Archived from groups: rec.audio.tech,comp.dsp (

More info?)

Jon Harris wrote:

> "jim" <"N0sp"@m.sjedging@mwt.net> wrote in message

> news:40ae9845_6@corp.newsgroups.com...

>

>>

>>Jon Harris wrote:

>>

>>

>>>>They *WON'T* have the same SNR. THAT'S THE POINT!

>>>

>>>Right! Halfscale with have 6dB less SNR than full-scale.

>>

>>Not necessarily. Its perfectly possible that the same bit pattern will result

>>from both input sequences or an even cleaner one from the half scale if the

>>full scale contains more noise.

>

>

> I thought we were talking about the difference between full-scale and half-scale

> sine waves?

>

>

>>I guess your assuming that quantization error

>>is the only source of noise in the original signal. Even if that were true the

>>difference in error from half-scale to full scale is insignificant compared to

>>the amount of error added when going down to one bit - so few bits will end up

>>different that you won't be able to hear the difference. If I understand

>>Randy's dither algo the amount of dither added decreases proportional to the

>>final bit depth.

>

>

> I think you have that backwards. The fewer bits, the more dither you need, so

> dither is inversely proportional to final bit depth.

>

>

>>Again that means when you get down to one bit so few bits

>>will be changed you won't be able to hear the difference.

>

>

> I think you are a bit confused about the way dithered quantization works. Given

> a clean input signal, properly dithered quantization to N bits adds noise based

> on N alone--it is NOT depended on the signal. Now if you quantize a signal

> identical to the first one in every way except 6dB quieter again to N bits

> (assuming the original signal has >> bits than N) you will have the same amount

> of quantization noise, but since the original signal was 6dB softer, you have

> 6dB worse SNR. It really works, I've tried it!

>

> I'll use my analog cassette tape analogy one more time. Imagine recording a

> very low level signal to a cassette tape. Then boost the gain on playback, say

> 60dB. You can still hear the recorded signal, but there is plenty of tape hiss.

> Now imagine doing the same thing again, except with the original signal 6dB

> softer. Boost by the same 60dB and the tape hiss is still at the same level,

> but the original signal is 6dB softer, hence 6dB worse SNR. Dithered

> quantization works THE SAME WAY! The dither (noise) creates a noise floor just

> like tape hiss. (The only difference is that the frequency response of the tape

> noise may be different than the dither noise. In fact, the digital designer can

> choose the sound of the noise floor using noise shaping.)

>

> BTW, that's one of the big breakthroughs about dither--it makes digital sound

> like analog!

Relative to full scale on 15 bits plus sign, one bit is 90 dB down.

bringing that back to full scale is the equivalent of sign extending

down to all the bits. It needs some mental contortions to decide what to

make of it.

Dither is properly done using an amplitude equal to the least-

significant bit. When only one bit is significant, does that mean adding

100% noise? If not, what?

Jerry

--

Engineering is the art of making what you want from things you can get.

¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