DSD to PCM... To dither or not to dither? 88.2kHz or 96kHz?

[Kuja65]

Hello everybody.

This is my first post here.


I have a bunch of DSD recordings that a firiend of mine did with his Korg MR recorder.

Some are made from vinyl and tape for archival purposes, some are live recordings (mostly of piano, voice and guitar variety).


In order that I can listen to them on my stereo,
he converted them to 24bit 88.2 kHz PCM using Weiss Saracon software.

He did two versions - without dithering (truncated) and with TPDF dithering.


Somehow I feel that the files with no dithering sound better - more open and natural?!

Dithered files seem to have less 3D space (very slightly) and some slight colder/metallic "overtone"?

Am I imagining this?



What is the correct way to perform DSD to 24bit PCM conversion?!

Do we need dithering or we can just truncate?


With PCM to PCM conversions, because of quantization errors,
dithering is needed in case of word length reduction, for example when going from 24bit to 16bit.

As i (wrongly?) understood,
dithering is NOT really needed when only sampling frequency rate is reduced,
going for example from 96kHz to 44.1 kHz with original 24 bits preserved?


What about DSD?

DSD is 1bit and has a sampling rate of 2.8224 MHz.

When converting from DSD to PCM 24bit 88.2kHz,
there is no simple word length reduction that will cause quantization errors,
as in going from higher PCM bit rate to lower?



So, the question is to dither or not to dither?


I tried searching on the forum, there are many DSD vs PCM discussions, but I couldn't find the exact and simple answer to this question.


Also, why is 88.2kHz suggested for DSD to PCM conversion?

Can we use 96kHz?






Thanks in advance,

Aleksandar


.

[wado1942]

2.8224/32=88.2. 1-bit DSD is said to have about 20-bit performance, so you chose the optimal PCM format. As for dither, the 1-bit signal is already heavily dithered with a lot of noise shaping. For the down-conversion (I assume you're doing this with software), I would probably use flat, unshaped dither but you can probably get away with no dither.

[24-96 Mastering]

Originally Posted by Kuja65

In order that I can listen to them on my stereo,
he converted them to 24bit 88.2 kHz PCM using Weiss Saracon software.

He did two versions - without dithering (truncated) and with TPDF dithering.

Somehow I feel that the files with no dithering sound better - more open and natural?!

Dithered files seem to have less 3D space (very slightly) and some slight colder/metallic "overtone"?

Am I imagining this?

It isn't unlikely that you're imagining that, yes. The effect of dithering at 24bit level, as opposed to truncation at 24bits, isn't very likely to produce an audible difference in realistic cases.
But best try for yourself: try to identify which is which, blindly, a few times, if you want to be sure.

What is the correct way to perform DSD to 24bit PCM conversion?!

Do we need dithering or we can just truncate?

Dithering is likely to produce a result sounding closer to the original. But again, at 24bit, coming originally from an analog source, this may very well not make an audible difference.

With PCM to PCM conversions, because of quantization errors,
dithering is needed in case of word length reduction, for example when going from 24bit to 16bit.

As i (wrongly?) understood,
dithering is NOT really needed when only sampling frequency rate is reduced,
going for example from 96kHz to 44.1 kHz with original 24 bits preserved?

That is incorrect, because sample rate conversion actually expands the word length / bit depth.

So if you start at 96kHz/16bit and change sampling rate to, say, 44.1 kHz, the newly created wave form will exceed 16 bit resolution.

Assuming you want to save the waveform back as a 16bit file, and want to prevent/randomize quantization error, you have to apply dither at the target resolution of 16bits.

As a rule of thumb: whenever you process digital audio, your bit depth expands and is no longer that of the original file.

Also, why is 88.2kHz suggested for DSD to PCM conversion?

Can we use 96kHz?

Sure, why not?

Thanks in advance,

Hope this helped. Also know that, for some reason, dithering is sometimes treated like a controversial topic on online forums.

[Bob Olhsson]

Broken dither routines can cause no dither to sound better! I'd never assume that software developers get it right.

