WIRELESS COMMUNICATIONS AND MOBILE COMPUTING
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
Published online 21 August 2009 in Wiley Online Library
(wileyonlinelibrary.com) DOI: 10.1002/wcm.833
Uplink carrier frequency offset estimation for WiMAX
OFDMA-based ranging
Yonghua Lin1∗,† , Da Fan1,2 , Qing Wang1 and Jianwen Chen1
1
IBM China Research Lab, Beijing 100094, China
2
Department of Electronic Engineering, Tsinghua University, Beijing 100084, China
Summary
Ranging is one of the most important processes in the mobile worldwide interoperability for microwave access
(WiMAX) standard, for resolving the uplink synchronization and near/far problems. In this paper, we focus on the
multi-user carrier frequency offset (CFO) estimation in both initial ranging and periodic ranging. After the analysis of
some existing ranging methods, we propose two algorithms based on the correlation properties of pseudo noise (PN)
sequences in time domain and frequency domain respectively. The root mean square error (RMSE) performance is
evaluated in both additive white Gaussian noise (AWGN) channel and multi-path fading channel. Simulation results
show that the proposed frequency-domain cross-correlation method performs better than the proposed time-domain
cross-correlation method, and is more robust to multi-user interference and residual timing offset. Copyright ©
2009 John Wiley & Sons, Ltd.
KEY WORDS: Ranging; OFDMA; WiMAX
1. Introduction
Broadband wireless access (BWA) systems have
attracted much attention in industry for providing
flexible and easy deploymenta solutions to wireless
high-speed communications. A technology developed
to fulfil these characteristics, standardized by IEEE,
is 802.16. It is commonly referred to as worldwide
interoperability for microwave access (WiMAX) [1].
Two versions of WiMAX have been defined, the first is
based on IEEE 802.16-2004 and is optimized for fixed
and nomadic access, the second version is designed to
support portability and mobility, and will be based on
the IEEE 802.16e amendment to the standard [2].
∗
†
One of the most interesting PHY modes supported
by WiMAX standard is orthogonal frequency division
multiple access (OFDMA) PHY mode. In OFDMA,
subcarriers are grouped into subchannels which are
assigned to multiple users for simultaneous transmissions. But OFDMA inherits the weakness of OFDM,
being sensitive to symbol timing offset (STO) and carrier frequency offset (CFO). STO is induced by the
round trip delay (RTD) between the subscriber station
(SS) and the base station (BS) [3], which results in intersymbol interference (ISI). CFO is caused by Doppler
effects and/or poor oscillator alignments. In such case,
it will destroy the orthogonality among different subcarriers, resulting in intercarrier interferences (ICI) and
Correspondence to: Yonghua Lin, IBM China Research Lab, Beijing 100094, China.
E-mail: linyh@cn.ibm.com
Copyright © 2009 John Wiley & Sons, Ltd.
UPLINK CARRIER FREQUENCY OFFSET ESTIMATION
multiuser interferences (MUI) [3]. Though each SS
could establish initial synchronization with the BS by
using the downlink preamble, the uplink signal arriving
at the BS may be plagued by residual synchronization
errors due to Doppler shifts and propagation delays [3].
If the users are not synchronized with the receiver, they
will interfere with each other and the BS will not be able
to recover individual signals of each user. In WiMAX,
the issues of uplink synchronization and near/far problems are addressed by a process called ‘ranging’, and
there are following kinds of ranging are defined:
Initial ranging (IR) for any ranging subscriber station
(RSS) that wants to synchronize to the system for the
first time,
handover (HO) ranging to support mobility and perform handoff from one access point (AP) to another,
periodic ranging (PR) to update and track variations
in STO and CFO,
bandwidth request (BR) ranging to request access to
the shared spectrum resource.
In OFDMA-based ranging, RSS randomly chooses
a ranging time-slot and a ranging code which is modulated by binary phase shift keying (BPSK) modulation.
Then the ranging transmission shall be performed during one OFDM symbol (in periodic ranging) or two
consecutive OFDM symbols (in initial ranging). More
consecutive symbols can also be used for increasing
the probability of code detection [2]. After separating colliding codes and extracting information about
timing, frequency and power, the BS will broadcast
the identified ranging codes with the needed adjustments information (e.g., timing, frequency, and power)
and a status notification (e.g., success, continue, and
abort). The ranging process at BS mainly includes STO
and CFO estimation, and power estimation. This paper
mainly considers the CFO estimation for initial ranging
and periodic ranging processes in WiMAX OFDMA
model. In 802.16d/e, it requires that a precision of 2%
subcarrier spacing should be maintained.
Frequency and timing recovery for single-user
OFDM has been widely discussed in many literatures
(e.g., References [3–6]). However, they cannot be used
in the uplink of a multi-user system which needs to
separate different users at the BS before their timing and frequency offsets are estimated respectively.
Some synchronization methods for multi-user OFDM
systems cannot separate different ranging users when
multiple users collide in the same ranging channel
(e.g., References [3,7–11]). Furthermore, these methods are only suitable for specific subcarrier-assignment
Copyright © 2009 John Wiley & Sons, Ltd.
1445
schemes, such as subband based or interleaved, and
cannot be used for WiMAX OFDMA uplink. Though
several ranging methods for WiMAX OFDMA model
have been analyzed (e.g., References [12–15]), only
code detection, timing offset estimation were discussed
except CFO estimation. In Reference [16], the frequency offset can be acquired by correlating FFT
output samples of two consecutive OFDMA symbols
at ranging subcarriers. However, this method will be
invalid in the case of multiple ranging users as it cannot
distinguish between them. Other ranging methods were
proposed in the literatures (i.e., References [17–19]).
However, their signal models are not consistent with
the current WiMAX standard. Hence, they are not considered in this work.
In this paper, we propose two CFO estimation
methods for WiMAX OFDMA-based initial ranging
and periodic ranging. The proposed algorithms are
evaluated using theoretical analysis and computer simulations over additive white Gaussian noise (AWGN)
and time-dispersive channel with multi-user interference. Simulation results show that the proposed
frequency-domain cross-correlation method performs
better than the proposed time-domain cross-correlation
method even in the case of multiple users simultaneously existing in one ranging time-slot. It is more
robust to residual timing offset resulting from the STO
estimation error.
The remainder of the paper is organized as follows. Section 2 introduces the signal model. The
proposed CFO estimation methods for initial ranging
and periodic ranging are discussed in Section 3 and
4, respectively. The effect from residual timing offset
is analyzed in Section 5. Performance analysis is provided in Section 6, and conclusions are finally drawn
in Section 7.
