I.
INTRODUCTION
Doppler Mitigation in
OFDM-Based Aeronautical
Communications
M. CENK ERTÜRK
University of Turkish Aeronautical Association and
University of South Florida
JAMAL HAQUE
WILFRIDO A. MORENO
HÜSEYIN ARSLAN, Senior Member, IEEE
University of South Florida
Doppler spectrum in an aeronautical channel is modeled with
dual-Doppler shift instead of classical Jakes model. Orthogonal
frequency division multiplexing (OFDM)-based systems are sensitive
to Doppler shifts/spread, since the time variation of the channel
causes inter-carrier interference (ICI). ICI analysis is provided here
for OFDM-based systems in the aeronautical channel. The effect of
ICI on the received signal is presented and its power is derived. As
compared with terrestrial channels, where ICI is generally overcome
by increasing the subcarrier spacing and bounding the normalized
Doppler frequency (NDF), we propose to mitigate the effect of
Doppler shifts in aeronautical channels, as follows. First, we use
parametric spectrum estimation methods to extract the Doppler
shifts by exploiting the predictable number of paths. Then, a
beamforming-based method is introduced to resolve the incoming
rays by separating them individually. Finally, paths are
Doppler-compensated and combined using diversity combining
techniques. Computer simulations are performed to provide
numerical results. It is shown that a mean square error (MSE)
performance of 1% is achieved with the parametric estimation
methods, and bit error rate (BER) performance approaching the
no-Doppler scenario is obtained with the beamforming-based
mitigation method.
Manuscript received August 4, 2011; revised March 3, 2012, August 23,
2012, December 14, 2012, January 17, 2013; released for publication
May 1, 2013.
IEEE Log No. T-AES/50/1/944776.
DOI. No. 10.1109/TAES.2013.110473.
Refereeing of this contribution was handled by M. Rice.
Author’s addresses: M. C. Ertürk, Dept. of Electrical and Electronics
Eng., University of Turkish Aeronautical Association, 10 Okul St.,
Etimesgut, Ankara, Turkey 06790. E-mail: (mcerturk@thk.edu.tr).
J. Haque, W. A. Moreno, H. Arslan, Dept. of Electrical Engineering,
University of South Florida, 4202 E. Fowler Ave., Tampa, FL 33620.
E-mail: (jhaque@mail.usf.edu).
0018-9251/14/$26.00 
C 2014 IEEE
120 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 50, NO. 1 JANUARY 2014
Aeronautical communications (AC) is an emerging
area in which aeronautical platforms are considered as a
part of the multi-tier network for future wireless
communication systems. Programs led by the National
Aeronautics and Space Administration (NASA), the
Federal Aviation Administration (FAA), and
EUROCONTROL all include the aeronautical platforms
as part of their multi-tier network [1, 2]. The driving
reasons for development of high data rate AC systems are:
1) the increase in data demand for air traffic control and
air traffic management due to the growth in air
transportation [3], 2) the need for low latency and low cost
services to provide in-flight multimedia access [4], and 3)
the potential to use AC systems as a backbone for
terrestrial communication networks [5]. AC systems can
provide service for ground networks, public safety,
military communications, and improved cockpit data
communications. To date most ground/aircraft cockpit
communications are done through voice only, which is
prone to language differences, accents, stresses, and
cultural barriers [6]. High data rate AC systems can
augment the cockpit verbal communication with video
and text to reduce cockpit communication errors.
Furthermore, there is a growing demand for high
speed data to meet commercial in-flight Internet
activities [7].
Orthogonal frequency division multiplexing
(OFDM)-based schemes have been adopted for several
current communication systems globally [8]. In an
OFDM-based system, a serial symbol stream is converted
into a parallel stream and each symbol is modulated with
different orthogonal subcarriers. Orthogonal subcarriers
and cyclic prefix (CP) usage provide robustness to
OFDM-based systems against the frequency selectivity of
wireless channel. However, OFDM-based systems have
relatively longer symbol durations compared with
single-carrier systems. Longer symbol duration leads to
weakness against the time variation of the channel, i.e.,
Doppler spread, which causes loss of orthogonality
between the subcarriers. If the orthogonality is not
preserved within an OFDM symbol duration, then there
will be inter-carrier interference (ICI).
ICI degrades channel estimation and symbol detection
performances of OFDM-based systems [9, 10]. If not
compensated ICI will cause an error floor for the symbol
detection. For example, in the terrestrial OFDM systems,
the channel model is generally based on the assumption
that the direction of arrival (DOA) of the signals at the
receiver are uniformly distributed, which yields to a
Doppler spectrum of the classical Jakes model. The
estimation of the channel and the compensation of the
channel effect on the received signal are computationally
complex in the Jakes Doppler spectrum scenario.
Therefore, ICI is generally overcome by increasing the
subcarrier spacing (decreasing the length of the OFDM
symbol) and bounding the normalized Doppler frequency
(NDF)1 which causes an error floor for symbol detection
in terrestrial communications [11].
As compared with the Jakes Doppler spectrum in
terrestrial communications, the aeronautical channel is
modeled with dual-Doppler shift [12]. The result of the
dual-Doppler shift is also ICI in an OFDM-based AC
system. However, in an aeronautical channel, as the
received signal has a dual path with corresponding
Doppler shifts, the Doppler spectrum can be interpreted as
a combination of two frequency offsets with
corresponding gains. Therefore, it is possible to estimate
and mitigate the effect of Doppler shifts by separating and
compensating the shifts individually.
In the literature OFDM channel estimation and ICI
compensation for the dual-Doppler shift are investigated
in [13–16]. In [13] a Kalman filter-based estimation
method with zero-forcing equalization is provided to
cancel the effect of ICI. In [14] a digital phase lock loop is
proposed to be used in order to track parameters of line of
sight (LOS) path with a maximum likelihood estimator
suggested to resolve the reflected path. The authors
propose a Kalman-based approach to provide more
accurate estimation, and to utilize an iterative cancellation
method for the ICI compensation. In [15] Doppler shift
compensation is suggested only for the LOS path, and
demodulation is performed in the presence of ICI. A
different version of OFDM, non-contiguous orthogonal
signal division multiplex (NCOSDM) is considered in
[16], where the number of subcarriers is changed
depending on the channel, to decrease the ICI and
maintain the system performance.
In this study an OFDM-based system2 is considered in
aeronautical environment. For the dual-Doppler shift
scenario, the effect of ICI on the received signal is
provided and its power is derived. Parametric spectrum
estimation methods are used to extract Doppler shifts by
exploiting the predictable number of paths, i.e., two [12]
or three [18]. The channel modeling order error and its
impact on the estimation performance is investigated. A
beamforming-based approach is proposed to separate the
paths based on DOA. Once the signals are separated then
conventional methods are used to compensate the Doppler
shift’s individually [19]. Estimation performance for
different modeling order errors, types of parametric
spectrum estimation methods, and Rice factors are
investigated through simulations. It is shown that an
average mean square error (MSE) performance of 1% is
achieved with the parametric estimation methods. Based
on various estimation errors and diversity combining
techniques, the impact of using a different number of
1 Note that this method also decreases the efficiency of OFDM-based
systems and is discussed in Section II-B.
2 We focus on multi-carrier system, i.e., OFDM in this study. However,
the methods investigated in this paper to estimate and resolve the
aeronautical channel effect on received signal can also be applied to
single-carrier systems. The reader is referred to [17] for investigation of
interference mitigation schemes in single-carrier systems.
ERTÜRK ET AL.: DOPPLER MITIGATION IN OFDM-BASED AERONAUTICAL COMMUNICATIONS
antenna elements on bit error rate (BER) performance is
investigated. We show that the proposed method can
achieve BER performance approaching the no-Doppler
scenario.
The remainder of this paper is organized as follows. In
Section II aeronautical channel is introduced and ICI
analysis for OFDM-based system is provided. Parametric
spectrum estimation techniques are introduced for the
estimation of Doppler shifts in Section III. In Section IV a
beamforming-based signal separation methodology is
used to compensate the Doppler shifts. Numerical results
for both estimation and compensation are presented in
Section V, followed by concluding remarks in Section VI.
II. AERONAUTICAL CHANNEL
The aeronautical environment poses numerous
challenges for developing a high data rate AC system, one
of which is the aeronautical channel [20]. Aeronautical
channels can be broken into three segments:
takeoff/landing, en route, and taxiing/parked. In this paper
we focus on the en route channel, which is generally
modeled as a two-ray channel model [12]. The remaining
two cases fall under the non-line of sight (NLOS)
dispersive Jakes models.
A. Doppler Spectrum in Aeronautical Channel
The en route channel presents different conditions
between air to ground and ground to air wireless links. For
each channel condition a progressive increase and
decrease of multipath and received Doppler spread of
arriving signals needs to be evaluated, as the aeronautical
platform moves from a flat surface area to a mountainous
area. In extremely rough environments the en route
channel might experience an intermittent loss of LOS
signal with increasing Doppler spread which will cause
channel Doppler spectrum closely following the classical
Jakes model [21]. However, the aeronautical channel is
generally modeled with a two-ray model, which consists
of narrow sparsely populated Doppler shifts [22–24].
Fig. 1 presents the two-ray Doppler spectrum in an
aeronautical channel. Both the direct and the reflected
paths have a narrow spread and random Doppler
frequencies. Therefore it can be assumed that two Doppler
Fig. 1. Doppler power spectrum in aeronautical channel.
121
shifts can uniformly span anywhere between −fd,max to The exact ICI power (PICI ) as a function of the
fd,max [12]. As a result Doppler spectral density for Doppler spectral density is given as [11]
aeronautical channel can be given as  fd,max

