Principles of Digital Communication

Chapter 5
Robert G. Gallager
October 10, 2004
ii
Contents
5 Channels, PAM, and QAM 1
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
5.2 Pulse amplitude modulation (PAM) . . . . . . . . . . . . . . . . . . . . . . . . . 3
5.2.1 Binary PAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
5.2.2 Multi-level PAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
5.2.3 PAM modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
5.2.4 PAM demodulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
5.3 The Nyquist criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
5.3.1 Band edge symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
5.3.2 Choosing p(tkT) as an orthonormal set . . . . . . . . . . . . . . . . . 10
5.3.3 Relation between PAM and analog source coding . . . . . . . . . . . . . 11
5.4 Modulation: baseband to passband and back . . . . . . . . . . . . . . . . . . . . 11
5.4.1 Double-sideband amplitude modulation . . . . . . . . . . . . . . . . . . . 11
5.4.2 Quadrature amplitude modulation (QAM) . . . . . . . . . . . . . . . . . . 13
5.4.3 QAM signal set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.4.4 QAM baseband modulation and demodulation . . . . . . . . . . . . . . . 15
5.4.5 QAM: baseband to passband and back . . . . . . . . . . . . . . . . . . . . 16
5.4.6 Implementation of QAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.5 Degrees of freedom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.5.1 Distance and orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.6 Carrier recovery in QAM systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.6.1 Tracking phase in the presence of noise . . . . . . . . . . . . . . . . . . . 21
5.6.2 Large phase errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.E Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
iii
iv CONTENTS
Chapter 5
Channels, PAM, and QAM
5.1 Introduction
In chapter 1, we discussed the reasons for separating a communication system into source coding
and channel coding layers connected by a binary interface (see Figure 5.1. Chapters 2 to 4
discussed source coding and we now begin to consider the channel coding layer. The channel
encoder converts a sequence of bits into a waveform for transmission over an analog channel,
preferably as eciently as possible i.e., at as high a rate as possible. The channel decoder
performs the inverse operation, recovering the transmitted bits with an acceptably low error
rate given the presence of noise introduced by the channel.
Input
E
Source
Encoder
E
Channel
Encoder
c
Channel
'
Source
Decoder
'
Channel
Decoder
Binary
Interface
'
Output
Figure 5.1: Separation of source and channel coding.
Channel coding is often viewed as being separated into two layers, one called digital coding and
the other called modulation as shown in Figure 5.2. Usage is by no means uniform here. The
word modulation alone is often used as a synonym for both the function of discrete coding and
modulation. These two functions are usually designed and implemented in a single physical unit,
sometimes functionally separated and sometimes functionally inter-twined into a single function
called coded modulation. In learning about channel encoding, however, it will be useful at rst
to separate these two functions.
We have shown the channel above as a one way device going from source to destination. Usually,
however, communication goes both ways, so that a physical location can send data to another
1
2 CHAPTER 5. CHANNELS, PAM, AND QAM
E
Discrete
Encoder
E
Modulation
c
Channel
'
Discrete
Decoder
'
Demodulation
Binary
Interface
Signal
constellation
Waveform
'
Figure 5.2: Separation of encoding into discrete coding and modulation.
location and also receive data from that remote location. A physical device that both encodes
data going out over a channel and also decodes oppositely directed data coming in from the
channel is called a modem (for modulator/demodulator). The communication in the two direc-
tions can usually be viewed as being approximately independent. The decoding at one location
is of course matched to the type of encoding done at the other, but we can separate this from
the encoding and decoding in the opposite direction; thus, as in Figures 5.1 and 5.2, we consider
communication in only a single direction.
There is an interesting analogy between analog source coding and channel coding. With analog
source coding, we started with an analog waveform, then represented the waveform by a sequence
of real or complex coecients in an orthogonal expansion. This sequence was then quantized into
a sequence of signals from a discrete alphabet, and nally the signals were encoded into a binary
sequence. With channel encoding in this analogy, we start at the opposite end by encoding a
sequence of bits into a sequence of signals
1
from a nite alphabet called a signal constellation.
The elements of this constellation are real or complex points in one or several dimensions. This
sequence of signal points is then mapped into a waveform by a process analogous to the inverse
of converting waveforms into sequences.
Initially we ignore digital encoding and look only at simple ways of mapping bits into signal
constellations and then mapping these signal constellations into waveforms. At the destination,
the received waveform (which is altered by noise from the transmitted waveform) is converted
back to a sequence of signal constellation points, and from there back to binary digits. We will
shortly nd that the noise plays a crucial role in this process, but for the present we ignore the
noise and consider only the process of converting bits to signal constellations to waveforms.
1
The words symbol and signal are essentially synonyms in the communication literature. We distinguish
them here by using symbol to refer to one of a distinguishable collection of objects whose particular names are
unimportant (such as binary digits) and signal to refer to one of a nite set of real numbers, complex numbers,
or vectors.
5.2. PULSE AMPLITUDE MODULATION (PAM) 3
5.2 Pulse amplitude modulation (PAM)
5.2.1 Binary PAM
The simplest type of modulation is binary pulse amplitude modulation
2
(binary PAM). With
binary PAM, binary digits enter the modulator at a xed rate 1/T. Each binary digit, 0 or 1
is then mapped into the signal constellation consisting of 1 and +1. It makes no dierence
whether the mapping is 0 1 and 1 1 or, alternatively, 0 1 and 1 1. The former
is more common, primarily because modulo two addition of binary digits then corresponds to
multiplication of the real numbers 1 and +1.
When we use binary digits, we usually do not think of them as real numbers; we could equally
well call them (heads, tails), or (Alice, Bob). For signals, however, the properties of the real (or
complex) numbers are important. In particular, these signals will be used as scalars (or perhaps
vectors) in generating waveforms.
Example 1 As a particularly simple example, a sequence u
1
, u
2
, . . . , of signals, each taking the
value 1 or +1, is generated at T-spaced intervals; each signal u
k
is multiplied by sinc(
t
T
k) to
obtain the channel waveform
u(t) =

k
u
k
sinc
_
t
T
k
_
. (5.1)
At the receiver, assuming no delay, no attenuation, and no noise, the waveform u(t) can be
sampled at time kT for each integer k, and the samples u(kT) = u
k
are converted back to
binary.
In communication, the time reference at the receiver is conventionally delayed by a xed amount
relative to that at the transmitter. This delay is equal to the sum of the propagation delay and
varous lter delays. With this convention, the no-delay assumption above makes sense. This also
explains why communication engineers often model lters between transmitter and receiver as
being non-causal. Finding the xed delay above is a signicant problem called timing recovery.
We ignore this problem for the time being.
The overall delay in a communication system is often important. This is true for voice communi-
cation and is particularly true, for example, when a communication system is used in a real time
control system. The above xed delay between transmitter and receiver time reference is only
one component of this delay. The discrete encoder, the discrete decoder, the processing, and the
demodulation add additional delay to the overall delay on the channel side of the binary inter-
face. Source encoding and decoding contribute additional delay to the overall communication
system.
Delay is one of the parameters of the quality of service requirement for a communication system.
Because of source/channel separation, the source delay and channel delay are often regarded as
separate requirements. Usually, however, the ltering delays associated with simple modulation
2
The terminology comes from analog amplitude modulation, where a baseband waveform is modulated up
to some passband for communication. For digital communication, the more interesting problem is turning a bit
stream into a waveform at baseband. We discuss modulating a baseband waveform to passband in subsequent
sections.
4 CHAPTER 5. CHANNELS, PAM, AND QAM
schemes are unimportant relative to other delays and the propagation delay cannot be changed;
thus we will not consider delay in any detail here.
We also usually measure amplitudes on a dierent scale at transmitter and receiver. The actual
power attenuation suered in transmission is a product of amplier gain, antenna coupling
losses, antenna directional gain, propagation losses, etc. The process of nding all these gains
and losses (and perhaps changing them) is called the link budget. Such gains and losses are
invariably calculated in decibels (dB). The number of decibels corresponding to a power gain
is dened to be 10 log
10
. Thus power losses correspond to negative dB and power gains to
positive dB. The use of a logarithmic measure of gain allows the various components of gain to
be added rather than multiplied.
The use of decibels rather than some other logarithmic measure such as natural logs or logs
to the base 2 is motivated by the ease of doing rough mental calculations. A factor of 2 is
10 log
10
2 = 3.010 dB, approximated as 3 dB. Thus 4 = 2
2
is 6 dB and 8 is 9 dB. Since 10
is 10 dB, we also see that 5 is 10/2 or 7 dB. We can just as easily see that 20 is 13 dB and so
forth.
It is important to remember that the gains expressed in dB are power gains. Thus if there is a
multiplicative gain of g in a signal, this corresponds to a gain g
2
in power, which corresponds
to 20 log
10
g dB.
The following table summarizes some dB power conversions which are well known to communi-
cation engineers. Note the ease of doing approximate mental calculations using integer valued
dB.
dB (to 1 place) dB (to 2 places)
1 0 0.00
1.25 1 0.97
2 3 3.01
2.5 4 3.98
e 4.3 4.34
3 4.8 4.77
5 4.97
4 6 6.02
5 7 6.99
8 9 9.03
10 10 10.00
The link budget in a communication system is largely separable from other issues, so we usually
normalize the amplitude scale at the transmitter to that at the receiver. We concentrate on the
received power level because that is where noise also appears. Additive noise in a channel is a
fundamental limitation to communication and arises from thermal eects and received radiation
at the receiver. We will later nd that signal to noise ratio, i.e., the ratio of the received signal
power to the noise power (properly interpreted) is the important factor here rather than either
the received signal power or noise power alone.
5.2.2 Multi-level PAM
Assume that an input data sequence arrives at a rate of R b/s and is converted, b bits at a time,
into a sequence of signals chosen from a signal set (alphabet, constellation) / = a
1
, a
2
, . . . a
M

