[spaark]

 
What is non-linear noise? What's the opposite?

Non-linear noise is noise produced by a non-linear system. If we're being pedantic, I suppose it's wrong to say the noise is non-linear, because it's the system that is.
 

  I'm not sure I would use the term linear or non-linear in this context, correlated or de-correlated would perhaps be more accurate and less confusing. The act of dithering is the act of de-correlating the quantisation error, which is essentially just a fancy and more precise way of saying the errors are turned into random noise. If we're going into the finer detail, I'm not sure the last sentence I've quoted of yours is entirely accurate; generally noise shaped dither would actually slightly increase the total amount of noise but of course significantly decrease the amount of audible noise.

Less confusing (or rather more insightful) perhaps, because the correlation of quantisation errors is the reason behind non-linearity, but not more accurate. I'm just getting to the crux of the matter. The reason dithering improves fidelity is because it eliminates the non-linearity i.e. the distortion, which is much more audible than white noise of the same energy.
 
I think you're right than it slightly increases the amount of noise as far as real noise shaping techniques are concerned, but we'd like an ideal one to simply distribute it.

 

[sarasa]

I've just came across this thread...
 
Quote:

[...]The result is that we have an absolutely perfect measurement of the waveform (2*) plus some noise.

 
From my understanding, I think that statement isn't correct, of course the Nyquist Theorem is right, but conditions to use the theorem weren't met since the measuring errors aren't expected, I do agree that the new waveform probably isn't that far away from the original one... but how far away? in most of these threads I usually don't find those errors expressed as a amplitude quantization function and someone who shows the math in order to really show if something is really improved or not...
 
I would expect that a higher frequency / bit accuracy could improve the relation to the original signal... I don't have any problem in being wrong, though...

 

[stv014]

Originally Posted by sarasa /img/forum/go_quote.gif
 
From my understanding, I think that statement isn't correct, of course the Nyquist Theorem is right, but conditions to use the theorem weren't met since the measuring errors aren't expected, I do agree that the new waveform probably isn't that far away from the original one... but how far away? in most of these threads I usually don't find those errors expressed as a amplitude quantization function and someone who shows the math in order to really show if something is really improved or not...
 
I would expect that a higher frequency / bit accuracy could improve the relation to the original signal... I don't have any problem in being wrong, though...

 
In the digital part of the reconstruction filter, image rejection for a 20 kHz tone at 44.1 kHz sample rate can be 120 dB or better without major difficulty (using a FIR filter with a length of <= 2 ms). There is some low-ish level imaging far outside the audio band after the analog reconstruction of the oversampled signal, though. However, this (other than being unlikely to be an audible issue) actually might not be improved by using a higher sample rate if it is still oversampled to the same rate anyway.
 
Dithered quantization simply adds noise to the signal. The amount of noise in the audio band depends on the bit depth (every additional bit improves it by 6.02 dB, but analog hardware usually becomes a limiting factor above about 20 bits, and 16 bits should normally be enough in practice for music), noise shaping - if any, and to some extent on the sample rate (because increasing it moves more of the same total noise power into the ultrasonic range).

 

[xnor]

 
From my understanding, I think that statement isn't correct, of course the Nyquist Theorem is right, but conditions to use the theorem weren't met since the measuring errors aren't expected, I do agree that the new waveform probably isn't that far away from the original one... but how far away? in most of these threads I usually don't find those errors expressed as a amplitude quantization function and someone who shows the math in order to really show if something is really improved or not...

 
It's quite simple actually.
 
You quantize an analog value x to x_q. This will add some quantization error e_q.
 
x_q = x + e_q    with an SNR = 20*log₁₀(x_rms / e_rms)
 
The error signal cannot escape the sampling theorem, so it is low-level noise from DC to Fs/2. You can theoretically perfectly reconstruct the signal including low-level noise given by all x_q values.

 

[gregorio]

  From my understanding, I think that statement isn't correct, of course the Nyquist Theorem is right, but conditions to use the theorem weren't met since the measuring errors aren't expected, I do agree that the new waveform probably isn't that far away from the original one... but how far away?

That's impossible to say because it depends on the design and construction of each individual ADC and DAC. In general though, the actual conversion process to and from digital is perfect or at least perfect to orders of magnitude more than the ear could possibly detect. So much so infact that the limiting factor of every ADC and DAC on the market is the analog input stage (in and ADC) and the analogue output stage (of a DAC) rather than the actual conversion process itself and this is true of even the best ADCs and DACs, employing the finest of components. Even very cheap ADCs and DACs these days have virtually linear responses and of course there are only a handful of companies which manufacture the actual digital conversion chips.
 
Quote:

  I would expect that a higher frequency / bit accuracy could improve the relation to the original signal...

To effectively implement the Nyquist Theorem, up to a point it can! But beyond that point the accuracy actually deteriorates simply due to the fact that the higher the sample rate the less time there is to process each sample. So, you can have high sample rates and few bits or lots of bits and a low sample rate but not both without compromising accuracy! In practise, modern ADCs take the former approach and initially sample at a frequency of many megahertz but with only a handful or so of bits. This initial sampling is then decimated down to the sample rate/bit depth selected by the user. Sample rates/bit depths from 16/44 up to 24/96 appear to fall into the optimum window for accuracy but with sampling rates beyond 24/96, accuracy deteriorates. I explained in a bit more detail at the end of this post.
 
G

 

[Don Hills]

   
...  At these very high sample rates and bit depths we start hitting the limits of the laws of physics in how fast we can perform the calculations required to implement a filter which reduces anti-aliasing to below the digital noise floor.  ...

 
Thank you for your explanation.

