Computer Communications 31 (2008) 4130–4135

Contents lists available at ScienceDirect

Computer Communications
journal homepage: www.elsevier.com/locate/comcom

Inter-carrier interference power of OFDM in a uniform scattering channel

Khoa N. Le *
Centre for Wireless Monitoring and Applications, Griffith School of Engineering, Griffith University, Gold Coast Campus, Qld 4222, Australia

a r t i c l e i n f o a b s t r a c t

Article history: This letter derives the upper and lower bounds on inter-carrier interference (ICI) power PICI of OFDM in a
Received 24 October 2007 uniform scattering channel. The bounds are computed as functions of fd T s product of the maximum
Received in revised form 21 August 2008 Doppler spread fd and symbol duration T s ; f frequency tracking and  mobile travelling direction. Insight-
Accepted 22 August 2008
ful discussions on the characteristics of P ICI are given. Future work is also outlined.
Available online 5 September 2008
Ó 2008 Elsevier B.V. All rights reserved.

Keywords:
Inter-carrier interference power
OFDM
Bounds
Doppler spread
Frequency tracking

1. Introduction model (BEM) in fast-varying time channels with satisfactory re-

Inter-carrier interference power (P ICI ) in OFDM wireless com-
munication systems is undesirable and should be reduced to the
minimum to improve transmission efficiency. It was shown in [1]
that the theoretical expression of PICI is difficult to interpret. Moti-
vated by this fact, the universal bound was derived in [1] which
specifies the upper limit of PICI for any Doppler spectrum Pðf Þ. From
a different view point, Ng and Dubey investigated the effectiveness
of directional and non-directional antennas in wireless OFDM sys-
tems to reduce P ICI using different scattering distributions. The sys-
tem comprises of a base station (BS) receiving and transmitting
signals from and to a mobile user equipment (UE) with random
scatterers between them. The scatterers form a scattering channel
which is characterised by their distribution pðhÞ. The P ICI as a func-
tion of the scattering channel between the BS and UE was also de-
rived. In addition, to compensate for Doppler spread and the
mobile travelling direction  and frequency tracking f was
employed.
OFDM and ICI have recently attracted attention from research-
ers world wide. A new algorithm for ICI reduction using ICI as a
diversity source and Weiner filter was proposed in [2] from which
the detection and computation times were improved. A two-stage
hybrid channel estimation and ICI cancellation structure for OFDM
systems were studied in [3] in which encouraging results were ob-
tained. A channel estimation and ICI cancellation algorithm with
low complexity were proposed in [4] based on a basis expansion
* Present address: Griffith School of Engineering, Griffith University, Parklands
Drive, Southport, Qld 4222, Australia. Tel.: +617 5552 9175; fax: +617 5552 8065.
E-mail address: K.Le@griffith.edu.au.
sults. Most pulses reported in the literature can either reduce the
out-of-band emission and ICI but not both. An optimal pulse to
effectively reduce both the out-of-band emission and ICI was pro-
posed in [5] with encouraging results. On this note, a new pulse
shape was proposed in [6] with additional analyses on bit error
rate (BER) of the corresponding OFDM system. Detailed compari-
sons with other pulses were also given. Inter symbol interference
(ISI) and ICI reduction algorithms for cooperative space–frequency
block-coded OFDM (SFBC-OFDM) networks employing amplify-
and-forward (AF) and decode-and-forward (DF) schemes were
shown in [7] in which new algorithms were practically tested
and better results were obtained. ICI was shown not only to cause
performance degradation but also phase noise in OFDM systems
[8]. Seo et al. [9] proposed a novel ICI reduction algorithm for ubiq-
uitous environments. Their findings showed that using the new
algorithm, BER and spectral efficiency of OFDM systems can be im-
proved. An extended H filter-based ICI estimator was proposed in
[10] from which the BER of OFDM systems can be significantly im-
proved under unknown-noise conditions. An effective algorithm to
combat ICI was proposed in [11] by employing linear filters and
null sub-carriers. The algorithm was shown to improve OFDM reli-
ability and more robust against Doppler spread. A two-step estima-
tor in estimating channel common phase errors in the first step and
removing ICI in the second step using a power series expansion of
the phase noise process for coded and uncoded OFDM systems was
reported in [12] with satisfactory results. Two methods for ICI
reduction were studied in [13]: (i) using an extended Kalman filter
(EKF), and (ii) employing the sequential Monte Carlo (SMC) called
sequential importance sampling (SIS). The two methods were
thoroughly assessed for different values of frequency offsets and