5.2. PULSE AMPLITUDE MODULATION (PAM) 5

of size M = [/[ = 2
b
. The signal rate is thus R
s
= R/b signals per second (signals/s, or Hz),
and the signal interval is T = 1/R
s
= b/R sec.
A standard M-PAM signal set (see Figure 5.3) consists of M = 2
b
d-spaced real numbers located
symmetrically about the origin; i.e.,
/ =
d(M1)
2
, . . . ,
d
2
,
d
2
, . . . ,
d(M1)
2
.
In other words, the signal points are the same as the representation levels of a symmetric M-level
uniform scalar quantizer.
a
1
a
2
a
3
a
4
a
5
a
6
a
7
a
8
d
E '
0
Figure 5.3: An 8-PAM signal set.
A sequence u
1
, u
2
, . . . , of signals, each taking a value from the set / = a
1
, . . . , a
M
, is generated
at T-spaced intervals. As before, each signal u
k
can be multiplied by sinc(
t
T
k) to form a channel
waveform u(t) =

k
u
k
sinc(
t
T
k).
If the incoming bits are independent equiprobable chance variables (which is well approximated
by eective source coding), then each signal u
k
is a sample value of a random variable U
k
that
is uniformly distributed over the alphabet /. Also the sequence U
1
, U
2
, . . . , is iid. As derived
in the exercises, the mean squared signal value, or energy per signal E
s
= E[U
2
k
] is then given
by
E
s
=
d
2
(M
2
1)
12
=
d
2
(2
2b
1)
12
.
For example, for M = 2, 4 and 8, we have E
s
= d
2
/4, 5d
2
/4 and 21d
2
/4, respectively.
For b greater than 2, 2
2b
1 is approximately 2
2b
, so we see that each increase in b by 1
increases E
s
by a factor of 4, which is 6 dB. Thus increasing the rate R by increasing b requires
impractically large energy for large b.
The spacing d between signal points could be decreased so as to increase b (and thus R) while
holding E
s
constant. However, when we later study noise and signal detection, we shall nd
that errors in multi-level PAM systems occur primarily between neighboring signal points and
that this error probability increases rapidly as d is reduced. Thus d is essentially determined by
noise considerations.
The relation between error probability and signal point spacing also helps explain why multi-
level PAM systems almost invariably use a standard M-PAM signal set. Because of the rapid
change of error probability with distance between nearest neighbors, we will see later that error
probability is dominated by the points that are closest together. Thus, if we constrain the signal
points to be separated by some minimal distance d between points, it can be seen that the
minimum energy E
s
for a given number of points M is achieved by the standard M-PAM set.
6 CHAPTER 5. CHANNELS, PAM, AND QAM
5.2.3 PAM modulation
A PAM modulator, for a given signal set /, is dened by the signal interval T and a real L
2
waveform p(t) as follows: the discrete sequence x
k
of points from / modulates the amplitudes
of a sequence of time shifts p(t kT) of the basic waveform (pulse) p(t) to create the
transmitted waveform u(t):
u(t) =

k
u
k
p(t kT). (5.2)
It may be helpful to visualize p(t) as the impulse response of a linear time-invariant lter. Then
u(t) is the response to a sequence of T-spaced impulses u
k
(tkT). Since the time-reference
at the receiver is dierent from that at the transmitter, however, we need not require that p(t)
be a causal lter. We will nd that the problem of choosing T and p(t) is largely separable from
that of choosing the signal set /. We will also nd that the choice of p(t) is the more challenging
and interesting problem.
We could choose p(t) to be sinc(
t
T
), but we will generalize this to a larger set of potential
waveforms. If we visualize generating the signal u
k
at time kT, then u(t) cannot be generated
in real time using the sinc function, since it starts at time . The sinc function can of course
be truncated in time, thus yielding a reasonable approximation with a delay of several signal
intervals between transmitter and receiver. Even assuming that a delay of several signal intervals
is not important, however, the sinc function drops o very slowly in time and thus requires a
large delay to achieve a reasonable approximation.
Rather than pursuing the relationship between delay and truncation errors for the sinc function,
we will look at the fundamental problem of choosing a modulation pulse p(t). This problem was
rst analyzed in 1928 in a classic paper by Harry Nyquist
3
. Before looking at Nyquists results,
however, we look at what happens at the demodulator.
5.2.4 PAM demodulation
For the time being we ignore the channel noise. Assume that the time reference and the am-
plitude scaling at the receiver have been selected so that the received waveform is the same as
the transmitted waveform u(t). This also assumes that the transmitted waveform has not been
corrupted in any way other than delay and attenuation.
The problem at the demodulator is then to retrieve the transmitted signals u
1
, u
2
, . . . from the
received waveform u(t) =

u
k
p(tkT). A PAM demodulator is determined by a signal interval
T (the same as at the modulator) and a real L
2
waveform q(t). The demodulator rst lters
the received waveform using an impulse response q(t) and then samples the output at T-spaced
sample times. That is, the received ltered waveform is
r(t) =
_

u()q(t ) d (5.3)
and the received samples are r(T), r(2T), . . . , .
3
Nyquists paper, Certain Topics in Telegraph Transmission theory, Trans. AIEE 47: pp. 627-644, 1928 is
perhaps the most signicant (and lucid) communication paper before Shannons work in 1948. The modern IEEE
was formed in 1963 as the combination of the AIEE (American Institute of Electrical Engineers) and the IRE
(Institute of Radio Engineers).
5.3. THE NYQUIST CRITERION 7
Our objective is to choose p(t) and q(t) so that r(kT) = u
k
for each k. If this objective is met,
we say that the PAM system has no intersymbol interference. If, instead, r(kT) is some non-
trivial linear combination of dierent input signals, then we say that intersymbol interference
has occured. We would also like these lters to be reasonable approximations to baseband lters
of bandwidth W = 1/(2T), and we would like them to decay relatively quickly in time. The
reader should verify that p(t) = q(t) = sinc(
t
T
) is one solution, except for the slow time-decay.
This problem of choosing lters to avoid intersymbol interference at rst appears to be somewhat
articial. First we have restricted the receiver to be a lter followed by a sampler. Exercise 2,
however, essentially shows that if the receiver is restricted to linear operations on the received
waveform, then there is no real loss of generality in restricting the operation to be a lter followed
by a T-spaced sampler.
The second articiality is that we are ignoring the noise, which turns out to be the fundamental
limitation on the bit rate. Thus we are ignoring a crucial piece of the problem. The reason we are
doing this is that there is a simple and elegant solution to the problem of avoiding intersymbol
interference, and we will be able to use this solution later when we bring noise into the picture.
Recall that u(t) =

k
u
k
p(t kT) and thus from (5.3)
r(t) =
_

k
u
k
p( kT)q(t ) d. (5.4)
Let g(t) be the convolution g(t) = p(t) q(t) =
_
p()q(t ) d and assume
4
that g(t) is L
2
.
We can then simplify (5.4) to
r(t) =

