This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts

for publication in the WCNC 2009 proceedings.

OFDM System Identification for Cognitive Radio

Based on Pilot-Induced Cyclostationarity
François-Xavier Socheleau∗ , Philippe Ciblat∗∗ and Sébastien Houcke∗
∗ Institut TELECOM; TELECOM Bretagne, UMR CNRS 3192 Lab-STICC, Université européenne de Bretagne,

Email: {fx.socheleau, sebastien.houcke}@telecom-bretagne.eu

∗∗ Institut TELECOM; TELECOM ParisTech,

Email: philippe.ciblat@telecom-paristech.fr

Abstract—In the context of cognitive radio, this paper ad-
dresses the challenge of OFDM system identification. We show
that it is possible to take advantage of pilot tone structures to
perform identification. Pilot subcarrier usage is of great interest
since it enables to discriminate OFDM systems that have the
same modulation parameters (subcarrier spacing, cyclic prefix
duration, etc.). The proposed method relies on the periodic re-
dundancy often induced between pilot symbols. We demonstrate
that most OFDM systems (Wifi, WiMAX, DVB-T, etc.) can be
identified thanks to this redundancy by conducting a hypothesis
test based on second order statistics. Detailed numerical examples
demonstrate the efficiency of the proposed identification criterion
in various kinds of environments.
I. I NTRODUCTION
for channel estimation and/or synchronization purposes, are
often replicated with regard to a pre-defined time/frequency
distribution. This property induces correlation between pilot
subcarriers that can be exploited in conjunction with the
periodicity of the time/frequency pilot mapping to perform
system identification. In that case, identification can be based
upon the knowledge of pilot structures without knowledge of
pilot symbols. We hereafter generalize the pilot correlation
property and adapt it to the context of cognitive radio under
the concept of pilot-induced cyclostationary (PIC) signatures.
We show that these signatures can be reliably identified thanks
to a hypothesis test based on 2nd order statistics. We here
suggest to take advantage of signatures created as a side-effect
of existing PIC structures to identify standards such as Wifi,

T HE increasing demand of new services constrains wire-

less systems to move towards a more intelligent control
of the limited spectrum resources. [1] laid the foundation of
WiMAX and DVB-T and also advocate to extend PIC use to
future cognitive systems.
The paper is organized as follows: Section II describes
the Cognitive Radio (CR) concept that is now considered as the pilot-assisted OFDM system model. Section III recalls
a key paradigm to evolve from the current fixed spectrum the cyclostationarity basis and introduces the pilot induced
usage to opportunistic spectrum sharing models tailored to cyclostationary (PIC) signature scheme with its corresponding
address the frequency resource scarcity issue. CR promises the identification algorithm. Identification performance is assessed
deployment of flexible wireless systems able to reconfigure through simulations in Section IV. Finally, conclusions are
themselves (i.e. adapt their modulation parameters, carrier presented in Section V.
frequency, power, etc.) with regard to the surrounding radio en-
vironment and regulatory policies. Spectrum sensing and espe- II. P ILOT- ASSISTED OFDM SYSTEM MODEL
cially system identification is therefore a crucial step towards
radio environment awareness. In this paper we focus on OFDM A. Signal model
based systems as it becomes the physical layer for many Assuming that an OFDM symbol consists of N subcarriers,
wireless standards [2]–[5]. Existing OFDM identification algo- the discrete-time baseband equivalent transmitted signal is
rithms are mainly based on modulation parameter estimation given by 
[6]–[9]. Considering the increasing interest in OFDM by the Es  
wireless designers, modulation parameters of such systems are x(m) = xd (m) + xt (m) , (1)
N
likely to become closer and closer. For instance, 3GPP/LTE [5]
and Fixed WiMAX [2] systems already have a subcarrier spac- where
ing only different from 4% which may prevent from getting an −1
 N n
accurate system identification based on the subcarrier spacing xd (m) = ak (n)e2iπ N (m−D−k(N +D))
estimation principle. To overcome this limitation, few methods k∈Z n=0
n∈I(k)
/
involving more particular signatures have been suggested in
[10]–[12]. These methods either add overhead, and thus reduce .g(m − k(N + D)),
systems capacity, or are only dedicated to OFDM systems with and
specific pilot tone configurations.   n
In this contribution, we propose a solution to address both xt (m) = bk (n)e2iπ N (m−D−k(N +D))
issues. By studying existing OFDM systems (Wifi, WiMAX, k∈Z n∈I(k) (2)
DVB-T etc.), it can be noticed that pilot symbols, used .g(m − k(N + D)).

978-1-4244-2948-6/09/$25.00 ©2009 IEEE

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 21, 2009 at 08:29 from IEEE Xplore. Restrictions apply.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings.

Es is the signal power and ak (n) are the transmitted data varies. Pilot tones are mapped onto certain subcarriers of
symbols at the n-th subcarrier of the k-th OFDM block. These each OFDM symbol such that I(k) = I where I is any
data symbols are assumed to be independent and identically subset of {0, · · · , N − 1} ∀k. The last arrangement is a
distributed (i.i.d), D is the CP length and m → g(m) is the circular configuration where the set of pilot subcarrier indexes
pulse shaping filter. I(k) denotes the set of pilot subcarrier periodically change such that I(k + K) = I(k) where K ∈ Z
indexes of the k-th symbol and bk (n) are the pilot symbols. and I(k) = ∅. Such a scheme is used under the assumption of
Let {h(l)}l=0,··· ,L−1 be the baseband equivalent discrete- fast fading channel and presents the interest of avoiding cases
time channel impulse response of length L. Unless otherwise where a given pilot subcarrier is attenuated by the channel for
stated, the channel is assumed to be time invariant. Notice that a period of several symbols. Note that some OFDM systems
impact of channel variations is discussed in Section IV. The (e.g. [4] and [13]) make a joint use of these arrangements.
received samples of the OFDM signal are then expressed as

L−1 III. P ILOT- INDUCED CYCLOSTATIONARITY (PIC)
m−τ
y(m) = e−i(2πε N +θ)
h(l)x(m − l − τ ) + η(m), (3) IDENTIFICATION
l=0 A. Background on cyclostationarity
where ε is the carrier frequency offset (normalized by the Two complex discrete time stochastic processes xk and
subcarrier spacing), θ the initial arbitrary carrier phase, τ yk , k ∈ Z, are said to exhibit joint second-order almost
the timing offset and η(m) a zero mean circularly-symmetric cyclostationarity in the wide sense if the cross correlation
complex-valued white Gaussian noise of variance σ 2 per com- function
plex dimension. The receiver does not know these parameters   ∗

Rxy (k, u) = E xk yk+u ,
({h(l)}l=0,··· ,L−1 , ε, θ, τ, σ 2 , ak (n), bk (n)).
where ∗ stands for complex conjugation, admits series repre-
sentation 
B. Pilot tone arrangements
Rxy (k, u) = α
Rxy (u)ei2παk
As depicted in Figure 1, three different types of pilot α∈Axy
tone arrangements are usually set up in OFDM systems to
meet the channel estimation requirements. The first one is the where
block type configuration used under the assumption of slow  1 
M

fading channel. Pilot tones are in that case mapped onto all
α
Rxy (u) = lim Rxy (k, u)e−i2παk (4)
M →+∞ 2M + 1
subcarriers of OFDM symbols within a specific period K so k=−M

that I(k) verifies is the cyclic cross correlation function (CCCF) and
  
{0, · · · , N − 1}, if k = mK (m ∈ Z)  
I(k) = Axy = α ∈ [−1/2; 1/2[ ∃u s.t. Rxy α
(u) = 0
∅ otherwise.
The second one, comb-type pilot configuration, is introduced is a countable set of cycle frequencies α [14]. Note that the
to satisfy the need for equalizing when the channel quickly CCCF is periodic in α with period 1.

Fig. 1. Examples of pilot tone arrangement: a) block type configuration, b)
comb-type configuration, c) circular configuration.
B. Signature generation
As illustrated in subsection II-B, the time/frequency pilot
tone distribution is always deterministic to meet the channel
estimation requirements. As the number of pilot tones is finite,
the deterministic characteristic of the pilot tone distribution
can be expressed as I(k + K) = I(k), K ∈ Z for any
combination of arrangements described in subsection II-B. For
the particular case of comb-type arrangement, note that K = 1.
Such a periodicity is a useful property that can be exploited
to induce cyclostationarity in OFDM frames through careful
choice of pilot symbols bk (n).
Let ck (n) be the k-th symbol on subcarrier n such that

bk (n), if n ∈ I(k)
ck (n) = (5)
ak (n) otherwise.
If the pilot tones are designed such that
bk (p) = bk+d(p,q) (q)eiϕ (6)
with d(p,q) ∈ Z and ϕ ∈ [−π; π[ then, as shown
in appendix A, processes {ck (p)}k and {ck (q)}k are

jointly cyclostationary with A(p,q) = Ack (p)ck (q) =

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 21, 2009 at 08:29 from IEEE Xplore. Restrictions apply.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings.


m−K/2
K ,m ∈ {0, 1, · · · , K − 1} where
 stands for signal gain, each term Yk (n) in Eq. (9) is normalized so that
integer flooring. Yk (n)
In the framework of OFDM system identification, joint Ỹk (n) =  , (10)
cyclostationary structure is of great interest to generate system  [Y (n)]
Var
specific signatures. Eq. (6) indicates that it is possible to design
where Var[.] denotes the variance and
a given signature S by choosing particular combinations of
 −1
p, q, d(p,q) and K. Such a signature S is then defined as M


  [Y (n)] = 1
Var
2
|Yk (n)| . (11)
 M
S= p, q, d(p,q) , K |A(p,q) = ∅ . (7) k=0

As examples, DVB-T, Wifi and WiMAX pilot structures can D. Decision statistics
be seen as (unintentional) PIC signatures where Considering that the system to be identified embeds a PIC
(p,q)
• DVB-T: K = 4, d = 0, 1, 2 or 3, (p, q) can be any signature defined by Eq. (7), our identification problem pre-
pilot tone pair. viously described boils down to a standard detection problem
(p,q)
• Wifi: K = 1, d = 0, (p, q) can be any pilot tone pair. that can be formulated as the following binary hypothesis test
• WiMAX: see section IV. ⎧
⎪
⎪ H0 : y(m) writes as in Eq. (3) without PIC structure
Note that the term eiϕ in Eq. (6) is here to bring flexibility ⎪
⎪
⎪
⎪ or with

PIC structure defined
 as
to PIC structures and to prevent coherent addition of pilot ⎪
⎨  

symbols which increases the peak-to-average power ratio. S = p , q , d(p ,q ) , K |A (p ,q ) = ∅
⎪
⎪ such that S = S.
⎪
⎪
⎪
C. Identification cost function ⎪ H1 : y(m) writes as 
⎪
⎩
in Eq. (3) with PIC structure
defined as S = p, q, d(p,q) , K |A(p,q) = ∅ .
Systems satisfying Eq. (1) and (7) have a (12)
(p,q)
periodic
 cross-correlation
 function R c (p,q) k, d = To decide the most likely hypothesis, we propose a detection
E ck (p)c∗k+d(p,q) (q) and can thus be discriminated by test constrained by the asymptotic false alarm probability
exploiting the cyclic cross-correlation function (CCCF) similar to what is suggested in [12]. The decision is made
by comparing JP IC to a positive threshold such that
 1
Rcα(p,q) (d(p,q) ) = lim . H1
2M + 1
M →+∞ JP IC < Λ,
>


M
  H0
E ck (p)c∗k+d(p,q) (q) e−i2παk . with Λ defined as
k=−M
FJP IC |H0 (Λ) = 1 − Pf a (13)
We propose to perform the system identification on the eval-
uation of the CCCF energy at cyclic frequencies α ∈ A(p,q) . where FJP IC |H0 is the cumulative distribution function (cdf)
In practice the symbols ck (n) are only accessible via the of JP IC when H0 holds and Pf a is the tolerated false alarm
observations Yk (n) expressed as probability. Moreover, we based our decision on the following
N −1
assumptions:
 1  nm Assumption 1: We assume that the number of possible
Yk (n) = √ y[k(N + D) + D + m]e−2iπ N .
N m=0 PIC signatures  are large enough to design OFDM systems
satisfying S S = ∅, ∀S = S.
Considering the channel frequency response and the time and Assumption 2: Under H0 , at a given subcarrier p, the
frequency offsets unknown at reception, we therefore have to symbols {ck (p)}k are assumed to be i.i.d.
consider the CCCF of (Yk (p), Yk (q)) instead of (ck (p), ck (q)). The relevance of the hypothesis test given by Eq. (12) can
If we assume that M OFDM symbols are available at recep- be easily justified by looking at the mean of the estimated
tion, the CCCF energy of yk (p) and yk (q) is evaluated thanks CCCF, defined in Eq. (9), under both hypotheses.
to the cost function defined as 


⎛ ⎞ E R̂Ỹα (p,q) d(p,q) = RỸα (p,q) d(p,q)
   
2
 
JP IC = ⎝ R̂Ỹα (p,q) d(p,q)  ⎠ (8) 0 under H0
(p,q)∈ξ α∈A(p,q) = μ(p,q)
K , ∀α ∈ A (p,q) under H1
where

 1 where μ(p,q) denotes a term proportional to the subcarrier
(p,q)
R̂Ỹα (p,q) d = . to noise ratio. It is then straightforward to see that the distance
M − d(p,q)
(p,q)
between the two asymptotic probability distribution functions
M −d
 −1 (9)
∗ −i2παk of JP IC is non null and makes possible the discrimination
Ỹk (p)Ỹk+d (p,q) (q)e
between the two hypotheses.
k=0
In order to find a relevant threshold Λ to perform the
and ξ = {(p, q)|A(p,q) = ∅ and d(p,q) + K ≤ M }. Note that identification, we hereafter derive the asymptotic statistical
in order to get the criterion JP IC less sensitive to the received behavior of JP IC under H0 .

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 21, 2009 at 08:29 from IEEE Xplore. Restrictions apply.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings.

As shown in appendix B, under hypothesis H0 , JP IC IV. S IMULATIONS

is a sum of weighted noncentral chi-square variables with A. Simulation context
a cumulative distribution function that can be expressed as
In the following, all the results are averaged over 1000
Laguerre series of the form
Monte Carlo runs. The asymptotic false alarm probability Pf a
x
e− 2ω xζ is fixed to 0.01. The Signal-to-Noise
 Ratio (SNR) is defined
FJP IC |H0 (x) = . as SNR(dB) = 10log10 Es /σ 2 . When notified, the propaga-
(2ω)ζ+1 Γ(ζ + 1)
 k!mk (ζ) (ζ + 1)x 
 tion channel simulated is a time-variant discrete-time channel
L
(ζ + 1)k k 2ων
, {hk (l)}l=0,··· ,L with L = D and an exponential decay profile
2 −l/β
k≥0 for its non-null component (i.e., E[|h" k (l)| ] = Ge for
 (ζ) l = 0, · · · , L and G is chosen such that l=0 E[|hk (l)|2 ] = 1).
L
∀ν > 0 and ω > 0,with ζ = card(A(p,q) ) and Lk the
The channel time variation is simulated using Jake’s model.
(p,q)∈ξ
k-th generalized Laguerre polynomial verifying For the simulations, uniformly distributed random ε and τ
were generated with −0.5 ≤ ε ≤ 0.5 and −0.5(N + D) ≤
k  
(ζ)   k+ζ (−x)m τ ≤ 0.5(N + D).
Lk (x) = .
k−m m!
m=0 B. PIC signatures identification performance
 ∞
Γ(x) = 0 tx−1 e−t dt and (.)k denotes the Pochhammer We here simulate a 512-subcarrier OFDM system with a
 subcarrier mapping scheme equivalent to the Mobile WiMAX
symbol defined as (x)k = Γ(x+k)
Γ(x) . The coefficients mk satisfy
DL-PUSC [13] one. We recall that it is designed with 60 pilot,
the following recurrent relation
360 data, 91 guard and 1 DC subcarrier. D is set to 64. K =
1
k−1 2 and joint cyclostationarity is induced on 1770 pilot pairs
mk = mj gk−j , k ≥ 1 with d(p,q) = 0 or 1. To limit the identification algorithm
k j=0
computational complexity, we test only 30 pairs among all
with admissible (p, q).
In Figure 2, we plot the correct detection probability versus
ω ζ+1
m0 =2(ζ + 1)ζ+1 . SNR in the context of AWGN channel for various observation
ζ +1−ν window lengths. We observe that the PIC criterion exhibits
!  ζ +1−ν
−card(A(p,q) )
excellent performance from -6 dB for a number of symbols as
ων +
2(M − d(p,q) ) small as 25. The identification rate is significantly improved
(p,q)∈ξ
as the number of available symbols increases. Figure 2 also
and depicts the impact of different synchronization assumptions on
 j the detection probability. On this figure, it can be seen that the
−ν
gj = + loss due to missynchronization can be very high (up to 6dB).
ζ +1−ν
 j This justifies the need for a synchronization method based on
 ν(2(M − d(p,q) )ω − 1) Eq. (14) as after correction, this loss is reduced to less than
card(A(p,q) ) ,
2(M − d(p,q) )ων + ζ + 1 − ν 0.5dB for M = 25. ε and τ were estimated by maximizing
(p,q)∈ξ
JP IC over a grid with a step of 0.2 over ε and 0.15(N + D)
j ≥ 1. over τ .
Note that the Laguerre expansion depends on ν and ω that
can be arbitrarily chosen. These parameter choice only have 1

