Advice on Bit Depth for recording

Audio doesn't work like graphics because of the way it gets decoded and turned back into an analog signal.

If your audio has (for example) 60db of dynamic range between its peak at its noise floor. It will be represented by the same number of bits (10), no matter where within the digital dynamic range it is recorded. At least as long as the analog noise floor is kept above the digital noise floor.

In that example, if that audio peaked at 0dbfs, the recorded noise floor would be at -60dbfs. If it peaked at -20dbfs, the noise floor would be at -80dbfs. In other words the bit depth does not add resolution that the original signal doesn't have to begin with.

With 16 bit, you only have 96db of dynamic range to deal with, which it pretty close to what you can get in the analog world. So that makes you have to try to maximize the range, so that the analog noise floor doesn't drop below the digital noise floor.

With 24 bit, you would be really hard pressed to get within 30db of the digital noise floor.
 
Well, I think he has two recorders - the Roland and the computer. By habit, I'd try to find out if the Roland just has "colored" analog out, etc.., or, is the digital file actually "colored", or, is it just the A/D conversion... hahaha
Not unlike troubleshooting a circuit - oh it is a circuit.

Then everything in the chain would have to be as close as possible for a fair comparison, from mic to monitors, gain staging and the position of the monitors in the room. Eliminating all that would seem to point to preamps or converters.

Might be interesting if the Roland tracks could be loaded into the computer for a side by side comparison.
 
Ya, comparing two, or more, analog outs is the usual easy first step. There would be that collection of test files that can be slung around every which way. I just set-up a monitor to test the MIC section. There are any number of stand-alone devices that can deliver reference material over analog and digital
 
As far as I'm concerned, photography is similar to the audio bit. You have 5 megapixels, 10, 12, 15...(whatever we're up to now), and then you have the large format camera that takes a much larger picture than your basic consumer camera. The more depth you have in a picture, the more quality you have. It's easier to make a large copy, without losing quality. Sure, you end up cutting it down again, when you print the final photo, because it's eventually cut down to an 8X10 or even smaller. Only when you keep it in the larger format and depth, do you keep that quality.

Recording is the same way. You have more depth with the 24 bit, but when you finalize it down to 16 bit for CD quality, you will probably get by, just like a snap shot will get you by, if you don't have your large format camera handy. If you are going to do a fair amount of bouncing of tracks, 24 bit might be the best way to go. In a pinch, 16 bit will be fine.
 
Once again, digital photography or graphics is not analogous to digital audio.

With graphics, it actually is resolution (how many variations in colors can be represented). With audio, it's dynamic range, you don't lose any gradations when you go from 24 bit to 16 bit. It simply moves the noise floor up, getting rid of any signal below -96db.

Sample rate could possibly be seen as resolution, but that isn't really accurate either. Sample rate only sets the highest frequency that can be recorded. As long as the sample rate is twice the frequency of the highest audio frequency in the audio, it will be decoded as a smooth analog wave.

With digital graphics, you have pixels. There is nothing in digital audio like that. That is the main reason that they don't work the same way and are really bad analogies for each other.
 
The word length can affect resolution, as can the rate. It doesn't mean that much of it is used.

With regular still photography, it's a static set of data. Beyond that, you can have a sampling rate for stuff that not static.

What pays off on 16x20 paper for me is the front end resolution at the lens(less so with the sensor or film) and finding a dither that puts that lonely dot right where I want it on the paper. If I'm thinking I'm going to waste paper and ink on a image.

Anyway, a typical color sensor is 12-bit times three. Or, 36-bit for my B&W. My sensors may be 10, 12, 14-bit ?? they all do white, but they can resolve the off-white better as the bit-depth goes up .

With the sensor conversion in camera, I'm JPEG, TIFF, and RAW. i almost always shoot JPEG and in color - 8-bit channels (RGB) x three for 24-bit. One could look at a Stereo recording about the same way with regards to soundstage made up of channels.

So, I'm fine with lower resolution in stills and in audio depending on the front end to be up to snuff. More is better, but I think one should define their needs. If I'm spending days transforming a single sound or image, I will want a larger data set.
 
With images more bits means more steps between darkest and lightest. The number of bits changes but dynamic range represented is always the same. With audio more bits means more dynamic range, each bit always representing the same difference in volume. There's a big difference between audio and graphics.

Audio vs. graphics.jpg
 
Also, with graphics, the display has a finite amount of pixels. With audio, it is converted into an analog waveform.
 
Well, the digital data has to be used to do different things. I can readily see better dynamic range in better film scanners. In either case, better dynamic range doesn't add up to anything if its not needed
 
bouldersoundguy said:
With images more bits means more steps between darkest and lightest. The number of bits changes but dynamic range represented is always the same. With audio more bits means more dynamic range, each bit always representing the same difference in volume. There's a big difference between audio and graphics.

With linear PCM audio, more bits means more steps between changes in volume. While it's true that each bit represents 6 dB dynamic range, at the level of the least significant bit you're essentially working off of 1 bit. Which gives you a range of values between 0 and 1 to quantize the power of the signal.

This will turn a sine wave into a square wave.

In 16 bit audio, if you hit a level of 0 dBfs then every bit in the word would hold a value of 1 for that sample. A 16 bit word length gives you 65,536 possible levels to quantize at. Bring the signal down to -6 and you're essentially using a 15 bit word length now. Each bit represents 6 dB so you can chop one off the top. Now you have 2^15 quantization levels or 32,768. That's half. Every time the signal drops 6 dB the resolution is cut in half. Very low signal levels will distort because of the dropoff in resolution. The distortion comes from rounding errors in the quantization process.

In 24 bit, you get an extra 50% dynamic range over 16, while the number of available quantization steps increases by a factor of 256. The theoretical range of 144 dB is beyond the mechanical limits of the rest of the signal chain, so it sort of filters out the low resolution end that causes a lot of the problems, and beefs up the working end that we're going to process in the DAW.
 
But if you took away the 16th bit and had a 15 bit word length, AND you used all 15 bits, the level would be 0dbfs. It would not be -6dbfs.

This is because more bits doesn't buy you more level, it buys you more space between , he loudest and the softest. It only buys you more possible levels because it adds them to the bottom of the dynamic range.

Man, I am not explaining this well...

Yes, more bits gives you more levels, but the number of levels is constant no matter what the level of the signal is.

In other words, if you have a signal with 60db of dynamic range, it will have the same resolution peaking at -20dbfs as it does peaking at -1dbfs.
 
But if you took away the 16th bit and had a 15 bit word length, AND you used all 15 bits, the level would be 0dbfs. It would not be -6dbfs.
No, the audio format is still 16 bits, you're just not using the last one because the signal doesn't have that much power.

In quantization, you're essentially taking a flawed approximation of an analog value. In the case of linear PCM, you're converting voltage (power) to a binary number (quantized approximation of power).

In Linear PCM, each discreet step in your word length corresponds to an equal slice of a volt. If you half the voltage, your signal will drop by approximately 6 dB. If you half the available quantization steps, your signal will drop by 6 dB. It's linear.

This is because more bits doesn't buy you more level, it buys you more space between the loudest and the softest.

More accurate space, yes.
 
Not more accurate space, just more space.

Dynamic range is different with graphics, it is more gradations (accuracy) between the least and the most. With audio it is more distance between the least and the most. The accuracy stays constant until you hit either limit.
 
I'm not going to pretend my little selections of digital stuff is going to match those early Telarc at 16/50, but those interested can snag reference files from 16/44 to DSD256 for the cost of download time;
http://www.2l.no/hires/

For the most part, recorded & processed on DXD PCM. For this style of recording, it might of been recorded on a 1-bit Sony, or, Tascam, though
 
I'm not going to pretend my little selections of digital stuff is going to match those early Telarc at 16/50, but those interested can snag reference files from 16/44 to DSD256 for the cost of download time;
2L High Resolution Music .:. free TEST BENCH

For the most part, recorded & processed on DXD PCM. For this style of recording, it might of been recorded on a 1-bit Sony, or, Tascam, though

WTF does this have to do with anything in this thread?
 
Just that 24-bit isn't any better than 1-bit for recording

No one was talking about DSD, which works on a completely different principal than PCM. We are discussing PCM bit depth.

Comparing DSD AND PCM to come to a conclusion about bit depth doesn't make any sense, because they are not related in any way. It's like comparing an apple and a tomato because they are both red. They aren't interchangeable.
 
As painfully slow, backwards, and Imperialist as Avid is, DSD is now part of the SPEC. Possibly, the poster only knows about how truly wonderful his sound card is, because it can do 24-bit : )

"Sampling, resolution 5-bit sigma/delta @ 5.645 or 6.144 MHz, 24-bit PCM
DSD sample rates 2.8224 and 5.6448 Mhz (64 and 128 fs) "

So, I don't mind making sure he hasn't lost track of the tech from the last decade. I have no plans or reason to record more than 24/96 (or to cassette for that matter). But I don't mind buying hi-res for music. DSD may be gone tomorrow ? Whatever.
 
Farview said:
In other words, if you have a signal with 60db of dynamic range, it will have the same resolution peaking at -20dbfs as it does peaking at -1dbfs.

So if you had those two signal in analog (-20dBu and -1dBu let's say) and you add 0.1 volts to both, would the -20 signal rise more in level or the -1 signal? Or would they both rise by the same amount?
 
I'm pretty sure the -20 would rise more. But that's the sort of thing that happens with logarithmic scales.
 
Back
Top