I’ve mentioned numerous times in blog posts and episodes of “Ask Joe” that you need to record at 24-bit. Then I realized that I’ve not written an article specifically on this topic.
When you’re recording digital audio, there are two main settings that you will come across at some point – bit depth and sample rate.
We won’t get into sample rate today. I’ll save that for a future article. To summarize, sample rate measures how many times per second the audio is “sampled,” or measured. Common sample rate values are 44.1 kHz, 48 kHz, 88.2 kHz, 96 kHz, and 192 kHz. Is higher better? Hmm…I have my own take on this, but that’s for another day. 
For today, just know that if you record at 44.1 kHz, that means the volume of the audio signal gets measured 44,100 times per second. When you put all these tiny measurements together, you get a waveform.
Okay, moving on.
What is bit depth?
Sample rate determines the frequency with which the system measures the volume of the audio. Bit depth determines how many different volume measurements the system has to work with.
In other words, if you think of the audio as being measured by a ruler, the bit depth is how many notches that ruler has. Some rulers only measure in whole inches (low bit rate), while others allow you to measure within one-sixteenth of an inch.
A ruler with more notches allows for a great number of measurement options, and therefore a more accurate measurement.
Comparing Bits to Inches
Okay, so the higher the bit depth the more individual measurements we can achieve. How does that relate specifically to recording?
If you’re recording at 44.1 kHz, then you’re telling your analog-to-digital converter (i.e. your audio interface) to take a volume measurement of the audio once every 1/44,100th of a second. How does it measure the volume? In bits.
1 bit = 6 dB
“dB” stands for decibel. It is logarithmic measurement of volume. If you increase the level of a signal by 6 dB, it will sound twice as loud.
So, for our converter to measure the signal at 2 bit rather than 1 bit, it needs to be twice as loud.
The 16-Bit Ruler
Digital systems don’t have “in between” measurements. Everything is cut and dry in the Land of Didge (shout-out to Slau).
So, the smallest unit of measuring volume in a digital system is 1 bit. There’s no 1.5 bit or 1.42983003 bit.
This means that in a 16-bit system, you have 16 notches on your ruler. 16 potential measurements for your audio. That may not sound like a lot, but keep in mind that these measurements are being taken thousands of times per second.
Alright, we said that 1 bit equals 6 dB of dynamic range (or volume). What is the dynamic range of a 16-bit recording? The answer is 96 dB.
96 dB? That’s great, right? Sure it is…in a perfect world.
Noise Floor
The problem you run into is noise. Every audio system out there has some amount of inherent noise.
In other words, no recording system is perfectly quiet. The electrical components generate a low-level noise. Each piece of your system contributes to the noise party. All of this noise adds up, and it’s called the noise floor.
This doesn’t even take into account any room noise that might get picked up by a microphone.
The noise floor essentially “steals” away some of your dynamic range. Let’s say that all the noise added together was 18 dB. That’s 3 bits.
Since this noise occupies the bottom 3 bits of your system, the level of your audio needs to be recorded ABOVE 3 bits (or 18 dB), or it will be lost in the noise. So instead of having 96 dB of dynamic range, you realistically only have 78 dB (or 13 bits).
The gap between your recorded signal and the noise floor is getting smaller. This means that if you don’t record your signal loud enough, you’ll end up hearing this noise in your recordings. On the flip-side, if you record your signal too loud (to stay well above the noise), you’re in danger of clipping.
24-Bit to the Rescue!
Enter 24-bit recording, super-hero cape blowing in the wind.
Now, instead of giving your converter 16 measurement options, you’re giving it 24. And if you kept your calculator our from earlier, then you know that 24 bits x 6 dB = 144 dB dynamic range!
An audiologist will tell you that our ears aren’t even capable of hearing a full 144 dB of dynamic range. However, having this much available dynamic range allows you to create greater separation between the recorded audio signal and the noise floor.
When you add in the 18 dB of noise we have in our make-believe system, and you drop the usable dynamic range down to 126 dB, you still have a TON of breathing room left.
Check out this diagram:
As you can see, the 16-bit system is still fairly close to the noise floor. The 24-bit system, however, towers above the noise floor, making it much less of an issue when recording.
