IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 47, NO.

3, MARCH 1999 365

Analysis of New and Existing Methods of Reducing Intercarrier

Interference Due to Carrier Frequency Offset in OFDM
Jean Armstrong

Abstract—Orthogonal frequency division multiplexing (OFDM)

is very sensitive to frequency errors caused by frequency differ-
ences between transmitter and receiver local oscillators. In this
paper, this sensitivity is analyzed in terms of the complex weight-
ing coefficients which give the contribution of each transmitter
subcarrier to each demodulated subcarrier. Previously described
windowing and self-intercarrier interference (ICI) cancellation
methods are analyzed in terms of these weighting coefficients.
New ICI cancellation schemes with very much improved perfor-
mance are described. A condition for orthogonality of windowing
schemes is derived in terms of the discrete Fourier transform
(DFT) of the windowing function.
Index Terms— Discrete Fourier transform, intercarrier inter-
ference, interference suppression, orthogonal frequency division
multiplexing, synchronization.

I. INTRODUCTION

O RTHOGONAL frequency division multiplexing

(OFDM) is being considered for data transmission
in a number of environments [1], [2]. One limitation of
OFDM in many applications is that it is very sensitive to
frequency errors caused by frequency differences between the
local oscillators in the transmitter and the receiver [3]–[5]. Fig. 1. Structure of an OFDM communication system.
Carrier frequency offset causes a number of impairments
including attenuation and rotation of each of the subcarriers subcarriers and the symbol period is . In the th symbol
and intercarrier interference (ICI) between subcarriers [4]. period, the complex values modulate the
A number of methods have been developed to reduce this subcarriers.
sensitivity to frequency offset, including windowing of the Systems using a number of different types of modulation
transmitted signal [6], [7] and use of self ICI cancellation of subcarriers within OFDM, such as phase shift keying
schemes [8]. (PSK) and quadrature amplitude modulation (QAM) have been
This paper analyzes in detail, for a perfect Nyquist channel, described in the literature. This analysis does not depend on
the ICI resulting from carrier frequency offset. Expressions the mapping of the data to be transmitted to the complex
are derived for each demodulated subcarrier at the receiver in values , and is therefore applicable to all forms
terms of each transmitted subcarrier and complex weighting of modulation which can be used within OFDM.
factors. Windowing and ICI cancellation schemes can be This analysis considers only the impairments due to carrier
related and described in terms of these complex weighting frequency offset. Other authors [9] have analyzed more general
factors. New ICI cancellation schemes which give greater models, but these do not clearly reveal the structure on which
ICI cancellation are developed. A new condition for the ICI cancellation depends. Frequency offset alone does not
orthogonality of windowing schemes is derived in terms of the cause intersymbol interference (ISI). Often a cyclic prefix is
discrete Fourier transform (DFT) of the windowing function. used in OFDM to eliminate the ISI and ICI caused by errors
in sampling time or distortion in the channel. This use of a
II. ANALYSIS OF ICI
cyclic prefix is not considered in this analysis.
A. Structure of OFDM System
B. Derivation of Expressions for the
Fig. 1 shows the structure of the OFDM communication
Complex Weighting Coefficients
system being considered. In this OFDM system there are
The signal at the output of the OFDM transmitter resulting
Paper approved by S. B. Gelfand, the Editor for Transmission Systems
of the IEEE Communications Society. Manuscript received March 5, 1997;
from the th transmitted symbol is given by
revised February 16, 1998.
The author is with the School of Electronic Engineering, La Trobe Univer-
sity, Bundoora, VIC 3083, Australia (e-mail: [email protected]).
(1)
Publisher Item Identifier S 0090-6778(99)01925-X.

0090–6778/99$10.00  1999 IEEE

366 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 47, NO. 3, MARCH 1999

where is the carrier frequency and is the impulse

response of the low-pass filter in the transmitter. At the
receiver, the signal is mixed with a local oscillator signal which
is above the correct frequency . Ignoring the effects of
noise, the demodulated signal is then given by

(2)

where is the combined impulse response of the channel

and of the transmitter and receiver filters. is the phase
offset between the phase of the receiver local oscillator and Fig. 2. Real and imaginary components of the complex weighting factors
the carrier phase at the start of the received symbol. 1
c0 1 1 1 cN01 for the case of fT=02: and N = 16 .
Assuming that satisfies the Nyquist criterion for sam-
ples taken at intervals and that is sampled at the where
optimum instants, then the samples input to the receiver DFT
are given by