2. Signal Model
Our system model is mainly based on the IEEE 802.16e
standard [2]. We consider an OFDMA uplink system
with N subcarriers. After assigning direct current (DC)
and guard subcarriers, the remaining subcarriers, Nd ,
are grouped into 0 subchannels. Each subchannel has
P = Nd /Q subcarriers. The subcarriers assigned to
each subchannel are not necessarily adjacent as they
are chosen randomly. Each user in uplink is assigned to
one or more subchannels. For OFDMA-based ranging,
one ranging time-slot is N2 (6 or 8) subchannels
by N1 (1, 2, 3, or 4) OFDMA symbols as shown in
Figure 1. In WiMAX ranging, IR and HO use the
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
1446
Y. LIN ET AL.
(u)
(u)
(u)
[XR,m (0), XR,m (1), · · · , XR,m (N − 1)]T , according to
(u)
XR,m (k)
Fig. 1. Ranging slot allocation.
=
Sm (r),
0,
if k = ind(r)
otherwise
(1)
where ind(r) is the index of the rth ranging subcarrier.
The length of the cyclic prefix (CP) is equivalent to Ng
samples, and assumed to be longer than the maximum
channel delay spread. After N-point inverse fast Fourier
transform (IFFT) and CP insertion at the transmitter,
the nth element of the time-domain ranging signal of
the uth ranging user is given by
for initial ranging
N−1
(u)
1
XR,0 (k)ej2πk(n−Ng )/N
N
k=0
n = 0, · · · , N − 1
s
(u)
xR (n) =
N−1
(u)
1
XR,1 (k)ej2πk(n−Ns )/N
N
k=0
n = Ns , · · · , 2Ns − 1
(2)
for periodic ranging
(u)
Fig. 2. Initial ranging and periodic ranging transmission. (a)
Initial ranging transmission over two consecutive OFDMA
symbols. (b) Periodic ranging transmission over three consecutive OFDMA symbols.
same signal structure, and PR and BR use another
signal structure. So in this paper, only IR and PR are
discussed. N2 is set to 6. N1 is set to 2 for IR and 3
for PR, respectively. A ranging channel is composed
of one or more groups of six adjacent subchannels. As
shown in Figure 2(a), for initial ranging, same ranging
code with length-R is modulated and transmitted in
the ranging channel during one ranging time-slot
composed of two OFDMA symbols. For periodic
ranging, the RSSs modulate three consecutive ranging
codes on the ranging subchannel for a period of three
OFDMA symbols (one code per symbol) as shown
in Figure 2(b). Moreover, a set of pseudo noise (PN)
sequences are selected as ranging codes.
In the ranging time-slot randomly selected by the uth RSS, the ranging code transmitted in the mth OFDM
symbol is, firstly, modulated by BPSK modulation,
which is denoted by Sm = [Sm (0), Sm (1), · · · , Sm (R −
1)]T , where m = 0, 1 for IR and m = 0, 1, 2 for PR.
The signal is then mapped onto the N subcarriers
(u)
of an OFDM symbol in frequency domain, XR,m =
Copyright © 2009 John Wiley & Sons, Ltd.
xR (n) =
N−1
1 (u)
XR,m (k)ej2πk(n−mNs −Ng )/N ,
N
k=0
n = mNs , · · · , (m + 1)Ns − 1
(3)
where m = 0, 1, 2, and Ns = N + Ng . Note that the
second symbol has a cyclic postfix instead of a CP in
initial ranging as shown in Figure 2(a). Similarly, the
data signal of the vth data subscriber station (DSS) is
denoted by
(v)
xD (n) =
N−1
1 (v)
CD,m (k)ej2πk(n−mNs −Ng )/N
N
k=0
n = mNs , · · · , (m + 1)Ns − 1
(4)
(v)
where m = 0, 1 for IR and m = 0, 1, 2 for PR, CD,m (k)
is the kth subcarrier signal transmitted by the vth DSS
in the mth OFDM symbol. Because different SSs will
have different location, the corresponding transmission
delays (dR,u for uth RSS and dD,v for vth DSS) in
units of OFDM samples are different. The maximum
possible relative delay dR,max for the RSS is the roundtrip transmission delay at the cell boundary, which can
be found from the knowledge of cell radius in practice.
The maximum delay dD,max for DSSs is determined by
the timing requirement of the ranging process.
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
UPLINK CARRIER FREQUENCY OFFSET ESTIMATION
We consider a multipath Rayleigh fading channel with L sample-spaced taps. The channel impulse
responses (CIRs) for uth RSS and vth DSS (denoted by
(u)
(v)
hR (l) and hD (l), l = 0, · · · , L − 1, respectively) are
assumed to be constant over one ranging time-slot and
nonzero only for l = 0, · · · , L − 1. Note that Ng for
RSSs should be designed such that Ng ≥ dR,max + L.
So the channel output samples for uth RSS and those
for vth DSS should be presented by
(u)
yR (n) =
L−1
hR (l)xR (n − l − dR,u )
L−1
hD (l)xD (n − l − dD,v )
(u)
(u)
(5)
(v)
(v)
(6)
l=0
(v)
yD (n) =
l=0
NR and ND are used to denote the number of RSSs and
DSSs accessing at the same time. Then the nth received
signal sample at the BS can be expressed as
y(n) =
N
R −1
u=0
(u)
yR (n) +
N
D −1
1447
ing the phase in time domain and frequency domain,
respectively.
We assume that NR is less than the number of initial
ranging codes Nc and the ranging code transmitted by
all RSSs are different.
3.1.
Cross-correlation in Time Domain
To detect the ranging codes and estimate the timing
offsets, one approach would to be cross-correlate the
received signal with all possible ranging codes in time
domain [12]. However, the CFO estimation scheme was
not addressed. In this section, a multi-user CFO estimation algorithm is proposed based on the results of code
detection and STO estimation in Reference [12]. It uses
the phase rotation of the cross-correlation between the
received signal and the time-domain reference ranging
signal. The cross-correlation is defined by
Ri (d) =
N
s −1
(i)
xR (n)y∗ (d + n)
n=0
(v)
yD (n) + z(n)
= Si + Ii + ID + Nz
(7)
(8)
v=0
where {z(n)} are independent and identically distributed (i.i.d.), circularly symmetric complex Gaussian noise samples with zero mean and variance σz2 .
In the following analysis, we will use the similar
method in References [5,15] to analyze the proposed
algorithms. Firstly, the channel is assumed to be nondispersive, and the received signals from different users
will only be affected by AWGN. Then, we will evaluate the performance of the proposed algorithms under
AWGN and time-dispersive channel, respectively. IR
has different symbol structure from PR, so we will first
introduce the CFO estimation methods for IR, and then
discuss their effectiveness for PR.
where
Si =
N
s −1
Ii =
N
R −1
(i)
(u)
xR (n) yR (d + n)
∗
u=0
u=i
N
s −1
N
D −1
N
s −1
(i)
(v)
xR (n) yD (n + d)
∗
n=0
ID =
v=0
Nz =
N
s −1
(i)
(i)
xR (n) yR (d + n)
n=0
n=0
(i)
xR (n)z∗ (n + d)
∗
(9)
n=0
3. CFO Estimation for OFDMA-based
Initial Ranging
In this section, two CFO estimation methods are proposed for WiMAX OFDMA-based initial ranging.
