Description

convolution encoder using matrix multiplication

All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.

Related Documents

Share

Transcript

APPENDIX 9
Matrices and Polynomials
The Multiplication of Polynomials
Let
α
(
z
) =
α
0
+
α
1
z
+
α
2
z
2
+
···
α
p
z
p
and
y
(
z
) =
y
0
+
y
1
z
+
y
2
z
2
+
···
y
n
z
n
betwo polynomials of degrees
p
and
n
respectively. Then, their product
γ
(
z
) =
α
(
z
)
y
(
z
) is a polynomial of degree
p
+
n
of which the coeﬃcients comprisecombinations of the coeﬃcient of
α
(
z
) and
y
(
z
).A simple way of performing the multiplication is via a table of which themargins contain the elements of the two polynomials and in which the cellscontain their products. An example of such a table is given below:(1)
α
0
α
1
z α
2
y
0
α
0
y
0
α
1
y
0
z α
2
y
0
z
2
y
1
z α
0
y
1
z α
1
y
1
z
2
α
2
y
0
z
3
y
2
z
2
α
0
y
1
z
2
α
1
y
2
z
3
α
2
y
2
z
4
The product is formed by adding all of the elements of the cells. However, if theelements on the SW–NE diagonal are gathered together, then a power of theargument
z
can be factored from their sum and then the associated coeﬃcientis a coeﬃcient of the product polynomial.The following is an example from the table above:(3)
γ
0
+
γ
1
z
+
γ
2
z
2
+
γ
3
z
3
+
γ
4
z
4
=
α
0
y
0
+ (
α
0
y
1
+
α
1
y
0
)
z
+ (
α
0
y
2
+
α
1
y
1
+
α
2
y
0
+)
z
2
+ (
α
1
y
2
+
α
2
y
1
)
z
3
+
α
2
y
4
z
4
.
The coeﬃcients of the product polynomial can also be seen as the products of the convolutions of the sequences
{
α
0
,α
1
,α
2
...α
p
}
and
{
y
0
,y
1
,y
2
...y
n
}
.1
D.S.G. POLLOCK: ECONOMETRICS
The coeﬃcients of the product polynomials can also be generated by asimple multiplication of a matrix by a vector. Thus, from the example, weshould have(3)
γ
0
γ
1
γ
2
γ
2
γ
v
=
y
0
0 0
y
1
y
0
0
y
2
y
1
y
0
0
y
2
y
1
0 0
y
2
α
0
α
1
α
2
=
α
0
0 0
α
1
α
0
0
α
2
α
1
α
0
0
α
2
α
1
0 0
α
2
y
0
y
1
y
2
.
To form the elements of the product polynomial
γ
(
z
), powers of
z
may beassociated with elements of the matrices and the vectors of values indicated bythe subscripts.The argument
z
is usually described as an algebraic indeterminate. Itsplace can be taken by any of a wide variety of operators. Examples are providedby the diﬀerence operator and the lag operator that are deﬁned in respect of doubly-inﬁnite sequences.It is also possible to replace
z
by matrices. However, the fundamentaltheorem of algebra indicates that all polynomial equations must have solutionsthat lie in the complex plane. Therefore, it is customary, albeit unnecessary,to regard
z
as a complex number.
Polynomials with Matrix ArgumentsToeplitz Matrices
There are two matrix arguments of polynomials that are of particularinterest in time series analysis. The ﬁrst is the matrix lag operator. Theoperator of order
T
denoted by(4)
L
T
= [
e
1
,e
2
,...,e
T
−
1
,
0]is formed from the identity matrix
I
T
= [
e
0
,e
1
,...,e
T
−
1
] by deleting the leadingvector
e
0
and by appending a column of zeros to the end of the array. Ineﬀect,
L
T
is the matrix with units on the ﬁrst subdiagonal band and with zeroselsewhere. Likewise,
L
2
T
has units on the second subdiagonal band and withzeros elsewhere, whereas
L
T
−
1
T
has a single unit in the bottom left (i.e.
S–W
)corner, and
L
T
+
rT
= 0 for all
r
≥
0. In addition,
L
0
T
=
I
T
is identiﬁed with the2
POLYNOMIALS AND MATRICES
identity matrix of order
T
. The example of
L
4
is given below:(5)
L
04
=
1 0 0 00 1 0 00 0 1 00 0 0 1