(3)

The result of the DFT of these samples is given by (8)

The decoded complex value therefore consists of a wanted

(4) component which is due to but which is subject to a
change in amplitude and phase given by , where

Substituting the value of from (3) and after some manip-

ulation, it can be shown that (9)

depends on the normalized frequency offset but is

independent of . In other words, all subcarriers experience
the same degree of attenuation and rotation of the wanted
component.
(5) In addition, the decoded complex value is subject to ICI.
This is the sum of components dependent on each of values
Using the properties of geometric series, this can alternatively . The contribution of each depends on the
be expressed as normalized frequency offset and on . It
does not depend directly on .
Fig. 2 shows the complex weighing factors
for the case of and . Note that
as the coefficients depend on the distance mod between the
subcarriers, there are only distinct coefficients, .
(6) The graphs are smooth, there are no sudden changes in the
weighting coefficients as the distance moves from 15 to 15,
If , then , and each decoded except between 1 and 0, and between 0 and 1.
complex value is simply the phase rotated version of the Fig. 3 shows how, for , the value of depends
transmitted value. The amount of rotation depends on the phase on the number of subcarriers in the OFDM system. For
offset between the transmitter and receiver local oscillators. there is little change in the power of the wanted subcarrier.
If , then ICI will occur and each output data symbol Because carrier frequency offset does not change the total
will depend on all of the input values. The analysis of ICI can power in the received signal, this also means that the total
be simplified by defining complex weighting coefficients, ICI power changes little with .
, which give the contribution of each of the input
values to the output value III. SELF ICI CANCELLATION SCHEMES
Zhao and Häggman [8] have described a method of reduc-
(7) ing sensitivity to frequency errors which they call self ICI
cancellation. The method maps the data to be transmitted onto
ARMSTRONG: ANALYSIS METHODS OF REDUCING INTERCARRIER INTERFERENCE IN OFDM 367

scheme gives no reduction in overall SNR compared with

normal OFDM. A disadvantage of the method is that it is
less bandwidth efficient than normal OFDM as only half as
many complex values can be transmitted per symbol.
The ICI cancellation scheme also results in a form of
windowing and the overall transmitted signal has a sinusoidal
envelope of period . This is because pairs of sinusoids of
frequency difference are being subtracted.

IV. WINDOWING

A. Windowing to Reduce Sensitivity to Linear Distortions

A number of authors [6], [7], [10], [11] have described
the use of windowing in OFDM. These applications can be
Fig. 3. c0 as function of number of subcarriers for 1 = 0 1.
fT : divided into two groups. In the first group, windowing is used
to reduce the sensitivity to linear distortions [10], [11]. In
adjacent pairs of subcarriers rather than onto single subcarriers, the second, windowing is used to reduce the sensitivity to
so that . frequency errors [6], [7]. In the first group, the signal at the
This results in cancellation of most of the ICI in the values output of the IDFT in the transmitter is cyclically extended.
. The windowing function shapes the cyclic extension, but the
For example, the decoded value for the zeroth carrier is original part of the signal remains unchanged. Where the only
given by impairments in the system are due to frequency differences
between the local oscillators, this form of windowing has no
effect on the system performance. This form of windowing
(10) will not be considered further.
The ICI now depends on the difference between the ad-
jacent weighting coefficients rather than on the coefficients B. Windowing to Reduce Sensitivity to Frequency Offset
themselves. As the difference between adjacent coefficients is
small, this results in substantial reduction in ICI. If adjacent The second form of windowing involves cyclically extend-
coefficients were equal, then the ICI would be completely ing by samples the time domain signal associated with each
cancelled. Thus this process can be considered as cancelling symbol. The whole of the resulting signal is then shaped
out the component of ICI which is constant between adjacent with the window function. Fig. 4(a) shows a block diagram
pairs of coefficients. ICI cancellation depends only on the of a typical system. Note that the transform in the receiver is
coefficients being slowly varying functions of offset. It does point whereas that in the transmitter is point. The
depend not on the absolute values of the coefficients and so inputs to the transmitter transform have been labeled
improves the performance for any frequency offset. so that windowing can be more readily
To maximize the overall SNR, the values analyzed and related to ICI cancellation. If , then
should be subtracted in pairs, because this results in the points of the received signal are used as input to the DFT, if
, then the signal corresponding to each symbol is zero
addition of the wanted signal components. This also further
padded at the receiver to give length . The outputs of the
reduces the ICI
DFT with even-numbered subscripts are used as estimates of
the transmitted data and the odd-numbered ones are discarded.
Because not all of the received signal power is being used in
generating data estimates, the method has a reduced overall
SNR compared with OFDM without windowing. The value of
(11)
the SNR loss depends on the form of windowing.
The remaining ICI depends on factors of the form A number of different windows, including the Hanning
. If the three weighting coefficients in each factor window [6], windows satisfying the Nyquist criterion [7], and
were linear functions of offset, for example, if and the Kaiser window [7] have been described in the literature.
, where is any constant as well as the gradient All of these windows give some reduction in the sensitivity
of the linear function, these factors would all be zero. This to frequency offset. But only Nyquist windows (of which the
cancellation could be considered as canceling out the compo- Hanning window is one particular example) have no ICI for
nent of ICI which is due to the linear variation in weighting the case of no frequency offset [7].
coefficient over groups of three adjacent coefficients.
When there is no carrier frequency offset C. Windows Which Preserve Orthogonality
(12) The conditions under which a window preserves orthog-
Thus, in the absence of other impairments, all of the received onality can be derived by considering the block diagram
power is decoded into wanted signal, and the ICI cancellation of Fig. 4(b). In the absence of noise and distortion, this is
368 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 47, NO. 3, MARCH 1999

