Why record higher than 44.1 khz?

[superspit]

yaaaayyyyyyy....LOGIC PREVAILS, yet again!!!
SSG and MM farking rule!

 

[ofajen]

Thats what I tried to elicit from you
But in all seriousness, why not if its there and its more accurate.

Any of the following may apply:

1) it doesn't add any relevant improvement in sound quality,
2) it bogs down your DAW with bigger files and more throughput, perhaps even degrading overall sound quality,
3) it unnecessarily increases your operational cost by increasing storage cost.

Of course, in those occasions where it really does make a relevant improvement and your system is up to the task, then it makes sense.

Cheers,

Otto

 

[aviv]

a 96kHz sample rate only ensures that the accurate reproduction range is increased to somewhere in the 40kHz range.

So 192khz is even tighter in the high end?
How much better can it get?
Seems abit mad, not the idea but just generally

 

[solaris1982]

That's it! Next time I make a cd, it's just going to be a single 700 MB sample.
According to this conversion rate, I'll sell... 5.8 billion copies! Man the .7 billion people worldwide who don't own a copy will feel so out of the loop!

Those .7 billion will probably just download it...as a 44.1 MP3

 

[SouthSIDE Glen]

So 192khz is even tighter in the high end?
How much better can it get?

NO. NO. NO. Neither 96kHz nor 192kHz is any "tighter" in the high end than 44.1kHz is. Every sample rate reproduces up to 20kHz exactly the same. The only difference is how far *past* the human hearing range their response extends. Theoretically speaking, 192kHz could reproduce up to about 90kHz. Problem is, that's some 5-6 times higher than the frequency range of the human ear and at least 4.5 times more than the equipment driving it.

Seems abit mad, not the idea but just generally

It is. It it little more than a marketing ploy, or at best, an attempt to cover weak converter design. And because there is such widespread misunderstanding amongst the public about how sample rate actually works, we wind up with never-ending threads like this one and constant battles between the "believers" and the ones busting their bubbles.

G.

 

[The REAL Jigsaw]

I record at 88.2khz and can honestly say it sounds better than 44.1khz. There's simply more data being captured from the signal.

 

[Farview]

I record at 88.2khz and can honestly say it sounds better than 44.1khz. There's simply more data being captured from the signal.

NO, your converters just happen to sound better at 88.2k. It's not the sample rate, it's the converter design.

Technically, if there is no useful sound above 22k, you are not capturing more data with a higher sample rate.

 

[SouthSIDE Glen]

I record at 88.2khz and can honestly say it sounds better than 44.1khz. There's simply more data being captured from the signal.

To reiterate:

Before anybody comes back and says, "but my stuff sounds better at this rate than at that rate", that may very well be true. But it's NOT because of the actual sample rate itself; it's because of some design flaw or idiosyncrasy in the converter itself that causes different performance level or coloration between frequency settings. And there no guarantee that Frequency A is always going to sound better than frequency B when you move to another converter just because it does on the one your using now; each model of converter can have it's own character.

Sure, there is more data being captured, but that extra data serves only to extend the frequency response; it does nothing in and of itself to increase the accuracy or quality of frequencies already covered by a lower sample rate. And, as Farview just mentioned, if there's no actual or usable analog data up in that range, then there's no useful extra data being grabbed by the extra samples.

No one is doubting you when you say that your rig sounds better at 88.2. That may very well be. And as long as everything works fine for you and sounds better at that rate, then by all means continue using it. All we're saying is that it's probably not the sample rate itself that's doing it; it just so happens that your converter sounds better at that speed. But replace your converter with another, or move to someone else's rig, and there's no guarantee that situation will hold.

G.

 

[superspit]

the mind-set of "the bigger the number the better quality"?.....yikes! (product manuafacture's love you guys!!)

 

[The REAL Jigsaw]

To Southside,

I hear you. That's all I meant by it, it sounds better at 88.2khz, so I keep my masters at that quality/resolution. But thanks for clearing up the issue.

 

[Soundmind??]

