When Is A Computer? (Part 8)
In a new series of When Is A Computer features, Andy Honeybone dissects 1984's favourite technology – word by word. This month deals with 'SAMPLING' and what it really means.
Market research is a term which conjurs visions of marauding ladies armed with clipboards. My own experience of the subject stems from being collared in Lewisham High Street and cajoled into turning a dial to indicate my 'interest factor' while watching videos of booze adverts. The only thing that kept me from leaving was the promise of a free gift. Having built up a fair thirst by the end of the session I was peeved to receive two miniscule screwdrivers which were to go rusty within the week.
The objective of a market survey is parallel to digitising a musical sound, and that is to gain information from a representative sample. This month's outpouring attempts to explain the workings of the new 'sampling' instruments which allow a natural sound to be recorded and played chromatically and polyphonically from a keyboard.
Unlike recording, the word 'sampling' implies that some of the input information will be ignored and so the process is more like a survey than a census. A movie camera does much the same thing by sampling one frame at a time. If the rate of sampling is too slow then the original cannot be reconstructed from the information available and the result is the musical equivalent of a jerky Charlie Chaplin silent film.
Before we get too immersed it might be as well to point out that the word 'sample' can refer to one digitised point (singular), or a whole collection of them as in 'trumpet sample' (plural). Sheep have the same problem.
For the current generation of sampling machines each (sub-) sample is stored as a digital word which means that first it must be converted from its original analogue form. Analogue and digital quantities have been discussed in previous articles but to summarise, analogue means voltages and digital means numbers. An analogue to digital converter is just a piece of hardware under the control of a computer. To capture a sound segment the process is: open sample gate for a few microseconds, close gate to trap sample in analogue memory (freeze it), start the analogue to digital conversion, wait until the end of conversion, get the digital value and store it in the next free computer memory location.
This process can repeat 30,000 times per second (fastest sampling rate on a Fairlight) and as if that isn't enough to make you say 'gosh' then cop this – at that rate, 16K of computer memory would be filled in just over half of a second. The advantage of such a high sampling rate is that because more information is gathered, higher fidelity is possible. It's all to do with bandwidth which is the range of frequencies that can be reproduced. The audio range extends to 20kHz and instruments with restricted bandwidths tend to sound dull and lacking in top.
Bandwidth is limited at two points in the audio-digital-audio chain. First, the audio analogue signal for conversion must not contain any frequency component (fundamental or harmonic) higher than the sampling frequency. Failure to comply with this requirement produces confusion (foldover) between the analogue signal and sampling frequency as to which one is doing the sampling. Imagine a short man and a tall woman – who holds the umbrella in the rain? This dichotomy (and I refer to sampling) is called aliasing and is prevented by a low-pass filter which has a cutoff frequency below the sampling rate.
The second point in the chain where bandwidth suffers is during replay after the digital to analogue converter when the reconstruction frequency must be removed. In this process, which is the opposite of sampling, a very steep slope, low-pass filter is used to attenuate the harmonics of the clock frequency at which the samples are being read-out of the converter. Because of the practical difficulties of separating the wheat from the chaff, a roughly 2:1 frequency ratio is imposed and the maximum recoverable frequency component (upper bandwidth limit) is half the clock rate.
When memory space is not infinite, sampling forces a compromise between sound fidelity and sound duration. The process is exactly the same as with a tape recorder in that using a faster tape speed gives better quality but uses up the tape more quickly. To capture ten seconds of a sustained Steinway bass note within a 16k waveform memory, the sampling rate would have to be cut to 1.6kHz and the resulting playback would have a bandwidth of 800Hz which would make the telephone sound like hi-fi.
So far the technology under discussion is the same as for digital echo units. The only difference is that the echo samples are continually advanced through memory to make way for new arrivals, and when the end is reached they are converted back to audio.