The correlation properties of PN codes are adopted
for extracting the desired ranging user and suppressing the interference signals of the other ranging
users and data users. Frequency offset will result in
phase rotation between two adjacent received samples
and at the same subcarrier between two successive
repeated symbols. CFO could be estimated by comparCopyright © 2009 John Wiley & Sons, Ltd.
In
the
above
equation,
Ns = N + N g ,
(i)
d = 1, 2, · · · , dmax,R , xR (n) is the ith time-domain
reference ranging signal, i = 1, 2, · · · , Nc , and (·)∗
represents conjugate. From Equation (9), Si is the
expected signal term, Ii is the interference term introduced by other ranging codes, ID is the interference
term introduced by DSSs’ signals, and Nz is the
Gaussian noise term. According to the correlation
property of PN codes, i.e., the cross-correlation of
different ranging codes is quite small compared with
auto-correlation, Ii is very low and approaching zero.
From Equations (2)–(4), since the random variables
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
1448
Y. LIN ET AL.
(i)
(v)
xR (n) and yD (n) are given by the sum of N i.i.d. ran(i)
(v)
dom variables with zero mean, both xR (n) and yD (n)
asymptotically become complex Gaussian with zero
mean for large N by the central limit theorem [20–22],
(i)
(v)
and the uncorrelated samples xR (n) and yD (n)
become independent Gaussian random variables, due
to the fact that uncorrelated Gaussian random variables
are statistically independent. Moreover, according to
the law of large numbers the mean of N i.i.d. random
variables with the same mean will tend to approach to
the expected value for large N. As a result, we have
the following approximation:
ID = Ns
Ns −1
1
(i)
(v)
xR (n) yD (n + d)
Ns
N
D −1
v=0
= Ns
v=0
φn = φ(0, n) − φ ξ (i) , d̂R,i + n
= −φ ξ (i) , d̂R,i + n
∗
n=0
N
D −1
(i)
E xR
(v)
· E ∗ yD
where f (i) is the CFO between the ith ranging user
and the uplink receiver. f is the subcarrier spacing,
and ξ (i) is defined as the normalized CFO (NCFO) of
the ith ranging user.
(i)
The phase rotation difference between xR (n) and
(i)
yR (n) is a function of the normalized CFO and their
time delay. After the correction of timing offset estimated using method in Reference [12], we assume there
are no other significant phase distortion effects. For the
ith ranging user, the phase rotation difference between
(i)
(i)
xR (n) and yR (d̂R,i + n) can be written as
(15)
(10)
≈0
Ns −1
1
(i)
xR (n)z∗ (n + d)
Ns
n=0
(i)
= Ns · E xR · E∗ (z)
N z = Ns ·
(11)
≈0
where E(·) represents expectation. Then Ri (d) can be
rewritten as
where d̂R,i is the estimated transmission delay (timing offset) of the ith RSS, n = 0, 1, · · · , Ns − 1. The
phase rotation difference can be used to determine the
normalized CFO ξ (i) .
The phase of Ri (d̂R,i ) (Equation (8)) represents
the sum of all phase rotation difference between the
(i)
time-domain reference ranging signal xR (n) and the
received samples y(d̂R,i + n). We have
∠Ri (d̂R,i ) = ∠
N
s −1
xR (n)y∗ (d̂R,i + n)
N
s −1
(i)
(i)
xR (n) yR (d̂R,i + n)
(i)
n=0
Ri (d) =
N
s −1
(i)
xR (n)y∗ (d + n)
≈∠
n=0
≈
N
s −1
n=0
n=0
(i)
∗
(i)
xR (n) yR (d + n)
(12)
(i)
(i)
N
s −1
(i)
2
xR (n) exp(φn )
n=0
For simplicity, we assume that the ranging users are
under AWGN channel. The received signal of the ith
ranging user can be rewritten as
yR (n) = xR (n − dR,i )ej2πξ
=∠
∗
(i) n/N
(13)
=∠
N
s −1
exp −φ ξ (i) , d̂R,i + n
=∠
N
s −1
exp
n=0
n=0
From Equation (13), the carrier frequency offset of the
ith ranging user, f (i) will result in a phase rotation
φ(f (i) , n)
= ∠ exp
−j2π(d̂R,i + n)ξ (i)
N
−jπηξ (i)
N
sin(πNs ξ (i) /N)
·
sin(πξ (i) /N)
(16)
φ f (i) , n = 2πf (i) n/Nf = 2πξ (i) n/N,
φ ξ (i) , n = φ(f (i) , n)
(14)
Copyright © 2009 John Wiley & Sons, Ltd.
where η = 2d̂R,i + Ns − 1. Then the normalized CFO
of the i-th ranging user, ξ (i) can be estimated by the
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
UPLINK CARRIER FREQUENCY OFFSET ESTIMATION
1449
Fig. 3. Block diagram of the proposed frequency offset estimator for initial ranging.
following equation:
ξ̂ (i) = −
can be rewritten as
∠Ri (d̂R,i ) · N
η·π
(17)
N
R −1
y(n + Ns ) =
(u)
yR (n + Ng )ej2πξ
(u)
u=0
+
3.2. Cross-correlation in Frequency Domain
(u)
(u)
(u)
(18)
where n = 0, · · · , N − 1, ξ (u) is the normalized CFO
of the uth ranging user. Then the received signal at BS
Copyright © 2009 John Wiley & Sons, Ltd.
(v)
yD (n + Ns ) + z(n + Ns )
v=0
Though several ranging methods based on crosscorrelation in frequency domain have been analyzed
(e.g., [13–15]), only code detection, timing offset estimation were discussed except CFO estimation. In
Reference [16], the frequency offset can be acquired
by correlating FFT output samples of two consecutive
OFDMA symbols at ranging subcarriers. However, this
method will be invalid in the case of multiple ranging users as it cannot distinguish between them. In
this section, we propose a novel CFO estimation algorithm as shown in Figure 3. The STOs of all ranging
users are first estimated by using the frequency-domain
cross-correlation method [16]. Then, the estimates of
all STOs are fed to CFO estimator to estimate the CFOs
of all ranging users. At last, the CFO of each ranging
user is estimated through multiplying the complex conjugate of the sum of the first FFT window by the sum of
the second FFT window. In initial ranging, the second
ranging symbol is created by repeating the first ranging symbol as shown in Figure 2(a). It is assumed that
the channel is constant during the transmission of the
two OFDMA symbols, which corresponds to ‘slowfading’ in the radio frequency channel. Without taking
into account the RSS’s transmission delay, the received
ranging signal has the following characteristics:
yR (n + Ns ) = yR (n + Ng )ej2πξ
N
D −1
(19)
After the CP removal and FFT processing, the
received signal y(n) in the ranging time-slot is converted into frequency-domain signal Ym (k), m = 0, 1.
The first OFDMA symbol will is given by
Y0 (k) = FFTN y(n + Ng )
=
N
R −1
u=0
+
(u)
FFTN yR (n + Ng )
N
D −1
v=0
(v)
FFTN yD (n + Ng )
+ FFTN z(n + Ng )
=
N
R −1
(u)
YR,0 (k) +
u=0
N
D −1
(v)
YD,0 (k) + Z0 (k)
v=0
(20)
Then the second OFDMA symbol is given as follows:
Y1 (k) = FFTN (y(n + Ns ))
=
N
R −1
u=0
(u)
(u)
ej2πξ FFTN yR (n + Ng )
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
1450
Y. LIN ET AL.
+
N
D −1
v=0
Y1′ (k) = e
(v)
FFTN yD (n + Ns )
=
e
j2πξ (u)
(u)
YR,0 (k) +
u=0
N
D −1
(i)
×e
(21)
where n
= 0, · · · , N − 1, k = 0, 1, · · · , N − 1, and
−j2πnk/N for any function f (n).
FFTN ≡ N−1
n=0 f (n)e
(u)
YR,m (k) is the received frequency-domain ranging signal in the mth OFDMA symbol from the uth ranging
(v)
user, and YD,m (k) denotes the received frequencydomain signal in the mth OFDMA symbol from the
vth data user, and m = 0, 1.
Assume dR,u is the timing offset between the
received signal from uth user and uplink receiver, and
Ỹ is the signal without timing offset to local receiver.
According to a timing offset in time domain corresponds to phase offset in frequency domain, Equations
(20) and (21) could be further described with the consideration of timing offset as
Y0 (k) =
e
N
D −1
Y1 (k) =
(u)
ỸR,0 (k)
(v)
YD,0 (k) + Z0 (k)
(u)
(u)
(22)
−j2πkdR,u
N
(u)
ỸR,0 (k)
j2πkd̂R,i
N
(v)
YD,1 (k) + e
j2πkd̂R,i
N
Z1 (k)
(25)
where d̂R,i is the estimated timing offset of the ith
RSS using the frequency-domain correlation method
in Reference [16].
To extract the signal of the ith user in frequency domain, we use the local reference ranging
(i)
(i)
(i)
(i)
code XR = [XR (0), XR (1), · · · , XR (N − 1)]T for
(i)
(i)
frequency-domain correlation (because XR,0 = XR,1 ,
(i)
let XR represent the ith RSS’s ranging signal in frequency domain). The results of the correlation have
the following vector expression:
T
(i)
= ỸR,0
+
ej2πξ e
ej2πξ
v=0
v=0
N
R −1
e
T (i)
E(d̂R,i )Y0 XR
u=0
+
j2πk(d̂R,i −dR,u )
N
N
D −1
+
−j2πkdR,u
N
N
R −1
(i)
u=0
u=i
(v)
YD,1 (k) + Z1 (k)
v=0
N
R −1
Y1 (k)
= ej2πξ ỸR,0 (k) +
+ FFTN (z(n + Ns ))
N
R −1
j2πkd̂R,i
N
(u)
(i)
XR +
N
R −1
u=0
u=i
N
D −1
(v)
E(d̂R,i )YD,0
v=0
ỸR,0 (k)
(u)
E(d̂R,i − dR,u )ỸR,0
T
N
D −1
(v)
YD,1 (k) + Z1 (k)
(26)
(23)
v=0
To elimate the impact of timing offset on the i-th user,
we will process Equations (22) and (23) as
Y0′ (k) = e
=
j2πkd̂R,i
N
(i)
ỸR,0 (k) +
+
N
D −1
e
j2πkd̂R,i
N
N
R −1
ej2πξ
u=0
u=i
e
j2πk(d̂R,i −dR,u )
N
+
(u)
ỸR,0 (k)
N
D −1
v=0
u=0
u=i
+
T (i)
E(d̂R,i )Y1 XR
(i)
(i)
= ej2πξ ỸR,0
Y0 (k)
N
R −1
(i)
XR
T (i)
(i)
XR + E(d̂R,i )Z0 XR
u=0
+
T
(u)
T
(i)
XR
(u)
E(d̂R,i − dR,u )ỸR,0
(v)
E(d̂R,i )YD,1
T
T
(i)
XR
T (i)
(i)
XR + E(d̂R,i )Z1 XR
(27)
(v)
YD,0 (k) + e
j2πkd̂R,i
N
Z0 (k)
v=0
(24)
Copyright © 2009 John Wiley & Sons, Ltd.
j2πd̂R,i
j2π(N−1)d̂R,i
N
)
where
E(d̂R,i ) = diag(1, e N , · · · , e
is a diagonal matrix to do timing correction for
the ith RSS in frequency domain, Ym = [Ym (0),
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
UPLINK CARRIER FREQUENCY OFFSET ESTIMATION
(u)
1451
Ym (1), · · · , Ym (N − 1)]T , ỸR,m = [ỸR,m (0), ỸR,m (1),
4.1.
(u)
(v)
(v)
(v)
· · · , ỸR,m (N − 1)]T , YD,m = [YD,m (0), YD,m (1), · · · ,
(v)
YD,m (N − 1)]T , Zm = [Zm (0), Zm (1), · · · , Zm (N −
1)]T , and m = 0, 1.
Because the proposed cross-correlation method in time
domain requires only one OFDM symbol, it could be
used for both initial ranging and periodic ranging without any adjustments.
(u)
(u)
In Equations (26) and (27), by considering the crosscorrelation property of PN codes in the second term, by
(i)
noting that XR (k) = 0 at data subcarriers in the third
term, and also by ignoring the impact of Gaussian noise
with zero mean in the fourth term, we can therefore
remove the last three terms on the right hand side of
the two equations. Then, the estimated ξ (i) could be
derived by
ξ̂ (i) =
arg
T (i)
T (i)
E(d̂R,i )Y1 XR / E(d̂R,i )Y0 XR
4.2.
Cross-correlation in Time Domain
Cross-correlation in Frequency Domain
In periodic ranging, not only the ranging codes transmitted in successive OFDM symbols are different
(i)
(i)
(i)
(XR,0 = XR,1 = XR,2 ), but also CP is adopted rather
than cyclic postfix.
After the CP removal and FFT processing of the
received samples y(n), the received frequency-domain
signal is as follows:
2π
N+Ng −1
(28)
N
R −1
Ym (k) =
n=Ng
Two CFO estimation methods for OFDMA-based
initial ranging have been analyzed in this session. To
sum up, the cross-correlation process in frequency
domain could provide two advantages over the crosscorrelation process in time domain. One is that, there
will be no interference from synchronized DSS’s
signal to the ranging channel, because they are orthogonal in frequency domain. The other one is, this
method makes better use of the auto-correlation/crosscorrelation properties of PN codes which are originally
sent over frequency domain [15].
