Recording at 44 vs 192

What's Hot
124»

Comments

  • NelsonPNelsonP Frets: 3347
    edited June 2018
    Cirrus said:

    My understanding is that you'd be right, if converting to half the sample rate just involved dropping every other sample point. Then you can be happy, because the samples you're left with are all real samples that were actually measured by the converter as you recorded them.
    Surely the shape of the wave being sampled would affect whether you can just drop every other sample point and end up with the same thing?

    Let's say that my sample frequency started at twice the frequency of the the tick marks in the examples below.
    You'd clearly hear a difference between the sine wave, the triangle wave and the square wave.
    Now, let's say I convert that to half the sample rate.
    Now all three waves would sound the same (and like the triangle wave), wouldn't they?
    But I suppose they also would have sounded like that if you'd sampled at the lower frequency in the first place.

    Thinking about it, there must be a therotical limit to the sample frequency which is ultimately determined by what can detected by human hearing. Why don't we all just use that frequency and be done with it? Ideally make it divisible by as many other nunbers as possible - say 60, 120, or 240 khz etc. EDIT: Actually this is pretty much what we have with 48, 96 etc. (I'm catching up slowly)

    And CDs can bugger off with their wacky sample rates.








    0reaction image LOL 0reaction image Wow! 0reaction image Wisdom
  • robinbowesrobinbowes Frets: 3021
    NelsonP said:

    Surely the shape of the wave being sampled would affect whether you can just drop every other sample point and end up with the same thing?



    No, the Nyquist-Shannon sample theorem states that:

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

    In other words, you can represent any signal accurately if you sample it at twice the maximum frequency of interest. The "shape of the wave" is not relevant.

    R.
    0reaction image LOL 0reaction image Wow! 2reaction image Wisdom
  • SporkySporky Frets: 27580
    edited June 2018
    NelsonP said:

    Let's say that my sample frequency started at twice the frequency of the the tick marks in the examples below.
    You'd clearly hear a difference between the sine wave, the triangle wave and the square wave.

    [snip]







    I think you need to be a bit careful here - it's easy to set up edge cases that make sampling look tricksy, but if you look at the starting point these edge cases often aren't very important.

    What's the fundamental frequency of each?

    If your starting sample rate is 96kHz then the fundamental of each is 12kHz, which is very, very high. All of these would sound exactly the same as the first harmonic in the triangle wave is 24kHz - above the human hearing threshold - and the first harmonic in the complex wave and the square wave is 36kHz, which is a lot above the human hearing threshold.

    If your sampling rate is lower then it's an odd question - I am happy to go through why, but it'll require that I do quite a bit of concentrating or the sums are likely to be off. I may even need to poke my tongue out of the corner of my mouth.
    "[Sporky] brings a certain vibe and dignity to the forum."
    0reaction image LOL 0reaction image Wow! 2reaction image Wisdom
  • NelsonPNelsonP Frets: 3347
    edited June 2018
    NelsonP said:

    Surely the shape of the wave being sampled would affect whether you can just drop every other sample point and end up with the same thing?



    No, the Nyquist-Shannon sample theorem states that:

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

    In other words, you can represent any signal accurately if you sample it at twice the maximum frequency of interest. The "shape of the wave" is not relevant.

    R.
    This is actually very clever.

    And it also means that once your sample rate gets above 40k you are never going to be in a siuation where you aren't represnting the signal accurately .

    I have learned something - i.e. 48 is best. Unless you plan to encode it as MP3 or put it on a CD. In which case use 44.1. Right?
    0reaction image LOL 0reaction image Wow! 1reaction image Wisdom
  • robinbowesrobinbowes Frets: 3021
    NelsonP said:
    NelsonP said:

    Surely the shape of the wave being sampled would affect whether you can just drop every other sample point and end up with the same thing?



    No, the Nyquist-Shannon sample theorem states that:

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

    In other words, you can represent any signal accurately if you sample it at twice the maximum frequency of interest. The "shape of the wave" is not relevant.

    R.
    This is actually very clever.

    And it also means that once your sample rate gets above 40k you are never going to be in a siuation where you aren't represnting the signal accurately .

    I have learned something - i.e. 48 is best. Unless you plan to encode it as MP3 or put it on a CD. In which case use 44.1.

    Right?

    It's at the core of most digital communication technologies, so yes - clever :)

    @Cirrus posted earlier that it's not as simple as using 44.1 for CD. I'm not saying I agree or disagree with him as I've not got time to check my rusty knowledge [1] but it sounds reasonable.

    R.

    [1] I have a B.Eng(Hons) in Electroacoustics, and did my final year project on Digital Filter Design, but I graduated in 1991!
    0reaction image LOL 0reaction image Wow! 0reaction image Wisdom
  • fastonebazfastonebaz Frets: 4065
    Thanks all for this interesting discussion. 
    0reaction image LOL 0reaction image Wow! 0reaction image Wisdom
  • CirrusCirrus Frets: 8481
    Well, I've been having fun hypothetical debates, but in practice, I tend to record at 48k. Moves the Nyquest up from 22 to 24k, which is only minor but I think it's a marginal problem anyway, and I'm happy that it moves that and the other limitations - cramping of EQ filters etc - that little bit further beyond what I can hear.

    I have worked with converters in the past where I could hear the difference at higher sample rates - an old Motu 828 in a studio I used 2003-2006 was *really* eye opening to me when I recorded a drum kit and was able to switch between the live mics coming through the desk and the monitored sound going through the A/D D/A; the converted sound was much flatter, lively metallic cymbal harmonics turned into sheets of white noise basically... and it wasn't as bad at 88.2 So I'm aware that sample rate *can* matter when the converter's not been designed properly. But I'm not superstitious about it: I try it, if it sounds worse I'll use a higher SR and if I can't tell the difference I'll use 48; especially these days when the CD is nothing but a sideshow as lossy codecs are totally happy converting straight from a 48k wav. And as already stated, if I do need a 44.1kHz file I'm happy to convert it using whatever decent SRC software is to hand.

    Dan Lavry, who makes converters, did a white paper where he argued that 60khz would be the ideal sample rate; the filters are moved up out of the audio range and can be gentler, there's only half an octave of audio that's above human hearing, and it's not so fast that the converters loose accuracy or the file sizes get silly.

    Here's my bottom line, anyway: Music's about soul and emotion, as long as they survive I'm happy.
    0reaction image LOL 0reaction image Wow! 0reaction image Wisdom
Sign In or Register to comment.