I've read somewhere that 88.2 will accurately record a dog whistle, soooo it is useful for more than just recording sounds no one can hear and 98% of playback systems can't reproduce. Of course, the microphone will probably limit the frequency response to around 20k, soooo. I believe the "industry standard" according to the pro tools folks is 48/24, so I wouldn't waste the file space to go beyond that. On possible difference may be in processing accuracy, which again, would probably be inaudible.

 

[SouthSIDE Glen]

I believe the "industry standard" according to the pro tools folks is 48/24

There is no "industry standard". But the 24-bit does come close these days; everybody pretty much understands the true advantage of 24-bit.

The 48khz, however, is only "standard" if dealing with digital audio for video, because 48k is the native rate for much of that medium - and even there, it's not set in stone. But for dealing strictly with audio, most pro engineers would rather just work at 44.1 than work at 48 only to have to add sample rate conversion as an extra stage before mastering for CDA.

On possible difference may be in processing accuracy, which again, would probably be inaudible.

I don't mean this as sounding like I'm picking on you, because I'm not, seriously. But this is a perfect example right here of the deeply ingrained and false bias in understanding that makes this subject so difficult for so many.

There is ZERO, nada, zed, no increase in any accuracy anywhere because of increased sample rate. None. It's natural to think so, but that's just not how it works.

Every sample remains the same, it's still 24-bit accuracy. Increase the number of samples, and accuracy of reproduction does not improve, there is only an increase in frequency response. a 10kHz signal sampled at 44.1 and sampled at 88.2 *will be identical* when re-integrated again - assuming exact performance in the two DACs. Same thing with 20kHz.

As far as the dog whistle, 44.1 will capture many dog whistles as well, depending upon the whistle itself. Dog whistles can be as low as 16-18kHz, or they can go as high as 22-23kHz; depending on the size and type of dog they have been designed for. The 16-18kHz ones are the ones that some people can hear of "feel", and also can be captured by a quality 44.1 converter. The 22-23kHz ones are the ones that no one can hear and that would require a higher sample rate to capture. Assuming you have a precision mic that can actually grab it, of course; a lab reference SDC might do it, maybe even a decent ribbon. But below that, and your mix is probably safe for dogs .

G.

 

[Bob's Mods]

Dead on SSG. There is simply no advantage to recording at anything other than the target frequency. I've tried this experiment many times. The only difference you'll hear is a quality shift introduced by the convertor itself. The same convertor at 44.1 may sound better or worse at 48, 88.2, 96 etc. The fidelity of 44.1 as the CD standard is fine and will not improve at higher sample rates.

 

[Steve Henningsgard]

Metaphor: a 100 megapixel camera with a shitty plastic lense will turn out enormously accurate shitty-looking pictures. However, a 1 megapixel camera with a phenomenal glass lense has the capability of recording amazingly detailed images. Get it?

 

[kubeek]

I must say that the Niquist theorem is not entirely true.
As the signal frequency approaches the half of sampling freq, its amplitude gets lower, or beats occur (not sure about the term, but it is when the amplitude, of the combined signal, pulses at the difference of the two frequencies..)
Try it and you will see.
So it is not true that with 44.1khz you can accurately sample 22.05 khz wave. A 22.05 will have its amplitude highly dependant on its phase lag to the samples. As the frequency lowers, this dependancy is lower and the ability to recreate the wave from the samples is more accurate.
I don't have cooledit here, but when I last tried the results were confirming this.

 

[SouthSIDE Glen]

I must say that the Niquist theorem is not entirely true.
As the signal frequency approaches the half of sampling freq, its amplitude gets lower, or beats occur (not sure about the term, but it is when the amplitude, of the combined signal, pulses at the difference of the two frequencies..)
Try it and you will see.
So it is not true that with 44.1khz you can accurately sample 22.05 khz wave. A 22.05 will have its amplitude highly dependant on its phase lag to the samples. As the frequency lowers, this dependancy is lower and the ability to recreate the wave from the samples is more accurate.
I don't have cooledit here, but when I last tried the results were confirming this.