an impact on the convergence speed and on the uniform con- 0.9

vergence property. Moreover, for computer implementation, 0.8

the Laguerre expansion has to be truncated. The number of 0.7

terms to consider within the series can be estimated using the 0.6

truncation error analytical expression given in [15].

det

0.5
P

0.4

E. Effect of synchronization errors 0.3

M=25, perfect knowledge of (ε,τ)
M=25,estimation of (ε,τ)
0.2 M=25,no synchronization
Timing missynchronization and/or frequency offset deterio- M=50, perfect knowledge of (ε,τ)

rates the observations Yk (n) as inter-symbol (ISI) and inter-

0.1 M=50,estimation of (ε,τ)
M=50,no synchronization
0
carrier (ICI) interferences occur. These interferences result in −16 −14 −12 −10 −8 −6
SNR (B)
−4 −2 0 2 4

attenuation of the cost function JP IC which has a bad impact

on identification performance as it reduces the distance be-
tween the distribution of JP IC |H1 and JP IC |H0 . To overcome
this performance loss, ε and τ can be estimated as
[ε̂, τ̂ ] = argmax JP IC .
(ε,τ )
Fig. 2. Effect of SNR and the observation window length as well as
synchronization assumptions on the correct detection probability.
(14) In Figure 3, we plot the correct detection probability versus
SNR when the frequency-selective channel is time-variant.

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 21, 2009 at 08:29 from IEEE Xplore. Restrictions apply.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings.

1 V. C ONCLUSION
0.9

0.8 In this paper, we introduced a new OFDM system iden-

0.7 tification algorithm to address part of the spectrum sensing
0.6 challenge inherent in cognitive radio. The method is based
on cyclostationary patterns embedded onto pilot tones. This
Pdet

0.5

0.4 has the main advantage of not adding any system overhead
0.3
and makes possible the discrimination of systems with similar
0.2
f =0Hz
d modulation parameters. Cyclostationary patterns presented in
f =100Hz

0.1
d
f =250Hz
d
this contribution consider existing pilot constraints so that
0
f =500Hz
d the method is compatible with most standard requirements.
−14 −12 −10 −8 −6 −4 −2 0 2
SNR (dB) Simulation results proved the efficiency of the proposed
identification criterion and showed its robustness to harsh
Fig. 3. Effect of Doppler spread on the correct detection probability (M = propagation environment as well as its gain compared with
25, β = 0.5D). state of the art.