For the Hanning window, which is a cosine roll-off window

with roll-off factor of o,

(14)

This window results in linear cancellation of the ICI in

the even-numbered outputs, , but each of
the odd numbered outputs depend on two of the inputs
. Therefore, these odd-numbered outputs
are discarded and are not used to contribute to the useful signal.
This windowing has the same performance with respect to ICI
as the self ICI cancellation scheme with linear cancellation
but has worse performance with respect to noise added in the
channel.

V. NEW HIGHER ORDER ICI CANCELLATION SCHEMES

The concept of self ICI cancellation can be extended. In the
(a) method of Zhao and Häggman [8], data is mapped onto pairs
of subcarriers. This results in cancellation of the component of
ICI due to the linear variation in weighting coefficients over
groups of three adjacent coefficients. By mapping data onto
larger groups of subcarriers, higher order ICI cancellation can
be achieved. For the general case of mapping onto groups
of subcarriers, the relative weightings of the subcarriers
in the group are given by the coefficients of the polynomial
expansion of . Using a similar analysis to that for the
(b)
linear case, it can be shown that when groups of subcarriers
Fig. 4. (a) Windowing to reduce sensitivity to frequency offset. (b) Simpli- are weighted in this way in both the transmitter and receiver,
fied block diagram of windowing.
the component of ICI which is due to the variation in weighting
coefficients described by a polynomial of order over
equivalent to Fig. 4(a). The combination of cyclic extension, groups of adjacent coefficients is cancelled.
windowing, and zero padding is equivalent to a cyclic exten- For example, by using groups of three subcarriers, cancel-
sion of length and windowing of length , where some lation of the component of ICI which is due to the cubic
of the window coefficients may be zero. The point IDFT variation in weighting coefficients over groups of five adjacent
followed by the cyclic extension can be further simplified coefficients can be achieved. In this case,
to an equivalent point IDFT in which all the inputs with At the receiver, the data
odd-numbered subscripts are zero. is estimated from weighted sums of the form
For no distortion and orthogonality of the wanted outputs,
that is those outputs with even subscripts, then for If the low-pass filter in the transmitter and the receiver are
even. For odd, can take any value, and may depend designed so that the receiver filter is matched to the transmitted
on any of the inputs as these values are discarded. filter, this form of weighting also results in overall matched
The windowing is a multiplication of the output of the IDFT filtering where the filtering is matched to the data which is
output values with the windowing values. A well-known being mapped onto the .
property of the DFT is that multiplication in the discrete time
domain is equivalent to circular convolution in the discrete VI. COMPARISON OF PERFORMANCE OF DIFFERENT
frequency domain and vice versa. Let be the METHODS OF REDUCING ICI DUE TO FREQUENCY OFFSET
point DFT of . Then is the circular
convolution of with . Using the Fig. 5 shows how the ratio of mean wanted power to mean
fact that for odd it can readily be shown that the uncancelled ICI power varies as a function of normalized
orthogonality condition is met if frequency offset for four different systems: standard OFDM
and OFDM with cancellation of the constant, linear, and cubic
components of ICI. The graphs are for , but graphs
for any would have almost identical form.
When ICI cancellation schemes are used, weighted groups
any value. (13) of subcarriers are modulated rather than individual subcarriers.
ARMSTRONG: ANALYSIS METHODS OF REDUCING INTERCARRIER INTERFERENCE IN OFDM 369

cancel higher order components. These have been shown

to give very great reductions in the ICI due to frequency
offset. These ICI cancellation methods map each complex
value to be transmitted onto weighted groups of subcarriers. A
disadvantage of these ICI methods is that fewer complex data
values are transmitted per symbol period. A condition for the
orthogonality of windowing schemes in terms of the DFT of
the windowing function has been derived.