fd
PICI = 1 − pfd (fd )sinc2 dfd . (5)

L−1
−fd,max f
pfd (fd ) = (hl ) δ(fd − fDl )
2
(1)
l=0 Using (1) in (5) the ICI power in two-ray aeronautical
channel can be given as
where hl , fDl are the path gains and Doppler frequencies,  
respectively (L = 2). PICI = 1 − |h0 |2 sinc2 (0 ) + |h1 |2 sinc2 (1 ) . (6)
Assuming |h0 | ∈ R is the amplitude of the LOS path and
B. ICI Analysis in Aeronautical Channel
the |h1 | ∈ R is the amplitude of the reflected path, then the
The ICI analysis is one of the key issues which has to power ratio, the so-called Rice factor, can be given as [12]
be completed to evaluate the performance of an OFDM

|h0 |2
system. The analysis will follow a general sample-spaced KRice (dB) = 10 × log10 . (7)
multipath fading channel model for analytical traceability. |h1 |2
x(n) are the discrete time samples of the transmitted
 The following normalization as a function of the Rice
OFDM signal, which can be given as: x(n) = N−1
j 2πkn
k=0
factor can be obtained: |h0 | = 1+K KRice
, |h1 | = 1+K1 Rice .
X(k)e N , 0 ≤ n ≤ N − 1, where X(k) is the symbol Rice
The ICI power caused by aeronautical channel is
transmitted on the kth subcarrier and N is the number of
illustrated in Fig. 2 for various 0 , 1 , and KRice values
subcarriers. Let 0 and 1 be the NDF for the first and the
fD 0 fD 1 span through the best to the worst case scenarios. It is
second ray, respectively, i.e., 0 = f and 1 = f ,
important to note that ICI power can be high even for very
where, fD0 and fD1 are the Doppler shifts due to two
small NDFs in dual-Doppler shift aeronautical channel
paths, and f is the OFDM subcarrier spacing. The
scenario when comparing with the terrestrial two-ray
general equation for L path channel can be given as
channel model given in [11].

L−1 ICI power in aeronautical channel will affect the
j 2πl (n−τl )
h(n) = hl e N δ(n − τl ) (2) detection of the symbols, since it can be seen as a
l=0 near-Gaussian noise [9]. In the terrestrial OFDM systems,
ICI is overcome by increasing the subcarrier spacing
where hl s are the path gains
with normalized overall (decreasing the length of the OFDM symbol) and
power, i.e., E [hl ] = 0, and L−1
l=0 E[|hl | ] = 1 and τl is
2
bounding the NDF [8, 11]. This method however has a
th
the delay corresponding to the l path as given in the twofold drawback: first, the system must be dynamic to
model of Doppler shifts with exponentials in [9]. change the system parameters depending on the level of
Assuming the receiver is synchronized to the first path Doppler, and second, more importantly, it reduces the
τ0 = 0, and τ1 = τ, i.e., τ being an integer, the received efficiency (data rate) of the system:
signal is
N log2 (Msub )
Ts +CP log2 (Msub )
ηOFDM = =
j 2π0 n j 2π1 (n−τ )
y(n) = h0 e N x(n) + h1 e N x(n − τ ) + w(n) (3) Nf 1 + CP f
(8)

where 0 ≤ n ≤ N − 1, and w(n) is additive white where N, Ts , CP, and f are the number of subcarriers,
Gaussian noise (AWGN) with one-sided spectral density symbol duration, CP size, and subcarrier spacing of the
of N0 . Taking the discrete Fourier transform (DFT) of (3) OFDM system, respectively, and Msub is the modulation
order of each subcarrier, i.e., Msub = 24 for 16-QAM.

N−1
Note, for a given CP size, increasing f decreases the
Y (k) = X(k)S(k, k) + X(m)S(m, k) + W (k) (4)
   efficiency of the OFDM system. Comparing with the
m=0,m=k
   Noise method of decreasing OFDM symbol duration to bound
ICI the effect of Doppler in terrestrial networks, since the
shape of the aeronautical channel Doppler has the
where the k th received symbol is related to the k th
particular characteristics defined in Section II-A, it is
transmitted symbol with coefficient S(k, k) and related to
possible to estimate the Doppler shifts with parametric
the other subcarrier’s symbols with S(m, k) [25]:
estimation techniques (Section III) and remove the effect
h0 sin (π(m − k + 0 )) j π (1−1/N)(m−k+0 ) of aeronautical channel on the received signal by
S(m, k) = e
Nsin(π(m − k + 0 )/N) beamforming techniques (Section IV).
  
First Ray III. PARAMETRIC DOPPLER ESTIMATION FOR
h1 sin (π(m − k + 1 )) j π (1−1/N)(m−k+1 ) −j 2πτ (m+1 ) AERONAUTICAL CHANNELS
+ e e N .
Nsin(π(m − k + 1 )/N) As discussed in the previous section, in order to
  
