Many sound engineers assert that digital audio does not sound as good as analogue audio, yet most recording is done today digitally. The differences between the two techniques are due to the constraints of technology and physics.
How Does Digital Recording Work?
Bits and Pieces
A binary number is either a 0 or a 1. The lowest useful level of encoding possible is 2-bit (Binary digITs).
There are four levels of loudness in 2-bit recording:
- 00 (quietest or least defined)
- 11 (loudest or most detailed)
4-bit encryption has 16 levels of loudness (the range of loudness from 0 to the loudest possible signal), 16-bit has 65,536 levels of loudness. The more bits a recording signal has, the more levels of loudness are available. The sound quality of a digital recording is dependent of the level of encoding used.
The sampling rate also affects the quality of the recording - the higher the sampling rate, the better the dynamic range of the recording. If a sampling rate is too low it will intercept a waveform too few times to make a proper analysis of the sound. The sample will not intercept the signal enough times to build an accurate representation of the sound recorded.
The normal frequency response of the human ear is 20 - 20,000Hz1, so ideally the sampling rate will be at least double the upper value, allowing it to capture and analyse the full (human) audible range.
This means that the sampling frequency will sample enough of the waveform of the highest possible pitch and 'guess' the correct frequency and timbre of the sound recorded.
8-bit sound sampled at 22,000Hz, usually used in computer games and on the web (because af space constrictions), sounds naff. A sampling rate of 200,000Hz and an encryption of 96-bit would be good, but there is no computer powerful enough, nor a medium efficient and big enough, to record a sound at that level.
How Digital Audio Is Recorded
This is going to be quite hard to explain, but here goes!
Digital audio works by sampling sound into binary code. To save space and simplify the process the sound is usually encoded into 16-bit PCM (pulse code modulation) and sampled at 44,100Hz. This is the configuration most commonly used for compact discs.
First the sound is sampled by using PAM (pulse amplitude modulation). The amplitude of the waveform sampled is measured by each sample taken.
After the sound is sampled it goes through a process called 'quantising'. Each set of indefinite samples collected is given a definite value, dependent on the bit rate of the sample (remember that a 16-bit sample will have 65,356 levels of clarity, or definition). The amplitude of each sample that is taken is either shortened or increased to fit in to the nearest level of clarity. There is an error ratio, usually in the range of half a level (this means a sample could either jump or slip a level). If any errors occur at bit rates of 8 or more and sample rates equal to or greater than 220,500Hz - as they frequently do at low sample rates and bit levels - you would not necessarily notice, as there are so many levels of clarity. When you have fewer levels of clarity to play around with, then half a level error is quite a lot, take the 2-bit for example. At anything less than 8-bit where there is supposed to be silence in a track there is noise. This is because the quantising process has in fact superimposed itself over the characteristics of the wave!
To stop the sound in lower bit rated tracks from appearing quite so prominently we use dither, a process which adds a small amount of noise to make the quantising process unpredictable. This means that instead of a fixed set of variables that the quantiser usually utillises to sort amplitudes, they change for each sample. A triangular sampler usually produces the best results.
The sample is now referred to as PCM.
Imagine a low frequency sine wave, also imagine eight samples taking place along the wave form. The samples measure the amplitudes of the waveform and, when processed, the digital waveform matches the original sound.
Now imagine the same conditions, but a frequency roughly four times higher than the original wave. The digital recording still sounds like the original, low frequency sound wave.
This is because although the sampling rate was the same the sample rate was too low to accurately predict the second, higher frequency correctly, and aliasing has taken place.
To beat this, a given fact is the Nyquist Theorem2. This holds that to reduce the likelihood of aliasing a sample rate should be at least two-and-a-half times the frequency as that of the highest sound you wish to record. This means that the digital process will sample a sound at least twice along its wavelength, giving an accurate sample.
How Do You Get Sound Back Out?
Ah, re-sampling and re-construction. This employs the use of a low-pass filter. Again using the Nyquist theorem we assume that the filter must have a frequency half that of the sampling rate. When a binary impulse is passed through the filter it produces a 'ripple'. Each time an impulse passes through another ripple is created. Each impulse has a different amplitude, so each ripple is of a different size. When the output of the filter creates a waveform it simply 'joins up' the 'ripples', like a piano note is made of the fundamental and its harmonics. The pitch is not perfect however, so the higher (sometimes barely audible) frequencies are reduced in amplitude. The recorded sound now is almost exactly the same as the original. Again low levels of noise, or dithering, is used to negate any aliasing that may occur.
Analogue Audio is Easy
Analogue audio, when recording on magnetic tape, transfers sound into a voltage and polarises the magnetic particles in a way which is (more or less) analogous to the sound produced. In other words, the sound is 'sampled' instantaneously, at an infinitely fast 'sampling rate'.
