Received November 16, 2015, accepted December 21, 2015, date of publication January 13, 2016,

date of current version March 4, 2016.

Digital Object Identifier 10.1109/ACCESS.2016.2518085

A Survey on High Mobility Wireless

Communications: Challenges,
Opportunities and Solutions
JINGXIAN WU1 , (Senior Member, IEEE), AND PINGZHI FAN2 , (Fellow, IEEE)
1 University of Arkansas, Fayetteville, AR 72703, USA
2 Key Laboratory of Information Coding and Transmission, Southwest Jiaotong University, Chengdu 610031, China
Corresponding author: J. Wu (wuj@uark.edu)
This work was supported by the National Basic Research Program of China (973 Program) under Grant 2012CB316100, the National
Science Foundation of US under Grant ECCS-1202075, the Fundamental Research Funds for the Central Universities under Grant
SWJTU12ZT02/11CX147/2682014ZT11, the National Science Foundation of China (NSFC) under Grant 61471302, and the 111 Project
Grant 111-2-14.

ABSTRACT Providing reliable broadband wireless communications in high mobility environments, such
as high-speed railway systems, remains one of the main challenges faced by the development of the next
generation wireless systems. This paper provides a systematic review of high mobility communications.
We first summarize a list of key challenges and opportunities in high mobility communication systems,
then provide comprehensive reviews of techniques that can address these challenges and utilize the unique
opportunities. The review covers a wide spectrum of communication operations, including the accurate
modeling of high mobility channels, the transceiver structures that can exploit the properties of high mobility
environments, the signal processing techniques that can harvest the benefits (e.g., Doppler diversity) and
mitigate the impairments (e.g., carrier frequency offset, intercarrier interference, channel estimation errors) in
high mobility systems, and the mobility management and network architectures that are designed specifically
for high mobility systems. The survey focuses primarily on physical layer operations, which are affected
the most by the mobile environment, with some additional discussions on higher layer operations, such
as handover management and control-plane/user-plane decoupling, which are essential to high mobility
operations. Future research directions on high mobility communications are summarized at the end of this
paper.

INDEX TERMS High mobility communications, fast time-varying fading, CFO, ICI, Doppler diversity,
mobility management.

I. INTRODUCTION Most current wireless communication systems are

Wireless communications in high mobility environments
have attracted considerable attentions during the past few
years, due to the large scale deployment of high speed rail-
way (HSR) systems, and the growing popularity of highway
vehicular communications systems and low altitude flying
object (LAFO) systems [1]–[6]. High mobility communi-
cations have been incorporated as an integral part of the
fifth generation (5G) communications [7], [8]. 5G systems
are expected to provide simultaneous broadband services to a
large number of users traveling at speeds up to 500 km/hr, the
speed achievable by HSR systems, at a data rate of 150 Mbps
or higher [9]. The speed is even higher for LAFO systems,
where the relative speed between transmitter and receiver can
be on the order of 1000 km/hr or higher.
designed for users with low or medium mobility, and high
mobility significantly limits the coverage area and transmis-
sion rate. The most popular high mobility communication
system is the global system for mobile for rail (GSM-R),
which supports a maximum data rate less than 200 kbps,
and the narrowband data service can only be used for train
operation controls. The fourth generation (4G) long term
evolution advanced (LTE-advance) systems are designed to
provide optimized services to users in the 0 to 15 km/hr range,
high performance services from 15 km/hr to 120 km/hr,
and functional services from 120 km/hr to 350 km/hr [10].
However, field tests at different countries show that
existing 4G systems can only provide a data rate on the
order of 2-4 Mbps to high speed trains. Future high mobility