k
u
k
g(t kT). (5.5)
This should not be surprising. The lters p(t) and q(t) are in cascade with each other. Thus
r(t) does not depend on which part of the ltering is done in one and which in the other; it is
only the convolution g(t) that determines r(t). Later, when we add channel noise, we will nd
that the individual choice of p(t) and q(t) becomes important.
We see from (5.5) that intersymbol interference will be avoided, i.e., r(kT) will equal u
k
for each
integer k and each sequence u
1
, u
2
, . . . , if the waveform g(t) has the property that g(0) = 1 and
g(kT) = 0 for each nonzero integer k. Waveforms with this property are said to be ideal Nyquist
or, more precisely, ideal Nyquist with interval T. We have already seen that sinc(
t
T
) is ideal
Nyquist with interval T, but we are now interested in nding approximately baseband-limited
functions that are ideal Nyquist but decay in time more rapidly than the sinc function.
As a simple example, the function rect(t/T) is ideal Nyquist with interval T. This function
decays in time quickly, but is not close to being baseband limited.
5.3 The Nyquist criterion
The ideal Nyquist property is determined solely by the T-spaced samples of the waveform g(t).
This suggests that the results about aliasing should be relevant. Let s(t) be the baseband-limited
4
By looking at the frequency domain, it is not dicult to construct a g(t) of innite energy from L2 p(t) and
q(t). When we study noise, however, we nd that there is no point in constructing such a g(t), so we ignore the
possibility.
8 CHAPTER 5. CHANNELS, PAM, AND QAM
waveform generated by the samples of g(t), i.e.,
s(t) =

k
g(kT) sinc(
t
T
k). (5.6)
We see that g(t) is ideal Nyquist with interval T if and only if s(t) = sinc(
t
T
), and thus if and
only if
s(f) = T rect(fT). (5.7)
From the aliasing theorem,
s(f) = l.i.m.

m
g(f +
m
T
) rect(fT). (5.8)
Combining (5.7) and (5.8), we get the Nyquist theorem,
Theorem 5.3.1 (Nyquist criterion) Let g(f) be L
2
and satisfy the condition
lim
|f|
g(f)[f[
1+
= 0 for some > 0. Then the inverse transform, g(t), of g(f) is
ideal Nyquist with interval T if and only if g(f) satises the Nyquist criterion for T, dened as
5
l.i.m.

m
g(f + m/T) rect(fT) = T rect(fT). (5.9)
Proof: From the aliasing theorem, the baseband approximation s(t) in (5.6) converges pointwise
and is L
2
. Similarly, the Fourier transform s(f) satises (5.8). If g(t) is ideal Nyquist, then
s(t) = sinc(
t
T
). This implies that s(f) is L
2
equivalent to T rect(fT), which in turn implies
(5.9). Conversely, satisfaction of the Nyquist criterion (5.9) implies that s(f) = T rect(fT).
This implies s(t) = sinc(
t
T
) implying that g(t) is ideal Nyquist.
There are many pulse shapes that satisfy the Nyquist criterion, and the question is how to
choose among them. We have seen that the sinc function appears to be desirable in minimizing
the bandwidth used by the waveform, but is awkward in terms of its slow decay in time. In
fact, any step discontinuity in g(f) causes g(t) to decay only as 1/t with increasing t. Similarly
a slope discontinuity in g(f) causes a 1/t
2
decay in g(t). Thus it is desirable to select g(t) to be
smooth but also to have small bandwidth.
5.3.1 Band edge symmetry
The nominal or Nyquist band associated with a PAM pulse g(t) with signal interval T is dened
to be W = 1/(2T). The actual baseband bandwidth B is dened as the smallest number B such
that g(f) = 0 for [f[ > B.
We discuss the various reasons why B is limited later, but for now simply note that if W is
much smaller than B, then W can be increased, thus increasing the rate R
s
at which signals
can be transmitted. It can also be seen from the Nyquist criterion that B W is necessary to
avoid intersymbol interference. Thus we want to choose g(t) in such a way that B exceeds W
by a relatively small amount. In particular, we now focus on the case where W B < 2W.
5
It can be seen that

m
g(f +m/T) is periodic and thus the rect(fT) could be essentially omitted from both
sides of (5.9). Doing this, however, would make the limit in the mean meaningless and would also complicate the
intuitive understanding of the theorem.
5.3. THE NYQUIST CRITERION 9
The assumption B < 2W means that g(f) = 0 for [f[ 2W. Thus for 0 f W, g(f +2mW)
can be non-zero only for m = 0 and m = 1. Thus the Nyquist criterion (5.9) in this positive
frequency interval becomes
g(f) + g(f 2W) = T for 0 f W. (5.10)
Since p(t) and q(t) are real, g(t) is also real, so g(f2W) = g

(2Wf). Substituting this in

(5.10) and letting = fW, (5.10) becomes
T g(W+) = g

(W). (5.11)
This is sketched and interpreted in Figure 5.4. The gure assumes the typical situation in which
g(f) is real. In the general case, the gure illustrates the real part of g(f) and the imaginary
part satises g(W+) = g(W).
g(W+)

B
T g(W)
g(f)

%
0
W B
T
f
Figure 5.4: Band edge symmetry illustrated for real g(f): For each , 0W,
g(W+) = T g(W). The portion of the curve for f W, rotated by 180
o
around
the point (W, T/2), is equal to the portion of the curve for f W.
Figure 5.4 makes it particularly clear that B must satisfy B W to avoid intersymbol inter-
ference. We then see that the choice of g(f) involves a tradeo between making g(f) smooth,
so as to avoid a slow decay in g(t), and reducing the excess of B over the Nyquist bandwidth
W. This excess is expressed as a rollo factor
6
, dened to be (B/W) 1, usually expressed as
a percentage. Thus g(f) in the gure has about a 30% rollo.
PAM lters in practice often have raised cosine transforms. The raised cosine frequency function,
for any given rollo between 0 and 1, is dened by
g

(f) =
_

_
T, 0 [f[
1
2T
;
T cos
2
_
T
2
([f[
1
2T
)

,
1
2T
[f[
1+
2T
;
0, [f[
1+
2T
.
(5.12)
The inverse transform of g

(f) can be shown to be (see Exercise 5.6)

g

(t) = sinc(
t
T
)
cos(t/T)
1 4
2
t
2
/T
2
, (5.13)
6
The requirement for a small rollo actually arises from a requirement on the transmitted pulse p(t), i.e., on
the bandwidth of the actual transmitted channel waveform, rather than on the cascade g(t) = p(t) q(t). The
tacit assumption here is that p(f) = 0 when g(f) = 0. One reason for this is that it is silly to transmit energy in
a part of the spectrum that is going to be completely ltered out at the receiver. We see later that p(f) and q(f)
are usually chosen to have the same magnitude, ensuring that p(f) and g(f) have the same rollo.
10 CHAPTER 5. CHANNELS, PAM, AND QAM
which decays asymptotically as 1/t
3
, compared to 1/t for sinc(
t
T
). In particular, for a rollo
= 1, g(f) is non-zero from 2W = 1/T to 2W = 1/T and g
1
(t) has most of its energy
between T and T. Rollos as sharp as 515% are used in current practice, leading to more
energy outside [T, T], but still decaying much faster than the sinc function.
The motivation for the raised cosine shape is that we would like most of the energy in g(f) to be
in the band [W, W], but at the same time we want g(f) to be smooth. It can be seen that the
raised cosine simply rounds o the step discontinuity in rect(
f
2W
) in such a way as to maintain
the Nyquist criterion while making g(f) continuous with a continuous derivitive.
5.3.2 Choosing p(tkT) as an orthonormal set
We have now seen that desirable choices among ideal Nyquist pulses are those for which g(f)
is a compromise between smoothness and limited rollo. As illustrated in gure 5.4, it is not
a serious additional constraint to restrict g(f) to be real and positive (why go negative or
imaginary in making a smooth transition from T to 0?). After the restriction g(f) 0, however,
there is still the question of choosing the transmit lter p(t) and the receive lter q(t) subject to
p(f) q(f) = g(f). When we study noise later, we will nd that there are important advantages
in choosing p(f) and q(f) to have the same magnitude
7
, i.e.,
[ p(f)[ = [ q(f)[ =
_
g(f) . (5.14)
The phase of p(f) can be chosen in an arbitrary way, but the requirement that p(f) q(f) = g(f)
0 means that q(f) = p

(f). In addition, if p(t) is real then p(f) = p

(f), which determines the

phase for negative f but leaves the phase for positive f arbitrary. It is convenient here, however,
to be slightly more general and allow p(t) to be complex. We will prove the following important
theorem:
Theorem 5.3.2 (Orthonormal shifts) Let g(f) 0 satisfy the Nyquist criterion for T and
let [ p(f)[ =
_
g(f). Then p(tkT); k Z is a set of orthogonal functions. Conversely, if
p(tkT); k Z is a set of orthogonal functions, then [ p(f)[
2
satises the Nyquist criterion.
Proof: Note that with q(t) = p

