7.

1 BASEBAND DATA TRANSMISSION IN WHITE GAUSSIAN NOI,E 331

(PSK), orfrequency-shift keying (FSK) if it is amplitude, phase, or frequency,respectively,

that is varied in accordance with the baseband signal. An important M-ary modulation
scheme, quadriphase-shift keying (QPSK), is often employed in situations in which band-
width efficiency is a consideration. Other schemes related to QPSK include offsetQPSK
and minimum-shift keying (MSK). These schemes will be discussed in Chapter 8.
A digital communication system is referred to as coherent if a local reference is
available for demodulation that is in phase with the transmitted carrier (with fixed phase
shifts due to transmission delays accounted for). Otherwise, it is referred to as noncoherent.
Similarly, if a periodic signal is available at the receiver that is in synchronism with the
transmitted sequence of digital signals (referred to as a clock), the system is referred to as
synchronous; if a signaling technique is employed in which such a clock is unnecessary
(e.g., timing markers might be built into the data blocks), the system is called asynchronous.
The primary measure of system performance for digital data communication systems
is the probability of error PE. In this chapter we will obtain expressions for PE for various
types of digital communication systems. We are, of course, interested in receiver struc-
tures that give minimum PE for given conditions. Synchronous detection in an additive
white Gaussian-noise background requires a correlation or a matchedjilter detector to give
minimum PE for fixed signal and noise conditions.
We begin our consideration of digital data transmission systems in Section 7.1 with
the analysis of a simple, synchronous baseband system that employs a special case of
the matched filter detector known as an integrate-and-dump detector. This analysis is
then generalized in Section 7.2 to the matched-filter receiver. Section 7.3 contains sev-
eral specializations of the results of Section 7.2 to common coherent digital signaling
schemes, while noncoherent schemes are analyzed in Section 7.4. After analyzing these
modulation schemes, which operate in an ideal environment in the sense that infinitebana-
width is available,we look at optimum signaling through bandlimited baseband channels in
Section 7.5. In Sections 7.6 and 7.7, the effects of multipath interference and signal fa~ing
on data transmissionare analyzed, and in Section7.8, the use of equalizing filtersto mitigate
the effects of channel distortion is examined. Finally, in Section 7.9, an old modulation
idea called multicarrier modulation (MCM), which has enjoyed a resurgence of attention,
is briefly described. A special case of MCM is known as orthogonal frequency-division
multiplexing (OFDM). .

7.1 BASEBAND DATA TRANSMISSION IN

WHITEGAUSSIAN NOISE

Consider the binary digital data communication system illustrated in Figure 7.2(a), in which
the transmitted signal consists of a sequence of constant-amplitude pulses of either A or - A
units in amplitude and T seconds in duration. A typical transmitted sequence is shown in
Figure 7.2(b). We may think of a positive pulse as representing a logic 1 and a negative pulse
as representing a logic 0 from the data source. Each T -second pulse is called a binit for
binary digit or, more simply, a bit. (In Chapter 10, the term bit will take on a new meaning.)
As in Chapter 6, the channel is idealized as simply adding white Gaussian noise
with double-sided power spectral density ~ No to the signal.A typicalsamplefunction
of the received signal plus noise is shown in Figure 7.2(c). For sketching purposes, it is
assumed that the noise is bandlimited, although it is modeled as white noise later when the
performance of the receiver is analyzed. It is assumed that the starting and ending times of
e]ch pulse are known by the receiver. The problem of acquiring this information, referred
to as synchronization, will not be considered at this time.
 - "'
332 CHAPTER 7 BINARY DATA TRANSMISSION

n(t): PSD = !NO

(a)

s(t)
"1" "0" "0" "0" "1"
A
I I
T 2T 3T 4T 5T
-A -

(b)

y(t)

-~ (c)
FIGURE 7.2 System model and waveforms for synchronou.s baseband digital data
Jransmission. (a) Baseband digital data communication system. (b) Typical transmitted
sequence. (c) Received sequence plus noise.

