Chap12 Spread Spectrum Signals

Digital Communications
Chapter 12 Spread Spectrum Signals for Digital

Communicaitons
Po-Ning Chen, Professor
Institute of Communication Engineering
National Chiao-Tung University, Taiwan
Digital Communications: Chapter 12 Ver 2010.12.25 Po-Ning Chen 1 / 87
12.1 Model of spread spectrum digital
communications system
What is spread spectrum communications?
A rough denition: The signal spectrum is wider than
necessary, i.e., 1]T.
Recollection: Sampling theorem
A signal of (baseband or single-sided) bandwidth W
base
can be
reconstructed from its samples taken at the Nyquist rate (=
2W
base
samples/second) using the interpolation formula
s(t) =

n=
s _
n
2W
base
_sinc _2W
base
_t
n
2W
base
__
Thus, T =
1
2W
base
.
However, for a signal consume W = W
pass
= 2W
base
Hz bandwidth
after upconvertion, we should put T =
1
W
.
Thus, T =
1
W
or W =
1
T
.
Since we have spectrum wider than necessary, we have
extra spectrum to make the system more robust.
Applications of spread spectrum technique
Channels with power constraint
E.g., power constraint on unlicensed frequency band
Channels with severe levels of interference
Interference by other users or applications
Self-interference due to multi-path propagation
Channels with possible interception
Privacy
Features of spread spectrum technology
Redundant codes (anti-interference)
Pseudo-randomness (anti-interception from jammers)
Or anti-interference in the sense that not to interfere
others
Model of spread spectrum digital com system
Usage of pseudo-random patterns
Synchronization
Achieved by a xed pseudo-random bit pattern
The interference (by other users) may be characterized as
an equivalent additive white noise
Two dierent interferences (from others)
Two types of modulations are majorly considered in this
subject.
PSK
This is mostly used in direct sequence spread spectrum
(DSSS), abbreviated as DS-PSK.
But some also use MSK in DSSS, abbreviated as
DS-MSK.
FSK
This is mostly used in frequency-hopped spread
spectrum (FHSS).
The FHSS will not be introduced in the lectures.
12.2 Direct sequence spread spectrum
signals
A simple spread spectrum sysstem
Chip interval: T
c
=
1
W
BPSK is applied for each chip interval.
Bandwidth expansion factor B
e
=
W
R
_ =
1T
c
1T
b
=
T
b
T
c
_
Number of chips per information bit L
c
=
T
b
T
c
In practice, the spread spectrum system often consists of an
encoder and a modulo-2 adder.
Encoder : Encode the original information bits (in a
pre-specied block) to channel code bits, say (7, 3) linear
block code.
Modulo-2 adder : Directly alter the coded bits by
modulo-2 addition with the PN sequences.
Example
1) Choose T
c
= 1 ms, T
ib
= 14 ms and T
cb
= 7 ms,
where
T
c
length of a chip
T
ib
length of an information bit
T
cb
length of a code bit
2) Use (6, 3} linear block code (3 information bits 6 code
bits)
100
010
001
100
010
001
-
generator
matrix
0
0
1
info
bits
=
0
0
1
0
0
1
code
bits
3) Use the repetition code for chip generation:
code bit 0 0000000

code bit 1 1111111
-- -
seven
chips
L
c
=
T
ib
T
c
= 14
4) XOR with the PN sequence:
Example (Revisited)
How about we combine Step 2) and Step 3)?
2&3) Use (n = 6 7, k = 3 1} linear block code (3 information
bits 42 code bits)
Combined DSSS system
a
i
= b
i
c
i
, i = 0, . . . , n 1 and each a
i
is BPSK-transmitted.
Performance analysis
Let g(t} be the baseband pulse shape of duration T
c
.
g
i
(t} =

g(t iT
c
} if a
i
= 0
g(t iT
c
} if a
i
= 1
for i = 0, 1, . . . , n 1
Then
g
i
(t} = (1 2a
i
}g(t iT
c
}
= |1 2(b
i
c
m,i
}|g(t iT
c
}
= |(1 2b
i
}p(t iT
c
}| |(1 2c
m,i
}g(t iT
c
}|
or |(2b
i
1}p(t iT
c
}| |(2c
m,i
1}g(t iT
c
}|
= p
i
(t} c
m,i
(t}
where p(t} =rectangular pulse of height 1 and duration T
c
.
Consequently,
channel symbol g
s
(t} =
n1
i =0
g
i
(t}
=
n1
i =0
p
i
(t}c
m,i
(t}
= _
n1
i =0
p
i
(t}__
n1
i =0
c
m,i
(t}_
= p
PN
(t} c
m
(t} where m = 1, 2, . . . , M
In implementation (e.g., spectrum), the DSSP channel
symbol is the modulo-2 addition between code bits/chips
and the PN chips, followed by a chip-based BPSK
modulation.
In analysis, DSSP channel symbol can be conveniently
expressed as a coded BPSK signal c
m
(t} multiplying a
randomly polarized sequence p
PN
(t}.
DSSS receiver design
For iT
c
t < (i + 1)T
c
,
r
i
(t) = p
i
(t)c
m,i
(t) + z(t)
where z(t) is the interference introduced mainly by other users and
also by background noise.
Since for iT
c
t < (i + 1)T
c
,
p
i
(t) p
i
(t) = |(2b
i
1)p(t iT
c
)| |(2b
i
1)p(t iT
c
)|
= 1
we have
c
m,i
(t) = |p
i
(t)c
m,i
(t)| p
i
(t)
= |r
i
(t) z
(
t)| p
i
(t)
= r
i
(t) p
i
(t) z(t) p
i
(t)
Conclusion: The estimator c
m,i
(t) can be obtained from
r
i
(t) p
i
(t) if the channel is interference free.
DSSS demodulator
In this gure, we drop subscript m for c
m,i
for convenience.
y
i
= Re_
T
b
0
|(2c
m,i
1}g
i
(t} + (2b
i
1}z(t}| g
i
(t}dt_
= (2c
m,i
1}Re|g
i
(t}, g
i
(t}| + (2b
i
1}Re|z(t}, g
i
(t}|
= (2c
m,i
1}2E
c
+ (2b
i
1}
i
where
i
= Re|z(t}, g
i
(t}|.
Recall that slide 2-24 has derived:
x(t}, y(t} =
1
2
Re{x
(t}, y
(t}} .
or
E
c
= g
passband
(t}, g
passband
(t} =
1
2
Re{g(t}, g(t}} =
1
2
E
g
.
y
i
= (2c
m,i
1}2E
c
+ (2b
i
1}
i
Assumptions:
z(t} is a baseband interference (hence, complex).
z(t} is a (WSS) broadband interference, i.e., PSD of z(t}
is
S
z
(f } = 2J
0
for |f |
W
2
.
z(t} Gaussian
(2b
i
1} is known to Rx
ML decision
m = arg min
1mM
|y 2E
c
(2c
m
1)|
2
= arg max
1mM
y, 2E
c
(2c 1) since |2c
m
1|
2
constant
= arg max
1mM
2E
c
n
i =1
(2c
m,i
1)y
i
= arg max
1mM
n
i =1
(2c
m,i
1)y
i
Suppose
linear code is employed, and
the transmitted codeword is the all-zero codeword.
m = arg max
1mM
n
i =1
(2c
m,i
1)|(2c
1,i
1)2E
c
+ (2b
i
1)
i
|
Pr|error| = Pr| m 1|
= Pr
i =1
(2c
1,i
1)|(2c
1,i
1)2E
c
+ (2b
i
1)
i
|
< max
2mM
n
i =1
(2c
m,i
1)|(2c
1,i
1)2E
c
+ (2b
i
1)
i
|
= Pr
2nE
c

n
i =1
(2b
i
1)
i
< max
2mM
_2E
c
n
i =1
(2c
m,i
1) +
n
i =1
(2c
m,i
1)(2b
i
1)
i
_
= Pr
2nE
c

n
i =1
(2b
i
1)
i
< max
2mM
_2E
c
(n 2w
m
) +
n
i =1
(2c
m,i
1)(2b
i
1)
i
_
= Pr _ min
2mM
_4E
c
w
m
2
n
i =1
c
m,i
(2b
i
1)
i
_ < 0_
where w
m
is the number of 1s in codeword m.
Let R
m
= 4E
c
w
m
2
n
i =1
c
m,i
(2b
i
1)
i
.
Note that R
m
given b is Gaussian with
mean E|R
m
|b| = 4E
c
w
m
and variance Var|R
m
|b| = 4w
m
E|
2
i
|.
We can have the union bound:
Pr|error|b| = Pr _ min
2mM
R
m
< 0_ b_
M
m=2
Pr |R
m
< 0| b|
=
M
m=2
Q
4E
c
w
m
_
4w
m
E|
2
i
|
=
M
m=2
Q
2E
c
w
m
_
w
m
E|
2
i
|
Since the upper bound has nothing to do with b, we have

Pr|error| =

b
Pr(b) Pr|error|b|
M
m=2
Q
2E
c
w
m
_
w
m
E|
2
i
|
.
i
= Re|z(t), g
i
(t)|
= Re_
(i +1)T
c
iT
c
z(t)g
(t iT
c
)dt_ = Re|
i
+
i
|
d
= Re_
T
c
0
z(t)g
(t)dt_
where
d
= means equality in their distribution.
Assumption:
i
and
i
are zero mean and uncorrelated.
E|
2
i
| =
1
2
E[|
i
+
i
|
2
=
1
2
E__
T
c
0
z(t)g
(t)dt_
2
_
=
1
2
T
c
0

T
c
0
E|z(t)z
(s)|g
(t)g(s)dtds
=
1
2
R
z
(t s)g
(t)g(s)dtds
E[|
i
+
i
|
2
=

g(s)R
z
(t s)ds_g
(t)dt
=

G(f )S
z
(f )e
2ft
df _g
(t)dt
=

|G(f )|
2
S
z
(f )df
E|
2
i
| =
1
2
|G(f )|
2
S
z
(f )df
= J
0
W2
W2
|G(f )|
2
df
2J
0
E
c
Pr|error|
M
m=2
Q _
2E
c
w
m
2w
m
E
c
J
0
_
=
M
m=2
Q
2E
c
w
m
J
0
=
M
m=2
Q
2(k]n}E
b
w
m
J
0
=
M
m=2
Q [
_
2R
c
b
w
m
)
where
R
c
= k]n code rate
b
= E
b
]J
0
signal-to-interference ratio
How about z(t} being narrowband interference?
Assumptions:
z(t} is a baseband interference (hence, complex).
z(t} is a (WSS) narrowband interference, i.e., PSD of
z(t} is
S
z
(f } =

