Chapter 8

Basic CDMA Concepts

8.1 INTRODUCTION

CDMA protocols constitute a class of protocols in which the multiple access property is
primarily achieved by means of coding. Each user is assigned a unique code sequence
used to encode the information-bearing signal. The receiver, knowing the code
sequences of the user, decodes a received signal after reception and recovers the
original data. Since the bandwidth of the code signal is chosen to be much larger than
the bandwidth of the information-bearing signal, the encoding process enlarges
(spreads) the spectrum of the signal and is therefore also known as spread-spectrum
modulation. The resulting encoded signal is also called a spread-spectrum signal, and
the CDMA protocols are often denoted as spread-spectrum multiple access (SSMA)
protocols.
It is the spectral spreading of the coded signal that gives the CDMA protocols
their multiple access capability. It is therefore important to know the techniques to
generate spread-spectrum signals and the properties of these signals.
The precise origin of spread-spectrum communications may be difficult to pin-
point because modern spread-spectrum communication is the outcome of developments
in many directions, such as high-resolution radars, direction finding, guidance,
correlation detection, matched filtering, interference rejection, jamming avoidance,
information theory, and secured communications [1–10].
The spread-spectrum modulation techniques were originally developed for use
in military radar and communication systems because of their resistance against
jamming signals and a low probability of detection. Only in recent years with new and

253
254

cheap technologies emerging and a decreasing military market have the manufacturers
of spread-spectrum equipment and researchers become interested in the civil
applications.
To qualify as a spread-spectrum modulation technique, two criteria must be
fulfilled [9]:

1. The transmission bandwidth must be much larger than the information bandwidth;
2. The resulting radio-frequency bandwidth is determined by a function other than the
information being sent (so the bandwidth is independent of the information signal).
This excludes modulation techniques like FM and pulse modulation (PM).

Therefore, spread-spectrum modulation transforms an information-bearing

signal into a transmission signal with a much larger bandwidth. This transformation is
achieved by encoding the information signal with a code signal that is independent of
the data and has a much larger spectral width than the data signal. This spreads the
original signal power over a much broader bandwidth, resulting in a low(er) power
density. The ratio of transmitted bandwidth to information bandwidth is called the
processing gain PG of the spread-spectrum system,

Bt
PG = (8.1)
Bi

where Bt is the transmission bandwidth and Bi is the bandwidth of the information-

bearing signal.
The receiver correlates the received signal with a synchronously generated
replica of the code signal to recover the original information-bearing signal. This
implies that the receiver must know the code signal used to modulate the data.
Because of the coding and the resulting enlarged bandwidth, spread-spectrum
signals have a number of properties that differ from the properties of narrowband
signals. The most interesting from a communication systems point of view are
discussed below. Each property is briefly explained with the help of illustrations, if
necessary, by applying direct-sequence spread-spectrum techniques.

1. Multiple access capability. If multiple users transmit a spread-spectrum signal at the

same time, the receiver can still distinguish between the users, provided each user
has a unique code that has a sufficiently low cross-correlation with the other codes.
Correlating the received signal with a code signal from a certain user will then only
despread the signal of this user, while the other spread-spectrum signals will remain
spread over a large bandwidth. Thus, within the information bandwidth, the power
of the desired user is much larger than the interfering power, provided there are not
too many interferers, and the desired signal can be extracted. The multiple access
capability is illustrated in Figure 8.1. In Figure 8.1(a), two users generate a spread-
spectrum signal from their narrowband data signals. In Figure 8.1(b) both users
 255

transmit their spread-spectrum signals at the same time. At the receiver only the
signal of user 1 is “despread” and the data recovered.

1
1&2
1

1
2

2 2
(a) (b)
Figure 8.1 Principle of spread-spectrum multiple access.

2. Protection against multipath interference. In a radio channel there is not just one
path between a transmitter and receiver. Due to reflections (and refractions), a
signal is received from a number of different paths. The signals of the different
paths are all copies of the transmitted signal but with different amplitudes and
phases. Adding these signals at the receiver is constructive at some of the
frequencies and destructive at others. In the time domain, this results in a dispersed
signal. Spread-spectrum modulation can combat this multipath interference;
however, the way in which this is done depends very much on the type of
modulation used. In the next section where CDMA protocols based on different
modulation methods are discussed, we show for each protocol how multipath
interference rejection is obtained.
3. Privacy. The transmitted signal can only be despread and the data recovered if the
code is known to the receiver.
4. Interference rejection. Cross-correlating the code signal with a narrowband signal
spreads the power of the narrowband signal, thereby reducing the interfering power
in the information bandwidth. This is illustrated in Figure 8.2. The spread-spectrum
signal (s) receives a narrowband interference (i). At the receiver the spread-
spectrum signal is despread while the interference signal spreads, making it appear
as background noise compared with the despread signal.
5. Anti-jamming capability, especially narrowband jamming. This is more or less the
same as interference rejection except the interference is now willfully inflicted on
the system. It is this property together with the next one that makes spread-spectrum
modulation attractive for military applications.
6. Low probability of interception (LPI) or covert operation. Because of its low power
density, the spread-spectrum signal is difficult to detect.
256

i s

s i

Figure 8.2 Interference rejection.

There are a number of modulation techniques that generate spread-spectrum

signals. We briefly discuss the most important ones:

Direct-sequence (DS) spread-spectrum. The information-bearing signal is multiplied

directly by a fast code signal.
Frequency hopping (FH) spread-spectrum. The carrier frequency at which the
information-bearing signal is transmitted is rapidly changed according to the code
signal.
Time-hopping (TH) spread-spectrum. The information-bearing signal is not transmitted
continuously. Instead, the signal is transmitted in short bursts where the times of the
bursts are decided by the code signal.
Chirp modulation. This kind of spread-spectrum modulation is almost exclusively used
in military radars. The radar continuously transmits a low-power signal whose
frequency is (linearly) varied (swept) over a wide range.
Hybrid modulation. Two or more of the above-mentioned spread-spectrum modulation
techniques can be used together to combine the advantages and, it is hoped, to combat
their disadvantages.

In Section 8.2, the above-mentioned modulation techniques are used to obtain

the multiple access capability that we want for CDMA (SSMA) protocols. Section 8.3
presents the code sequences and the properties of these sequences in detail.

8.2 SPREAD-SPECTRUM MULTIPLE ACCESS

We can classify the SSMA or CDMA protocols in two different ways: by concept or by
modulation method. The first classification gives us two protocol groups, averaging
systems and avoidance systems. The averaging systems reduce the interference by
averaging the interference over a wide time interval. The avoidance systems reduce the
interference by avoiding it for a large part of the time.
Classifying by modulation gives us five protocols, direct-sequence (or pseudo-
noise), frequency hopping, time-hopping, protocols based on chirp modulation, and
hybrid methods. Of these, the first (DS) is an averaging spread-spectrum protocol,
 257

while the hybrid protocols can be averaging protocols depending on whether DS is used
as part of the hybrid method. All the other protocols are of the avoidance type. Table
8.1 summarizes both ways of classification.

Table 8.1
Classifying SSMA Protocols
DS TH FH Chirp Hybrid
Averaging x x
Avoidance x x x x

In the following sections, CDMA protocols are discussed where a division has
been made that is based on the modulation technique.