0140-3664/$ - see front matter Ó 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.comcom.2008.08.022
 K.N. Le / Computer Communications 31 (2008) 4130–4135 4131

signal-to-noise ratios. A two-stage precoder/equalizer was pro- The upper and lower bounds on P ICI with respect to Pðf Þ given in
posed in [14] to effectively combat ICI from which it was found Eqs. (3)–(5) were derived in [1] and are given here as
that an improve of about 10% on the BER could be achieved. A a1 ð2pfd T s Þ2 a2 ð2pfd T s Þ4 a1 ð2pfd T s Þ2
space–frequency joint processing technique with low complexity  6 PICI 6 ; ð6Þ
12 360 12
was shown in [15] to suppress ICI and other interference in 2D.
The ICI power as a function of the correlation of phase noise, chan- where a1 and a2 are dependent on Pðf Þ. For the classical model:
nel correlation and frequency offset was derived in [16] using cor- a1 ¼ 1=2 and a2 ¼ 3=8, for the uniform model: a1 ¼ 1=3 and
relative coding. Loose bounds on the ICI power were briefly a2 ¼ 1=5, and for the two-path model: a1 ¼ a2 ¼ 1.
discussed. A new ICI-self-cancellation-scheme for OFDM systems Since a1 6 1, the universal bound on PICI was also derived [1]
was assessed in [17] in which improved performance was noted. and is given in Eq. (7) as
An ICI cancellation method based on ICI matrix properties for ð2pfd T s Þ2 p2 ðfd T s Þ2
UWB-OFDM systems was discussed in [18]. Improved performance PICI 6 ¼ : ð7Þ
12 3
was obtained for indoor environments and normalised frequency
shifts. ICI reduction mechanism was considered as a combinatorial It should be noted that the universal bound given in Eq. (7) is
optimization problem in [19] in which two relaxation methods applicable for any Pðf Þ. The main aim of this paper is to derive
were proposed with encouraging results. the P ICI bounds by evaluating 1  rðT s xÞ as a function of pðhÞ which
Even though the theoretical expression of P ICI in terms of fd T s , f is characterised by the scattering channel between the BS and the
and  was given, it is difficult to interpret and to visualise the UE.
behaviour of the PICI . In fact, the final compact analytical expres-
sion of P ICI cannot be obtained which motivated further work on 3. Bounds based on pðhÞ
deriving bounds on PICI . It should be noted that the bounds re-
ported in [1] as functions of fd T s were derived based on Doppler From [21], the correlation function of the Jakes/Clarke 2D model
spectrum Pðf Þ. The bounds derived in this letter as functions of with an AoA pdf pðhÞ is given as [20]
fd T s and f are based on the scattering channel using its angle-of-ar- Z þp
1
rival (AoA) probability density function (pdf) pðhÞ. Since fd T s and f rðT s xÞ ¼ pðhÞ exp½j2pfd T s x cosðhÞdh; ð8Þ
2p p
are important for the BS and the UE to compensate for Doppler
spread respectively, it is necessary to examine bounds on P ICI as where pðhÞ ¼ A ¼ 1=2bm ¼ 6=p [20] and bm is the antenna beam
functions of fd T s and f, thus yielding more insight on PICI . width.
The letter is organised as follows. Section 2 gives background on To incorporate the effects of the UE travelling direction  and
the universal bound derived in [1]. Section 3 derives the upper and the frequency tracking f to compensate Doppler spread in the
lower bounds on PICI for a uniform scattering channel. Special cases channel, Eq. (8) can be rewritten as [20]
when f ¼ 0 are also given. Section 4 further gives insight on the Z þp
1
new bounds and compares them with the universal bound. Expres- rðT s xÞ ¼ pðhÞ expfj2pfd T s x½cosðh  Þ  fgdh: ð9Þ
2p p
sions of fd T s and f to obtain a zero lower bound and to also lower
the upper bound are derived. Section 5 concludes the main findings From Eq. (9), we obtain
of the letter and also outlines possible further work. Z þp
1
1  rðT s xÞ ¼ pðhÞh1  expfj2pfd T s x½cosðh  Þ  fgidh
2p p
2. Bounds based on Pðf Þ Z þp
1
¼ pðhÞh1  cosf2pfd T s x½cosðh  Þ  fgidh:
The PICI of an OFDM system with an infinite number of sub-car- 2p p
riers is defined as [1,20]
ð10Þ
Z þ1 Z þ1
PICI ¼ ð1  jxjÞ½1  rðT s xÞdx ¼ 1  ð1  jxjÞrðT s xÞdx; ð1Þ In addition, we have
1 1
u2 u4
where rðT s xÞ is the correlation function of the transmitted signal cosðuÞ ¼ 1  þ  ; ð11Þ
2! 4!
xðtÞ.
To determine the upper and lower bounds on P ICI , the limits of thus
1  rðT s xÞ as given in Eq. (1) are required. In terms of Pðf Þ, u2 u4 u2
1  rðT s xÞ can be given as  6 1  cosðuÞ 6 ; ð12Þ
2! 4! 2!
Z fd
1  rðT s xÞ ¼ Pðf Þ½1  expðj2pfT s xÞdf which are the bounds on 1  cosðuÞ. It should be noted that the
fd bounds given in Eq. (12) are only approximate as the most domi-
Z fd nant terms in the infinite series of cosðuÞ are used. Thus, even
¼2 Pðf Þ½1  cosð2pfT s xÞdf : ð2Þ though tight bounds on PICI were given in [1], there still exists some
0
small degree of inaccuracy in the approximation which is also ap-
There are three main types of Doppler spectrum Pðf Þ: (1) classi- plied to the bounds derived in this letter. From Eq. (1), the PICI as
cal, (2) uniform, and (3) two-path, which are given as follows: a function of fd T s ,  and f is given as
8 Z
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 1
< if jf j < fd ;
P1 ðf Þ ¼ p fd 2
1ðf =fd Þ
ð3Þ PICI ¼ ð1  jxjÞ
: 1
Z þp 
0 otherwise;
( 1
1  pðhÞh1  cosf2pfd T s x½cosðh  Þ  fgidh dx:
2f d
if jf j < fd ; 2p  p
P 2 ðf Þ ¼ ð4Þ
0 otherwise; ð13Þ
1 From Eqs. (10) and (12), substituting u ¼ 2pfd T s x½cosðh  Þ  f,
P 3 ðf Þ ¼ ½dðf þ fd Þ þ dðf  fd Þ: ð5Þ
2 the upper and lower bounds on PICI can be derived as follows.
4132 K.N. Le / Computer Communications 31 (2008) 4130–4135

