Haykin_ch09_pp3.

fm Page 557 Friday, January 4, 2013 4:58 PM

9.13 Spread Spectrum Signals 557

1
Input Serial- 2 Digital-
Forward 64-point Parallel- Transmitted

…
binary error- 16-QAM to- to- to-
data 47 inverse signal
correction modulator parallel serial analog
stream 48 FFT
encoder converter converter converter

36 48 12 Subcarriers 250 Data stream

Mbit/s Mbit/s MHz 250 kHz kHz 16 MHz
(a)

Forward Estimate
Analog- Serial-to- 64-point Parallel- 16-QAM
Received error- of original
to-digital parallel FFT to-serial de-

…
signal correction binary
converter converter algorithm converter modulator
decoder data stream

(b)

Figure 9.31 Block diagram of the typical implementation of an OFDM, illustrating the transmission
of binary data at 36 Mbit/s.

implementation. However, OFDM suffers from the so-called PAPR problem. This
problem arises due to the statistical probabilities of a large number of independent
subchannels in the OFDM becoming superimposed on each other in some unknown
fashion, thereby resulting in high peaks. For a detailed account of the PAPR problem and
how to mitigate it, the reader is referred to Appendix G.

9.13 Spread Spectrum Signals

In previous sections of this chapter we described different methods for mitigating the
effect of multipath interference in signaling over fading channels. In this section of the
chapter, we describe another novel way of thinking about wireless communications, which
is based on a class of signals called spread spectrum signals.14
A signal is said to belong to this class of signals if it satisfies the following two
requirements:
1. Spreading. Given an information-bearing signal, spreading of the signal is
accomplished in the transmitter by means of an independent spreading signal, such
that the resulting spread spectrum signal occupies a bandwidth much larger than the
bandwidth of the original information-bearing signal: the larger the better.
2. Despreading. Given a noisy version of the transmitted spread spectrum signal,
despreading (i.e., recovering the original information-bearing signal) is achieved by
correlating the received signal with a synchronized replica of the spreading signal in
the receiver.
Haykin_ch09_pp3.fm Page 558 Friday, January 4, 2013 4:58 PM

558 Chapter 9 Signaling over Fading Channels

In effect, the information-bearing signal is spread (increased) in bandwidth before its

transmission over the channel, and the received signal at the channel output is despread
(i.e., decreased) in bandwidth by the same amount.
To explain the rationale of spread spectrum signals, consider, first, a scenario where
there are no interfering signals at the channel output whatsoever. In this idealized scenario,
an exact replica of the original information-bearing signal is reproduced at the receiver
output; this recovery follows from the combined action of spreading and despreading, in
that order. We may thus say that the receiver performance is transparent with respect to
the combined spreading–despreading process.
Consider, next, a practical scenario where an additive narrowband interference is
introduced at the receiver input. Since the interfering signal is introduced into the
communication system after transmission of the information-bearing signal, its bandwidth
is increased by the spreading signal in the receiver, with the result that its power spectral
density is correspondingly reduced. Typically, at its output end, the receiver includes a
filter whose bandwidth-occupancy matches that of the information-bearing signal.
Consequently, the average power of the interfering signal is reduced, and the output SNR
of the receiver is increased; hence, there is practical benefit in improved SNR to be gained
from using the spread spectrum technique when there is an interfering signal (e.g., due to
multipath) to deal with. Of course, this benefit is obtained at the expense of increased
channel bandwidth.

Classification of Spread Spectrum Signals

Depending on how the use of spread spectrum signals is carried out, we may classify them
as follows:
1. Direct Sequence-Spread Spectrum
One method of spreading the bandwidth of an information-bearing signal is to use
the so-called direct sequence-spread spectrum (DS-SS), wherein a pseudo-noise
(PN) sequence is employed as the spreading sequence (signal). The PN sequence is
a periodic binary sequence with noise-like properties, details of which are
presented in Appendix J. The baseband modulated signal, representative of the
DS-SS method, is obtained by multiplying the information-bearing signal by the
PN sequence, whereby each information bit is chopped into a number of small
time increments, called chips. The second stage of modulation is aimed at
conversion of the baseband DS-SS signal into a form suitable for transmission over
a wireless channel, which is accomplished by using M-ary PSK, discussed in
Chapter 7. The family of spread spectrum systems so formed is referred to simply
as DS/MPSK systems, a distinct characteristic of which is that spreading of the
transmission bandwidth takes place instantaneously. Moreover, the signal-
processing capability of these systems to combat the effect of interferers,
commonly referred to as jammers be they friendly or unfriendly, is a function of
the PN sequence length. Unfortunately, this capability is limited by physical
considerations of the PN-sequence generator.
2. Frequency Hop-Spread Spectrum
To overcome the physical limitations of DS/MPSK systems, we may resort to
alternative methods. One such method is to force the jammer to occupy a wider
Haykin_ch09_pp3.fm Page 559 Friday, January 4, 2013 4:58 PM