ACKNOWLEDGMENT
The author wishes to thank the anonymous reviewers and
Dr. J. M. Badcock, Prof. P. M. Grant, Dr. K. A. Seaton, and
Fig. 5. Wanted signal power/uncancelled ICI power as a function of Prof. E. R. Smith for their helpful comments and suggestions.
=Deltaf T .
REFERENCES
The graphs describe the performance for any modulation [1] B. Hirosaki, S. Hasegawa, and A. Sabato, “Advanced group-band
scheme which meets the condition that the data being mapped modems using orthogonally multiplexed QAM technique,” IEEE Trans.
onto the weighted groups are independent identically dis- Commun., vol. 34, pp. 587–592, June 1986.
[2] M. Alard and R. Lassalle, “Principles of modulation and channel coding
tributed random variables and when is integral so that for digital broadcasting for mobile receivers,” EBU Tech. Rev., no. 256,
all subcarriers are used. They do not describe exactly the pp. 168–190, 1987.
case where different weighted groups of subcarriers are being [3] H. Sari, G. Karam, and I. Jeanclaude, “Channel equalization and carrier
synchronization in OFDM systems,” presented at 1993 Tirrenia Int.
modulated with different average powers, for example where Workshop on Digital Communications, Tirrenia, Italy, Sept. 1993.
spectral shaping is used, or where there is correlation between [4] T. Pollet, M. Van Bladel, and M. Moeneclaey, “BER sensitivity of
the variables modulating different groups of subcarriers. For OFDM systems to carrier frequency offset and Wiener phase noise,”
IEEE Trans. Commun., vol. 43, pp. 191–193, Feb./Mar./Apr. 1995.
not integral, the ICI is less for subcarriers close to the [5] P. H. Moose, “A technique for orthogonal frequency division multi-
unused subcarriers. The graphs shows the worst case ICI. All plexing frequency offset correction,” IEEE Trans. Commun., vol. 42,
pp. 2908–2914, Oct. 1994.
of the methods give a higher signal-to-uncancelled-ICI ratio [6] M. Gudmundson and P.-O. Anderson, “Adjacent channel interference in
than standard OFDM. an OFDM system,” in IEEE 46th Vehicular Technology Conf., Atlanta,
GA, Apr. 1996, pp. 918–922.
[7] C. Muschallik, “Improving an OFDM reception using an adaptive
VII. CONCLUSION Nyquist windowing,” IEEE Trans. Consumer Electron., vol. 42, Aug.
1996.
An analysis of the effect of frequency errors in OFDM has [8] Y. Zhao and S.-G. Häggman, “Sensitivity to Doppler shift and carrier
been presented. The ICI due to carrier frequency offset has frequency errors in OFDM systems—The consequences and solutions,”
in IEEE 46th Vehicular Technology Conf., Atlanta, GA, Apr. 1996, pp.
been described in terms of complex weighting coefficients. 1564–1568.
The self ICI cancellation schemes and windowing schemes [9] M. Russell and G. L. Stuber, “Interchannel interference analysis of
described by other authors have been analyzed in terms of the OFDM in a mobile environment,” in IEEE 45th Vehicular Technology
Conf., July 1995, 1996, pp. 820–824.
complex weighting factors. It is shown that the ICI cancellation [10] S. B. Weinstein and P. M. Ebert, “Data transmission by frequency-
scheme and cosine roll-off windowing, with a roll-off factor division multiplexing using the discrete Fourier transform,” IEEE Trans.
of one, cancel the component of ICI due to the linear variation Commun. Technol., vol. COM-19, pp. 628–634, Oct. 1971.
[11] L. J. Cimini, “Analysis and simulation of a digital mobile channel using
of weighting coefficients over groups of three coefficients. orthogonal frequency division multiplexing,” IEEE Trans. Commun.,
New self ICI cancellation schemes have been derived which vol. COM-33, pp. 665–675, July 1985.