Digital audio encoding of aperiodic sounds (noise)?

I am referring only to external aperiodic noises picked up by a mic; not something that occurs within the audio components, power supplies or the digital conversion itself. So for example, banging a nail into wood, popping a balloon, manually sawing wood etc. I have tried to search the web with little success. I understand the concept of how analog digital converters (ADC) encode periodic signals, error correction, interpolation, averaging. Of course those are mathematically quite apt with periodic sounds. But how does all that work with aperiodic sounds? Thanks in advance for any specifics.

 

Is this all theoretical or is there a practical usage of this info you're looking into possibly implementing?

Though I don't have knowledge to present in this case, I would assume companies dealing with Foley overdubbing would likely have some answers for you, at least in terms of what gear to use. The why might be a bit tougher to come by but if the end result is all that matters,...

 

Are you asking how very short sounds are captured? Like drum hits?
48000 snapshots a second seems to do a pretty good job. A sound would need to be extremely brief in duration to escape that, not including the extended reverb created from an incidental transient in most environments. Maybe I'm not understanding your question...

 

No I am not asking how short sounds are encoded. Digital audio encoding has an elaborate algorithm devoted to encoding periodic sounds including interpolations and error correction and the math makes sense with them. How does that encoding process handle aperiodic sounds that don't meet the assumptions underlying periodic sounds?

It may be that the algorithms are so powerful mathematically that they capture them just as accurately. Or maybe they don't. I just can't find any discussion of this rather basic point.

 

The simple side of my mind says “a sound is a sound” but that doesn’t add much to the discussion.

If you’re looking to get deep into it I’d suggest trying to address the AES/EBU or SMPTE crowd through FB or similar. I’m sure you’d be able to find info and conversation through those channels.

 

Bumping this. I'm sorry to bother you @Archimago but might you know where there is any discussion of the thread topic?

 

I don't believe digital encoding make any assumptions about whether the sounds are periodic. It shouldn't make any difference. The assumptions are about the maximum frequencies to be encoded. Error correction is unrelated to the encoding.

 

Thanks for your reply. I understand that the digital encoder will process any sound below the filter frequency. However I am talking about the interpolation algorithms to connect the samples. My understanding, which could well be wrong, was that the interpolation equations established by Shannon dealt with the reconstruction of sinusoidal waves of whatever complexity and superposition. But if something is not sinusoidal then what?

 

I think there may be some cases of non-sinisoidal waves near the sampling rate that might not be fully reconstructable, but that wouldn't be where most of the sound is, and wouldn't likely be audible to most/all human ears. But I'm not a mathematician. Even sounds like balloons popping and wood sawing are going to be mostly sinisoidal in nature.

 

Zappa used his synclavier for recording several sounds he used later in compositions, most of which treated to the point where they were unrecognizable in use (see, Civilization, Phaze III ).

For a more primative version, see Bernie Krause (and don't ask why this is a rectangular jpg):

 

For a non-digital application, see:

 

For an even more primative example, see:

 

My understanding is that encoding analog phenomenon has nothing to do with periodicity. For example, DC can be digitized accurately and there is nothing periodic about that signal. Perhaps you can share with us what your understanding of how ADCs encode periodic signals, especially that part about error correction.

My understanding of digitizing is that you look at the signal at any given moment and assign a code to it, doesn't matter what it was in the past, doesn't matter what it will be in the future.

 

I think Al Gator may have answered the question mostly above. Yes I understand about the digital sampling encoding - I was asking about the interpolation of the samples of non-sinusoidal waves.

 

This video explains it, using a square wave as an example of a non-sinusoidal waveform:

 

That's a great explanation - thanks for posting it. I remember studying Fast Fourier Transforms way back in my college days, it was some of the hardest math (and programming) I ever did. I was more than happy to leave them behind and just enjoy their results.

 

Math Rock.

 

I’ve wondered about this also. Dithering is a statistical process and requires repeated samples. But I’ve never found any rigorous discussion of how many samples (or how much time) is required for what level of spurious response improvement. I’ve seen discussions about the quantizing noise inherent in adc conversion and the reduction in it due to dithering but nothing of the issue of dithering effectiveness on non period waveforms, yet the only explanations I’ve seen of dithering assume periodicity.

 

But the kind of music that humans listen to necessarily consists of periodic waveforms, otherwise it wouldn't sound much like music. Even the human voice itself consists of periodic waveforms. So dithering is designed to improve the reproduction of music and voice. When you're dealing with non-musical, non-periodic waveforms, adding dither isn't going to be of any benefit.

 

A good ADC which encode in DSD128, with a built-in 1 bit delta-sigma modulator (actually two, one for each channel), take 5.6 millions "delta snapshots" per second from the analogue signal. It records every each beat of wings of a hummingbird. Nothing related with some hipotetical filtering of the "periodic" or "aperiodic" sounds. The resulting noise of the encoding, which have nothing in common with the original analogue sound, is thrown out in UHF, beyond 60 Khz, and in the subsonic range up to 10 Hz.

I don't understand your question.

 

Well yes and no. Thee are quite a few noise elements in all music production and in human speech along with the period waveforms..

 

I'm afraid I don't understand your reply. I never talked about filtering of periodic or aperiodic sounds, merely the interpolation algorithm. Yes if you sample at higher and higher rates than it becomes closer to continuous, but there still is interpolation. I was referring principally to Redbook ADCs in the OP.

 

Just as the mastering engineers of DVDs and Blu-ray discs have different processing methods available for film, videotape, cartoons, and mixed content to get the highest quality with the source material they're using, so to do digital audio mastering engineers have different methods and techniques to get the highest quality regardless if they're dealing with highly dynamic orchestral music, the latest Justin Bieber album, or chiptune music recorded from a Commodore 64.

And remember that with proper dithering, digital audio effectively has infinite resolution -- the only difference sampling rate and bit depth make are in terms of the highest frequencies that can be reproduced and the level of the noise floor, respectively.

 

I'm sorry but you are not answering the question.

 

I already did when I posted Technology Connections' video. The video specifically addresses the characteristics of the Nyquist Theorem when it comes to reproducing non-sinusoidal waveforms such as square waves.

The discussion then turned to the topic of dithering, and that's what I was addressing. There are different types of dithering to suit different types of audio content, and in some cases no dithering at all may be preferable.