3.1. Upper bound (UB) values of A obtained by using a1 and a2 are slightly different which
indicates that there does not exist a unique value for A to exactly
The UB on P ICI is given in Eq. (14) as match Pðf Þ and pðhÞ, but only approximate. To assess the effective-
Z ( Z þp ) ness of these values, bounds corresponding to A ¼ 2=3 and
1
1 ð2pfd T s xÞ2 ½cosðh  Þ  f2 A ¼ 8=15 for the uniform model; A ¼ 2 and A ¼ 8=3 are plotted
PICI 6 ð1  jxjÞ pðhÞ dh dx
1 2p p 2! in Figs. 1–4, respectively. From Figs. 1 and 2, for the uniform model,
(Z ) Z þp  even though the approximate solution is out-of-bound, both tight
1
ð1  jxjÞð2pfd T s xÞ2 1
¼ dx pðhÞ½cosðh  Þ  f2 dh bounds are smaller than the universal bound which suggests that
1 2! 2p p
A ¼ 2=3 and A ¼ 8=15 can be used for the matching with the upper
 ( 2
) !
A ð2pfd T s Þ ð2pfd T s Þ2 ð1 þ 2f2 Þ tight bound being replaced by the universal bound. The slightly
¼ f1 þ 2f2 g ¼ ða1 Þ ; ð14Þ out-of-bound approximate solution is mainly caused by integra-
2 12 12
tion errors because the integral with respect to h cannot theoreti-
where a1 ¼ A=2. cally be evaluated. From Fig. 3, for the two-path model, it
appears that A ¼ 2 is suitable for the matching because the approx-
3.2. Lower bound (LB) imate solution is out-of-bound, and both upper and lower tight
bounds are less than the universal bound. In fact, Fig. 3 is very sim-
The LB on PICI is given in Eq. (15) as ilar to Fig. 5. On the other hand, from Fig. 4, it can be suggested that
(Z ) Z  A ¼ 8=3 cannot be suitably used for the matching as the approxi-
1
ð2pfd T s xÞ2 1 þp
PICI P ð1  jxjÞ dx pðhÞ½cosðh  Þ  f2 dh mate solution is out-of-bound and both tight bounds are larger
1 2! 2p p
(Z ) Z 
1
ð2pfd T s xÞ4 1 þp
 ð1  jxjÞ dx pðhÞ½cosðh  Þ  f4 dh
1 4! 2p p Uniform model matching with A = 2/3
! ! 0
ð2pfd T s Þ2 ð1 þ 2f2 Þ 2p4 ðfd T s Þ4
¼ ða1 Þ 
12 45
! ! −10
2
Apð3=4 þ 6f þ 2f Þ 4
ð2pfd T s Þ2 ð1 þ 2f2 Þ
 ¼ ða1 Þ Tight upper bound being replaced
2p 12
! −20 by universal bound
  4 2 4
3A ð2pfd T s Þ ð3=4 þ 6f þ 2f Þ