4. Algorithm Analysis for Periodic
Ranging
In periodic ranging, the SS can send a transmission in
one of the following ways [2]:
(a) Modulating one ranging code on the ranging subchannel for a period of one OFDMA symbol.
(b) Modulating three consecutive ranging codes on the
ranging subchannel for a period of three OFDMA
symbols (one code per symbol).
To perform the frequency offset estimation in frequency domain, we need more than one consecutive
symbols in algorithms. So the system will select (b) for
transmission, as shown in Figure 2(b). The algorithm
analysis in Section 3 is based on the symbol structure of
initial ranging. It is necessary to verify its effectiveness
in periodic ranging with different symbol structure.
Copyright © 2009 John Wiley & Sons, Ltd.
+
(u)
yR (n + mNs )
u=0
N
D −1
(v)
yD (n + mNs ) + z(n + mNs )
v=0
× e−j2πnk/N
=
N
R −1
(u)
YR,m (k) +
u=0
N
D −1
(v)
YD,m (k) + Zm (k)
v=0
(29)
(u)
(v)
where m = 0, 1, 2, YR,m (k) and YD,m (k) are the mth
frequency-domain OFDM signal of the uth RSS and
the vth DSS, respectively.
Considering the timing offset, Equation (29) could
be further described as
Ym (k) =
N
R −1
e
−j2πkdR,u
N
(u)
ỸR,m (k)
u=0
+
N
D −1
e
−j2πkdD,v
N
(v)
ỸD,m (k) + Zm (k)
(30)
v=0
where Ỹ is the signal without timing offset to local
receiver.
From Equation (30), the timing-offset correction of
the ith ranging user is given by
Ym′ (k) = e
=
j2πkd̂R,i
N
Ym (k)
(i)
ỸR,m (k) +
N
R −1
e
j2πk(d̂R,i −dR,u )
N
(u)
ỸR,m (k)
u=0
u=i
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
1452
Y. LIN ET AL.
+
N
D −1
e
j2πk(d̂R,i −dD,v )
N
(v)
ỸD,m (k) + e
j2πkd̂R,i
N
+ ejθ0 e
Zm (k)
j2π(mNs +Ng )ξ (i)
N
N−1
(i)
(i)
(l)cl−k ,
(l)HR,m
XR,m
l=0
l=k
v=0
(31)
To extract the signal of the ith user in frequency domain, we use the local reference rang(i)
ing code XR,m for frequency-domain correla(i)
(i)
(i)
(i)
tion. XR,m = [XR,m (0) XR,m (1) · · · XR,m (N − 1)]T
(i)
and XR,m (k) = 0 at data subcarriers. The vector
expression of frequency-domain correlation for periodic ranging is given by
(i)
[E(d̂R,i )Ym ]T XR,m
T (i)
(i)
= ỸR,m XR,m
+
N
R −1
E
u=0
u=i
+
N
D −1
v=0
N−1
(i)
XR,m
(i)
1 1 − ej2πξ
=
j2πξ (i)
N
1−e N
(v) T (i)
E(d̂R,i − dD,v )ỸD,m XR,m
=e
(32)
(i)
(n)e−j2πkn/N
ỹR,m
n=0
=
N−1
n=0
× ejθ0 e
=
N−1
j2πnl
1 (i)
(i)
(l)e N
X (l)HR,m
N l=0 R,m
j2π(n+mNs +Ng )ξ (i)
N
e
−j2πnk
N
N−1
1 jθ0 j2π(mNs +Ng )ξ(i) (i)
(i)
N
(l)
XR,m (l)HR,m
e e
N
l=0
×
N−1
e
j2πn(l−k+ξ (i) )
N
n=0
=
N−1
1 jθ0 j2π(mNs +Ng )ξ(i) (i)
(i)
N
(l)cl−k
XR,m (l)HR,m
e e
N
l=0
= ejθ0 e
j2π(mNs +Ng )ξ (i)
N
N−1
1 j2πnξ(i)
e N
N
n=0
(u) T
d̂R,i − dR,u ỸR,m
ỸR,m is the signal without timing offset to local
receiver. Following Reference [23], its element could
be written as
(i)
(k) =
ỸR,m
where θ0 is the phase rotation between the phase of
the receiver local oscillator and the carrier phase at
the start of the received signal, ξ (i) is the ith rang(i)
ing user’s normalized CFO, and HR,m (l) denotes the
channel frequency response on the lth subcarrier of
the ith ranging user during the qth OFDMA block.
The complex weighting coefficients, c0 and cl−k are
given by
c0 =
T (i)
+ E d̂R,i Zm XR,m
(i)
(33)
(i)
(i)
(k)c0
(k)HR,m
XR,m
Copyright © 2009 John Wiley & Sons, Ltd.
cl−k =
jπ(N−1)ξ (i)
N
sin πξ (i)
(i)
N · sin πξN
N−1
1 j2πn(l−k+ξ(i) )
N
e
N
n=0
(i)
1 1 − ej2π(l−k+ξ )
j2π(l−k+ξ (i) )
N
N
1−e
(i)
(i)
jπ(N−1)(l−k+ξ ) sin π(l − k + ξ )
N
=e
(i) )
N · sin π(l−k+ξ
N
=
(34)
Note that the coefficients have the following periodical property, cl−k = cN+l−k . The first term in Equation
(33) represents an attenuated and rotated version of the
wanted ranging signal. The second term corresponds
to the inter-carrier interference (ICI) resulting from
carrier frequency offset.
From Equation (34), the frequency shifting will result
in a change in amplitude and phase of the wanted
ranging signal given by c0 , which depends on the normalized frequency offset ξ (i) but is independent of k. In
other words, all subcarriers experience the same degree
of attenuation and rotation of the wanted ranging signal.
Statistically, the ICI term in Equation (33) can be modeled as Gaussian noise since it is the result of an addition
of N − 1 random variables (Central Limit Theorem),
where N is generally large [20,21,24]. So Equation (33)
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
UPLINK CARRIER FREQUENCY OFFSET ESTIMATION
1453
can be rewritten as
In CFO estimation for periodic ranging, different
N−1
ranging codes for the two consecutive symbols will
be used when correlation in frequency domain.
In CFO estimation for periodic ranging, to calculate the normalized CFO ξ (i) , the factor 1/2π(1 + g)
should be used instead of 1/2π.
(i)
ỸR,m (k) =
(i)
ỹR,m (n)e−j2πkn/N
n=0
= ejθ0 e
j2π(mNs +Ng )ξ (i)
N
(i)
(i)
XR,m (k)HR,m (k)c0
(i)
+ VR,m (k)
(35)
(i)
where VR,q (k) is the additional noise modeled by the
ICI disturbance on the kth subcarrier.
