Dsd?

Jan 6, 2019 at 11:45 AM Post #16 of 56
Sorry, higher dynamic, higher SNR, better channel separation, clearer of all notes, much deeper and wide

1. How is 6dB dynamic range higher than 96dB dynamic range?
2. How does having a smaller dynamic range and more distortion give you better channel separation and clearer/deeper/wider notes?

The answers are obviously: "It isn't and it doesn't!!" - So, how are you hearing the exact opposite of what's actually happening??

G
 
Jan 6, 2019 at 1:10 PM Post #17 of 56
Sorry, higher dynamic, higher SNR, better channel separation, clearer of all notes, much deeper and wide .... seems like a re-master version from studios or even better. That is dsd against pcm.
(I prefer original track than re-mastering bcs I dont like neutralization or eqing, boosting... make it like a HDR imagine.
Above is just comparision)

I cant get a dsd and pcm version of a same track same recording and mastering ... so I listen by this way :
Same track (wav) -> pcm playback -> sp1000 dac mode -> kse1500 (1)

And

Same track wav-> dsd playback(this is capability, pcm to dsd realtime) ->sp1000 dac mode -> kse1500 (2).
what you describe cannot possibly be about DSD vs PCM as formats. that much is now clear. it's either your DAC conversion thingy that massacres one of your 2 listening scenarios, or your own brain playing tricks on you. a proper listening test(blind test) and/or some measurements of the output could help you know which one it is.
 
Jan 6, 2019 at 1:22 PM Post #18 of 56
Sorry, higher dynamic, higher SNR, better channel separation, clearer of all notes, much deeper and wide ....

To reiterate...

1) Check your equipment, particularly the DAC to verify that your PCM playback is working properly.
2) Set up a line level matched, direct A/B switched blind test comparing the same recording and mastering in both formats.

Do these two simple things and I'll talk to you about it.
 
Jan 7, 2019 at 5:41 AM Post #19 of 56
1. How is 6dB dynamic range higher than 96dB dynamic range?
2. How does having a smaller dynamic range and more distortion give you better channel separation and clearer/deeper/wider notes?

The answers are obviously: "It isn't and it doesn't!!" - So, how are you hearing the exact opposite of what's actually happening??

G

While I certainly do not disagree with what you (and others) have to say in reply to the OP, I have a question on the 6db dynamic range.

I've never really took the time to learn a lot about DSD. I understand what you mean by the 6db as DSD is only 1 bit, but surely it has much higher dynamic range than that in practice, often claimed to be around 120db.

How does it achieve that?
 
Jan 7, 2019 at 7:42 AM Post #20 of 56
I understand what you mean by the 6db as DSD is only 1 bit, but surely it has much higher dynamic range than that in practice, often claimed to be around 120db.
How does it achieve that?

Noise-shaped dither.

Digital dynamic range is defined by the range between the peak signal level and the digital noise floor, and the digital noise floor is determined by the amount of quantisation error. As each "bit" of data doubles the number of available values (binary numbers), so the amount of quantisation error is halved and the digital noise floor is reduced by 6dB. However, because digital quantisation error can be calculated mathematically (statistically), it can be manipulated using probability functions. The quantisation error can be randomised (dithered) and the resultant white noise product can be manipulated (moved), with a mathematical process called "Noise-Shaped Dither". Although DSD only has one bit (and therefore only 6dB of dynamic range), it also has a very high sample rate (and therefore a very high Nyquist Point) that provides a very wide ultrasonic audio freq band where we can move/redistribute all that dither noise. So while technically we've still only got 6dB of dynamic range, the digital noise (floor) which defines the dynamic range is all in the ultrasonic freq band and is therefore inaudible. Using very aggressive noise-shaping algorithms we can reduce the digital noise to a level that provides up to about 120dB of dynamic range within the audible frequency band but remember, exactly the same amount of noise is still there, it's just been moved to the ultrasonic freq band. This raises three points for consideration:

1. If we're going to talk about DSD dynamic range being perceptually 120dB (due to digital noise floor redistribution with noise-shaped dither), then we have to compare apples to apples and talk about CD with noise-shaped dither applied as well. With noise-shaped dither there is no difference between CD and DSD, both provide the same 120dB dynamic range! If we ignore noise-shaped dither then CD has 96dB dynamic range and DSD has just 6dB.