8.2.1 DS

In the DS-CDMA protocols, the modulated information-bearing signal (the data signal)
is directly modulated by a digital code signal. The data signal can be either an analog or
digital signal. In most cases it will be a digital signal. What we often see in the case of a
digital signal is that the data modulation is omitted and the data signal is directly
multiplied by the code signal and the resulting signal modulates the wideband carrier. It
is from this direct multiplication that the DS-CDMA protocol gets its name.

Data
Data Data
Data Wideband
Wide-band
modulator code
code
modulator modulation
modulation

Carrier Code
Code
generator generator
generator generator

Figure 8.3 Block diagram of a DS-SSMA transmitter.

In Figure 8.3, a block diagram of a DS-CDMA transmitter is given. The binary data
signal modulates a radio frequency (RF) carrier. The modulated carrier is then
modulated by the code signal. This code signal consists of a number of code bits or
“chips” that can be either +1 or −1. To obtain the desired spreading of the signal, the
chip rate of the code signal must be much higher than the chip rate of the information
signal. For the code modulation various modulation techniques can be used but usually
258

some form of PSK such as BPSK, differential BPSK (D-BPSK), quadrature PSK
(QPSK), or MSK is employed.
If we omit the data modulation and use BPSK for the code modulation, we get
the block diagram given in Figure 8.4.

Binary data
Binary data Wide-band
Wideband
X modulator
modulator

Code
Code Carrier
Carrier
generator generator
generator generator

Figure 8.4 Modified block diagram of a DS-SS transmitter.

Data signal

Code signal

Data signal x code signal

BPSK-modulated signal

time
Figure 8.5 Generation of a BPSK-modulated spread-spectrum signal.

The DS-SS signal resulting from this transmitter is shown in Figure 8.5. In this figure,
ten code signals per information signal are transmitted (the code chip rate is 10 times
the information chip rate), so the processing gain is equal to 10. In practice, the
processing gain is much larger (in the order of 102 to 103).
After transmission of the signal, the receiver (seen in Figure 8.6) uses coherent
demodulation to despread the spread-spectrum signal using a locally generated code
sequence. To perform the despreading operation, the receiver must not only know the
 259

code sequence used to spread the signal, but the codes of the received signal and the
locally generated code must also be synchronized.
This synchronization must be accomplished at the beginning of the reception
and maintained until the whole signal is received. The synchronization/tracking block
performs this operation. After despreading, a data-modulated signal results and after
demodulation the original data are recovered.

Data
Code
Code Data
Data Data
demodulator demodulator
demodulator demodulator

Code
Code Code
Code Carrier
Carrier
synchr./
synchr. generator generator
tracking
/tracking generator generator

Figure 8.6 Receiver of a DS-SS signal.

In the previous section, a number of advantageous properties of spread-spectrum

signals were mentioned. The most important of those properties from the viewpoint of
CDMA protocols is multiple access capability, multipath interference rejection,
narrowband interference rejection, and, with respect to secure/private communication,
LPI. We explain these four properties in relation to DS-CDMA.

• Multiple access. If multiple users use the channel at the same time, multiple DS
signals will overlap in time and frequency. At the receiver, coherent demodulation
is used to remove the code modulation. This operation concentrates the power of the
desired user in the information bandwidth. If the cross-correlation between the code
of the desired user and the code of the interfering user is small, coherent detection
will only put a small part of the power of the interfering signals into the information
bandwidth.
• Multipath interference. If the code sequence has an ideal autocorrelation function,
the correlation function is zero outside the interval [ −Tc,Tc] where Tc is the chip
duration. This means that if the desired signal and a version that is delayed for more
than 2Tc are received, coherent demodulation treats the delayed version as an
interfering signal, putting only a small part of the power in the information
bandwidth.
• Narrowband interference. The coherent detection at the receiver involves a
multiplication of the received signal by a locally generated code sequence.
260

However, as we saw at the transmitter, multiplying a narrowband signal with a

wideband code sequence spreads the spectrum of the narrowband signal so that its
power in the information bandwidth decreases by a factor equal to the processing
gain.
• LPI. Because the direct sequence signal uses the whole signal spectrum all the time,
it has a very low transmitted power per hertz. This makes it very difficult to detect a
DS signal.

Apart from the above-mentioned properties, the DS-CDMA protocols have a

number of other specific properties that we can divide into advantageous (+) and
disadvantageous (−) behavior:

+ The generation of the coded signal is easy. It is done by simple multiplication.

+ Since only one carrier frequency has to be generated, the frequency synthesizer
(carrier generator) is simple.
+ Coherent demodulation of the spread-spectrum signal is possible.
+ No synchronization among the users is necessary.
− It is difficult to acquire and maintain the synchronization of the locally generated
code signal and the received signal. Synchronization has to take place within a
fraction of the chip time.
− For correct reception, the locally generated code sequence and the received code
sequence must be synchronized within a fraction of the chip time. This combined
with the nonavailability of large contiguous frequency bands practically limits the
spread bandwidth to 10 to 20 MHz.
− The power received from users close to the base station is much higher than that
received from users further away. Since a user continuously transmits over the
whole bandwidth, a user close to the base will constantly create a lot of interference
for users far from the base station, making their reception impossible. This near-far
effect is solved by applying a power control algorithm so that all users are received
by the base station with the same average power. However, this control proves to be
quite difficult.

8.2.2 FH

In FH-CDMA protocols, the carrier frequency of the modulated information signal is

not constant but changes periodically. During time intervals T, the carrier frequency
remains the same but after each time interval the carrier hops to another (or possibly the
same) frequency. The hopping pattern is decided by the code signal. The set of
available frequencies the carrier can attain is called the hop-set.
The frequency occupation of an FH-spread-spectrum system differs
considerably from a DS-spread-spectrum system. A DS system occupies the whole
frequency band when it transmits, whereas an FH system uses only a small part of the
bandwidth when it transmits but the location of this part differs in time.
 261

Suppose an FH system is transmitting in frequency band 2 during the first time

period (see Figure 8.7). A DS system transmitting in the same time period spreads its
signal power over the whole frequency band so the power transmitted in frequency band
1 will be much less than that of the FH system. However, the DS system transmits in
frequency band 1 during all time periods, while the FH system only uses this band part
of the time. On average both systems will transmit the same power in the frequency
band.
In a DS system, the narrowband interference is reduced by a factor of PG. In an
FH system however, it is the (raw) chip error rate that is reduced by PG. For strong
interference this can be a significant difference, where a single interferer may
“overload” a DS receiver but only cause a single chip error now and then in an FH
system (which the forward error control (FEC) scheme may easily handle).
The difference between FH-spread-spectrum and DH-spread-spectrum
frequency usage is illustrated in Figure 8.7.

Frequency Frequency
frequency

frequency

FH DS time
time
Time Time

Figure 8.7 Time/frequency occupancy of FH and DS signals.

The block diagram for an FH-CDMA system is given in Figure 8.8.

