Introduction to Sampling

Part 1

Part One of Newslink's Sampling Course by David Marshall

Sampling and samplers are integral to modern music. Samplers (like the Roland S-series) and sample playback modules (like the Roland U-series) are a commonplace and accepted part of everyday life, not only for engineers and keyboard players, but increasingly for drummers, guitarists, writers and producers. As usual, nobody explains the technical terms: either you're in and you know the jargon or you're on the outside feeling baffled and inadequate. As usual, there's less to it than you might think: sampling is nothing more than a way of recording sound.

Analogue methods, like magnetic tape or vinyl discs, record the continuous sound. Analogue in this case means proportional, each movement in the wave shape being mirrored by a corresponding movement of the magnetised particles on the tape or the grooves on the disc. Digital (ie numeric) methods represent a moving wave as a series of separate numbers: the technology is that of the digital computer, and much of the terminology is borrowed from computers. The demystification of the jargon merely requires an understanding of how the process works.

Fig. 1 SAMPLE RATE The frequency of waveform measured in Hertz (Cycles per second). To satisfy memory and speed considerations this rate should be fairly low, while still giving good quality results at the end. Nyquist's Theory suggests a Sample Rate of at least double the highest frequency to be sampled. Below that threshold the original waveform cannot be rescued from the Sampled data. The theoretical upper limit for human hearing is 20kHz (20,000 cycles/sec) - following Nyquist's principles Compact Discs (a form of sampled sound) offer excellent quality with a frequency of 44.1kHz. Roland's new S-770 sampler goes even further with sample rates of 24kHz, 44.1kHz and 48kHz.

Sampling takes a continuous sound wave and selectively records a series of 'snapshots' of the sound as it changes (it's similar to the way a movie camera records a moving picture as a series of frames each of which is a separate still picture). The separate 'snapshots' or samples are converted into numbers which, like map reference points, represent the shape of the sound wave at any given moment. This conversion of a continuous (analogue) sound wave into discrete numbers (digits) is done by the ANALOGUE to DIGITAL CONVERTOR (ADC).

The series of numbers is stored in the sampler's memory until playback. When a film is played back the speed at which the separate images change (25 per second) is fast enough to trick the eye into thinking it is receiving a continuous moving image. When a sampler plays back a recording, the missing segments of the sound wave have to be supplied by its internal software/technology. The DIGITAL to ANALOGUE CONVERTOR (DAC) reads the sampled information; on the basis of the stored samples it then interpolates the continuous wave.

Fig.2 BIT RESOLUTION The waveform is measured against a series of discrete values and given the value of the nearest. In other words it is QUANTISED. The more levels available the more accurate the reproduction will be. The resolution of these measurements depends on the number of binary bits employed. A 1-bit system can have only two levels (0 or 1 negative or positive polarity). A 2-bit system can have four degrees (00 and 01 give two levels of negative polarity while 10 and 11 give two levels of positive polarity). The early Fairlight Samplers were 8-bit (256 levels), while tile S-770 is 16-bit, giving 65,536 levels.

It sounds ideal - if only it did. Unfortunately the interpolated end result depends largely on the quality of the original information. The two most important factors in determining the quality of sampling are SAMPLE RATE and BIT RESOLUTION.
Sample Rate (fig 1) is simply the number of times per second that samples are taken. For acceptable results this must be often enough to capture all the nuances of a rapidly changing waveform.

The second consideration is the accuracy of measurement of the separate samples. Measuring the length of a car in whole metres, and rounding any fraction up or down, could have potentially disastrous results if used as a criterion for parking in a confined space. However, increasing the resolution one hundredfold and rounding up or down to the nearest centimetre should be well within the tolerance limits of the average driver. The same applies with Bit Resolution (fig 2), which determines the scale of measurement when translating an analogue wave into a numerical or digital format for storage and processing within a sampler. 8- bit resolution is at the low end of tolerability with a scale of 256 values, while today 16-bit (65,536-level) resolution is beginning to be regarded as a professional minimum.

Fig 3 shows how the transfer of information from the analogue to digital domain results in an approximation of the original waveform, the shaded area representing the approximated data. If the signal were to be converted straight back into an analogue waveform without correction this 'approximated' data would be heard as noise.

Fig.3 QUANTISATION PROBLEMS The shaded grey area is correct data. The red shaded area is either missing from the original waveform or has been added to it. This produces QUANTISATION NOiSE. In low resolution systems this would be heard as a low pitched signal. As the resolution increases it tends towards a more random 'white' or 'pink' noise element. becoming inaudible for practical purposes in professional standard equipment.

The combination of a low bit-resolution and a low sample rate leaves vary few reference points. In this example, sampling at 8Hz with 3-bit resolution, only the intersections of the grid pattern can be stored as digital values. The resultant waveform is shown as a heavy line) is vastly different from the original (shown as a thin line).

The greater the sample rate and the finer the resolution, the better are the chances of recreating the original waveform. However, there are some practical limitations: increasing the sample rate or bit resolution increases the storage memory needed and slows the processing power. The equation between quality and speed is determined largely by the cost and availability of technology: although modern samplers can now offer greater power at less expense than their predecessors there is still a large gap between the performance of machines at the upper and lower ends of the market. As usual you gets what you pays for.

This article has dealt with the principles of sampling and playing back a single 'one shot' sound at a fixed pitch. Most people will want more from a sampler than that. Subsequent articles will look at the problems of playing samples at different pitches; at the use of looping to produce sustaining tones; and at how the sampler can become a creative rather than imitative musical instrument.

Basic Synthesis 1 | Basic Synthesis 2 | Advanced Synthesis 1 | Advanced Synthesis 2| Sampling 1 | Sampling 2 | A History of Sampling