(t),
g(kT) =
_

p()q(kT ) d =
_

p()p

( kT) d. (5.15)
Letting p
k
be the function p(t kT), we see that the above equation is g(kT) = p
0
, p
k
). From
the Nyquist criterion theorem, g(t) is an ideal Nyquist pulse and thus p
0
, p
k
) is 0 for k ,= 0
and 1 for k = 0. Also by shifting the variable of integration in (5.15), we see that p
j
, p
k+j
) = 0
for k ,= 0 and 1 for k = 0. In other words, p
k
is an orthonormal set. Conversely, (5.15) is
satised for g(f) = [ p(f)[
2
. If p
k
is an orthonormal set, then g(kT) = 0 for k ,= 0 and 1 for
k = 0. Thus g(t) is ideal Nyquist and g(f) satises the Nyquist criterion.
Given this orthonormal shift property for p(t), the PAM transmitted waveform u(t) =

k
u
k
p(tkT) is simply an orthonormal expansion. Retrieving the coecient u
k
then cor-
responds to projecting u(t) onto the one dimensional subspace spanned by p
k
. Note that this
7
A function p(t) satisfying (5.14) is often called square root of Nyquist, although it is the magnitude of the
transform that is the square root of the transform of an ideal Nyquist pulse.
5.4. MODULATION: BASEBAND TO PASSBAND AND BACK 11
projection is accomplished by ltering u(t) by q(t) and then sampling at time kT. The lter
q(t) is called the matched lter to p(t). We discuss these lters later when noise is introduced
into the picture.
Note that we have restricted the pulse p(t) to have unit energy. There is no loss of generality
here, since the input signals u
k
can be scaled arbitrarily and there is no point in having an
arbitrary scale factor in both places.
5.3.3 Relation between PAM and analog source coding
The main emphasis in PAM modulation has been that of converting a sequence of T-spaced
signals into a waveform. Similarly, the rst part of analog source coding is often to convert
a waveform into a T-spaced sequence of samples. The major dierence is that with PAM
modulation, we have control over the PAM pulse p(t) and thus some control over the class of
waveforms. With source coding, we are stuck with whatever class of waveforms describes the
source of interest.
For both systems we can dene the nominal bandwidth W = 1/(2T) and dene B as the actual
baseband bandwidth of the waveforms. In the case of source coding, we have seen that B W is
a necessary condition for the sampling appoximation

k
u(kT) sinc(
t
T
k) to perfectly recreate
the waveform u(t). We used the aliasing theorem and the T-spaced sinc weighted sinusoid
expansion to analyze the squared error if B > W.
For PAM, on the other hand, the necessary condition for the PAM demodulator to recreate
the initial PAM sequence is B W. We saw that with B > W, we can use aliasing to our
advantage, creating an aggregate pulse g(t) that is ideal Nyquist. There is considerable choice in
such a pulse, and it is chosen by using contributions from both f < W and f > W. Finally we
saw that the transmission pulse p(t) for PAM can be chosen so that its T-spaced shifts form an
orthonormal set. The sinc functions have this property, but many other waveforms with slightly
greater bandwidth have the same property but decay much faster with t.
5.4 Modulation: baseband to passband and back
The discussion of PAM was focused on transmitting a waveform u(t) of T-spaced signals in
the baseband frequency range [f[ B
u
where B
u
is slightly larger than the Nyquist bandwidth
W
u
=
1
2T
. Most communication systems, however, operate in some passband of frequencies both
appropriate to the physical medium and permitted by regulatory agencies. Thus the waveforms
we have been considering usually require translation up to an appropriate passband before
transmission. Typically, many such transmissions are multiplexed together, each in a separate
passband, creating the need to keep each waveform in a tightly constrained frequency passband.
5.4.1 Double-sideband amplitude modulation
Modulation of a PAM waveform to passband can essentially be accomplished by multiplying
the baseband PAM waveform u(t) by a sinuosoid, say e
2ifct
. This does not quite work since it
results in a complex waveform, whereas only real waveforms can actually be transmitted. Thus
12 CHAPTER 5. CHANNELS, PAM, AND QAM
we also multiply u(t) by the complex conjugate of e
2ifct
, resulting in the modulated waveform
x(t) = u(t)[e
2ifct
+ e
2ifct
] = 2u(t) cos(2f
c
t), (5.16)
x(f) = u(f f
c
) + u(f + f
c
). (5.17)
As illustrated in Figure 5.5, u(t) is both translated up in frequency by f
c
and also translated
down by f
c
. Since x(t) must be real, x(f) = x

(f), and the negative frequencies cannot be

avoided. More to the point, the negative frequency portion of the baseband waveform u(t) is
now a positive frequency portion of x(t), going from f
c
B
u
to f
c
. Thus the baseband bandwidth
B
u
has now become a passband bandwidth of 2B
u
.

i
i
ii u(f)
f B
u
1
T
E '

i
i
ii x(f)
f
f
c
1
T
E '
fcBu fc+Bu

i
i
ii x(f)
f
c
1
T
E '
0
Figure 5.5: Frequency domain representation of a baseband waveform u(t) shifted up to a
passband around the carrier f
c
. Note that the baseband bandwidth B
u
of u has become the
passband bandwidth 2B
u
of x.
In the communication eld, the bandwidth of a system is universally dened as the range of
positive frequencies used in transmission. Since transmitted waveforms are real, the negative
frequency part of those waveforms is determined by the positive part and is not counted. This
is consistent with our earlier baseband usage, where B
u
is the bandwidth of the baseband
waveform u(t) in Figure 5.5, and with our new usage for passband waveforms where B = 2B
u
is the bandwidth of x(f).
The passband modulation scheme described by (5.16) is called double-sideband amplitude mod-
ulation. The terminology comes not from the negative frequency band around f
c
and the
positive band around f
c
, but rather from viewing [f
c
B
u
, f
c
+B
u
] as two sidebands, the upper,
[f
c
, f
c
+B
u
], coming from the positive frequency components of u(t) and the lower, [f
c
B
u
, f
c
]
from its negative components. Since u(t) is real, these two bands are redundant and either could
be reconstructed from the other.
Double-sideband modulation is quite wasteful of bandwidth since half of the band is redundant.
Redundancy is often useful for added protection against noise, but such redundancy is usually
better achieved through digital coding.
The simplest and most widely employed solution for using this wasted bandwidth
8
is quadrature
amplitude modulation (QAM) which is described in the next subsection. We will see that PAM at
passband is appropriately viewed as a special case of QAM, and thus we discuss the demodulation
of PAM from passband to baseband at the same time as the demodulation of QAM.
8
An alternate approach is single-sideband modulation. Here either the positive or negative sideband of a
double-sideband waveform is ltered out, thus reducing the transmitted bandwidth by a factor of 2. This used to
be quite popular for analog communication but is harder to implement for digital communication than QAM.
5.4. MODULATION: BASEBAND TO PASSBAND AND BACK 13
5.4.2 Quadrature amplitude modulation (QAM)
QAM is very similar to PAM except that with QAM the baseband waveform u(t) is chosen to
be complex. The complex QAM waveform u(t) is then shifted up to passband as u(t)e
2ifct
.
This waveform is complex and is converted into a real waveform for transmission by adding its
complex conjugate. The resulting real passband waveform is then
x(t) = u(t)e
2ifct
+ u

(t)e
2ifct
. (5.18)
Note that the passband waveform for PAM in (5.16) is a special case of this in which u(t) is real.
The passband waveform x(t) in (5.18) can also be written in the following equivalent ways:
x(t) = 2'u(t)e
2ifct
(5.19)
= 2'u(t) cos(2f
c
t) 2u(t) sin(2f
c
t) . (5.20)
The factor of 2 in (5.19) and (5.20) is an arbitrary scale factor. Some authors leave it out,
(thus requiring a factor of 1/2 in (5.18)) and others replace it by

2 (requiring a factor of
1/

2 in (5.18)). This scale factor causes additional confusion when we look at the energy in
the waveforms. With the scaling here, |x|
2
= 2|u|
2
. Using the scale factor

2 solves this
problem, but introduces many other problems, not least of which is an extraordinary number of

