Oh !
May-be a little time more to think about the difference between a musical signal, constantly changing in frequencies and levels in a pretty random form (from an electronic point of view) and pure sinusoidal signals ?
I understand why we need to add dither each time we go from analog to digital, or we change sampling frequencies or number of bits while we make measurements.
No, you can back out dither by reducing the resolution of the digital signal, say by attentuating it by > 6 dB when there is 1 bit of dither.
That is a halfways-decent generality, but not a 100% truth. Dither is cheap and generally harmless so it is wise to just put it in. It is often added unintentionally because most audio processing may add some more dither.
People say stuff like that, but its not the constant changes in music that make the difference, it is the simple fact that there are a lot of different frequencies present that makes things significantly different.
The actual rule is that dither needs to be added when ever resolution is decreased.
yes, but natural random noise in a recording studio is, at least, an order of magnitude higher than any dither.
Not to mention mikes and their preamps. But if you're going from 24 bit recording to 16 bit distribution, you don't just truncate.
Exactly.
Until we compute in 32-bits (gain, filtering ...), it is OK. Going to 16-bits but also to 24-bits asks for dithering.
2) analog noise differs from digital noise. 16 bits have to be dithered even if the original sound was digitized from tape.
Last edited:
It is a bit of a peripheral point in all its mootness, but as a matter of fact, when you pick a nit and loose a bit, SN goes down with 6dB too, and it is in this bit of resolution that the effect of the dithering resided. So, you can back out dither.
However, the 1 bit truncated signal has now become 'de-dithered', literally, and in order to smooth over the re-introduced quantization noise, you need to re-dither again before the signal can be of any use.
Last edited:
I don't know about "any dither" but that is generally true for most practical dither.
It can be instructive to try to measure the residual noise (usually room tone) in real world recordings.
Typically the background noise on a real world recording is no better than -70 dB, but a few remarkable exceptions exist.
So, your order of magnitude estimate is in pretty good company.
Most real world recordings are pre-dithered by their background noise.
Just about any analog tape has enough self-noise that transcribing it to 16 bit digital with a gain setting that properly aligns the dynamic range of the tape with that of the digital domain it is being transcribed to need not have extra dither added.
However, the caveat "Transcribing it to 16 bit digital with a gain setting that properly aligns the dynamic range of the tape with that of the digital domain"
is not a global rule, slam dunk or generally given fact, so adding dither is good cheap insurance.
Most modern converters are Sigma-Delta and they are generally dithered, anyway.
???
room tone is not noise, it needs to be dithered when any change is made to it, changing less than 6 db does not negate the need for dither.
dither needs (mathematically) to be decorrelated, otherwise it generates birdies and other spectral lines which may be called distortion.
There are those I respect that feel that even dithering from 32 bit float to 24 bit is audible when taken in the aggregate (ie mixed).
Cheers
Alan
room tone is not noise, it needs to be dithered when any change is made to it, changing less than 6 db does not negate the need for dither.
dither needs (mathematically) to be decorrelated, otherwise it generates birdies and other spectral lines which may be called distortion.
There are those I respect that feel that even dithering from 32 bit float to 24 bit is audible when taken in the aggregate (ie mixed).
Cheers
Alan
Yes, dither is 'good cheap insurance' that tends to hide potential distortion artifacts. Analog tape got MORE LINEAR with lower and lower operating level, what does digital do? Just the opposite, so hiding the non-linearity is a comforting thing.
Now, I will go back to the measurements I am making with my SR-1 analyzer, whether or not it is dithered or not. I never thought to find out.
Now, I will go back to the measurements I am making with my SR-1 analyzer, whether or not it is dithered or not. I never thought to find out.
Yes, and, while a lot of people complaigned about the sound of the first CD, i used a sigma-delta DAC that I prefered a lot at this time.
My purpose is not to desagree with Dithering (the negative impact is neglecting) but to moderate the idea that, with real music, its absence can be a real cause of audible artifacts.
Last edited:
Completely incorrect. Either you believe this, in which case you have some learning to do, or you know it's incorrect and you're just trolling.
It really is a sad thing to see things like this from someone with experience and background.
George, this isn't the most mathematically precise explanation. I've posted some nice papers without (IMHO) overly difficult math. I guess no one pays any attention. Missed above, yes JC harbors the totally incorrect misconceptions that linger, I guess will linger, ad nauseum.
Well totally off topic, my batch of Russian caviar just came in today, perfect for this evening's celebrations.
So taking a risk on eBay, paid off!
So with the Champagne, shrimp and caviar among the other bits I am off for a great evening.
My best wishes to all of you for the coming year. May you, your hearing and equipment all improve next year.
So taking a risk on eBay, paid off!
So with the Champagne, shrimp and caviar among the other bits I am off for a great evening.
My best wishes to all of you for the coming year. May you, your hearing and equipment all improve next year.
Thanks Pavel and Scott. Even I can understand the difference between noise in the analog realm of things and what is happening every time you transfer a digital signal. The math may be over my head but the concept of dithering for the lsb I do comprehend from all of your conversations.
Is there ever a reason not to use dithering in the digital domain, any caveats about that? What difference is there between multi-bit transferring and single bit as far as dithering goes?
Is there ever a reason not to use dithering in the digital domain, any caveats about that? What difference is there between multi-bit transferring and single bit as far as dithering goes?
In all seriousness, I can recommend some good basic texts, if you are truly interested.
In one case, you compare a value to a scale, in an other, you measure successively a difference between two values, accumulating little random errors (noise) ?
- Status
- Not open for further replies.