Digital Signal Processing Reference
In-Depth Information
1
0
−1
0
5
10
15
20
25
30
(a) Sample of ROM
1
0
−1
0
5
10
15
20
25
30
(b) Waveform at (a), Decimated by a Factor of 2.5, in Modulo−32 fashion
Figure 3.39:
(a) One cycle of a 32-sample sinusoid, interpolated values (every 2.5 th sample) marked with
diamonds; (b) Net decimated output sequence, formed by extracting the interpolated samples (marked
with diamonds) from the upper plot.
Additionally, small amplitudes which would be adequately encoded using a large number of bits may
actually be encoded as zero when using a small number of bits, leading to a squelch effect which can be
disconcerting.
In
μ
-Law compression, the interesting result is that smaller signal amplitudes wind up being
quantized (after reconstruction) with smaller amplitude differences between adjacent quantized levels
than do larger signal amplitudes. As a result, when the signal amplitude is low, accompanying quantization
noise is also less, and thus is better masked.
The script (see exercises below)
LVxCompQuant(TypeCompr,AMu,NoBits)
affords experimentation with
μ
-law compression as applied or not applied to an audio signal quantized
to a specified number of bits. Passing
TypeCompr
as 1 results in no
μ
-law compression being applied, the
audio signal is simply quantized to the specified number of bits and then converted back to an analog signal
having a discrete number of analog levels. Passing
TypeCompr
as 2 invokes the use of
μ
-law compression
prior to quantization. Upon reconversion to the analog domain and decompression (or expansion), the
spacing between adjacent quantization levels decreases along with sample amplitude, greatly improving
the signal-to-noise ratio. The parameter
AMu
is the
μ
-law parameter and 255 is the most commonly