Second Ray mitigate the effect of ICI, the Doppler shifts need to be

122 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 50, NO. 1 JANUARY 2014
 Fig. 2. ICI power for various 0 , 1 , and KRice values.
estimated. For spectrum estimation, if the signal process
(i.e., modeling order) is known as in aeronautical channel,
the parametric methods outperform the nonparametric
methods, since the parametric methods try to understand
the process [26]. Parametric methods are categorized as
autoregressive (AR), moving average (MA),
autoregressive moving average (ARMA), and harmonic. In
this study harmonic process is chosen since we use a pilot
(single tone) as a training symbol. In parametric methods
the frequencies of interest are extracted through a method
known as eigendecomposition of the autocorrelation
matrix. The autocorrelation matrix is decomposed into two
subspaces, the signal subspace and the noise subspace,
which can be shown for aeronautical channel as follows.
Let us assume a training tone is generated by encoding
only one subcarrier of an OFDM symbol, i.e., X(k) = 1
j 2πρn
for k = ρ and X(k) = 0 for all other k, i.e., x(n) = e N .
Then the received signal can be given using (3) as
y(n) = h0 ej nω0 + h1 ej (n−τ )ω1 + w(n) (9)
where, ωi s are the Doppler-shifted pilot tone frequencies
in radians for i = 0, 1 where, ωi = 2π (Ni +ρ) , hl s are the
path gains with normalized overall power, i.e., E [hl ] = 0,
and w(n) is AWGN with one-sided spectral density of N0 .
Assuming the variance of w(n) as σw2 , the autocorrelation
matrix (size of MA × MA ) can be presented similar to the
example given in [26, pp. 455–458]:
Ry = Vs Ds VH
s + Vn Dv Vn
H
(10)
which is projecting a vector into signal and noise
subspaces, respectively. Note that Ds and Dv are diagonal
matrices that contain the eigenvalues of autocorrelation
matrix due to signal λi = λsi + σw2 and noise λi = σw2 ,
respectively. Vs = [v1 , v2 , . . . vp ] are the eigenvectors of
ERTÜRK ET AL.: DOPPLER MITIGATION IN OFDM-BASED AERONAUTICAL COMMUNICATIONS
Ry with their respective eigenvalues greater than σw2 (p is
the modeling order), and Vn = [vp+1 , vp+2 . . . vMA ] is the
group of eigenvectors of Ry that spans the noise subspace
and consists of eigenvalues equal to σw2 .
1) Multiple Signal Classification (MUSIC) Method:
Let Ry be the MA × MA autocorrelation matrix of x(n)
with MA > p + 1, and assume the eigenvalues are
arranged in decreasing order with their corresponding
eigenvectors. The eigenvectors are divided into two
groups, the p signal eigenvectors with the largest
eigenvalue, and the MA − p noise vectors that mostly
have eigenvalue equal to σw2 . The eigenvectors of Ry will
have a length of MA , and each of the noise subspace
eigenfilters will have MA − 1 roots (zeros). p of the roots
will lie on the unit circle at the frequencies of the complex
exponentials with the eigenspectrum exhibiting sharp
peaks at the frequencies of interest. The remaining
(MA − p − 1) zeros may lie anywhere, and may cause
spurious peaks in the eigenspectrum. However, the
MUSIC algorithm takes care of these spurious peaks by
means of averaging, as follows
1
PMU (ej w ) = M   (11)
A eH vi 2
i=p+1
where e = [1, ej ω , ej 2ω , . . . , ej (MA −1)ω ]T . Thus, finding
the angles of the roots of PMU , i.e., ω̂i , allows to estimate
the Doppler shift due to each ray. Finally, subtracting the
known transmitted tone, the ˆi can be extracted as
N ω̂i
ˆi = − ρ, i = 0, 1. (12)
2π
2) Eigenvector Method: The eigenvector method is
slightly different from the MUSIC algorithm, since a
compensation of each eigenvector is performed with its
123
associated eigenvalue, and therefore this method produces
less spurious peaks:
1
PEV (ej w ) = M   (13)
A 1 eH vi 2
i=p+1 λi
where λi is the eigenvalue associated with each
eigenvector vi . The angles of the roots of PEV also make
the Doppler frequencies and the estimation can be done as
in (12).
3) Minimum Norm Algorithm Method: The minimum
norm algorithm uses a single vector a that is constrained
to lie in the noise subspace, that is,
1 separating the paths with beamforming, and then
PMN (ej w ) =   (14)
eH a2
where a = Vn VH n v is the projection matrix that projects an
arbitrary vector v on to the subspace. The z-transform of a
−k
will be A(z) = M−1 k=0 a(k)z , where zk for k = p + 1,
. . . , MA − 1 are the spurious roots. The minimum norm
algorithm constrains the selection of roots to minimize the
effect of spurious roots. Determining the roots of PMN
allows us to estimate ˆi as in (12).
4) Parametric Modeling Sensitivity: The parametric
Doppler estimation techniques are driven by prior
knowledge of the signal process. Depending on the
modeling technique, each of the AR, MA, and ARMA
incur a different modeling error. Splitting of a single
spectral peak into two or more peaks occurs when the
incoming signal is overmodeled, i.e., when the modeling
order p is too large. If the signal process is known, and an
appropriate model with a known modeling order is
selected, a higher resolution spectrum estimation can be
achieved within a short signal duration. Since the
aeronautical channel is based on a two-ray model for the
nominal case based on [12], we are assuming two Doppler
shifts. However, as discussed in [18], the number of paths
might be greater than two. This issue lies within the scope
of this paper and is investigated with various modeling
errors via simulations in Section V.
IV. BEAMFORMING-BASED SIGNAL SEPARATION
FOR AERONAUTICAL DOPPLER MITIGATION
ICI due to dual or multiple Doppler shifts can be
compensated by a simple use of beamforming separating
the arriving paths. Once the paths are separated, the
Doppler shifts can be compensated individually with
conventional methods [19]. By taking advantage of
estimated individual shifts in the previous section, we find
the DOA and the weights of the antennas to separate the
paths for individual processing. The signals are further
combined to improve the receiver performance using
diversity combining schemes. See Fig. 3 for the proposed
aeronautical receiver block diagram.
The spatial filtering, also known as beamforming,
describes how an array of antenna elements combined
with signal processing can either block or direct the
radiation or reception of signals in the desired signal [27].
124 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 50, NO. 1 JANUARY 2014
Fig. 3. Aeronautical receiver block diagram.
In an aeronautical channel the Doppler shifts due to a wide
range of DOAs can be used as an advantage for first
compensating and combining them with diversity
techniques.
A linear array with evenly spaced sensors is used with
m being the index of each element, i.e., m = 0, 1, 2, . . . ,
M – 1. If the angle of the signal arriving at the sensor is θ
and each of the M sensors is d distance apart, the received
signal in the adjacent sensor travels a difference of
fc
c
d cos(θ), where fc is the signal carrier frequency and c
is the propagation speed. The phase difference between
any two elements with md distance is −m fcc d cos(θ) [28].
The received signal y(n) at mth sensor elements will be
fc
ym (n) = y(n)ej m c d cos(θ)
(15)
since the signal y(n) will experience the antenna array
factor due to the geometry of the sensor at reception. Let
us assume that we multiply the mth sensor element by a
weighting factor. The beamform to receive the signal for
l th path will be

