Unfortunately, although this process is generally understood, it is not common practise to employ correct dithering procedures at these stages. In fact, practising engineers are often under the misapprehension that dither is only needed once in a system, or that if there is sufficient noise in a recording then further dither will not be needed. This is an important misunderstanding that has so often led to lost quality in recordings.
518 provides two additive triangular probability-distribution dither forms – flat and high-pass. The essential difference between them is that H.P. dither has a lower subjective noise-floor, being about 3.5dB less audible than additive white dither.
Both spectra are illustrated in Appendix 1.
Noise Shaping with dither
Noise-shaping is a technique in which a filtered version of the quantisation error plus dither is fed back and subtracted from the quantiser input. By this technique, the total noise can be redistributed, moving the noise from spectral regions where the listener is sensitive to other regions where they are less audible.
In 518 we provide four different shapers; they are carefully selected for their different properties, each optimised for a different application.
Dynamic Range
The dynamic range of a digital channel is defined by the maximum level – often referred to as full-scale or 0dBFS – and by the quantisation noise floor which is itself determined by the sample word-length. A normal TPDF dithered quantisation introduces a benign uncorrelated addition to the noise-floor.
A background to this and a method of calculating dynamic range is given in the paper [2] referred to on page 31 of this manual.
From a users point of view, the noise-floor is determined by the number of bits; the subjective noise-floor is determined by the spectrum of the noise, i.e. the shaper used.
The most important application where maximising subjective dynamic- range occurs is in a word-size reduction. Here the correct strategy for maximising dynamic range is to use the Gain feature of 518 to raise the overall recording so that it nearly reaches full-scale. (This can be done by playing the recording and monitoring with a peak-hold indicator the maximum level; then replay increasing the gain sufficient to raise the peak to say -1dBFS).
Further maximising of the subjective dynamic-range can occur by minimising the impact of the added noise using a shaper. For example, if a 20-bit original recording achieved a peak level of -4dBFS, then transferring it to a CD using Gain +3dB and Shape B could result in a 19- bit subjective dynamic range on the CD. Transferring it without 518, i.e. by truncation, will reduce the recording to 15 bit with unpleasant low-level artefacts.
Pre and De-emphasis
The use of pre and de-emphasis as signal-processing means of optimising the subjective dynamic range of analogue channels, should be quite familiar ideas to audio engineers. In particular, pre and de-emphasis have been used in channels where the analogue noise level showed an increasing level with frequency, e.g. magnetic tape, shellac or vinyl grooves and Frequency Modulation. In all these cases, a well- documented property of music and speech is exploited, namely that for material microphoned at normal listening positions, and of acoustic