The function of the receiver is to decide whether the transmitted signal was A or - A
during each bit period. A straightforwiITd way of accomplishing this is to pass the signal
plus noise through a lmvpass predetection filter, sample its output at some timewithin each
"')"-second mterval, and detemune the sign of the sample. If the sample is greater than zero,
the declSlon is made that + A was transmitted. If the sample is less than zero, the decision is
that - A was transmitted. With such a receiver structure, however, we do not take advantage
of everything known about the signal. Since the starting and ending times of the pulses are
known, a better procedure is to compare the area of the received signal-plus"noise wavefonn
(data) with zero at the end of each signaling interval by integrating the received data over
the T -second signaling interval. Of course, a noise component is present at the output of the
integrator, but since the input noise has zero mean, it takes on positive and negative values
with equal probability. Thus the output noise component has zero mean. The proposed re-
ceiver structure and a typical waveform at the output of the integrator are shown in Figure 7.3
where to is the start of an arbitrary signaling interval. For obvious reasons, this receiver is
referred to as an integrate-and-dump detector;
The question to be answered is: How well does this receiver perform, and on what
parameters does its performance depend? As mentioned previously, a useful criterion of
performance is probability of error, and it is this we now compute. The output of the
integrator at the end of a signaling interval is
'o+T
V =
l
[s(t) + n(t)] dt
/0

+ AT + N if+ A is sent
= (7.1)
I - AT + N if - A is sent

:"''':1
 7.' BASEBAND DATA TRANSMISSION IN WHITE GAUSSIAN NOISE 333

~ f+TO& ~'=~+T
(a)
> 0: choose +A
< 0: choose-A

Signal plus noise

AT \ Signal

to
-AT

(b)
FIGURE 7.3 Receiver structure and integrator output. (a) Integrate-and-dump receiver.
(b) Output from the integrator.

where N is a random variable defined as

to+T
N =
I10
n(t)dt (7.2)

Since N results from a linear operation on a sample function from a Gaussian process, it is
a Gaussian random variable. It has mean

to+T to+T
E{N}=E { )10 n(t)dt } =)10 E{n(t)}dt=O (7.3)
since n(t) has zero mean. Its variance is therefore E~'n(\-) \'\(i)'S':" I\~,O')