9.13 Spread Spectrum Signals 559

spectrum by randomly hopping the input data-modulated carrier from one frequency
to the next. In effect, the spectrum of the transmitter signal is spread sequentially
rather than instantaneously; the term sequentially refers to the pseudo-randomly
ordered sequence of frequency hops. This second type of spread spectrum in which
the carrier hops randomly from one frequency to another is called frequency hop-
spread spectrum. A commonly used modulation format used herein is that of M-ary
FSK, which was also discussed in Chapter 7. The combination of the two
modulation techniques, namely frequency hopping and M-ary FSK, is referred to
simply as FH/MFSK. Since frequency-hopping does not cover over the entire spread
spectrum instantaneously, we are led to consider the rate at which the hops occur. In
this context, we may go on to identify two basic kinds of frequency hopping, which
are the converse of each other, as summarized here:
• First, slow-frequency hopping, in which the symbol rate of the M-ary FSK signal,
denoted by Rs, is an integer multiple of the hop rate, denoted by Rh; that is, several
symbols of the input data sequence are transmitted for each frequency hop.
• Second, fast-frequency hopping, in which the hop rate Rh is an integer multiple of
the M-ary FSK symbol rate Rs; that is, the carrier frequency will change (i.e.,
hop) several times during the transmission of one input-data symbol.
The spread spectrum technique of the FH variety is particularly attractive for
military applications. But, compared with the alternative spread spectrum technique,
DS/MPSK, the commercial use of FH/MFSK is insignificant, which is especially so
in regard to fast frequency hopping. The limiting factor behind this statement is the
expense involved in the employment of frequency synthesizers, which are basic to
the implementation of FH/MFSK systems. Accordingly, the FH/MFSK will not be
considered further.

Processing Gain of the DS/BPSK

Before closing this section on spread spectrum signals, it is informative to expand on the
improvement in SNR gained at the receiver output, mentioned earlier on. To this end,
consider the simple case of the DS/BPSK, in which the binary PSK, representing the
second stage of modulation in the transmitter, is coherent; that is, the receiver is
synchronized with the transmitter in all of its features. In Problem 9.34, it is shown that the
processing gain of a spread spectrum signal compared to its unspread version is
T
PG = -----b (9.125)
Tc
where Tb is the bit duration and Tc is the chip duration. With PG expressed in decibels, in
Problem 9.34 it is also shown that
10 log10 (SNR)O = 10 log10 (SNR)I + 10 log10 (PG) dB (9.126)
where (SNR)I and (SNR)O are the input SNR and output SNR, respectively. Furthermore,
recognizing that the ratio TbTc is equal to the number of chips contained in a single bit
duration, it follows that the processing gain realized by the use of DS/BPSK increases
with increasing length of a single period of the PN sequence, which was emphasized
previously.
Haykin_ch09_pp3.fm Page 560 Friday, January 4, 2013 4:58 PM

560 Chapter 9 Signaling over Fading Channels

9.14 Code-Division Multiple Access

Modern wireless networks are commonly of a multiuser type, in that the multiple
communication links within the network are shared among multiple users. Specifically,
each individual user is permitted to share the available radio resources (i.e., time and
frequency) with other users in the network and do so in an independent manner.
Stated in another way, a multiple access technique permits the radio resources to be
shared among multiple users seeking to communicate with each other. In the context of
time and frequency domains, we recall from Chapter 1 that frequency-division multiple
access (FDMA) and time-division multiple access (TDMA) techniques allocate the radio
resources of a wireless channel through the use of disjointedness (i.e., orthogonality) in
frequency and time, respectively. On the other hand, the code-division multiple access
(CDMA) technique, building on spread spectrum signals and benefiting from their
attributes, provides an alternative to the traditional techniques of FDMA and TDMA; it
does so by not requiring the bandwidth allocation of FDMA nor the time synchronization
needed in TDMA. Rather, CDMA operates on the following principle:

The users of a common wireless channel are permitted access to the channel
through the assignment of a spreading code to each individual user under the
umbrella of spread spectrum modulation.

