intersant .. ik vond dit nog .. wat volgens mij wel aannemelijk klinkt :
Dan Lavry, of Lavry converter fame says:
http://recforums.prosoundweb.com/in...ebbdf07292160b9
A “purely” digital attenuation (via computation, not via digitally controlled hardware)? One should be very careful with it. Say for example I have a CD (16 bits) and I wanted to lower the volume by 6dB. The digital way of doing it is to multiply the digital signal be 1/2 and that will amount to shifting all the bits by one so that the most significant bit (MSB) data is now the MSB-1 location and so on… You are now playing the same music with a 15 bit digital audio… You see, the “new” MSB is fixed at zero, and the old LSB (least significant bit) has no where to shift to (there is no 17 bits, only 16).
The same concept is true for say -10dB, which is not a multiple of power of 2. It shifts one bit (as above) and than you lose about 4 more db on the LSB side… If you want to attenuate by 50dB, you have less than 6 bit quality…
And of course the other issue is dither. When you do such digital attenuation, you are in fact reducing word length so dither is needed. In fact the further you go the more it is needed. That will of course cost you in the dynamic range department. If you do not dither, you will end up with noise modulation and distortions.
Simply put, with a rather crude example, if you have 24 bits digital material, and the noise it say -110dB (about 18 bits), but the digital bus is 100 bits wide, multiplying and/or shifting by say 10 bits is great, but the noise “moved from bit 18 to bit 28” (This is very crudely stated but stated to illustrate the point). At the end you have to remove the lower 76 bits, and end up with 24 bits. The release format does not allow you to have 28 bits to express the same dynamic range, so you lost 4 bits. The top 10 bits are now set to 0, and you 24 bits is gone, It is now at a 14 bits performance level… That is in theory. What happens if your DA can not track the lower bits bits of that 24 word? Say it is a 20 bit DA, than you lose 4 more nits and you are now at 10 bit digital audio. With a 16 bit DA there is almost no audio left. Of course that 10 bits example is 60dB attenuation, pretty extream, but it does illustarte the point...
In my view, analog attenuation is not problem free either, nor is it a walk in the park, but it beats the pure digital attenuation “blind folded with hands tied behind the back”. One (that is not a golden ear) may be able to “get by” with a small range of digital attenuation - each 6dB attenuation cost you 1 bit.
BR
Dan Lavry