Bit depth
From DNBWiki
Bit depth refers to the number of bits used to store the amplitude of the sample in a digital audio stream. Bit depths in common use include:
- 8 bit - 80s synths, samplers, video games, Commodore etc
- 16 bit - CD quality
- 24 bit - increasingly common level used for I/O in the pro-audio and "prosumer" worlds
- 32 bit (floating point) - common level used internally with most modern software and digital hardware
- 64 bit - also used internall in high-end hardware/software DSP products
Contents |
[edit] Bit depth explained
Bits refer to "binary digits" - the number is simply a count of how many bits are used to store each sample's amplitude. 16 bit means the largest binary number available is 1111111111111111, or 65,535 in decimal. Thus, 0db (the loudest possible sound) is 65,535 and 65,534 different levels of "quieter" are available. This equates to a dynamic range of roughly 96db.
In turn, with 24 bits available, our maximum value is 111111111111111111111111 or 16,777,215 in decimal. Thus we have 16,777,215 different volumes available, which is roughly 144db of dynamic range.
[edit] Floating point
The same trend does not continue with 32 bit, as 32 bit audio systems are generally floating point.
This is basically like scientific mode on a calculator (1.045681 E12). The first 24 bits stay as they are, but instead of representing an absolute volume, they represent a volume compared to some arbritary norm. the last 8 bits are used as a sort of 'scale factor' to determine how loud the sound actually is in the grand scheme of things. (To be technical, the number stored is the result of multiplying the 24 bit mantissa by the 8 bit exponent. See Wikipedia for details.) This gives us a dynamic range of something ridiculous like 1000db.
[edit] The advantage of higher bit depths
So why use higher bit depths? After all, CDs are only 16 bit, and that's "good enough" isn't it?
[edit] Usage of 24 bit I/O
Well, yes and no. The dynamic range of an orchestra, for example, is often greater than a CD's 96db. As such it is not possible to completely capture an orchestral performance without resorting to compression.
Furthermore, it is not always possible to record everything very close to a perfect "hot" level (near 0db). Every few decibels takes you away from the highest resolution of the digital audio representation (found in the most signficant bits), and pushes your signal down into the lower significant bits and noisefloor, resulting in less accurate sampling.
[edit] Usage of 32 bit internal DSP
The benefit of using floating point 32 bit systems during DSP is that it is very difficult to lose data (as a result of the signal being quiet or attenuated) or clip data (as a result of the signal being loud or boosted). Even with significant gain operations we are nowhere near the 1000db edges of our range, so mathematical accuracy is preserved. For example, when you see clip channels in a 32 bit mixer, you are not actually clipping them since unless the master is clipping (see grid thread).
