How It Works - Digital Audio (Part 2)
In the second instalment of our 'How It Works' series, engineer/musician David Mellor focuses on the operational principles of digital audio systems.
If it's digital, then it has got to be great. Right? Not necessarily so. David Mellor pulls his digit out and starts talking numbers...
Digital audio is going through its adolescent stage at the moment and, as we all know, this can be a difficult time. Although research had been going on for several years at such institutions as the BBC in this country, Soundstream in the USA, and Denon in Japan, digits did not come into the public eye until 1981 when the first digitally recorded classical music LPs were issued. Congratulations go to Decca for being first.
To understand why the transition to digital was, and is, necessary we must first look at the problems of good old-fashioned analogue.
'Analogue' comes from the word 'analogy', which my dictionary defines as "Agreement, similarity, process of reasoning from parallel cases". An analogue recording is one which represents sound vibrations in air in another form, such as variations in magnetism in a length of tape, or wiggles in a groove on a plastic disc. There are other means of analogue recording which we do not need to go into, but they all work in a similar way.
The problem which all recording systems need to cope with is the extremely high sensitivity of the human ear - if we were bats then our problems would be much greater! The loudest sounds the ear can cope with - say a jet engine at 100 metres - are around 120 decibels (120dB) louder than the quietest sounds we can hear. In plain language, the loudest sounds are one million times louder than the quietest! Translated onto disc, if the stylus moved 0.1mm for a medium level sound, then it would have to be capable of resolving a movement of one thousandth of this for a very quiet sound, and also be able to cope with a megawiggle of 100 metres for jumbo jet volumes. Clearly, this is impossible. How big would a 12-inch single have to be?
The solution is to restrict the dynamic range to about half - around 60 to 70dB. This is usually acceptable for music reproduction, but it does mean that on any halfway decent system you can clearly hear the background noise of the medium (disc or tape) intruding into what you want to hear. Please don't expect me to say that digital solves this problem completely, because it doesn't, at least not yet. Analogue has, however, other problems which you cannot get rid of - either in the real world or in any flight of technological fancy you may care to take.
Suppose you made a photocopy of a black and white photograph. You would not expect it to be as clear as the original. Make a copy of the first copy and it would look worse still. The same problem exists with analogue audio. Make a copy of a tape, and it will contain all the nasties of the original plus nasties of its own - noise and distortion. And every copy down the line will get worse.
Take the example of a multitrack recording issued on LP, or 'black disc' as it has come to be known. The original is the multitrack master, which is mixed down (ie. copied) onto stereo tape. One. This tape is so valuable, having taken hours and hours of studio time to create, that a copy is taken of it to use in further production. Two. In the disc cutting room, a certain amount of equalisation is necessary to enable a satisfactory cut. Different tracks need different amounts, so a copy is made with all the changes from which the cut is made. It mounts up doesn't it, and we are not even halfway yet!
The equalised production master, as it is called, is then copied onto the master lacquer disc. This is copied three times mechanically to make the stampers which press out the records that go into the shops. According to my arithmetic, this makes at least eight generations of copies, each copy having the faults of the last and bringing new ones of its own. It's a wonder there is any music left by the time the record gets on to your turntable!
To sum up my opinion of analogue, therefore, it is not much good in the first place, and copying only makes it worse. No point in beating about the bush is there?
Hold up both hands in front of you. How many fingers can you see? I make the answer ten.
Isn't it a funny coincidence that we should have ten fingers and we also count in tens? Isn't it intriguing as well that we sometimes call fingers 'digits'?
Computers, however, do not have fingers so they can count in any system they like, the easiest way for them to do it being binary, where you only have two digits - 0 and 1. Computer hackers will know all about this of course and I am not going to pander to them by rabbiting on about it. All I have to say is that the point of having just two digits is that they can be easily represented in material form. For example, 0 could correspond to -5 volts, and 1 correspond to +5 volts in an electrical circuit. Alternatively, 0 could be a light spot on a surface, 1 could be a dark spot. Getting rid of in-between levels makes it dead easy to store and retrieve information. Digital information is always stored using this binary system, the only problem being how to apply it to something useful, like audio.
Figure 1 shows the Mellor Digital Industries 2-bit digital audio system. Here the incoming analogue audio is analysed into four discrete levels.
As you can see, the angular-looking digitised waveform vaguely resembles the smooth original analogue version.
