bit-depth

I recently wrote an article about 24-bit audio recording where I said I would write something about 16-bit recording (using a 16 bit interface of some sort).  So as promised, here is that article.
But in keeping with what we’re all about at Home Brew Audio, the explanation is intended to help you make better recordings and will be explained in a way that keeps it simple.  I used to work for a guy who always said “explain it so that my grandmother could understand it.”  I know some pretty darned sharp grandmothers, so I won’t phrase it like that.  But I know what he meant.  There is no shortage of material out there talking about digital recording and bits, but there are precious few (I couldn’t find any, but I assume there are some;)) I would call written so that normal people could understand.  Most of the time my eyes started glazing over before I got to the 2nd paragraph; and I have a bachelor of science degree and a masters degree.  Basically If I can’t understand it without slipping into a coma, I assume most people are also going to have trouble.  So I do my best to write non-coma-inducing articles.  That said, let’s talk about bits!
In Why You Might Want a 24 Bit Audio A-D Converter, I went over the basics of converting analog audio (sound moving through the air) into digital audio, something a computer can understand and manipulate (ones and zeros).  The metaphor there was from the movie Tron when the MCP converts Kevin Flynn from a regular person (like analog audio) into a digital person made up of ones and zeros.
The reason I want to talk about 16-bit audio is that it is probably the most common type of digital audio.  Also, we’ve been listening to it for decades since CDs use 16-bit audio.  The number of bits is sometimes referred to as “bit depth.”  Whatever.  What you need to know is that we’re talking about the length of words here…digital words.  I referred to “ones and zeros” above because computer language is all math and the highest single digit is “1.”
How Bit Depth is Like Champagne
Think about human math.  The highest single digit is 9.  Now imagine if our math had only 1 digit available.  If I try to text a friend how many bottles of champagne I have for the party (we have ten), I would not be able to convey that information very well. I’d have to say something like “I have 9 bottles and I also have 1 bottle.”  Silly huh?  If I could use 2 digits,  now I can say “I have 10 bottles.”  I can convey more information.  But that’s only good up to 99 bottles.  To say I had any number between 100 and 999 bottles (woo hoo, now that’s a party!), I’d need another digit.  And so on.  The more digits I have available, the more and better information I can convey.

I’d have to say something like “I have 9 bottles and I also have 1 bottle.”  Silly huh?

Here is how this relates to digital audio.  During analog-to-digital conversion, sound is captured or “sampled” thousands of times per second.  The converter has to be able to describe how loud or quiet (amplitude) the audio from each sample is.  And since computers use binary language, ones and zeros, they don’t get the luxury of 9 “levels” per digit like us.  They only have 2 levels per digit, a 1 or a 0.  So converters need several digits to accurately represent that amplitude.  The more digits available, the longer the “digital word” and the wider the dynamic range that can be accurately represented and played back on our systems.  Each digit is actually called a “bit” (hey remember the Tron reference and the bit character that can only say “yes” or “no”?) and we need a long enough digital word to reproduce really quiet sounds as well as the loud ones.  If the word is short, say 8 bits long, our audio would theoretically have quite a lot of noise – though not nearly as much as most people think (see our post – A Common Misconception About Bit Depth In Digital Audio). In fact, at 8 bits the low-level hiss would still be waaaay quieter than even cassette tapes!
So how much dynamic range do we need?  Dynamic range is the difference between the quietest sound and the loudest sound.  If your digital converters can’t translate enough dynamic range for a vocal recording, that narrow dynamic range will bottom out (at the digital noise floor) before it can deal with the quietest parts of, say, an audio recording.  So instead of a nice intimate vocal you get an ugly noise.
How do you know how much dynamic range you need?  
In theory, the more the better.  But in real life you want the “real” (or “analog”) noise, stuff like computer drive noise, electronics, and ambient room sounds to be louder than the digital noise floor.  In most systems these days, you can go all the way down to about -80 to -90 dB or so before you just can’t get the “real” noise any quieter (remember that in digital audio, Zero dB is the loudest point or “digital ceiling,” so loudness is depicted as negative numbers).  So in theory, as long as the digital noise floor is lower than -90dB, you will be more than OK.
So how do you know what number of bits in a digital word will give you enough digital dynamic range?
Here’s the basic rule-of-thumb.  For every bit (digit) added to a digital word, you get 6 dB of added dynamic range.  So if your converters use 8-bit conversion, you get 48 dB of dynamic range (8 x 6).  That means you can record audio whose loudest peak is just under the 0 dB mark, and whose quietest part is right around -48dB.  Anything quieter than -48 dB won’t be converted properly and will sound a mess!  Just garbage-y awfulness.  A 16-bit converter gets you 96 dB of dynamic range, which is well beneath our theoretical analog noise floor of -80 to -90 dB.
The Rest is History 
With the above knowledge, the audio industry sort of adopted 16-bit audio as its standard, which among other things is why music CDs are 16-bit audio.  For most people recording in home recording studios, 16-bit audio will probably be good enough.  Your computer can also handle 16-bit audio files faster than higher bit words (24, 48 and 96 are other common bit lengths in audio).
There is more to tell about the properties and relative advantages of using 16, 24 or even higher bit rates.  Then there’s the whole issue of sampling frequency (you’ve probably heard or seen the term 44.1 KHz bandied about).  I will address these in other articles, giving each the Home Brew Audio treatment so “regular people” can understand the concepts. In fact, the one about sampling frequency is right here.
Cheers!
Ken