Vous êtes sur la page 1sur 6

Performance Analysis and Simulation of Cyclic

Code-Shift Keying
Chi-Han Kao1, Clark Robertson1, and Kyle Lin2
1
Electrical and Computer Engineering Department
2
Operations Research Department
Naval Postgraduate School, Monterey, CA 93943-5121
Since CCSK is non-orthogonal, it is complicated to evaluate
Abstract—Cyclic code-shift keying (CCSK) is the baseband its performance analytically. Previously, the evaluation has
symbol modulation scheme used by Joint Tactical Information been done by simulation. Therefore, an analytic evaluation of
Distribution System (JTIDS), the communication terminal of CCSK performance in terms of probability of symbol error is
Link-16. Since CCSK is non-orthogonal, an analytic evaluation of
nontrivial. In this paper, an analytic upper bound on the
its performance in terms of probability of symbol error is
nontrivial. In this paper, an analytic upper bound on the probability of symbol error of CCSK is derived for the 32-chip
probability of symbol error of CCSK is derived for the 32-chip CCSK sequence chosen for JTIDS. The probability of symbol
CCSK sequence chosen for JTIDS. The probability of symbol error obtained with the analytic upper bound is compared with
error obtained with the analytic method is compared with that that obtained by Monte Carlo simulation in AWGN. In
obtained by Monte Carlo simulation for additive white Gaussian addition to the 32-chip CCSK sequence chosen for JTIDS, a
noise. The results show that the analytic method yields a tight
new 32-chip CCSK sequence with a smaller maximum off-
upper bound. In addition to the 32-chip CCSK sequence chosen
for JTIDS, a new 32-chip CCSK sequence with a smaller peak cross-correlation value is presented and evaluated both
maximum off-peak cross-correlation is obtained and evaluated analytically and by Monte Carlo simulation.
both analytically and by Monte Carlo simulation. The results
obtained for the new CCSK sequence compare favorably with the II. CYCLIC CODE-SHIFT KEYING
sequence chosen for JTIDS.
A. CCSK Symbol Modulation
Index Terms—cyclic code-shift keying (CCSK), Joint Tactical
Information Distribution System (JTIDS), Link-16, probability
For JTIDS, data modulation consists of CCSK and MSK.
of symbol error. CCSK provides M-ary baseband modulation, where each 5-bit
symbol is represented by a 32-chip sequence. MSK provides
I. INTRODUCTION chip modulation for transmission over the channel. The 32-
chip CCSK sequence (S0) chosen by JTIDS is shown in Table
C yclic code-shift keying (CCSK) is the baseband symbol
modulation scheme used by Joint Tactical Information
Distribution System (JTIDS), the communication terminal of
I. As is seen, thirty-two sequences are derived by cyclically
shifting S0 to the left between one and 31 times to obtain a
unique sequence for all possible combinations of five bits.
Link-16. Based on a time-division multiple access
architecture, Link-16 provides a secure and ostensibly jam- TABLE I 32-CHIP CCSK SEQUENCE CHOSEN FOR JTIDS.
resistant data link so that near real-time tactical information 5-bit
32-chip CCSK sequence chosen for JTIDS
can be exchanged among different platforms. JTIDS is symbol
essentially a hybrid direct sequence/frequency-hopping 00000 S0 = 01111100111010010000101011101100
(DS/FH) spread spectrum system that uses a (31,15) Reed- 00001 S1 = 11111001110100100001010111011000
Solomon code for forward error correction coding, CCSK for 00010 S2 = 11110011101001000010101110110001
baseband symbol modulation, and minimum-shift keying ⋮ ⋮
(MSK) for chip modulation. The performance of hybrid 11111 S31 = 00111110011101001000010101110110
DS/FH spread spectrum systems with various modulation
schemes in different types of interference and fading has been
investigated in [1]-[5], but only [2] attempts to evaluate the B. CCSK Symbol Demodulation
performance of JTIDS’s waveform analytically in additive At the CCSK symbol demodulator, the determination of
white Gaussian noise (AWGN). However, the results which 5-bit symbol was received is accomplished by
presented in [2] are based on the overly optimistic assumption computing the cross-correlation between the received 32-chip
that the cross-correlation values for the CCSK symbol sequence and all possible thirty-two sequences. The decision
demodulator are independent. Actually, it can be shown that is made by choosing the 5-bit symbol corresponding to the
these cross-correlation values are not independent. branch with the largest cross-correlation value. For example, if
symbol 0 is sent in the absence of noise, the cross-correlation
yields
978-1-4244-2677-5/08/$25.00 ©2008 IEEE
⎧ 32, i=0 N = 0 (column 2), H = 4 occurs for ℜ7 , ℜ16 , and ℜ25 .
ℜi = ⎨ (1)
h
⎩ i ≤ H , 1 ≤ i ≤ 31 Second, when N = 1 , ℜ0 decreases by two, while ℜi for
where i is the number of the cross-correlation branch, hi is 1 ≤ i ≤ 31 can either increase or decrease by two, depending
the off-peak cross-correlation value, and H is the maximum on where the chip error has occurred (for column 3, the first
off-peak cross-correlation value. For the 32-chip CCSK chip is assumed to be in error, whereas for column 4, the fifth
sequence chosen for JTIDS, hi = −4 , 0, or 4 for i = 1, 2,...,31 , chip is erroneous) and, as a result, H = 6 . Third, when
and, hence, H = 4 . In this case, the decision made at the N = 2 , ℜ0 decreases by four, while ℜi for 1 ≤ i ≤ 31 can
CCSK symbol demodulator is that symbol 0 was received either increase by four, decrease by four, or stay the same.
since ℜ0 = 32 is the largest. Note that (1) shows that CCSK is Fourth, the location of H varies based on the location and the
number of the chip errors in the received sequence. For
not orthogonal since the hi s have values other than zero.
example, when N = 2 and the chip errors occur at the fifth
C. Cross-correlation Properties of CCSK and the tenth chip location of the received sequence (column
To analyze the probability of symbol error of CCSK, an 5), ℜ7 = 8 is the maximum; however, if the chip errors occur
understanding of the nature of the cross-correlation properties at the first and the second chip location of the received
of CCSK sequences is needed. Specifically, we need to sequence (column 6), H = 4 instead of H = 8 occurs for
investigate the cases where the received sequence consists of branches 5, 12, 16, 20, and 25. Five pairs of CCSK symbols
N chip errors for 0 ≤ N ≤ 32 . For N = 0, 1, and 2, the cross- are converted into baseband waveforms and listed in Table III
correlation results given that symbol 0 is sent are shown in to illustrate why this occurs.
Table II for two specific cases each when N = 1 and N = 2.
TABLE III FIVE PAIRS OF THE 32-CHIP CCSK BASEBAND WAVEFORM.
Symbol 0 -1 1 1 1 1 1 -1 -1 1 1 1 -1 1 -1 -1 1 -1 -1 -1 -1 1 -1 1 -1 1 1 1 -1 1 1 -1 -1
TABLE II PARTIAL CROSS-CORRELATION RESULTS FOR THE 32-CHIP CCSK Symbol 7 -1 1 1 1 -1 1 -1 -1 1 -1 -1 -1 -1 1 -1 1 -1 1 1 1 -1 1 1 -1 -1 -1 1 1 1 1 1 -1
SEQUENCE CHOSEN FOR JTIDS. Symbol 0 -1 1 1 1 1 1 -1 -1 1 1 1 -1 1 -1 -1 1 -1 -1 -1 -1 1 -1 1 -1 1 1 1 -1 1 1 -1 -1
ℜi N =0 N =1 N =1 N =2 N =2 Symbol 16 -1 -1 -1 -1 1 -1 1 -1 1 1 1 -1 1 1 -1 -1 -1 1 1 1 1 1 -1 -1 1 1 1 -1 1 -1 -1 1
Symbol 0 -1 1 1 1 1 1 -1 -1 1 1 1 -1 1 -1 -1 1 -1 -1 -1 -1 1 -1 1 -1 1 1 1 -1 1 1 -1 -1
i=0 32 30 30 28 28
Symbol 25 1 1 -1 1 1 -1 -1 -1 1 1 1 1 1 -1 -1 1 1 1 -1 1 -1 -1 1 -1 -1 -1 -1 1 -1 1 -1 1
i =1 0 2 -2 -4 0 Symbol 0 -1 1 1 1 1 1 -1 -1 1 1 1 -1 1 -1 -1 1 -1 -1 -1 -1 1 -1 1 -1 1 1 1 -1 1 1 -1 -1
i=2 0 2 2 4 0 Symbol 1 1 1 1 1 1 -1 -1 1 1 1 -1 1 -1 -1 1 -1 -1 -1 -1 1 -1 1 -1 1 1 1 -1 1 1 -1 -1 -1
Symbol 0 -1 1 1 1 1 1 -1 -1 1 1 1 -1 1 -1 -1 1 -1 -1 -1 -1 1 -1 1 -1 1 1 1 -1 1 1 -1 -1
i=3 -4 -2 -2 -4 -4
Symbol 3 1 1 1 -1 -1 1 1 1 -1 1 -1 -1 1 -1 -1 -1 -1 1 -1 1 -1 1 1 1 -1 1 1 -1 -1 -1 1 1
i=4 0 2 -2 0 0
i=5 0 2 -2 0 4
i=6
Comparing two waveforms chip-by-chip for each symbol
-4 -6 -6 -8 -4
i=7 4 2 6 8 0
pair in Table III, we see that for the first three symbol pairs
i =8 0 2 -2 0 0 there are 18 same-sign chip pairs and 14 different-sign chip
i=9 -4 -2 -2 0 -4 pairs. For the fourth symbol pair, there are 16 same-sign chip
i = 10 -4 -2 -2 0 0 pairs and 16 different-sign chip pairs. For the fifth symbol
i = 11 0 -2 -2 -4 -4 pair, there are 14 same-sign chip pairs and 18 different-sign
i = 12 0 2 2 4 4 chip pairs. In general, we can divide the 32-chip waveform of
i = 13 0 -2 2 0 0 each symbol pair into two groups. The first group has
i = 14 0 -2 2 4 -4 ( 32 + hi ) 2 same-sign chip pairs, while the second group has
i = 15 -4 -2 -2 -4 0
i = 16 4 2 2 0 4
( 32 − hi ) 2 different-sign chip pairs. Given this observation, we
i = 17 -4 -6 -2 -4 -4 conclude the following. First, when N = 1 , ℜi for 1 ≤ i ≤ 31
i = 18 0 -2 -2 0 0 decrease by two when the chip error occurs in the first group
i = 19 0 -2 2 0 -4
and increase by two when the chip error occurs in the second
i = 20 0 2 -2 -4 4
group. Second, when N = 2 , ℜi for 1 ≤ i ≤ 31 decrease by
i = 21 0 -2 -2 0 -4
i = 22 -4 -2 -6 -4 0 four when both chip errors occur in the first group, increase by
i = 23 -4 -6 -2 0 -8 four when both chip errors occur in the second group, and stay
i = 24 0 2 -2 -4 0 the same if one chip error occurs in the first group and the
i = 25 4 6 2 0 4 other occurs in the second group.
i = 26 -4 -2 -2 -4 0 Observing the first three symbol pairs in Table III, the
i = 27 0 -2 2 0 -4 probability that a chip error occurs in the first group given that
i = 28 0 2 2 0 0
N = 1 is 18 32 , while the probability that a chip error occurs
i = 29 -4 -2 -6 -4 0
i = 30 0 -2 -2 0 0 in the second group given that N = 1 is 14 32 . Let X 1 and
i = 31 0 -2 -2 -4 0 X 2 represent the number of chip errors in the first and second
group, respectively. The preceding conditional probabilities
From Table II, several properties are observed. First, when for a single chip error can also be obtained from
⎛18 ⎞ ⎛14 ⎞ possible value of ℜ7 ranges from −12 to 20; that is, it is
⎜ ⎟⎜ ⎟ possible to have either a tie when ℜ0 = ℜ7 = 16 or to have a
1 0 18
P { X 1 = 1 X 1 + X 2 = 1} = ⎝ ⎠ ⎝ ⎠ = (2)
⎛ 32 ⎞ 32 symbol error when ℜ0 < ℜ7 .
⎜ ⎟
⎝1⎠ Based on the observations from N = 0 to N = 8 , CCSK
and from cross-correlation properties can be summarized. First, the
⎛18 ⎞⎛ 14 ⎞ value of ℜ0 is given by
⎜ ⎟⎜ ⎟ ℜ0 = 32 − 2 N , (5)
0 1 14
P { X 1 = 0 X 1 + X 2 = 1} = ⎝ ⎠⎝ ⎠ = . (3)
⎛ 32 ⎞ 32 where 0 ≤ N ≤ 32 . Second, the value of ℜi is given by
⎜ ⎟
1
⎝ ⎠ ℜi = hi + 2 ( N − 2q ) , 0 ≤ q ≤ N , 1 ≤ i ≤ 31. (6)
Note that the distributions in (2) and (3) are known as the Third, the value of ℜi is in the range of
hyper-geometric distribution. The general expression for the
hi − 2 N ≤ ℜi ≤ hi + 2 N , 1 ≤ i ≤ 31. (7)
hyper-geometric distribution is given by [6]
Fourth, ℜi is a hyper-geometric random variable with a
⎛ n1 ⎞⎛ n2 ⎞
⎜ ⎟⎜ ⎟ conditional probability mass function (pmf)
q j − q⎠
P { X 1 = q X 1 + X 2 = j} = ⎝ ⎠⎝ , (4) ⎛ ( 32 + hi ) 2 ⎞⎛ ( 32 − hi ) 2 ⎞
⎛ n1 + n2 ⎞ ⎜ ⎟⎜ ⎟
⎜ ⎟
P {ℜi = hi + 2 ( N − 2q) N = j} = ⎝
q ⎠⎝ j − q ⎠ . (8)
⎝ j ⎠
⎛ 32 ⎞
where X 1 and X 2 are independent binomial random ⎜ ⎟
⎝ j⎠
variables, n1 is the total number of independent trials for X 1 , For example, when h7 = 4 , N = 1 , and q = 1 , we obtain
and n2 is the total number of independent trials for X 2 . In our
P {ℜ7 = 2 N = 1} = 18 32 , which is the same result as (2).
case, n1 = ( 32 + hi ) 2 , n2 = ( 32 − hi ) 2 , and N = X 1 + X 2 .
Last, the CCSK symbol demodulator does not make a symbol
In addition to the cases of N = 0, 1, and 2, the cross- error when N ≤ 6 ; that is,
correlation results for ℜ0 and ℜ7 from N = 5 to N = 8 P {symbol error N = j} = 0, 0 ≤ j ≤ 6. (9)
given that symbol 0 is sent are shown in Table IV, where q is
Note that (9) is the conditional probability of symbol error for
denoted as the number of chip errors in the first group, and CCSK when N ≤ 6 . Recalling the principle of computing
N − q is denoted as the number of chip errors in the second probabilities by conditioning, we get [7]
group. As can be seen, there is no symbol error when N ≤ 6 . P { X } = ∑ P { X N = j}P { N = j} . (10)
j
TABLE IV PARTIAL CROSS-CORRELATION RESULTS FOR ℜ0 AND ℜ7 . If X denotes a CCSK symbol error, then the probability of
q
N N −q ℜ0 ℜ7 Remark CCSK symbol error is obtained from (10) as
5 5 0 22 -6 No error 32
4
3
1
2
22
22
-2
2
No error
No error PS = ∑ P {symbol error N = j}P { N = j} . (11)
2 3 22 6 No error j =0
1 4 22 10 No error
0 5 22 14 No error For JTIDS, N is a binomial random variable with a pmf
6 6 0 20 -8 No error
5 1 20 -4 No error ⎛ 32 ⎞
P { N = j} = ⎜ ⎟ Pc j (1 − Pc ) , j = 0,1,...,32
32 − j
1 5 20 12 No error (12)
0 6 20 16 No error ⎝ ⎠j
7 7 0 18 -10 No error
6 1 18 -6 No error where Pc is the probability of chip error at the output of the
1 6 18 14 No error
0 7 18 18 Tie MSK chip demodulator. To evaluate (11), the conditional
8 8 0 16 -12 No error
7 1 16 -8 No error probabilities of symbol error for 0 ≤ N ≤ 32 are required; that
1 7 16 16 Tie
0 8 16 20 Error is, the problem remaining is to obtain P {symbol error N = j}
for 7 ≤ N ≤ 32 .
When N = 7 , there are eight possibilities based on the D. Conditional Probability of Symbol Error of CCSK
value of q but only four are shown. Now, ℜ0 = 18 , and the
Based on the CCSK cross-correlation properties in (5)
possible value of ℜ7 ranges from −10 to 18; that is, it is through (9), the conditional probability of symbol error given
possible to have a tie when ℜ0 = ℜ7 = 18 . In this case, the that N = 7,8,...,32 can be obtained case by case; however,
CCSK symbol demodulator can make an error by choosing except for N = 7 , an upper bound is obtained rather than an
symbol 7. exact result. First, from Table IV, when N = 7 and symbol 0
When N = 8 , there are nine possibilities based on the value is sent, we have ℜ0 = 18 , while the possible values of ℜ7 are
of q , but only four are shown. Now, ℜ0 = 16 , and the in the set of [ −10, − 6, − 2, 2, 6, 10, 14, 18] . Since a symbol
error occurs when ℜ7 ≥ ℜ0 , we need to find the conditional In addition to ℜ7 , ℜ16 , and ℜ25 , we need to consider other
probability that ℜ7 = 18 given that N = 7. From (8), branches that have an off-peak cross-correlation value hi = 0
P {ℜ7 = 18 N = 7} = 1.02 × 10−3 = P {ℜi = 18 N = 7} (13) (such as ℜ1 ) since these branches can also have a cross-
where i = 16 and 25 since ℜ7 , ℜ16 , and ℜ25 have the same correlation value of 16 when N = 8 . From (8),
maximum off-peak cross-correlation value. Next, P {ℜ1 = 16 N = 8} = 1.224 × 10−3 , (26)
P {S7 chosen N = 7} = P {S7 chosen N = 7, ℜ7 = 18} × and the conditional probability that S1 is chosen is
(14)
P {ℜ7 = 18 N = 7} = (1 2 ) (1.02 × 10−3 ) , P {S1 chosen N = 8} = P {S1 chosen N = 8, ℜ1 = 16} ×
where the factor 1 2 is due to the tie when ℜ0 = ℜ7 . By P {ℜ1 = 16 N = 8} ≤ (1 2 ) P {ℜ1 = 16 N = 8} (27)
enumerating all possibilities, it is not possible to have a tree- = 6.118 × 10−4
way tie among ℜ0 , ℜ7 , and ℜ25 when N = 7. Therefore, where the factor 1 2 is an upper bound and is obtained when
ℜ0 = ℜ1 = 16. From (22), (23), (25), and (27), we get
P {S7 chosen N = 7} = P {Si chosen N = 7} (15)
31
where i = 16 and 25. As a result, P {symbol error N = 8} = ∑ P {Si chosen N = 8}
31 i =1
P {symbol error N = 7} = ∑ P {Si chosen N = 7} = P {S7 chosen N = 8} + P {S16 chosen N = 8} + (28)
i =1 (16) P {S 25 chosen N = 8} + 18P { S1 chosen N = 8}
= 3P {S7 chosen N = 7} = 1.53 × 10−3. ≤ 0.0207
When N = 8 , we have ℜ0 = 16 , and the possible values of where the factor 18 in (28) is due to the fact that there are
ℜ7 are in the set of [ −12, − 8, − 4, 0, 4, 8, 12, 16, 20] . Since a eighteen branches (including ℜ1 ) that have hi = 0 .
symbol error occurs when ℜ7 ≥ ℜ0 , we need to find the Repeating the above process, we obtain the remaining
conditional probabilities of symbol error for 9 ≤ N ≤ 32 . The
conditional probabilities that ℜ7 = 16 and ℜ7 = 20 given that overall conditional probabilities of symbol error for the CCSK
N = 8 , respectively. From (8), sequence chosen for JTIDS are listed in the second column of
P {ℜ7 = 20 N = 8} = 2.855 × 10−4 = P {ℜi = 20 N = 8} (17) Table VI. Since the non-zero values (except N = 7 ) are upper
bounds, the analytic conditional probabilities of symbol error
where i = 16 and 25, and
are denoted as ζ UBi . Note that when N ≥ 11 , the upper bound
P {ℜ7 = 16 N = 8} = 5.873 × 10−3 = P {ℜi = 16 N = 8} (18)
becomes very loose and exceeds one. When this occurs, the
where i = 16 and 25. Next, we have upper bound is given as one.
