|
Multirate signal processing has been a growing field
for more than one decade. The analysis-synthesis systems have
found many applicationssuch as speech, image coding,
transmultiplexing, adaptive filteringand spectral analysis. One
of the issues recognized as being importantto the successful
application of these systems is the computational complexity. A
significant solution to reducing the computational complexity is
based on the sinusoidal modulated filter banks. Traditionalblock
transforms, including the DFT, DHT, DCT, and DST, and the
extendedlapped transform are all examples of the sinusoidal
modulated filterbanks. In this dissertation, we pay our
attention to implenentationand application of the sinusoidal
modulated filter banks. We propose two implementations
for the modulated filter banks: 1. We present a unified second
order structure for computation of filter banks with
arbitrary length of basis functions. For the regularity
and parallelism of the structure, the new scheme really
provides a low cost hardware implementation and is especially
suitable for VLSI technologies. 2. For the M-band cosine
modulated filter banks, the analysis and synthesis
sections are efficiently implemented with two M-point of
DFTs and some extra operations. In addition, two main
applications of the DFT and DCT filterbanks are discussed in
detail: 1. The DFT filter bank is used for fast FIR filtering
in the transform domain. From a measure of computational
complexity, the proposed implementation has been proved to
have better performance than the overlap-add and overlap-
save algorithms. 2. The analysis filters of the DCT filter
bank are converted into a unified second filter form which
requires only two multipliers and five adders to implement.
The structure is used as a prefilter for efficient digital
filter design. The proposed prefilter structure has been
proved to be an efficient one for the design of lowpass,
bandpass and highpass digital filters simutaneously.
|