Hacker News Reader

Comment by rrrooss | Hacker News Reader

There has been a lot of talk about quantative aspects of but little about the qualitative. What does a 18khz square wave look like wher recorded and reproduced in 44.1 kHz. How is phase affecte?. Phase of an audio signal reaching the ears helps you perceive distance and position. How many samples are used to represent the frequencies at the higher end of the audible range. What is the quantization error difference between 16 and 24. And if the author can't hear the difference between 4bit audio and 12bit audio I question what he/she is listening on. The aliasing would be HUGE.

View on HN

An 18 kHz square wave sampled at 44 kHz looks like an 18 kHz sine wave, everything after the fundamental frequency is well outside the Nyquist limit and will have been thrown away by anti-aliasing filters. And furthermore you couldn't hear it even if it wasn't. Fourier decomposition of a square wave gives the sum of odd multiples of the fundamental, the next frequency is 3 x 18 kHz = 54 kHz.

Okay now actually look at the response on Matlab. As you get closer to the nyquist freq there are less samples to describe the wave. And while you get a bastadised 18k signal it's a far throw from what you put in. Anf while most are crying your splitting hairs the frequency response and dynamic range arguments pale in comparison to making the exact wave you put into the encoder come out. On the right system with the right recording the tiniest nuance in a room reverb helps trick your brain into believing the sound as actually happening. And this is not important to all listeners. But people saying that it makes no difference and is not important as a format to release music in are thinking only of their current needs and experiences. If you have actually heard a good recording on a good system in a good room you will know what I'm talking about. If listening experience has been laptop speakers, headphones and your mum and dads mini system then I totally agree any more than 320kbs mp3s are overkill. But to say that high res formats gave no place in consumer land is to show your lack of understanding of different people and different needs. In the age of iTunes you surely you buy the album and download it in whatever format u see fit. If u think audiophiles are wackos go get ya 320's. I would like to get it as I came fr the studio. The best it can be is as it came from the studio.

First off, forget about MP3; I'm not arguing for a lossy standard.

You can't get an 18 kHz square wave out of a system with 44 kHz sampling. You need at least 1 harmonic before it'll even LOOK square, and that requires a frequency response out to 54 kHz, ie. a sampling frequency of 108 kHz. You CERTAINLY won't find one in a reverb tail, even assuming you had a generator for one in the first place (you might JUST get one from a cymbal crash, but I don't think the physics works)

The point being, your source material can't contain an 18 kHz square wave either since it's been through a studio production system with the same antialiasing filters.

Since you know nothing about me but seem to be making assumptions anyway, here's some background. I've worked in broadcast audio; I own studio recordings in 24 bit / 192 kHz (Linn release of Mozart's Requiem, studio master series). I also own studio equipment that can actually play it. Audiophiles are, by and large, cash cows for companies with no scruples.

Good an audio nerd. I am making the point about the square wave close to the nyquist to point out the short comings of a format for accurately reproducing an input. Square waves in the real world are rare but I am arguing for a format that produces the most accurate representation of the intended signal. Imagine the situation where i have my guitar cab set up and I have a square wave (distortion) coming out of it and I far mic it up so as to capture the room a give a feeling of space. To really feel like your there you would want the resulting complex wave made up of the 18k direct sound from the cab and the room response a recording medium that can't do that accurately is second rate especialy when the formats are out there. And the higher the sample rate the further from the nyquist that 18k is and the more samples that can be used to describe the resulting wave an the more convinced my brain is that sound is real.

...right up to about 20 kHz, whereafter YOU CAN'T HEAR THEM. Hence 44 kHz sampling.

Seriously, A/B test this, you might be surprised.

Also if you think that's anything like a square wave coming out of a guitar speaker (or that that is even desirable in the most case), I've got a bridge to sell you. And yes, I do play.

http://www.eirec.com/DPimages/digisqwvtest.jpg

Okay here is a picture of what I'm trying to explain. And the author of this picture used a frequency much further inside human hearing range. This is transient response test I guess. My main argument is for the verbatim capture of the input wave. It will make the sound at 10k but it isn't the same wave that went in.

> What does a 18khz square wave look like wher recorded and reproduced in 44.1 kHz.

It looks like an 18khz sine wave, possibly with slightly reduced amplitude depending on the anti-aliasing filter rolloff and fc, but not enough to be audible (18khz isn't audible for a large portion of the population anyway).

> How is phase affecte?.

Probably delayed a bit by the anti-aliasing filter.

> Phase of an audio signal reaching the ears helps you perceive distance and position.

Not at 18khz, the wavelength is too short for your ears to notice any realistic group delay. High frequency localization is most done by ILD and effects caused by ear shape.

> What is the quantization error difference between 16 and 24.

Quantization error in a DAC just defines the noise floor. 16 and 24bit DACs are usually within a few dB of each other in dynamic range, it's really not audible.

> And if the author can't hear the difference between 4bit audio and 12bit audio I question what he/she is listening on. The aliasing would be HUGE.

What does aliasing have to do with bit depth?

http://www.eirec.com/DPimages/digisqwvtest.jpg Here is an example of higher sampling rates being useful INSIDE the human hearing range. This example shows transient response. Frequency response and dynamic range are low hanging fruit in high res audio debates. To argue that you can't hear it is a mute point when accuracy is the point, this is a reference quality format. Temporal resolution is the next area to persue if accuracy is of concern. If file size is of concern then high res is not aplicable. What is everyone arguing here? That if they any hear it on their setup it is of no use?

Reply on news.ycombinator.com