Digital Signal Processing Reference
In-Depth Information
The rhs of (2.218) is recognized as the geometric series representation of the
time signal point c
n+m
so that
c
n+m
= (Φ
′
m
|Φ
′
n
) = S
n,m
.
(2.219)
Thus, in the basis{Υ
k
}(1≤k≤K), the time signal c
n+m
appears as the
matrix element of vectors |Φ
′
n
) and (Φ
′
m
|. Since, in general, c
n+m
= 0 for
arbitrary nonnegative integers n and m, functions |Φ
′
n
) and (Φ
′
m
| are not
orthogonal to each other and, as they stand, they do not form a basis set.
The matrix element between|Φ
′
n
) and (Φ
′
m
|from (2.219) represents, in fact,
the socalled overlap S
n,m
≡(Φ
′
m
|Φ
′
n
). This overlap S
n,m
is nothing but the
signal point c
n+m
by virtue of (2.219). Note that the set of K vectors from
(2.216) could be orthogonalized by using, e.g., the GramSchmidt or Lanczos
orthogonalization. However, orthogonalization is a convenience, but not a
necessity in the expansion methods.
Although they do not represent a basis, the K−dimensional vectors
{Φ
′
1
, Φ
′
2
, Φ
′
3
,..., Φ
′
K
}from (2.216) can be easily shown to be linearly indepen
dent
3
. By contrast, the augmented set of vectors{Φ
′
1
, Φ
′
2
, Φ
′
3
,..., Φ
′
K
, Φ
′
K+1
}
is linearly dependent and so is any other sequence with more added Φ
′
−terms,
{Φ
′
1
, Φ
′
2
, Φ
′
3
,..., Φ
′
K+m
}(m = 1, 2,...). This follows from the wellknown fact
that the necessary and su
cient condition for a set of vectors to be linearly
dependent is that the Gram determinant with the elements S
n,m
= (Φ
′
m
|Φ
′
n
)
is equal to zero
det{S
n,m
}= 0
(n,m = 1, 2,...,K + 1).
(2.220)
Due to (2.218), the lhs of (2.220) is equal to the Hankel determinant H
K+1
(c
0
).
Therefore (2.220) can be rewritten as H
K+1
(c
0
) = 0
det{S
n,m
}= H
K+1
(c
0
) = 0
(n,m = 1, 2,...,K + 1).
(2.221)
This coincides with the condition (2.204). Therefore, we see that the previous
checkings for the fulfillment of the condition (2.204), by which the exact num
ber K of the harmonic components in the time signal c
n
can be determined, is
entirely equivalent to verifying whether the vectors{Φ
′
1
, Φ
′
2
, Φ
′
3
,..., Φ
′
K
, Φ
′
K+1
}
are linearly dependent. Hence, this is yet another way of arriving to the same
criterion (2.204) for determining the exact number K of true resonances.
Recall that, by reference to (2.198), zerovalued amplitude d
K+m
for m > 1
represents a signature of spuriousness in the time signal c
n
with precisely K
3
Let S ={ξ
1
, ξ
2
, ξ
3
, ..., ξ
M
}be a finite set of M distinct elements in a linear spaceM. The
set S is said to be linearly dependent if there exists a final set of scalars{a
1
, a
2
, a
3
, ..., a
M
}
(not all zero) such that
P
m=1
a
m
ξ
m
= 0. Conversely, the set S will be linearly independent,
if it is not linearly dependent. In other words, when a set S is linearly independent, then
for any choices of distinct elements{ξ
1
, ξ
2
, ξ
3
, ..., ξ
M
}in S and scalars{a
1
, a
2
, a
3
, ..., a
M
},
the relation
P
m=1
a
m
ξ
m
= 0 could be fulfilled if and only if all the M values of the scalars
{a
m
}
m=1
are simultaneously equal to zero, a
m
= 0 (m = 1, 2, 3, ..., M ).
Search WWH ::
Custom Search