Sampling and Quantization - Digital Speech: Coding for Low Bit Rate Communication Systems - page 35

Table 3.2 Step size multiplier values for 2,
3, and 4 bit quantizers [9]
Adaptation multiplier values
Previous
o/p levels 2bit 3bit 4bit

L1 0.60 0.85 0.80

L2 2.20 1.00 0.80

L3 1.00 0.80

L4 1.50 0.80

L5 1.20

L6 1.60

L7 2.00

L8 2.40

The recommended step size multiplier values [9] do not, in general, consti-
tute critical target values. As can be seen from Table 3.2 [9], the middle values
are fairly constant. What is critical, however, is that the step size increase
should be more rapid than its decrease. This is very important for preventing
quantizer overload.

3.3.6 DifferentialQuantizer
In a differential quantizer, the final quantized signal, r(n) is the difference
between the input samples x(n) and their estimates x p (n) .

r(n) = x(n) − x p (n) (3.28)

and

p

x p (n) =

1 ˆ

x(n − k)a k (3.29)

k =

k) th sample
and p is the number of previously quantized samples considered in the
estimation process.
The reason for this preprocessing stage to form the prediction residual
(prediction error signal) before quantization is that, in speech signals, there is
a strong correlation between adjacent samples and, hence, by removing some
of the redundancies that speech signals possess, the signal variance is reduced
before quantization. This reduces the quantization noise by employing a
smaller quantizer step size . Block diagrams of typical adaptive differential
quantizers are shown in Figures 3.7 and 3.8.

where a k is the weighting used for the previously quantized (n −