Digital, real-drum sound technology has made great strides in the last year but on analysis the surprise is why nobody ever thought of it sooner (answer: because the sound storage Read Only Memories (ROMs) were too small and too expensive). In comparison to the problems involved with applying the technique to a keyboard, each drum sound is relatively short, there are no sustain plateaux in the envelopes and there are no key release worries. Generally, drum samples are read out at the same frequency at which they were sampled in order to preserve the original pitch.
A great deal of playing around with the sample readout frequency is required to stretch one sampled voice over a four or five octave keyboard. For higher pitches the problem is that the duration of the voice sample is greatly shortened and provision has to be made for continuously looping around for the duration of a key press. The overall envelope is no longer natural in origin but is supplied by an envelope generator and voltage controlled amplifier which may be actual or logical (ie, exists in software only).
It is unfortunate that the start of a note often contains the vital information which identifies its source. Brass sounds start with a rasp, strings begin with a grate, and organs wheeze into life. All these nuances have to be crammed into the two second or less sample space available in the present generation of machines. Multi-sampling is possible where half-octave related samples can be supplied and this reduces the amount of stretching necessary. Such sophistication is required if acoustic instruments need to be reproduced with a high degree of realism. The distortion of natural sounds seems to be the selling point rather than the pitfall of these machines.
The Emulator is the only instrument to date which relies totally on sound sampling for its voices. The keyboard drives voltage-controlled oscillators, but instead of their outputs being processed by filters and amplifiers as in subtractive synthesis, they are used to clock the voice sample from computer memory. This arrangement takes care of reading the samples at different speeds to give different pitches.
Normally, within a computer system, information held in memory can only be accessed by the central processing unit (CPU) which is the posh name for the microprocessor. The micro performs a job similar to a warehouse's, fetching a value from a certain address, storing it in one of its internal registers, awaiting further instructions and finally sending that value to its destination.
The method is very slow because all transactions are through a central 'bottle-neck'. If you can imagine eight voice circuits, each trying to access the same stored voice samples via the micro at a speed fast enough to give a good bandwidth, then you will appreciate that it just can't be done.
A data transfer technique which entirely bypasses the micro goes by the name of Direct Memory Access (DMA). This process is to CPU transfer what shop lifting is to mail order. A DMA transfer halts the processor, dives straight into the memory, prises the value from an address and finally releases the bewildered micro to carry on where it left off. By this method a sampling instrument can deliver the goods with a simple micro whose main purpose is to run the disk drive.
A keyboard which is receiving a great deal of attention is the Kurzweil which promises to be a pre-sampled, polyphonic, touch-sensitive instrument specialising in natural sounds. The technical problems are formidable and a great deal of ingenuity must have gone into compressing the necessary data to manageable proportions.
Much research has been undertaken in this area because of the widespread interest in speech synthesis. Dynamic compression is a good starting point and obviously if any part of the waveform is found to repeat, then it need only be stored the once. Although waveforms may not be symmetrical they can be considered so, and then only half the data has to be stored. Another measure is to store only the difference between samples. Crafty dodges like these can reduce storage to one quarter of that necessary for direct conversion.
A technique by the name of Pulse Code Modulation (PCM) is making an appearance in digital musical instruments from Technics (SX-K250) and Roland (TR 909). A pulse code is exemplified by Morse Code and with PCM a pattern of pulses is generated from each sample which falls within a given range of values (quantisation). The data can then be transmitted along one line and the original modulating signal recovered by a decoder. The BBC uses PCM to multiplex their high quality broadcasts to distant transmitters and the technique is commonly used in digital recording.
The Emulator may cost a quarter of the price of a Fairlight but that's still a tidy sum. Those of you waiting for a budget model will be pleased to hear that circuit technology is advancing rapidly with 32-bit microprocessors and single chips capable of holding 32k of data. Unfortunately, the huge demand for microcomputer components has created a chip shortage and prices seem to be going back up. When and what this will bring is anyone's guess.
Feature by Andy Honeybone
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!