This statement is testimony to what we said in the first paragraph of Section 9.13, namely
that spread spectrum signals provide a novel way of thinking about wireless
communications.
To elaborate on the way in which CDMA distinguishes itself from FDMA and TDMA
in graphical terms, consider Figure 9.32. Parts a and b of the figure depict the ways in
which the radio resources are distributed in FDMA and TDMA, respectively. To be
specific:
• In FDMA, the channel bandwidth B is divided equally among a total number of K
users, with each user being allotted a subband of width B/K and having the whole
time resource T at its disposal.
• In TDMA, the time resource T is divided equally among the K users, with each user
having total access to the frequency resource, namely the total channel bandwidth B,
but for only T/K in each time frame.
In a way, we may therefore think of FDMA and TDMA as the dual of each other.
Turning next to Figure 9.32c, we see that CDMA operates in a manner entirely
different from both FDMA and TDMA. Graphically, we see that each CDMA user has full
access to the entire radio resources at every point in time from one frame to the next.
Nevertheless, for the full utilization of radio resources to be achievable, it is necessary that
the spreading codes assigned to all the K users form an orthogonal set.
In other words, orthogonality is a common requirement to the FDMA, TDMA, and
CDMA, each in its own specific way. However, this requirement is easier to implement
practically in FDMA and TDMA than it is in CDMA.
In an ideal CDMA system, to satisfy the orthogonality requirement, the cross-
correlation between any two users of the system must be zero. Correspondingly, for this
Haykin_ch09_pp3.fm Page 561 Tuesday, January 8, 2013 11:17 AM

9.14 Code-Division Multiple Access 561

K
k
Bt Bt = B Bt = B 2
K 1
…
B = B t /M k
1 2 … k … K
…

2
1
Tt = T Tt Tt = T
T = T t /M
(a) (b) (c)

Figure 9.32 Resource distribution in (a) FDMA, (b) TDMA, and (c) CDMA. This figure shows the
essence of multiple access as in Figure 1.2 with a difference: Figure 9.32 is quantitative in its
description of multiple-access techniques.

ideal condition to be satisfied, we require that the cross-correlation function between the
spreading sequences (codes) assigned to any two CDMA users of the system must be zero
for all cyclic shifts in time. Unfortunately, ordinary PN sequences do not satisfy the
orthogonality requirement because of their relatively poor cross-correlation properties.
Accordingly, we have to look to alternative spreading codes to satisfy the orthogonality
requirements. Fortunately, such an endeavor is mathematically feasible, depending on
whether synchrony of the CDMA receiver to its transmitter is required or not. In what
follows, we describe the use of Walsh–Hadamard sequences for the synchronous case and
Gold sequences for the asynchronous case.

Walsh–Hadamard Sequences
Consider the case of a CDMA system, for which synchronization among users of the
system is permissible. Under this condition, perfect orthogonality of two spreading
signals, cj(t) and ck(t), respectively assigned to users j and k for different time offsets,
namely

R jk   =  Cj  t  Ck  t –   dt = 0 for j  k (9.127)
–

reduces to

R jk  0  =  Cj  t  Ck  t  dt = 0 for j  k and  = 0 (9.128)
–

where the asterisk denotes complex conjugation. It turns out that, for the special case
described in (9.128), the orthogonality requirement can be satisfied exactly, and the
resulting sequences are known as the Walsh–Hadamard sequences (codes).15
Haykin_ch09_pp3.fm Page 562 Friday, January 4, 2013 4:58 PM

562 Chapter 9 Signaling over Fading Channels

To construct a Walsh–Hadamard sequence, we begin with a 2  2 matrix, denoted by

H2, for which the inner product of its two rows (or two columns) is zero. For example, we
may choose the matrix

H2 = + 1 +1 (9.129)
+1 –1
the two rows of which are indeed orthogonal to each other. To go on and construct a
Walsh–Hadamard sequence of length 4 using H2, we construct the Kronecker product of
H2 with itself, as shown by
H4 = H2  H2 (9.130)
To explain what we mean by the Kronecker product in a generic sense, let A = {ajk} and
B = (bjk} denote m  m and n  n matrices, respectively.16 Then, we may introduce the
following rule:
The Kronecker product of the two matrices A and B is made up of an mn  mn
matrix, which is obtained from the matrix A by replacing its element ajk in
matrix A with the scaled matrix ajk B.

EXAMPLE 8 Construction of Hadamard–Walsh H4 from H2

For the example of (9.129) on matrix H2, applying the Kronecker product rule, we may
express the H4 of (9.130) as follows:

+ 1  H2 + 1  H2
H4 =
+ 1  H2 – 1  H2

+ 1 +1 +1 +1 (9.131)

= + 1 –1 +1 –1
+ 1 +1 –1 –1
+ 1 –1 –1 +1

The four rows (and columns) of H4 defined in (9.131) are indeed orthogonal to each other.
Carrying on in this manner, we may go on to construct the Hadamard–Walsh sequences
H6, H8, and so on.

In practical terms, a synchronous CDMA system is achievable provided that a single

transmitter (e.g., the base station of a cellular network) transmits individual data streams
simultaneously, with each data stream being addressed to a specific CDMA user (e.g.,
mobile unit).

Gold Sequences
Whereas Walsh–Hadamard sequences are well suited for synchronous CDMA, Gold
sequences, on the other hand, are well suited for applications in asynchronous CDMA;