2s in equations. At one level, scaling is a trivial matter, but since it is done inconsistently in
the literature, we should get used to handling it in a consistent way. One intuitive advantage
of the convention here, as illustrated in Figure 5.5 is that the positive frequency part of x(t) is
simply u(t) shifted up by f
c
.
We now provide a more detailed explanation of QAM systems. A QAM modulator has three
parts (see gure 5.6). The rst is to map the incoming binary digits into complex signals. The
second is to map the sequence of complex signals into a baseband waveform, and the third is
to modulate the baseband waveform to passband. The demodulator, not surprisingly, performs
the inverse of these operations in reverse order, rst mapping the received bandpass waveform
into a baseband waveform, then recovering the sequence of signals, and nally recovering the
binary digits. We discuss each of these operations in turn.
Input
Binary
E
Signal
encoder
E
Baseband
modulator
E
Baseband to
passband
c
Channel
Baseband
Demodulator
'
Passband to
baseband Output
Binary
'
Signal
decoder
' '
Figure 5.6: QAM modulator and demodulator.
14 CHAPTER 5. CHANNELS, PAM, AND QAM
5.4.3 QAM signal set
The input data sequence arrives at a rate of R b/s and is converted, b bits at a time, into
a sequence of complex signals u
k
chosen from a signal set (alphabet, constellation) / of size
M = [/[ = 2
b
. The signal rate is thus R
s
= R/b signals per second, and the signal interval is
T = 1/R
s
= b/R sec.
In the case of QAM, the transmitted signals u
k
are complex numbers u
k
C, rather than real
numbers. Alternatively, we may think of each signal as a real 2-tuple in R
2
.
A standard (M

)-QAM signal set, where M = (M

)
2
is the Cartesian product of two
M

-PAM sets; i.e.,

/ = (a

+ ia

) [ a

, a

,
where
/

= d(M

1)/2, . . . , d/2, d/2, . . . , d(M

1)/2.
The signal set / thus consists of a square array of M = (M

)
2
= 2
b
signal points located
symmetrically about the origin, as illustrated below for M = 16.
t t t t
t t t t
t t t t
t t t t
dE '
The minimum distance between two-dimensional points is denoted by d. Also the average
energy per two-dimensional signal, which is denoted E
s
, is simply twice the average energy per
dimension:
E
s
=
d
2
[(M

)
2
1]
6
=
d
2
[M 1]
6
.
In the case of QAM there are clearly many ways to arrange the signal points other than on a
square grid as above. For example, in an M-PSK (phase-shift keyed) signal set, the signal points
consist of M equally spaced points on a circle centered on the origin. Thus 4-PSK = 4-QAM.
For large M it can be seen that the signal points become very close to each other on a circle so
that PSK is rarely used for large M. On the other hand, PSK has some practical advantages
because of the uniform signal magnitudes.
As with PAM, the probability of detection error is primarily a function of the minimum distance
d. We will also see that E
s
is linear in the signal power of the passband waveform. In wireless
systems the signal power is limited both to conserve battery power and to meet regulatory
requirements. In wired systems, the power is limited both to avoid crosstalk between adjacent
wires and to avoid non-linear eects.
For all of these reasons, it is desirable to choose signal constellations that approximately minimize
E
s
for a given d and M. One simple result here is that a hexagonal grid of signal points achieves
smaller E
s
than a square grid for large M and xed minimum distance. Unfortunately, nding
5.4. MODULATION: BASEBAND TO PASSBAND AND BACK 15
the optimal signal set to minimize E
s
as a function of M is a messy and ugly problem, and the
minima have few interesting properties or symmetries. We will not spend further time on this
other than a few exercises and will usually simply assume a standard (M

)-QAM signal
set.
5.4.4 QAM baseband modulation and demodulation
A QAM baseband modulator is determined by the signal interval T and a complex L
2
waveform
p(t). The discrete-time complex sequence u
k
of signal points modulates the amplitudes of a
sequence of time shifts p(tkT) of the basic pulse p(t) to create a complex transmitted signal
u(t) as follows:
u(t) =

kZ
u
k
p(tkT). (5.21)
As in the PAM case, we could choose p(t) to be sinc(
t
T
), but for the same reasons as before,
we usually prefer p(t) to decay faster than the sinc function with increasing t, and thus for p(f)
to be a continuous function that goes to zero rapidly but not instantaneously as f increases
beyond 1/(2T). As with PAM, we dene
1
2T
to be the nominal baseband bandwidth of the
QAM modulator.
We assume for the moment that the process of conversion to passband, channel transmission,
and conversion back to baseband, is ideal, recreating the baseband modulator output u(t) at the
input to the baseband demodulator. The baseband demodulator is determined by the interval
T (the same as at the modulator) and a complex L
2
waveform q(t). The demodulator lters
u(t) by q(t) and samples the output at T-spaced sample times. Denoting the ltered output by
r(t) =
_

u()q(t ) d,
we see that the received samples are r(T), r(2T), . . . . Note that this is the same as the PAM
demodulator except that real signals and waveforms have been replaced by complex signals and
waveforms. As before, the output r(t) can be represented as
r(t) =

k
u
k
g(t kT),
where g(t) is the convolution of p(t) and q(t). As before, r(kT) = u
k
if g(t) is ideal Nyquist,
namely if g(0) = 1 and g(kT) = 0 for all non-zero integer k. The Nyquist criterion is valid
whether or not g(t) is real.
The band-edge symmetry condition of Figure 5.4, on the other hand, does require g(t) to be
real. For the same reasons as with PAM, however, g(t) is usually chosen to be real anyway (even
if p(t) and q(t) are complex), and g(f) is usually chosen to be both real and positive and often
to be a raised cosine function.
Finally, as discussed with PAM, p(f) is usually chosen to satisfy [ p(f)[ =
_
g(f). With this
choice, the set of time shifts p(tkT) form an orthonormal set of functions. With this choice
also, the baseband bandwidth of u(t), p(t), and g(t) are all the same. Each has a nominal
baseband bandwidth given by
1
2T
and each has an actual baseband bandwidth that exceeds
1
2T
by some small rollo factor.
16 CHAPTER 5. CHANNELS, PAM, AND QAM
In summary, QAM baseband modulation is virtually the same as PAM baseband modulation.
The signal set for QAM is of course complex, and the modulating pulse p(t) can be complex,
but the Nyquist results about avoiding intersymbol interference are unchanged.
5.4.5 QAM: baseband to passband and back
Next we discuss modulating the complex QAM baseband waveform u(t) to the passband wave-
form x(t). Alternative expressions for x(t) are given by (5.18), (5.19). and (5.20) and the
frequency representation is illustrated in Figure 5.5.
As with PAM, u(t) has a nominal baseband bandwidth W
u
=
1
2T
. The actual baseband band-
width B
u
exceeds W
u
by some small rollo factor. The corresponding passband waveform x(t)
has a nominal passband bandwidth W = 2W
u
=
1
T
and an actual passband bandwidth B = 2B
u
.
We will assume in everything to follow that B/2 < f
c
. This implies that u(t)e
2ifct
is actually
constrained to positive frequencies, and thus that the Fourier transform u(ff
c
) does not overlap
with u(f+f
c
).
We can view the modulation to passband as a two step process. First u(t) is translated up in
frequency by an amount f
c
, resulting in a complex passband waveform u
p
(t) = u(t)e
2ifct
. Next
u
p
(t) is converted to the real passband waveform x(t) = u

p
(t) + u
p
(t).
Assume for now that x(t) is transmitted to the receiver with no noise and no delay. In principle,
the received x(t) can be modulated back down to baseband by the reverse of the two steps used
in going from baseband to passband. That is, rst x(t) must be converted back to the complex
positive passband waveform u
p
(t), and then u
p
(t) must be shifted down in frequency by f
c
.
Mathematically, we can retrieve u
p
(t) from x(t) simply by ltering x(t) by a complex lter h(t)
such that

h(f) = 0 for f < 0 and

h(f) = 1 for f > 0. This lter is called a Hilbert lter. We
note that h(t) is not an L
2
function, but it can be converted to L
2
by making

