That's the obsolete article. It's very recommended to read these latest articles:
What does vinyl's sound differ from CD physically? And why significant amount of audiophiles consider CD losses? In the article I have tried to get to the base of the distinction reasons of a digital sound of CD with frequency of digitization 44000 Hz and vinyl, as the analog (Indissoluble) form of a wave (oscillogram) changes by transformation of a sound to digital and back.
Today we should discuss: whether it is necessary to pass, in general, to digital sources with high frequencies of digitization (96, 192 and more kHz) when we already have very popular format of CD 44.1 kHz, or all this sensation - only dexterous marketing courses of manufacturers of the audio equipment trying constantly to increase the sales.
Now see to frequency specter. In one previous article I spoke, that any signal can be presented as the sum of sinusoids. But here there are two moments:
1. The periodic or cyclically repeating signal of any form can be presented, as the sum of sinusoids (harmonics) with different phases (speaking roughly, time delays) and frequencies, multiple (to increase on 2, 3 and so on) the basic frequency of this signal. This frequency depends on duration of a cycle of repetition.
2. Not-periodic or not-repeating signal (it is any audiorecord) has an infinite continuous spectrum (the infinite amount of sinusoids «nestled to each other» with level, generally, falls down with growth of frequency).
Having accepted for true that average person does not hear above 20 kHz, we confidently reject all sinusoids which have frequency above 20 kHz (ultrasound), and we consider, that our perception of a sound loses nothing.
Let's assume, that we really do not hear ultrasound above 20 kHz, but it plays the pernicious role in black business of distortion initial (analog) sound. And now we shall look, as it occurs.
Our initial (analog) signal has the acyclic form (its waveform never repeats) and, hence, it have an infinite continuous spectrum. The sinusoids which are being above 20 kHz we have named all ultrasound (in this article). And on sound card's input it is necessary to understand as ultrasound not only a signal of acoustic wave which practically loses all the ultrasonic components directly in microphone or in guitar's pickup, but also various noises (high frequencies) had added to ultrasound components of initial signal at passage of amplifier's electric circuits, ADC and e.t.c (even noises from your computer).
What occurs now at analog-digital conversion of a signal with spectrum from 0 Hz to infinity without acceptance of special measures? All ultrasonic sinusoids with frequencies above half of sampling rate are moved to heard range.
On digital processing's slang it refers spectrum «mirrored». And instead of a clean sound from 0 to 22 kHz (half sampling rate) we have a mix pure sound and the deformed («mirrored») ultrasound (spectrum's part, located above 22 kHz). The ultrasound are moved by frequency to area below half of sample rate. Ultrasound spectrum are mirrored: top frequencies appear below, and bottom - above.
As example, you can imagine a duet of a baritone and the tenor. They both sing in A-major tonality . And now imagine, that you will hear, if the baritone continues to sing in A-major, and tenor starts to sing in G-sharp, moreover and words sings on the contrary.
Before digitizing a signal, by means of the analog filter of low frequencies (which passes through itself only low frequencies) delete all ultrasonic components. After make conversion to digital form.
However, ideal filters - are not present. And in the near future will not be. And that spectrum which turns out after a filtration, all the same contains ultrasonic components. Them power (levels of sinusoids on corresponding frequencies) till 24-30 kHz (depends on quality of the filter and the price of a sound card) smoothly falls down. Therefore filter does not eradicate ultrasound completely.
Then all these components with frequencies above 22 kHz, impudently pass through the filter, are kept in digital record and become audible to us, cynically being added to our virgin sound from 0 to 22 kHz.
How to act in this case? Good news: in filters available for us, ultrasonic components above 30-40 kHz on its output can be neglected in a kind of their low power.
The filter of low frequencies has such parameter, as cut frequency. It is the maximal frequency above which the filter, theoretically, does not pass a signal (sound). But actually this border is rather conditional. Attenuation of a signal passing through the filter decreases below some value above cut frequency. Frequency characteristic of the filter of low frequencies (dependence of attenuating signal level by frequency) has rather smooth form, gradually more and more attenuate a signal with growth of frequency.
The frequency characteristic of the filter as the snow hill. If at it was available sharp border (on cut frequency) the hill would come to an end with breakage, stopping almost instantly. But our real hill (as well as best analog filter) too smoothly decreases
and we also smoothly have move down all further and further in area of ultrasound.
The above frequency of input sinusoid from filter's cut frequency, the is worse filter passes signal. Therefore total power of frequencies 30-40 kHz (the sum powers of all sinusoids this frequenc range) much less than total power of sinusoids in a range 22-30 kHz.
For this reason if frequencies 30-40 kHz also will be displaced in area of heard frequencies will be less appreciable, than those which frequencies lay from 22 (half of sample rate of CD) up to 30 kHz. And with growth of frequency the filter works all better and better.
Therefore we need to increase only sampling frequency so, that total power of ultrasonic frequencies, placed above half of frequency of digitization and passed through the filter, was very small.
Power of these parasitic sinusoids after the filter as we have considered above, with growth of frequency falls down. And above frequency 48 kHz (half from 96 kHz) the ultrasound practically is absent. Hence, during analog-digital conversion at sample rate 96 kHz already, practically, will nothing in a heard spectrum of frequencies.
But now a following question: why, if all ok with 96 kHz, we need to climb on sampling frequencies 192 kHz and above?
In article was mentioned about "squaring" (artifacts), arising at digital signal because the signal is measured with time's breaks and value of measurements - too is discrete. In digital-analog convertors we look "trembling" of phase (jitter) also. In what it is shown? Measurements are done not strictly in regular intervals, and with some delays or advancings in time.
The reason of jitter becomes instability clock generator ADC and DAC of sound card. This generator sets the moments of measurements. Ambient electrical noises force jitter too.
If the ADC during record and the DAC during playing would have absolutely identical jitter (time distances between measurements would be identical at record and at playing) there would be no trouble. But time position of sample (phase) shivers and deviates. It, approximately, how to sing, sitting on a cart moving by cobblestone road. Or to pour expensive cognac to wine-glass with shivering hands.
"Squaring" and a jitter generate noise (distortions of a signal at digitazing). The more bit length of ADC sample and DAC (16, 24, 32 bits and more), the more close the restored value to the original. The more sample rate (less time distance between measurements), the more measurements is placed on signal.
Due to a plenty of measurements, on the average it is more exact (with smaller noise) is recorded and restored. It approximately as if you shoot at center of target from pistol can get 2 bullets from 6. If in your hands the automatic device, having missed sixty bullets, will hit the twenty. If who will survive from two bullets from 20 - precisely die. Thus, 192 kHz gets in center of target a little bit better, than 96, and furthermore - 44.
Whether application of high sample rate is necessary or not? Question ambiguous. Improvement of quality of sound - is complex question. It depend by sample frequency, and quality of the low frequencies filter, and stability of the clock generator of sampling rate, and accuracy of level's measurement / restoration by ADC / DAC. Mechanical increase of sample rate will not give a improvement at absence of work at other directions. Also it can appear, that the analog output of a professional sound card with the maximal sample rate 48 kHz sounds better, than at built in soundcard with frequency of digitization 192 kHz.
Author of original text and translation: Audiventory (a.k.a. musiclabo)