2. For noise-shaped dither to do it's job (randomise quantisation error) with statistical perfection, it has to be applied at a level equivalent to about 1.5 - 2 bits of data. IE. With 16bit CD, it's applied at a level equivalent to bits 14-16. The actual dynamic range of noise-shaped 16bit is therefore around 86dB (although perceptually 120dB). Hopefully you see the problem here? DSD does not have a spare 1.5 - 2 bits of data to apply noise-shaped dither, it only has 1 bit of data to start with! Therefore, the required amount of dither cannot be applied and the quantisation error cannot be fully randomised, leaving us with some signal distortion. So while the dynamic range of CD and DSD is effectively the same, DSD has much higher distortion. Although bare in mind this is relative, "much higher distortion" should still be well below audibility.

3. Because this is all based on the mathematics of how digital audio works, it ONLY applies to the purely digital noise floor (caused by quantisation error). Noise-shaping does not work on (or affect in anyway) the noise-floor of the analogue or acoustic signal. So, the noise floor of the recording venue and the self noise of the mics and other analogue equipment in the recording and reproduction chains is still all there. In practice, it's this acoustic and analogue noise floor which defines the dynamic range of digital recordings and typically it's at least 1,000 times higher in level than the digital noise floor (with noise-shaped dither).

G
 
Last edited:
Jan 7, 2019 at 8:18 AM Post #21 of 56
Noise-shaped dither.

Digital dynamic range is defined by the range between the peak signal level and the digital noise floor, and the digital noise floor is determined by the amount of quantisation error. As each "bit" of data doubles the number of available values (binary numbers), so the amount of quantisation error is halved and the digital noise floor is reduced by 6dB. However, because digital quantisation error can be calculated mathematically (statistically), it can be manipulated using probability functions. The quantisation error can be randomised (dithered) and the resultant white noise product can be manipulated (moved), with a mathematical process called "Noise-Shaped Dither". Although DSD only has one bit (and therefore only 6dB of dynamic range), it also has a very high sample rate (and therefore a very high Nyquist Point) that provides a very wide ultrasonic audio freq band where we can move/redistribute all that dither noise. So while technically we've still only got 6dB of dynamic range, the digital noise (floor) which defines the dynamic range is all in the ultrasonic freq band and is therefore inaudible. Using very aggressive noise-shaping algorithms we can reduce the digital noise to a level that provides up to about 120dB of dynamic range within the audible frequency band but remember, exactly the same amount of noise is still there, it's just been moved to the ultrasonic freq band. This raises three points for consideration:

1. If we're going to talk about DSD dynamic range being perceptually 120dB (due to digital noise floor redistribution with noise-shaped dither), then we have to compare apples to apples and talk about CD with noise-shaped dither applied as well. With noise-shaped dither there is no difference between CD and DSD, both provide the same 120dB dynamic range! If we ignore noise-shaped dither then CD has 96dB dynamic range and DSD has just 6dB.

2. For noise-shaped dither to do it's job (randomise quantisation error) with statistical perfection, it has to be applied at a level equivalent to about 1.5 - 2 bits of data. IE. With 16bit CD, it's applied at a level equivalent to bits 14-16. The actual dynamic range of noise-shaped 16bit is therefore around 86dB (although perceptually 120dB). Hopefully you see the problem here? DSD does not have a spare 1.5 - 2 bits of data to apply noise-shaped dither, it only has 1 bit of data to start with! Therefore, the required amount of dither cannot be applied and the quantisation error cannot be fully randomised, leaving us with some signal distortion. So while the dynamic range of CD and DSD is effectively the same, DSD has much higher distortion. Although bare in mind this is relative, "much higher distortion" should still be well below audibility.

3. Because this is all based on the mathematics of how digital audio works, it ONLY applies to the purely digital noise floor (caused by quantisation error). Noise-shaping does not work on (or affect in anyway) the noise-floor of the analogue or acoustic signal. So, the noise floor of the recording venue and the self noise of the mics and other analogue equipment in the recording and reproduction chains is still all there. In practice, it's this acoustic and analogue noise floor which defines the dynamic range of digital recordings and typically it's at least 1,000 times higher in level than the digital noise floor (with noise-shaped dither).

G

Thanks for the clear explanation G, much appreciated.

A couple other questions, though moving a bit (no pun intended) away from DSD.

I know it is not noise shaping, but how does noise reduction work with pre-emphasis which many early CDs had? I know it is part of redbook specs, mainly because (I think) the early video adapters (and DACs) operated in 14 bits, but why would it have been necessary with dithered 14 bits anyway?

Is the early CD pre-emphasis the same thing (in concept) as noise reduction systems like Dolby, DBX and the RIAA pre-emphasis on vinyl?

Thanks
 
Jan 7, 2019 at 8:21 AM Post #22 of 56
Thank you G.
I am learning quite a bit from your posts recently - not in way that would secure me a passed exam in sound ingeneering - but rather in a layman’s term kind of way that helps my brain understand some of the more complex parts to digital audio.