8 360
dB

! ! −30
2
ð2pfd T s Þ ð1 þ 2f2 Þ ð2pfd T s Þ4 ð3=4 þ 6f2 þ 2f4 Þ
¼ ða1 Þ  ða2 Þ ;
12 360
−40
Universal Bound
ð15Þ
Lower Bound
where a2 ¼ 3A=8. Upper Bound
−50 Approximate Solution

3.3. Exact solution

−60
0 0.1 0.2 0.3 0.4 0.5
The exact solution of the PICI is given by Eq. (16) as
Z  Z  fT
1 þp d s
A
PICI ¼ ð1  jxjÞ 1  cos½2pfd T s xðcosðh  Þ  fÞdh dx;
1 2p p Fig. 1. Uniform model matching with A ¼ 2=3 and f ¼ 0.
ð16Þ
whose exact analytical expressions cannot be obtained for general
values of  and f. Thus, approximations are employed to simulate Uniform model matching with A = 8/15
the exact P ICI given by Eq. (16). Since the exact expression for P ICI 0
cannot be theoretically obtained, this shows the necessity of bounds
on P ICI which provide insight on its characteristics and on how to −10
effectively control it.
Tight upper bound being replaced
−20
3.4. Matching Pðf Þ and pðhÞ by universal bound

From Eqs. (14) and (15) with f set to 0, using the values of a1 −30
and a2 worked out in [1], it is possible to estimate A so that pðhÞ
dB

can be correspondingly matched with the predefined Pðf Þ models. −40

(1) Classical model: a1 ¼ 1=2 ¼ A=2 ) A ¼ 1 and a2 ¼ 3=8 ¼

−50
3A=8 ) A ¼ 1. Universal Bound
(2) Uniform model: a1 ¼ 1=3 ¼ A=2 ) A ¼ 2=3 and a2 ¼ 1=5 ¼ Lower Bound
−60 Upper Bound
3A=8 ) A ¼ 8=15. Approximate Solution
(3) Two-path model: a1 ¼ 1 ¼ A=2 ) A ¼ 2 and a2 ¼ 1 ¼
3A=8 ) A ¼ 8=3. −70
0 0.1 0.2 0.3 0.4 0.5
fT
d s
It is clear that for the classical model, A ¼ 1 is consistently obtained
by using both a1 and a2 . For the uniform and two-path models, the Fig. 2. Uniform model matching with A ¼ 8=15 and f ¼ 0.
 K.N. Le / Computer Communications 31 (2008) 4130–4135 4133

Two−path model matching with A = 2 PICI bound as a function of fdTs with ζ = 0

0 0

−10 −10

Universal Bound
Approximate solution is slightly out−of−bound Lower Bound
−20 −20 Upper Bound
Approximate Solution

dB −30
dB

−30

−40 −40

Universal Bound
Lower Bound
−50 Upper Bound −50
Approximate Solution

−60 −60
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
fdTs fT
d s

Fig. 5. Approximate P ICI and bounds as functions of fd T s with f ¼ 0 and A ¼ 6=p.

Fig. 3. Two-path model matching with A ¼ 2 and f ¼ 0.

   
Two−path model matching with A = 8/3 3 ð2pfd T s Þ2 360 ð2pfd T s Þ4
10 PICI P  ; ð18Þ
p 12 301:6 360

0 corresponding to a2  1:19 which is approximately the two-path

model.
−10 To assess the effectiveness of the bounds on PICI given in Sec-
tions 3.1 and 3.2 and to compare them with the universal bound
Approximate solution is slightly out−of−bound
−20 derived in [1], Eqs. (17) and (18) are plotted in Fig. 5 with f ¼ 0
from which it is clear that the bounds are consistent with the uni-
dB

−30 versal bound given in Eq. (7). The approximate solution to Eq.
(16) is also consistent with these bounds which validates Eqs.
(14) and (15). It should also be noted that the PICI upper bound
−40 Universal Bound is always present in OFDM systems as long as fd T s is different
Lower Bound
Upper Bound from zero, i.e. the system is in operation. Thus, it is not possible
−50
Approximate Solution to completely eradicate PICI in OFDM systems. However, the PICI
lower bound can theoretically be reduced to zero under specific
−60 conditions. From that, the UB can be re-evaluated to see whether
0 0.1 0.2 0.3 0.4 0.5
fT it is lower than the theoretical UB. It should be noted that if the
d s
UB can be lowered, then the P ICI can be more effectively con-
Fig. 4. Two-path model matching with A ¼ 8=3 and f ¼ 0. trolled than in the case when the LB is lowered. From Eq. (15),
the lower bound can be eradicated if the following condition is
met:
than the universal bound. It should also be noted that because the
tight bounds in this letter are derived as functions of f and fd T s 1 þ 2f2 p2 ðfd T s Þ2 ð3=4 þ 6f2 þ 2f4 Þ
with the universal bound as a special case with f ¼ 0, thus, it is ¼ ;
2 15
not surprising that some values of A can result in tight bounds
which are not only out-of-bound with respect to the approximate yielding
solution but also larger than the universal bound as can be clearly sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
seen in Figs. 1–4. 15ð1 þ 2f2 Þ
fd T s ¼ : ð19Þ
2p2 ð3=4 þ 6f2 þ 2f4 Þ
4. Discussion
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0:16 59:22ðfd T s Þ2 þ 15 þ 225  1; 184:35ðfd T s Þ2 þ 2; 922:27ðfd T s Þ4
For A ¼ 6=p and f ¼ 0, the upper bound given in Eq. (14) f :
becomes fd T s
  ð20Þ
3 ð2pfd T s Þ2
PICI 6 ; ð17Þ The frequency tracking parameter f as a function of fd T s can be sim-
p 12
ilarly estimated as given in Eq. (20)Using Eqs. (19) and (20), the UB
can be re-evaluated as follows:
which is consistently less than the universal bound given in Eq. (7).
The approximate value of a1 from Eq. (17) is 0:95 which approxi-  2
5 1 þ 2f2
mately corresponds to the two-path model. Under the same condi- UBðLB¼0Þ ðfÞ ¼ ; ð21Þ
2ð0:75 þ 6f2 þ 2f4 Þ
tion, the lower bound given in Eq. (15) becomes
4134
and
2 in Eq. (7) is a special case of Eq. (14) with f ¼ 0. It should be noted
UBðLB¼0Þ ðfd T s Þ ¼ 2:52  6:68ðfd T s Þ þ 0:0168
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 22; 500  118; 435ðfd T s Þ2 þ 292; 227ðfd T s Þ4 :
To assess the effectiveness of the new conditions given in Eqs.
(19) and (20), Eqs. (21) and (22) are plotted against f and fd T s to-
gether with the theoretical UB given in Eq. (14) with A ¼ 1 in
Fig. 6 from which it is clear that the UB can be approximately low-
ered by maximum 2 dB for f P 0:856 and by maximum 1 dB for
fd T s P 0:68. For other values of A such as 6=p, 8=15, 2=3, 2 and
8=3, the above ranges of f and fd T s remain almost unchanged. Thus,
it can be suggested that for f P 0:856 or fd T s P 0:68, the UB on the
PICI can be reduced which improves the efficiency of OFDM sys-
tems. It should also be noted that the condition f P 0:856 means
that the mobile UE has to employ substantial frequency correction;
fd T s P 0:68 means that the system has to endure larger Doppler
spread or employ faster sampling intervals than the typical value
of about 0.5. Thus, the above conditions can be only be achieved gi-
