Basic question about 16 vs 24 bit

[Jesse Skeens]

Ok i should know this stuff for now but i just want to make sure I'm on the right track.

Got a MidiMan Flying Cow 24bit converter today. Was checking out the bit depth meter inWavelab when I noticed my noise floor with master fader on the console down, was right around -96db or 16bit. Raise the fader on the board and its more like -60db (analog effects and stuff making all that noise)

Now my question is, are the 16 bits from 0 down to -96 on a 24 bit converter te same as a 16 one? Meaning that until I get below -96 reocrding at 16 won't be any worse than 24?

I always heard you could record on 24bit at a lower level since you had more bits to play with. But since the noise floor is so high on my system would it really matter anyways? Seems like those extra 24bits are hiden anyways. Unless I made a digital fade in Wavelab I never see the bit meter go below 16 when using a 24bit file. So it makes me wonder what the deal with 24bit is. I always thought it would give me more resolution in the upper region of my signal, like it owuld bemore detailed, but now that I think I understand it better thats not the case>

Jesse

[Professor]

OK, first off that isn't a 'bit meter' it is a VU meter that is giving you a representation of the audio signal strength in deciBels. The theoretical 96 dB of dynamic range in a 16 bit system is based on the total numeric range of the 16 measurement. If the digital system itself generates no noise then it has a 96dB range, but that is never the case and functionally a 16bit system has 85-90dB of dynamic range. The theoretical yield for 24bit is 144dB of range but functionally the best systems are only hitting around 124dB of range. That range is only limited by where the noise floor of the digital system begins to creep in.

In a 24 bit system you still want to record a strong signal to disc, and although you can turn up the signal more in 24 bits, that noise floor is still down there and will come up with the rest of the track.

A lot of people talk of the 'extra bits' in 24 bit or say they are 'not using the last 2-4 bits anyway' when they record at a lower level but that simply isn't true. The difference is spread evenly across the entire audio signal because each additional bit increase the resolution of the entire signal, not the headroom of the track. It is literally like trying to use graph paper with squares that are 4 inches across, then switching to 2" squares, then 1" then 1/2" then 1/4" then 1/8" then 1/16" then 1/32" then 1/64". If you are trying to draw a wave on the big 16bit, 4" squares it is hard to graph (and you can only place your dots in the corners of the boxes). With the 24bit, 1/64 squares you draw a much more detailed picture of the audio. Practically speaking this does not become a noticeable difference in the volume capabilities of the system but in the perception of depth in the recording. The softs are just as soft and the louds are just as loud, but the soft sounds hiding out behind the loud stuff are sudeenly noticeable, and the space around the instrument is bigger. When switching from 24bit to 16bit a mix will typically sound as though it has been flattened out from front to back and kind of squashed into a flat picture rather than a 3D image. Obviously this is easier to hear on some types of recordings than others.

So while the difference between 16bit and 24bit is quite tremendous sonically it is not so obvious as a volume difference and it is certainly not something that would get metered while recording. I've never heard of a recorder that measured the bits being used per sample on a visual meter - because every sample uses all 24bits and they are flying by at least 44,100 times per second. If any system did meter the 'bit rate' it wouldn't ba able to display it since your monitor only refreshes 60-120 times per second.

Hopefully the difference between 16 and 24 will be audible to you in recording and mixing. If not, then just keep recording in 24bit because eventually you will hear it. Even if you can never hear it, maybe someone else will hear it and think you are a better engineer for it.

One last great explanation I remember from the Panasonic boys explaining the 32bit internal processing of the DA7 console is: If you add 4+4 the answer is 8 and you can represent all the numbers as single digits, but if you multiply 4x4 the product is 16 and you need another digit to represent the number. When you are altering things in the computer, how many digits to you need to multiply the big six digit numbers?

Either way, stick with 24 bit, you'll be glad you did.