Various values of maximum Doppler frequencies fd have been R EFERENCES

inspected. We can see that our algorithm is quite robust to
Doppler spread below 100Hz and that performance degrades
up to 4dB for fd = 500Hz (at 3GHz, this corresponds to
a relative velocity of 180kph). However, note that even for
fd = 500Hz, the identification performance is still excellent
in the standard operating SNR range in a mobile environment.
C. PIC vs m-sequence identification
We here compare the PIC method with the state of the
art pilot structure based identification algorithm described in
[12]. This algorithm relies on the property of m-sequences that
modulate pilots of systems like [3] and [2]. As shown in [12],
fixed WiMAX can be identified thanks to m-sequences but
can also be identified using the PIC criterion described in this
paper. Indeed, from [2], we get K = 1 and d(p,q) = 0 with ξ
defined as all possible pair of pilot tones (that is card(ξ) = 28),
D is set to 32.
Figure 4 compares both methods and clearly indicates that
for fixed WiMAX identification, the PIC criterion outperforms
(up to 4dB) the m-sequence based algorithm.
[10] P.D Sutton, K.E. Nolan, and L.E. Doyle, “Cyclostationary Signatures
1
[11] K. Maeda, A. Benjebbour, T. Asai, T. Furuno, and T. Ohya, “Recogni-
0.9
0.8
0.7
[12] F.X Socheleau, S. Houcke, A. Aissa-El-Bey, and P. Ciblat, “OFDM
0.6
Pdet
[1] J. Mitola III, Cognitive Radio An Integrated Agent Architecture for Soft-
ware Defined Radio, Ph.D. thesis, KTH Royal Institute of Technology,
Stockholm, Sweden, 2000.
[2] IEEE Std. 802.16, Part 16: Air Interface for Fixed Broadband Wireless
Access Systems, 2004.
[3] IEEE Std. 802.11a, Part 11: Wireless LAN Medium Access Control
(MAC) and Physical Layer (PHY) specifications High-speed Physical
Layer in the 5 GHz Band, 1999.
[4] ETSI EN 300 744 V1.5.1, Digital Video Broadcasting (DVB); Framing
structure, channel coding and modulation for digital terrestrial televi-
sion, 2004.
[5] 3GPP TR 25.814, Physical Layer Aspects for Evolved UTRA, v.2.0.0,
2006.
[6] T. Yucek and H. Arslan, “OFDM Signal Identification and Transmission
Parameter Estimation for Cognitive Radio Applications,” in IEEE Glob.
Telecom., Nov. 2007, pp. 4056–4060.
[7] A. Bouzegzi, P. Ciblat, and P. Jallon, “Maximum Likelihood based
methods for OFDM intercarrier spacing characterization,” in IEEE Symp.
on Personal, Indoor and Mobile Radio Communications, Sept. 2008.
[8] A. Bouzegzi, P. Jallon, and P. Ciblat, “A fourth-order based algorithm
for characterization of OFDM signals,” in IEEE Workshop on Signal
Processing Advances in Wireless Communications, Jul. 2008.
[9] M. Oner and F. Jondral, “Air interface recognition for a software radio
system exploiting cyclostationarity,” in IEEE Int. Conf. on Personal,
Indoor and Mobile Radio Communications, Sept. 2004.
in Practical Cognitive Radio Applications,” IEEE Journal Sel. Areas in
Comm., vol. 26, no. 1, pp. 13–24, Jan. 2008.
tion Among OFDM-Based Systems Utilizing Cyclostationarity-Inducing
Transmission,” in IEEE Symp. on New Frontiers in Dynamic Spectrum
Access Networks, 2007, pp. 516–523.
system identification based on m-sequence signatures in cognitive radio

0.5 context,” in IEEE Conf. on Personal, Indoor and Mobile Radio

Communications, Sept. 2008.
0.4
[13] IEEE Std. 802.16, Part 16: Air Interface for Fixed and Mobile Broad-
0.3
band Wireless Access Systems, Amendment 2: Physical and Medium
0.2 Access Control Layers for Combined Fixed and Mobile Operations in
License Bands and Corrigendum 1, 2005.
0.1 m−sequence
PIC
[14] W. A. Gardner, A. Napolitano, and L. Paurac, “Cyclostationarity: Half
0
−14 −12 −10 −8 −6 −4 −2 0 2 4 6 a century of research,” Signal processing, vol. 86, no. 4, pp. 639–697,
SNR (dB) Apr. 2006.
[15] A. Castano Martinez and F. Lopez Blazquez, “Distribution of a sum
Fig. 4. Identification performance comparison between PIC and m-sequence of weighted noncentral chi-square variables,” Sociedad de Estadistica e
criteria (M = 25, β = 0.5D). Investigacion Operativa test, vol. 14, no. 02, pp. 397–415, 2005.
[16] P.J. Brockwell and R.A. Davis, Time Series: theory and analysis,
Springer, 1991.

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 21, 2009 at 08:29 from IEEE Xplore. Restrictions apply.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings.

A PPENDIX A Once again due to Assumption 1 and due to the circularity of

(ck (p), ck (q)) CCCF DERIVATION symbols (PSK or QAM), we get
 
The CCCF of ck (p) and ck (q) is defined as
E Ỹki (p)Ỹk∗i +d(p,q) (q)Ỹk∗j (p)Ỹkj +d(p,q) (q)