Similar to the analysis of Equations (26) and (27),
we can ignore the effect of the other ranging users,
data users and noise, so the CFO of the ith ranging
user could be estimated by
1
2π(1 + g)
T (i)
(i)
ỸR,0
E d̂R,i Y1 XR,1
· arg
T (i) · (i)
E d̂R,i Y0 XR,0
ỸR,1
ξ̂ (i) =
So, this proposed CFO estimation method with crosscorrelation in frequency domain could be used in both
initial ranging and periodic ranging scenario.
5. Effect of Residual Timing offset
After the imprecise timing correction due to the STO
estimation error, the residual timing offset will cause
an impact on CFO estimation. Here, we will provide a
rough estimation of this impact.
T
(i)
XR,0
T (i)
XR,1
1
2π(1 + g)
N−1
(i)
HR,0 (k)
E d̂R,i Y1 T X(i)
k=0
R,1
· arg
·
E d̂R,i Y0 T X(i) N−1
(i)
R,0
HR,1 (k)
=
5.1.
Cross-correlation in Time Domain
From Equation (13), after the timing correction with
STO estimation error, the received signal of the ith
ranging user is given by
(i)
(i)
yR (n + d̂R,i ) = xR (n + d̂R,i − dR,i )ej2πξ
(i)
= xR (n + dR,i )ej2πξ
(i) (n+d̂ )/N
R,i
(i) (n+d +d )/N
R,i
R,i
(38)
k=0
(36)
where g = Ng /N. We assume the channel is flat.
Moreover, the neighboring OFDMA symbols have the
same channel frequency response on the same subcar(i)
(i)
riers, i.e., HR,0 (k) = HR,1 (k), k = 0, 1, · · · , N − 1.
Then Equation (36) can be rewritten as
ξ̂ (i)
T (i)
E d̂R,i Y1 XR,1
1
· arg
=
T (i)
2π(1 + g)
E d̂R,i Y0 X
R,0
(37)
Comparing the method for initial ranging in Equation
(28) and that for periodic ranging in Equation (37),
there will be only minor differences in processing.
For periodic ranging channel, the CP between
the first and second OFDM symbol should be
removed.
Copyright © 2009 John Wiley & Sons, Ltd.
where dR,i = d̂R,i − dR,i is the STO estimation error.
Comparing Equations (38) and (13), when dR,i = 0,
the STO estimation error leads to the received signal
(i)
out of synchronization with the reference signal xR (n),
which impairs the ranging code’s orthogonality in time
domain. On the other hand, the residual timing offset introduces a new phase rotation 2πξ (i) dR,i /N.
Therefore, this method will cause some performance
degradation.
Figure 4 shows the auto-correlation of the timedomain ranging signal versus residual STO values for
STO = 10 and NCFO = 0.2. It can be seen that the
auto-correlation performance will degrade sharply with
only one sample’s timing error.
5.2.
Cross-correlation in Frequency Domain
With perfect timing correction, the absolute values of
the first items in Equations (26), (27), and (32) are
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
1454
Y. LIN ET AL.
Fig. 4. Auto-correlation performance of the time-domain
ranging signal with residual STO values.
Fig. 5. Normalized results of frequency-domain correlation
with residual STO values.
denoted by
residual STO values for STO = 10 and NCFO = 0.2.
It can be seen that the result of frequency-domain correlation will decrease quickly when the residual STO
is larger than 4.
P1 =
(i)
ỸR,0
T
(i)
XR,0
(39)
But if there is residual timing offset caused by STO
estimation error, P1 will become
P1′ =
=
(i) T (i)
E d̂R,i − dR,i ỸR,0 XR,0
(i)
E(dR,i )ỸR,0
T
(i)
XR,0
6.1. Simulation Parameters
(40)
For simplicity, we assume that the channel is flat
and has a constant gain of one. Subcarriers of the
ranging channel is contiguous starting from 0, i.e.,
k = 0, 1, · · · , R − 1, where R is the length of ranging
codes. Using Equation (35), the ratio of P1′ to P1 is
P1′
P1
c0 ·
=
R−1
k=0
e
R−1
n=0
=
j2πkdR,i
N
c0
| sin(πRdR,i /N)|
|R · sin(πdR,i /N)|
(41)
Equation (41) could be explained as, with the residual timing offset, the signal of the ith ranging user
will be decreased to P1′ /P1 of the original signal. For
example, when N = 1024, R = 144 and dR,i is 5,
P1′ /P1 = 0.3636. So, timing offset correction is very
important in CFO estimation. Figure 5 shows normalized results of frequency-domain correlation with
Copyright © 2009 John Wiley & Sons, Ltd.
6. Simulation Results
In the simulation, the OFDMA system parameters are
selected from Reference [2]. The uplink bandwidth
is 3 MHz, the subcarrier frequency spacing f is
3.28 kHz, N = 1024. We use QPSK format for DSS.
Within the 1024 subcarriers, there are 92 guard subcarriers on the left-side band and 91 on the right, and
1 DC subcarrier residing on index 512. The remaining
840 subcarriers are partitioned into Q = 35 subchannels. The ranging channel is composed of six adjacent
subchannels and spanning 144 subcarriers per OFDMA
symbol. The performances of proposed algorithms are
evaluated in both AWGN channel and time-dispersive
channel. For the time-dispersive channel, we use one of
the Stanford University Interim (SUI) channel models
defined by IEEE 802.16 working group for assessing
technologies for broadband fixed wireless applications
[25]. SUI-3 channel model with three paths is considered in our simulation. The number of sample-spaced
channel taps, L, is set to 4. Channels of different
users are generated independently. We consider a cell
radius of 5 km which gives the maximum transmission delay (round trip) dR,max ≈ 34 s = 114 samples.
Ng is set to 128 samples satisfiying the condition
dR,max < Ng − L. The timing requirement based on
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
UPLINK CARRIER FREQUENCY OFFSET ESTIMATION
Reference [2] is that all uplink OFDM symbols should
arrive at the BS within an accuracy of ±25% of the minimum guard-interval or better. In Reference [2], Ng can
be 1/4, 1/8, 1/16, or 1/32 of N, and hence, the timing offset should be within ±8 samples. So the dD,max equals
to eight samples in our case. Similarly, the frequency
offset should be within a tolerance of maximum 2% of
the subcarrier spacing.
We assume that Nc = 128 ranging codes are assigned
for initial ranging and the maximum number of RSSs
in one ranging time-slot, NR , is set to 15. Because
six subchannels are allocated to ranging subchannel,
the maximum number of DSSs accessed simultaneously, ND , is 29. To quantify the performance of the
CFO estimation, the normalized root mean square error
(NRMSE) of the estimates is used, and it is defined as
ENRMSE
NR
1
(u)
(u) 2
=
ξ̂ρ − ξρ
NR
1455
Fig. 6. The RMSE performance of the CFO estimation versus
SNR over AWGN channel. The number of RSSs is NR = 1,
2, 5, 15, respectively.