 

[Matvei]

So, I am about to buy the Fostex HP-A4 DAC which comes out next week in Japan, and I had some questions relating to this thread in terms of PCM vs DSD. For reference, the DAC supports up to 24bit 192kHz in PCM, and up to 5.6MHz!!! in DSD

It comes with a new application called Fostex Audio Player to allow playback of these file types on the fly.

If I understand this thread correctly, 24 bit PCM is useless, as are 96kHz and 192kHz sampling rates. Therefore, HD Audio is snake oil.

I'm not quite sure I understand DSD. It was said in this thread and others as well by Giorgio that although SACD shouldn't sound any better than redbook, they often do because they have better masters. In that case, wouldn't that mean the only use of the DSD feature is to play back (mostly illegally) ripped SACDs? And what on earth is the difference between 2.8MHz and 5.6MHz DSD? Is there any point to using this crap or are these features all extraneous?

 

[gregorio]

So, I am about to buy the Fostex HP-A4 DAC which comes out next week in Japan, and I had some questions relating to this thread in terms of PCM vs DSD. For reference, the DAC supports up to 24bit 192kHz in PCM, and up to 5.6MHz!!! in DSD

If I understand this thread correctly, 24 bit PCM is useless, as are 96kHz and 192kHz sampling rates. Therefore, HD Audio is snake oil.

If all you are doing is playing commercially released recordings, then yes, 24bit is useless. It's not useless if you are recording or mixing music though. Again, 96kHz may have some uses during recording and mixing but not so much for playback. 192kHz is useless for virtually everything, including recording, mixing and playing back commercial recordings!
 

I'm not quite sure I understand DSD. It was said in this thread and others as well by Giorgio that although SACD shouldn't sound any better than redbook, they often do because they have better masters. In that case, wouldn't that mean the only use of the DSD feature is to play back (mostly illegally) ripped SACDs? And what on earth is the difference between 2.8MHz and 5.6MHz DSD? Is there any point to using this crap or are these features all extraneous?

DSD uses 1bit rather than 16bit or 24bit. The downside of this is huge amounts of unwanted noise but with very high sampling rates there is plenty of space above the audible spectrum to move this noise, thereby making it inaudible. The SACD standard has a sampling rate of 2.8mHz and therefore a theoretical audio range of 1.4mHz, which is way more than enough to account for any difficulties related to reconstruction filters or other required process. This DAC obviously has a mode which exactly doubles that and indeed there are apparently some recordings available in this format and one or two DAC manufacturers are now offering a quad DSD oversampled rate of 11.2mHz. It appears that these oversampled DSD rates which have appeared are purely for marketing (snake oil) purposes. On the basis that if something has a higher number in it's specs, it's easier to market it as better, even if it isn't or even if it's actually worse!
 
Your basic premise is correct, look for the quality of the recording rather than for the distribution format. In some cases the best you will find will be a standard CD (16/44) in other cases it might be SACD or 24/96. You can always take a good master which may happen to be in 24/96 format, convert to a lossless 16/44 format to save space on say your smartphone and be secure in the knowledge that you're not loosing anything audible and are therefore listening to the highest quality available!
 
G

 

[MrMateoHead]

I shared this link in a different thread:
 
http://www.npr.org/blogs/therecord/2013/09/11/219727031/what-does-a-song-that-costs-5-sound-like
 
You can get some of the history of DSD from that article, and sense of why it is being revived today. The conclusion would appear that people are willing to pay more for "high quality", though what is and is not high quality is the hard part. I, for one, don't want to pay $5 for a file download and $50 for an album, but that's just me. The file sizes are atrocious, and there is no guarantee that a 'warm analog like' sound will be heard on each listen. The analog / digital snake oil bid drives me nuts.
 
When I complain about "sound quality" sometimes, I am not complaining about a specific format. What I am complaining about is clipped recordings, or recordings that seem stripped of dynamic range or for whatever reasons are harsh and nasty. An Mp3 has the same subjective properties (to my ears) as the CD it was ripped from, so I am not convinced that new formats and new equipment = better listening. Good headphones have made me aware just how far my recordings range (from amazing to disappointing). Are there ANY standards or guidelines for making good recordings these days?

 

[xnor]

Are there ANY standards or guidelines for making good recordings these days?

 
Standards don't help if your "boss" tells you to "make it louder" and work at a faster pace.

 

[Copperears]

I remember years ago someone writing about the fact that even an unplugged resistor sitting quietly inside a desk drawer is generating self-noise of about 2-3db.

Per resistor.

I decided at that point (1992?) that all the folderol about any playback above 16-bit/44kHz was just that, folderol, especially as I've never lived in an anechoic chamber entirely sealed from street/refrigerator/lightbulb/HVAC noise, any of which probably squash the dynamic listening range to about 35-40db.

And when I've gotten close to that, with IEMs, there's the - what - 20db of white noise that is my blood coursing through my veins and arteries.

Vampires may not experience that, but then again I've heard tell they have a terrible time with tinnitus.




Sent from my iPad using Tapatalk

 

[Copperears]

Absolutely fascinating!

http://youtu.be/j6PpDQ6miBg




Sent from my iPad using Tapatalk

 

[xnor]

I remember years ago someone writing about the fact that even an unplugged resistor sitting quietly inside a desk drawer is generating self-noise of about 2-3db.

Per resistor.

 
Ahhh.. that is what raised the noise floor in my system from -100 dB to over +1000 dB!!! Now I can rest in peace.

 

[Copperears]

Did I say something ludicrous, again?! Damn......

Resistors were bigger back then...... I guess they're smaller and quieter now..... hmmm.... I'll keep trying....

Perhaps I misremembered what I read? Impossible.....

 

[bigshot]

I use Magic Pebbles to combat my noise floor problem.