Probably also helps that I am not rummaging through your posts in order to find some loophole;-)
 
Jan 8, 2019 at 6:53 AM Post #23 of 56
[1] I know it is not noise shaping, but how does noise reduction work with pre-emphasis which many early CDs had? [1a] I know it is part of redbook specs, mainly because (I think) the early video adapters (and DACs) operated in 14 bits, but why would it have been necessary with dithered 14 bits anyway?
[2] Is the early CD pre-emphasis the same thing (in concept) as noise reduction systems like Dolby, DBX and the RIAA pre-emphasis on vinyl?

1. Basically, it's just a calibrated EQ boost applied to the recording and then a calibrated inverse filter (EQ cut) applied during reproduction. In the case of CD pre-emphasis, it was a high freq boost/cut, +10dB by 20kHz. TBH, I'm only going on what I've read about the history of digital audio, I have no personal experience of pre-emphasis. It's use was already long dead when I started in the sound engineering business (beginning of the '90's).
1a. Before redbook was ratified, 14bit was expected to be the bit depth of CD but redbook ended-up being specified as 16bit. Although CD was launched to consumers in 1983, specifications for it started the mid '70's and the redbook specs were published in 1980. Now it doesn't sound like much of a time difference but it was a monumental time for digital technology. As companies started to realise digital tech was going to revolutionise society, there was a massive R&D push to not only invent/design chips with greater functionality but also produce them in mass numbers for public consumption. In other words, the redbook specs were published according to the digital tech that existed in the late '70's but by the time CDs and CD players actually started being manufactured for consumers, the available tech had already vastly improved. For example, in the late '70's there was no digital dither or digital filters; dither was effectively just added analogue white(ish) noise, not the statistically perfect, mathematical representation of it that's possible in the purely digital domain and anti-alias and reconstruction filters were also all analogue, plus there was no oversampling. These analogue filters were relatively very "noisy", as it''s pretty much impossible to create a very steep analogue filter without significant artefacts. However, by the time consumer CD players were released, they were pretty much all capable of 2 times oversampling, allowing cheaper and much less noisy analogue filters and within a few years digital dither was available and the filters also became digital. In other words, much of what is "digital audio" was, at the time when rebook was agreed and published, still analogue and "Emphasis" was included to help reduce the noise of these analogue processes. However, much of the need for "emphasis" no longer existed by the time CDs actually hit the streets and by the mid '80's there was no benefit at all (and was effectively dead). BTW, noise-shaping first became available in the early '90's (Sony's "Super bit-mapping", I believe) and was standard practice by the late '90's.

2. Yes, in concept it was exactly the same thing, although the actual EQ (curve) applied is different.

Probably also helps that I am not rummaging through your posts in order to find some loophole:wink:

I appreciate that, as it's always a potential issue when explaining something in layman's terms. I myself could find loopholes in what I've stated, not because it's intrinsically wrong but because any simplification or "layman's terms" is by definition an interpretation and can therefore be misinterpreted by the reader. Disputes can therefore arise which effectively derail the discussion because they're often disputes about the choice of wording and semantics, rather than the actual facts.

G
 
Jan 8, 2019 at 7:55 AM Post #24 of 56
That is sadly what I see as well. I guess it is fairly obvious to people when they’ve spent a little time round this part of the site, but to a guy like myself who basically went through High school and University in the same manner, it was obvious from the get-go. It is a matter of sneakily luring the conversation into relativity..in such a way that renders facts completely useless. Then it effectively ends up being a competition about semantics.
 
Jan 8, 2019 at 8:22 AM Post #25 of 56
1. Basically, it's just a calibrated EQ boost applied to the recording and then a calibrated inverse filter (EQ cut) applied during reproduction. In the case of CD pre-emphasis, it was a high freq boost/cut, +10dB by 20kHz. TBH, I'm only going on what I've read about the history of digital audio, I have no personal experience of pre-emphasis. It's use was already long dead when I started in the sound engineering business (beginning of the '90's).
1a. Before redbook was ratified, 14bit was expected to be the bit depth of CD but redbook ended-up being specified as 16bit. Although CD was launched to consumers in 1983, specifications for it started the mid '70's and the redbook specs were published in 1980. Now it doesn't sound like much of a time difference but it was a monumental time for digital technology. As companies started to realise digital tech was going to revolutionise society, there was a massive R&D push to not only invent/design chips with greater functionality but also produce them in mass numbers for public consumption. In other words, the redbook specs were published according to the digital tech that existed in the late '70's but by the time CDs and CD players actually started being manufactured for consumers, the available tech had already vastly improved. For example, in the late '70's there was no digital dither or digital filters; dither was effectively just added analogue white(ish) noise, not the statistically perfect, mathematical representation of it that's possible in the purely digital domain and anti-alias and reconstruction filters were also all analogue, plus there was no oversampling. These analogue filters were relatively very "noisy", as it''s pretty much impossible to create a very steep analogue filter without significant artefacts. However, by the time consumer CD players were released, they were pretty much all capable of 2 times oversampling, allowing cheaper and much less noisy analogue filters and within a few years digital dither was available and the filters also became digital. In other words, much of what is "digital audio" was, at the time when rebook was agreed and published, still analogue and "Emphasis" was included to help reduce the noise of these analogue processes. However, much of the need for "emphasis" no longer existed by the time CDs actually hit the streets and by the mid '80's there was no benefit at all (and was effectively dead). BTW, noise-shaping first became available in the early '90's (Sony's "Super bit-mapping", I believe) and was standard practice by the late '90's.