(42)
ρ=1 u=1
(u)
where is the total number of Monte Carlo tests, ξρ is
(u)
the normalized CFO of the uth ranging user, and ξ̂ρ is
(u)
the estimate of ξρ . The subscript (·)ρ denotes the index
of the Monte Carlo test. The NRMSE is computed
after averaging over all participating ranging users for
10 000 independent Monte Carlo tests.
In each simulation run, the timing offset for RSSs
and DSSs are generated randomly from the interval
[0, dR,max ] and [0, dD,max ], respectively. The normalized CFOs of RSSs and DSSs uniformly distribute in
the range of [−0.2, 0.2] and [−0.02, 0.02]. In order to
evaluate the effect of data users on the performance of
proposed algorithms, we consider the following conditions of 0 DSS, 15 DSSs, and 29 DSSs in one ranging
time-slot.
improves as SNR increases. But both methods’ performances degrade as the number of RSSs increases. The
powers of the expected ranging signal and interferences
increase as the SNR increases, which resulting in no
obvious performance improvements of the two methods. The NRMSE performances of the two methods
are still less than 2% especially in multiple RSSs condition, which satisfies the timing requirement defined
in Reference [1].
Figure 7 depicts the NRMSE performance of the frequency offset estimation as a function of SNR in ‘SUI3’ time-dispersive channel. The condition is the same
as that in Figure 6. From Figure 7, the time-domain
6.2. Performance of Frequency
Offset Estimation
Figure 6 shows the NRMSE performance of the frequency offset estimation as a function of SNR in
AWGN channel. We consider the condition of precise
timing and 0 DSS in one ranging time-slot. The number
of RSSs is NR = 1, 2, 5, 15,respectively. In fairness,
the SNRs of all RSSs are the same. From Figure
6, The proposed frequency-domain cross-correlation
(CC) method performs better than the time-domain
cross-correlation method. When the number of RSSs
is NR = 1, the proposed two methods have the similar performances and the NRMSE of each method
Copyright © 2009 John Wiley & Sons, Ltd.
Fig. 7. The RMSE performance of the CFO estimation versus
SNR over ‘SUI-3’ time-dispersive channel. The number of
RSSs is NR = 1, 2, 5, 15, respectively.
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
1456
Y. LIN ET AL.
Fig. 8. The RMSE performance of the CFO estimation versus number of RSSs over AWGN channel. SNR = 20 dB.
cross-correlation method can not meet the system
requirement and the performance of frequency-domain
cross-correlation method also declines slightly. For the
frequency-domain cross-correlation method, in order
to satisfy the system requirement that the normalized
CFO must be less than 2%, the number of RSSs in one
ranging time-slot can not be more than 5. The following
simulations will give the NRMSE performance curves
as the number of RSSs changes under different channel models, we can obtain the exact number of RSSs
supported by the two methods in one ranging time-slot.
Figure 8 shows the NRMSE performance of the frequency offset estimation versus the number of RSSs
over AWGN channel. The performances of proposed
methods are evaluated in the absence/presence of residual timing offset. We assume that all SSs have the same
SNR of 20 dB and the residual timing offsets are i.i.d.
for different RSSs in the range of [−8, 8].
From Figure 8, it can be seen that the frequencydomain cross-correlation method is more robust to
the DSSs’ interferences than the time-domain crosscorrelation method. When there is no residual timing
offsets, the NRMSEs of the two CFO estimation
methods are less than 2% even in multiple RSSs
and DSSs condition, such as NR = 15 and ND = 29.
But the method of frequency-domain cross-correlation
has smaller NRMSE than that of time-domain crosscorrelation. The key reason is that the PN code is
Copyright © 2009 John Wiley & Sons, Ltd.
transmitted in frequency domain originally and thus
its auto-correlation/cross-correlation properties will be
impaired in time domain.
It can also be seen from Figure 8 that the NRMSE performance of the time-domain cross-correlation method
degrades, and cannot satisfy the system requirement
when there are residual timing offsets induced by
the STO estimation error. However, the proposed
frequency-domain cross-correlation method still has a
much better performance and is robust to the residual
timing offset. For example, the NRMSE is less than 2%
when the number of RSSs is less than 5 in one ranging
time-slot.
Figure 9 depicts the NRMSE performance of the frequency offset estimation versus the number of RSSs
over ‘SUI-3’ time-dispersive channel. The effect from
residual STO is also evaluated at SNR = 20 dB. The
error floor in Figure 9 clearly shows the performance degradation in the time-dispersive channel as
compared with the corresponding curves in AWGN
channel. Moreover, the time-domain cross-correlation
method can not meet the system requirement. It
is mainly because the derived result for the timedomain cross-correlation method is only suitable for
a flat fading channel, it cannot directly apply to
the time-dispersive channel environment. Without the
need of this assumption, the frequency-domain crosscorrelation method gives a much better performance
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
UPLINK CARRIER FREQUENCY OFFSET ESTIMATION
1457
Fig. 9. The RMSE performance of the CFO estimation versus number of RSSs over ‘SUI-3’ time-dispersive channel. SNR =
20 dB.
under the time-dispersive channel. For the frequencydomain cross-correlation method, in order to satisfy
the system requirement that the normalized CFO must
be less than 2%, the number of RSSs in one ranging time-slot can not be more than 8 in the absence
of residual STO while 2 in the presence of residual
STO, respectively. Since the probability of 3 or more
RSSs choosing the same ranging slot is small [18], the
proposed frequency-domain cross-correlation method
is also suitable for time-dispersive channel.
7. Conclusion
We have presented two multi-user CFO estimation
algorithms for WiMAX OFDMA-based initial ranging and periodic ranging. The proposed algorithms
are based on the correlation properties of the PN
codes in time domain and frequency domain, respectively. The NRMSE performance is evaluated for
both AWGN channel and time-dispersive channel. The
simulation results show that the frequency domain
cross-correlation method is more robust to the interferences from the other ranging users and data users
and also performs better than the time-domain crosscorrelation method over multi-path fading channel and
in the presence of residual timing offset. Hence, the proCopyright © 2009 John Wiley & Sons, Ltd.
posed algorithms can be used in practical WiMAX BS
receivers and other OFDMA-based wireless systems.
References
1. WiMAX Forum. Fixed, nomadic, portable and mobile applications for 802.16-2004 and 802.16e WiMAX networks. White
Paper, November 2005.
2. IEEE P802.16Rev2/D5. Part 16: Air Interface for Broadband
Wireless Access Systems. June 2008.
3. Morelli M, Kuo C-CJ, Pun M-O. Synchronization techniques
for orthogonal frequency division multiple access (OFDMA): a
tutorial review. In Proceedings of the IEEE, vol. 95, no. 7, July
2007; 1394–1427.
