Vous êtes sur la page 1sur 16

SUB BAND CODING

S.I.Rosaline II sem/ COS

Introduction
Compression schemes depend on data characteristics. Vector Quantization source output has high degree of clustering. Differential Encoding source output sample to sample difference is small. Scalar Quantization source output is random But, most source outputs exhibit a combination of characteristics.
4/6/2014 Rosaline 2

Introduction(contd..)
Decomposing source blocks into different frequency band requires TRANSFORM CODING. Disadvantage Blocking (generation of coding artifacts at the block edges). Overcome by LOT( Lapped Orthogonal Transform).

4/6/2014

Rosaline

Example

4/6/2014

Rosaline

Consider a sequence {xn } as shown above. Inference - fast sample to sample variation - slow varying trend (dotted lines) To extract trend avg. samples in a moving window smooths out rapid variations, makes evident the slow trend. Avg {xn } , window size -2 generate new sequence {yn } .

4/6/2014

Rosaline

Now {yn } easily encoded compared with {xn }. However, we want to encode {xn } . Hence done by, encoding {yn } by difference equation {zn }. Decomposing analysis Recomposing synthesis implemented using discrete time filters.

4/6/2014

Rosaline

FILTERS
Filter - A system that isolates different frequency components. Low pass filter, high pass filter, Band pass filter cut off frequency. Magnitude Transfer function - ratio of magnitude of input & output of the filter as a function of frequency.

4/6/2014

Rosaline

Ideal & Practical Magnitude Response Low Pass Filter

4/6/2014

Rosaline

Realistic filters ripple occurs in pass band & stop band. Filters digital in nature. Hence analog s/g undergoes sampling first. Sampling Nyquist theorem 2f0 . If Nyquist rule violated aliasing occurs original signal reconstruction is impossible anti aliasing filter used. Input output relation of the filter is:

4/6/2014

Rosaline

Filters used in Sub Band Coding

4/6/2014

Rosaline

10

Filter in sub band coding Quadrature Mirror Filter. High pass impulsive response is Filters are symmetric : Filters with fewer taps are less efficient than that with more taps. Example Johnston eight tap filter & Smith Barnwell eight tap filter.

4/6/2014

Rosaline

11

4/6/2014

Rosaline

12

Basic Sub Band Coding Algorithm

4/6/2014

Rosaline

13

Analysis Source output passed thro analysis filter bank Passband of the filters can be overlapping or non overlapping. Reducing the number of samplesdecimation/downsampling. Amount of decimation depends on the ratio of the bandwidth of the filter output to the filter input. After decimation, the output is encoded using ADPCM, PCM, or vector quantization.
4/6/2014 Rosaline 14

Non Overlapping & Overlapping filter banks

4/6/2014

Rosaline

15

Synthesis
Quantization and coding Quantized & coded coefficients used to reconstruct the original signal. (Bit Allocation) Encoded samples Decoded Upsampled passed thro a bank of reconstruction filters output added to get the final output. Three major components analysis & synthesis filters, Bit allocation, Encoding Scheme.
4/6/2014 Rosaline 16

Vous aimerez peut-être aussi