I was wondering why the new soundblaster soundcards delivering 24 bit A/D and D/a conversion only yield a 100 db S/N ratio according to the published specifications. The extra eight bits should significantly increase the signal to noise ratio compared to the standard 16 bits.
Also the dynamic range sould be around 144 db, I think, 6 times 24.

Regards

Enrico

Pick up a book on semi-conductor physics & thermal noise (especially)

Thank you for your reply,
I see , so only costly professional equipment can deliver 24 woth low thermal noise A/D and D/A conversion.

The practical limits, without cryogenic cooling, seems to be around -115dB. To approach this -115dB limit is more expensive than entry level cards but not exceedingly so today.

I see, it is an interesting piece of data, however, it is still far from the theoretical value of 144 db. Do you know which value can be reached with crygenic cooling?
and is crygenic cooling employed in high-end professional equipment, like say: the best production studios?
Anyway, for affordable consumer equipment the environmental noise seems not to be the only limitation to appreciate the extended dynamic range offered by the extra bits, also internal noise plays a key role.

I very much doubt that any music hardware at any level is cooled beyond the fans and heatsinks you find in an standard PC. The only digital cryogenic cooling I have any knowledge of at all is (very expensive) scientific laboratory equipment where extremely low noise levels can have some real benefit, such as when communicating with a satellite out around Saturn or listening to galaxies 500 million light years away.

I don't know the limit but it is more a practical matter than anything else. The cooler one get the equipment, the lower goes the internal thermal noise, but the relationship might well not be linear. When you get down to the last little bit above absolute zero, where various materials become super conducting, you will also be approaching the noise reduction limits.

115dB is considerably more dynamic range than any realistic recording or playback setup can utilize. In most cases the 96+dB possible on a CD is too much to realize; it is too expensive to get a listening room quiet enough. Very few CDs utilize the full range available and quite a bit of music could not provide it anyway. The internal noise of the best audio equipment is just not a realistic limitation to any kind of music enjoyment.

I see, do you think that in a consumer system dithering is not necessary because the internal noise is sufficient to to ramdomize the effects of quatization?

my opinion
I suspect that for things such as transfers from LPs, 78s, and cassettes the background noise is always high enough that the low levels signals for which dither is advantageous cannot exist, therefore dithering or not makes no difference.

For original digital recordings that end up on CD, it is probably the case, as I suggested in my last post, that little music utilizes the full dynamic range. If there are no passages that are low enough, I think dither cannot make any difference. Many quality home music systems are optimized for characteristics other than minimum system noise; extremely low level passages that would benefit from dither may often not be separable from the systme noise,even if the environment is quiet enough.

However, the benefits of dither can be readily demonstrated in test environments. Music productions often utilize slow a fade off to zero, even if it did not actually happen that way in the performance. Dither will make a difference. Whether or not you will hear the diffference depends upon the quality and noise level of your listening space, the quality, such as noise and distortion levels, of your equipment, and how much attention you happen to be paying to such low level parts.

I would say there is probably never anything to fear from dither -- reasonaably done, anyway. There are various kinds of dither and it is used with many different noise shaping configuratins. For some of them the dither will be marginally audible by itself, but mostly this is not so.

Maybe also 24 bit quatization doen't require dither because of the greater resolution so 24 bits can used in the A/D converter, than, in a second stage 24 to 16 bit conversion is permormed with the help of dither.

In theory there's still a difference.
y = quantize(x + white_noise)
is better than
y = quantize(x) + white_noise

However, in practice a recorded 24 bit signal usually contains already a lot of noise. In these cases one does not need to dither since the quantized-only version will be fine due to the in-signal noise. Anyhow, you'll be on the safe side if you do dithered quantization even on 24 bits signals.

Sebi

So we can conclude that it is better to use dither anyway. A digital speudoramdom generated dither has the additional advantage not to increase the noise floor any farther, because it is normally substracted from the least significant bit of the encoded signal.

That doesn't make sense to me. We are talking about non-subtractive dithering, right ?

Sebi

The dither normally employed definitely adds noise but noise shaping assures that most of it is in the higher frequencies, not readily heard. Its not that it isn't there, that the result isn't noisier than without, it is rather that the benefits generally outweigh the added noise.

In CoolEdit/Audition, and therefore I assume in most audio editors, there is a varied choice of dither types and noise shaping curves. By applying them to a selection of digital silence it is easy to see and analyze the results. What I normally use for LP transfers (where it probably isn't actually necessary, but just in case) raises the noise from none to RMS AVG = -89dB, well above the soundcard noise floor of -98dB or the theoretical minimum for CD. With closed headphones I can just barely hear that with the amplifier at an insane volume level, far above where I could ever listen to any music without great pain.

I heared that Window 2000 SP2 or higher limits resolution of 24 bit sound cards to 16 bit samples (=96 dB). So use Windows 2000 SP1.

Um I think you are confused. Win2k SP2 is perfectly able to output 24 bit audio via WDM drivers. It is possible that he infamous kmixer could screw things up though if it is allowed to. However in most cases where the users actually care about 24 bit audio they would be using ASIO not WDM completely bypassing the Windows sound system.

Well, I took the idea of subtracting dither from the encoded signal from the book "Principle of Digital Audio " by Ken C. Pohlmann Howard W. Sams & Company page 73

I quote: [ An analog dither signal necessarily decreases thè S/N ratio of thè digi-tization System. Blesser has proposed a System with a digitai dither signal that would preserve thè S/N ratio, as shown in figure 4-9. Noise is a random valued signal, which may be simulated by generating a quickly changing pseudo-random sequence of digitai data. This can be accomplished with a series of shift registers and a feedback network comprised of EXCLUSIVE OR gates. That sequence can be input into a D/A converter to produce analog noise, and thè signal added to thè audio signal to achieve thè benefit of dither. Then, following thè signal A/D converter, it may be digitally subtracted from
Fig. 4-8.

SAMPLE AND HOLD CIRCUIT


ANALOG-TO-DIGITAL CONVERTER —•<! j - DIGITAL OUTPUT



DIGITAL-TO-ANALOG CONVERTER DIGITAL PSEUDO-RANDOM NOISE GENERATOR


thè audio signal, leaving thè dynamic range of thè originai signal. The am-plitude of thè dither is not criticai. A further benefit is that inaccuracies in thè A/D converter are similarly randomized. Digital dithering is also an im-portant consideration in digitai signal processing computation, to decrease round-off error.
]

What do you think of that?

This has been "proposed;" has it ever been implemented?

The mechanics of this proposal are easy enough to follow but the consequence is much less clear to me. While I have some feeling for the way 'normal' dither works, I don't understand how this helps low level resolution after is is "digitally subrtacted form the audio signal" -- unless it is understood that 'normal' dither will be added before the D to A process.

I believe I do see the part about randomizing the A/D "inaccuracies." I think this implies a 'ladder' type of converter rather than a delta-sigma type.

I think I can understand how the effects of quatization remain ramdomized afer the substraction of the ramdom sequence from the encoded signal. However, today sigma-delta one bit noise shaping encoders are based on different principles.

Thank you to all for your valuable replies