IPB

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
SACD decoder options
Anakunda
post Dec 13 2011, 18:14
Post #1





Group: Members
Posts: 465
Joined: 24-November 08
Member No.: 63072



Hello,

perhaps somebody can explain meaning of SACD decoder options.

ASIO Driver mode: PCM / DSD (Autodetect) difference
and
DSD2PCM mode: Multistage (fixed point) / Multistage (floating point) / Direct (Floating point, 30kHz LF) difference

I'm utterly aimed to get highest quality for converter..

What's the native sampling frequency for SACD? I have 4 choices which 44.1 is too low, but there are higher. I intuiitionally choosed 88.2kHz but not sure if this still isnot downsampling the undecoded audio. I need a value that won't up/downsample the audio any way and extracts in the same reate as original. Thankyou.
Go to the top of the page
+Quote Post
lvqcl
post Dec 13 2011, 18:33
Post #2





Group: Developer
Posts: 3399
Joined: 2-December 07
Member No.: 49183



QUOTE
ASIO Driver mode: PCM / DSD (Autodetect) difference

I'm almost 100% sure that your soundcard doesn't support DSD input stream, so there's no difference.

QUOTE
DSD2PCM mode: Multistage (fixed point) / Multistage (floating point) / Direct (Floating point, 30kHz LF)

floating point is faster (at least here, on Intel Core2)

QUOTE
What's the native sampling frequency for SACD?

2822400 Hz.
Go to the top of the page
+Quote Post
Anakunda
post Dec 13 2011, 18:37
Post #3





Group: Members
Posts: 465
Joined: 24-November 08
Member No.: 63072



QUOTE (lvqcl @ Dec 13 2011, 18:33) *
QUOTE
DSD2PCM mode: Multistage (fixed point) / Multistage (floating point) / Direct (Floating point, 30kHz LF)

floating point is faster (at least here, on Intel Core2)


And what's the impact on audio quality between the 3 modes, or is the output always same regardless the mode?
Go to the top of the page
+Quote Post
NiPP0
post Apr 7 2012, 01:07
Post #4





Group: Members
Posts: 12
Joined: 7-April 12
Member No.: 98484



Currently I use the 'Multi-Stage (Double Precision)' which was added in v0.5.3 of the foobar SACD plugin. I assume this is an enhanced version of of 'Mutli-Stage (Fixed Point)' mode (read that this was the best quality method).

I'm looking to convert to 176.4/24 FLAC, but I am unsure as to what the SACD plugin does in regard to dithering...

1. Does the SACD plugin apply dithering, or do I need to do it afterwards in foobar's output options?
2. What dithering method does the plug-in use? Does it vary based on the DSD2PCM mode selected?
3. Does it dither if the DSD2PCM mode is set to 'Floating Point' (set foobar's output to Wave64 with 'Auto' depth, still got a 24-bit file?)

Any help would be greatly appreciated,

NiPP0

Edit: Also, are all SACD's 2822.4kHz or are some 5644.8kHz?

This post has been edited by NiPP0: Apr 7 2012, 01:11
Go to the top of the page
+Quote Post
SoNic67
post Apr 7 2012, 05:29
Post #5





Group: Members
Posts: 108
Joined: 3-February 11
Member No.: 87877



SACD is an optical support that holds an "encrypted" version of DSD audio. The same as DVD/CD are optical supports for PCM audio.
So there is no point of asking what is the sampling frequency for SACD - that's the disc itself.
More about that on wiki.
As quality goes, the DSD (resulted from SACD) is equivalent with PCM 24 bit 96kHz... So the 176.4kHz might be overkill, but some might say 88.2kHz is not "complete". As in it "goes" only to 40kHz instead of 50kHz biggrin.gif
In my view, 24 bit at 88.2kHz will get the maximum quality from DSD on today's DACs (that are slightly less performant at 192kHz rates).

