Digital audio

Music by numbers with Shiuming Lai...

"Digital" has become something of an essential tag in the world of consumer electronics during the past decade. Although much of the thunder of its entry into the domestic market was stolen by Compact Disc, this was done so by revolutionising a firmly established application - that of audio reproduction. The fact is, home computers, which existed as something relatively new for some years before CD, and virtually all computers that have ever been conceived, are also digital. It's only since the early 1990s that some aspects of home computing technology have caught up with, or surpassed, the specifications set out for their equivalents in more specialised hardware. PC expansion cards began by offering CD-standard audio, followed by Atari's effort, which not only improved on this technically but integrated it into the Falcon030 computer architecture.

More recently, the leaps and bounds made in microprocessor performance have helped bring recordable digital video to the masses. Check out the home movie making section of your local electronics emporium and you'll see a revival of digital fever taking shape, while Digital Versatile Disc (DVD) has been running around like a headless chicken over regional coding and audio standards. Digital television broadcasting is set to make its debut this year too, so with the furore surrounding this new era, you could be forgiven for thinking digitally-upgraded technology is not exactly a mature concept. If you did, you might also be surprised to learn audio recording and playback equipment of this nature was first demonstrated to the general public, as far back as 1967.

In this first article of a small series, we'll be looking at the principles of digital audio.

Digital recording

Analogue recording of any type is a physical representation of some quantity which is hence analogous to its real world counterpart. In audio, this is the straight translation of air pressure waves to a storage medium, be it magnetic tape, or vinyl disc. The waves directly influence the writing mechanism (vibrating a diamond or sapphire tip while it cuts a groove, in the case of vinyl mastering) so it is desirable, for the highest possible sound quality, to ensure the signal path throughout the system is as clean as possible.

Upon playback, many problems can arise from the dependency on the physical condition of the recording medium and accuracy of the transport mechanism. Tape is prone to stretching and drop-outs, vinyl is useless once scratched and fluctuations in the running speed of either manifest themselves in the form of inconsistent tuning. On top-end equipment these problems have been almost engineered around but at no insignificant cost.

Digital audio was developed to bypass these restrictions of analogue systems, achieving this by first converting the sound into a discrete structure, so it may be stored logically as opposed to physically. The idea is staggeringly simple and removes a lot of previous worry in the areas of storage and transmission. The critical factors governing the fidelity of sound signal reproduced are now more firmly concentrated at two points: the quality of analogue-to-digital (A/D: acquisition) and digital-to-analogue (D/A: reconstruction) conversion processes. This is what computer manufacturers, Atari included, usually don't tell you when claiming "better-than" or "CD-quality" sound. It may hold true while the sound is in the digital domain but what you get to hear depends most fundamentally on the individual device performing that conversion. No amount of techno-wizardry can compensate the ill effect of inferior A/D:D/A, so for professional use, this is why engineers and musicians spend extra money on quality offboard converters.

Information integrity

Literally anything which can store a sequence of numbers can be used to store digital audio (or video for that matter) and that includes the human brain, marks on a sheet of paper, even bags of peanuts, though more commonly a magnetic or optical disc. Once in the digital domain, it's feasible to use lower grade conductors for transmission, yet without affecting the
** italics ** presence ** /italics **
of digital data. That's the great advantage of digital storage - at the lowest level, base two (binary numbers) a piece of information is there or it isn't. Likewise in a digital tape recorder, even if the signal is coloured by noise on the tape, within a threshold, the digital levels of logical one and zero can be easily fully recovered and the tape noise doesn't enter the system (figure 1). It is this quality which gives digital technology its robustness.

This brings us neatly to duplication without degradation, another digital exclusive. If you trace a picture by hand, then trace another copy from that, and another from that, by the time you reach, say, the thousandth generation copy, no matter how carefully you traced, it would be notably different from the original image. On the other hand, if you traced it on graph paper, you could theoretically make infinite generations of copies, all identical to the first. You could also transmit it to the other side of the globe by whatever means, such as dictating the coordinates of filled squares over the telephone, and still end up with the identical picture at the other end (provided the same scale of graph paper is used).

Analogue-to-digital conversion

Dubious acts of electronically "stealing" portions of other people's work in the record industry have familiarised the term "sampling" to many a music lover. This generically refers to the process of converting a sound into digital form, but technically is more specific than that. Sound waves have dimension in time and magnitude, both of which are continuous variables. Sampling is the conversion of continuous time to discrete time, while at each sampling interval the level of the sound wave (amplitude) must be measured and converted to a discrete value which can then be stored in a digital memory device such as found in your computer. This is known as quantisation, though not to be confused with the sequencing term which does refer to time divisions.

