home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Tech Arsenal 1
/
Tech Arsenal (Arsenal Computer).ISO
/
tek-04
/
ddj0492.zip
/
WAVELET.ZIP
/
WAVEDEMO.MAN
< prev
next >
Wrap
Text File
|
1991-11-19
|
6KB
|
101 lines
WAVEDEMO User's Manual / Tip Sheet
Rather than a full fledged user's manual, I'm providing this
manual / tip sheet to help the user get started with experimenting
with WAVEDEMO.EXE. The commands in the program are more or less
self explanitory. Study the help panels before you proceed to use
WAVEDEMO.
Upon toggling out of the help panels (using the F1 key), the
control panel shows alpha and beta set to zero. With alpha and
beta equal, the Haar wavelet coefficients are generated. The
default sinewave of 312 data points in the input graph display
is shown. All other graphs are zero. The values of alpha and
beta are set by (A)lpha and (B)eta, respectively. Entering in a
new alpha or beta automatically recalculates the wavelet filter
coefficients. The range of alpha and beta is -π ≤ α,ß < π.
Changing the number of levels with (L)evels will clear any
existing transform coefficients in both the decomposition and
reconstruction displays. The (I)nput data and (R)ead coefficient
operation also have the same effect. The (Z)ero coefficients
operation will zero only those coefficients of the transform type
displayed. The coefficients of one transform type may be copied
to the other using the (C)opy operation.
The (I)nput and (O)utput operations read and write signal
data files, respectively. (I)nput loads the data into the input
graph display for wavelet decomposition. (O)utput stores the
reconstructed data in the output graph display. The name
requested at the prompt must follow standard DOS conventions. I
chose the .DAT extension for simplicity. The file SINE.DAT is a
sine wave of the same frequency as the default sine wave in
WAVEDEMO but with a length of 256 samples. AH.DAT is a 256-point
sampled voice waveform of the phone 'AH' as in "hot". Note: when
creating an input file the number of data points in the file must
be an integral multiple of the unit interval for the wavelet
transform. That is, if J levels of transform are to be created,
J
then the length of the unit interval is 2 data points.
The (W)rite and (R)ead operations store and load wavelet
transform coefficients files. (W)rite stores the wavelet
transform coefficients (Level 0 approximation and detail
coefficients) from the decomposition graph displays. The size of
the transform coefficient set and the values of alpha and beta
used to generate the wavelet filter coefficients are also stored.
(R)ead load the wavelet transform coefficient into the
reconstruction graph display. The alpha and beta values are
loaded and the wavelet filter coefficients are regenerated.
(P)rune uses ZeroTreeDetail to zero the upper levels of
detail coefficients. The prompt request the number of upper
levels to set to zero. Performing the inverse wavelet transform
upon the modified data yields a partial reconstruction of the
original signal. The partial reconstruction is based only on the
coarse details and the Level 0 approximation coefficients.
Experiments in data compression can be conducted by picking the
wavelet systems whose approximation (scaling function) level most
completely matches the original signal. The minimizing the mean
squared error (MSE) value can assist in this operation. To save
the partial reconstruction coefficients, use the (C)opy operation
to move them over to the decomposition display.
(V)iew coefficients allows for numerical examination of the
input and output signal data points and wavelet transform
coefficients for both decomposition and reconstruction. (V)iew
allows examination of the five additional coefficients at the end
of each coefficient and signal data array. These coefficients
handle overlap effect caused by the wavelet filters running off
the end of the data set.
The files PHI.COF and PSI.COF are wavelet coefficient files
that allow viewing of the wavelet scaling function, φ, and the
primary wavelet, [phi], respectively. Since phi and psi are
recursively defined, they can not be viewed as functions in the
traditional sense. They can viewed by performing the inverse fast
wavelet transform on an impulse (as single coefficient non-zero,
all others zero) in either the approximation or the detail
coefficients of level 0. This is equivalent to performing the
inverse Fourier transform on an impulse to obtain a pure sinewave
at a given frequency. PHI.COF contains the a single non-zero
coefficient in the approximation level, while PSI.COF contains the
a single non-zero coefficient in the detail level.
The values of alpha and beta must be set after (R)EADing the
appropriate file into the reconstruction display. After
(E)xecution of the inverse transform, phi or psi will be displayed
in the output graph display. Note that with wavelets of order six
(six coefficients), there is some distortion and/or truncation of
phi and psi. This is caused by data end effects and spreading of
the function outside the range of the graph. An inverse transform
larger than 256 point would solve these problems.
Hope you enjoy WAVEDEMO and find it educational! - Mac A. Cody