Finally you say something concrete! Patrick Dixon;289480 Wrote: > > The LSB in a 24-bit PCM signal does indeed represent approximately > -144dB wrt to full scale. However, in the real world, you have to > allow some headroom in the recording to prevent (excessive) clipping, > so the actual music you are listing to may well be (say) -20dB wrt full > scale too. The LSB however, stays exactly where it is at -144dB wrt > full scale, but only (say) -144dB minus -20dB down from the actual > music volume - ie -124dB wrt the music
It doesn't make any difference where it is wrt music if it is inaudible even without music. > In addition, you are implementing a digital volume control, so you are > reducing the music volume still further within the available PCM range, > maybe by (say) 20-30dB. Once again the LSB stays where it is at -144dB > wrt full scale, but now your actual music level is also (say) -40dB or > -50dB below full scale. The difference between the actual music volume > and the LSB rounding error is now -104dB or -94dB which is considerably > less than the -144dB claimed elsewhere in this thread. You are making precisely the error I suspected - you are confusing S/N with absolute magnitude of noise. If the signal itself is too quiet to be audible, it doesn't matter much what the S/N is, does it, Patrick? > Obviously the actual level of the rounding error wrt to the real music > level will vary depending on the recording and the amount of volume > control reduction used, but in the right circumstances - with a decent > system and a decent recording, it is possible to hear the effect. Let's check this yet again, shall we? We'll use the most generous possible numbers for you. Suppose full-scale at the SB/preamp output is +-2V, and suppose you have an amplifier with a gain of 30. As a sanity check, a full-scale sine wave would then draw 450 watts/channel into 4 ohm speakers, which is about right. Let's suppose you have a system capable of handling that and you turn the volume up to max. Now, take a signal which oscillates between 000...01 and 000...00 at (say) 2KHz (right where it's easiest to hear). That's the largest signal a rounding error can ever make, and I've made it as audible as possible by putting it in the sensitive range of human hearing. How loud is that, going through the above system? Well, if full scale is 2V, this signal is 72dB down in voltage (so about 2E-7V), or 144dB down in power. So the power going into each speaker is 450E-14.4 = 2E-12 watts: two trillionths of a watt of power. Suppose the speakers are quite sensitive: 100dB SPL for one watt at one meter. Then the sound produced by our error signal at one meter is 100-144+10log(450) = -17.5, which matches quite nicely with what I found before. The absolute threshold of human hearing in an absolutely quiet sound booth is 0dB SPL (that's why it's called 0dB). So this signal - played without music - is nearly 20 dB below the threshold. To make it just barely at threshold - in an absolutely silent room with no music on top of it - you'd need a total analogue gain of *300*. Full scale would be 600V, and the amplifier would need to supply *45,000 watts/channel*). You're wrong, Patrick, and that "bug" you filed - and wasted SD's time with - was a figment of your imagination. Reading through those threads it's quite clear they knew that, but they're a good company and they accommodate their customers. -- opaqueice ------------------------------------------------------------------------ opaqueice's Profile: http://forums.slimdevices.com/member.php?userid=4234 View this thread: http://forums.slimdevices.com/showthread.php?t=45736 _______________________________________________ audiophiles mailing list audiophiles@lists.slimdevices.com http://lists.slimdevices.com/lists/listinfo/audiophiles