kHz, with a low pass filter. iZotope 64-bit SRC is not involved at all in the process. DSD to PCM converter implementation is done based on Maxim Anisiutkin open source libdsd2pcm library v.0.6.4: https://sourceforge.net/projects/sacddecoder/files/foo_input_sacd/ for those who are able to parse the code and see what it's doing in details.

--

Krzysztof Maj

http://mkrzych.wordpress.com/

"Music is the highest form of art. It is also the most noble. It is human emotion, captured, crystallised, encased… and then passed on to others." - By Ken Ishiwata

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share



×
×
  • Create New...