, L
4
=
0 0 0 01 0 0 00 1 0 00 0 1 0
,L
24
=
0 0 0 00 0 0 01 0 0 00 1 0 0
, L
34
=
0 0 0 00 0 0 00 0 0 01 0 0 0
.
Putting
L
T
in place of the argument
z
of a polynomial in non-negative powerscreates a so-called Toeplitz banded lower-triangular matrix of order
T
. Anexample is provided by the quadratic polynomials
α
(
z
) =
α
0
+
α
1
z
+
α
2
z
2
and
y
(
z
) =
y
0
+
y
1
z
+
y
2
z
2
. Then, there is
α
(
L
3
)
y
(
L
3
) =
y
(
L
3
)
α
(
L
3
), which iswritten explicitly as(6)
α
0
0 0
α
1
α
0
0
α
2
α
1
α
0
y
0
0 0
y
1
y
0
0
y
2
y
1
y
0
=
y
0
0 0
y
1
y
0
0
y
2
y
1
y
0
α
0
0 0
α
1
α
0
0
α
2
α
1
α
0
.
The commutativity of the two matrices in multiplication reﬂects their polyno-mial nature. Such commutativity is available both for lower-triangular Toeplitzand for upper-triangular Toeplitz matrices, which correspond to polynomialsin negative powers of
z
.The commutativity is not available for mixtures of upper and lower trian-gular matrices; and, in this respect, the matrix algebra diﬀers from the corre-sponding polynomial algebra. An example is provided by the matrix version of the following polynomial identity:(7) (1
−
z
)(1
−
z
−
1
) = 2
z
0
−
(
z
+
z
−
1
) = (1
−
z
−
1
)(1
−
z
)Putting
L
T
in place of
z
in each of these expressions creates three diﬀerentmatrices. This can be illustrated with the case of
L
3
. Then, (1
−
z
)(1
−
z
−
1
)gives rise to(8)
1 0 0
−
1 1 00
−
1 1
1
−
1 00 1
−
10 0 1
=
1
−
1 0
−
1 2
−
10
−
1 2
,
whereas (1
−
z
−
1
)(1
−
z
) gives rise to(9)
1
−
1 00 1
−
10 0 1
1 0 0
−
1 1 00
−
1 1
=
2
−
1 0
−
1 2
−
10
−
1 1
.
3
D.S.G. POLLOCK: ECONOMETRICS
A Toeplitz matrix, in which each band contains only repetitions of the sameelement, is obtained from the remaining expression by replacing
z
by
L
3
in2
z
0
−
(
z
+
z
−
1
), wherein
z
0
is replaced by the identity matrix:(10)
2
−
1 0
−
1 2
−
10
−
1 2
=
−
1 1 0 00
−
1 1 00 0
−
1 1
−
1 0 01
−
1 00 1
−
10 0 1
.
Example.
It is straightforward to derive the dispersion matrices that are foundwithin the formulae for the ﬁnite-sample estimators from the corresponding au-tocovariance generating functions. Let
γ
(
z
) =
{
γ
0
+
γ
1
(
z
+
z
−
1
)+
γ
2
(
z
2
+
z
−
2
)+
···}
denote the autocovariance generating function of a stationary stochasticprocess. Then, the corresponding dispersion matrix for a sample of
T
consec-utive elements drawn from the process is(11) Γ =
γ
0
I
T
+
T
−
1
τ
=1
γ
τ
(
L
τ T
+
F
τ T
)
,
where
F
T
=
L
T
is in place of
z
−
1
. Since
L
T
and
F
T
are nilpotent of degree
T
,such that
L
qT
,F
qT
= 0 when
q
≥
T
, the index of summation has an upper limitof
T
−
1.
Circulant Matrices
In the second of the matrix representations, which is appropriate to afrequency-domain interpretation of ﬁltering, the argument
z
is replaced by thefull-rank circulant matrix(12)
K
T
= [
e
1
,e
2
,...,e
T
−
1
,e
0
]
,
which is obtained from the identity matrix
I
T
= [
e
0
,e
1
,...,e
T
−
1
] by displacingthe leading column to the end of the array. This is an orthonormal matrixof which the transpose is the inverse, such that
K
T
K
T
=
K
T
K
T
=
I
T
. Thepowers of the matrix form a
T
-periodic sequence such that
K
T
+
qT
=
K
qT
forall
q
. The periodicity of these powers is analogous to the periodicity of thepowers of the argument
z
= exp
{−
i2
π/T
}
, which is to be found in the Fouriertransform of a sequence of order
T
.4

Search

Similar documents

Tags

We Need Your Support

Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks