Direct Sequence Spreading

Gene W. Marsh

I. A General Description of Direct Sequence Spreading

A. The standard view of a communication syste m

Channel Channel
Encoder Modulator Channel Demodulator Decoder

Fig. 1. A block diagram for a Standard Communication System.

B. A PN-Spread Communication System

Channel Channel
Encoder Modulator Channel Demodulator Decoder

PN Generator PN Generator

Fig. 2. A block diagram for a PN-spread Communication System.

1. To approach channel capacity, it is desirable to make the signal more noise-

like.Therefore, we introduce a random number generator, to make the modu-
lation appear random
C. A closer look at the modulator/demodulator .1
1. In the standard modulator pair shown in Figure 3, a bit determines whether
the transmit filter or its inverse is emitted every T seconds.
The waveform then passes through the channel, and is corrupted by noise.
The resulting signal is passed through a matched filter, and sampled every T
seconds.
1
a) T is known as the bit time and R = --- is known as the bit rate.
T
2. In the CDMA system shown in Figure 3, a slightly different thing happens.
A bit still enters the system every T seconds, but it is now multiplied by a

1. If you like your bits to be from the set { 0, 1 } , then you can replace bi and ci with 2bi 1 and 2ci 1 ,
below.

Wireless Communication - Page 1 of 16

Direct Sequence Spreadin g

Modulator Demodulator
bi { 1, 1}
xi ( t ) zi ( t ) yi ( t ) ri
g(t) Channel g(T t)
T
Fig. 3. The Standard Model

Modulator
bi { 1, 1 }
xij ( t ) zij ( t )
g(t) Channel

Demodulator
cij { 1, 1}
zij ( t ) yij ( t ) r ij sij di
g(Tc t) sij
0 j Lc
Tc

cij { 1, 1}

Fig. 4. The CDMA Model

faster moving, random sequence every Tc .

The result is sent through the channel, sampled, and the sampled signal is
multiplied by the corresponding random bit. The result is then summed.
a) Clearly things that move faster in time are wider in frequency. Hence the
name spreading.
b) Each of the short transmitted symbols is known as a chip.

Page 2 of 16
 Direct Sequence Spreadin g

1
c) The time Tc is then known as a chip time, while Rc = ----
T-c is known as
the chip rate or spreading rate.
d) The spreading bits cij are assumed to be i.i.d Bernoulli random variables
1
over { 1, 1} with parameter --2- .
T Rc
e) Clearly, Lc = ----
Tc - = -----
R is the number of chips per bit. This is often
referred to as the bandwidth expansion factor.

II. Performance issues

A. Does this little stunt cost us anything?
1. Let us examine the performance of each system assuming the channel simply
corrupts the signal using zero mean, white, additive Gaussian noise n ( t )
2
N0
with variance = ----- 2- , which is independent of the data symbols.
a) In the standard system, we have

yi(t) = bi g ( t ) *g(T t) + n ( t ) *g(T t)

After sampling, we find that E di bi = 2bi Eb and

N0 N0
di bi = -----
2

2
2- G ( f ) df = -----
2- Eb where Eb = g ( t ) 2 dt .

If we assume that there is no ISI, then successive samples of the noise are
independent. The threshold for decisions is placed at 0. Assume zeros
and ones are equally likely.
We can then calculate the probability of error as follows:

Eb 2E b
P ( 'e bi = 1) = P ( di > 0 bi = 1) = Q ----------------- = Q --------
-
N0 N0
------E
2 b

Eb 2Eb
P ( 'e bi = 1 ) = P ( di < 0 bi = 1) = 1 Q ----------------- = Q ---------
N0 N0
----- 2 b- E

Page 3 of 16
Direct Sequence Spreadin g

2Eb
P ( 'e ) = Q --------- (1)
N0
b) In the CDMA system, for each chip, we have

y ij(t) = bi cij g ( t ) *g(Tc t) + n ( t ) *g(Tc t)

The rij are then independent, with a Gaussian distribution such that

N0 N0
= bi cij E c and rij bi, c ij = -----
2

2
E rij bi, cj 2
- G ( f ) df = -----
-
2 Ec

where Ec = g ( t ) 2 dt .