Figure 8.8 shows the block diagram of an FH-CDMA transmitter and receiver.
The data signal is baseband-modulated on a carrier. Several modulation techniques can
be used for this but it does not really matter which one is used for the application of
frequency hopping. Usually FM modulation is used for analog signals and frequency
shift keying (FSK) modulation for digital signals. Using a fast-frequency synthesizer
controlled by the code signal, the carrier frequency is converted up to the transmission
frequency.
The inverse process takes place at the receiver. Using a locally generated code
sequence, the received signal is converted down to the baseband-modulated carrier. The
data are recovered after (baseband) demodulation. The synchronization/tracking circuit
ensures that the hopping of the locally generated carrier synchronizes to the hopping
pattern of the received carrier so that correct despreading of the signal is possible.
262

Data Baseband Up Down Data Data

modulator converter converter demodulator

Synchr.
tracking

Code Frequency
Frequency Code
generator syntheziser
syntheziser generator

Figure 8.8 Block diagram of an FH-CDMA transmitter and receiver.

Within the FH-CDMA protocols, a distinction is made based on the hopping

rate of the carrier. If the number of hops is (much) greater than the data rate, it is called
fast-frequency-hopping (F-FH) CDMA protocol. In this case, the carrier frequency
changes a number of times during the transmission of 1 bit, so that 1 bit is transmitted
in different frequencies. If the number of hops is (much) smaller than the data rate, it is
called slow-frequency-hopping (S-FH) CDMA protocol. In this case, multiple bits are
transmitted at the same frequency.
The occupied bandwidth of the signal on one of the hopping frequencies
depends not only on the bandwidth of the information signal, but also on the shape of
the hopping signal and the hopping frequency. If the hopping frequency is much smaller
than the information bandwidth (which is the case in S-FH), then the information
bandwidth is the main factor that decides the occupied bandwidth. If however, the
hopping frequency is much greater than the information bandwidth, the pulse shape of
the hopping signal will decide the occupied bandwidth at one hopping frequency. If this
pulse shape is very abrupt (resulting in very abrupt frequency changes), the frequency
band will be very broad, limiting the number of hop frequencies. If we make sure that
the frequency changes are smooth, the frequency band at each hopping frequency will
be about 1/Th times the frequency bandwidth, where Th is equal to the hopping
frequency. We can make the frequency changes smooth by decreasing the transmitted
power before a frequency hop and increasing it again when the hopping frequency has
changed.
As was done with the DS-CDMA protocols, we discuss the properties of FH-
CDMA with respect to multiple access capability, multipath interference rejection,
narrowband interference rejection, and probability of interception.

• Multiple access. It is quite easy to visualize how the F-FH and S-FH CDMA
protocols obtain their multiple access capability. In the F-FH protocol, 1 bit is
transmitted in different frequency bands. If the desired user is the only one to
 263

transmit in most of the frequency bands, the received power of the desired signal is
much higher than the interfering power and the signal is received correctly.
In the S-FH protocol, multiple bits are transmitted at one frequency. If the
probability of other users transmitting in the same frequency band is low enough, the
desired user is received correctly most of the time. For those times that interfering
users transmit in the same frequency band, error-correcting codes are used to recover
the data transmitted during that period.

• Multipath interference. In the F-FH CDMA protocol the carrier frequency changes a
number of times during the transmission of 1 bit. Thus, a particular signal frequency
is modulated and transmitted on a number of carrier frequencies. The multipath
effect is different at the different carrier frequencies. As a result, signal frequencies
that are amplified at one carrier frequency will be attenuated at another carrier
frequency and vice versa. At the receiver, the responses at the different hopping
frequencies are averaged, thus reducing the multipath interference. This is not as
effective as the multipath interference rejection in a DS-CDMA system but it still
gives quite an improvement.
• Narrowband interference. Suppose a narrowband signal is interfering on one of the
hopping frequencies. If there are PG hopping frequencies (where PG is the
processing gain), the desired user will (on the average) use the hopping frequency
where the interferer is located 1/PG percent of the time. The interference is therefore
reduced by a factor PG.
• LPI. The difficulty in intercepting an FH signal lies not in its low transmission
power. During a transmission, it uses as much power per hertz as does a continuous
transmission. But the frequency at which the signal is going to be transmitted is
unknown and the duration of the transmission at a particular frequency is quite
small. Therefore, although the signal is more readily intercepted than a DS signal, it
is still a difficult task to perform.

Apart from the above-mentioned properties, FH-CDMA protocols have a number of

other specific properties that we can divide into advantageous (+) and
disadvantageous (−) behavior:

+ Synchronization is much easier with FH-CDMA than with DS-CDMA. With FH-
CDMA, synchronization has to be within a fraction of the hop time. Since spectral
spreading is not obtained by using a very high hopping frequency but by using a
large hop-set, the hop time will be much longer than the chip time of a DS-CDMA
system. Thus, an FH-CDMA system allows a larger synchronization error.
+ The different frequency bands that an FH signal can occupy do not have to be
contiguous, because we can make the frequency synthesizer easily skip over certain
parts of the spectrum. Combined with the easier synchronization, this allows much
higher spread-spectrum bandwidths.
264

+ Because FH-CDMA is an avoidance spread-spectrum system, the probability of

multiple users transmitting in the same frequency band at the same time is small. If
a user far from the base station transmits, it is received by the base station even if
users close to the base station are transmitting, since those users are probably
transmitting at other frequencies. Thus, the near-far performance is much better than
that of DS.
+ Because of the larger possible bandwidth a FH system can employ, it offers a higher
possible reduction of narrowband interference than a DS system.
− A highly sophisticated frequency synthesizer is necessary.
− An abrupt change of the signal when changing frequency bands lead to an increase
in the frequency band occupied. To avoid this, the signal turned off and on when
changing frequency.
− Coherent demodulation is difficult because of the problems in maintaining phase
relationship during hopping.

8.2.3 TH

In the TH-CDMA protocols, the data signal is transmitted in rapid bursts at time
intervals determined by the code assigned to the user.

Buffer Buffer

Data Slow in Data Fast in Data

Data
fast out modulator demodulator slow out

Code Carrier Carrier Code

generator generator generator generator

Figure 8.9 Block diagram of a TH-CDMA transmitter and receiver.

The time axis is divided into frames and each frame is divided into M time slots.
During each frame the user will transmit in one of the M time slots. Which of the M
time slots is transmitted depends on the code signal assigned to the user. Since a user
transmits all of its data in 1, instead of M time slots, the frequency it needs for its
transmission has increased by a factor M. A block diagram of a TH-CDMA system is
given in Figure 8.9.
 265

Frequency

Time

Figure 8.10 Time-frequency plot of the TH-CDMA protocol.

Figure 8.10 shows the time-frequency plot of the TH-CDMA systems.

Comparing Figure 8.10 with Figure 8.7, we see that the TH-CDMA protocol uses the
whole wideband spectrum for short periods instead of parts of the spectrum all of the
time.
Following the same procedure as for the previous CDMA protocols, we discuss
the properties of TH-CDMA with respect to multiple access capability, multipath
interference rejection, narrowband interference rejection, and probability of
interception.