-Jeremy

[Jesse Skeens]

thanks for the reply. when i said wavelab has a bit meter though i meant it, checkit out sometime. it tells you how many bits are being passed at any given time. you can use it to check to make sure your sending device is pushing 24 or 16 bits. for instance if you are only passing 16 then the bottom blocks dont ever light.

what doesnt make sense though is when you say 24 bits has smaller squares. i thought each bit was 6db so how can 6db be smaller than 6db since in both 16 and 24 bit its the same value for what each bit stands for?

my understnading is that the only difference with 16 and 24 is that when you get to lower signal levels there are more bits to keep going down to. where as in 16bit once you get below -96db then everything is quantized at the last bit. where as 24 lets you keep doing down as it has more bits to spare.

jesse

[helstab]

professor:

Thank you for that explination. I UNDERSTAND! wow thanks a lot. GREAT explination!

[MoreSpaceEcho]

what doesnt make sense though is when you say 24 bits has smaller squares. i thought each bit was 6db so how can 6db be smaller than 6db since in both 16 and 24 bit its the same value for what each bit stands for?

my understnading is that the only difference with 16 and 24 is that when you get to lower signal levels there are more bits to keep going down to. where as in 16bit once you get below -96db then everything is quantized at the last bit. where as 24 lets you keep doing down as it has more bits to spare.

jesse

hey jesse. i'm trying to figure this stuff out too. in 16 bit, there's 65,536 quantization steps. every bit you add doubles the number of steps, so 17 bt would be 131,072, etc etc, 24 bit is ~16,777,000 steps. my understanding is that the top 16 bits in a 24 bit system are the same as a regular 16 bit system, and all the extra quantization steps are added for bits 17-24. Which is what i'm not getting...why have 8 million quantization steps between -138 and -144 when we only have 65 thousand between 0 and -96? maybe i have this all wrong, i dunno. i asked this question on brad's forum at PSW yesterday, if i get a reply i'll report back here.

cheers,
-scott

[mrhawkinson]

what doesnt make sense though is when you say 24 bits has smaller squares. i thought each bit was 6db so how can 6db be smaller than 6db since in both 16 and 24 bit its the same value for what each bit stands for?

The '1 Bit = 6 dB' rule refers to quantization noise, not the number of quantization levels in a signal. What that means is, the noise caused by chopping an analog signal into 'stair steps' will be lower at higher word lengths.

Quantization LEVELS, the number of 'stair steps' available, are spread evenly so at higher word lengths loud signals are more detailed just like quiet ones.

my understnading is that the only difference with 16 and 24 is that when you get to lower signal levels there are more bits to keep going down to. where as in 16bit once you get below -96db then everything is quantized at the last bit. where as 24 lets you keep doing down as it has more bits to spare.

It is very helpful to learn to think of the data itself in terms of % of full scale rather than dB. There is a huge collision going on in your thinking between signal loudness, quantization level, and quantization noise.

LOUDNESS
In the digital world, loudness is entirely a function of how close a signal comes to full scale. A snare hit that peaks at 95% of full scale will be the same loudness whether it's quantized to 16 or 24 bits.

QUANTIZATION LEVEL (WORD LENGTH)
Word length determines how many 'stair steps' are available between complete silence and full scale. The greater the word length, the greater the precision of the output compared with the input.

QUANTIZATION NOISE
Quantization noise is the result of the coarseness of the 'stair steps' in a digital waveform -- the greater the difference between the input and the output, the noisier the playback will sound. At lower word length, the errors are bigger. The magnitude of the error is the same at all levels, which means the error is more significant at lower levels. This is why the 'noise floor' is lower at higher word lengths.


The only place the dreaded 'dB' enters the scene is when someone creates an interface for a piece of software. The unit is used to describe gain changes and available headroom because it can be a handy way to think about apparent volume. But behind the curtain everything is happening in terms of percentage of full scale.

To tie this into the content of the original post on the thread: if you can afford to disk-space and system-performance wise, do all your recording at 24 bit. Leave yourself 2 or 3 dB of headroom, especially if you are going to mix the recordings with other tracks. Definitely do all your mixing and processing at 24 bit. Close that stupid, meaningless bit meter!

