The Fairlight Explained (Part 5)
More on Kim Ryrie's CMI, the instrument that started it all, from the pen of Fairlight-user Jim Grant.
Sampling may be the CMI's most talked-about feature, but as this article shows, defining sounds using harmonic information can be just as dramatic.
So far we have discussed only one method of actually creating sounds with the Fairlight - sampling. This aspect of sound formation is probably the single most important feature of the CMI, and was certainly the focus of public attention when the machine was announced. However, the ability to specify sound by means of harmonic information can not only result in some very interesting sounds, it's also rather useful in an educational environment.
Two display Pages, 4 and 5, allow the construction of waveforms by harmonic data. They deal with exactly the same information but present it to the musician in different ways.
First of all, though, let's clear up a little mystery that's been evading us for some months - the Mode switch. Actually, this is very simple and is therefore something of an anti-climax. When a voice operates in Mode 1, only the first 32 segments of waveform RAM (4k bytes) are used to represent the sound. An unlooped Mode 1 sound will stop at the 32nd segment, even though another 96 segments of RAM exist. In order to compensate for shorter note event time as played on the keyboard, each of the 32 segments is looped several times before moving on to the next segment: this maintains a fairly constant net event length for any pitch. Mode 4 uses the entire waveform RAM (all 128 segments of it) and is always used for sampling since long, high bandwidth sounds need lots of numbers to represent them.
So what's the use of Mode 1? Well, calculating a time waveform from harmonic data can be quite time-consuming, especially if the supplied data is detailed and enables subtle nuances of sound to be generated. However, more often that not, only a simple waveform is required, and to calculate the RAM waveform for all 128 segments when a short loop is all that's needed is rather wasteful, to say the least. There's no hard and fast rule about which Mode a sound should be in: the choice is entirely the musician's. However, using a voice as the destination for sampling data always results in all 128 segments being overwritten, even if the voice selected is Mode 1.
Figure 1 shows a typical Page 5 display. This page displays the harmonic overtone series as a set of 32 'faders' similar to those on a graphic equaliser. Each fader is logarithmic in nature and has a range of zero to 255, allowing a good degree of control over harmonic amplitudes and thus enabling the application of a Fourier type harmonic series.
As an example, Figure 2 shows a square wave generated by the CMI, computed from the values of Fourier components shown on the faders. The resultant waveform is visually very similar to the real thing, and perhaps more importantly, sounds indistinguishable.
However, the power of the CMI lies in its ability not only to compute a complex waveform from a set of Fourier components so that it can be played on the keyboard, but also to compute a different waveform for each segment. Every segment has a unique Page 5 display, so while a Mode 1 sound has 32 sets of faders, a Mode 4 one will have 128! The current segment number is indicated on the display, and this allows a synthesised sound to change drastically throughout its duration, simply by the user filling each segment with a different waveform calculated from its own Page 5 fader settings.
In fact, the technique of using different waveform segments as the sound progresses is very much the domain of PPG synthesis. Generally speaking, these progressions are known as wavetables, and in the PPG, a sound consists of a set of 64 waveforms that reside initially in EPROM (they are transferred to RAM on power-up) which are read out sequentially when a key is pressed. The idea behind this system was to circumvent the need for filters by constructing wavetables that held a set of representative waveforms of, say, the classic filter sweep. Unfortunately, this results in a very hard, metallic sound, as the sound changes abruptly from one waveform to another, slightly different one. It's still a good sound, but in the interests of flexibility, PPG have chosen to incorporate the usual VCFs and ADSRs as well as extensive wavetable modulation.
The CMI is also capable of this form of synthesis to a limited degree, using the loop controls on Page 7. For example, suppose we had filled all 128 segments with waveforms that change very slightly as we progress through the waveform (see Figure 3). Now, if the loop controls were set up as shown in Figure 4 (this is a Page 7 display), moving CNTRL1 on the music keyboard would result in a different timbre when the note was played. Using this technique allows for some expressive playing, since the principle is rather akin to varying the filter frequency control on a synthesiser, the only difference being that the actual timbres can be radically different from one segment to the next.
To increase the timbral movement within a sound, CNTRL2 can be patched to 'LOOP LENGTH' on Page 7, resulting in sections of different waveforms being read out repeatedly. Since the waveform data is computed by the CMI, it's always constructed so that the waveform fits exactly into one segment, thereby overcoming looping problems.
Well, with all this talk of 'Fourier components' and the like, some of you may reasonably be thinking 'what's it got to do with music?' The answer, of course, is not much. Only scientists and engineers delight in quantifying the world which our senses seem to handle perfectly adequately. However, in order to express ourselves explicitly and unambiguously about a wide variety of concepts (some of which may be abstract) we need to use the language of mathematics. Fourier analysis and synthesis are mathematical statements about something which is not intuitively obvious: the fact that any truly periodic waveform can be decomposed into an infinite sum of sinewaves, usually called harmonics. Similarly, any periodic waveform can be constructed from the sum of an infinite number of sinewaves. The sinewaves have frequencies that are related to the fundamental of the waveform in such a way that the second harmonic lies at twice the fundamental frequency, the third harmonic lies at three times the fundamental, and so on. The fundamental itself is often referred to as the first harmonic.
Of course, obtaining a reasonable representation of a desired waveform does not require an infinite number of sinewaves: more than 16 is enough to give a good approximation. The Fairlight uses a maximum of 32 harmonics, which enables most waveforms to be synthesised with a fair degree of accuracy. Figures 5 to 8 show the development of a square wave by successively adding further harmonics and using Page 5 to compute the resultant waveform. The square wave doesn't contain any even harmonics (2, 4, 6 and so on) and we can see that the lower harmonic numbers set the basic square shape while the higher ones fill in the bumps and sharpen the edges. Even with all the odd harmonics the Fairlight can compute, the square wave is still not visually perfect, but it sounds OK.
Feature by Jim Grant
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!