• Multiple access. The multiple access capability of TH-SS signals is acquired in the
same manner as that of the FH-SS signals, namely by making the probability of
users’ transmissions in the same frequency band at the same time small. In the case
of time hopping, all transmissions are in the same frequency band so the
probability of more than one transmission at the same time is small. This is again
achieved by assigning different codes to different users. If multiple transmissions
do occur, error-correcting codes ensure that the desired signal can still be
recovered.
If there is synchronization among the users, and the assigned codes are such that no
more than one user transmits at a particular slot, then the TH-CDMA protocol
reduces to a TDMA protocol where the slot in which a user transmits is not fixed
but changes from frame to frame.
• Multipath interference. In the TH-CDMA protocol, a signal is transmitted in
reduced time. The signaling rate therefore increases and dispersion of the signal
will lead to overlap of adjacent bits much sooner. Therefore, no advantage is
gained with respect to multipath interference rejection.
• Narrowband interference. A TH-CDMA signal is transmitted in reduced time. This
reduction is equal to 1/PG where PG is the processing gain. At the receiver we only
266

receive an interfering signal during the reception of the desired signal. Thus, we
only receive the interfering signal 1/PG percent of the time, reducing the interfering
power by a factor PG.
• LPI. With TH-CDMA, the frequency at which a user transmits is constant but the
times at which a user transmits are unknown and the durations of the transmissions
are very short. Particularly when multiple users are transmitting, this makes it
difficult for an intercepting receiver to distinguish the beginning and end of a
transmission and to decide which transmissions belong to which user.

Apart from the above-mentioned properties, the TH-CDMA protocols have a

number of other specific properties that we can divide into advantageous (+) and
disadvantageous (−) behavior:

+ Implementation is simpler than that of FH-CDMA protocols.

+ It is a very useful method when the transmitter is average-power limited but not
peak-power limited since the data are transmitted is short bursts at high power.
+ As with the FH-CDMA protocols, the near-far problem is much less of a problem
since TH-CDMA is an avoidance system, so most of the time a terminal far from
the base station transmits alone, and is not hindered by transmissions from stations
close by.
− It takes a long time before the code is synchronized, and the time is short in which
the receiver has to perform the synchronization.
− If multiple transmissions occur, a lot of data bits are lost so a good error-correcting
code and data interleaving are necessary.

8.2.4 Chirp Spread Spectrum

Although chirp spread spectrum is not yet adapted as a CDMA protocol, for the sake of
completeness a short description is given here.

f f2
2

BB

f f1
1
TT

t1
t1 tt22

Figure 8.11 Chirp modulation.

 267

A chirp spread-spectrum system spreads the bandwidth by linear frequency

modulation of the carrier. This is shown in Figure 8.11.
The processing gain PG is the product of the bandwidth B over which the
frequency is varied and the duration T of a given signal waveform:

PG = BT (8.2)

8.2.5 Hybrid Systems

The hybrid CDMA systems include all CDMA systems that employ a combination of
two or more of the above-mentioned spread-spectrum modulation techniques. If we
limit ourselves to DS, FH, and TH modulations, we have four possible hybrid systems:
DS/FH, DS/TH, FH/TH, and DS/FH/TH.
The idea of the hybrid system is to combine the specific advantages of each of
the modulation techniques. If we take, for example, the combined DS/FH system, we
have the advantage of the antimultipath property of the DS system combined with the
favorable near-far operation of the FH system. Of course, the disadvantage lies in the
increased complexity of the transmitter and receiver. For illustration purposes, we give
a block diagram of a combined DS/FH CDMA transmitter in Figure 8.12.

Data Up
Up converter
converter

Code Code Frequency

generator generator synthesizer

Code clock

Figure 8.12 Hybrid DS-FH transmitter.

268

The data signal is first spread using a DS code signal. The spread signal is then
modulated on a carrier whose frequency hops according to another code sequence. A
code clock ensures a fixed relation between the two codes.

8.3 DESIGN OF PSEUDONOISE SEQUENCES

The objective of this section is to present an overview of the design and properties of
code sequences for CDMA systems. The choice of the type of code sequence for
CDMA systems is important with respect to the resistance against both multipath and
multiuser interference.

8.3.1 Basics

To combat these types of interference, two properties are very important: (1) Each code
sequence in the set must be easy to distinguish from a time-shifted version of itself, and
(2) each code sequence in the set must be easy to distinguish from (a possibly tune-
shifted version of) every other signal in the set.
The first property is important with respect to multipath propagation effects that
occur in mobile outdoor and indoor radio environments The second property plays an
important role with respect to the multiple access capability of the communications
system. In this chapter, we describe three types of code sequences: maximum length,
Gold, and kasami.
Before describing the construction of these code sequences and discussing the
correlation properties with respect to the requirements mentioned above, we present in
the next section some basic definitions of correlation functions and some information
about shift register sequences. Finally, some numerical examples of the aforementioned
code sequences are given.

Correlation functions
Primarily because system implementation is simpler, code sequences used for CDMA
communications systems are required to be periodic. If T is the period of the code
sequence denoted as X(t), then periodicity implies that X(t) = X(t+T) for all time
instants t and for each code sequence X in the set. For the code symbols of X we use the
notation Xm. Furthermore, we assume that the code symbols are {−1,1} The distinction
between code sequences X(t) and Y(t) is measured in terms of the correlation function
defined as

rX ,Y (τ ) = ∫ X (t )Y (t + τ )dt
0
(8.3)

This expression is shown as the cross-correlation function if X ≠ Y and as the

autocorrelation function if X = Y. Code sequence of interest are those consisting of a
number of time-limited pulses called chips. Then, the signal X(t) is written as
 269

∞
X (t ) = ∑ a ϕ (t − iT )
i =−∞
i
x c (8.4)

where axi is the ith code symbol user of x, ϕ(t) is the basic pulse wave form, and Tc is
the time duration of the pulse, termed chip duration and

Tc

∫ ϕ (t − iT )ϕ (t − jT )dt = 0
0
c c if i ≠ j (8.5)

If the delay τ is taken as a multiple of the chip duration (τ = lTc), then (8.3) is written as

N −1
rX ,Y (l ) = λ ∑ a ix a xi+1 (8.6)
i =0

where

Tc

λ = ∫ ϕ 2 (t )dt (8.7)
0

which equals Tc if ϕ(t) is a rectangle pulse duration Tc with amplitude.

In a DS-CDMA system, each data bit is multiplied with a user-specific code
sequence. For the current data bit and the previous data bit of a user with code sequence
Y, we use the notation by0 and by-1 , respectively.
Now consider the situation in a multipath radio environment where X is the code
sequence of the reference user and Y is the code sequence of a interfering user (or a
delayed part of code sequence X). The length of a code sequence is denoted by Nc. In
the receiver of the reference user a partial correlation operation is performed. This
situation is shown in Figure 8.13
Now we define the following two partial correlation functions
τ
R XY (τ ) = ∫ X (t )Y (t − τ )dt
0
(8.8)

N c Tc
Λ
R XY (τ ) = ∫τ X (t )Y (t − τ )dt (8.9)

b y0 by–1
Nc–1 0 Nc–1 0 Y

X
Nc–1 I–1 0
270

Figure 8.13 Cross-correlation of code sequence X of the reference user, and code sequence Y
being the code sequence of an interfering user or a delayed version of the reference
code sequence.