J
av
W
1
= 2J
0
{
W
W
1
) for |f |
W
1
2
0 otherwise
where J
av
= 2WJ
0
.
All the derivations remain unchanged except
E|
2
i
| =
1
2
|G(f }|
2
S
z
(f }df
=
J
av
2W
1
W
1
2
W
1
2
|G(f }|
2
df
The value of E|
2
i
| hence depend on the spectra of g(t} and
the location of the narrowband jammer.
Example 12.2-1
E|
2
i
| =
J
av
2W
1
W
1
2
W
1
2
|G(f )|
2
df =
J
av
E
c
W
1
2
2
sinc
2
(x)dx
J
av
E
c
W
1
= J
av
E
c
T
c
= 2J
0
E
c
where we use x = fT
c
and = W
1
T
c
=
W
1
W
in the derivation.
How about z(t} being CW jammer?
Assumptions:
z(t} is a CW (continuous wave) interference (hence,
complex).
z(t} is a (WSS) CW (continuous wave) interference, i.e.,
PSD of z(t} is
S
z
(f } = J
av
(f }
E|
2
i
| =
1
2
|G(f }|
2
S
z
(f }df
=
J
av
2
|G(0}|
2
= 2J
0
E
c
for Example 12.2-1
From the above discussion, we learn that
Under narrowband jammer, the DSSS performance
depends on the shape of g(t}.
For example (Example 12.2-2), if g(t} =
_
4E
c
T
c
sin{
t
T
c
) for
0 t < T
c
, then we obtain
Pr|error|
M
m=2
Q
2
4
R
c
b
w
m
=
M
m=2
Q [
_
(2.4674}R
c
b
w
m
)
One half cycle sinusoidal g(t} performs about 0.9dB
better than rectangular g(t}.
Alternative expression for union bound
Since J
av
= 2J
0
W = 2J
0
]T
c
and P
av
= E
b
]T
b
,
b
=
E
b
J
0
=
P
av
T
b
J
av
T
c
]2
=
2L
c
J
av
]P
av
Pr|error|
M
m=2
Q [
_
2R
c
b
w
m
) =
M
m=2
Q
4
L
c
R
c
w
m
J
av
]P
av
(M 1}Q
4
L
c
J
av
]P
av
min
2mM
R
c
w
m
where
J
av
P
av
Jamming-to-signal power ratio
L
c
Processing gain
min
2mM
R
c
w
m
Coding gain (Recall w
1
= 0}
Interpretation
Processing gain:
Theoretically, it is the number of chips per information
bit, which equals the bandwidth expansion factor B
e
.
Practically, it is the gain obtained via the uncoded DSSS
system (e.g., uncoded BPSK DSSS) in comparison with
the non-DSSS system (e.g., BPSK Q(
2
b
)).
So, it is the advantage gained over the jammer by the
processing of spreading or expanding the bandwidth of
the transmitted signal.
Coding gain
It is the advantage gained over the jammer by a proper
code design.
Example. Uncoded DSSS: Assume we use (n, 1) code.
Then,
R
c
=
1
n
, M = 2
1
= 2, w
1
= 0, w
2
= n.
Hence, coding gain = min
2mM
R
c
w
m
=
1
n
n = 1 = 0 dB.
Denition: Jamming margin
The largest jamming-to-signal power ratio that
achieves the specied performance (i.e., error rate)
under xed processing gain and coding gain.
Example 12.2-3
Problem: Find the jamming margin to achieve error rate 10
6
with L
c
= 1000 and uncoded DSSS.
For M = 2 (uncoded DSSS), the union bound is equal to the
exact error.
Answer:
Pr|error| = Q
4
L
c
J
av
]P
av
R
c
w
2
= Q _
4
1000
J
av
]P
av
_ 10
6
Then, J
av
]P
av
= 22.5 dB.
Example 12.2-3 (revisited)
Problem: Given that
b
= 10.5 dB satises Q(
2
b
) = 10
6
, nd
the jamming margin to achieve error rate 10
6
with L
c
= 1000 and
uncoded DSSS.
Answer:
Pr|error| = Q _
4
L
c
J
av
]P
av
min
2mM
R
c
w
m
_ = 10
6
Then,
2
L
c
J
av
]P
av
min
2mM
R
c
w
m
= 10.5 dB
or equivalently,
10log
10
(2) dB+L
c
dB+ min
2mM
R
c
w
m
dB(J
av
]P
av
) dB = 10.5 dB.
Thus,
3 dB+30 dB+0 dB(J
av
]P
av
) dB = 10.5 dB (J
av
]P
av
) dB = 22.5 dB

Spectrum analysis
We now demonstrate why it is named spread spectrum system!
Assume the uncoded DSSS system where all-zero and all-one
codes are used.
Then
channel symbol g
s
(t) = p
PN
(t) c(t) + z(t)
where
c(t) =

n=
I
n
s(t nT
b
) with s(t) =

g(t mod T
c
) 0 t < T
b
0 otherwise
and
I
n
1
n=
zero-mean i.i.d.
From slide 3-115,
S
c
(f ) =
1
T
b
S
I
(f )|S(f )|
2
=
1
T
b
|S(f )|
2
where S
I
(f ) =
k=
R
I
(k)e
2kfT
b
= 1.
Assume g(t) rectangular pulse of height 1]
T
b
and duration T
c
(hence,

T
b
0
s
2
(t)dt = 1). Then (cf. slide 31 by replacing T
b
with
T
c
and letting E = 1]2),
S
c
(f ) =
1
T
b
{T
b
sinc
2
(T
b
f )) = sinc
2
(T
b
f )
Similarly,
p
PN
(t)c(t) =

i =
(2b
i
1)p(t iT
c
)I
i modn
s(t (i modn)T
b
)
d
=
T
c
T
b
i =
(2b
i
1)
1
T
c
p(t iT
c
)
where here 2b
i
1
i =1
and (2b
i
1)I
i modn
i =1
actually have the
same distribution. Then from slide 3-115,
S
pc
(f ) =
1
T
c
_
T
c
T
b
T
c
P(f )
2
=
1
T
b
{T
c
sinc
2
(T
c
f )) =
1
L
c
sinc
2
(T
c
f )
Recovered symbol at the receiver end:
p
PN
(t}g
s
(t} = p
2
PN
(t} c(t} + p
PN
(t}z(t}
= c(t} + p
PN
(t}z(t}
This indicates that the PSD of the new noise is:
S
pz
(f } =

