Ramiro Jordan
COPYRIGHT
Copyright 1992, University of New Mexico. All rights
reserved.
*********************************************************************************
Documentation for avs module dminorm
INPUT
image pointer to VIFF structure containing image
data to be processed.
order order of the linear combiner.
freq number of computed spectral values.
kdim used to determine the dimension of the noise
subspace. If it is a positive, non-zero
integer, then it represents the actual value
of the noise subspace dimension. If it is
zero, then dminorm will use the Akaike (AIC)
criteria to determine the noise subspace
dimension. If it is -1, then the minimum
description length (MDL) criteria will be
used.
auto_type type of auto correlation to be used. 0
specifies a biased estimate, 1 specifies an
unbiased estimate, 2 specifies an FFT based
estimate.
arith_type type of arithmetic to be used on complex
data. 0 specifies scalar arithmetic and 1
specifies vector arithmetic.
psd_type specifies the centering of the power spectral
density estimate. 0 specifies not centered,
1 specifies centered.
procdir process direction: 0 indicated vector
oriented processing, 1 indicates band
oriented processing.
OUTPUT
image pointer to VIFF structure containing image
data after processing.
Return Value: 1 on success, 0 on failure.
DESCRIPTION
An implementation of the Minimum Norm for spectral estima-
tion. Minimum Norm is a spectral estimation technique based
on eigendecomposition of the autocorrelation matrix of a
sampled random process. The Minimum Norm algorithm attempts
to eliminate the effects of spurious zeros, which result in
spurious peaks in the power spectrum, by pushing them inside
the unit circle.
The algorithm involves generating estimates of the eigen-
values and eigenvectors of a Toelitz and Hermitian auto-
correlation matrix estimate. The eigenvectors are then
sorted by ascending order of their corresponding eigen-
values. The eigenvectors corresponding to the k smallest,
approximately equal eigenvalues are selected. k is the
dimension of the noise subspace, and can be specified by the
user or may be computed using the Akaike (AIC) or Minimum
Description Length (MDL) information theoretic criteria.
The Minimum Norm power spectrum is computed by solving an
ordinary linear prediction problem.
For a more theoretical discussion of the Minimum Norm algo-
rithm see: Optimum Signal Processing by Sophocles
Orfranidis, McGraw-Hill, 1988; or Eigenvector-Based
Parametric Modeling of Time-Domain Signals: Application to
Nuclear Magnetic Resonance Spectroscopy by Glen Abousleman
(Master's thesis, University of New Mexico, Dept. of EECE,
1990).
dminorm does not work on explicit location data and will
return an error when such a file is encountered. No output
file will be created. dminorm works only on VFF_TYP_FLOAT
or VFF_TYP_COMPLEX data files.
Input/Output Files
The input file must be of type VFF_TYP_FLOAT or
VFF_TYP_COMPLEX. The output file returned will be of size
specified by the user. If the data is of type
VFF_TYP_COMPLEX and complex aithmetic is selected then the
output file will be of type VFF_TYP_FLOAT.
Process direction
Data can be processed with either band or vector orientation
(-d option). The default data processing direction is in
the vector direction (-d = 0). This means that data stored
in multiband format will be processed as a set pixel loca-
tion. The direction of processing can be changed so that
each band is processed as a signal (instead of each vector
being a signal) by setting the -d option to 1.
SEE ALSO
dminorm(1), intro(3), vipl(3), verror(3), vutils(3)
RESTRICTIONS
works only on VFF_TYP_COMPLEX and VFF_TYP_FLOAT.
dminorm does not work on explicit location data and will
return an error when such a data file is encountered.
AUTHOR
Jeremy Worley, Ramiro Jordan, Glen Abousleman
COPYRIGHT
Copyright 1992, University of New Mexico. All rights
reserved.
*********************************************************************************
Documentation for avs module dmusic
INPUT
image pointer to VIFF structure containing image
data to be processed.
order order of the linear combiner.
freq number of computed spectral values.
kdim used to determine the dimension of the noise
subspace. If it is a positive, non-zero
integer, then it represents the actual value
of the noise subspace dimension. If it is
zero, then dmusic will use the Akaike (AIC)
criteria to determine the noise subspace
dimension. If it is -1, then the minimum
description length (MDL) criteria will be
used.
auto_type type of auto correlation to be used. 0
specifies a biased estimate, 1 specifies an
unbiased estimate, 2 specifies an FFT based
estimate.
arith_type type of arithmetic to be used on complex
data. 0 specifies scalar arithmetic and 1
specifies vector arithmetic.
psd_type specifies the centering of the power spectral
density estimate. 0 specifies not centered,
1 specifies centered.
procdir process direction: 0 indicated vector
oriented processing, 1 indicates band
oriented processing.
OUTPUT
image pointer to VIFF structure containing image
data after processing.
Return Value: 1 on success, 0 on failure.
DESCRIPTION
An implementation of the MUSIC (Multiple Signal Classifica-
tion) for spectral estimation. MUSIC is a spectral estima-
tion technique based on eigendecomposition of the autocorre-
lation matrix of a sampled random process.
The process involves generating estimates of the eigenvalues
and eigenvectors of a Toelitz and Hermitian autocorrelation
matrix estimate. The eigenvectors are then sorted by
ascending order of their corresponding eigenvalues. The
eigenvectors corresponding to the k smallest, approximately
equal eigenvalues are selected. k is the dimension of the
noise subspace, and can be specified by the user or may be
computed using the Akaike (AIC) or Minimum Description
Length (MDL) information theoretic criteria.
k power spectrum estimates corresponding to the k selected
eigenvectors are computed and averaged to form the MUSIC
power spectrum estimate.
For a more theoretical discussion of the MUSIC algorithm
see: Optimum Signal Processing by Sophocles Orfranidis,
McGraw-Hill, 1988; or Eigenvector-Based Parametric Modeling
of Time-Domain Signals: Application to Nuclear Magnetic
Resonance Spectroscopy by Glen Abousleman (Master's thesis,
University of New Mexico, Dept. of EECE, 1990).
dmusic does not work on explicit location data and will
return an error when such a file is encountered. No output
file will be created. dmusic works only on VFF_TYP_FLOAT or
VFF_TYP_COMPLEX data files.
Input/Output Files
The input file must be of type VFF_TYP_FLOAT or
VFF_TYP_COMPLEX. The output file returned will be of size
specified by the user. If the data is of type
VFF_TYP_COMPLEX and complex aithmetic is selected then the
output file will be of type VFF_TYP_FLOAT.
Process direction
Data can be processed with either band or vector orientation
(-d option). The default data processing direction is in
the vector direction (-d = 0). This means that data stored
in multiband format will be processed as a set pixel loca-
tion. The direction of processing can be changed so that
each band is processed as a signal (instead of each vector
being a signal) by setting the -d option to 1.
SEE ALSO
dmusic(1), intro(3), vipl(3), verror(3), vutils(3)
RESTRICTIONS
works only on VFF_TYP_COMPLEX and VFF_TYP_FLOAT.
dmusic does not work on explicit location data and will
return an error when such a data file is encountered.
AUTHOR
Jeremy Worley, Ramiro Jordan, Glen Abousleman
COPYRIGHT
Copyright 1992, University of New Mexico. All rights
reserved.
*********************************************************************************