4J1'~~~~
.."" i_u
var{N} = E{N2} = E ([(OHn(t)dtfl ~-6ca L \,.J..m-. ~r-\"d'f
IO+T IO+T == Rl-\;-o')
= l l 10
IO+T
10
E{n(t)n (a)) dt ,da
tVr wa,~,
~
~~~)=Z:~
~\o

I
=
l l 10 10
IO+T
- No8 (t -
2
a) dt da ~, R",E!:-a)=~. ~-(

where we have made the substitUtion E{n(t)n(a)} = !No8(t - a). Using the sifting prop-
erty of the delta function, we obtain
1
var{N} =
l1
10
IO+T
-Noda
2

= -NoT
2 (7.4)

Thus the pdf of N is

e-~2/NOT
IN (1/) = (7.5)

where 1/is used as the dummy variable for N to avoid confusion with n(t).
 I I
I
!
334 CHAPTER 7 BINARYDATATRANSMISSION

IN (T/)

P (Error IA Sent) P (Error I-A Sent)

=P (AT+N<O) =P (-AT + N>O)

FIGURE 7.4 Illustration of error

T/
-AT 0 AT probabilities for binary signaling.

There are two ways in which errors occur. If + A is transmitted, an error occurs if
AT + N < 0, that is, if N < - AT. From (7.5), the probability of this event is
-AT

i
_ry2jNoT
(7.6)
P(errorlA sent) = P(EIA) = -00 ~dlJ
T{NoT

which is the area to the left of IJ = -AT in Figure 7.4. Letting u = ,J21J/NoT, we can
write this as2

P(EIA) = 1 00 e-u2j2 du ~ Q (7.7)

,J2A2T/Nov'2ii ( j2A2T
No )
The other way in which an error can occur is if - A is transmitted and - A T + N > O.The
probability of this event is the same as the probability that N > AT, which can be written
as

00 e-ry2jNoT fj, (WT (7.8)

P(EI"" A) = 1AT../iiNOf dlJ= Q ( V N;- )
which is the area to the right of IJ = AT in Figure 7.4. The average probability of error is

PE = P(EI + A)P(+A) + P(EI - A)P(-A) (7.9)

Substituting (7.7) and (7.8) into (7.9) and noting that P(+A) + PC-A) = 1, where peA)
is the probability that + A is transmitted, we obtain

(7.10)
PE~Q(N)
Thus the important parameter is A2T / No. We can interpret this ratio in two ways.
First, since the energy in each signal pulse is .

tO+T
Eb = l to
A2dt = A2T (7.11)

we see that the ratio of signal energy per pulse to noise power spectral density is

A2T Eb
z---- (7.12)
- No - No

2See Appendix G for a discussion and tabulation of the Q-function.

',-
-
 -~

7.1 BASEBAND DATA TRANSMISSION IN WHITE GAUSSIAN NOISE 335

1.0
5 X 10-1

5 X 1O~2

t 10-2
ri:: 5 X 10-3

10-3
5 X 10-4

10-4
-10 -5 0 5 10 FIGURE 7.5 ~ for antipodal baseband
10 logloZ digital signaling.

where Eb is called the energy per bit because each signal pulse (+A or -A) carries one
bit of infonnation. Second, we recall that a rectangular pulse of duration T seconds has
amplitude spectrum ATsincTf and that Bp = liT is a rough measure of its bandwidth.
Thus
A2 A2
z= =- (7.13)
No (lIT) NoBp
can be interpreted as the ratio of signal power to noise power in the signal bandwidth. The
bandwidth Bp is sometimes referred to as the bit-rate bandwidth. We will refer to z as
the signal-to-noise ratio (SNR). An often-used reference to this signal-to-noise ratio in the
digital communications industry is "Eb-over-No."
A plot of PE versus z is shown in Figure 7.5, where z is given in decibels. Also shown
is an approximation for PE using the asymptotic expansion for the Q-function:

e-U2/2 u 1 (7.14)
Q (u) ~ u./iii'
Using this approximation,
"" e-z
PE = 2",r.:::;'
z z
TC
1 (7.15)

which shows that PE essentially decreases exponentially with increasing z. Figure 7.5 shows
that the approximation of (7.15) is close to the true result of (7.10) for z > 3 dB.

EXAMPLE 7.1
Digital data are to be transmitted through a baseband system with No = 10-7 W/Hz and the received
signal amplitude A = 20 mY. (a) If 103 bits per second (bps) are transmitted, what is PE? (b) If
10" bps are transmitted, to what value must A be adjusted in order to attain the same PE as in part (a)?
 - ,

336 CHAPTER 7 BINARYDATATRANSMISSION

Solution: To solve part (a), note that

I
Z- -A2T - (0.02)2(10-3) -4
- No - 10-7 - (7.16) I

Using- (7.15), PE ~ e-4!2~

A2(1O 4)/(10-7)
= 2.58 x
= 4,whichgIvesA = 63.2mY.
10-3. Part (b) is solved by finding A such that
. I

EXAMPLE7.2
The conditions are the same as in the preceding example, but a bandwidth of 5000 Hz is available.
I

(a) What is the maximum data rate that can be supported by the channel? (b) Find the transmitter I
power required to give a probability of error of 10-6 at the data rate found in part (a).

I
Solution: (a) Since a rectangular pulse has Fourier transform

n(tIT) - T sinc(fT) I

we take the signal bandwidth to be that of the first null of the sinc function. Therefore, lIT = I
5000 Hz, which implies a maximum data rate of R = 5000 bps. (b) To find the transmitter power to

give PE = 10-6, we solve I

10-6 = Q[J2A2TINo] = Q[hZ] (7.17)
I
Using the approximation (7.14) for the error function, we need to solve

6 - e-Z
10- = 2.j1rZ
iteratively. This gives the result

z ~ 10.54 dB = 11.31

Thus, A2T 1No = 11.31, or

A2 = (11.31)NoIT = 5.65 mW (actually y2 x 10-3)

This corresponds to a signal amplitude of approximately 75.2 mY. . I

7.2 BINARY SYNCHRONOUS DATA

TRANSMISSION WITHARBITRARY
SIGNAL SHAPES
In Section 7.1 we analyzed a simple baseband digital communication system. As in the I

case of analog transmission, it is often necessary to utilize modulation to condition a digital

message signal so that it is suitable for transmission through a channel. Thus, instead of
the constant-level signals considered in Section 7.1, we will let a logic I be represented by
SI (t) and a logic 0 by S2 (t). The only restriction on SI (t) and S2(t) is that they must have
finite energy in a T -second interval. The energies of SI (t) and S2(t) are denoted by

to+T
EI ~ l to
s~(t) dt (7.18)