M−1
yB,l (n) = ym (n)wm,l , l = 0, 1, . . . , L − 1 (16)
m=0
where yB,l (n) is the outcome of the signal spatially
processed through wm,l weights of M sensor elements for
L arriving paths. Therefore, L rays of signal each
associated with angle of arrivals θl will be received at the
array of sensors. If the angles are known, the respective
array weights can be calculated as
fc
wm,l = e−j m c d cos(θl )
. (17)
Each of the l paths will have its associated array of
weighting factors wm,l , l = 0, . . ., L – 1, based on DOA.
Given that the Doppler frequencies have been estimated
with (12), and the speed of the mobile platform (v) is
known, the arriving angle of different paths can be
calculated as3
−1 fDl c
θl = cos . (18)
fc v
3 Inthe current model θ l , l = 0, 1, is assumed to be the projection of
velocity vector on the direction of propagation. Therefore, the DOA of
the signal cannot be from upper 180◦ s for the aeronautical node and
lower 180◦ s for the ground node since there will be no propagation from
those directions.

Fig. 4. Beamforming radiation patterns of two sensor elements.
To provide multi-beamforming to separate each
incoming ray, each arriving angle will be used to generate
a different array of weights using (17). For an aeronautical
two-ray channel, there will be two, i.e., L = 2, arriving
signals with different DOA [12]. For the particular case of
paths arriving from angle 20◦ and 60◦ , two beamforming
weights are generated. Fig. 4 shows the radiation pattern
for the respective DOA with two antenna elements. Note
that the beam for the first signal will attenuate the signal
with angle of arrival 20◦ , and allow the second signal with
angle of arrival 60◦ to pass through. Similarly, the second
beam will attenuate the signal with angle of arrival 60◦ .
For the aeronautical two-ray channel, there will be two
such spatially processed signals, yB,0 (n) and yB,1 (n) (note
that two processing chains are assumed to be available for
processing each path separately as the weights of two
paths are different.). The signals are further combined to
improve the performance of the receiver. In this paper we
selected selective combining (SC) and maximal ratio
combining (MRC) for investigation.
1) Selection Combining: In SC the signal with highest
signal power is selected for further processing [29]:

1 l = l0
βl = (19)
0 o/w
where β l s are weights for each branch, and l0 is the branch
with the highest received signal power. Thus for the
aeronautical channel the SC equation will be
j 2π0 n j 2π1 n
ySC (n) = yB,0 (n)β0 e N + yB,1 (n)β1 e N (20)
where β l s are defined in (19).
2) Maximum Ratio Combining: In MRC the
beamformed aeronautical signals are cophased. The
diversity combining branch weights for this method can be
given as [30]
βl = al ej θl , l = 0, 1, . . . , L − 1
E[yB,l (n)]
where al = N0
E[yB,l (n)] and N0 denote the mean
,
signal amplitude in the l th branch and noise power. Note
that the gain on each branch becomes proportional to the
ERTÜRK ET AL.: DOPPLER MITIGATION IN OFDM-BASED AERONAUTICAL COMMUNICATIONS
TABLE I
Simulation Parameters
Carrier (fc ) 1000 Mhz (L-band)
Aeronautical node speed (v) 800 km/hr
Bandwidth (B) 800 kHz
Subcarrier number (N) 128
Subcarrier spacing (f ) 6.25 kHz
OFDM symbol duration (Ts ) 160 μs
OFDM CP size (TCP ) 1/4 Ts , 40 μs
Modulation QPSK (uncoded)
Channel Rician two-ray model
1st and 2nd Ray angles Uniform distr. {0, 2π }
Maximum delay τ max 33 μs [12]
Antenna separation 0.3m
Antenna number (M) 2, 4, 6
Autocorrelation matrix length (MA ) 50, 75, 100, 125
signal amplitude in this method. The MRC output can also
be given as in (20) with using weights as in (21).
V. NUMERICAL RESULTS
Computer simulations are performed based on an
OFDM system with uncoded quadrature phase shift
keying (QPSK) modulation in a two-ray channel model as
defined in Section II. The parameters used in simulations
are summarized in Table I. Doppler estimation simulation
with a signal-to-noise ratio (SNR) of 20 dB,4 and a
randomly and uniformly distributed dual-Doppler shift
ranging from −f to + f (i.e.,i ∼ U (−1, 1), i = 0, 1)
is performed for the aeronautical channel.5
Fig. 5 presents the MSE performance for the first and
second paths separately for various Rice factors (KRice )
and OFDM symbol numbers6 with an autocorrelation
matrix length of MA = 100. As the KRice increases, the
MSE of the LOS path decreases; on the other hand
the MSE of reflected path increases since the power of the
reflected path becomes smaller. Although the MSE of the
reflected path increases with the KRice , since its power
4 We studied different SNR levels, however, there is no significant
performance improvement for range of SNR ∈ {0dB, 20dB}. Therefore,
we only provide SNR = 20 dB. Note that this behavior is due to
generating the training tone only for one subcarrier for the estimation.
The number of subcarriers in a training OFDM symbol is important as
the MSE will change with noise level for increased subcarriers. In the
current simulation scenario, the cross effect of individual paths on the
estimation is dominant compared with noise. The scope of this paper is to
investigate the effect of two Doppler shifts, and the optimization of the
number of training subcarriers depending on the noise level to provide
better estimation is left as a future work.
5 We also investigated the MSE with various constant Doppler scenarios.
As the two Doppler shifts become closer, their separability becomes
harder – the complexity and latency for the estimation increases, thus
MSE performance becomes worse. For the sake of brevity, we provide
the results only with randomly and uniformly distributed Doppler
frequencies to find the average value for MSE (given in (22)) as
(21)
discussed in [12].
6 The derivations for estimations of Doppler shifts are provided for only
one OFDM symbol for analytical tractability. However, to provide better
estimation, multiple OFDM symbols and various autocorrelation matrix
lengths (MA ) are investigated in simulations.
125
 Fig. 5. MSE performance of individual paths for various Rice factors. a) LOS path. b) Reflected path.