After the multiplier, the sij clearly have mean E sij bi, cij = bi E c and
2
variance 2(sij bi, cij) = (rij bi, cij) . In particular, we should note that
the sij do not depend on the cij at all.
Now, if we sum the Lc chips corresponding to a particular bit, we find
that di is again Normal, with conditional mean and variance given by
2
N0

E di bi = bi Lc Ec and di bi = 2- Lc Ec . The probability of error
-----
is therefore still

2Lc E c 2E b

P ( 'e ) = Q --------------
N0 - = Q --------
N0- (2)

Rc
where E b = Lc Ec = -----E c .
R
c) Therefore, spreading has not cost us anything, except some complexity in
the modulator/demodulator.
(1) In general, this is only true for direct spreading in coherent commu-
nications systems.
(2) It should surprise no one that this is true, since, if you look carefully
at our picture, all we have done is modify the transmit filter, and
then made a matched filter for the receiver.

Page 4 of 16
 Direct Sequence Spreadin g

III. Pseudo-Noise (PN) Sequences

A. M-Sequences (Maximal Length Shift Register Sequences )
1. From Galois field theory, we have the notion of an M-sequence.
a) Let us consider Galois Field 2. This is composed of the elements 0 and 1
with addition defined by exclusive-or and multiplication defined by
and.
n
b) Let n be any positive integer. Then we can define a Galois Field with 2
elements by considering all possible polynomials of degree n 1 or less.
n
Addition is defined by polynomial addition modulo x + 1 , and multipli-
n
cation is polynomial multiplication modulo x .
n
c) For any such GF ( 2 ) , it is possible to find an element, , such that, if
n k n
0 GF ( 2 ) , then = for some 0 k < 2 . Thus, you can cycle
through all of the elements of the field by multiplying repeated multipli-
cation or division by . These elements are the primitive elements of the
field, and are represented by the primitive polynomials.
d) There are tables of primitive polynomials.
e) This sequence of n -bit numbers can be used to generate a sequence of
n
2 1 bits with useful properties (to be discussed later).
f) For any primitive polynomial, there are two ways to generate this bit
sequence. I will illustrate this by example. Let n = 5 . The primitive
5 2
polynomial for GF ( 32) is p ( x ) = x + x + 1 .
(1) The xor into the middle method is shown in Figure 5. In general,

out
5 4 3 2 1
x x x + x x

Fig. 5. A Galois Configuration

this is the easiest way to implement an M-sequence in software.

Page 5 of 16
Direct Sequence Spreadin g

The sequence produced by this machine is shown in Table 1.

Page 6 of 16
 Direct Sequence Spreadin g

Table 1: Sequence Generated by the Galois Configuratio na

State Output State Output State Output State Output

(Hex) Bit (Hex) Bit (Hex) Bit (Hex) Bit

1f 1 14 0 12 0 15 1
1d 1 a 0 9 1 18 0
1c 0 5 1 16 0 c 0
e 0 10 0 b 1 6 0
7 1 8 0 17 1 3 1
11 1 4 0 19 1 13 1
1a 0 2 0 1e 0 1b 1
d 1 1 1 f 1 1f 1
a. States are listed sequentially down the columns

(2) The shift around method is shown in Figure 5. This is the pre-
ferred method for implementing an M-sequence in hardware. The
sequence produced by this implementation is shown in Table 1.
Note that the output sequence shown here is the reverse of the one
in Table 1.
4 3 2 1 0 out
x x x x x

+
Fig. 6. A Fibonacci Configuration

Page 7 of 16
Direct Sequence Spreadin g

Table 2: Sequence Generated by the Fibonacci Configuratio na

State Output State Output State Output State Output

(Hex) Bit (Hex) Bit (Hex) Bit (Hex) Bit

1f 1 1b 1 2 0 1a 0
f 1 1d 1 1 1 d 1
7 1 e 0 10 0 6 0
3 1 17 1 8 0 13 1
11 1 b 1 4 0 19 1
18 0 15 1 12 0 1c 0
c 0 a 0 9 1 1e 0
16 0 5 1 14 0 1f 1
a. States are listed sequentially down the columns.

2. Some nice properties of M-sequences.

a) Even numbers of ones and zeros are produced.
b) The sequence produced is relatively uncorrelated with shifts of itself.
c) It is possible to generate the output bit in a different way.
(1) Associate with each bit in the register a 0 or a one i.e., create a mask
for the register.
(2) To compute a bit, and the mask with the register. The parity of
the result is the next output bit. Note: Do not destroy the contents of
the register, as you need it to compute the next state.
(3) The sequences generated in this way are shifted versions of the
original sequence. Indeed, all shifts of the sequence can be gener-
ated in this fashion.
d) Given a starting time and a clock rate, there are fast algorithms for com-
puting the current state of the shift register.
3. Thus, M-sequences work well for generating our spreading bits. Indeed, this
is the most common way to do it.
B. Acquisition of a Spread Spectrum Signal
1. Suppose we are generating our spreading codes using M-sequences. How do
we get the sequences synchronized in the transmitter and the receiver?
a) Technically, this is known in the industry as acquiring the signal.
b) At Qualcomm, we referred to this as making the jump to hyperspace.