ven that the system has the ability to operate under extreme con-
ditions which is the trade-off of UB reduction on PICI .
Further observations and insight are as follows. From Eq. (20),
to ensure that f P 0 then 0 6 fd T s 6 1:0066. There are two points
which can be noted: (i) there are no conditions imposed on Eq.
(19) which makes it more convenient than Eq. (20); and (ii) Eq.
(19) can be used at the BS and Eq. (20) can be employed at the
UE to compensate for Doppler spread and also to lower the P ICI .
Generally, to completely eradicate the lower bound on PICI , Eqs.
(19) and (20) can be used to estimate fd T s using a chosen value
of f or vice versa and independent of pðhÞ, which is useful for OFDM
system design.
Another point to note is that the bounds on PICI are independent
of . This means that the travelling direction of the UE does not
change the system’s characteristics as expected. However, the
instantaneous values of P ICI are dependent on fd T s ,  and f as can
0
Normalised UB (dB)
−0.5
−1
−1.5
−2
0.8
0
Normalised UB (dB)
−1
−2
−3
−4
0.65
Fig. 6. Magnified UBs on P ICI as functions of f and fd T s when A ¼ 1.
K.N. Le / Computer Communications 31 (2008) 4130–4135
be seen in Eq. (16). It is also evident that the universal bound given
that the universal bound is applicable for any Doppler spectrum
Pðf Þ, but not for any AoA pdf pðhÞ. From Eq. (14), it is clear that
PICI increases quadratically with f which is typically less than 1,
ð22Þ yielding a slower increasing rate in the P ICI . For extreme cases
where f P 1, the P ICI rapidly increases which suggests that f should
be kept less than unity to effectively control the PICI .
It should also be clear that the bounds on P ICI are dependent on
pðhÞ which is characterised by the scattering channel between the
BS and the UE. The advantage of this is that scattering channels can
be assessed by using their scatterer distributions such as Gaussian
[22,23] and hyperbolic [24]. The disadvantage of this is that the
upper universal bound on P ICI independent of pðhÞ cannot be de-
rived. However, if pðhÞ is uniform, the bounds given in Eqs. (14)
and (15) can be easily estimated. For non-uniform pðhÞ, the estima-
tion of bounds on PICI is difficult and only approximate expressions
can be obtained.
5. Conclusion
Bounds on P ICI have been derived in this letter based on the
scattering channel’s AoA pdf pðhÞ as functions of fd T s and the fre-
quency tracking parameter f. For A ¼ 6=p and f ¼ 0, the uniform
scattering channel can approximately be modelled as the two-path
Doppler spectrum. Matching between the Doppler spectrum and
scattering channel’s AoA pdf has been examined in which it was
shown that exact matching can be achieved when A ¼ 1. For the
uniform and two-path models, only approximate values of A can
be used. It has been shown that the universal bound derived based
on Doppler spectrum Pðf Þ is a special case of the bound formulae
given in this letter. The upper bound on P ICI is quadratically propor-
tional to f which implies that f should be kept less than unity to
effectively reduce P ICI . Further work can also be carried out by
examining the effects of non-uniform scattering channels such as
Gaussian and hyperbolic.
Magnified UB as a function of ζ when LB = 0, A = 1
Modified UB(LB = 0)
Theoretical UB
0.85 0.9 0.95 1
ζ
Magnified UB as a function of fdTs when LB = 0, A = 1
Modified UB(LB = 0)
Theoretical UB
0.7 0.75 0.8 0.85 0.9 0.95 1
fdTs
 K.N. Le / Computer Communications 31 (2008) 4130–4135
Acknowledgement
The author thank the anonymous referees for useful comments
which improved the paper.
References
[1] Y. Li, L.J. Cimini, Bounds on the interchannel interference of OFDM in time-
varying impairments, IEEE Transactions on Communications 49 (3) (2001)
401–404.
[2] H. Song, J. Kim, S. Nam, et al., Joint doppler-frequency diversity for OFDM systems