P {S7 chosen N = 8} = P {S7 chosen N = 8, ℜ7 = 20} ×
P {ℜ7 = 20 N = 8 } + P {S7 chosen N = 8, ℜ7 = 16} × (19) III. PROBABILITY OF SYMBOL ERROR FOR CCSK
P {ℜ7 = 16 N = 8 } Now, combining (11) and (12), we obtain an upper bound
where for the probability of symbol error for CCSK as
P {S7 chosen N = 8, ℜ7 = 20} = 1,
32
⎛ 32 ⎞
(20) PS < ∑ ζ UB j ⎜ ⎟ Pc j (1 − Pc )
32 − j
(29)
and j =0 ⎝ j⎠
1 1 where Pc is the probability of chip error at the output of the
≤ P {S7 chosen N = 8, ℜ7 = 16} ≤ (21) MSK chip demodulator. MSK can be considered as a special
3 2
where the factor 1 2 is obtained when ℜ0 = ℜ7 = 16 , and the case of offset quadrature phase-shift keying (OQPSK) with
sinusoidal pulse shaping. When a coherent matched filter or
factor 1 3 is obtained when ℜ0 = ℜ7 = ℜ25 = 16 since it is correlator is used to recover the chips, MSK has the same
possible to have a three-way tie when N = 8 . Using (17), performance as BPSK, QPSK, and OQPSK [8]; that is,
(18), (20), and (21) with the upper bound in (19), we obtain ⎛ 2 Ec ⎞
P {S7 chosen N = 8} ≤ 3.222 × 10−3. Pc = Q ⎜
⎜ N ⎟⎟
(22) . (30)
⎝ 0 ⎠
Note that
Since Es = 5Eb = 32 Ec , we can rewrite (30) as
P {S7 chosen N = 8} = P {S25 chosen N = 8} (23)
⎛ 10 Eb ⎞
Pc = Q ⎜
⎜ 32 N ⎟⎟
since S7 and S25 have similar properties; however, . (31)
⎝ 0 ⎠
P {S7 chosen N = 8} ≠ P {S16 chosen N = 8} (24)
The actual JTIDS waveform is received noncoherently at the
since P {S16 chosen N = 8, ℜ16 = 16} = 1 2 ; that is, there is not chip level, but in this paper the performance of a JTIDS-type
waveform with coherent detection is evaluated in order to
a three-way tie when N = 8 and ℜ16 = 16 ; As a result, ascertain the performance possible if coherent chip
P {S16 chosen N = 8} = 3.222 × 10−3. (25) demodulation were practical. The analysis presented in this
paper can easily be modified to evaluate performance with and are shown in the third column of Table VI.
noncoherent chip demodulation. Modifying (29), we obtain the simulation result for the
Substituting (31) into (29), we obtain the results shown in probability of symbol error of CCSK in AWGN from
Figure 1. To compare the difference between orthogonal 32
⎛ 32 ⎞
PS = ∑ ζ SIM j ⎜ ⎟ Pc j (1 − Pc ) .
32 − j
signaling and quasi-orthogonal CCSK, the probability of (32)
j =0 ⎝ j⎠
symbol error for 32-ary orthogonal signaling is also shown in
the figure. As expected, the performance of 32-chip CCSK is To compare the difference between the analytic upper
inferior to that of 32-ary orthogonal signaling by about 2 dB at bound and the simulation, both results are shown in Figure 2.
As is seen, the analytic result given in (29) is a tight upper
PS = 10−5 ; however, the advantage of using CCSK is that only
bound.
one detector branch is required to recover the original symbol
0
instead of thirty-two detector branches. 10
Simulation
-1
0 10 Upper bound
10
32-ary orthogonal signaling -2
-1 10
10 32-chip CCSK, upper bound
-3
-2 10
10
-4
-3 10
10

Ps
-5
-4 10
10
Ps

-6
-5 10
10
-7
-6 10
10
-8
-7 10
10
-9
-8 10
10 0 2 4 6 8 10 12 14
E b/No (dB)
-9
10
0 2 4 6 8 10 12 14 Fig. 2. Probability of symbol error for the 32-chip CCSK sequence chosen for
E b/No (dB) JTIDS in AWGN: Monte Carlo simulation versus analytic upper bound.
Fig. 1. Probability of symbol error in AWGN: the 32-chip CCSK sequence
chosen for JTIDS versus 32-ary orthogonal signaling. V. NEW CCSK SEQUENCE
Recall that the 32-chip CCSK sequence chosen for JTIDS
IV. SIMULATION
has a maximum off-peak cross-correlation value H = 4 .
To check the analytic upper bound derived in the last Intuitively, the performance of CCSK can be improved if the
section, a Monte Carlo simulation with stratified sampling [9] maximum off-peak cross-correlation value is smaller than
is written to obtain the conditional probabilities of symbol four. Based on this idea, a search algorithm was created and a
error for 7 ≤ N ≤ 32 . This simulation is implemented in a new 32-chip CCSK sequence was found. This new sequence
manner similar to that of finding the analytical upper bound; and its cyclically shifted versions are shown in Table V.
that is, the simulation is done case-by-case for different N .
Starting from N = 7 and given that symbol 0 is sent, the TABLE V NEW 32-CHIP CCSK SEQUENCE.
major steps of the simulation are as follows. First, in each 5-bit
New 32-chip CCSK sequence
iteration, (i) generate randomly a 32-chip sequence with seven symbol
chip errors relative to the original 32-chip sequence for 00000 S0 = 10111010001111010010000001100110
symbol 0 to model noisy 32-chip sequence, (ii) cross-correlate 00001 S1 = 01110100011110100100000011001101
the noisy 32-chip sequence with all of the 32 local sequences 00010 S2 = 11101000111101001000000110011010
to yield 32 cross-correlation values ℜ0, ℜ1,..., ℜ31 , (iii) calculate ⋮ ⋮
11111 S31 = 01011101000111101001000000110011
the probability of symbol error based on the following rules: if
ℜi > ℜ0 for 1 ≤ i ≤ 31 , the conditional probability of symbol
Given that symbol 0 is sent in the absence of noise, this new
error is one; if ℜi = ℜ0 for 1 ≤ i ≤ 31 , the conditional
sequence has a maximum off-peak cross-correlation value
probability of symbol error is τ (τ + 1) , where τ is the total H = 0 . With the same approach used to evaluate the CCSK
number of ties; if ℜi < ℜ0 for 1 ≤ i ≤ 31 , the conditional sequence chosen for JTIDS, this new sequence is evaluated
probability of symbol error is zero. Second, repeat the above both analytically and by Monte Carlo simulation to obtain the
iteration 10,000 times and calculate the average conditional conditional probabilities of symbol error ζ UB '
j
and ζ SIM
'
j
,
probability of symbol error. Last, repeat the above process for respectively. The results are shown in the fourth and the fifth
N = 8 through N = 32 . The overall simulation results for the column of Table VI. Note that the new sequence allows for
conditional probabilities of symbol error are denoted as ζ SIM j seven instead of six chip errors in the received sequence
without making a symbol error.
0
10
TABLE VI CONDITIONAL PROBABILITIES OF SYMBOL ERROR: THE CCSK
Simulation
SEQUENCE CHOSEN BY JTIDS VERSUS THE NEW CCSK SEQUENCE. -1
10 Upper bound
N= j ζ UB j
ζ SIM j
ζ UB
'
j
ζ SIM
'
j -2
10
0 0 0 0 0 -3
10
1 0 0 0 0
# # # # # 10
-4

Ps
5 0 0 0 0 -5
10
6 0 0 0 0
7 0.0015 0.0015 0 0 10
-6

8 0.0207 0.0194 0.0147 0.0143


-7
9 0.1166 0.1126 0.1040 0.1025 10
10 0.4187 0.3669 0.4023 0.3550 -8
10
11 1.0 0.7093 1.0 0.7140
12 1.0 0.9351 1.0 0.9367 10
-9

13 1.0 0.9953 1.0 0.9956 0 2 4 6 8 10 12 14


E b/No (dB)
14 1.0 1.0 1.0 0.9999
15 1.0 1.0 1.0 1.0 Fig. 4. Probability of symbol error for the new CCSK sequence in AWGN:
analytical upper bound versus Monte Carlo simulation.
# # # # #
32 1.0 1.0 1.0 1.0
VI. CONCLUSION
In this paper, an analytic upper bound on the probability of
Replacing ζ SIM j with ζ SIM
'
in (32), we obtain the
j
symbol error of CCSK is derived for the CCSK sequence
probability of symbol error for the new CCSK sequence in chosen for JTIDS. The probability of symbol error obtained
AWGN. To compare the difference between the CCSK with the analytic upper bound was compared to the probability
sequence chosen for JTIDS and the new CCSK sequence, both of symbol error obtained by Monte Carlo simulation for
simulation results are shown in Figure 3. As can be seen, the AWGN. The results show that the analytic method yields a
results obtained with the new sequence are only slightly better tight upper bound. In addition to the CCSK sequence chosen
than those obtained with the original JTIDS sequence since for JTIDS, a new CCSK sequence with a smaller maximum
performance is determined at the symbol level rather than the off-peak cross-correlation value is introduced and evaluated
chip level. In essence, for practical values of Pc , the first non- both analytically and by Monte Carlo simulation. The results
zero term in (32) is not dominant. obtained for this new sequence compare favorably to the
0
sequence chosen for JTIDS.
10
JTIDS sequence
-1
10 New sequence REFERENCES
-2 [1] M. B. Pursely, T. C. Royster, IV, and M. Y. Tan, “High-rate direct-
10
sequence spread spectrum,” Proc. IEEE Military Commun. Conf., vol. 2,
-3
10
pp. 1101-1106, 2003.
[2] H. Wang, J. Kuang, Z. Wang, and H. Xu, “Transmission performance
-4
10 evaluation of JTIDS,” Proc. IEEE Military Commun. Conf., vol. 4, pp.
2264-2268, 2005.
Ps

-5
10 [3] F. J. Block, “Comparison of jamming robustness of airborne networking
waveforms,” Proc. IEEE Military Commun. Conf., vol. 4, pp. 2119-
2125, 2005.
-6
10

-7
[4] M. B. Pursley, and T. C. Royster, “High-rate direct-sequence spread
10 spectrum with error-control coding,” IEEE Trans. Commun., vol. 54, no.
-8
9, pp. 1693-1702, Sept. 2006.
10 [5] L. Sadiq and A. H. Aghvami, “Performance of a coded hybrid spread
-9 spectrum communication system in the presence of partial band noise
10
0 2 4 6 8 10 12 14 and multiple access interference,” Proc. IEEE Military Commun. Conf.,
E b/No (dB) vol. 3, pp. 817-821, 1988.
[6] S. M. Ross, “Introduction to Probability Models”, 8th ed., San Diego,
Fig. 3. Probability of symbol error in AWGN (simulation): the new CCSK
Academic Press, pp. 99, 2003.
sequence versus the CCSK sequence chosen for JTIDS.
[7] S. M. Ross, “Introduction to Probability Models”, 8th ed., San Diego,
Academic Press, pp. 119, 2003.
Replacing ζ UB j with ζ UB
'
in (29), we obtain the upper [8] S. Pasupathy, “Minimum shift keying: a spectrally efficient
j
modulation,” IEEE Commun. Mag., vol. 17, Issue 4, pp. 14-22, Jul.
bound of the probability of symbol error for the new CCSK 1979.
sequence in AWGN. To compare the difference between the [9] S. M. Ross, “Simulation,” 3rd ed., San Diego: Academic Press, pp. 158-
167, 2002.
analytic upper bound and the simulation, both results are
shown in Figure 4. As before, the analytic results given by
(29) yield a tight upper bound.