In a 24-bit system, you don’t need to record the levels super-hot, because you’re signal is not nearly as likely to drop down into the noise floor. This leads to better sound quality, less noise, and less stress when recording.
Do you record at 24-bit? Leave a comment.
[Photo by internets_dairy]
You explained that well, I esp. like the ruler analogy. I’m looking at a 16bit recording device and could you tell me…
What was the bit depth equivalent of the recording device used for Pet Sounds or Abbey Road? Is there a bit depth equivalent to those tape machines? What was the noise floor on those machines? (not the noise of the mics, preamp & etc they used, but the recording tape machine)Curious. Thanks.
Great Discussion! I think that chart is a little misleading though. It makes it seem as though recording at 24 bit will give you much more headroom. I don’t believe this is the case. To my knowledge recording at 24 bit will give you increased resolution and low level detail (ala more dynamic range of course) – If you have a noisy signal, switching to 24 bit will not pick up less noise. Make sense?
…and I just read Kim’s post above mine. She put it so much more eloquently. Thanks.
Jay, 24-bit does actually give you 144 dB of dynamic range vs 96 dB. So that’s a significant difference. Whether that difference is “headroom” or not is more of just a matter of semantics, since digital systems don’t have “headroom” like analog systems do. There’s either no signal, signal, or clipping.
I’ve done an experiment with a nice hand-held recorder where I switched it back and forth between 16 and 24. There was more audible noise at 16-bit.
That’s not a super scientific experiment, but it was definitely intriguing.
Good clear writing, well done.
Unfortunately, it’s not 100% correct. Noise floor added by analogue equipment is relative to nominal (either 0dBfs or somewhere like -18dBfs, depending on how you view the world).
So if your analogue equipment has a collective noise floor at -72dBfs (as per your example), it will be at -72dBfs regardless of whether you digitise at 16 bit or 24 bit.
Where 24 bit recording has an advantage, though, is when your analogue noise floor is lower than -96dBfs. Most good mic preamps, for example, will have over 100dB dynamic range.
In other words, when recording at 16 bit, your highest noise floor is actually coming from the digitisation process itself. When recording at 24 bit, this digital noise floor is now far below the noise floor from the other analogue equipment in the chain.
I’ve written a bit more about this issue here:
http://kimlajoie.wordpress.com/2009/06/29/bit-depth/
http://kimlajoie.wordpress.com/2009/09/03/dynamic-range-and-headroom/
Hope that helps!
-Kim.
Great, thanks!
wow its great to see some folks getting it in. For me . I have been a Cubase user since day one, the 32 bit setting works great when using plug ins and I’d have to work real hard to overload my system. The biggest and best use of sampling rates is 88.2 it just adds the air to a mix as Dolby on Dolby off did when using tape. (yes I am that old)Its very overlooked . If your using Protools , I have noticed is really makes a difference in your final mixes, even when converted to 320 mp3’s. I don’t think we as audio professionals and those who record at home, can argue about sample rates etc, when we have the music buying public buying tracks from Jay Z . which are made from 8 channel stems that are 320 mp3’s. It’s more “what” your recording than the “how” of the music, sound scape, etc. I recently mastered a Grammy nominated project that was done on a very old PC, with one keyboard work station, and a single mic. Lets just make good quaility music that stands up next to anything ” major label”. Nuff said!
Joe,
Thanks for the insightful post. It totally de-mystified the whole bit depth and has me racing for the studio now to hear the difference for my self.
I really appreciate you covering this topic as it seems to be just as important as signal chain, gear, and talent. I’m the type of guy that needs a little logic to become a believer and when I ask most people of the subject they just cop out and say ‘Cause that’s the way it is’ now I know!
You’ve made a believer out of me. Really love the site! Thought I’d leave a comment so you knew to keep up the good work!
Thanks,
-Ox
Great article, totally on board with you, but here is an advanced idea / question:
db has 2 modes of perception. Mathematical and audible. In the physical world the “doubling” of sound has a math based formula where as 3db the “sound energy” or power of the waves doubles. This is an “on paper calculation”.
On the other hand it takes more (and even this is variable due to room, barometric pressure, humidity, etc) to hear “the doubling” of the sound.
How does the math you discuss relate to these imperfect human characteristics?
One other thing you need to put into the mix is dithering and how LSB’s are handled. I think you will touch on this (or may have already).