Page 8 of 16
 Direct Sequence Spreadin g

2. This is the hardest problem in any spread spectrum system.

3. Here is how it is done.
a) First, tie the state of the shift register to absolute time. For instance, you
can declare it to have the all ones state at midnight on January 1, 1992,
and that you will clock it at a 1 MHz rate.
b) Now, suppose someone is sending you a signal, and you know what mask
he is using. You need a clock. Check it, and determine what the state of
the shift register is at this point in time, and initialize your shift register to
that value.
c) Now, you need to accumulate a sufficient number of chips. If you accu-
mulate N chips, the average value of the signal will be NEc , and the vari-
N0
ance will be NEc ------
2 .
d) After accumulating N chips, assume that your clock is off by one register
position, update it accordingly, and accumulate N new chips at this new
time hypothesis.
e) In the end, if you have an n -bit shift register, you should accumulate
n
2 1 different hypothesis, one for each time offset. The hypothesis with
the highest absolute value is declared to be the correct offset, the shift
register is initialized to that value, and demodulation can begin.
f) Using the statistics of the signal, it is possible to compute the probability
that you select the wrong time offset. Buy choosing a large enough value
of N , it is possible to make this acceptably small.
g) Just in case, after declaring a hypothesis to be correct, we wait for a bit to
see if the tracking loops take hold. This is a good way of detecting a bad
acquisition.

IV. Applications of Spread Spectrum

A. Low probability of Intercept (LPI) Communication s
1. If we simply take the original waveform and scale it down in time by Lc ,
then we decrease its energy by a factor of Lc .

1 1
Ec = g ( Lc t ) 2 dt = ----
L-c g ( t ) 2 dt = -----E
Lc b

After despreading, the energy has increased by a factor of Lc , as we saw
above. Therefore, we have spread the same energy over a a wider spectrum.
This makes the signal harder to distinguish from the noise background. It is

Page 9 of 16
Direct Sequence Spreadin g

used in aircraft communications so make the signal hard to detect. It would

be a shame for a stealth bomber to be given away by a radio transmission.
a) This is the root of my earlier comment about making the signal more
noise like.
2. This same thing can be seen by looking at the signal in time, where we have
divided the signal energy up over a sequence of smaller chips.
3. In most spread systems, the signal is buried in noise.
B. Code Division Multiple Access (CDMA)
1. Consider a system where we have 2 signals. These signals use the same code
bits, but one uses a time shifted version of the other, i.e. the second user is
+ n and l = ( j + n ) modL .
shifted in time by n chips. Let k = i + j----------
- c
Lc
The undesired user adds some interference. We are going to assume that
interference is Gaussian. We then have

yij(t) = bi cij g ( t ) *g(Tc t) + bk ckl g ( t ) *g(Tc t) + n ( t ) *g(Tc t) (3)

This will in turn give:

E r ij bi, cj = bi cij E c

2
N0
2
rij bi, cij = 2- Ec + E c
-----

E sij bi, cij = b i Ec

2
2(s ij bi, cij) = (r ij bi, cij)

E di bi = b i L c Ec

N0
di bi = -----
2 2
2- Lc Ec + Lc Ec

2
( L c Ec ) Eb
P ( 'e ) = Q ------------------------------------- = Q ------------------
- (4)
N0 N0
-----
2
- Lc E c + L c E c 2
------ + E c
2

a) Note that we assumed that the two users were lined up in time. In general,
this is not true, but it is the worst case. In that sense, the above is an upper

Page 10 of 16
 Direct Sequence Spreadin g

bound on what happens.

b) We also assumed that the other users interference was Gaussian. For a
long enough spreading sequence, this is not unreasonable.
2. Other user looks like white noise to the current bit, contributing an energy
equal to their chip energy. Thus, his interference is controlled.
3. It is possible to demodulate both users. We can use this to make a multiple
access scheme called Code Division Multiple Access (CDMA).
a) Let each user use either
(1) a unique spreading sequence, different from and uncorrelated with
all other users;
(2) the same spreading pattern, delayed by some number of chips from
his fellows.
b) Every user in the system can demodulate his own signal. The other users
simply raise the noise floor. For N users, we can bound the probability of
error as

