Topic: 44.1 or 48 khz for microtrack???

[digitalcandy926]

which should I use and what is the difference?  thanks in advance

[KLowe]

what are you using this music for?

if to be synched with a video then 48.

if for personal enjoyment then 44.1 (you will NEVER hear a difference b/t 44 and 48)....and 48's just make for larger files and more work in post

It is more important to record in 24 bit so you can set the levels conservatively without having to worry about clipping.

-one mans opinion

[prof_peabody]

what are you using this music for?

if to be synched with a video then 48.

if for personal enjoyment then 44.1 (you will NEVER hear a difference b/t 44 and 48)....and 48's just make for larger files and more work in post

It is more important to record in 24 bit so you can set the levels conservatively without having to worry about clipping.

-one mans opinion

If you are going to record in 24 bit you are going to need to dither the recording down to 16 bit.  Since you are forced to dither you may as well record at 48 khz - this will give you a more detailed representation of the waveform.  If you are doing bit truncation to go from 24 bit to 16 bit you may as well be recording in 16 bit because of the artifacts that are introduced by truncation.

some info on sample rates can be found here:

http://en.wikipedia.org/wiki/Sampling_rate

[DSatz]

KLowe asked the right first question: What's the purpose of the recording? In general he laid the main issues out very well, and I recommend his reply.

> If you are going to record in 24 bit you are going to need to dither the recording down to 16 bit.  Since you are forced to dither you may as well record at 48 khz - this will give you a more detailed representation of the waveform.

That assumes a few things which we don't know to be true, e.g. that the MicroTrack recorder has no other difference in performance at the two sampling rates other than bandwidth (if that is indeed different, which we also don't know). And the general claim that higher sampling rates give a "more detailed" representation of an audio waveform is based on an incorrect understanding of how digital audio works.

A system with a higher sampling rate can handle higher audio frequencies without aliasing, but it won't provide any "more detail" about those audio frequencies which it can handle, i.e. those which are less than half the sampling rate. If you want more low-level detail, maybe you can improve your signal-to-noise ratio--perhaps by using more bits per sample if the dynamic range of the A/D channel was the limiting factor. But the sampling rate has no effect on that at all.

The commonly-used visual abstractions of the sampling process may be at fault here, since they don't account for what happens when the samples are converted back into analog signals. If all you ever consider is that collection of stored digital samples, it will seem "intuitively obvious" that having more of them--closer together to each other in time--would convey more detail. Indeed it would seem as if only a fool could miss that fact. But that "intuitively obvious" conclusion is exactly what is mistaken; it takes the sampling process out of context.

When there's an absolute limit on the highest frequencies that can be handled (as in any linear PCM system), and an absolute limit on the highest levels that can be handled (as in any linear PCM system), there is correspondingly an absolute limit on how rapidly a signal can change within one sampling interval. By mathematical definition, when there is no significant signal energy at or above one-half the sampling frequency, then the sampling intervals are close enough to each other that the system can already handle the biggest, fastest changes that are mathematically possible within its amplitude and bandwidth limits.

Putting the sampling points closer together just doesn't make the system any better at doing that. Once the ability is sufficient, it is sufficient. Having $300 in your bank account is enough to pay a $250 water bill if that is the only expense to be considered. Having $600 in the bank is nicer than having only $300, but objectively it's not any better at paying a $250 water bill (again, if that's the only expense to be considered). As far as the water company is concerned, a $250 check drawn from an account with exactly $250 in it is just fine--though I hope that your bank isn't like mine is about such matters ...

--best regards

[rowjimmytour]

KLowe asked the right first question: What's the purpose of the recording? In general he laid the main issues out very well, and I recommend his reply.

> If you are going to record in 24 bit you are going to need to dither the recording down to 16 bit.  Since you are forced to dither you may as well record at 48 khz - this will give you a more detailed representation of the waveform.

That assumes a few things which we don't know to be true, e.g. that the MicroTrack recorder has no other difference in performance at the two sampling rates other than bandwidth (if that is indeed different, which we also don't know). And the general claim that higher sampling rates give a "more detailed" representation of an audio waveform is based on an incorrect understanding of how digital audio works.