Fig. 6. MSE performance for various autocorrelation matrix lengths Fig. 7. MSE performance for various methods and OFDM symbols
and OFDM symbols for MUSIC. with autocorrelation matrix length MA = 100.

decreases, the effect of it also decreases on the ICI power.

This can also be tracked from (6) and Fig 2. Therefore, we
investigate the rest of our results with an average MSE
with weighting factors as follows
 L−1
 1 | i − ˆ
 i |2
Average MSE = E |hi |2
L i=0 |i |2
In our simulations KRice is randomly changed with a
uniform distribution between 2 – 20 dB to simulate an en
route scenario [12, 14, 15]. Fig. 6 shows the MSE
performance of the MUSIC algorithm for various numbers
of OFDM symbols and lengths of autocorrelation matrix
MA . Note that with using 12 – 14 OFDM symbols for
estimation, the MSE is below 1%. The increase of MA and
OFDM symbols shows a trade-off between the increase in
the number of computations and the increase in the
latency. Simulations show that various MSE performances
can be achieved depending on OFDM symbols and MA . In
Fig. 7 MUSIC, EV, and minimum norm algorithms are
studied for a given autocorrelation matrix length, i.e., MA
= 100. Although the performance curve of the three
 algorithms is relatively close, the EV method shows a
. (22) slightly better performance of estimating the dual-Doppler
shift using fewer training OFDM symbols.
Fig. 8 shows the performance of the MUSIC
algorithm when the modeling order is chosen incorrectly.
Using the example where the number of incoming rays is
L = 2, we investigate the effect of modeling order p = 2,
3, 5. The incorrect model order, i.e., p = 3 causes an
average of 5% MSE difference while the two-ray model is
used, which will impact the overall system BER
performance (see Fig. 10 for the effect of estimation
error on BER performance.). Therefore, techniques to
estimate modeling parameters, i.e., order, need to be
considered [31]. Channel estimation techniques that
have the capability to estimate channel multipaths or