Assuming that τ is a multiple of the chip duration Tc implying τ = lTc, these partial
correlation functions are written as

l −1
R xy (l ) = ∑ a xi a iy−1 (8.10)
i =0

Λ N c −1

R xy (l ) = ∑a a
l =1
l
x
l −1
y (8.11)

If X = Y, then (8.10) and (8.11) are the partial autocorrelation functions and in the case
X ≠ Y, (8.10) and (8.11) are the partial cross-correlation functions.
With respect to the sign of by0 and by–1, there are two possibilities: either by0 and
–1
by have the same sign or they have the opposite sign. For the first case we have the
periodic correlation function:

Λ N c −1

θ xy (l ) = Rxy (l ) + R xy (l ) = ∑a a
m= 0
l
x
l −1
y (8.12)

Also, θxy(l) satisfies

θx , y ( N c − l ) = θ y , x (l ) (8.13)

which implies that θxy(l) is an even function with respect to N.

If by0 and by–1 have the opposite sign, then we have an a-periodic correlation
function

Λ Λ l −1 N c −1

θ xy (l ) = R xy (l ) − R xy (l ) = ∑ a a i
x
i −1
y − ∑a a i
x
i −1
y (8.14)
i =0 n=l

which is odd with respect to Nc, that is,

Λ Λ
θ x , y ( N c − l ) = − θ y , x (l ) (8.15)
 271

For analysis and comparison of code sequences it is convenient to define peak

correlation functions. For a set χ of periodic sequences, the peak periodic cross-
correlation magnitude θc is defined as

{
θc = max θxy (l ):0 ≤ l ≤ N c − 1, x ∈ χ , y ∈ χ , x ≠ y } (8.16)

The peak out-of-phase-periodic autocorrelation magnitude is defined as

{
θa = max θx (l ) :0 ≤ l ≤ N c − 1, x ∈ χ } (8.17)

Now the larger of θc and θa is denoted by θmax.

Λ
Analogously, we define the peak a-periodic cross-correlation magnitude θ c and
Λ
the peak out-of-phase a-periodic autocorrelation magnitude θ a , respectively

Λ
Λ 
θ c = max  θ xy (l ):0 ≤ l ≤ N c − 1, x ∈ χ , y ∈ χ , x ≠ y  (8.18)
 

and

Λ
Λ 
θ a = max  θ x (l ) :0 ≤ l ≤ N c − 1, x ∈ χ  (8.19)
 

Λ Λ Λ
The larger of θ c and θ a is denoted as θ max .
The two requirements mentioned in the introduction are now equivalently
reformulated: (1) For each sequence in the set, both the periodic out-of-phase
autocorrelation function and the a-periodic out-of-phase autocorrelation function must
be small for 1≤ l ≤ Nc–1. (2) For each pair of sequences x and y, both the periodic
autocorrelation and the a-periodic cross-correlation function must be small for all l.

Linear Shift Registers

Code sequences must have noiselike properties to meet the requirements mentioned in
the preceding section. However, because of implementation problems, the code
sequences are generated according to a periodic deterministic scheme. Periodic code
sequences having noiselike properties are called PN sequences. A proper way to
generate these code sequences is by using linear binary shift registers. An example for
such a register is presented in Figure 8.14. In this configuration, an XOR operation is
performed on the contents of register 2 and register 0 and the result is fed back to the
272

input of register 4. The shift direction of the shift register is from left to right (4-3-2-1-
0).
The register in Figure 8.14 with five sections generates a code sequence with length
Nc=25–1=31. We show in the next section that the feedback connections are chosen
very carefully in order to generate a code sequence with satisfactory correlation
properties. In general, the configuration of a linear binary shift register of n sections is
described by a generator polynomial, which is a binary polynomial of degree n. The
number n is the number of sections of the shift register:

h( x ) = h0 x n + h1 x n −1 +...+ hn −1 x + hn (h ∈{0,1})
i (8.20)

+
.

4 3 2 1 0

h(x) = x5 + x2 + 1

Figure 8.14 Two-tap linear binary shift register.

We use the notation used by Sarwate and Pursley [11] where h0 = hn = 1 and hi = 1 for i
≠ 0 and i ≠ n if there is a feedback connection from the ith cell. It is convenient and
conventional to represent the polynomial h(x) by a binary vector h = (h0, h1, ..., hn) and
to express this vector in octal notation. For example, the shift register in Figure 8.14 is
represented by generator polynomial h(x) = x5 + x2 +1. The binary representation of this
polynomial is 100101 and the octal representation is 45. If the sequence generated by
h(x) is denoted as u, then a circular shifted version of u is denoted as Tu where i is the
number of shifts. For instance, if u = 10011101 then Tu = l001110. Note that TNu =
T0u = u.

8.3.2 PN Codes

Three basic classes of code sequences suitable for CDMA applications are discussed in
this section:

• maximum length sequences;

• Gold sequences;
• Kasami sequences.
 273

Maximum length sequences

Maximum length code sequences are generated by a single linear shift register. As the
name suggests, maximum length code sequences are precisely the sequences of
maximum possible period (Nc = 2"-1) from an n-stage binary shift register with linear
feedback. To generate a maximum length code sequence, the generator polynomial
must be a primitive polynomial of degree n. Before explaining the term primitive
polynomial, first the following terms are defined:

Irreducible polynomial
This is a polynomial that can not be decomposed into other polynomials. Example: The
polynomial x2 + 2x +1 is not irreducible since x2 + 2x +1 = (x + 1)(x +1). The
polynomial x2 + x +1 is irreducible.

Exponent
The smallest positive integer p for which the polynomial h(x) is a divisor of (1 – xp) is
called the exponent of h(x).

Primitive polynomial
A primitive polynomial of degree n is an irreducible polynomial with exponent 2n –1.
It has been proven that the periodic autocorrelation for a maximum length
sequence u is given by


N l = 0 mod N c
θu (l ) =  c (8.21)
 − 1 l ≠ 0 mod N c

Thus, binary maximum length-sequence has a two-valued periodic autocorrelation

function.
Golomb [12] observed that if n ≠ 0 mod 4, there exists pairs of maximum length
sequences with three-valued cross-correlation functions, where the three values are
{−1,− t(n), t(n) − 2}with

 ( n + 1)
1+ 2 2 n odd
t ( n) =  ( n+2 )
(8.22)
1 + 2 2 n even

A cross-correlation function taking on these values is called a preferred three-valued

cross-correlation function and the corresponding pair of maximum length sequences
(polynomials) is called a preferred pair of maximum length sequences. For this
preferred pair of maximum length sequences we find

θ a = θ c = t ( n) (8.23)
274

Figures 8.15 to 8.17 show preferred pairs of maximum length sequences for periods 31,
63, and 127, respectively.

Figure 8.15 Preferred pairs of maximum length sequences of period 31. The vertices of every
triangle form a maximal connected set.

It is clear from these figures that only very small sets of maximum length
sequences can have good periodic cross-correlation properties. For multiple access
systems, it is desirable to obtain larger sets of sequences of period Nc = 2n – 1 which
have the same bound θ≤ t(n) on the peak periodic cross-correlation as do the
maximal connected sets. Although there are no analytical results that give the values of
Λ
the maximal a-periodic cross-correlation θ c for maximum length sequences generated
Λ
by preferred pairs, Massey and Uhran [13] have obtained bounds on θ c . They state that
if the maximum of the periodic cross-correlation and the periodic autocorrelation θmax
equals t(n), then the bound for the a-periodic cross-correlation is

 n −1
2 + 2 2 + 1
n
Λ n even
θc =  (8.24)
 2 n −1 + 2 2 + 1
( n −1 )
n odd

We did not find a bound in the literature for out-of-phase a-periodic autocorrelation.
 275

Figure 8.16 Preferred pairs of maximum length sequences of period 63. Every pair of adjacent
vertices is a maximal connected set.