Eb
P ( 'e ) Q --------------------------------------
- (5)
N0
2- + ( N 1 ) Ec
-----
c) It is theoretically possible to approach Shannon capacity using this
approach.
(1) If I remember right, theoretically is the correct word, because the
demodulator needs to demodulate each signal, and subtract off its
effect before it demodulates the next signal.
C. Anti-Jam (AJ) Communications
1. Jamming is the transmission of a signal in order to degrade a communication
system.
2. Design of a communication system which is robust against jamming can be
viewed as a game, where the transmitter tries to minimize the bit error rate,
and the jammer tries to maximize the bit error rate.
3. Often, the jammer is power limited. In this case, what the transmitter wants
to do is to force him to spread his power over the widest possible bandwidth,
so that he wastes as much of his power as possible.
4. Now, consider a channel with no noise, over which we send a BPSK signal.
Suppose the jammer transmits a single loud tone at the carrier frequency. In a
normal system, we can imagine the tone being loud enough that it would
dominate our own signal, and we would demodulate that rather than the
received signal. However, passing it through the despreader modulates that

Page 11 of 16
Direct Sequence Spreadin g

signal, making it look like wide band noise. Our own signal, on the other
hand, becomes despread, making it look like a tone. The spread system per-
forms better than the standard system in such a situation.
5. Direct sequence spreading is not the best or most common technique used for
AJ modems. Most of them use non-coherent FSK and hopping rather than
spreading.
D. Position Location
T
1. To do time tracking for the system above, one takes 2 samples, one at --4- and
3T e l
one at ------
4 . Call them s ij and sij . The time tracking metric is then computed
e
2
l
2
as TM = sij sij . If we are sampling too late, then
0j<L 0j<L
c c

this metric will be positive, and if we are sampling too early, then this metric
will be negative.
2. If we are off by a fraction of a chip, this metric drops appreciably. Therefore,
it is possible to do time tracking down to a fraction of a chip.
1
a) For example, in CDMA cellular, we track to -- 8- of a chip. We could do
better, but this is deemed sufficient.
3. If the spreading sequence is tied to absolute time (more on this below), it is
possible to use this feature to measure the time for a signal to travel from
source to destination. Given several independent transmitters, one can do tri-
angulation.
4. GPS works on a scheme similar to this.
E. Multi-path Mitigation
1. First, realize that we only really need concern ourselves with fading on the
chip level.
a) The wider your signal bandwidth, the less serious the effect of fading.
This is because fades only have a certain bandwidth in the frequency
domain. If your signal is wider than that bandwidth, it is unlikely that a
fade will occur which will crush all of the signal energy.
b) The accepted way to combat fading is with diversity. In this case, direct
sequence spreading gives you time diversity. Since a fade lasts for only a
finite amount of time, it is possible that only a fraction of the chips that
make up one bit will be faded.
2. Consider a case where we have 2 signals arriving with different delays, and
suppose that the difference in the delays, is such that Tc < . Let

Page 12 of 16
 Direct Sequence Spreadin g

+ n and l = ( j + n) modL . We then have

= nTc + c , k = i + j----------
Lc - c

yij(t) = bi cij g ( t ) *g(Tc t) + bk ckl g ( t + ) *g(Tc t) + n ( t ) *g(T t) (6)

This will in turn give:

E r ij bi, cj = bi cij E c

2
N0

rij bi, c ij = ----- 2
2- Ec + E

E sij bi, cij = b i Ec

2
2(s ij bi, cij) = (r ij bi, cij)

E di bi = b i L c Ec

N0
di bi = -----
2 2
2- Lc Ec + Lc E

2
( Lc E c )
P ( 'e ) = Q --------------------------------------
-
N0 2
-----
2 Lc Ec + Lc E
-

Now, we know that E E c , so that

L c Ec Eb
P ( 'e) Q ------------------
- = Q ------------------
- (7)
N0 N0
-----
2 - + E c -----
2 - + E c
3. What is the significance of this?
a) Any signal arriving at a time offset greater than one chip looks like white
noise to the current bit. Thus, ISI is controlled.
b) It is possible to demodulate the late path as well. This helps improve our
performance. Thus, the ISI can help you. This is a benefit a narrowband
system cannot provide.