Figure 2 shows how a digital representation is merely an approximation, exaggerated here for clarity. For this series we will only consider linear quantisation, where the quantisation interval Q is equal for the entire range. Non-linear quantisation systems based on logarithmic companding curves ("A-law" or "mu-law", there are two main types) are used for data bandwidth reduction, or increasing dynamic range depending how you look at it, and are beyond the scope of this article.

To achieve high quality with digital recordings let's go back to our graph paper analogy. Like a graphics display, we talk of resolution when discussing the extent of information depth recorded by some digital system: the smaller the discrete components (squares on paper or pixels on screen) the more accurately you can represent a continuous quantity. The audio sampling interval should be small so it must be done many times very quickly, the frequency being measured in Hertz (Hz, events per unit time, s (seconds)); it's also necessary for the quantisation interval to be small. Compact disc uses a sampling frequency of 44.1KHz (44,100 samples per second) with 16-bit quantisation resolution, giving 65,536 possible discrete amplitudes - Ed McGlone's explanation in AC#1 p27 "How do they do that?" explains this well. Dynamic range, mentioned earlier, is determined by the quantisation resolution, and is simply the range between the smallest and largest signal strengths generated by audio equipment, measured in decibels (dB), the unit of sound pressure change. In digital systems each extra bit of resolution increases dynamic range by a theoretical 6dB (varying with the quality of the DAC), so the linear 8-bit sound of the STe has a dynamic range of 48dB, while the Falcon's 16-bit CODEC chip has a range in excess of 90dB, enabling it to handle more subtle signal nuances and produce a far more realistic sound.

Another aspect of sound reproduction we must consider is signal-to-noise ratio (also dB). All recorded sound has some background noise present originating from the recording and playback process. The higher the resolution of a digital system, the better its S/N ratio, as the signal masks out the noise.

Nyquist's theorem

An important rule for digital recording, this stipulates the minimum sampling frequency to be used, should be at least twice that of the highest frequency component in the sound you're recording. The reasoning is quite simple - a single cycle of any kind of wave, by definition, consists of two complementary phases, so it must be sampled at least twice in order to capture some information about both phases. In figure 3, you can see a sampling frequency equal to the sound being recorded only manages to capture information about one phase.

In practice, when recording real music, which typically contains a huge range of frequency components, failure to comply with Nyquist's theorem induces a phenomenon called aliasing. This is a difference frequency noise relating the sampling and sound source frequencies (heard as an unpleasant whining or tinniness in the offending ranges). Look for the sampling tips boxout.

Watch this space

Next time will be a more in-depth look at sampled sound and the STe.



Terminology

"Sample" at the lowest level means a single word (number of bits) representing an amplitude measurement of a sound - although a collection of these samples describing a sound over time t is also called a sample. The latter is generally what people mean when talking of samples.
 
 

Sampling tips

All digital recording systems need to filter the sound before A/D to remove frequencies above the Nyquist limit. This job is delegated to what are known as low-pass anti-aliasing filters; inexpensive 8-bit sampling hardware for the ST incorporates such filters but they usually cut off quite low (just 5KHz in some cases). When sampling sounds for STe/TT computers in particular, it's advisable to use a 16-bit sampler if you have access to one, and convert the sounds down to 8-bit. The enhanced sound chips of those machines are capable of a frequency bandwidth well beyond that offered by many 8-bit samplers.

HiSoft Systems' Replay 16 is still available new, though you can just as easily use the facilities of a Falcon or modern multimedia PC.

If you make extensive use of sample CDs, invest in a CD-ROM drive and ExtenDOS Pro, which allows SCSI dump of the data on a CD. Equally applicable to Falcon owners, this method provides the best possible quality because it bypasses the inevitable degradation which occurs from traversing the analogue/digital boundary, even if 16-bit is used throughout. Before rushing off to buy the cheapest CD-ROM drive you can get, beware that not all drives support the command set for this feature, so check first. Ask for CD-DA or Digital Audio Support. There is still no ratified standard for this command set as far as SCSI-2 but ExtenDOS supports most current implementations from major manufacturers like NEC, Pioneer, Sony and Toshiba.
 
 
 

[Home] [Info] [Back] [Next]