126 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 50, NO. 1 JANUARY 2014

Fig. 8. MUSIC method modeling sensitivity with autocorrelation
matrix length MA = 100.
echos can be used for better estimation of modeling order
[32, 33].
The paths are separated using beamforming for M
sensor elements as described in Section IV. Fig. 9 shows
BER performance for different sensor numbers and
combining techniques together with no Doppler channel
(M = 6) scenario as a benchmark. As the number of
antenna elements increases, narrower beams are formed
for the incoming paths; therefore BER performance
increases significantly. In addition, when two separated
signals are combined using MRC compared with SC, there
is a 4 dB increase in performance. Another simulation is
run to estimate the BER performance with respect to
estimation error. Fig. 10 shows the respective performance
for 1% to 5% MSE for MRC with M = 6. It can be
concluded that the impact of Doppler shifts in the
aeronautical channel on the OFDM-based system can be
mitigated with parametric spectrum estimation and
beamforming. Moreover, the performance loss due to
frequency estimation error can easily be compensated by
beamforming and diversity combining the multiple paths.
Fig. 9. BER performance of beamforming-based signal separation and
diversity combining.
ERTÜRK ET AL.: DOPPLER MITIGATION IN OFDM-BASED AERONAUTICAL COMMUNICATIONS
Fig. 10. BER performance with various estimation errors for MRC with
M = 6 antennas.
VI. CONCLUSION AND FUTURE WORK
In this paper an OFDM-based system is analyzed
particularly for aeronautical environment and it is shown
that ICI can be mitigated by exploiting the dual-Doppler
shift characteristics of the aeronautical channel. As the
number of paths is predictable, parametric spectrum
estimation algorithms are used to estimate the Doppler
shifts. Simulations using the MUSIC, EV, and minimum
norm algorithms show that the estimation can be done
efficiently with an MSE performance less than 1%. It is
shown that increasing the autocorrelation matrix size, or
number of OFDM symbols, for estimation, increases the
MSE performance but can cause an increase in the number
of computations, and the latency of the estimation,
respectively. We show when modeling order is not
correctly selected, the estimation performance degrades
dramatically. For the compensation of ICI using the
estimated Doppler shifts, we first find the DOAs of the
paths. Based on DOA and estimation errors, we show that
beamforming with a different number of antenna elements
can create beams with resolutions that are capable of
separating these Doppler affected paths. The separated
signals are first compensated for single Doppler shift and
then diversity combining techniques are used to improve
the system BER performance. Numerical results show that
BER performance approaches the no-Doppler scenario
with the beamforming-based mitigation method. The
estimation and compensation methods proposed in this
paper can easily be applied to single-carrier systems. The
future direction of this study includes recursive estimation
of the dual-Doppler frequency shift using fewer OFDM
symbols and lower autocorrelation matrix.
REFERENCES
[1] Schnell, M. and Scalise, S.
NEWSKY – Concept for NEtWorking the SKY for civil
aeronautical communications.
IEEE Aerospace and Electronic Systems Magazine, 22, 5
(May 2007), 5–29.
[2] NASA-ACAST Future aeronautical communication
infrastructure technology investigation. NASA Contractor
127
 Report CR-2008-215144, Apr. 2008. [Online.]
http://acast.grc.nasa.gov/main/projects/
[3] Kerczewski, R.
Aeronautical communications research and development
needs for future air traffic management applications.
In Proceedings of the IEEE Aerospace Conference,
(AEROCONF) Mar. 2002, pp. 1169–1176.
[4] Sakhaee, E. and Jamalipour, A.
The global in-flight internet.
IEEE Journal on Selected Areas in Communication,
24 (Sept. 2006), 1748–1757.
[5] Lai, J.
Broadband wireless communication systems provided by
commercial airplanes. U.S. Patent 6 285 878, Sept. 4, 2001.
[6] Connell, L.
Pilot and controller communications issues. In Proceedings of
the Methods and Metrics of Voice Communications
Workshops, NASA-Ames Research Center, 1995.
[7] Jahn, A. and Holzbock, M.
Evolution of aeronautical communications for personal and
multimedia services.
IEEE Communications Magazine (July 2003), 104–112.
[8] Prasad, R.
OFDM for Wireless Communication Systems. Norwood, MA:
Artech House, 2004.
[9] Robertson, P. and Kaiser, S.
Analysis of the loss of orthogonality through Doppler spread