When migrating from one bit depth to another or processing in an different bit-depth environ, there are bits “added” and “dropped”. The LSB (least significant bit/s) are the ones to be dropped. The “32 bit processing” can frequently cause audio havoc as you get poor dithering and bad bit handling. The concept of “floating bits” enters the picture (Slau begins to hit on this above). A topic of great discussion. How and what floats, when are they added, how, and when are they dropped in the signal path.
Sound trivial? Not so much – Any time you change sample / bit / and other attributes you diminish the quality of your sound (fact).
Joe you kick major butt on these topics. But anyone who digs on this article should think about the digital signal path and work really hard to keep a consistent rate/depth for the whole signal chain. A 48 bit internal may not be better then 24bit.
This is but the beginning, look here for some real freaky stuff (I could teach a semester on this):
http://en.wikipedia.org/wiki/Equal-loudness_contour
F-R-E-A-K-Y!
Joe you roxxor the tex-nology.
Ah…the “Fletcher Munson” curves. Mix sounds just right at 85 dB? Turn it down and watch it crash and burn.
Marc said: “Any time you change sample / bit / and other attributes you diminish the quality of your sound (fact).”
Not quite. You can up-sample (go from 16 to 48 bits, say) with no quality loss. Why? Because you don’t need to do any rounding. You’re essentially just adding “.000000000″ to the end of the number, so if in 16 bits you have a sample of, say, 4325, in 48 bits it’s basically 4325.000000000 (yes I’m simplifying). You can then convert that back to 16 bits by lopping off the decimal point & all those zeroes, and you get exactly the same 4325 you started with.
The problem comes in if you have (e.g. after EQ processing) a 48 bit number like 4325.345823748. When you go to chop off the decimal point, you are losing information & detail. So generally downsampling (decreasing the number of bits) does lose quality, while upsampling (increasing bits) does not.
Bonus info: upsampling does not increase sound quality, by the way. I’ve been asked this many times. Aside from certain “sound repair” processes, you can’t really get quality back once it’s gone.
~peter in oakland
So does recording at 24 bits eat up more processor power during recording and mixing? Because my out-dated hardware has a problem with too many 16-bit tracks once I start to apply plugins.
Perhaps, Joe, you could write an article on making the most of available system resources?
It will theoretically require more processing, but I doubt it’ll make a noticeable difference in available processing power. You WILL notice an improvement in audio quality.
I actually did an entire series of posts on preserve processing power. Here are the links:
Preserve Processing Part 1 – Bus Several Tracks Through the Same Effect
Preserve Processing Part 2 – Commit to Plug-in Settings
Preserve Processing Part 3 – Offline Processing
It probably won’t matter for mixing, because (as Joe pointed out) most plugins operate internally at even higher bit rates (48 bit is probably most common; 32 & even 64 bit are out there too). So your 16 bit files get changed to 48 bit, just like my 24 bit files do. The actual work of the compressor/EQ/whatever ends up being the same.
For tracking it probably wouldn’t make much difference to the processor, but it’s 1.5 times as much work for your hard drive. So if your disk is up to spec (7200 rpm, and internal, firewire, or esata) you’d probably be able to get away with it.
~Peter in oakland
Another great article Joe,
I run every session I have (barring voiceovers) at 24bit/88.2. I’ve found that my finished mixes have more air to them when compared to 16bit/44.1. Now I won’t say that it is because of the bit rate or the sample rate as there are two variables that I am changing at the same time. I have noticed that 24 bit (in pro tools) sounds much better when you leave plenty of headroom on the individual tracts. In the 16bit world we always wanted to “maximize (y)our bits” The battle was always to get your levels as hot as possible without clipping. This was always super stressful when recording, and then a pain when mixing because all of the plug-ins would clip. In 24 bit land just get a decent signal to pro tools, allow anywhere from 6 to 12 db or headroom, and you are golden.
It really is freeing once you realize you can give up trying to get the meter within 2 pixels of the clip light while recording.
Hi Joe,
LOL Hollaback!