The unfortunate truth is that there are very very few people who really know audio, digital signal processing, computer programming and the underlying math involved. The few I've met also tend to roll their eyes at most commercial audio software.

My point is that the thing to be highly skeptical of is practice, i.e. implementation rather than theory.

[Kuja65]

Thanks guys!

I tried listening to those files again, and now it seems that there is not really that much difference between them (if any). :-)

Originally Posted by Bob Olhsson

Broken dither routines can cause no dither to sound better! I'd never assume that software developers get it right.

The unfortunate truth is that there are very very few people who really know audio, digital signal processing, computer programming and the underlying math involved. The few I've met also tend to roll their eyes at most commercial audio software.

My point is that the thing to be highly skeptical of is practice, i.e. implementation rather than theory.

My friend used Weiss Saracon for DSD to PCM conversion and dithering.

From what I have read, this software is highly regarded and should be one of the best available?

Or is it not?

[DSD_Mastering]

Originally Posted by Kuja65

Thanks guys! My friend used Weiss Saracon for DSD to PCM conversion and dithering.

From what I have read, this software is highly regarded and should be one of the best available?

Or is it not?

I wouldn't say it's the best. It's just another sample rate converter.

We like the Pyramix and SBM sample rate converters.

[ross.k2186]

I would also not fully trust the software (unless it's a good brand) as well as this is quite a substantial mathematical prcoccess. Converters here seem to be very important as It is taking an analogous signal and computing it into a numerical value so it's integrity is crucial.

Also 88.2 for DSD sound right as it is an integer related to the sum of DSD; 2.8224/32=88.2.

In terms of converting this to a PCM file say 16 bit 44.1, well 88.2 Khz is a whole multiple integer of 44.1 so the maths is a lot easier of the software and digital D/A converts
Converting to 16 bit would need to be dithered I believe as the word-length is larger then the intended use of 16 bit.

[Spartacus]

Originally Posted by ross.k2186

In terms of converting this to a PCM file say 16 bit 44.1, well 88.2 Khz is a whole multiple integer of 44.1 so the maths is a lot easier of the software and digital D/A converts.

This is not true anymore.

[24-96 Mastering]

Originally Posted by Sounds Great

I know little about the conversion process of these various formats, but this statement has me curious, how could such a conclusion be made?

By measuring dynamic range within the audible band. AFAIK, below 20kHz, 1 bit DSD noise floor is specced to be around -120 dB, which would correspond to noise floor in a dithered 21 bit LPCM system.

[ross.k2186]

Originally Posted by Spartacus

This is not true anymore.

I am interested in how this is untrue to benefit my own knowledge?. Thank you

[ross.k2186]

Originally Posted by 24-96 Mastering

By measuring dynamic range within the audible band. AFAIK, below 20kHz, 1 bit DSD noise floor is specced to be around -120 dB, which would correspond to noise floor in a dithered 21 bit LPCM system.

Thats a good way of going about it, I mean it seems way more logic to do a calculation of the amplitude range/dynamic range under 20 khz. I like this one here .

[UnderTow]

Originally Posted by nortide99

There should be no such question if you are talking about going from DSD to PCM, because DSD [1 bit] cannot actually be dithered.

Also, simply going from 96kHz to 44.1kHz, while staying at 24 bit, as in your example, should should not require any dither step. Dithering would only be a step if you were going from, e.g., 24 bit to 16 bit.

As Robin already pointed out, it all depends on what the sample rate conversion software is doing and how many bits it uses internally. Hopefully with something as expensive as Saracon one would hope it would either do the right dithering itself where needed or give it as a user option. The latter seems to be what the OP is suggesting as he is asking the question.

The Saracon manual should provide the necessary information to the user to make the right choice. (At how many bits the internal calculations are performed and whether there is automatic dither enabled or not). As it supports 64 bit IEEE float files I suspect the internal calculations are done at 64 bit. So, if my guess is right, yes you need to dither the output to 16/24/32 bit.

Alistair

[24-96 Mastering]

Originally Posted by Sounds Great

Ok and thank you, but it seems to me that this would only address dynamic range performance.