2. Yes, in concept it was exactly the same thing, although the actual EQ (curve) applied is different.



I appreciate that, as it's always a potential issue when explaining something in layman's terms. I myself could find loopholes in what I've stated, not because it's intrinsically wrong but because any simplification or "layman's terms" is by definition an interpretation and can therefore be misinterpreted by the reader. Disputes can therefore arise which effectively derail the discussion because they're often disputes about the choice of wording and semantics, rather than the actual facts.

G
Thanks again for the explanation - the historical context is quite interesting too.

With regard to pre-emphasis generally, I've never got my head around why boosting the frequencies on recording and then reversing the boost on playback results in noise reduction.
 
Jan 8, 2019 at 9:48 AM Post #26 of 56
So in the end whatever is "over" flac 16 is just not worth it if the flac file is correctly coded, this is just good to know cause i'm going to save so much space on my hard disks
for me personally, anything over 24bit is not worth it at all.
16 is great, fantastic! belissimo! however, i do notice a slight change from 16 to 24. mostly to do with volume sensitivity. i've noticed personally that songs in 24bit require more volume and can get louder without sounding scratchy. but thats just me and my setup and my ears. i could be wrong.
 
Jan 8, 2019 at 12:31 PM Post #27 of 56
Thanks again for the explanation - the historical context is quite interesting too.

With regard to pre-emphasis generally, I've never got my head around why boosting the frequencies on recording and then reversing the boost on playback results in noise reduction.
I believe it's a simple concept of SNR. let's say your system will cause some noise around 50hz at a known moment in the recording or playback process. you would boost the signal around 50hz before reaching that moment, then the noise would come as it is, but thanks to the boost, the SNR at 50hz is going to be better(the signal was boosted but not the noise). then latter on, at a position in the chain that doesn't generate a loud 50hz noise anymore, we reverse the EQ to get back our properly tuned music. taking the signal back to it's original amplitude, and the noise down by the amount of boost we first created.
 
Jan 8, 2019 at 2:30 PM Post #28 of 56
for me personally, anything over 24bit is not worth it at all.
16 is great, fantastic! belissimo! however, i do notice a slight change from 16 to 24. mostly to do with volume sensitivity. i've noticed personally that songs in 24bit require more volume and can get louder without sounding scratchy. but thats just me and my setup and my ears. i could be wrong.
I've exactly the same problem, volume with flac 24 drops down
 
Jan 8, 2019 at 6:20 PM Post #29 of 56
I believe it's a simple concept of SNR. let's say your system will cause some noise around 50hz at a known moment in the recording or playback process. you would boost the signal around 50hz before reaching that moment, then the noise would come as it is, but thanks to the boost, the SNR at 50hz is going to be better(the signal was boosted but not the noise). then latter on, at a position in the chain that doesn't generate a loud 50hz noise anymore, we reverse the EQ to get back our properly tuned music. taking the signal back to it's original amplitude, and the noise down by the amount of boost we first created.
Thanks for that explanation. I can understand that boosting the signal just before reaching that moment, and probably most relevant in the mastering domain, but how does that work with a more general pre-emphasis, eg a linear boost/cut on some early CDs or the RIAA curve on LP playback (or Dolby NR for that matter)? For example, if we boost all the higher frequencies to deal with tape hiss, why wouldn't the hiss also be boosted and remain as is when reversed?
 
Jan 8, 2019 at 7:44 PM Post #30 of 56
For example, if we boost all the higher frequencies to deal with tape hiss, why wouldn't the hiss also be boosted and remain as is when reversed?
I may be wrong, but I think the point is that you would boost the higher frequencies before storing on the medium, so later when you get the signal - plus noise - back from the medium and you reverse the boost, you achieve better SNR.
 

Users who are viewing this thread

Back
Top