This post has been edited by SoNic67: Apr 7 2012, 06:20
Go to the top of the page
+Quote Post
Anakunda
post Apr 7 2012, 09:00
Post #6





Group: Members
Posts: 465
Joined: 24-November 08
Member No.: 63072



QUOTE (SoNic67 @ Apr 7 2012, 06:29) *
As quality goes, the DSD (resulted from SACD) is equivalent with PCM 24 bit 96kHz... So the 176.4kHz might be overkill, but some might say 88.2kHz is not "complete". As in it "goes" only to 40kHz instead of 50kHz biggrin.gif


Then why must it be resampled to 88.2, 176.4 or even 352.8kHz? Why can't the reader keep the original sampling? Is 352.8kHz any sense when 176.4 is overkill already?
Go to the top of the page
+Quote Post
lvqcl
post Apr 7 2012, 09:48
Post #7





Group: Developer
Posts: 3399
Joined: 2-December 07
Member No.: 49183



Do you know what the orginal samplerete is?
Go to the top of the page
+Quote Post
Sandrine
post Apr 7 2012, 12:03
Post #8





Group: Members
Posts: 319
Joined: 2-July 10
Member No.: 81991



Guys, the discussion you're having here is somewhat off track. DSD and PCM format are not mathematically comparable. When converting from DSD to PCM there is no dithering, but it IS a lossy process. AFAIK, there is no resampling done in the first step but the conversion is table-based. Some of the modes will convert to a specific sample rate and then resample to other sample rates. I can't tell you off the top of my head which modes do what right now, though.

You would be much better off asking these questions on the plug-in page at sourceforge as the author doesn't have time to write in the hydrogenaudio forums.
Go to the top of the page
+Quote Post
NiPP0
post Apr 7 2012, 21:39
Post #9





Group: Members
Posts: 12
Joined: 7-April 12
Member No.: 98484



Edit 2: Can anybody confirm that Multi-Stage (Double Precision) is the best quality as of v0.5.3, I'm basing my assumption on the fact it has the highest CPU usage (~30% on an i7@4.0GHz) and was just added (can't see why dev's would have added it if it was using more CPU but was lower quality).

This post has been edited by NiPP0: Apr 7 2012, 21:43
Go to the top of the page
+Quote Post
NiPP0
post Apr 7 2012, 21:44
Post #10





Group: Members
Posts: 12
Joined: 7-April 12
Member No.: 98484



2822.4/24=117.6... 117.6KHz isn't a rate so we have to do 2822.4/32=88.2. I don't think the plug-in outputs 32-bit, so it would have to be converted to 24-bit. This would keep about 75% (24/32) of the DSD's audio data. Going to 96/24 wouldn't be a good idea, so 176.4/24 would be the next step up. This would be overkill by about 50% (2822.4/176.4=24, 24/16=1.5).

I'm sure DSD -> PCM conversion algorithms will keep being improved, so while I planned on doing 176.4/24 FLAC I'm thinking it would be better to do 88.2/24 and keep the .DFF/.DSD/SACD .ISO. Remember that even at 352.8/24 conversion can't be considered lossless.

When editing an image (lossless or not) I prefer to down-sample the final output slightly rather than up-sample it. This is based on the fact the original image may have slightly less than optimum spatial resolution, so downsampling gives me the feeling that the ouput is 'honest' (contains as much data as it should).

Any ideas on the dithering question?

NiPP0

This post has been edited by NiPP0: Apr 7 2012, 22:16
Go to the top of the page
+Quote Post
lvqcl
post Apr 7 2012, 23:58
Post #11





Group: Developer
Posts: 3399
Joined: 2-December 07
Member No.: 49183



QUOTE (NiPP0 @ Apr 8 2012, 00:44) *
2822.4/24=117.6... 117.6KHz isn't a rate so we have to do 2822.4/32=88.2. I don't think the plug-in outputs 32-bit, so it would have to be converted to 24-bit. This would keep about 75% (24/32) of the DSD's audio data.

So you divided a samplerate by a bit depth?
Go to the top of the page
+Quote Post
Anakunda
post Apr 8 2012, 00:02
Post #12





Group: Members
Posts: 465
Joined: 24-November 08
Member No.: 63072



I think mine foobar shows an astronomic samplerate when I open SACD ISO, so it must be divided by something. I don't even think music album could fit on SACD with so a high samplerate.
Go to the top of the page
+Quote Post
SoNic67
post Apr 8 2012, 00:39
Post #13





Group: Members
Posts: 108
Joined: 3-February 11
Member No.: 87877



QUOTE (NiPP0 @ Apr 7 2012, 16:44) *
2822.4/24=117.6...

That's why I said that 88.2kHz is not quite there.
Actually the definition of DSD is 1-bit with 64 times the CD samplerate (CD is PCM at 16 bit depth). So if you want precise calculation, you have 28822.4/16=176.4kHz@16 bit.
It makes no sense to divide to 32 bit since there are no real DAC's that can take advantage of that PCM bitdepth. Sabre, Texas Instruments and Wolfson have each one 32 bit DAC cip in production, but the datasheet performances are not better than previous 24 bit DAC cips. Actually, the higest quality are some "24 bit" DAC's. Even those reference DAC chips don't attain more than 20-21 bit of real resolution (the lowest 3-4 bit are just electrical noise). To make things worse, the best performance of the same top-of-the-line DAC cips happens at 96kHz (not at 192kHz). THD+N specs tell the story.
So, keeping the resulting PCM at 24 bit 88.2kHz is IMO the best solution.

