4 Signal Space Representation of Waveforms

Signal space (or vector) representation of signals (waveforms) is a very ef-

fective and useful tool in the analysis of digitally modulated signals. In fact,
any set of signals is equivalent to a set of vectors.
4.1 Review of Vector Space Concepts
Denition 4.1. The inner product of two (potentially complex-valued)
n-dimensional vectors u and v is dened as
u, v = v
H
u
where ()
H
denotes the Hermitian transpose operator which performs
transposing operation and then conjugation.
4.2. Some properties of the inner product
u, v = v, u

z, w + w, z = 2 Re {z, w} = 2 Re {w, z}.

Denition 4.3. Two vectors u and v are orthogonal if u, v = 0.
More generally, a set of m vectors v
(k)
, 1 k m, are orthogonal if

v
(i)
, v
(j)
_
= 0 for all 1 i, j m, and i = j.
Denition 4.4. The norm of a vector v is denoted by v and is dened
as
v =
_
v, v
which in the n-dimensional space is simply the length of the vector.
Denition 4.5. A set of m vectors is said to be orthonormal if the vectors
are orthogonal and each vector has a unit norm.
4.6. Given two vectors u and v, we can decompose v into a sum of two
vectors, one a multiple of u and the other orthogonal to u.
(a) proj
u
(v) =
v,u
u,u
u is the orthogonal projection of v onto u.
(b) v proj
u
(v) is the component of v orthogonal to u.
13
Example 4.7. Let v = (5, 5)
T
and u = (0, 5)
T
.
4.8. A vector may also be represented as a linear combination of orthogonal
unit vectors or an orthonormal basis {e
i
, 1 i n}, i.e.,
v =
n

i=1
v
i
e
i
where, by denition, a unit vector has length unity and v
i
is the projection
of the vector v onto the unit vector e
i
, i.e.,
v
i
= v, e
i

4.9. Gram-Schmidt orthogonalization procedure for constructing a

set of orthonormal vectors from a set of n-dimensional vectors v
(i)
, 1 i
M.
(a) Arbitrarily select a vector from the set, say, v
(1)
. Let u
(1)
= v
(1)
.
Normalize its length to obtain the rst vector, say,
e
(1)
=
u
(1)
_
_
u
(1)
_
_
.
(b) Select an unselected vector from the set, say, v
(2)
.
Subtract the projection of v
(2)
onto u
(1)
:
u
(2)
= v
(2)
proj
u
(1)
_
v
(2)
_
= v
(2)

v
(2)
, u
(1)
_

u
(1)
, u
(1)
_
u
(1)
= v
(2)

_
v
(2)
, e
(1)
_
e
(1)
.
Then, we normalize the vector u
(2)
to unit length:
e
(2)
=
u
(2)
_
_
u
(2)
_
_
.
14
(c) Continue by selecting an unselected vector from the set, say, v
(3)
and subtract the projections of v
(3)
into u
(1)
and u
(2)
:
u
(3)
= v
(3)
proj
u
(1)
_
v
(3)
_
proj
u
(2)
_
v
(3)
_
= v
(3)

v
(3)
, u
(1)
_

u
(1)
, u
(1)
_
u
(1)

v
(3)
, u
(2)
_

u
(2)
, u
(2)
_
u
(2)
= v
(3)

_
v
(3)
, e
(1)
_
e
(1)

_
v
(3)
, e
(2)
_
e
(2)
.
Then, we normalize the vector u
(3)
to unit length:
e
(3)
=
u
(3)
_
_
u
(3)
_
_
.
By continuing this procedure, we construct a set of N orthonormal
vectors, where
N min(M, n).
Example 4.10. Consider four vectors: v
(1)
= (1, 1, 0)
T
, v
(2)
= (1, 1, 0)
T
,
v
(3)
= (1, 1, 1)
T
and v
(4)
= (1, 1, 1)
T
.
15
4.2 Signal Space Concepts
As in the case of vectors, we may develop a parallel treatment for a set of
signals.
Denition 4.11. The inner product of two generally complex-valued sig-
nals x
1
(t) and x
2
(t) is denoted by x
1
(t), x
2
(t) and dened by
x
1
(t), x
2
(t) =
_

x
1
(t)x

2
(t)dt.
The signals are orthogonal if their inner product is zero.
The norm of a signal is dened as
x(t) =
_
x(t), x(t) =
_
E
x
where E
x
is the energy in x(t).
A set of m signal is orthonormal if they are orthogonal and their norms
are all unity.
4.12. Similar to 4.9, the Gram-Schmidt orthogonalization procedure
can be used to construct a set of orthonormal waveforms from a set of nite
enerygy signal waveforms: x
i
(t), 1 i m.
Once we have constructed the set of, say N, orthonormal waveforms
{
k
(t)}, we can express the signals {s
i
(t)} as linear combinations of the

