ECE 461 Fall 2006

September 20, 2006

Digital Modulation
After possible source and error control encoding, we have a sequence m
n
of message symbols
to be transmitted on the channel. The message symbols are assumed to come from a nite
alphabet, say 0, 1, . . . , M 1. In the simplest case of binary signaling, M = 2. Each symbol
in the sequence is assigned to one of M waveforms s
0
(t), . . . , s
M1
(t).
Memoryless modulation versus modulation with memory. If the symbol to waveform mapping is
xed from one interval to the next, i.e., m s
m
(t), then the modulation is memoryless. If the
mapping from symbol to waveform in the n-th symbol interval depends on previously transmitted
symbols (or waveforms) then the modulation is said to have memory.
For memoryless modulation, to send the sequence m
n
of symbols at the rate of 1/T
s
symbols
per second, we transmit the signal
s(t) =

n
s
mn
(t nT
s
) . (1)
Linear versus nonlinear modulation. A digital modulation scheme is said to be linear if we
can write the mapping from the sequence of symbols m
n
to the transmitted signal s(t) as
concatenation of a mapping from the sequence m
n
to a complex sequence c
n
, followed by a
linear mapping from c
n
to s(t). Otherwise the modulation is nonlinear.
Linear Memoryless Modulation
In this case, the mapping from symbols to waveforms can be written in complex baseband as:
s
m
(t) =

c
m
e
jm
g(t) , m = 1, 2, . . . , M , (2)
where g(t) is a real-valued, unit energy, pulse shaping waveform.
The signal s
m
(t) can be represented by a point in the complex plane, i.e., the signal space
corresponding to a symbol interval is a 1-d (complex) space with basis function g(t).
PSfrag replacements
I I
Q Q
s
m

c
m

m
s
1
s
2
s
3
s
4
s
M
Representation of s
m
(t) Signal constellation
c _V.V. Veeravalli, 2006 1
In real passband,
s
m
(t) = Re[

2s
m
(t) e
j2fct
] =

2c
m
cos(2f
c
t +
m
)g(t) . (3)
As we saw in class, the signal energy is the same in both the real passband and complex baseband
domains and equals c
m
.
The average symbol energy for the constellation is given by
c
s
=
1
M
M

m=1
c
m
. (4)
The average bit energy for the constellation (assuming that M = 2

, for some integer ) is given

by
c
b
=
c
s
log
2
M
=
c
s

. (5)
The distance between signals s
k
and s
m
is d
k,m
= |s
k
s
m
|, and the minimum distance is given
by
d
min
= min
k,m
d
k,m
. (6)
A measure of goodness of the constellation is the ratio
=
d
2
min
c
b
. (7)
Note that is independent of scaling of the constellation.
Some commonly used signal constellations are:
Pulse Amplitude Modulation (PAM). Information only in amplitude:

m
= 0 and

c
m
= (2m 1 M)
d
2
, m = 1, 2, . . . , M . (8)
We can compute as a function of M. For example, = 4 for M = 2.
Phase Modulation or Phase Shift Keying (PSK). Information only in phase:

m
=
2m
M
and c
m
= c , m = 1, 2, . . . , M . (9)
We showed in class that
d
min
=

2c

1 cos
2
M

= = 2 log
2
M

1 cos
2
M

For QPSK, = 4 (as in BPSK).

Quadrature Amplitude Modulation (QAM). Information in phase and amplitude. We can
design constellations to maximize for a given M. Rectangular constellations are convenient
for demodulation. For rectangular 16-QAM, = 1.6.
c _V.V. Veeravalli, 2006 2
Orthogonal Memoryless Modulation
Here the signal set is given by
s
m
(t) =

c g
m
(t), m = 1, 2, . . . , M (10)
where g
m
(t) are (possibly complex) unit energy signals, i.e., |g
m
(t)| = 1.
The correlation between signals s
k
(t) and s
m
(t) is given by:

km
=
s
k
(t), s
m
(t))
c
= g
k
(t), g
m
(t)) (11)
There are two kinds of orthogonality:
Orthogonality only in the real component of the correlation, i.e. Re
km
= 0, for k ,= m.
This form of orthogonality is enough for coherent demodulation.
Complete orthogonality, i.e.,
km
= 0, for k ,= m. This is required for noncoherent demodula-
tion.
Examples of orthogonal signal sets
Separation in time:
g
m
(t) = g (t (m 1)T
s
/M) (12)
where g(t) is such that g(t kT
s
/M), g(t mT
s
/M)) =
km
. For example, g(t) = p
Ts/M
(t), a
rectangular pulse of width T
s
/M.
This signal set is completely orthogonal. We can also create a signal set of twice the size which
satises orthogonality only in the real component of the correlation by adding jg
m
(t) to the
above signal set as we saw in class.
Separation in frequency:
g
m
(t) = e
j2(m1)
f
t
p
Ts
(t) (13)
It is easy to show that

km
= sinc[T
s
(k m)
f
] e
jTs(km)
f
(14)
and that
Re
km
= sinc[2T
s
(k m)
f
] . (15)
Thus the smallest value of
f
such that
km
= 0, for k ,= m, is 1/T
s
, and such that Re
km
=
0, for k ,= m, is 1/2T
s
.
Separation in time and frequency: One way to do this is to pick g
m
(t) to be the Walsh-
Hadamard functions on [0, T
s
] as we saw in class.
c _V.V. Veeravalli, 2006 3