Page 13 of 16
Direct Sequence Spreadin g

V. CDMA Cellular
A. Capacity of the CDMA System2
1. The capacity of the system is determined by the capacity of the mobile-to-
cell link.
a) This is principally because the mobiles must use smaller transmitters,
and cannot synchronize their signals.
2. Let us begin by realizing that the capacity of the system is controlled by the
signal to noise ratio needed to achieve an acceptable link.
a) Since we have a wideband system, it is possible to use powerful coding
techniques with little penalty. Because of this, the CDMA system only
Eb

requires ------ = 7 dB .
N 0 desired
3. Assume the following:
a) The system is interference limited, i.e. that the noise from other users in
our own cell is much greater than the background thermal noise.
b) Assume that we only care about users in our own cell.
c) All signals are power controlled so that the reach the cell with equal
power.3
4. Under these assumptions, we see from (5) that the signal-to-noise ratio is just
Eb Rc
-------------------------
- = R ( N 1) .
-----------------------
( N 1) E c

Eb
a) We need this to be equal to -----
- .
N0 desired
Rc 1
b) Solving gives N -----
R --------------------------
Eb .
------
N0 desired
5. The assumption that we are limited only by the users in our own cell is too
generous. We can mitigate this by appropriate scaling.
a) With an omnidirectional antenna, the number of users is decreased by
about a factor of F = 0.6 .
b) In the actual system, the cells use sectorized antennas. Each sector has a

2. This is a back-of-the-envelope approximation. It is relatively easy to understand, and gives approximate

results. Actual capacity is determined through simulation and testing .
3. This is crucial for achieving good capacity in the CDMA system .

Page 14 of 16
 Direct Sequence Spreadin g

field of view of 120 . Due to overlap between the sectors, this only buys
back a factor of G = 2.55 .
c) Note that both of these numbers are empirical.
6. In a CDMA system, it is possible to have a variable rate transmission, so that
one does not transmit when there is no data.
a) Speech does not occur 100% of the time in a conversation.
b) The amount of dead time varies with the language spoken. For English,
the empirical number is d = 0.4 .
c) By using a variable rate vocoder, and not transmitting during the silent
times, we get to take advantage of this, and our signal to noise ratio
increases appropriately.
7. Therefore, the total capacity of the system is approximately
Rc 1 1
N ----- ----------------------------
R Eb d- F G
-- (8)
------
N0 desired

8. For the CDMA system, R = 9600 Hz and Rc = 1.2288 MHz .

9. Using the numbers above, this gives us about 98 CDMA channels in a
1.25MHz bandwidth.
10. After blocking is considered, this yields about a factor of 20 increase in the
number of calls per cell.
a) This estimate varies considerably depending on who is doing it, as most
of the data is empirical.
B. Power Control and the Near/Far Problem
1. One of the principle assumptions made in the capacity calculation above was
that all mobiles reach the cell at the same power level. This assumption is
absolutely essential to operation of the system.
2. As mobiles are spread out over the entire cell, this is somewhat problemati-
cal, to say the least.
3. In order to combat this, we had to institute a method of power control. Very
simply, it works as follows.
a) the mobile and the cell both monitor their average received signal-to-
noise ratio (SNR). The mobile communicates his received SNR to the
cell.
b) The cell adjusts his output to provide an acceptable SNR to the cell.
c) If the SNR received at the cell is too low, the cell tells the mobile to
increase his power.

Page 15 of 16
Direct Sequence Spreadin g

d) If all cells controlling the mobile tell him to increase his power, he does.
Otherwise, he decreases his power.
e) In this way, as a mobile leaves one cell, he will begin to pick up another.
Both cells will tell him to increase his power, so that both can hear.
f) As he moves into the new cell, it will tell him to decrease his output, and
he will. Eventually, the old cell can no longer hear him.
g) This is the basic mechanism for a handoff, but the reality is much more
complicated.
C. A Final Note
1. There is a lot more that can be said about CDMA cellular. Due to time pres-
sure, I will stop here. [4] provides a good general description of what CDMA
cellular is and how it works.

VI. References
[1] John G. Proakis. Digital Communications. McGraw-Hill Book Company, New York,
1983.
[2] Marvin K. Simon, Jim K. Omura, Robert A. Sholtz, Barry K. Levitt. Spread Spectrum
Communications, Volume 1. Computer Science Press, Rockville, Maryland, 1985.
[3] Richard E. Blahut. Theory and Practice of Error Control Codes. Addison-Wesley Pub-
lishing Company, Reading, Massachusetts, 1983
[4] An Overview of the Application of Code Division Multiple Access (CDMA) to Digital
Cellular Systems and Personal Cellular Networks. Document Number EX60-10010.
Qualcomm, Inc., San Diego, CA, May 21, 1992.

Page 16 of 16