This post has been edited by SoNic67: Apr 8 2012, 00:42
Go to the top of the page
+Quote Post
kode54
post Apr 8 2012, 00:56
Post #14





Group: Admin
Posts: 4619
Joined: 15-December 02
Member No.: 4082



QUOTE (Anakunda @ Apr 7 2012, 16:02) *
I think mine foobar shows an astronomic samplerate when I open SACD ISO, so it must be divided by something. I don't even think music album could fit on SACD with so a high samplerate.

The component which supports playback of SACD ISOs does not report the actual playback sample rate. It does report the playback rate during actual playback, though.
Go to the top of the page
+Quote Post
lvqcl
post Apr 8 2012, 09:42
Post #15





Group: Developer
Posts: 3399
Joined: 2-December 07
Member No.: 49183



QUOTE (SoNic67 @ Apr 8 2012, 03:39) *
Actually the definition of DSD is 1-bit with 64 times the CD samplerate

Agree.

QUOTE (SoNic67 @ Apr 8 2012, 03:39) *
So if you want precise calculation, you have 28822.4/16=176.4kHz@16 bit.

And this is absolutely meaningless.
Go to the top of the page
+Quote Post
SoNic67
post Apr 10 2012, 04:03
Post #16





Group: Members
Posts: 108
Joined: 3-February 11
Member No.: 87877



I ment to say that information carried is the same... 1bit@64x44.1kHz or 16bit@176.4kHz or 24bit@110kHz.
Noise shaping takes care of the audio band, pushing the noise over 50kHz.

This post has been edited by SoNic67: Apr 10 2012, 04:07
Go to the top of the page
+Quote Post
NiPP0
post Apr 13 2012, 15:00
Post #17





Group: Members
Posts: 12
Joined: 7-April 12
Member No.: 98484



Sorry to keep asking, but does dithering need to be applied by foobar or does the DSD2PCM algorithm take care of it (as well as re-sampling)?
Go to the top of the page
+Quote Post
lvqcl
post Apr 13 2012, 15:58
Post #18





Group: Developer
Posts: 3399
Joined: 2-December 07
Member No.: 49183



I tested it (ver. 0.5.3), and it outputs samples in 32-bit float format (as it should btw). So it doesn't perform dithering itself.
Go to the top of the page
+Quote Post
NiPP0
post Apr 14 2012, 17:59
Post #19





Group: Members
Posts: 12
Joined: 7-April 12
Member No.: 98484



QUOTE (lvqcl @ Apr 13 2012, 15:58) *
I tested it (ver. 0.5.3), and it outputs samples in 32-bit float format (as it should btw). So it doesn't perform dithering itself.

I'm getting 24-bit output (with the converter depth set to 'Auto') regardless of the DSD2PCM mode I select (Int/Float)... tried outputting to Wave64 as well as standard WAV.
Using SACD Decoder v0.5.3/v0.5.4 and foobar2000 v1.1.11, no DSPs running.

When I look under one of the loaded SACD track's properties I get...
Samplerate: 2822400 Hz
Bits per sample: 24

Thanks,

NiPP0

This post has been edited by NiPP0: Apr 14 2012, 18:00
Go to the top of the page
+Quote Post
NiPP0
post Apr 29 2012, 22:38
Post #20





Group: Members
Posts: 12
Joined: 7-April 12
Member No.: 98484



Done some more reading... from what I gather DSD -> PCM is comparable to analog -> PCM and doesn't need to be dithered. Please correct me if I am wrong.
Go to the top of the page
+Quote Post
Anakunda
post Sep 5 2012, 10:35
Post #21





Group: Members
Posts: 465
Joined: 24-November 08
Member No.: 63072



Can I know how does the new decoding mode (Installed FIR) work, and what's the difference against existing modes, and how to choose a proper filter....

30kHz_641.txt
40kHz_641.txt
50kHz_641.txt
60kHz_641.txt

The main thing, does Installed FIR promise yet higher quality of conversion than multistage double precission?

This post has been edited by Anakunda: Sep 5 2012, 10:39
Go to the top of the page
+Quote Post

Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 1st October 2014 - 22:00