4. Schmidl TM, Cox DC. Robust frequency and timing synchronization for OFDM. IEEE Transactions on Communications
1997; 45: 1613–1621.
5. van de Beek JJ, Sandell M, Borjesson PO. ML estimation of timing and frequency offset in OFDM systems. IEEE Transactions
on Signal Processing 1997; 45: 1800–1805.
6. Wang HM, Yin QY, Meng YK, et al. Adaptive joint estimation of symbol timing and carrier frequency offset for OFDM
systems. In Proceedings of IEEE International Conference on
Communications (ICC), June 2007; 3034–3039.
7. van de Beek JJ, Borjesson PO, Boucheret ML, et al. A time
and frequency synchronization scheme for multiuser OFDM.
IEEE JJournal on Selected Areas in Communications 1999; 17:
1900–1914.
8. Morelli M. Timing and frequency synchronization for the uplink
of an OFDMA system. IEEE Transactions on Communications
2004; 52: 296–306.
9. Barbarossa S, Pompili M, Giannakis GB. Channel-independent
synchronization of orthogonal frequency division multiple
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm
1458
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
Y. LIN ET AL.
access systems. IEEE JJournal on Selected Areas in Communications 2002; 20(2): 474–486.
Cao Z, Tureli U, Yao YD. Deterministic multiuser carrierfrequency offset estimation for Interleaved OFDMA uplink.
IEEE Transactions on Communications 2004; 52: 1585–1594.
Fan D, Cao Z. Carrier frequency offset estimation for interleaved OFDMA uplink based on subspace processing. Journal
Of Electronics (China) 2007; 24(4): 433–438.
Fu X, Minn H. Initial uplink synchronization and power control
(ranging process) for OFDMA systems. In Proceedings of IEEE
Global Telecommunications Conference (GLOBECOM) 2004;
6: 3999–4003.
Krinock J, Singh M, Paff M, et al. Comments on OFDMA
ranging scheme described in IEEE 802.16ab-01/01r1. IEEE
802.16abc- 01/24, August 2001.
Kim KN, Kim JH, Kim SC, Cho SJ. The scheme to improve the
performance of initial ranging symbol detection with common
ranging code for OFDMA systems. The Eighth International
Conference on Advanced Communication Technology, Korea,
vol. 1, February 2006; 183–188.
Mahmoud HA, Arslan H, Ozdemir MK. An efficient initial
ranging algorithm for WiMAX (802.16e) OFDMA. Computer
Communications 2009; 32(1): 159–168.
Zhou Y, Zhang ZY, Zhou XW. OFDMA initial ranging for
IEEE 802.16e based on time-domain and frequency-domain
approaches. In International Conference on Communication
Technology, Guilin, China, November 2006; 1–5.
Fu X, Li Y, Minn H. A new ranging method for OFDMA systems. IEEE Transactions on Wireless Communications 2007;
6(2): 659–669.
Zeng J, Minn H. A novel OFDMA ranging method exploiting
multiuser diversity. In Proceedings of IEEE Global Telecommunications Conference (GLOBECOM), November 2007;
1498–1502.
Zeng J, Minn H. An investigation into initial ranging method
for mobile OFDMA systems. In Proceedings of IEEE Sarnoff
Symposium, April 2008; 1–5.
Durrett R. Probability: Theory and Examples (3rd edn). Duxbury
Press, 2004.
Walpole RE, Myers RH, Myers SL, Ye K. Probability & Statistics for Engineers & Scientists. Prentice Hall, 2006.
Ochiai H, Imai H. Performance analysis of the deliberately
clipped OFDM signals. IEEE Transactions on Communications
2002; 50(1): 89–101.
Armstrong J. Analysis of new and existing methods of reducing intercar-rier interference due to carreier frequency offset in
OFDM. IEEE Transactions on Communications 1999; 47(3):
365–369.
Speth M, Fechtel SA, Fock G, Meyr H. Optimum receiver design
for wireless broadband systems using OFDM, Part I. IEEE
Transactions on Communications 1999; 47(11): 1668–1677.
Erceg V, Hari KVS, Smith MS, et al. Channel Models for fixed
wireless applications. Contribution IEEE 802.16a-03/01, June
2003.
Authors’ Biographies
Yonghua Lin is a Research Staff Member in System Software and Networking
Department at the IBM China Research
Lab. She received B.S. and M.S. degrees
in Information and Communication from
Xi’an Jiaotong University in 2000 and
2003, respectively. She subsequently
joined IBM at China Research Lab,
where she has worked on multiple
Copyright © 2009 John Wiley & Sons, Ltd.
projects related to multicore processors (general multicore
processor and network processors) and networking (high-end
router, IPTV media gateway, and mobile base station). She is
leading an IBM Research group for appliance and infrastructure for wireless access network in next generation. She is an
author or coauthor of 20 patents and 10 technical papers in
related domain. Ms. Lin is a member of Institute of Electrical
and Electronics Engineers and the Association for Computing
Machinery.
Da Fan received the B.S. and M.S.
in Communication Engineering from
PLA Information Engineering University, Zhengzhou, China, and Ph.D. in
Electronic Engineering from Tsinghua
University, Beijing, China, in 2000,
2003, and 2009, respectively. As an
intern, currently he is working at IBM
China Research Laboratory (CRL).
His research interests include wireless communications
and mobile networks, broadband multi-carrier/OFDM
techniques, dynamic resource allocation and management,
wireless MAN (WiMAX/802.16). He has published more
than 20 papers on Communication and Signal Processing
fields, and held two patents.
Qing Wang, obtained Ph.D. from
School of EEE , Nanyang Technological University, Singapore in 2003. Her
research field is Statistical Signal Processing, especially for sinusoidal signal
detection and estimation. She got M. E.
and B. E both from Northwestern Polytechnical University, China in 1999 and
1992. The majors are Control Engineering and Electronic Engineering respectively. She joined IBM
China Research Lab (CRL) in 2004. Before joined CRL,
she was an Electronic Engineer in an institute for 4 years
(1992–1996). In CRL, she works on wireless network cloud,
WiMAX basestation design, wireless communication system
on general IT platform, IMS system design, and IPTV system design. Her research interests include software defined
radio, wireless broadband access, array signal processing,
media accelerating on multicore system, and streaming data
processing. She is an IEEE member from 2003 and a member
of Communication Society.
Jianwen Chen received the B.S. in
information and electronics engineering
from Huazhong University of Science
and Technology, Wuhan, China in 2002.
He received the Ph.D. in electronics
engineering from Tsinghua University,
Beijing, China in 2007. His research
interests focused on video compression
algorithms and video processing systems, wireless transmission algorithms and systems. He
joined IBM China Research Center, Beijing, China in 2007.
His current major focus is on wireless video algorithms
and systems, programming model and parallel computation
framework design and so on.
Wirel. Commun. Mob. Comput. 2010; 10:1444–1458
DOI: 10.1002/wcm