Gold Sequences
One important class of periodic sequences that provides larger sets of sequences with
good periodic cross-correlation is the class of Gold sequences. A set of Gold sequences
of period N = 2n−1 consists of (Nc + 2) sequences for which the maximum periodic
cross-correlation θc = t(n) with t(n) given by (8.22). A set of Gold sequences is
constructed from appropriately selected maximum length sequences. Suppose a shift
Λ Λ
register polynomial f(x) factors into h(x) h( x ) where h(x) and h( x ) have no factors in
common. Then the set of all sequences generated by f(x) is just the set of all sequences
of the form a ⊕ b, where a is some sequence generated by h(x), b is some sequence
Λ
generated by h( x ) , and a and b are not necessarily nonzero sequences. Now suppose
Λ
that h(x) and h( x ) are two different primitive binary polynomials of degree n that
generate the maximum length sequences u and v, respectively, of period Nc = 2n −1. If y
Λ
denotes a nonzero sequence generated by f(x) = h(x) h( x ) , then either
276

Figure 8.17 Preferred pairs of maximum length sequences of period 127. Every set of six
consecutive vertices is a maximal connected set.

y = T iu (8.25)

or

y = T jv (8.26)

or

y = T iu ⊕ T jv (8.27)

where 0 ≤ i, j ≤ Nc −1 and where, as before, T i u ⊕ T jv denotes the sequence whose kth

element is ui+k ⊕ vj+k. From this it follows that y is some phase of some sequence in the
set G(u,v) defined by

Λ
G ( u, v ) = {u, v, u ⊕ v, u ⊕ Tv, u ⊕ T 2
v ,..., u ⊕ T N c −1v} (8.28)

Note that G(u, v) contains Nc+ 2 = 2n + 1 sequences of period Nc.

Figure 8.18 shows a possible configuration to generate a Gold sequence of
length Nc = 63. The octal representation of the shift registers producing u and v are 141
Λ
and 163, respectively (h(x) = 1+ x5 + x6 and h( x ) = l + x + x4 + x5 + x6).
 277

In 1969, Gold [14] published the construction G(u,v). In particular, he showed

that if (u,v) is any preferred pair of maximum length sequences, then G(u,v) has peak
periodic correlation parameter θc = t(n). For all the Gold codes that have θc = t(n), the
bound for the a-periodic cross-correlation is

 n −1
 2 + 2 +1
n
Λ 2
n even
θc =  ( n − 1)
(8.29)
 2 n −1 + 2 2 + 1 n odd


We did not find a bound in the literature for out-of-phase a-periodic autocorrelation.

h(x) = 1+x5 + x6
+

5 4 3 2 1 0

+ + + +

5 4 3 2 1 0

Λ
4 5 6
h( x ) = l + x + x + x + x

Figure 8.18 Shift register configuration for set of Gold codes with length Nc = 63.

Kasami Sequences
Let n be even and u denote an maximum length sequence of period Nc = 2n −1
generated by h(x). Consider the sequence w of period 2n/2 −1, which is generated by the
Λ
polynomial h( x ) whose roots are the s(n)th powers of the roots of h(x), where
s( n ) = 1 + 2
n
2
(8.30)
278

Λ
Furthermore, since h( x ) is shown to be a polynomial of degree n/2, w is a maximum
length sequence of period 2n/2 −1. Now consider the sequences generated by the
Λ
polynomial h(x) h( x ) of degree 3n/2. Clearly, any such sequence must be of one of the
forms T iu, T jw or T iu ⊕ T jw, where 0 ≤ i ≤ 2n −1 and 0 ≤ j≤ 2n/2 −1. Thus, any
Λ
sequence y of period 2n −1 generated by h(x) h( x ) is one of the sequences in the set
G(u,w) defined by

{u, w, u ⊕ w, u ⊕ Tw, u ⊕ T }
Λ
G ( u , w) = w,..., u ⊕ T 2 w
2 2n
(8.31)

This set of sequences is called the small set of Kasami sequences denoted by Ks(u) in
honor of Kasami, who discovered that the correlation functions for sequences belonging
to Ks(u) take on the values in the set {−1, −s(n), s(n) −2} with s(n) given by (8.30).
Consequently, for the set Ks(u),

θ max = s(n) = 1 + 2
n
2
(8.32)

Notice that θmax for the set Ks(u) is approximately one half of the value of θmax achieved
by the sets discussed previously. On the other hand, Ks(u) contains only 2n/2 = (Nc+1)1/2
sequences, while the sets of Gold code sequences contain Nc +2 sequences. An example
for a shift register configuration that produces a small set of Kasami codes of length Nc
= 63 is presented in Figure 8.19. The octal representation of the shift registers
Λ
producing u and v are 103 and 15, respectively, h(x) = 1 + x + x6 = 1000011 and h( x )
= 1 + x2 + x3 = 1101.
Besides the small sets of Kasami sequences, there are also large sets of Kasami
sequences. Let n be even and let h(x) denote a primitive binary polynomial of degree n
that generates the maximum length sequence u. Let w = u[s(n)] denote a maximum
length sequence of period 2n/2 −1 generated by the primitive polynomial h ′ (x) of degree
Λ
n/2 and let h( x ) denote the polynomial of degree n that generates u[t(n)].
Λ
Then, the set of sequences of period Nc generated by h(x) h’(x) h( x ) , called the
large set of Kasami sequences and denoted by KL(u), is as follows:
1. If n = 2 mod 4, then

2 2 − 2 
n

{
K L (u) = G (u, v )U  U T i w ⊕ G (u, v )  } (8.33)
 i = 0 
where v = u[t(n)] and G(u,v) is defined in (8.28).
 279

h(x) = 1 + x + x6
+

5 4 3 2 1 0

+ +

2 1 0

Λ
2 3
h( x ) = 1 + x + x

Figure 8.19 Shift register configuration for the small set of Kasami codes with length Nc = 63.

2. If n = 0 mod 4 then

2 2 − 2 i 
n

{
K L (u) = H t ( n ) (u)U  U T w ⊕ H t ( n ) (u)  }
 i =0 

U{v ⊕ T k w:0 ≤ j ≤ 2,0 ≤ k < (2 )}

−1
( j) n
1
3
2
(8.34)

where v(t) is the result of decimating T i(u) by t(n) and Ht(n)(u) is defined as

u, u ⊕ v ( 0) , u ⊕ Tv ( 0) , ..., u ⊕ T c 3 −1v ( 0) 

N

 
H t ( n ) (u) =  u, u ⊕ v (1) , u ⊕ Tv (1) , ..., u ⊕ T 3 −1v (1) 
Nc
(8.35)
u, u ⊕ v ( 2 ) , u ⊕ Tv ( 2 ) , ..., u ⊕ T N c 3 −1v ( 2 ) 
 

In either case, the correlation functions for KL(u) take on the values in the set {−1,−t(n),
t(n) −2, −s(n), s(n) −2} and θmax = t(n). If n = 2 mod 4 then KL(u) contains 2n/2(2n + 1)
sequences while if n = 0 mod 4, KL(u) contains 2n/2(2n + 1) sequences.
An example of a shift register configuration that produces a large set of Kasami
codes of length Nc = 63 is presented in Figure 8.20. The octal representation of the shift
280

registers producing the sequences u, v, and w are 103, 15, and 147, respectively, (h(x) =
Λ
1 + x + x6 , h( x ) = 1 + x2 + x3, and h ′ (x) = 1 + x + x2 + x5 + x6).