Only dynamic range was referenced in wado1942's comparison

[UnderTow]

Originally Posted by Sounds Great

I didn't see anything about dynamic range.

"20 bit performance". Bit depth determines the dynamic range.

Alistair

[24-96 Mastering]

Originally Posted by Sounds Great

I didn't see anything about dynamic range.

Bit depth in a linear PCM system determines dynamic range.
When wado1942 wrote "DSD is said to have about 20-bit performance", dynamic range is the only thing we can assume from that figure.

[wado1942]

I was referring to dynamic range. The 1-bit Delta slope at 2.822.4MHz yields about the same dynamic range as one would expect from a 20-bit linear 44.1KHz signal. I was specifying it in terms of bits to show that dither probably isn't needed when converting to 24-bit 88.2KHz. It probably wouldn't hurt either.

[Spartacus]

Originally Posted by ross.k2186

I am interested in how this is untrue to benefit my own knowledge?. Thank you

Math in SRC algorithms is not as simple as just a fraction of your sample rate.
Many software or converter chips for exemple do it with more complex math, upsampling and proper filter to achieve a perfect result.
You can do a search here and find more and more users aware of this and using the good tool to perform SRC whatever your sample rate is.

[Bob Olhsson]

It's all about how the filtering is done.

My understanding from what Andy Moorer wrote is that an even multiple can get by with single rather than double precision but that's hardly about lots simpler math.

[Noooway]

Sorry but this is a digression. I want to cut a mix/master to vinyl on an old Scully but are not really sure what type of dither to use on 24-bit wav masters. Should I use conventional TPDF or some type of noise shaping dither? I have been using the dither in Sonnox Limiter, is this fine or should I look at something else for good dither algorithms? And when I´m first at it, should I preferable let the digital masters stay in original 96 kHz and just use a nice LPF on the highs to not fry the cutter head?

Thank you!

[Noooway]

Originally Posted by swim

I don't know what you are talking about doing, but you will get some advice on this board, that is to always add some dithering, or just noise, no matter what just to make sure in case you need it for headroom.

You'll find out from reading this board that dither is the most important thing for mastering.

Sorry if my post was not clear enough. I want to make a final master in 24-bit to transfer to a Scully Lathe cutter. This master will only be used to transfer to vinyl. So no more editing or converting to 16-bit on the files. So what dither is best, or at least a recommendation, for a final 24-bit master, some noise shaped or TPDF dither?
I always use TPDF saving any 24-bit file that I will edit or convert further as I believe Bob Olhsson one time recomended. Of course I have been searching the forum for the best way to dither a final 24-bit master, but I can´t seem anyone answere that specific. And the reason I ask is because I wonder if some type of dither could be bad for a lathe cutter in some way or another. Just to make sure :-)

[wado1942]

For anything other than 16-bit down-conversion, I'd use flat, TPDF dither. Far more important than that is making sure there's no clipping or excessive compression. That will harm a vinyl release more than dither by a long shot.

[UnderTow]

Originally Posted by Noooway

I always use TPDF saving any 24-bit file that I will edit or convert further as I believe Bob Olhsson one time recomended.

Perfect for anything 24 bit including the final master.

And the reason I ask is because I wonder if some type of dither could be bad for a lathe cutter in some way or another. Just to make sure :-)

You might want to start a separate thread for this specific question to catch the attention of the vinyl experts.

Alistair

[dcollins]

Originally Posted by Noooway

I always use TPDF saving any 24-bit file that I will edit or convert further as I believe Bob Olhsson one time recomended. Of course I have been searching the forum for the best way to dither a final 24-bit master, but I can´t seem anyone answere that specific. And the reason I ask is because I wonder if some type of dither could be bad for a lathe cutter in some way or another. Just to make sure :-)

I guess I'll be the first to point out that d*ther at the 24 bit level is not audible even in an all-digital chain anyway, and the surface noise of vinyl is roughly 1000dB higher, so.......just use TPDF like you would for everything else and move on.


DC