486 Chapter 15
itself. The error can be referred to as quantizing distortion, granulation noise, or simply
quantizing noise.
One obvious nonlinearity will occur when the input signal amplitude drops just below
the threshold for the fi rst quantizing level. At this stage the quantizing output will
remain at zero and all knowledge of the size of the signal will be lost. The remedy is
to add a small amount of noise to the signal prior to quantizing ( Figure 15.27 ). This
deliberate additional noise is known as dither noise. It does reduce the dynamic range
by an amount that depends on its exact characteristics and amplitude but the damage is
typically 3 dB. One virtue is that as the original input signal amplitude is reduced below
the 1 quantizing level thresholds (q) the noise is still present and therefore, by virtue
of the intrinsic nonlinearity of the quantizer, so are the sidebands that contain vestiges of
the original input signal. Thus the quantizer output must also contain information about
the original input signal level even though it is buried in the noise. However, the noise
is wideband and a spectral plot of the reconstituted waveform will show an undistorted
signal component standing clear of the wideband noise.
12 (^12) 50 0 50 100
(a)
150 200 250 300
10
10
8
8
6
4
4
6
2
2
0 Time
Amplitude
Figure 15.27(a) : Adding a small amount of random noise to the signal prior to quantizing
can help disguise the otherwise highly correlated quantizing noise. Aided by the binary
modulation action of the quantizer, the sidebands of the noise are spread across the whole
audio band width and to a very great degree their correlation with the original
distortion signal is broken up. In this illustration the peak-to-peak amplitude of the
noise has been set at ±1.5 q.