As there are only four different levels available it is not a very precise reproduction, but it will serve as an example. Translated into binary, the four levels would be 00, 01, 10 and 11. The waveform could therefore be stored as the following sequence of binary numbers:
01 10 11 11 01 00 01 01 10
We could never retrieve the original waveform because we have made such drastic approximations. If we had used more than two bits (Binary digITS), then we could come much closer, at the cost of more storage space for the information.
Modern full-specification digital equipment uses 16-bit 'words'. This means that 65,536 different levels can be distinguished, in comparison to the four different levels of my 2-bit example.
There is a simple rule which links numbers of bits to signal-to-noise ratio. Simply, each extra bit increases the signal-to-noise by 6dB. Put another way, if you multiply the number of information levels by two, then you halve the noise level of the digitised signal. My 2-bit system thus has a signal-to-noise ratio of a mere 6dB, while a 16-bit system has a theoretical ratio of 96dB (the bigger the better). I deliberately said 'theoretical' because these figures can be made worse by poor design, or improved by clever mathematical manipulation, which I shall comment briefly upon later.
In a digital system, each time the analogue-to-digital convertor ('ADC' as it is called) measures the height of the incoming analogue signal, it is said to have taken a sample. This is not sampling as in Emulators or S900's, because it takes place on a much smaller time-frame. Audio frequencies, as you know, range from 20Hz or so to 20,000Hz, depending on how good your hearing is. It is logical, therefore, to assume that to digitally encode (record) a frequency of 20,000Hz, you will have to take samples at a frequency somewhat higher than this.
Leaving out the complicated mathematics (I am told there is a chap at the University of Cambridge who understands it!), it can be proved that to digitally encode any frequency, you must sample at twice that frequency or greater. So, to encode an audio signal up to 20,000Hz you must sample at 40,000Hz at least. In other words, take 40,000 samples every second.
In practice, the sampling frequency (often referred to as the sampling rate) has to be greater than this to facilitate the filtering out of unwanted frequencies from the final audio output. How much greater depends very much on the filter designer's art.
There is nothing mysterious or magical about digital audio. In a few paragraphs we have specified the main parameters of a professional system. If we need a signal-to-noise ratio of 96dB, then we must use 16-bit samples. If a frequency response up to 20,000Hz is required, the sampling rate must be above 40,000Hz. In fact, the Compact Disc system takes 16-bit samples at a frequency of 44,100Hz, which is not too different to this.
This brings me to the point I was making in the introduction about digital not being wonderful as of right. If fewer than 16 bits are used then the signal-to-noise ratio will not be as good. Likewise, if the sampling rate is too low, then the frequency response will be correspondingly poor. It is up to the manufacturer to decide these parameters. Sticking a label saying 'digital' on the product will not automatically make it good.
 Compact Disc
What can I say? In my opinion it is the best thing to hit audio since Ian Dury's rhythm stick! It is not the specification of the system that makes it so wonderful, indeed many would say that it could be improved upon considerably. The important point is that the binary numbers on that little silver disc are the same numbers that the producer approved in the recording studio when the track was mixed. No matter how many copies are made, the numbers stay the same. With a CD you are, in effect, hearing the original master, rather than a pale imitation.
At this stage I ought to climb down a little from my disparaging comments about analogue systems such as tape, or it might appear that I regard anything non-digital with something worse than I contempt. Analogue magnetic tape has been a mature medium since the mid-Fifties and in the last thirty years has been improved almost to perfection. At all stages of development good producers have made use of its characteristics, good and bad, and it is true to a large extent to say that "the medium is the message". The distortions produced by analogue recording methods are an integral part of the 'sound'. We have a different medium now with digital and our sound will also be different - but we can now let everyone hear an exact copy of the original.
There are enough compact disc players and discs in circulation to spare me having to explain what they look like and how they are tracked by laser and last forever and provide the utmost in user auditory satisfaction and....
What you probably did not know, is that digital information is recorded onto a spiral path of dots on the compact disc which the machine can track. Encoded into that datastream is a system of error-correction which cynics might say allows the manufacturers to get away with sub-standard product. Error-correction is necessary in all digital systems or else, instead of the odd whoosh or click you get from old-fashioned black disc, you might get something very nasty indeed.
The digital codes stored on the compact disc contain extra bits which can be used to check whether the data has been corrupted in any way. If this proves to be the case, the original data can often be reconstituted exactly from spare codes in the datastream. If the damage is too bad, then a system of interpolation (electronic guesswork) tries to make good the lost data as best it can.