h(f) have the
value 0 except in the positive passband [
B
2
+f
c
,
B
2
+f
c
] where it has the value 1. We can then
easily retrieve u(t) from u
p
(t) simply by a frequency shift. Figure 5.7 illustrates the sequence of
operations from u(t) to x(t) and back again.
u(t)
E n
d
c
e
2ifct
E
u
p
(t)
2'
E
x(t)
Hilbert
lter
E
u
p
(t)
n
d
c
e
2ifct
E
u(t)
. .
transmitter
. .
receiver
Figure 5.7: Baseband to passband and back.
5.4.6 Implementation of QAM
From an implementation standpoint, the baseband waveform u(t) is usually implemented as two
real waveforms, 'u(t) and u(t). These are then modulated up to passband by multiplica-
tion by in-phase and out-of-phase carriers as in (5.20), i.e.,
x(t) = 2'u(t) cos(2f
c
t) 2u(t) sin(2f
c
t).
There are many other possible implementations, however, such as starting with u(t) given as
magnitude and phase. The positive frequency expression u
p
(t) = u(t)e
2ifct
is a complex multi-
5.4. MODULATION: BASEBAND TO PASSBAND AND BACK 17
plication of complex waveforms which normally is implemented by 4 real multiplications rather
than the two above used to form x(t) directly. Thus forming u
p
(t) rst is primarily a means of
gaining insight rather than a component of implementation.
The baseband waveforms 'u(t) and u(t) are easier to generate and visualize if the mod-
ulating pulse p(t) is also real. Our discussion of the Nyquist criterion showed that this is not a
fundamental limitation. In this case,
'u(t) =

k
'u
k
p(
t
T
k),
u(t) =

k
u
k
p(
t
T
k).
Letting u

k
= 'u
k
and u

k
= u
k
, the transmitted passband waveform becomes
x(t) = 2 cos(2f
c
t)
_

k
u

k
p(tkT)
_
2 sin(2f
c
t)
_

k
u

k
p(tkT)
_
.
If the QAM signal set is a standard QAM set, then

k
u

k
p(tkT) and

k
u

k
p(tkT) are parallel
baseband PAM systems. They are modulated to passband by double-sideband modulation by
quadrature carriers cos 2f
c
t and sin 2f
c
t. These are then summed (with the usual factor of
2), as shown in Figure 5.8. This realization of QAM is called double-sideband quadrature-carrier
(DSB-QC) modulation
9
.
u

E

k
u

k
(tkT)
E
lter
p(t)
E

k
u

k
p(tkT)
n
d
c
sin 2f
c
t T
u

E

k
u

k
(tkT)
E
lter
p(t)
E

k
u

k
p(tkT)
n
d
c
cos 2f
c
t
c
n
+
E
x(t)
Figure 5.8: DSB-QC modulation
We have seen that u(t) can be recovered from x(t) by a Hilbert lter followed by shifting down
in frequency. A more easily implemented but equivalent procedure starts by multiplying x(t)
both by cos(2f
c
t) and by sin(2f
c
t). This results in
Using the trigonometric identities 2 cos
2
() = 1 + cos(2), 2 sin() cos() = sin(2), and
2 sin
2
() = 1 cos(2), this can be rewritten as
x(t) cos(2f
c
t) = 'u(t) +'u(t) cos(4f
c
t) u(t) sin(4f
c
t), (5.22)
x(t) sin(2f
c
t) = u(t) 'u(t) sin(4f
c
t) u(t) cos(4f
c
t). (5.23)
9
The terminology comes from analog modulation where two real analog waveforms are modulated respectively
onto cosine and sine carriers. For analog modulation, it is customary to transmit an additional component of
carrier from which timing and phase can be recovered. As we see shortly, no such additional carrier is necessary
here.
18 CHAPTER 5. CHANNELS, PAM, AND QAM
To interpret this, note that multiplying by cos(2f
c
t) =
1
2
e
2ifct
+
1
2
e
2ifct
both shifts x(t) up
and down in frequency by f
c
. Thus the positive frequency term in x(t) gives rise to a baseband
term and a term at 2f
c
, and the negative frequency term gives rise to a baseband term and a
term at 2f
c
. Filtering out the double frequency terms then yields 'u(t). The interpretation
of the sine multiplication is similar.
Now assume that the baseband modulation lter p(t) is real. Then 'u(t) =

u

k
p(tkT)
and u(t) =

u

k
p(tkT) are parallel baseband PAM modulations. If we further choose
p(t) so that p(tkT); k Z is an orthonormal set (i.e., so that [ p(f)[
2
satises the Nyquist
criterion), then the baseband demodulation for both the real and imaginary parts of u(t) should
use the lter q(t) = p(t). This lter serves both to lter out the double frequency terms and
to serve as the baseband demodulator. The resulting DSB-QC receiver is shown in Figure 5.9.
u

E
T spaced
sampler
E
receive lter
q(t)
E n
d
c
sin 2f
c
t
E
u

E
T spaced
sampler
E
receive lter
q(t)
E n
d
c
cos 2f
c
t
E
E
x(t)
Figure 5.9: DSB-QC demodulation
Using cosine and sine (i.e., quadrature) carriers above, followed by ltering out the double
frequency terms, is equivalent to using a Hilbert lter followed by shifting the resulting complex
waveform down in frequency. The other stages of Figure 5.9, i.e., the parallel lters q(t) followed
by sampling to get u

k
and u

k
is specic to DSB-QC, which requires both a real lter p(t)
and a standard QAM signal set. Otherwise the real and imaginary parts of u(t) would be jointly
involved in nding u

k
and similarly u

k
.
5.5 Degrees of freedom
We found that with PAM, we can choose real signals separated by T and transmit them in the
baseband bandwidth W = 1/(2T). Thus, over a long interval T
0
, and in a baseband bandwidth
W, we can transmit 2WT
0
real signals using PAM.
Using QAM, with T-spaced signals, we need a passband bandwidth W = 1/T. However, each
signal is complex and consists of a real and imaginary part, each of which can be independently
selected. Thus, again, over a long interval of time T
0
, we can select an arbitrary set of 2WT
0
real signals.
The above argument seems a little aky since W is dened dierently for PAM and QAM. To
obtain a more reasonable comparison, consider an overall large baseband bandwidth W
0
broken
into m passbands each of bandwidth W
0
/m. Using QAM in each band, we can transmit 2W
0
T
0
5.5. DEGREES OF FREEDOM 19
real signals in a long interval T
0
. With PAM used over the entire band W
0
, we again send 2W
0
T
0
real signals in a duration T
0
. We see that in principle, QAM and baseband PAM are equivalent
in terms of the number of degrees of freedom that can be used to transmit real signals. As
pointed out earlier, however, PAM when modulated up to passband uses only half the available
degrees of freedom.
Recall that when we were looking at T-spaced truncated sinusoids and T-spaced sinc weighted
sinusoids, we argued that the class of real waveforms occupying a time interval (T
0
/2, T
0
/2)
and a frequency interval (W
0
, W
0
) has about 2T
0
W
0
degrees of freedom for large W
0
, T
0
. What
we see now is that baseband PAM and passband QAM each employ about 2T
0
W
0
degrees of
freedom. In other words, these simple techniques essentially use all the degrees of freedom
available in the given bands.
5.5.1 Distance and orthogonality
We have shown how to modulate a complex QAM baseband waveform u(t) up to a real passband
waveform x(t) and how to retrieve u(t) from x(t) at the receiver. We have also discussed signal
constellations that minimize energy for given minimum distance. Finally, by using a modulation
waveform p(t) with orthonormal shifts, we have seen that the energy dierence between two
baseband signal waveforms, say u(t) =

u
k
p(t kT) and v(t) =

k
v
k
p(t kt) is related to
the energy dierence in the signal points by
|u v|
2
=

k
[u
k
v
k
[
2
.
We now must explore what happens to this energy dierence at passband.
We have already pointed out that the energy |x|
2
in the passband waveform x(t) is twice that
in the corresponding baseband waveform u(t). Next suppose that x(t) and y(t) are the passband
waveforms arising from the baseband waveforms u(t) and v(t) respectively. Then
x(t) y(t) = 2'u(t)e
2ifct
2'v(t)e
2ifct
= 2'[u(t)v(t)]e
2ifct
.
Thus we see that x(t) y(t) is the passband waveform corresponding to u(t) v(t), so
|x(t) y(t)|
2
= 2|u(t) v(t)|
2
.
This says that for QAM, distances between waveforms are preserved (aside from the factor of 2
in energy or