in OFDM systems. In Proceedings of the IEEE Global
Telecommunications Conference (GLOBECOM), Dec. 1999,
pp. 701–706.
[10] Cheon, H. and Hong, D.
Effect of channel estimation error in OFDM-based WLAN.
IEEE Communications Letters, 6, 5 (May 2002), 190–192.
[11] Li, Y. and Cimini, L. J.
Bounds on the interchannel interference of OFDM in
time-varying impairments.
IEEE Transactions on Communications, 49, 3 (Mar. 2001),
401–404.
[12] Haas, E.
Aeronautical channel modeling.
IEEE Transactions on Vehicular Technology, 51,
2 (Mar. 2002), 254–264.
[13] An, J., Wu, Y., and Liu, G.
Delay and Doppler shift joint tracking method for OFDM
based aeronautical communication systems. In Proceedings of
the International Conference on Wireless Communications,
Networking and Mobile Computing, Oct. 2008, pp. 1–4.
[14] Xiangyuan, B., Wei, X., and Ronghua, Z.
A novel channel estimation scheme for wideband OFDM
aeronautical communication. In Proceedings of the
International Conference on Wireless Communications,
Networking and Mobile Computing, Sept. 2009, pp. 1–4.
[15] Wu, X., Zhang, J., and Liu, Z.
Performance analysis of TCI elimination by information aid
for aviation mobile broadband communication. In
Proceedings of the IEEE Vehicular Technology Conference
(VTC), Sept. 2008, pp. 1–5.
[16] Zhang, C, Han, H., and Yu, H.
NC-OSDM transmission for ICI cancelation in high speed
mobile systems. In Proceedings of the Second International
Workshop on Cognitive Radio and Advanced Spectrum
Management, May 2009, pp. 50–53.
[17] Popescu, O. et al.
Interference mitigation in aeronautical telemetry systems
using Kalman filter.
IEEE Transactions on Aerospace and Electronic Systems, 43,
4 (Oct. 2007), 1624–1630.
[18] Rice, M., Davis, A., and Bettweiser, C.
Wideband channel model for aeronautical telemetry.
IEEE Transactions on Aerospace and Electronic Systems, 40,
1 (Jan. 2004), 57–69.
[19] Van De Beek, M. J. J. and Borjesson, P.
ML estimation of the time and frequency offset in OFDM
system.
IEEE Transactions on Signal Processing, 45 (July 1997),
1800–1805.
[20] Erturk, M. C., Haque, J., and Arslan, H.
Challenges in aeronautical data networks. In Proceedings of
the IEEE Aerospace Conference (AEROCONF), Mar. 2010,
pp. 1–4.
[21] Sen, I. and Matolak, D.
The 5-GHz airport surface area channel: Part II, Measurement
and modeling results for small airports.
IEEE Transactions on Vehicular Technology, 57, 4 (July
2008), 2027–2035.
[22] Bello, P. A.
Aeronautical channel characterization.
IEEE Transactions on Communications, COM-21, 11 (May
1973), 548–563.
[23] Elnoubi, S.
A simplified stochastic model for the aeronautical mobile
radio channel. In Proceedings of the IEEE Vehicular
Technology Conference (VTC), vol. 2, 1992,
pp. 960–963.
[24] Hoeher, P. and Haas, E.
Aeronautical channel modeling at VHF-band. In Proceedings
of the. IEEE Vehicular Technology Conference (VTC), vol. 4,
1999, pp. 1961–1966.
[25] Haque, J., Erturk, M., and Arslan, H.
Aeronautical ICI analysis and Doppler estimation.
IEEE Communications Letters, 15, 9 (Sept. 2011), 906–909.
[26] Hayes, M. H.
Statistical Digital Signal Processing and Modeling. Hoboken,
NJ: Wiley, Feb. 2006, pp. 455–458.
[27] Veen, B. D. V. and Buckley, K. M.
Beamforming: A versatile approach to spatial filtering.
IEEE ASSP Magazine, 5, 2 (Apr. 1998), 4–24.
[28] Godara, L. C.
Application of antenna arrays to mobile communications:
Part II, Beam-forming and direction-of-arrival considerations.
Proceedings of the IEEE, 85, 8 (Aug. 1997),
1195–1245.
[29] Kong, N. and Milstein, L.
Average SNR of a generalized diversity selection combining
scheme.
IEEE Communications Letters, 3, 3 (Mar. 1999), 57–59.
[30] Shah, A. and Haimovich, A.
Performance analysis of maximum ratio combining and
comparison with optimum combining for mobile radio
communications with co-channel interference.
IEEE Transactions on Vehicular Technology, 49 (Mar. 2000),
1454–1463.
[31] Rezek, S. and Roberts, I. A.
Parametric model order estimation: A brief review.
Model Based Digital Signal Processing Techniques in the
Analysis of Biomedical Signals (Digest 1997/009), IEE
Colloquium on the Use of Biomedical Signals (Ref. 1997/009),
Apr. 1997,
pp. 3/1-3/6, DOI 10.1049/ic:19970061.
[32] Vanderveen, M., van der Veen, A.-J., and Paulraj, A.
Estimation of multipath parameters in wireless
communications.
IEEE Transactions on Signal Processing, 46, 3 (Mar. 1998),
682–690.
[33] Vanderveen, M., Papadias, C., and Paulraj, A.
Joint angle and delay estimation (JADE) for multipath signals
arriving at an antenna array.
IEEE Communications Letters, 1, 1 (Jan. 1997), 12–14.