Hey, I just wanted to point out something that I think is important to clarify. Rather than “16 notches,” a binary system can represent many thousands of steps. Much like the ratios of decibels, the bit depth is logarithmic. For example, with only 4 bits (for values of 1, 2,4 and 8), it’s possible to represent 16 discrete numbers (or levels). The real strength is in the doubling of each value with each additional bit. So, with only four more bits, we can now represent as many as 256 values. That’s a 16-fold increase in resolution for only 4 more bits. One can really begin to see the power in resolution as we climb the bit tree. At 16 bits, we’re in the 65/532 column and can represent any value up to double that number. Adding another 8 bits to bring us up to 24 gives us an astonishing number of discrete values.
One other thing I wanted to point out is the issue of 32-bit values. Some folks have the notion that there are systems capable of recording at higher resolution than 24-bit. Currently, 24 bits are all we have in terms of converters. What happens beyond those converters is simply processing that occurs at a higher resolution in software. In fact, Pro Tools works internally at 48 bits. Thing is, when we’re coming in and out of the real world, it’s a maximum of 24 bits.
Keep up the great work and I’ll be listening intently for your upcoming podcast.
Slau
Thanks buddy. Great clarifications.
Yep I was going to point out the same thing. Most simply, a 16 bit system has about 65,000 notches, while a 24 bit system has 16,000,000. So it’s a *much* more sensitive & detailed “ruler.”
(the math: 2 to the 16th power is 65,536; 2 to the 24th power is 16,777,216)
~peter in oakland
(from memphis originally, by the way — i see you’re in smyrna, joe)
If 16-bit is good and 24-bit is better, then why are 1-bit recorders such as the Korg MR-1000 considered superior?
Good question. 1-bit recorders operate completely differently from 16- and 24-bit. 16 and 24-bit recorders use PCM (pulse-code modulation). 1-bit recorders use DSD (direct-stream digital), which operates at a super-high sample rate (like 5.4 MHz). Rather than measuring each sample on a “ruler,” these measure the first sample, then it simply asks itself “Is the next sample louder or softer than the previous?” Those are the only two options, so a 1-bit system allows you to measure each sample relative to the previous one.
With such a huge sample rate, you end up with ENORMOUS files. Computers/hard drives couldn’t stream such massive files in a multi-track setting, which is mostly why it hasn’t made its way into the major DAW platforms.
All that to say, 1-bit and 24-bit converters are completely different animals.
I don’t know much about it, but I was reading a bit (haha) on wikipedia the other day:
http://en.wikipedia.org/wiki/Direct_Stream_Digital
It’s an interesting idea, and maybe it does sound better, but nobody really has the playback gear for it, so your music would almost certainly get dumped back to PCM formats anyway (WAV, MP3).
Thanx Joe for sharing your knowledge.
Well explained for beginners and intermediate users.
Since I’m recording at 24-bit, I noticed there’s more room for further editing purposes.
On the other hand, I know that the default sample rate/bit depth for an audio CD is 44.1/16. So the question is when I bounce to disk/record to disk , what happens exactly?
Is there any lose of frequesncies after coverting to Wave ACD format? In other words, what’s the difference between 16 and 24 when making an audio cd out of protools session.
Thanx again for helping beginners to learn more.
Hey Al, I’ll cover this in a future post. When converting down from 24-bit to 16-bit, you need to apply dither. Essentially, you’re lopping off 8 bits, but you get the benefits of recording the audio and mixing down at 24-bit.
^^
I look forward to this post. That conversion and dither have always sort of baffled me.
I stumbled upon your website late last night, and I’ll tell ya…I wish I would’ve found it a year ago. The wealth of knowledge and suggestions here is astonishing. Thanks Joe and everyone that contributes with comments.
Thanks Cush! I’m glad you found the site! Welcome.
24BFL!!! That’s ‘24-bit for life’ — i’m thinking of getting a tattoo:)
When I switched from a 16-bit interface to a 24-bit interface, I definitely noticed a significant difference. Not so much in just listening back to a single recorded track, but in how multiple tracks came together in a mix. The noise floor in each track seems to get added together, and you end up with something pretty noticeable.
Not sure if ProTools has a similar plug-in, but in Logic there’s the “Bitcrusher”, that effectively reduces your bit rate. I think it sounds pretty awful as an effect, but it’s an educational tool on what bit depth means, as you listen to a very clean sounding track get noisier and noisier.
Ha. Removing bits is a fun way to screw up a sound. Great for sound design. Awful for everything else.