using hybrid interference cancellation in time-varying multipath fading
channels, IEEE Transactions on Vehicular Technology 57 (2008) 635–641.
[3] Q. Jiang, J. Speidel, C.M. Zhao, A joint OFDM channel estimation and ICI
cancellation for double selective channels, Wireless Personal Communications
45 (1) (2008) 131–143.
[4] L.K. Zou, Q. Chang, C.D. Xiu, et al., Channel estimation and ICI cancellation for
OFDM systems in fast time-varying environments, IEICE Transactions on
Communications E91B (4) (2008) 1203–1206.
[5] W. Qin, Q.C. Peng, On the design of optimum pulse in OFDM systems, IEICE
Electronic Express 5 (8) (2008) 260–264.
[6] V. Kumbasar, O. Kucur, ICI reduction in OFDM systems by using improved SINC
power pulse, Digital Signal Processing 17 (6) (2007) 997–1006.
[7] D. Sreedhar, A. Chockalingam, Interference mitigation in cooperative SFBC-
OFDM, EURASIP Journal on Advances in Signal Processing 2008 (125735)
(2008) 25735.
[8] F. Septier, Y. Delignon, A. Menhaj-Rivenq, et al., Non-pilot-aided sequential
monte carlo method to joint signal, phase noise, and frequency offset
estimation in multicarrier systems, EURASIP Journal on Advances in Signal
Processing 2008 (612929) (2008) 12929.
[9] J.W. Seo, S.S. Yeo, K.S. Park, et al., A robust and secure time-domain ICI
canceller for OFDM based ubiquitous systems in time-varying multipath
channels, Journal of Supercomputing 45 (1) (2008) 29–43.
[10] Y. Jing, F.L. Yin, Z. Chen, An H-infinity filter based approach to combat inter-
carrier interference for OFDM systems, IEEE Communications Letters 12 (6)
(2008) 453–455.
4135
[11] Y. Long, L.L. Kuang, J.H. Lu, An ICI mitigation method for high speed mobile
OFDM, IEICE Transactions on Communications E91B (5) (2008) 1661–1664.
[12] F. Munier, T. Eriksson, A. Svensson, An ICI reduction scheme for OFDM system
with phase noise over fading channels, IEEE Transactions on Communications
56 (7) (2008) 1119–1126.
[13] A.P. Palamides, A.M. Maras, A Bayesian state–space approach to combat inter-
carrier interference in OFDM systems, IEEE Signal Processing Letters 14 (2007)
677–679.
[14] Y. Fu, W.A. Krzymien, C. Tellambura, A technique for multiuser and intercarrier
interference reduction in multiple-antenna multiuser OFDM downlink, IEEE
Transactions on Wireless Communications 6 (2007) 3493–3497.
[15] M. Huang, Y.Z. Li, S.D. Zhou, et al., Joint space–frequency iterative interference
suppression in multiuser MIMO-OFDM systems, IEEE Communications Letters
11 (2007) 802–804.
[16] W. Bai, Y. Kim, Effects of correlative coding on OFDM systems for intercarrier–
interference suppression, IEEE Transactions on Vehicular Technology 56
(2007) 3732–3738.
[17] Y.H. Peng, Y.C. Kuo, G.R. Lee, et al., Performance analysis of a new ICI-self-
cancellation-scheme in OFDM systems, IEEE Transactions on Consumer
Electronics 53 (2007) 333–1338.
[18] L.N. Qi, Z.L. Gan, H.B. Zhu, ICI cancellation in UWB-OFDM systems based on ICI
matrix properties, Journal of Applied Sciences 53 (2007) 570–576.
[19] Y.H. Zhang, W.S. Lu, T.A. Gulliver, Integer QP relaxation-based algorithms for
intercarrier–interference reduction in OFDM systems, Canadian Journal of
Electrical and Computer Engineering 32 (2007) 199–205.
[20] W.T. Ng, V.K. Dubey, Effect of employing directional antennas on mobile OFDM
system with time-varying channel, IEEE Communications Letters 7 (4) (2003)
165–167.
[21] W.C. Jakes, Microwave Mobile Communications, Wiley, New York, 1974.
[22] R. Janaswamy, Angle and time of arrival statistics for the Gaussian scatter
density model, IEEE Transactions on Wireless Communications 1 (3) (2002)
488–497.
[23] D.D.N. Bevan, V.T. Ermolayev, A.G. Flaksman, I.M. Averin, Gaussian channel
model for mobile multipath environment, EURASIP Journal on Applied Signal
Processing 2004 (9) (2004) 1321–1329.
[24] K.N. Le, A new formula for the angle-of-arrival probability density function in
mobile environment, Signal Processing 87 (6) (2007) 1314–1325.