The dynamic range of 16 bits

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Pedja said:
You will never, never get signal, offset or anything written as 1/2LSB. I have never tried CoolEdit, but from the screenshots already posted here it is obvious it oversamples – why and how, by default or not, I do not have a clue and it is not important for now, but screenshots showed bigger resolution than those original files had. It is not hard to understand this way it can easily and happily move something up or down for the level that matches to the 16bit’s 1/2LSB. The point is that you’ll never, never save anything of it into the 16 bit file.

I have some wav files considering this… if I do not post it in the next hour it means I went to sleep, but will post them tomorrow.

I don't have any idea how Cool Edit is coded. "oversample" does not appear in the help files. It may attempt to draw a smooth curve through the data points (oversample, more or less), or it may actually use floating point data and save integer data. I cannot save since my demo copy ran out months ago (this is only about the third time I've used it). I have no doubt that it will save the 0`s and 1`s that were on the screen that corresponded to a 1/2LSB amplitude square wave with a 1/2LSB offset.


Rob M said:
Okay, let me revise that.

:D

I believe KYW attempted something similar many pages back, but also had problems. He attempted to use an offset of 1/2LSB on a 0dBFS signal, then scale the result by -96dB. That actually scaled the offset as well....

Regards,

Mark Broker
 
Re Fred Dieckmann : The the number of possible DAC output voltage levels for the smallest possible output signal.


Using proper dither the lowest signal is below the dynamic range!

See attched picture...


Cheers

hpW
 

Attachments

  • 110db.gif
    110db.gif
    9.3 KB · Views: 260
Konnichiwa,

jcx said:
Which is Right? – I vote neither

If you want to move beyond pinheaded angel counting and address practical audible resolution and 16 bit word length issues you quickly discover that digital domain manipulation must be done at high bit resolution and dithering used to create the final 16 bit representation. Using dither and noise shaping, audible resolution in 16/44 CD digital audio can easily exceed 96 dB – with the greater advantages of sub-lsb linearity and decorrelation of quantization noise – only people shipwrecked on desert islands, working with bamboo and coconuts have any excuse to be limited to undithered or 16 bit resolution editing for even 16/44 mastering

(see next page in first link)

http://www.digido.com/ditheressay.html

Well, dither can encode signals below the theoretical limit (whichever we accept as such), but only by raising the noisefloor to at least -90.3db. Now if I where to go and claim that the signal below the noisefloor must be considered as being part of the dynamic range for an analogue system I'd be harshly taken to task. So I would think the "dither" escape from the trap is not really one.

BTW, do I think that a properly dithered 16-Bit digital recording sounds better? Sure. Does it have anything approaching 96db dynamic range? Don't be daft. At least not if you keep things on level pegging with the charaterisation of analogue systems, to which these digital systems are ragulary compared to....

If you remove the comparability you can of course come up with almost any number you like, just define your conditions, exactly the same way my PC Speaker Setup at work is rated "200 Watt". Of course, in the small print it says of course PMPO - Peak Maximum Output - RMS is more like 20..25W with only one channel being given signal.

And exactly that was the very beginning point back in 1997. I contended that thet 96db dynamic range claimed for CD was in effect a marketing figure with no comparability to those figures used to charaterise analogue systems.

And I cannot fail to notice that still no-body has shown a wave file that would be recognised by any Pro-Audio Digital Level meter or indeed any piece of software as haing a signal level of -96dbfs.

Sayonara
 
Do you really talk about the same thing?

Digital noise is one step, not zero, one step minus and one step plus. One steg compared to fulll signal is 16 x 6 db + 2. I think we all can agree on that noone has measured 90 dB at the most from a 16 bit system.

The question is: How noisy is a 100% digital signal? How small is the smallest signal. The smallest signal is only one step but this signal has also half a step DC-level.
 
Was I wrong if I thought that the dither was meant to stay out of this story…?

File with 1LSB amplitude is here: maximum sample value = 1, minimum sample value = 0. Rob already uploaded something like this, but (no offence) this is probably more convenient. I am pretty convinced that any software analysis will never show this has anything but -90.3 peak. Since as long as this is 16bit format, none offset could be put into this, file is converted to the 172kHz/24bit format. After conversion it normally shows the same peak value. Then data were centered on the zero axis and… These three files are here zipped, so you can proceed to fiddle... Maybe you can put it back to 16 bit format…

Again, ignoring files, programs and stuff, it is only important how you define dynamics: as the smallest possible signal amplitude referenced to the peak amplitude or as the smallest possible change in the amplitude referenced to the peak-to-peak amplitude.

Pedja
 

Attachments

  • 16bit-1lsb.zip
    7.3 KB · Views: 41
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.