S
p
(f } S
z
(f }
=

S
p
(s}S
z
(f s}ds = 2J
0

S
p
(s}ds
= 2J
0
1
T
c
|P(s}|
2
ds = 2J
0
T
c
sinc
2
(T
c
s}ds
= 2J
0
where for simplicity we let S
z
(f } = 2J
0
for f R.
Summary
Multiplication of p
PN
(t} = spreading the power over the
bandwidth of p
PN
(t} (So that the transmitted signal is
hidden under the broadband interference.)
Multiplication twice of p
PN
(t} will recover the original
signal.
The spreading fraction is approximately equal to the
processing gain.
Modulator: Transmit p
PN
(t}c(t}
Demodulator: Based on r (t}p
PN
(t} = c(t} + z(t}p
PN
(t}
Further performance enhancement by coding
Coding gain = min
2mM
R
c
w
m
(Recall w
1
= 0}
Use (n
1
, k} code as the outer code, and (n
2
, 1} repetition code
as the inner code, where n = n
1
n
2
.
Then
Coding gain = min
2mM
R
c
w
m
= min
2mM
k
n
1
n
2
n
2
w
(out)
m
= min
2mM
R
(out)
c
w
(out)
m
The use of the inner code here is to align the length of the outer
code n
1
to the length of the PN sequence n.
Since the inner code is the binary repetition code, the bit error rate
of the outer code is the symbol error rate of the inner code p,
where under broadband interference,
p = Q _
_
2R
(in)
c

(in)
b
w
(in)
2
_ For M = 2, we have equality, not .
= Q
2
1
n
2
n
2
E
c
J
0
n
2
= Q
2
1
n
2
n
2
(k]n)E
b
J
0
n
2
= Q _
_
2
b
R
(out)
c
_ = Q _
2
2L
c
J
av
]P
av
R
(out)
c
_ (cf. slide 35)
Then the symbol error rate of the entire system is
P
e

n
1
m=t+1
_
n
1
m
_p
m
(1 p)
n
1
m
2
k
m=2
|4p(1 p)|
w
m
2
-- -
Cherno bound
where t = (d
min
1)]2 and d
min
is the minimum Hamming
distance among outer codewords.
Golay (24, 12} (outer) code
Example. Use Golay (24, 12) outer code and set L
c
= 100.
We need to rst determine n
2
based on n
1
= 24.
12T
b
= nT
c
= n
1
n
2
T
c
= 24n
2
T
c
n
2
=
12T
b
24T
c
=
1
2
L
c
=
1
2
100 = 50.
Then p = Q {
_
2
2100
J
av
P
av
12
24
) = Q {
_
200
J
av
P
av
).
P
e

24
m=4
_
24
m
_p
m
(1 p)
24m
759|4p(1 p)|
4
+ 2576|4p(1 p)|
6
+ 759|4p(1 p)|
8
+|4p(1 p)|
12
.

Golay (24, 12} code
Weight number of codewords
0 1
8 759
12 2576
16 759
24 1
Appendix: Hard-decision versus soft-decision
The performance usually improves 3 dB by using soft-decision.
12.2-2 Some applications of DS spread
spectrum signals
Code division multiple access (CDMA)
If each user has its own PN sequence (with good properties), then
many DSSS signals are allowed to occupy the same channel
bandwidth.
r (t) = p
(1)
(t)c
(1)
(t) + p
(2)
(t)c
(2)
(t) + + p
(N
u
)
(t)c
(N
u
)
(t) + z(t)
- - -
z(t)
p
(1)
(t)r (t) = c
(1)
(t) + p
(1)
(t) z(t)
How to determine the number of users (capacity)?
Each user is a broadband interference with power P
av
(cf. slide 8)
P
av
J
av
=
P
av
(N
u
1)P
av
=
1
N
u
1
.
By this, we can obtain for L
c
= 100 and Golay outer code and
P
e
10
6
, N
u
= 41 .
12.2-3 Eect of pulsed interference on DS
spread spectrum systems
Types of interferences
CW jammer S
z
(f } = J
av
(f }
Broadband interference S
z
(f } = 2J
0
for |f | W]2
Pulsed interference
z
p
(t} = z
(t}(t}
where z
(t} is a broadband interference with