Which is Better?
The reason why digital audio is thought by some to be worse than analogue is that a digital signals 'loses' most of the signal that is recorded, it simply records changes in voltage in a given period of time (1/44,100th of a second) with a certain amount of clarity (16-bit or with 65,536 levels of perception).
The problem with analogue recording, on the other hand, is that you will always incur noise from the equipment you are recording. Noise comes from mechanical pieces moving, hum from cables etc. The information is carried as a change in voltage down a cable and is analogous to the original waveform. The distance the cable travels degrades the signal, and as the information is a part of the signal, this is also means the sound is degraded.
Digital signals do not rely on the quality of signal between devices as the information is sent as a definite off or on (0 or 1) and so is a separate entity to the signal.
Compare this with analogue and you see the pros and cons of both systems. Analogue technology records all the signal all of the time - this means that all the information included in the sound is recorded, but it is subject to degradation, because of noise. Digital recording captures the sound without (much) noise, but the depth of sound is found wanting.
Perhaps it would be useful to think of analogue as taking an infinite (an amount pointless to measure) number of samples a second at an infinite level of clarity, but the quality of the signal is affected by the way the information is transferred and how that information is read/written/interpreted. A digital signal, although it has a better delivery system, only records the 'bullet points' of the signal rather than the 'full conversation'.
Perhaps it is even easier to think of the two systems of water delivery. A well could represent digital audio, the bucket size the bit depth, the length of string the amount of times you are able to bring the bucket to the top of the well. A stream could represent analogue audio. The stream constantly flows but there is never an adequate way to retain the water, as collecting the water usually stirs up silt - muddying the water/signal.
Copies, Copies, Copies
Imagine copying a video (home recording, of course!), the copy looks quite good, in fact almost indistinguishable from the original. Say you lose the original and you need another copy, you copy the copy. The second copy is not as good as the first, and nowhere as clear as the master tape. Imagine doing this 500 to 1000 times. Even using top of the line equipment degradation occurs. Now imagine copying a high quality MPEG-1 layer file in exactly the same way. No appreciable loss in quality. Why? (This example deals with video rather than audio recording, but the principles are largely the same.)
Perhaps the most obvious benefit of digital audio is the fact that the information recorded is generally very hard to corrupt when compared to analogue, and a good deal quicker. Why is this?
Remember that digital information is measured as pulses of voltage, either off or on. There is very little chance of a misunderstanding between one piece of digital kit and another, as they deal with definite values.
This 'understanding' between digital machines means that a copy of a master recording will be a clone of the original recording (well not exactly, but enough to fool you, more later).
As the information is binary it very simple to simply copy the information in bulk - an extremely fast process not needing to be done in real time.
Analogue is another matter. The information contained within an analogue system is bulky and unwieldy. Basically whatever machine is recording it has to wait for a signal from the output device before it can record the signal. This can only happen in real time.
Even if connections between analogue machines are the same as between digital machines, the sound is still only part of the signal (not, as in digital, a separate entity) and still suffers from interference. This interference is caused by proximity to magnetic fields, 'dirty' power even loose connections. Digital signals are still subject to these fluctuations, but as digital relies on pulses of information, rather than a steady stream of information, digital signals rarely incur any penalties (as long as the pulses are powerful enough to bigger than the background noise).
A copy of a digital signal is extremely accurate, as we've discussed. However, it will include some incredibly minute mistakes. These mistakes could be due to any circumstance, buffer underrun, energy spikes, dust on laser and so on. Whatever can go wrong will. Unlike analogue copying, a digital copy will be specially coded in a technique first formulated and performed by Philips (after all they did invent the CD). Basically samples are not sent in sequential order, instead they are sent and received in an apparently random order in a process known as interleaving. This means that if the sample is damaged in some way then the next sequential sample is in another physical space to the first and so on. The CD player is then able to 'fill in gaps' - a kind of 'persistence in hearing'.
Why the Digital Revolution?
The reasons why there has been a digital revolution are:
Digital information is easy to store, and make changes to.
Digital technology is easy to store, cheap to buy and easy to use.
The techniques used to encode digital signals mean that the information carried by the signal (0's and 1's) is forever separated from the signal itself. This means no whirr from recording/playback equipment, no noise from the disc etc.
Record companies are able to charge large amounts for a 'new' technology - even though they now cost less to produce and distribute than vinyls/cassettes.
New advances are now able to be made using digital technology, therefore increasing sales of new playback devices and utilities as the defunct media are put out to pasture. The companies make more money more quickly (and then blame everyone else when the market collapses under its own weight).
CDs look nice and can be used as coasters or frisbees when they get scratched.
It is relatively easy to copy, edit and pirate digital information.
Digital equipment rarely wears out, compared to analogue.