Λ
6
h( x ) = 1 + x + x

5 4 3 2 1 0

h(x) = 1 + x2 + x3
2 1 0 +

+ + +

5 4 3 2 1 0

2 5 6
h ′ (x) = 1 + x + x + x + x

Figure 8.20 Shift register configuration for the large set of Kasami codes with length Nc = 63.

Numerical Examples
Table 8.2 presents a number of examples of PN code sequences. Three types of
sequences are considered. Gold sequences (G), small-set Kasami sequences (KS), and
large-set Kasami sequences (KL). The first column denotes the code length Nc, the third
column shows the maximum number of code sequences per set, and in the fourth
column the values that periodic cross-correlation can take are given. The last three
columns show the octal representation of the polynomials describing the linear binary
shift registers used to generate the desired code sequence. Table 8.2 only shows some
examples for pseudo-random code sequences.
 281

Table 8.2
Examples of Gold Codes, Small Sets of Kasami Codes, and
Large Sets of Kasami Codes and Their Basic Properties

Nc Code No. Periodic Pol. 1 Pol. 2 Pol. 3

Cross-Correlation
31 G 33 {−9, −1, −7} 45 67 -
63 G 65 {−17, −1,15} 141 163 -
KS 8 {−9, −1,7} 103 15 -
KL 520 {−17, −9, −1, 103 15 147
7,15}
127 G 129 {−17, −1,15} 211 277 -
255 KS 16 {−17, −1,15} 435 023 -
KL 4111 {−33,−17, −1, 435 023 675
15,31}
511 G 513 {−33, −1,31} 1021 1333 -

G, Gold codes; KS, small sets of Kasami codes; KL, large set of Kasami codes.

From Table 8.2 it is clear that for a code sequence with length Nc = 63, the small
set of Kasami codes has better correlation properties than the set of Gold codes.
However, the number of code sequences in the set is less (8 instead of 33 in the case of
Gold codes). The large set of Kasami sequences has the same cross-correlation
properties as the set of Gold codes, but the number of code sequences in the set is
higher (520).
We plotted the correlation functions of some Gold and Kasami code sequences.
For the autocorrelation we used the code generated by the unshifted registers. The
cross-correlation function was produced by considering one code generated by
unshifted registers and one code generated by shifting one of the registers 1 chip period.
Figures 8.21(a) and 8.21(b) show the periodic autocorrelation and cross-
correlation of a Gold code with length Nc=127. We considered the codes x = u ⊕ w and
y = u ⊕ T1w. It is clear that the periodic out-of-phase autocorrelation and cross-
correlation have the values {–17, –1, 15}. Figures 8.22(a) and 8.22(b) show the a-
periodic out-of-phase autocorrelation and cross-correlation for the same code sequence.
The results of these figures are consistent with Table 8.2.
It is not possible to find preferred pairs with sequence-length Nc = 255, and
hence Gold codes with length Nc = 255 having a three-valued periodic cross-correlation
function do not exist. Figures 8.23(a) and 8.23(b) show the periodic autocorrelation
and cross-correlation for a code sequence generated by the XOR-sum of two not
preferred maximum length sequence with length Nc = 255.
282

Autocorrelation Cross-correlation

Chip delay Chip delay

(a) (b)

Figure 8.21 (a) Periodic autocorrelation and (b) cross-correlation for Gold code with code length
Nc = 127. Generator polynomials 211 and 277 (octal).

Cross-correlation
Autocorrelation

Chip delay Chip delay

(a) (b)

Figure 8.22 (a) a-Periodic autocorrelation and (b) cross-correlation for Gold code with code
length Nc = 127. Generator polynomials 211 and 277 (octal).

Autocorrelation Cross-correlation
 283

Chip delay Chip delay

(a) (b)

Figure 8.23 (a) Periodic autocorrelation and (b) cross-correlation for Gold code with Nc = 255.
Generator polynomials 717 and 765 (not preferred sequences).

Autocorrelation Cross-correlation

Chip delay Chip delay

(a) (b)

Figure 8.24 (a) Periodic autocorrelation and (b) cross-correlation functions for small-set Kasami
code with Nc=255. Generator polynomials 435 and 023.

Code sequences with length Nc=255 that have better correlation properties are
the Kasami sequences. Figures 8.24(a) and 8.24(b) show the autocorrelation and cross-
correlation of a small set of Kasami codes. Figures 8.25(a) and 8.25(b) show the
autocorrelation and cross-correlation of a large set of Kasami codes.
284

Figure 8.25 (a) Periodic autocorrelation and (b) cross-correlation functions for large Kasami
code with Nc = 255. Generator polynomials 435, 023, and 675.

8.3.3 Random Wave Approximation

In a multipath propagation environment such as the indoor radio environment, there is

overlap between shifted data bits of the reference user and of other users. This situation
is depicted in Figure 8.26.
This implies that the receiver suffers from interference caused by both the
desired signal and by the signals from other users. The interference power is determined
by the characteristics of the multipath channel (time delay, phase, and amplitude) and
by the correlation properties of the code sequences used.

b11 t=T b10 t=0 b1-1

Path 1 = j, user 1

b11 b10 b1-1

Path 1 < j, user 1

Figure 8.26 Overlap of bits (and code sequences) due to multipath interference.
 285

The ACF determines the amount of self-interference, while the cross-correlation

between the desired code sequence and code sequences of other users determines the
amount of multiuser interference. To simplify these performance calculations, an
approximation of the code sequence characteristics is applied. The starting point for this
approximation is the expression for the correlation variance defined as

ψ a = E b1−1 R(τ ) + b10 R(τ ) 

Λ
(8.36)
 

It is worth mentioning here that the interference power is related to this variance. In
(8.36), b1−1 and b10 are, respectively, the previous and the next bit with values {−1,1}
Λ
having equal probability of occurrence. The functions R(τ) and R (τ ) are the partial
correlation functions. If the delay is a multiple of the chip duration (τ = pTc), then the
expressions for the discrete correlation functions are

p
1
R ( p) =
Nc
∑a a i
1
i− p
k (8.37)
i =0

Λ N c −1
1
R ( p) =
Nc
∑a a i
1
i− p
k (8.38)
i = p +1

where aki is the ith code symbol of the sequence of user k and Nc is the length of the
code sequence. To compute the variance in (8.36), we average over all possible time
delays and over all possibilities for b1−1 and b10. The time delays in a multipath
environment are obviously not multiples of the code symbol duration. However, if we
know the discrete correlation function, then the correlation for a non discrete time delay
is very easy to calculate by considering that the correlation function is linear between
two discrete time delays, as shown in Figure 8.27.
To find an expression for the variance in (8.36), we approximate the Gold and
Kasami sequences by a random code sequence consisting of Nc code symbols. For a
random code sequence, the code symbols aki = 1 and aki = −1 occur with equal
probability. Assuming that the data bits are –1 or 1 with equal probability, the variance
in (8.36) is written as

2 2
1 
E  R(τ ) + R(τ )  + E  R(τ ) − R(τ ) 
Λ 1 Λ
ψa = (8.39)
2   2  

To simplify this expression further, it is instructive to observe one chip of normalized

duration overlapped by another chip with delay τ. This situation is depicted in Figure
8.28.
286

TC 2TC 3TC 4TC

Figure 8.27 Part of a correlation function.

aki

2τ
0 1
i-1
ak ak i

Figure 8.28 Overlap of two chips, aki and aki-1.