A system with a higher sampling rate can handle higher audio frequencies without aliasing, but it won't provide any "more detail" about those audio frequencies which it can handle, i.e. those which are less than half the sampling rate. If you want more low-level detail, maybe you can improve your signal-to-noise ratio--perhaps by using more bits per sample if the dynamic range of the A/D channel was the limiting factor. But the sampling rate has no effect on that at all.

The commonly-used visual abstractions of the sampling process may be at fault here, since they don't account for what happens when the samples are converted back into analog signals. If all you ever consider is that collection of stored digital samples, it will seem "intuitively obvious" that having more of them--closer together to each other in time--would convey more detail. Indeed it would seem as if only a fool could miss that fact. But that "intuitively obvious" conclusion is exactly what is mistaken; it takes the sampling process out of context.

When there's an absolute limit on the highest frequencies that can be handled (as in any linear PCM system), and an absolute limit on the highest levels that can be handled (as in any linear PCM system), there is correspondingly an absolute limit on how rapidly a signal can change within one sampling interval. By mathematical definition, when there is no significant signal energy at or above one-half the sampling frequency, then the sampling intervals are close enough to each other that the system can already handle the biggest, fastest changes that are mathematically possible within its amplitude and bandwidth limits.

Putting the sampling points closer together just doesn't make the system any better at doing that. Once the ability is sufficient, it is sufficient. Having $300 in your bank account is enough to pay a $250 water bill if that is the only expense to be considered. Having $600 in the bank is nicer than having only $300, but objectively it's not any better at paying a $250 water bill (again, if that's the only expense to be considered). As far as the water company is concerned, a $250 check drawn from an account with exactly $250 in it is just fine--though I hope that your bank isn't like mine is about such matters ...

--best regards

Dsatz thanks once again for the great post and explainging the most complex stuff in the simplest way. Recently I switched from recording 24/48 to 24/44.1 thinking in the same frame that you posted. Now if I was to shoot video or do audio for some one who was shooting then I would go back but for now I am going to continue 24/44.1.
Peace

[prof_peabody]

KLowe asked the right first question: What's the purpose of the recording? In general he laid the main issues out very well, and I recommend his reply.

> If you are going to record in 24 bit you are going to need to dither the recording down to 16 bit.  Since you are forced to dither you may as well record at 48 khz - this will give you a more detailed representation of the waveform.

That assumes a few things which we don't know to be true, e.g. that the MicroTrack recorder has no other difference in performance at the two sampling rates other than bandwidth (if that is indeed different, which we also don't know). And the general claim that higher sampling rates give a "more detailed" representation of an audio waveform is based on an incorrect understanding of how digital audio works.

A system with a higher sampling rate can handle higher audio frequencies without aliasing, but it won't provide any "more detail" about those audio frequencies which it can handle, i.e. those which are less than half the sampling rate. If you want more low-level detail, maybe you can improve your signal-to-noise ratio--perhaps by using more bits per sample if the dynamic range of the A/D channel was the limiting factor. But the sampling rate has no effect on that at all.

The commonly-used visual abstractions of the sampling process may be at fault here, since they don't account for what happens when the samples are converted back into analog signals. If all you ever consider is that collection of stored digital samples, it will seem "intuitively obvious" that having more of them--closer together to each other in time--would convey more detail. Indeed it would seem as if only a fool could miss that fact. But that "intuitively obvious" conclusion is exactly what is mistaken; it takes the sampling process out of context.

When there's an absolute limit on the highest frequencies that can be handled (as in any linear PCM system), and an absolute limit on the highest levels that can be handled (as in any linear PCM system), there is correspondingly an absolute limit on how rapidly a signal can change within one sampling interval. By mathematical definition, when there is no significant signal energy at or above one-half the sampling frequency, then the sampling intervals are close enough to each other that the system can already handle the biggest, fastest changes that are mathematically possible within its amplitude and bandwidth limits.

Putting the sampling points closer together just doesn't make the system any better at doing that. Once the ability is sufficient, it is sufficient. Having $300 in your bank account is enough to pay a $250 water bill if that is the only expense to be considered. Having $600 in the bank is nicer than having only $300, but objectively it's not any better at paying a $250 water bill (again, if that's the only expense to be considered). As far as the water company is concerned, a $250 check drawn from an account with exactly $250 in it is just fine--though I hope that your bank isn't like mine is about such matters ...