k
(t). Thus, we may write
x
i
(t) =

k
x
i,k

k
(t).
Based on the above expression, each signal may be represented by the vector
(or sequence)
x
(i)
= (x
i,1
, x
i,2
, . . . , x
i,N
)
T
,
or, equivalently, as a point in the N-dimensional (in general, complex) signal
space.
Denition 4.13. As discussed in 4.12, a set of M signals {x
i
(t)} can be
represented by a set of M vectors
_
x
(i)
_
in the N-dimensional space. The
corresponding set of vectors is called the signal space representation, or
constellation, of {x
i
(t)}.
16
If the original signals are real, then the corresponding vector repre-
sentations are in R
N
; and if the signals are complex, then the vector
representations are in C
N
.
Figure 2 demonstrates the process of obtaining the vector equivalent
from a signal (signal-to-vector mapping) and vice versa (vector-to-
signal mapping).
36
Digital Communications
r--s--'-'---< X)-------.
s2
1-----=-----..-( X )-------._
}--s(t)
L___s_:.:_N __ -1 X}-----_/
(a)
s(l)
(b)
FIGURE 2.2-2
Vector to signal (a), and signal to vector (b) mappings.
EXAMPLE 2.2-4. Let us obtain the vector representation of the four signals shown in
Figure 2.2-l(a) by using the orthonormal set of functions in Figure 2.2-1(b). Since
the dimensionality of the signal space is N = 3, each signal is described by three
components. The signal s1 (t) is characterized by the vectors 1
= ( ,.Ji, 0, 0)
1
Similarly,
the signals s2(t), s
3
(t), and s4(t) are characterized by the vectors s
2
= (0, ,.Ji, oy,
s 3
= ( ,.Ji, 0, 1)
1
, and s4 = ( -.Ji, 0, 1)
1
, respectively. These vectors are shown in
Figure 2.2-3. Their lengths are lls,ll = .Ji, lls2ll = .Ji, lls3ll = -J3, and Jls4ll = -J3,
and the corresponding signal energies are Ek = llsk 11
2
, k = 1, 2, 3, 4.
We have demonstrated that a set of M finite energy waveforms {sm(t)} can be rep-
resented by a weighted linear combination of orthonormal functions {n(t)} of dimen-
sionality N :::; M. The functions {n(t)} are obtained by applying the Gram-Schmidt
orthogonalization procedure on {sm(t)}. It should be emphasized, however, that the
functions {n(t)} obtained from the Gram-Schmidt procedure are not unique. If we
Figure 2: Vector to signal (a), and signal to vector (b) mappings.
17
Example 4.14. Consider the four waveforms illustrated in Figure 3.
Chapter Two: Deterministic and Random Signal Analysis
2
-I
I
V'll----------,
2
c/>2(1)
I
-..[2
I
21
-V'l
FIGURE 2.2-1
-I
(a)
-I
(b)
2
2
Gram-Schmidt orthogonalization of the signal {sm(t), m = 1, 2, 3, 4} and the corresponding
orthonormal basis.
The energy in the mth signal is simply the square of the length of the vector or, equiv-
alently, the square of the Euclidean distance from the origin to the point sm in the
N -dimensional space. Thus, any signal can be represented geometrically as a point in
the signal space spanned by the orthonormal functions {cf>n(t)}. From the orthonormality
of the basis it also follows that
(2.2-47)
This shows that the inner product of two sigmtls is equal to the inner product of the
corresponding vectors.
35
Figure 3: Four signals for orthogonalization in Example 4.14
4.15. From the orthonormality of the basis, we have
(a) the inner product of two signals is equal to the inner product of the
corresponding vectors:
x
i
(t), x
j
(t) =
_
x
(i)
, x
(j)
_
.
(b) E
x
(k) = x
k
(t)
2
=
_
_
x
(k)
_
_
2
.
4.16. It should be emphasized, however, that the functions {
k
(t)} ob-
tained from the Gram-Schmidt procedure are not unique. If we alter the
order in which the orthogonalization of the signals {x
i
(t)} is performed,
the orthonormal waveforms will be dierent and the corresponding vector
representation of the signals {x
i
(t)} will depend on the choice of the or-
thonormal functions {
k
(t)}. Nevertheless, the dimensionality of the signal
space N will not change, and the vectors x
(i)
will retain their geometric
conguration; i.e., their lengths and their inner products will be invariant
to the choice of the orthonormal functions {
k
(t)}.
18