2 in distance) in going from baseband to passband. Thus distances are preserved
in going from signals to baseband waveforms to passband waveforms and back. We will see later
that the error probability caused by noise is essentially determined by the distances between the
set of passband source waveforms. This error probability is then simply related to the choice of
signal constellation and the discrete coding that precedes the mapping of data into signals.
This preservation of distance through the modulation process is the crux of the signal space
viewpoint of digital communication. It gives a very practical focus to viewing waveforms as
elements of the L
2
inner product space.
There is unfortunately a complication in this very nice story. The set of baseband waveforms
constitutes a complex inner product space u : R C whereas the set of passband waveforms
is best viewed as constituting a real inner product space x : R R. The transformation
x(t) = 'u(t)e
2ifct
is not linear, since, for example, iu(t) does not map into ix(t) (unless
20 CHAPTER 5. CHANNELS, PAM, AND QAM
u(t) = 0. In fact, the notion of a linear transformation does not make sense when we view the
transformation as going from complex L
2
to real L
2
since the scalars are dierent in the two
spaces.
Example 2 As an important example, suppose the QAM modulation waveform is a real
waveform p(t) with orthonormal shifts. The corresponding passband waveform is then
2p(t) cos(2f
c
t). As shown in the exercises,

2p(tkT) cos(2f
c
t); k Z is an orthonor-
mal set (still assuming that p(f) = 0 for [f[ f
c
). Unfortunately, although p(tkT); k Z
spans the space of complex baseband source waveforms,

2p(tkT) cos(2f
c
t); k Z does
not span the corresponding set of passband waveforms. It is necessary to add the additional set

2p(tkT) sin(2f
c
t); k Z of waveforms, which, together with the cosine terms, forms an
orthonormal set that spans the space of passband source waveforms.
Another way to look at this example is to observe that modulating the baseband function
u(t) into the positive passband function u
p
(t) = u(t)e
2ifct
is somewhat easier to under-
stand in that the orthonormal set p(tkT); k Z is modulated to the orthonormal set
p(tkT)e
2ifct
; k Z, which can be seen to span the space of complex positive frequency pass-
band source waveforms. The additional set of orthonormal waveforms p(tkT)e
2ifct
; k Z
is then needed to span the real passband source waveforms. We then see that the sine, cosine
series is simply another way to express this. In the sine, cosine formulation all the coecients in
the series are real, whereas in the complex exponential formulation, there is a real and complex
coecient for each term, but they are pairwise dependent. It will be easier to understand the
eects of noise in the sine, cosine formulation.
In the above example, we have seen that each orthonormal function at baseband gives rise to
two real orthonormal functions at passband. It can be seen from a degrees of freedom argument
that this is inevitable no matter what set of orthonormal functions are used at baseband. For a
nominal bandwidth W, there are 2W real degrees of freedom per second in the baseband com-
plex source waveform, and this breaks up into 2 real degrees of freedom for each orthonormal
baseband waveform. At passband, we have the same 2W degrees of freedom per second, but
with a real orthonormal expansion, there is only one real degree of freedom for each orthonor-
mal waveform. Thus there must be two passband orthonormal waveforms for each baseband
orthonormal waveform.
The sine, cosine expansion above generalizes in a nice way to an arbitrary set of complex or-
thonormal baseband functions. Each complex function in this baseband set generates two real
functions in an orthogonal passband set. This is expressed precisely in the following theorem
which is proven in the exercises.
Theorem 5.5.1 Let
k
(t) : k Z be an orthonormal set limited to the frequency band
[B/2, B/2]. Let f
c
be greater than B/2 and for each k Z let

k,1
(t) = '
_
2
k
(t)
2ifct
_
,

k,2
(t) =
_
2
k
(t)
2ifct
_
.
This is an orthogonal set of functions, each of energy 2. Furthermore, if u(t) =

k
u
k

k
(t),
then the corresponding passband function x(t) = 2'u(t)e
2ifct
is given by
x(t) =

k
'u
k

k,1
(t) +u
k

k,2
(t).
5.6. CARRIER RECOVERY IN QAM SYSTEMS 21
5.6 Carrier recovery in QAM systems
Consider a QAM receiver and visualize the passband-to-baseband conversion as multiplying the
positive frequency passband by the complex sinusoid e
2ifct
. If the receiver has a phase error
(t) in its estimate of the phase of the transmitted carrier, then it will instead multiply the
incoming waveform by e
2ifct+i(t)
. We assume in this analysis that the time reference at the
receiver is perfectly known, so that the sampling of the ltered output is done at the correct
time. Thus the assumption is that the oscillator at the receiver is not quite in phase with the
oscillator at the transmitter. Another point of view here is that the carrier frequency is usually
orders of magnitude higher than the baseband bandwidth, and thus a small error in timing is
signicant in terms of carrier phase but not in terms of sampling. The carrier phase error will
rotate the correct complex baseband signal u(t) by (t); i.e., the actual received baseband signal
v(t) will be
v(t) = e
i(t)
u(t).
If (t) is slowly time-varying relative to the response q(t) of the receiver lter, then the samples
r(kT) of the lter output will be
r(kT) e
i(kT)
u
k
,
as illustrated in Figure 5.10. The phase error (t) is said to come through coherently. This
phase coherence makes carrier recovery easy in QAM systems.
t t t t
t t t t
t t t t
t t t t
$W
y
g
#

gy
z
$X
Figure 5.10: Rotation of constellation points by phase error
As can be seen from the gure, if the phase error is small enough, and the set of points in the
constellation are well enough separated, then the phase error can be simply corrected by moving
to the closest signal point and adjusting the phase of the demodulating carrier accordingly.
There are two complicating factors here. The rst is that we have not taken noise into account
yet. When the received signal y(t) is x(t) +n(t), then the output of the T spaced sampler is not
the original signals u
k
, but rather a noise corrupted version of them. The second problem is
that if a large phase error ever occurs, it can not be corrected. For example, in Figure 5.10, if
(t) = /2, then even in the absence of noise, the received samples always line up with signals
from the constellation (but of course, not the transmitted signals).
5.6.1 Tracking phase in the presence of noise
The problem of deciding on or detecting the signals u
k
from the received samples r(kT) in
the presence of noise is a major topic that we will take up shortly. Here, however, we have the
22 CHAPTER 5. CHANNELS, PAM, AND QAM
added complication of both detecting the transmitted signals and also tracking and eliminating
the phase error.
Fortunately, the problem of decision making and that of phase tracking are largely separable.
The reason is that the oscillators used to generate the modulating and demodulating carriers
are relatively stable and have phases which change quite slowly relative to each other. Thus the
phase error with any kind of reasonable tracking will be quite small, and thus the data signals
can be detected from the received samples almost as if the phase error were zero. The dierence
between the received sample and the detected data signal will still be non-zero, mostly due to
noise but partly due to phase error. However, the noise has zero mean (as we understand later)
and thus tends to average out over many sample times. Thus the general approach is to make
decisions on the data signals as if the phase error is zero, and then to make slow changes to
the phase based on averaging over many sample times. This approach is called decision directed
carrier recovery. Note that if we track the phase as phase errors occur, we are also tracking the
carrier, in both frequency and phase.
In a decision directed scheme, assume that the received sample r(kT) is used to make a decision
d
k
is made on the transmitted signal point u
k
. Also assume that d
k
= u
k
with very high
probability. The apparent phase error for the kth sample is then the dierence between the
phase of r(kT) and the phase of d
k
. Any method for feeding back the apparent phase error to
the generator of the sinusoid e
2ifct+i(t)
in such a way as to slowly reduce the apparent phase
error will tend to produce a robust carrier recovery system.
In one popular method, the feedback signal is taken as the imaginary part of r(kT)d

k
. If the
phase angle from d
k
to r(kT) is
k
, then
r(kT)d

k
= [r(kT)[[d
k
[ e
i
k
,
so the imaginary part is [r(kT)[[d
k
[ sin
k
[r(kT)[[d
k
[
k
, when
k
is small. Decision-directed
carrier recovery based on such a feedback signal can be extremely robust even in the presence
of substantial distortion and large initial phase errors. With a second-order phase-locked carrier
recovery loop, it turns out that the carrier frequency f
c
can be recovered as well.
5.6.2 Large phase errors
A problem with decision-directed carrier recovery and with many other approaches is that the
recovered phase may settle into any value for which the received eye pattern (i.e., the pattern of
a long string of received samples as viewed on a scope) looks OK. With (MM)-QAM signal
sets, as in Figure 5.10, the signal set has four-fold symmetry, and phase errors of 90

, 180

, or 270

are not detectable. Simple dierential coding methods that transmit the phase (quadrantal)
part of the signal information as a change of phase from the previous signal rather than as an
absolute phase can easily overcome this problem. Another approach is to resynchronize the
system frequently by sending some known pattern of signals. This latter approach is frequently
used in wireless systems where fading sometimes causes a loss of phase synchronization.
5.E. EXERCISES 23
5.E Exercises
(1) (PAM) A discrete memoryless source emits binary equiprobable symbols at a rate of 1000
symbols per second. The symbols from a one second interval are grouped into pairs and
sent over a bandlimited channel using a standard 4-PAM signal set. The modulation uses
a signal interval 0.002 and pulse p(t) = sinc(t/T).
(a) Suppose that a sample sequence u
1
, . . . , u
500
of transmitted signals includes 115 ap-
pearances of 3d/2, 130 appearances of d/2, 120 appearances of d/2, and 135 appear-
ances of 3d/2. Find the energy in the corresponding transmitted waveform u(t) =

