32kHz vs. 48kHz - What's the difference
 

What's the difference between these 2 audio types. I think that 12 bit goes with 32kHz and 16bit goes with 48kHz, but I'd really just like to know what the difference is and if there is an advantage to recording in one format over the other. Any help is appreciated! Thanks.

Tim

48 Khz/16-bit audio has a higher fidelity (wider frequency response, up to a theoretical 24000 Hz, lower quantization noise
due to more bits) than 32 KHZ/12-bit audio. In practice, there are other limitations, such as the input circuit, which will probably make the two formats more equal. Also, if you record your audio in the 32 KHz mode, many units allow an overdub of two more stereo tracks, enabling you to store four tracks of audio. A few high end units allow similtaneous recording of all four tracks.

Sound is an analog pressure wave. To get it into the digital realm, we have to sample it. The 32 kHz vs 48 kHz (or 44.1 kHz on CDs) determine the number of samples per second. This determines the highest frequency that you can represent accurately when sampling. There is a fancy word, the Nyquist frequency that basically means you will be able to sample frequencies up to half of the sample frequency. So you are limited to 16 kHz high end on 32 kHz sampling and 24 kHz high end on 48 kHz sampling.

Now, this is the top frequency that those sampling frequencies can represent. Realistically, you want to leave a little wiggle room, as a 48 kHz sampling frequency would only use 2 samples to represent a sound at 24 kHz. If those samples aren't at the two peaks, there would be no sound. This is why recording studios sample at 96 kHz. This effectively gives you a better representation of the sound on the highend.

As far as bit width, this determines the accuracy of the sampling. A 12-bit sampling can represent 4096 discrete voltage levels. A 16-bit sampling represents 65536 discrete voltage levels.

I did a quick web search and here is a 4-page PDF file that wilk give you a better understanding of the factors in digital sound, include sampling frequency and aliasing:

http://birds.cornell.edu/brp/PDFs/AppA_DigitalSound.pdf

Practically speaking, you should nearly always use 16bit mode. Some consumer cams come out of the factory in 12bit mode. Perhaps the only case where 12bit mode is useful is when recording 4 channels on the XL1s with the MA100 adapter.

For music, 48khz sampling is advantageous over 44.1khz since it reproduces high frequencies better. You can check this yourself by saving a good music recording in 48khz and 32khz. For voice 48khz doesn't really help since the human voice doesn't have meaningful sounds that go that high frequency (too lazy to check this myself though, could be wrong). Jay Rose's article "home on the ranges" has some information on sound frequencies and I think some discussion of 48khz versus 96khz.

Wow, thanks for the informative answers. I'll think I'll just put my cam on 16 bit for now on just because it simply can't hurt anything at the very least. I know I won't be doing 4 channel recording

I used to sell computer-based digital oscilloscopes for Tektronix. From an engineering measurement standpoint, we always wanted a very high sample frequency.

Our rule of thumb was:

That the Nyquist Theorem states that you can accurately measure the frequency of a symmetrical sine wave of one-half the sampling frequency.

To measure more complex waves, you need to sample at at least 10X the highest expected frequency. Much more is better.

That's why the sample rates are getting up to near 200 Khz on pro equipment.

I haven't a clue why 48 kHz audio sounds so good given what I just wrote.

12-bit sounds pretty bad for music. The VX-1000 was a 12-bit 32Khz audio system IIRC.

>That the Nyquist Theorem states that you can accurately >measure the frequency of a symmetrical sine wave of one-half >the sampling frequency.
>To measure more complex waves, you need to sample at at >least 10X the highest expected frequency. Much more is better.

By more complex waves, you probably mean that the waves have partials which measure above the highest expected frequency. The Nyquist theorem still holds for the highest partial of interest in the wave.


>12-bit sounds pretty bad for music.

Twelve bit is not bad for music, provided the system actually delivers 12 bit performance. I've heard some music played back on a good 12-bit system (a Sequential Circuits Prophet 2000 sampling unit) and couldn't readily distinguish any difference
between its playback and that of a commercial CD player.
12 bit, 32 KHZ audio recorded and played on a good DAT machine sounds good.

Many systems are quoted for
the digital to analog converter resolution, even though the converter itself doesn't deliver performance anywhere near that.
Many "16-bit" codecs (dual channel A/D and A/D in
a single package) were sold for computer motherboards. The spec sheet also listed an 80 dB dynamic range. 80 dB ?!? Why,
that's only a 13-bit dynamic range. (~6 dB / bit) Above and beyond the limitations of the converters, the supporting circuitry is an issue.

How can that be? A waveform at 1/2 the sampling frequency is going to be sampled twice. In fact, to even measure the frequency at that ratio, one has to trust that the waveform is symmetrical. That is not enough to describe any complexity at all. And because the sampling is not synchronous, the problem is even worse if the sound gets anywhere near a square-wave as it can for distorted instruments.

I know it works reasonably well but my understanding still has a hole or two in it.

>To measure more complex waves, you need to sample at at >least 10X the highest expected frequency.

Mike,

I think this "rule of thumb" 10x may simply be a guideline to record complex waveforms at a *base* frequency. Since square waves have partials that are multiples of its base frequency, the sampling rate should be increased to accomodate the harmonics of interest. Distorted waves also have frequencies that are higher than the base frequency. When harmonics enter the picture, the new, "frequency of interest" is higher than than the base frequency of the waveform. Hence, the 10x guideline. I don't think you have any holes in your understanding.

There are other limitations, such as the quality of the anti-aliasing filter. Oversampling is used to reduce the complexity and increase the accuracy of the anti-aliasing filter.

Here's my guess:
A- Not everyone can hear the high frequencies (in the 20khz range) all that well.
B- Other sounds in the lower frequencies may mask higher frequencies sounds, so your brain doesn't actually hear the high frequencies.
C- Not all sounds have that many high frequencies. see jay rose's article "home on the ranges" over at dv.com.

>I haven't a clue why 48 kHz audio sounds so good given what I just wrote.

Mike,

Do you think that 48 KHZ audio sounds much better than 32 KHZ mode on the same system?

Personally, I don't think that my vx2000 or cheapie JVC camcorder audio sounds much different in 32 KHZ/16-bit mode. For a while, older 32 KHZ/12-bit systems had A/D converters that operated at noise floors a few bits above the 12-bit specification.
When you switch a machine built to handle 16-bits, 48 KHZ into
32 KHZ/12 bit mode, you will probably take full advantage of the
time and level resolution of that system.

I wasn't comparing 32 and 48 Khz sampling. Just commenting on the quality of sound resulting from 48 Khz sampling

Shoot, I cannot hear much difference between 22 Khz and 44 Khz Smartsound disks. I assume the differences are there.

Too many years in submarines/ working with machiner / pounding metal / shooting firearms, all without ear protection has killed the higher levels of sound for me.

However, I can't hear what I'm missing!! :-)))