M
     
 1
Rcα(p,q) (u) = lim E ck (p)c∗k+u (q) e−i2παk . = E Ỹki (p)Ỹk∗j (p) E Ỹk∗i +d(p,q) (q)Ỹkj +d(p,q) (q) .
M →+∞ 2M + 1
k=−M
From Assumption 2, this term is different from zero only
The data symbols ak (n) being i.i.d and bk (p) = if ki = kj . Moreover, as shown in Eq. (10), Ỹk (n) is
bk+d(p,q) (q)eiϕ , it follows that expressed as a ratio of two random variables. The variance
   estimator introduced in Eq. (11) being consistent, it converges
E ck (p)c∗k+d(p,q) (q) = σb2 e−iϕ δ[k − K − k0 ]
almost surely to a constant denoted vn so that, thanks to
∈Z
the asymptotic theory developed in [16], Ỹk (n) converges in
where δ[.] is the Kronecker delta, σb2 is the variance of distribution to Y (n)/√vn . Thus, for ki = kj and a time
symbols bk (n) and k0 is the index of the first OFDM symbol invariant propagation channel
embedding pilot tones. Hence,  
2 −i(2παk0 +ϕ)    E Ỹki (p)Ỹk∗i +d(p,q) (q)Ỹk∗j (p)Ỹkj +d(p,q) (q)
σ e m    
Rcα(p,q) (d(p,q) ) = b δ α− . E |Yki (p)|2 E |Yki (q)|2
K K =
m∈Z
vp vq
(p,q)
Therefore,
 for α ∈ [−1/2; 1/2[, R
α
c (p,q) (d ) =
 0 iff α ∈ = 1
m−K/2
, m ∈ {0, 1, · · · , K − 1} .  
K
since vn = E |Yk (n)|2 and thanks to assumption 2. There-
fore, the asymptotic covariance is expressed as
A PPENDIX B
(p,q)
JP IC |H0 CDF DERIVATION 1
M −d −1
[C]i,j =  e−i2πk(αi −αj )
We first derive
the expectation and variance of the CCCF M − d (p,q) 2
R̂Ỹα (p,q) d(p,q) defined in Eq. (9).   k=0

 sin π M − d(p,q) (αi − αj ) −iπ(αi −αj )(M −d(p,q) −1)
The expectation of R̂Ỹα (p,q) d(p,q) is given by =  2 e .

 M − d(p,q) sin(π(αi − αj ))
E R̂Ỹα (p,q) d(p,q) (16)
 −1  
(p,q)
M −d From Eq.  (16) we canα deduce  that the correlation between
1 ∗ −i2παk
= (p,q)
E Ỹk (p)Ỹk+d (p,q) (q) e R̂Ỹαi(p,q) d(p,q) and R̂Ỹ j(p,q) d(p,q) for αi = αj is upper
M −d bounded by

k=0
(p,q)   1
= RỸ (p,q) d
α
.  
[C]i,j  ≤  2
  M −d (p,q) sin(π(αi − αj ))
From Assumption 1, E ck (p)c∗k+d(p,q) (q) = 0 so that 1

 ≤  2 .
E R̂Ỹα (p,q) d(p,q) = 0. (15) M − d(p,q) sin(π/K)
 2
To compute the covariance, we introduce the covariance matrix which implies that as long as K  π M − d(p,q) , the
defined as cyclic cross-correlation coefficients are asymptotically mutu-
  ally independent. Thus, thanks to the central limit theorem and
C = E (R − E{R})(R − E{R})H according to Eqs. (15) and (16)

  
where the superscript H stands for transpose conjugate and D 1


 R̂Ỹα (p,q) d(p,q) |H0 −→ CN 0,
M − d(p,q)
R = R̂Ỹα0(p,q) d(p,q) , R̂Ỹα1(p,q) d(p,q) , · · · ,

 T D
αcard(A )−1 where −→ indicates the convergence in distribution. The
R̂Ỹ (p,q) (p,q) d(p,q) . cyclic cross-correlation coefficients estimate being asymptotic
uncorrelated Gaussian variables, it follows that Ỹ |H0 can be
If we now focus on each element [C]i,j =
  (p,q)
αj   ∗  expressed as a sum of weighted central chi-square variables.
E R̂Ỹαi(p,q) d R̂Ỹ (p,q) d(p,q) , we have: Therefore, the characteristic function of JP IC |H0 is expressed
as
1
[C]i,j =  2 . !  it
card(A(p,q) )
M − d(p,q) ψJP IC |H0 (t) = 1− .
M − d(p,q)
M −d(p,q)
 −1   (p,q)∈ξ
E Ỹki (p)Ỹk∗i +d(p,q) (q)Ỹk∗j (p)Ỹkj +d(p,q) (q) The inversion of this characteristic function using the results
ki ,kj =0
presented in [15] concludes the proof.
.e−i2π(ki αi −kj αj ) .

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 21, 2009 at 08:29 from IEEE Xplore. Restrictions apply.