S
z
(f } = S
z
(f }] for some 0 < < 1 and (t} is a
0-1-valued random pulse of duration T
b
, which equals 1
with probability .
Hence, for uncoded DSSS (no coding gain),
when (t} = 0, the system is error free,
when (t} = 1, the system suers broadband interference
with
Pr|error| = Q
4
L
c
(J
av
]}]P
av
= Q
4
(W]R}
(2J
0
W]}](E
b
R}
= Q
2
E
b
J
0

The system error under pulsed interference is
P
e
(} = (1 } 0 + Q
2
E
b
J
0
= Q
2
E
b
J
0
.
What is the that maximizes P
e
from an enemys
standpoint?
dP
e
(}
d
= 0
0.71
E
b
J
0
if E
b
]J
0
0.71
1 if E
b
]J
0
< 0.71
and
P
e
(
0.083
E
b
J
0
if E
b
]J
0
0.71
= Q _
_
2
E
b
J
0
_ if E
b
]J
0
< 0.71
Worst-case pulse jamming: =
; hence it is not a constant

on the dotted line.
Summary
The DSSS system performs poor under burst-in-time
jammer, not under burst-in-frequency jammer (CW
jammer).
For example, at the requirement of P
e
= 10
6
, the
Jamming margin will be increased around 40 dB in
comparison with the CW jammer.
Cuto rate
Performance index
Usual measure: The required SNR for a specied error
rate
Analytically convenient measure: Cuto rate
Denition 1 (Cuto rate)
Fix code rate R
c
and use (n, nR
c
} code. The maximum R
0
that satises
P
e
2
n(R
0
R
c
)
is called the cuto rate (which is a function of R
c
).
Interpretation: If R
c
< R
0
, then P
e
0 as n 0.
Sample derivation of cuto rate
Give
Channel symbol 1 s
1
= |s
1,1
, s
1,2
, . . . , s
1,n
|
Channel symbol 2 s
2
= |s
2,1
, s
2,2
, . . . , s
2,n
|
where s
m,j
=
E
c
.
From slide 4-44,
P
2
= Q _
d
12
2N
0
_.
Now suppose we randomly assign each of s
m,j
independently
(random coding) with
Pr|s
m,j
=
_
E
c
| = Pr|s
m,j
=
_
E
c
| =
1
2
.
Then Pr|d
2
12
= 4dE
c
| = {
n
d
)2
n
for integer 0 d n.
Using Q(x}
1
2
e
x
2
2
e
x
2
2
yields:
E|P
2
| =
n
d=0
_
n
d
_2
n
Q
2dE
c
N
0
d=0
_
n
d
_2
n
e
dE
c
N
0
= 2
n
{1 + e
E
c
N
0
)
n
= 2
n1log
2
1+e
E
c
N
0
The union bound for M-ary random code gives

E|P
M
| (M 1}E|P
2
| ME|P
2
| = 2
nR
c
2
n1log
2
1+e
E
c
N
0
= 2
n(
R
0
R
c
)
where

R
0
= 1 log
2
{1 + e
E
c
N
0
) .
Since E|P
M
| 2
n(
R
0
R
c
)
, there must exist a code with
P
M
2
n(
R
0
R
c
)
and hence
R
0

R
0
= 1 log
2
{1 + e
E
c
N
0
) .
As it turns out, this lower bound of cuto rate is tight! So,
R
0
=