--best regards

DSatz - you misinterpreted part of what I wrote.  Take a 5000 Hz wave - record it at 44.1 and 48 khz.  The 48 khz sampling will have a more detailed representation of the waveform (given all else is equal with the ADC).  This means your DAC, on playback, does less interpolation.  It's better the record at 48 khz, if you record at 24 bit and plan to dither. 

Basically, it DOES provide more detail because you will have more samples defining the waveform at 48 khz - this is a very basic concept is digital signal processing.  I can point you to some books and references if you need more information. 

[Ozpeter]

I think the Prof is pulling our legs...  c'mon, I think you should revisit digital audio basic theory.

[DSatz]

Ozpeter, prof_peabody and I have exchanged a few private messages over the past day or so. He is very courteous and sincere in wanting to help me grasp what seems so obvious to him.

From his message posted above, it's clear that he thinks of the sampled waveform as if the analog signal's value was defined only at the discrete moments in time for which a sample has been stored. From that premise, his conclusion is indeed quite obvious (more samples = greater accuracy), and quite a few people share it, including some prominent audio professionals. The premise is mistaken since the signal is still defined at all points even though it is being represented in the short term by discrete-time samples. But let's take the belief seriously and see what predictive value it offers.

THD+N is a straightforward indication of the difference between input and output, stated as a percentage of the input signal's amplitude. When the output of an audio recorder fails to deliver whatever was presented to its input, that difference can be characterized as either noise or distortion; THD+N measurements aren't even concerned with which one is which; any difference for any reason is counted equally.

If prof_peabody's belief is correct, the THD+N of (say) a 1 kHz sinusoid recorded at (let's say) -10 dBFS on a 96 kHz, 24-bit digital recorder should be less than half the THD+N of the same signal as recorded on a 44.1 kHz, 24-bit digital recorder. Yet the THD+N of a digital recorder actually turns out to be a function of its bit depth, not its sampling rate. If you believe as he does, this outcome must be baffling. A 96 kHz recording is a "more accurate" rendering of the original analog waveform, so why does the analog output of the system not resemble its input any more closely than in a comparable 44.1 kHz recording? Why is this supposed "greater accuracy" not measurable, nor audible, nor visible on an oscilloscope, nor discernable by any other objective means?

I think that many people are fooled by their visualizations of the digital samples, and by failing to understand what happens when the discrete-time signal is converted back to a continuous, analog signal. I'd like to point out that to prof_peabody's credit, he clearly recognizes how important that is--many people do not, unfortunately.

Once the sampling rate is already greater than twice the highest frequency of interest, a "more accurate" recording of an analog signal cannot be achieved simply by increasing the sampling rate--a statement which proceeds directly from Shannon's sampling theorem (60 years old this year). Shannon's work isn't intuitively obvious and it's not visually convincing, either, without some math. Its only virtue, really, is that it explains, correctly predicts, and allows people to design linear, discrete-time systems such as digital audio recorders.

--best regards

[Ozpeter]

Thanks for that.  I confess to feeling guilty that my response to him above was a bit blunt.   But the net is groaning with misunderstandings about the fundamentals of digital audio, and I can't help thinking "here we go again" if what appears to be urban myths about digital audio get perpetuated.

I've found that one of the easiest ways of visualising how the samples get turned into waveforms is to load up a file in Adobe Audition (or Cool Edit) and zoom right in to the point where you can pull the samples around with the mouse.   You can clearly see how the samples get interpreted.  Other such programs would doubtless show the same.  (Handy for demonstrating the phenomenon of "intersample peaks" where the samples are below full scale, but the resulting waveform goes over the top).

The following is a pdf of a discussion on Harmony Central which concerns some of the sample rate and bit depth issues which often arise.  Actually there were two threads there that I downloaded, hence two links.  It's long and contains some, er, robust discussion, but anyone wanting to get their head round this stuff might find it illuminating - I did!  And apologies if I've posted it already elsewhere on this forum - I know I posted it somewhere...

http://www.fileden.com/files/2007/9/22/1451533/Bits%20and%20samples%201.pdf

http://www.fileden.com/files/2007/9/22/1451533/Bits%20and%20samples%202.pdf