-
8/16/2019 09bit059-121104112517-phpapp01
1/13
Subband Coding
‐09bit059
(Mihika Shah)
-
8/16/2019 09bit059-121104112517-phpapp01
2/13
Introduction
• Vector quantization is used when the source output shows high
degree of clustering.
• Scalar quantization is used when the source output is linear in
nature.
• When the source has well defined characteristic then we can choose
a particular compression scheme.
• But in
real
world,
most
source
exhibit
a combination
of
characteristics. So subband coding helps us to solve this problem.
-
8/16/2019 09bit059-121104112517-phpapp01
3/13
What
is
Subband
Coding?
• A Source output is decomposed into its constituents. And each
constituent is
encoded
and
decoded
separately
to
improve
compression performance.
• It separates the source output into bands of different frequency
using digital filters.
•Different filters are used like low pass filter or high pass filter.
-
8/16/2019 09bit059-121104112517-phpapp01
4/13
Filters• A system that blocks certain frequency components is called a
filter.
• Filters that only pass components below certain frequency f0 are
called low pass filters.
• Filters that
block
all
components
below
certain
frequency
f0
are
called high pass filters.
• Filters that pass components that have frequency content above
certain frequency
f1
and
below
frequency
f2
are
called
band
pass
filters.
-
8/16/2019 09bit059-121104112517-phpapp01
5/13
This is
the
most
frequently
used filter banks which
consists of a cascade of stages
and each stage contains both
low pass and high pass filter.
-
8/16/2019 09bit059-121104112517-phpapp01
6/13
Basic Subband Coding Algorithm
It consists of three phases:
1. Analysis2. Quantization and Coding
3. Synthesis
-
8/16/2019 09bit059-121104112517-phpapp01
7/13
Analysis
Fig‐1. Block diagram of subband coding system
-
8/16/2019 09bit059-121104112517-phpapp01
8/13
(Contd…)
• Source output is passed through either non‐overlapping or
overlapping filters.
• The range
of
frequencies
at
the
output
is
less
than
the
range
of
frequencies at the input to the filter.
• The process of reducing the number of samples is called
decimation or downsampling.
• The amount of decimation depends on the ratio of the bandwidth
of the
filter
output
to
filter
input.
•If the bandwidth of filter output is 1/M of the bandwidth of filter
input then the decimation will be of M.
-
8/16/2019 09bit059-121104112517-phpapp01
9/13
Quantization and Coding
• Allocation of bits between the subbands is the main issue.
• Bits are allocated according to the measure of information content
between subbands.
• When the information content of bands is very different, bit
allocation procedure has a significant impact.
-
8/16/2019 09bit059-121104112517-phpapp01
10/13
Synthesis
• Quantized and coded coefficients are used for the reconstruction of
the original signal.
• From each subband, encoded samples are decoded at the receiver.
• Decoded values are unsampled by inserting an appropriate number
of 0’s
between
samples.
• The unsampled signals are passed through a bank of reconstruction
filters when the number of samples per second has been brought
back to
the
original
rate.
• Summation of the reconstruction filter gives the final reconstructed
output.
-
8/16/2019 09bit059-121104112517-phpapp01
11/13
Applications
• Speech Coding
• Audio Coding
• Image Compression
-
8/16/2019 09bit059-121104112517-phpapp01
12/13
(Contd…)
• The Moving Picture Experts Group ( MPEG ) has proposed an
audio coding scheme which is based on subband coding.
• There are three layers in which layer 1 and layer 2 both use a
bank of 32 filters. Input is splitted
into 32 bands, each with the
bandwidth of f/64, where f is the sampling frequency.
• 32000 samples per second, 44100 samples per second and 48000
samples per second are the allowable frequencies.
-
8/16/2019 09bit059-121104112517-phpapp01
13/13
Thank you