R
0
.
R
c
=
k
n
(information) bits/chip; So R
0
is measured in bits/chip.
R
0
is usually in the shape of 1 log
2
(1 +
}, where
e
E
c
N
0
soft-decision decoding (as just derived)
_
4p(1 p} hard-decision decoding
given p = Q(
_
2E
c
]N
0
}
For worst-case pulsed interference, Omura and Levitt
(1982) derive
e
E
c
N
0
soft-decision with knowledge of jammer state
min
0
|e
2E
c
1 + e
2
E
c
N
0
|
soft-decision with no knowledge of jammer state
_
4p(1 p}
hard-decision with knowledge of jammer state
_
4p(1 p}
hard-decision with no knowledge of jammer state
where p = Q [
_
2E
c
]N
0
) (and N
0
= J
0
).
The receiver may know the jammer state (side information) by
measuring the noise power level in adjacent frequency band.
Observations from Omura and Levitt
When R
0
< 0.7 bits/chip (e.g., E
c
]N
0
< 0dB),
soft-decision in AWGN (curve (0)) performs identically to
soft-decision with jammer state information (curve (1)).
When jammer state is known, the worse-case pulsed
jammer has no eect on soft-decision system performance.
When R
0
< 0.4 bits/chip (e.g., E
c
]N
0
< 0dB),
hard-decision with jammer state information (curve (2))
performs identically to soft-decision with no jammer state
information (curve (4)).
Knowing the jammer state information does not help
improving the hard-decision system performance.
Big question: Why (3) performs worse than (4)?
Without jammer state information, the reception y is
untrustworthy.
The soft-decision based on
y 2E
c
(2c
m
1}
2
=
n
i =1
(y
i
2E
c
(2c
m,i
1}}
2
may eliminate the correct codeword at the time when a
wrong codeword gives a slightly larger y2E
c
(2c
m
1}
2
due to one very dominant (y
i
2E
c
(2c
m,i
1}}
2
.
However, the hard-decision based on
d
Hamming
(r, c} =
n
i =1
(r
i
c
i
}
can limit the dominant aection from any single bit,
and makes the decision based more on the entire
receptions.
One can use a quantizer (or a limiter) to achieve the
same goal and improves the performance of the
hard-decision decoding.
The limiting action from quantizers or limiters ensures
that any single bit does not heavily (and dominantly) bias
the corresponding decision metric.
12.2-5 Generation of PN sequences
Properties of (deterministic) PN sequences
Rule 1: Balanced property
Relative frequencies of 0 and 1 are each (nearly) 1/2.
Rule 2: Run length property
Run length (of 0s and 1s) are as expected close to a
fair-coin ipping.
1/2 of run lengths are 1; 1/4 of run lengths are 2; 1/8 of
run lengths are 3 . . . etc.
Rule 3: Delay and add property
If the sequence is shifted by any non-zero number of
elements, the resulting sequence will have an equal
number of agreements and disagreements with the
original sequence.
Example of PN sequences
Maximum-length shift-register sequences (n = 2
m
1, k = m}
code
Also named m-sequences.
Maximum-length shift-register sequence
(n, k} = (2
m
1, m}
By its name, the codewords are
the sequential output of m-stage
shift-register with feedback.
The maximum length of codewords
is 2
m
1 because the register
contents can only have 2
m
1
possibilities.
Polynomial representation of m-sequences
The code can be specied by
g(p} = 1 +
1
p +
2
p
2
+ +
m1
p
m1
+ p
m
based on its structure.
Vulnerability of m-sequences
Suppose the enemy knows the number of shift registers, m.
Then (2m 1} observations are sucient to determine
1
,
2
, . . . ,
m1
.
a
m+1
= a
1
+
1
a
2
+ +
m1
a
m
a
m+2
= a
2
+
1
a
3
+ +
m1
a
m+1
a
2m1
= a
m1
+
1
a
m
+ +
m1
a
2m2
Possible solutions:
Frequent change of (
1
,
2
, . . . ,
m1
}.
Combination of several m-sequences in a non-linear way
(without changing the necessary properties).
Periodic autocorrelation and crosscorrelation
function
Periodic autocorrelation function
R
b
(j } =
n
i =1
(2b
i
1}(2b
i +j
1}
Periodic crosscorrelation function
R
b
b
(j } =
n
i =1
(2b
i
1}(2
b
i +j
1}
For m-sequences:
R
b
(j } =

n j = 0
1 1 j < n
but R
b
b
(j } may be large!
Although it is possible to select a small subset of m-sequences
that have relatively smaller cross-correlation peak values, the
number of sequences in the set is usually too small for CDMA
applications.
Gold sequences (Gold 1967-1968)
Gold and Kasami proved that there exists certain pairs of
m-sequences with crosscorrelation function taking values in
{1, t(m}, t(m} 2}, where
t(m} =