-p

[pscottm]

great info..

but none of it means shit if you're using a crap converter. remember, a good converter sounds better at 16/44.1 than a crap one at 24/96. i saw a dvd player at wall-mart (forgive me) w '24bit 96k fidelity' written on the front, and it was $49.95. ok, now tell me these numbers actually mean something.


edit: sorry, typed too fast, made no sense!

[joel hamilton]

I am 100% with scott on this one.

Poopy converters sound poopy even if they ARE filling up your hard drive with 96,000 24 bit words a second.

[Ethan Winer]

Jeremy,

That was a great explanation, but I need to correct one point. You said, "The difference is spread evenly across the entire audio signal ... like trying to use graph paper with squares that are 4 inches across, then switching to 2" squares <

In most cases the increased resolution is irrelevant, and the only benefit to using more than 16 bits is a lower noise floor.

The kind of resolution you're talking about directly relates to harmonic and intermodulation distortion. In a decent digital setup these are already so far below the distortion of your mikes and speakers as to not matter. Yes, more resolution can yield a cleaner sound, but only if the resolution is so bad that distortion is audible. So I'll agree the 16 bits is better than 8 bits for reasons of resolution, but not 24 versus 16.

--Ethan

[heylow]

Since most of the tech stuff is being covered here, I'd just like to add that Wavelab does IN DEED have a bit meter and it is INDEED useless as far as I can tell.

Full scale 0 is full scale zero (FS) no matter whether you're at 16 or 24 bits. Full scale zero (peaking on your digital meters) is like +22dB VU (on your analog gear) whereas -15dB FS is generally equal to 0dB VU (on your analog gear). That number can change depending on the studio and their personal calibration preferences. I believe many of the "big dudes" in Nashville stick to the -18FS dB = 0VU dB idea.....some go as far as -12FS to equal 0VU.

My simple take on the whole thing is that with 24 bits, there is not a need to "bit smash" as there is much greater resolution into the "noise floor". In fact, many people believe that converters have a sweet spot. I know that in my recordings, if I observe the general guidelines I laid out above, which would be (for me) -15FS = 0VU with my peaks coming up to maybe -6FS, I find I get WAY more open sounding recordings than if I jam every light on that meter. This could be either converters sweet spots or maybe even more probably, analog outboard sweetspots.

I hope that makes sense. I didnt mean to get that far into it....what I was trying to illustrate is, basicly, get to know your gear, get to know your sweet spots and know what you are trying to achieve and dont worry about silly distractions like "bit meters".


heylow

[Jesse Skeens]

hey jesse. i'm trying to figure this stuff out too. in 16 bit, there's 65,536 quantization steps. every bit you add doubles the number of steps, so 17 bt would be 131,072, etc etc, 24 bit is ~16,777,000 steps. my understanding is that the top 16 bits in a 24 bit system are the same as a regular 16 bit system, and all the extra quantization steps are added for bits 17-24. Which is what i'm not getting...why have 8 million quantization steps between -138 and -144 when we only have 65 thousand between 0 and -96? maybe i have this all wrong, i dunno. i asked this question on brad's forum at PSW yesterday, if i get a reply i'll report back here.

cheers,
-scott

that is what i was thinking too, why have all that extra resolution down below. let me know what you find.

[Jesse Skeens]

what doesnt make sense

Quantization LEVELS, the number of 'stair steps' available, are spread evenly so at higher word lengths loud signals are more detailed just like quiet ones.


-p

so are you saying there are more star steps from say -60db to full scale on a 24bit system compared to a 16bit one?

[mrhawkinson]

so are you saying there are more star steps from say -60db to full scale on a 24bit system compared to a 16bit one?

Yes, there are 256 times as many.

-p

[MoreSpaceEcho]

so are you saying there are more star steps from say -60db to full scale on a 24bit system compared to a 16bit one?

Yes, there are 256 times as many.

-p

Cite your sources!

I question you in the friendliest manner. I'm really curious about this.

thanks
scott

[mrhawkinson]

Cite your sources!

Here's a great 6-part article on the subject, from the archives of the fine UK recording mag Sound on Sound:

http://www.soundonsound.com/sos/may98/a ... gital.html

Parts 1 and 2 of the article relate really directly to what's discussed in this thread.

-p