128 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 50, NO. 1 JANUARY 2014
 M. Cenk Ertürk received his B.Sc. (first class honors) and M.Sc. degrees in electrical
engineering from Nigde and Bilkent Universities, Turkey in 2005 and 2008, and his
Ph.D. degree from University of South Florida, Tampa, FL in 2012.
From 2006 to 2009 he was with the Defense and Security Research Group in the
Scientific and Technological Research Council of Turkey (TUBITAK, Ankara, Turkey)
where he was involved in several communication projects servicing the improvement of
Turkey’s defense technologies and from 2011 to 2012, he was with the Docomo USA
Communications Laboratories, Palo Alto, CA, where he was involved in system level
designs of future radio access systems. His research interests include system level
design of heterogeneous networks, aeronautical communication networks, and
baseband signal processing in wireless communication networks.

Jamal Haque received his B.S. and M.S. and Ph.D degrees in electrical engineering from the University of South
Florida, Tampa, FL.
He has over sixteen years of telecommunication and aerospace products design and development experience in the
area of communication and signal processing. His research interests include wireless systems, OFDM-based systems in
high mobile platform, synchronization, channel estimation, cognitive, software defined radio, and channel coding.
Dr. Haque holds twenty US patents.

Wilfrido A. Moreno is an associate professor in the Department of Electrical Engineering at the University of South
Florida and director for the Ibero American Science and Technology Educational Consortium at USF, (USF-ISTEC). He
is a founding member of the Center for Microelectronics Research, (CMR), where he serves as the supervisor for the
System Integration and Laser Processing Laboratory. His current research interest is oriented towards system integration
by providing off-the-shelf hardware/software solutions to industrial application in the areas of digital signal processing,
communications, industrial controls, microelectronics, biomedical engineering, and multimedia solutions including
voice and video over IP applied to distance learning solutions. Since becoming an assistant professor in the fall of 1994,
he has been successful in obtaining funding from local industries and federal agencies that have supported over seventy
masters students and ten doctoral students. Additionally, three research laboratories have been established with
state-of-the-art equipment and software tools in the areas of digital signal processing, rapid system prototyping and
within the last year, a wireless sensor networks laboratory.
Dr. Moreno has published over eighty technical papers and has participated in several International conferences and
seminars around the world.
Hüseyin Arslan received his Ph.D. degree in 1998 from Southern Methodist University
(SMU), Dallas, TX.
From January 1998 to August 2002, he was with the research group of Ericsson
Inc., NC, where he was involved with several project related to 2G and 3G wireless
cellular communication systems. Since August 2002, he has been with the Electrical
Engineering Department of the University of South Florida. In addition he has worked
as part time consultant for various companies and institutions including Anritsu
Company, The Scientific and Technological Research Council of Turkey-TUBITAK,
Lecroy, and XG technologies. His research interests are related to advanced signal
processing techniques at the physical layer, with cross-layer design for networking
adaptivity and quality of service (QoS) control. He is interested in many forms of
wireless technologies including cellular, wireless PAN/LAN/MANs, fixed wireless
access, and specialized wireless data networks like wireless sensor networks and
wireless telemetry. His current research interests are on UWB, OFDM based wireless
technologies with emphasis on WIMAX and IMT-advanced, and cognitive and software
defined radio.
Dr. Arslan has served as technical program committee chair, technical program
committee member, session and symposium organizer, and workshop chair in several
IEEE conferences. He is a member of the editorial board for Wireless Communication
and Mobile Computing Journal and Research Letters in Communications.

ERTÜRK ET AL.: DOPPLER MITIGATION IN OFDM-BASED AERONAUTICAL COMMUNICATIONS 129