BINARY POLYNOMIAL TRANSFORMS FOR NONLINEAR SIGNAL
PROCESSING
Sos Agaian
Jaakko Astola, Karen Egiazarian, Rusen Oktem
Electrical Engineering Department
Signal Pro cessing Lab oratory
UniversityofTexas at San Antonio
Tamp ere UniversityofTechnology
San Antonio, Texas, USA
Tamp ere, FINLAND
ABSTRACT ing computers or systems suitable for simultaneous arith-
metical or logical op erations.
In this pap er weintro duce parametric binary Rademacher
functions of twotyp es. Based on them using di erent kinds
2. BINARY POLYNOMIAL FUNCTIONS AND
of arithmetical and logical op erations we generate a set of
MATRICES
binary p olynomial transforms (BPT). Some applications of
BPT in nonlinear ltering and in compression of binary
2.1. Rademacher Functions and Matrices
images are presented.
Let a and b b e arbitrary integers. We form the classes
r (t; a; b) and s (t; a; b) of real p erio dic functions in the
n n
1. INTRODUCTION
following way. Let
Sp ectral analysis is a p owerful to ol in communication, sig-
r (t; a; b) s (t; a; b) 1; (1)
0 0
nal/ image pro cessing and applied mathematics and there
are many reasons b ehind the usefulness of sp ectral analysis.
(
n
S
First, in many application s it is convenient to transform a
2 1
m 1 m
; + ); a; for t 2 [
n n
n+1
2 2
m=0
2
problem into another, easier problem; the sp ectral domain
r (t; a; b)=
S n
n+1
2 1
m 1 m+1
b; for t 2 [ + ; ):
n n is essentially a steady-state viewp oint; this not only gives
n+1
2 2
m=0 2
insight for analysis but also for design. Second, and more
(2)
imp ortant: the sp ectral approach allows us to treat entire
and
(
classes of signals whichhave similar prop erties in the sp ec-
1
a; for t 2 [0; );
n+1
2
tral domain; it has excellent prop erties such as fast algo-
1 1
s (t; a; b)= n =0;1;2; :::
n+1
b; for t 2 [ ; )
n
n+1
2 rithms, high energy compaction prop erty of the transform
2
0; otherwise for t 2 [0; 1);
data.
(3)
However, there are some problems in signal/ image pro-
Note that
cessing (esp ecially if the input data is binary) where they
1
cannot b e applied directly.
); a; for t 2 [0;
2
r (t; a; b)=s (t; a; b)= :
1 1
1
Wemay prop erly ask what is a basic transform for bi-
b; for t 2 [ ; 1)
2
nary signal/image pro cessing which plays the same role for
it as sp ectral transform for linear pro cessing.
DEFINITION 2.1 : The functions r (t; a; b) are called
n
Similarly wemay ask what is a basic transform for non-
Rademacher (a; b)-functions of I-type. The functions
linear signal/image pro cessing which plays the same role for
s (t; a; b) are called Rademacher (a; b)-functions of II-type.
n
it as sp ectral transform for linear pro cessing.
We will develop a compact representation of signals for
Note that when a = 1 and b = 1; the Rademacher
later use in digital logic and nonlinear signal analysis. Our
(a; b)-functions of I-typ e coincide with the Rademacher func-
approach will b e based on intro duction of parametric bi-
tions [5]. Let us consider some prop erties of the generalized
nary Rademacher functions (BRF); on generating a set of
Rademacher functions de ned ab ove.
transforms using Rademacher functions and di erent kinds
of arithmetical and logical op erations. Emphasis is placed
PROPERTY 2.1 : There are the following representa-
on the two classes of transforms, intro duced in this pap er:
tions of the Rademacher (a; b)-functions of the I-typ e:
transforms generated by BRFs using only one op eration
(arithmetical or logical) and transforms generated by BRFs
1) r (t; a; b)=a+(b a)c (t);t2[0; 1); (4)
n+1 n+1
using two op erations (arithmetical or logical). We present
where c are the co ecients from
n+1
the basic prop erties of these transforms.
1 1
This development is motivated also by p otential appli-
X X