500
k=1
u
k
sinc(
t
T
k) as a function of d.
(b) What is the bandwidth of the waveform u(t) in part (a)?
(c) Assume iid signal probabilities equal to the relative frequencies in (a). Find E[U
2
k
].
(d) Find E
__
U
2
(t) dt

where U(t) is the random waveform

500
k=1
U
k
sinc(
t
T
k).
(e) Now suppose that the binary source is not memoryless, but is instead generated by a
Markov chain where
Pr(X
i
=1 [ X
i1
=1) = Pr(X
i
=0 [ X
i1
=0) = 0.9.
Assume the Markov chain starts in steady state with Pr(X
1
=1) = 1/2. Using the mapping
(00 a
1
), (01 a
2
), (10 a
3
), (11 a
4
), nd E[A
2
k
] for 1 k 500.
(f) Find E
__
U
2
(t) dt

for this new source.

(g) For the above Markov chain, explain how we could change the above mapping to reduce
the expected energy without changing the separation between signal points.
(2) (Nyquist) Suppose that the PAM modulated baseband waveform u(t) =

k=
u
k
p(tkT)
is received. That is, u(t) is known, T is known, and p(t) is known. We want to determine
the signals u
k
from u(t). We assume we must use only linear operations. That is, we
wish to nd some waveform d
k
(t) for each integer k such that
_

u(t)d
k
(t) dt = u
k
.
(a) What properites must be satised by d
k
(t) such that the above equation is satised no
matter what values are taken by the other signals, . . . , u
k2
, u
k1
, u
k+1
, u
k+2
, . . . ? These
properties should take the form of constraints on the inner products p(t kT), d
j
(t)). Do
not worry about convergence, interchange of limits, etc.
(b) Suppose you nd a function d
0
(t) that satises these constraints for k = 0. Show that
for each k, a function d
k
(t) satisfying these constraints can be found simply in terms of
d
0
(t).
(c) What is the relationship between d
0
(t) and a function q(t) that avoids intersymbol
interference in the approach taken in the notes (i.e., a function q(t) such that p(t) q(t) is
ideal Nyquist).
You have shown that the lter/sample approach in the notes is no less general than the
arbitrary linear operation approach here. Note that, in the absence of noise and with a
known signal constellation, it might be possible to retrieve the signals from the waveform
using non-linear operations even in the presence of intersymbol interference.
24 CHAPTER 5. CHANNELS, PAM, AND QAM
(3) (PAM) Consider standard M-PAM and assume that the signals are used with equal prob-
ability. Show that the average energy per signal E
s
= U
2
k
is equal to the average energy
U
2
= d
2
M
2
/12 of a uniform continuous distribution over the interval [dM/2, dM/2], mi-
nus the average energy (U U
k
)
2
= d
2
/12 of a uniform continuous distribution over the
interval [d/2, d/2]:
E
s
=
d
2
(M
2
1)
12
.
(4) (Nyquist) Let v(t) be a continuous L
2
waveform and dene g(t) = v(t) sinc(
t
T
).
(a) Show that g(t) is ideal Nyquist with interval T.
(b) Find g(f) as a function of v(f).
(c) Give a direct demonstration that g(f) satises the Nyquist criterion.
(d) If v(t) is baseband limited to B, what is g(t) baseband limited to?
Note: The usual form of the Nyquist criterion helps in choosing waveforms that avoid
intersymbol interference with prescribed rollo properties in frequency. The approach above
show how to avoid intersymbol interference with prescribed attenuation in time and in
frequency.
(5) (Nyquist) Consider a PAM baseband system in which the modulator is dened by a signal
interval T and a pulse p(t), the channel is dened by a lter h(t), and the receiver is dened
by a lter q(t) which is sampled at T-spaced intervals. The received waveform, after the
receive lter q(t), is then given by r(t) =

k
u
k
g(t kT) where g(t) = p(t) h(t) q(t).
(a) What property must g(t) have so that r(kT) = u
k
for all k and for all choices of input
u
k
? What is the Nyquist criterion for g(f)?
(b) Now assume that T = 1/2 and that p(t), h(t), q(t) and all their Fourier transforms are
restricted to be real. Assume further that p(f) and

h(f) are given by
p(f) =
_
_
_
1, [f[ 0.5;
1.5 t, 0.5 < [f[ 1.5
0, [f[ > 1.5

h(f) =
_

_
1, [f[ 0.75;
0, 0.75 < [f[ 1
1, 1 < [f[ 1.25
0, [f[ > 1.25
0
1
2
3
2
p(f) 1
0
3
4
5
4

h(f)
1
Is it possible to choose a receive lter transform q(f) so that there is no intersymbol interfer-
ence? If so, give such a q(f) and indicate the regions in which your solution is non-unique.
(c) Redo part (c) with the modication that now

h(f) = 1 for [f[ 0.75 and

h(f) = 0 for
[f[ > 0.75.
(d) Explain the conditions on p(f)

h(f) under which intersymbol interference can be avoided

by proper choice of q(f) (you may assume, as above, that p(f),

h(f), p(t), and h(t) are all

real).
5.E. EXERCISES 25
(6) (Nyquist) For the special case = 1, T = 1, verify the formula in (5.13) for the inverse
transform of the raised cosine frequency function in (5.12). Hint: As an intermediate step,
verify that g
1
(t) = sinc(2t) +
1
2
sinc(2t + 1) +
1
2
sinc(2t 1). Sketch g
1
(t), in particular
showing its value at mT/2 for each m 0.
(7) (QAM) (a) Let
1
(t) and
2
(t) be orthonormal complex waveforms. Let
j
(t) =
j
(t)e
2ifct
for j = 1, 2. Show that
1
(t) and
2
(t) are orthonormal for any f
c
.
(b) Suppose that
2
(t) =
1
(t T). Show that
2
(t) =
1
(t T) if f
c
is an integer multiple
of 1/T.
(8) (QAM) (a) Assume B/2 < f
c
. Let u(t) be a real function baseband limited to B/2 and let
v(t) be a pure imaginary function limited to B/2. Show that the corresponding passband
functions, 'u(t)e
2ifct
and 'v(t)e
2ifct
are orthogonal.
(b) Give an example where the functions in part (a) are not orthogonal if B/2 > f
c
.
(9) (Passband expansions) Assume that p(tkT) : kZ is a set of orthonormal functions.
Assume that p(f) = 0 for [f[ f
c
).
(a) Show that

2p(tkT) cos(2f
c
t); kZ is an orthonormal set.
(b) Show that

2p(tkT) sin(2f
c
t); kZ is an orthonormal set and that each function
in it is orthonormal to the cosine set in part (a).
(10) (Passband expansions) Prove Theorem 5.5.1. Hint: First show that the set of functions

k,1
(f) and

k,2
(f) are orthogonal with energy 2 by comparing the integral over neg-
ative frequencies with that over positive frequencies. Indicate explicitly why you need
f
c
> B/2.
(11) (Carrierless amplitude-phase modulation (CAP)) We have seen how to modulate a base-
band QAM waveform up to passband and then demodulate it by shifting down to baseband,
followed by ltering and sampling. This exercise explores the interesting concept of elim-
inating the baseband operations by modulating and demodulating directly at passband.
This approach is used in one of the North American standards for Asymmetrical Digital
Subscriber Loop (ADSL)
(a) Let u
k
be a complex data sequence and let u(t) =

k
u
k
p(tkT) be the corresponding
modulated output. Let p(f) be equal to

T over f [3/(2T), 5/(2T)] and be equal to 0
elsewhere. At the receiver, u(t) is ltered using p(t) and the output y(t) is then T-space
sampled at time instants kT. Show that y(kT) = u
k
for all k Z. Dont worry about the
fact that the transmitted waveform u(t) is complex.
(b) Now suppose that p(f) =

T rect(T(f f
c
)] for some arbitrary f
c
rather than f
c
= 2/T
as in part (a). For what values of f
c
does the scheme still work?
(c) Suppose that 'u(t) is now sent over a communication channel. Suppose that the
received waveform is ltered by a Hilbert lter before going through the demodulation
procedure above. Does the scheme still work?