2
(m+1)2
+ 1 m odd
2
(m+2)2
+ 1 m even
Example. Gold sequence with m = 10.
Periodic crosscorrelation function values
{1, 2
(m+2)2
1, 2
(m+2)2
1} = {1, 65, 63}
Generation of Gold sequences
Two m-sequences with periodic crosscorrelation function in
1, t(m), t(m) 2 are called preferred sequences.
Existence of two preferred sequences has been proved by
Gold and Kasami.
Let |a
1
, a
2
, . . . , a
n
| and |b
1
, b
2
, . . . , b
n
| be the selected
preferred sequences. Then
Gold sequences =
|a
1
, a
2
, . . . , a
n
|
|b
1
, b
2
, . . . , b
n
|
|a
1
b
1
, a
2
b
2
, . . . , a
n1
b
n1
, a
n
b
n
|
|a
1
b
2
, a
2
b
3
, . . . , a
n1
b
n
, a
n
b
1
|
|a
1
b
n
, a
2
b
1
, . . . , a
n1
b
n2
, a
n
b
n1
|
This gives (n + 2) Gold sequences in which some of them are

no longer maximal length sequences. The autocorrelation
function values are also in 1, t(m), t(m) 2.
Example.
Construct n = 31 Gold sequences.
Select two preferred sequences:
g
1
(p} = 1 + p
2
+ p
5
g
2
(p} = 1 + p + p
2
+ p
4
+ p
5
Welch bound
Theorem 1
Give a set of M binary sequences of length n. Then the peak
crosscorrelation function value among them is lower-bounded
by
n
M 1
Mn 1
When M 1,
n
M 1
Mn 1
n
M
Mn
=

n.
For Gold sequences (n = 2
m
1 2
m
),
peak cross = t(m} =

2
(m+1)2
+ 1 m odd
2
(m+2)2
+ 1 m even
=

2
m
+ 1 m odd
2
2
m
+ 1 m even
=

n m odd
2
n m even
Therefore, Gold sequences do not achieve the Welch
bound.
Kasami sequences
A set of M = 2
m2
sequences of length n = 2
m
1 for any m
even.
It is formed by the following procedure.
1
Pick an m-sequence a = |a
1
, a
2
, . . . , a
n
|.
2
Since n = 2
m
1 = (2
m2
1)(2
m2
+ 1), we can
fragament a into (2
m2
+ 1)-bit blocks.
|a
1
, . . . , a
2
m2
+1
- - -
block 1
, a
2
m2
+2
, . . . , a
2(2
m2
+1)
---
block 2
, a
22
m2
+3
, . . .|
3
Let
b = |a
k
, a
2k
, . . . , a
(2
m2
1)k
, a
k
, a
2k
, . . . , a
(2
m2
1)k
, . . .|
where k = 2
m2
+ 1.
Kasami sequences =
|a
1
, a
2
, . . . , a
n
|
|a
1
b
1
, a
2
b
2
, . . . , a
n
b
n
|
|a
1
b
2
, a
2
b
3
, . . . , a
n
b
1
|
|a
1
b
2
m2
1
, a
2
b
2
m2 , . . . , a
n
b
2
m2
2
|
The o-peak autocorrelation and crosscorrelation function

values are in {1, (2
m2
+ 1}, 2
m2
1} and the Welch bound
is achieved (at a price of much less number of sequences can
be used!)

Chap12 Spread Spectrum Signals

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chap12 Spread Spectrum Signals

Uploaded by

Copyright:

Available Formats

Digital Communications

Chapter 12 Spread Spectrum Signals for Digital

code bit 0 0000000

Since the upper bound has nothing to do with b, we have

Digital Communications: Chapter 12 Ver 2010.12.25 Po-Ning Chen 39 / 87

Digital Communications: Chapter 12 Ver 2010.12.25 Po-Ning Chen 48 / 87

(t} is a broadband interference with

Digital Communications: Chapter 12 Ver 2010.12.25 Po-Ning Chen 56 / 87

; hence it is not a constant

The union bound for M-ary random code gives

This gives (n + 2) Gold sequences in which some of them are

The o-peak autocorrelation and crosscorrelation function

You might also like