Topic: 24 bit > 16 bit

[boojum]

I am going back to my earlier statement that I am not sure whether Ohm's Law or the Nyquist Theorem is less understood and more discussed.  We can really hold forth on these two.  I just record at 24/48 and let it go at that.  Since I will soon have the option to do it in FLAC.  I suppose that 24/96 would use about the same amount of space as 24/48 in WAV.  But unless there is a strong technical reason to do so, I will probably stay with 24/48.    As usual, YMMV.     

[SparkE!]

Well, in theory with a 96 kHz samples per second recording you get 96/2 = 48000 Hz upper frequency limit, in practice with the pre AD brickwall filters about 45 kHz upper limit. But, like I tried to point out, as practically all microphones, ALL reproducers and ALL HUMANS cut off at around 20000 Hz at the latest, there is no use, point, need nor any sense trying to record something that does not even enter the recording chain. And if it enters, does not get out. And, if by some freak phenomenon, would get out, only bats would hear it.

There is no additional benefits to recording at 96 kHz sampling rate exept higher cut off limit. It does not reveal any "hidden detail" or "unveil" the sound. Just that also the frequences we can not hear can be recorded (if the mic were good enough, and it is not).

---------
Against stupidity and ignorance the Gods themselves contend in vain.

Dang it, Petrus! There is no such thing as a brickwall filter.  You're right that you want everything below 48 kHz and nothing above 48 kHz when recording at a 96 kHz sample rate, at least from a purely theoretical, mathematical point of view.  It's not possible in practice.  And people don't use a 45 kHz cutoff frequency either.  If you did that, you wouldn't be even close to being attenuated sufficiently at 48 kHz.  How many poles do you think people use in a typical anti-aliasing filter?  You only get about 6 dB per octave per pole as a rule of thumb.  Using that approximation, you'd only get about 1/2 dB per pole at 48 kHz with a 45 kHz cutoff frequency and that's just the Bode approximation.  In practice, you won't even get that much.  In order to get any appreciable attenuation at 48 kHz from your anti-aliasing filter, you have to use a much lower cutoff frequency.

Let me put this in perspective.  It's hard to get components whose tolerances will allow you to use well designed filters with more than about 8 poles in them when using typical opamp, resistor and capacitor-based filters.  With a 20 kHz cutoff frequency and a .1 dB Chebyshev design, I recall that you can only expect about 100 dB of attenuation clear out at 96 kHz.  At 48 kHz, I don't remember the exact numbers, but it's more like 55 dB of attenuation and you need to be happy with it because that's about the best you can do.

Now that means that you get about 55 dB of attenuation at fs/2 when you use fs=96kHz.  That helps a lot.  If you use the same filter, but a fs of 48 kHz, you only get a few dB of attenuation at fs/2.  That's the true benefit of using the higher sampling frequency.

On the other hand, I still agree with your assertion that you don't gain much by using the higher sample rate and you damn sure use twice the memory to record the same signal when you double the sampling rate.  What it comes down to is our ability to perceive the improvement in recording quality by using the higher sample rate.  Most people can't tell the difference.

I'm glad that you're on here preaching the virtues of recording at reasonable sample rates, but please don't justify your opinions with technical misinformation.  Seriously, I appreciate that you're preaching common sense recording practices.  Just please, be more careful to use real facts to justify your advice.

[echo1434]

Just my view...

Regardless of noise floors or whatnot, 24-bit recording is closer to real life. As was already discussed here, 24-bit allows for over 16 million amplitude levels, while 16-bit only allows 65,536 (a 256-fold difference).

Well, real life (as well as analog) has infinite amplitude possibilities. So of course 16 million is a lot closer to infinity than 65,000. This helps solves one of the problems of how high quality analog (reel-to-reel, vinyl) can still be superior to digital formats.

That, and the extra robustness for editing, I think it's an easy choice. Whether one can truly hear the difference is obviously going to stir a big debate, and it also depends on the quality of your stereo system, as well as your ears.

