Proakis-50210 book August 3, 2001 13:39

Section 7.1 Geometric Representation of Signal Waveforms 341

signals is then described and their performance on the AWGN channel is evaluated in
terms of the probability of error. The various modulation methods are compared on
the basis of their performance characteristics, their bandwidth requirements and their
implementation complexity.

7.1 GEOMETRIC REPRESENTATION OF SIGNAL WAVEFORMS

The Gram-Schmidt orthogonalization procedure may be used to construct an orthonor-
mal basis for a set of signals. In this section, we develop a geometric representation of
signal waveforms as points in a signal space. Such a representation provides a compact
characterization of signal sets for transmitting information over a channel and simplifies
the analysis of their performance. Using vector representation, waveform communi-
cation channels are represented by vector channels. This reduces the complexity of
analysis considerably.
Suppose we have a set of M signal waveforms sm (t), 1 ≤ m ≤ M which are to
be used for transmitting information over a communication channel. From the set
of M waveforms, we first construct a set of N ≤ M orthonormal waveforms, where
N is the dimension of the signal space. For this purpose, we use the Gram-Schmidt
orthogonalization procedure.

Gram-Schmidt Orthogonalization Procedure. We begin with the first

waveform s1 (t), which is assumed to have energy E 1 . The first waveform of the or-
thonormal set is constructed simply as
s1 (t)
ψ1 (t) = √ (7.1.1)
E1
Thus, ψ1 (t) is simply s1 (t) normalized to unit energy.
The second waveform is constructed from s2 (t) by first computing the projection
of s2 (t) onto ψ1 (t), which is
! ∞
c21 = s2 (t)ψ1 (t) dt (7.1.2)
−∞

Then, c21 ψ1 (t) is subtracted from s2 (t) to yield

d2 (t) = s2 (t) − c21 ψ1 (t) (7.1.3)

Now, d2 (t) is orthogonal to ψ1 (t), but it does not possess unit energy. If E2 denotes the
energy in d2 (t), then the energy normalized waveform that is orthogonal to ψ1 (t) is
d2 (t)
ψ2 (t) = √ (7.1.4)
E2
! ∞
E2 = d22 (t) dt (7.1.5)
−∞
Proakis-50210 book August 3, 2001 13:39

342 Digital Transmission through Gaussian Noise Channel Chapter 7

In general, the orthogonalization of the kth function leads to

dk (t)
ψk (t) = √ (7.1.6)
Ek
where
k−1
"
dk (t) = sk (t) − cki ψi (t) (7.1.7)
i=1
! ∞
Ek = dk2 (t) dt (7.1.8)
−∞
and
! ∞
cki = sk (t)ψi (t) dt, i = 1, 2, . . . , k − 1 (7.1.9)
−∞
Thus, the orthogonalization process is continued until all the M signal waveforms
{sm (t)} have been exhausted and N ≤ M orthonormal waveforms have been constructed.
The N orthonormal waveforms {ψn (t)} form a basis in the N -dimensional signal space.
The dimensionality N of the signal space will be equal to M if all the M signal
waveforms are linearly independent; i.e., if none of the signal waveforms is a linear
combination of the other signal waveforms.
Example 7.1.1
Let us apply the Gram-Schmidt procedure to the set of four waveforms illustrated√ in
Figure 7.1(a). The waveform s1 (t) has energy E1 = 2, so that ψ1 (t) = s1 (t)/ 2. Next
we observe
√ that c21√= 0, so that ψ1 (t) and s2 (t) are orthogonal. Therefore, ψ2 (t) =
s2 (t)/ E√2 = s2 (t)/ 2. To obtain ψ3 (t), we compute c31 and c32 , which are c31 = 0 and
c32 = − 2. Hence,
√
d3 (t) = s3 (t) + 2ψ2 (t)
Since
√ d3 (t) has unit energy, it follows that ψ3 (t) = d3 (t). Finally, we find that c41 =
2, c42 = 0, c43 = 1. Hence,
√
d4 (t) = s4 (t) − 2ψ1 (t) − ψ3 (t) = 0
Thus, s4 (t) is a linear combination of ψ1 (t) and ψ3 (t) and, consequently, the dimen-
sionality of the signal set is N = 3. The functions ψ1 (t), ψ2 (t), and ψ3 (t) are shown in
Figure 7.1(b)
Once we have constructed the set of orthogonal waveforms {ψn (t)}, we can
express the M signals {sm (t)} as exact linear combinations of the {ψn (t)}. Hence, we
may write
N
"
sm (t) = smn ψn (t), m = 1, 2, . . . , M (7.1.10)
n=1
where
! ∞
smn = sm (t)ψn (t) dt
−∞
Proakis-50210 book August 3, 2001 13:39

Section 7.1 Geometric Representation of Signal Waveforms 343

s1(t) s3(t)