The routines for doing this are quite sophisticated and are clever enough to know if the task is beyond them. If so, the output from the player would be muted. If this occurs on any disc you have, you should take the disc back to the shop and complain. They are usually very understanding about this, but I often find that if I take back a faulty disc then the replacement is faulty in exactly the same way. This would happen if a batch of discs had been pressed by a faulty stamper.
I do not want to go into technicalities relevant only to scientists and mathematicians, but there is one word used in relation to compact disc players that deserves attention - oversampling.
Like 'overstrung' in relation to pianos, this is a 'goodness value' term used by manufacturers when they are fairly certain that no-one knows what it means. 'Oversampling' is a technique developed by Philips to put one over on their awfully clever Japanese competitors. Put simply, it means you can squeeze out an extra 2-bits' worth of performance by creating extra samples which correspond exactly to what they ought to be, taking into account the samples that are actually present. Note that you are not getting any extra information from the disc; that information was not there in the first place. What is happening is that the laws of physics are being fooled into putting the noise that exists, into a part of the audio spectrum that you can't hear. That's clever! It's a good trick and it works - but do not be conned into thinking that it works magic. That only happens in advert copywriters' imaginations.
There are very few companies in the public eye who are actively involved in developing digital systems. Sony, of course, is the best known of those that are. They currently have three systems that we should know about.
 1600 Format.
The Sony 1600 was the first commercial video-based 16-bit recorder, and continues in its 1610 and 1630 incarnations. It is not possible to record digital signals on ordinary tape because the information rate is too high. 16 bits, 44,100 times a second works out to over 700,000 bits per second - and that is only for one channel with no error-correction bits included. The obvious answer was to use a video recorder, which is capable of this high bandwidth, to store the information.
A 1600 system therefore comes in two boxes - the processor and the video recorder. In order to get the digital signal onto videotape - normally the professional U-Matic variety - the processor has to format the binary digits into a pseudo-video signal. You can actually plug this into a video monitor and watch the bits go by on the screen. Fascinating.
 F1 Format.
The Sony F1 system is very similar but can work with domestic video recorders, Betamax being the norm. There is a price to pay, and that is a reduction in the capabilities of the error-correction process. Nevertheless, the F1 is a full 16-bit recorder, and with good tape it can provide fully professional results at a reasonable price. There are companies - not Sony as far as I am aware - who make conversion boxes so that you can copy your F1 tapes, in the digital domain, onto a 1600 system for compact disc mastering. I would not advise you to spend any money on this system unless you envisage a financial gain in the very short term, because waiting in the wings is...
Instead of using a separate video recorder and digital processor, why not combine them both in one box, and tailor the recorder section to the specific needs of the audio?
Enter R-DAT, short for Rotary-head Digital Audio Tape. This is Sony's plan to conquer the world and consign the faithful old Philips compact cassette system to the dustbin. R-DAT uses cassettes, smaller than those we are familiar with, and a recorder with a rotary-head like a video recorder to store 16-bit digital audio at a sampling rate of up to 48kHz. I have heard it in operation and it sounds very good.
No one knows when R-DAT is due to be released to the public, or what final form it may take, but watch out. I think it is going to be big. Bigger than...
Short for Digital Audio Stationary Head. This format, developed by Sony and Studer in co-operation, is meant to be the upcoming professional format, though the stereo version has not caught on too well yet. There are political differences between Sony and Studer which make life awkward for the would-be purchaser. The multitrack version is doing a lot better with sales of several hundred machines around the world. However, another Japanese company is making Sony just a little hot under the collar with its infant...
Mitsubishi's PROfessional DIGItal format is another fully-fledged pro system using stationary heads. Both stereo 2-track and 32-track recorders are available.
It is difficult to know what Sony are going to do with their R-DAT system. It is not inconceivable that they will bring out a multitrack recorder which would use several cassettes. Expand as necessary. I don't think anyone in the UK really knows yet, so we will just have to wait and see. Whatever happens, we are undoubtedly going to see a steady trend towards digital audio, and the gradual decline of analogue, both in the professional and domestic audio markets. This will not happen overnight, so don't sell your Walkman just yet.
Now you know something about what is going on inside these little digital beasties, you should be in a better position to judge whether to jump into the market now, or hold on a while. One thing is for certain, digital is the future... you can count on it.
Feature by David Mellor
Previous article in this issue:
Next article in this issue:
mu:zines is the result of thousands of hours of effort, and will require many thousands more going forward to reach our goals of getting all this content online.
If you value this resource, you can support this project - it really helps!