I believe it sounds like you're referring to aliasing. Without going into great detail, this is an artifact related to the fact that there is no such thing as a brick wall filter that cuts frequencies at exactly the target frequency. This is not the fault of the Nyquist theorem itself, but rather a physical limitation in our ability to build physical circuits that follow the idea precisely.

It is largely for this reason that we sample at 44.1Khz instead of a simple 40kHz - it's an attempt to reduce or eliminate aliasing in the audible range that can happen because of the physical limitations. We cannot simply put a brick wall low-pass filter up at 20kHz that passes nothing more, there's always going to be a soft knee curve to the filter. The idea is that by pushing the sampling rate a bit above the Nyquist ideal, we are pushing any aliasing and beat frequencies up and away from the <20kHz range and into the effective range of the low pass filter.

This is also, BTW, where some, if not much, of the coloration differences in converter circuitry and circuit design comes in; the design of the LPF section. The slope of the filter's curve and the efficiency of the filter can make a difference in how well it handles the kinds of problems you describe.

Again, this is not because of a problem with Nyquist itself or the sampling rate itself, but because of how well the circuit designed to handle it is designed.

G.

 

[kubeek]

When in digital, no circuits matter.
Yes you are right that the thing I described is aliasing, but that doesn´t change anything on that Nyquist theorem isn´t really correct.
On Wikipedia it says this:

If a function f(t) contains no frequencies higher than W cps, it is completely determined by giving its ordinates at a series of points spaced 1/(2W) seconds apart.

I understand this that if I sample for example a 22000hz wave with 44.1khz sampling, the samples will determine the 22k wave.
But when I look at the samples (which as I remember won´t look like a 22khz sine wave), and sample the exact wave like the one I can see, (in the way that I create a signal which will produce the same samples), I will obtain the same series of samples as above.
That means to me, that the function is not really determined by the samples, because I have two different input waves, which produce the same samples.

So is Nyquist right or not? I say no.

 

[kubeek]

Here is how a 22000hz sine wave, 20ms long, sampled at 44.1khz, looks like. (generated in digital domain)
Yes there definitely is 22khz content, but also an added 50hz amplitude modulation.

 

[Farview]

When in digital, no circuits matter.

The signal is only stored in digital. In order to use the signal it has to be converted back to analog where circuits do matter. You also have the analog path on the way to the converter to consider. There is also the sample clock...

Yes you are right that the thing I described is aliasing, but that doesn´t change anything on that Nyquist theorem isn´t really correct.
On Wikipedia it says this:

I understand this that if I sample for example a 22000hz wave with 44.1khz sampling, the samples will determine the 22k wave.
But when I look at the samples (which as I remember won´t look like a 22khz sine wave), and sample the exact wave like the one I can see, (in the way that I create a signal which will produce the same samples), I will obtain the same series of samples as above.
That means to me, that the function is not really determined by the samples, because I have two different input waves, which produce the same samples.

So is Nyquist right or not? I say no.

You are getting confused between the theory and actual practice. The theory assumes a brick wall low pass filter that doesn't exist. The theory is correct, the implementation is a compromise.

Here is how a 22000hz sine wave, 20ms long, sampled at 44.1khz, looks like. (generated in digital domain)
Yes there definitely is 22khz content, but also an added 50hz amplitude modulation.

But, even if you could hear 22k, it would not be there. The anti-aliasing filter on the DA would take all that out.

Get an oscilloscope and play this file into it, tell me if you see anything. You are looking at a computer algorithm guessing what the wav would look like on an ocilloscope. It's not real, nothing is real until it hits the analog world.

 

[kubeek]

You are getting confused between the theory and actual practice. The theory assumes a brick wall low pass filter that doesn't exist. The theory is correct, the implementation is a compromise.

The theoretical brickwall will be after 22050hz, but the wave is 22000hz, so it won´t be affected by that brickwall. Which means I don´t need to take it into account because it cuts only all frequencies higher and those are not present in sine wave.

You say the 22000hz won´t be there, or the 50hz?

The digital-analog conversion doesn´t play any role, because the theorem says that the original function is determined by the samples, and I have shown that there exist at least two different input waveforms that produce the same samples. Therefore the function cannot be determined by the samples.