1 1

0 2 t 0 1 3 t
!1

s3(t) s4(t)

1 1

0 1 2 t 0 3 t
!1

(a) Original signal set

"1(t) "3(t)
1
1
2

0 2 t 2 3 t

"2(t)
1
2

1 2 t
1
!
2
(b) Orthonormal waveforms

Figure 7.1 Application of Gram-Schmidt orthogonalization procedure to signals

{si (t)}.

and
! ∞ N
sm2 (t) dt = 2
"
Em = smn (7.1.11)
−∞ n=1

Based on the expression in Equation (7.1.7), each signal waveform may be rep-
resented by the vector
sm = (sm1 , sm2 , . . . , sm N ) (7.1.12)
or equivalently, as a point in N -dimensional signal space with coordinates {smi , i =
1, 2, . . . , N }. The energy of the mth signal waveform is simply the square of the length
of the vector or, equivalently, the square of the Euclidean distance from the origin to
the point in the N -dimensional space. We can also show that the inner product of two
Proakis-50210 book August 3, 2001 13:39

344 Digital Transmission through Gaussian Noise Channel Chapter 7

signals is equal to the inner product of their vector representations; i.e.,

! ∞
sm (t)sn (t) dt = sm · sn (7.1.13)
−∞

Thus, any N -dimensional signal can be represented geometrically as a point in the

signal space spanned by the N orthonormal functions {ψn (t)}.
Example 7.1.2
Let us determine the vector representation of the four signals shown in Figure 7.1(a) by
using the orthonormal set of functions in Figure 7.1(b). Since the dimensionality of the
signal space is N = 3, each signal is described by three components, which are obtained
by projecting each of the four signal waveforms
√ on the three orthonormal
√ basis functions
√
ψ1 (t),√ψ2 (t), ψ3 (t). Thus, we obtain s1 = ( 2, 0, 0), s2 = (0, 2, 0), s3 = (0, − 2, 1),
s4 = ( 2, 0, 1). These signal vectors are shown in Figure 7.2.

Finally, we should observe that the set of basis functions {ψn (t)} obtained by the
Gram-Schmidt procedure is not unique. For example, another set of basis functions
that span the three-dimensional space is shown in Figure 7.3. For this basis, the signal

"2(t)

s2

2
2
s1 "1(t)

3
3 s4
s3 Figure 7.2 Signal vectors corresponding
"3(t) to the signals si (t), i = 1, 2, 3, 4.

"1(t) "3(t)

1 1

0 1 t 0 2 3 t

"2(t)

0 1 2 t

Figure 7.3 Alternate set of basis functions.

Proakis-50210 book August 3, 2001 13:39

Section 7.2 Pulse Amplitude Modulation 345

vectors are s1 = (1, 1, 0), s2 = (1, −1, 0), s3 = (−1, 1, 1), and s4 = (1, 1, 1). The reader
should note that the change in the basis functions does not change the dimensionality
of the space N , the lengths (energies) of the signal vectors, or the inner product of any
two vectors.
Although the Gram-Schmidt orthogonalization procedure is guaranteed to gener-
ate an orthonormal basis for representation of the signal set, in many cases it is simpler
to use a method based on inspection to generate the orthonormal basis. We explore this
method in the problems at the end of this chapter.

7.2 PULSE AMPLITUDE MODULATION

In pulse amplitude modulation (PAM), the information is conveyed by the amplitude
of the transmitted signal. Let us first consider PAM signals that are appropriate for
baseband channels.

Baseband Signals. Binary PAM is the simplest digital modulation method.

In binary PAM, the information bit 1 may be represented by a pulse of amplitude A
and the information bit 0 is represented by a pulse of amplitude −A, as shown in
Figure 7.4. This type of signaling is also called binary antipodal signaling. Pulses are
transmitted at a bit rate Rb = 1/Tb bits/sec, where Tb is called the bit interval. Although
the pulses are shown as rectangular, in practical systems, the rise time and decay time
are nonzero and the pulses are generally smoother. The pulse shape determines the
spectral characteristics of the transmitted signal as described in Chapter 8.
The generalization of PAM to nonbinary (M-ary) pulse transmission is relatively
straightforward. Instead of transmitting one bit at a time, the binary information se-
quence is subdivided into blocks of k bits, called symbols, and each block, or symbol, is
represented by one of M = 2k pulse amplitude values. Thus with k = 2, we have M = 4
pulse amplitude values. Figure 7.5 illustrates the PAM signals for k = 2, M = 4. Note
that when the bit rate Rb is fixed, the symbol interval is
k
T = = kTb (7.2.1)
Rb
as shown in Figure 7.6.

S2(t)
S1(t)

A
0 Tb t

0 Tb t !A
Figure 7.4 Binary PAM signals.