For the two chips aki and akj−1 there are two possibilities: either they have the same sign
or they have opposite sign. Both situations occur with the same probability in case of a
random sequence. The overlap area denoted here by AO determines the contribution of
chip aki to the correlation function (normalized on the code length Nc). In the case of
equal bits, the overlap area, normalized on the code length N is 1/N, while in the case of
unequal bits, the part between 0 and τ cancels the part between τ and 2τ, implying that
the overlap area is (1−2τ)/Nc. In terms of the overlap area:

1
A0 = if a ki−1 = a ki
Nc
(8.40)
 287

1 − 2τ
A0 = if a ki−1 = a ki
Nc

Now, the expectation of A02 is calculated by averaging over all possible delays

1 (1 − 2τ ) 2
1

E[ A ] =
1 1 2
2
0
2 Nc 2 + ∫
2 0 Nc 2 dτ =
3N c 2
(8.41)

Now, the final simplified expression for the variance in (8.35) is

ψ a = N c E[ A 2 ] =
2
(8.42)
3N c

Before applying this approximation, we verify the approximation for a number of

practical Gold sequences and Kasami sequences. Therefore, a large number of these
code sequences has been generated, and for each combination of code sequences the
expectation of the correlation variance has been calculated by averaging over all
possible delays. The results are shown in Figures 8.29(a) to 8.29(e). To explain these
figures, we use the Gold code of length 31 as an example. The plot shows a number of
bars and a line. The line represents the approximated value for the correlation variance
given by (8.42), and each bar represents the simulated correlation variance between
two code sequences of the same length. Earlier, we found for example that it is
possible to generate 65 code sequences of length 63. For all figures, we only show 50
bars for clarity of the plot. In each plot we first observe a large peak, which is actually
the average correlation variance of the autocorrelation function. All the other bars
present the average correlation variance of cross-correlations with other codes. By
observing Figures 8.29(a) to 8.29(e), it is reasonable to conclude that the approximation
of (8.36) is rather good. To strengthen this conclusion, quantitative results are shown in
Table 8.3.
Code correlation variance Code correlation variance

Code number Code number

(a) (b)
288

Code correlation variance

Code correlation variance

Code number Code number

(c) (d)
Code correlation variance

Code number
(e)

Figure 8.29 (a) Gold code (Nc = 31), (b) Kasami code (Nc = 31), (c) Gold code (Nc = 127), (d)
Kasami code (Nc = 255), and (e) Gold code (Nc = 511).

Table 8.3
Quantitative Comparison of Simulation Results and Approximation
for Correlation Variance for Several Practical Code Sequences
 289

Approximation Sample Mean ms Sample Standard Deviation σS

2
ψa =
3N c
Gold code 0.010582 0.010406 0.002182
Nc = 63
Kasami code 0.010582 0.0119 0.005654
Nc = 63
Gold code 0.005249 0.005553 0.000722
Nc = 127
Kasami code 0.002614 0.0025691 0.000289
Nc = 255

.
8.3.4 Conclusions

In this section, a description was given of three types of pseudo-noise sequences

suitable for DS-CDMA systems. Gold codes and Kasami code sequences are especially
suitable, since it is possible to generate a relatively large number of code sequences
with bounded periodic correlation functions. The basic properties of the code sequences
were discussed and shift register configurations to generate the sequences were shown.
The periodic correlation functions are important when the desired data bit overlaps with
a previous or next data bit having the same sign. In the situation where the previous or
next overlapping bit has the opposite sign, the so-called a-periodic correlation function
is very important. For this a-periodic correlation function, well-defined bounds are
found in literature. However, by simulation, we saw that the peaks in the a-periodic
correlation function can be much larger than the bounded periodic correlation function.
Furthermore, it was shown that Gold code sequences and Kasami code
sequences cannot be generated with arbitrary length. In the first place, the length of a
code sequence can always be written as 2n − l where n is an integer. This implies that
we can generate sequences with lengths 31, 63, 127, 255, and so forth. Gold sequences
cannot be generated for integer values n being a multiple of 4. This implies that we
cannot generate Gold sequences with length Nc = 255 since this would imply n = 8,
which is a multiple of 4. On the other hand, Kasami sequences cannot be generated for
odd values of n. For example, Kasami code sequences with length Nc = 127 and Nc =
511 are not possible. Code sequences with these lengths must be generated by a Gold
shift register generation.
PASCAL software was developed to generate Gold sequences and Kasami
sequences. For simplification of performance calculations, an approximation of the
correlation variance is desirable. A very simple expression for this correlation variance
results if we approximate real code sequences of length Nc by random sequences of
length Nc. This approximation has been verified by simulation and the results show that
the approximation is quite good.
290

REFERENCES

[1] R. Prasad, CDMA for Wireless Personal Communications, Norwood, MA:

Artech House, 1996.
[2] Special issue on spread-spectrum communication, IEEE Trans. Commun., Vol.
COM-30, May 1982.
[3] M.K. Simon, J.K. Omura, R.A. Scholtz, and B.K. Levitt, “Spread-spectrum
communications, Vol. I, II, III, Comp. Sci., 1985.
[4] R.A. Scholtz, “The spread-spectrum concept,” IEEE Trans. Commun., Vol.
COM-25, pp. 748–755, August 1977.
[5] D.J. Torrieri, Principles of Secure Communication Systems, Norwood, MA:
Artech House, 1985.
[6] G.R. Cooper and C.D. McGillem, Modern Communication and Spread-
Spectrum, New York: McGraw-Hill Company, 1986.
[7] S.G. Glisic and P.A. Leppanen (Eds.), Code Division Multiple Access
Communications, Boston: Kluwer Academic Publishers, 1995.
[8] A.J. Viterbi, CDMA Principles of Spread-Spectrum Communications, Reading,
MA: Addison-Wesley Publishing Company, 1995.
[9] R.C. Dixon, Spread-Spectrum Systems, New York: John Wiley & Sons, 1984.
[10] A.M. Viterbi and A.J. Viterbi, “Erlang capacity of a power controlled CDMA
system,” IEEE J. Selected Areas in Comm., Vol. 11, pp. 892–899, August 1993.
[11] D.P. Sarwate and M.B. Pursley, “Crosscorrelation properties of pseudo-random
and related sequences,” Proc. IEEE, Vol. 68, No. 5, May 1980.
[12] S.W. Golomb, Shift Register Sequences, San Francisco, CA: Holden-Day, 1967.
[13] J.L. Massey and J.J. Uhran, Jr., “Final report for multipath study,” Contract
NASS-10786, Univ. Notre Dame, IN, 1969.
[14] R. Gold, “Optimal binary sequences for spread spectrum multiplexing,” IEEE
Trans. Inf. Theory, Vol. IT-13, pp. 619–621, October 1967.