Databases Reference
In-Depth Information
canonical correlation matrix K , we find
IK
KI
D0
0D
IK
KI
D H
0D T
0
H
H
E
=
=
=
E
(3.101)
DD H
D D T
DKD T
=
.
(3.102)
D KD H
This shows that D is unitary and DKD T
=
K . The latter can be true only if D ij =
0
whenever k i =
k j . Therefore, D is diagonal and unitary if all circularity coefficients are
distinct. Since K is real, the corresponding diagonal entries of all nonzero circularity
coefficients are actually
1. On the other hand, components with identical circularity
coefficient cannot be separated.
±
x 2 ] T . The first component x 1 is the signal-
space representation of a QPSK signal with amplitude 2 and phase offset
Example 3.5. Consider a source x =
[ x 1 ,
π/
8, i.e.,
2e j π/ 8
2je j π/ 8
x 1 ∈{±
. The second component x 2 , independent of x 1 , is the signal-
space representation of a BPSK signal with amplitude 1 and phase offset
, ±
}
π/
4, i.e.,
e j π/ 4
x 1 ∈{±
}
. Hence,
40
01
00
0j
R xx =
R xx =
and
.
In order to take x into canonical coordinates, we use the strong uncorrelating transform
01
10
2 0
01
01
1
2
F xx R 1 / 2
A xx =
=
=
.
xx
0
We see that F xx is a permutation matrix, R 1 / 2
xx is diagonal, and the product of the two is
monomial. The circularity coefficients are k 1 =
0. Note that the circularity
coefficients carry no information about the amplitude or phase of the two signals.
Now consider the linear mixture y
1 and k 2 =
=
Mx with
j
1
M
=
.
2
j1
+
j
and R yy =
MR xx M T , we compute the SVD of the coherence
MR xx M H
With R yy =
R yy R T / 2
R 1 / 2
yy
UKV H . The unitary Takagi factor F yy is then obtained
matrix C
=
=
yy
U ( V H U ) 1 / 2 . The circularity coefficients of y are the same as those of x .
In order to take y into canonical coordinates, we use the strong uncorrelating transform
(rounded to four decimals)
as F yy =
0
.
7071
2
.
1213j
0
.
7071
+
0
.
7071j
F yy R 1 / 2
A yy =
=
.
yy
0
.
7045
0
.
0605j
0
.
3825
0
.
3220j
We see that
0
0
.
7071
0
.
7071j
A yy M
=
0
.
3825
0
.
3220j
0
 
Search WWH ::




Custom Search