Anyway, I'm sure higher bit depths/sample rates are going to become commonplace in the future, so why not take a step in that direction now and not regret it later?

[boojum]

This is a real hijack, so you might want to just skip it.

Just my view...
<snip>

 So of course 16 million is a lot closer to infinity than 65,000.

<snip>

I am not sure, but I think they are equally close if I have any understanding of the Law of Large Numbers.  Infinity is so beyond any concept we might have that the difference between 65K and 16M is insignificant.  I think of the three infinities I know of, Aleph sub 1, Aleph sub 2 and Aleph sub 3 (yes, three infinities) they are pretty big: Alpeh sub 1 is the sum of all the points on a plane.  A point has no dimension.  Aleph sub 2 is the sum of all the odd and even numbers and Aleph sub 3 is the sum of all the possible curves in the world.  If there is a mathematician around, please help me here.  I studied US History so I am way beyond my depth.  But 65K and 16M are just about equal by the terms of infinity.  As counterintuitive as it sound.

Theoretical math is freaky.  The square root of four is plus or minus two.  The fourth root of sixteen is plus two, minus two, and two twos whose signs we do not yet know.  Sheesh.  See why I studied history???

L8R

[chunga1]

 

[echo1434]

Ok, maybe the way I used the term "infinity" was a bit imprecise. Perhaps I should speak in terms about how "what can be perceived" relates to infinity. I read somewhere that approximately 20-bit is highest bit depth that a human can differentiate, but I don't know the evidence that claim was founded upon.

But at least we can all agree that 16-bit audio is more or less "pretty good". So if you expanded one of its properties by 256-fold (by producing 24-bit audio), that would be a marked technical improvement, no?

[Petrus]

SparkE, I am just trying to get these things to laymans levels, a perfect brickwall filter does not exist, sad but true. If we need perfect (unheardable) filtering of sub-samplerate frequences, OK, use 48 kHz and not 44.1 kHz, that keeps the minuscule time domain problems at around 20 kHz at bay. 96 kHz is an overkill. But the final products are often 44.1 kHz anyway (CDs...), so what is the point.

This "when a note plays" stuff: If the system plays perfectly the audible frequency range, also the timing information is perfect as far as we can hear it. This talk is about "timing" and "notes" not being in full sync is total BS inveted by golden ear belivers who lost their hobby when turntable tweaking went out of fashion. It has no scientific base and no test evidence to prove it.

About time coherence: If there were some timing problems in these systems, they would have to in the order or 1/50000 sec. (why else would people demand sample rates of 96 kHz and above?). Half wavelength at that frequency is 3.3 mm. By shifting your head by 1/8 of an inch while listening would throw the image out of whack, or what? We all know perfectly well it does not happen. Hearing is not that presice, audible range from about 16 to 20000 hertz contains all the information we humans can use, time, amplitude, transients, everything. There is nothing out there.

All this has nothing to do with bit depths. As previously said, 24 is convenient compared to 16 when recording and editing, for the final product 16 bits is plenty enough. Just to remember the original question.

Echo1434: you have one major flaw in your thinking of analog versus digital. Even 16 bit system has infinite values for the final output; lowpass filtering after the D/A conversion smooths out the waveform, there are no 65000 steps there in the sound you listen. And besides, in 24 bit systems those imaginary 16 million steps do not replace the 65000, they reside outside of that first 65000 step area, because adding 8 more bits gives just more dynamic range. Not "resolution", 12, 14, 16 bits define the waveform perfectly and steplessly within their dynamic range windows. Adding more bits adds dynamic range, it gives only more resolution to the most quiet sounds, loud sounds are already perfecly taken care off. Graphic representations of digitized waveforms give a wrong idea about the workings of the system.

[SparkE!]

SparkE, I am just trying to get these things to laymans levels, a perfect brickwall filter does not exist, sad but true.

OK, sorry about that.  I just saw brickwall filters used in your description in a way that indicated that they were used in practice.  Then you mentioned a compromise that was starting your filter at 45 kHz instead of trying to get it all right at 48 kHz.  At that point my BS detector went off and I pulled the handle on the whole house sprinkler system.

But seriously, the main thing that higher sampling rates does for you is to help ensure that out-of-band signals stay out-of-band.  A properly designed recorder whose sample rate is 44.1 kHz will convert more of a 34.1 kHz signal to a 10 kHz alias signal than a properly designed recorder whose sample rate is 96 kHz will.  As long as there are no ultrasonic signals present, you don't have to worry, but air handling systems and other mechanical systems can have frequency components well into the ultrasonic range.  So can the output of the PA system where you are recording if someone installs ultrawide tweeters as part of the PA system.  Granted, this is not the most common situation, but when it exists, you will get a real and noticeable improvement by using a higher sampling rate, then downconvert to a lower sample rate for storage on media intended for use with playback systems.

[gratefulphish]

As a result of this thread, I have now started recording at 24/48 vs. 24/96.  It makes life a lot easier, having each set fit into a 2Gb file, instead of putting the parts together, working on it, splitting it up.  Converting, putting back together, etc.   

But as I was thinking about it, it raised another question.  If we are recording in 44.1, then as I understand it, the samples that we are recording are essentially all within the audible range, except for maybe a few at the very high end.  When we go to 96, from all of the prior responses, it seems that all of the extra samples are in the area above the audible range.  So then the question arose about sample rate conversion.  What exactly happens when we convert from 96K to 44.1?  Is it just removing all of the samples that never would have been there if the original recording was done in 44.1, or is it actually selecting from some all of the 96K samples, including some of which may be in the beyond auditory realm.  If so, then it would seem to me that we might actually be losing useful samples, compared to just recording at 44.1 to begin with.  I hope the question is at least somewhat clear.  It just seems that we could potentially be losing some of data from the audible range, if a significant portion of the converted samples from a 96K recording are outside that range.

[Petrus]

Downconverting from 96 to 44.1 kHz sample rate we basically arrive to the same situation we would have had if we had recorded in 44.1 kHz in the first place. Theoretically there might be some loss (rounding errors) compared to an original 44.1 kHz file, what is certain is that the quality can not be better than an original 44.1 file. So there is nothing to gain from recording with oversampling and downconversion. The only exeption could be heavy editing/slowing down etc. Using 24 is often helpfull, even when downconverting to 16 bits.

Downconverting from 96 to 48 basically we just throw away every other sample, no rounding errors there, end result should be identical to an original 48 kHz file.

[echo1434]

So there is nothing to gain from recording with oversampling and downconversion. The only exeption could be heavy editing/slowing down etc.

I'd say I pretty much agree with this, but this is still kind of assuming that the end product will be CD. I don't really use CD anymore, and I'm guessing that format will eventually go by the wayside.

In the future, more audio players will support the higher sample rates, so that's what's keeping me recording at 48K, for whatever tiny benefit it gives.

[SparkE!]

Downconverting from 96 to 44.1 kHz sample rate we basically arrive to the same situation we would have had if we had recorded in 44.1 kHz in the first place. Theoretically there might be some loss (rounding errors) compared to an original 44.1 kHz file, what is certain is that the quality can not be better than an original 44.1 file. So there is nothing to gain from recording with oversampling and downconversion. The only exeption could be heavy editing/slowing down etc. Using 24 is often helpfull, even when downconverting to 16 bits.

Downconverting from 96 to 48 basically we just throw away every other sample, no rounding errors there, end result should be identical to an original 48 kHz file.

Actually, you do gain from using the higher sample rate.  The gain comes in the form of lower levels of ultrasonic frequencies that are unintentionally aliased into the audible frequency spectrum.  I'll agree that there is nothing to gain if you don't have ultrasonics present in the recording environment though.