2169-3536
2016 IEEE. Translations and content mining are permitted for academic research only.
450 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 4, 2016
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
systems are expected to provide services to high mobility
users at a data rate on the order of hundreds of Mbps or
higher [9]. Existing technologies developed primarily for
low or medium mobility users cannot be readily applied
to high mobility scenarios with speed up to 500 km/hr or
higher, where the fast movement of the wireless terminals
will seriously degrade the system performance and make the
system nonfunctional. Therefore, it is imperative to develop
new theories and technologies that are designed specifically
for high mobility environments, and they should be able
to cope with the challenges and opportunities from high
mobility communications.
A. KEY CHALLENGES OF HIGH MOBILITY
COMMUNICATIONS
High mobility of the wireless communication terminals
imposes numerous challenges on the modeling, design,
analysis, and evaluations of next generation communication
systems. In the mean time, high mobility also provides unique
opportunities that can be exploited to facilitate system designs
and to improve system performance. Below we provide a
list of key challenges faced by the design of high mobility
communication systems, and some opportunities that can
benefit system designs.
1) FAST TIME-VARYING FADING
One of the main distinguishing factors of high mobility com-
munications is the fast time-variation of the fading channel
caused by the large Doppler spread. For a wireless terminal
operating at a speed of 500 km/hr and a frequency of 5 GHz,
the maximum Doppler frequency is fD = 2, 314 Hz,
which corresponds to a channel coherence time of
approximately 216 µs. It should be noted that most current
wireless communication systems are designed to operate for
fD on the order of hundreds of Hz. In addition, the movement
speeds of the wireless terminals will change with respect
to time, and this results in time-varying Doppler spreads
and non-stationary fading coefficients. The non-stationary
properties of the fading channel, along with the fast changing
scattering environment, make it a challenging task for the
accurate modeling and analysis of high mobility channels.
2) CHANNEL ESTIMATION ERRORS
In high mobility systems, due to fast fading variation, it is
more difficult to accurately estimate, track, and predict the
fast time-varying fading coefficients. Channel estimation is a
non-trivial task in high mobility systems, and the estimation
performance might have significant impacts on the overall
system performance. In many high mobility systems, chan-
nel estimation errors are inevitable and they will seriously
degrade system performance. The common assumption of
perfect channel state information (CSI) adopted for the design
of low mobility systems is no longer valid for high mobil-
ity systems. Thus the designs of high mobility transceivers
need to take into considerations of the properties of channel
estimation errors. Alternatively, non-coherent systems that do
VOLUME 4, 2016
not require the knowledge of CSI can be developed for high
mobility systems.
3) DOPPLER DIVERSITY
Even though fast time-varying fading will degrade system
performance, the short coherence time and fast variation
of the fading channel also provide Doppler diversity,
which can be exploited to improve system performance.
Doppler diversity has been studied extensively for systems
with perfect CSI. However, channel estimation errors are
non-negligible in high mobility systems and they may have
significant impact on achievable Doppler diversity orders.
A higher Doppler spread means a high Doppler diversity
order, but also in general a higher channel estimation mean
squared error (MSE). Thus there is a tradeoff between
Doppler diversity and channel estimation errors, and it is
important to identify the optimum Doppler diversity designs
in the presence of imperfect CSI.
4) CARRIER FREQUENCY OFFSET
The Doppler shift at the received signal will cause a mismatch
between the frequencies of the oscillators at the transmitter
and receiver, and this is called carrier frequency offset (CFO).
It should be noted that, even for systems without Doppler
shifts, small CFO may also occur due to the instability of
oscillators in wireless transceivers. In multi-carrier systems
such as orthogonal frequency division multiplexing (OFDM),
CFO will destroy the orthogonality among the subcarriers,
and introduce inter-carrier interference (ICI). ICI will
seriously degrade the performance of OFDM systems.
In addition, since the Doppler shifts in high mobility systems
change with respect to time, the CFO is time-varying, which
is difficult to track and compensate.
5) INTERCARRIER INTERFERENCE
ICI in high mobility OFDM systems are caused by two
factors: CFO and doubly-selective (both time-selective and
frequency-selective) fading. Conventional OFDM systems
are designed primarily for systems experiencing quasi-static
frequency selective fading, where the channel remains
constant within one OFDM symbol period. In high mobility
systems with doubly selective fading, the channel changes
inside one OFDM symbol, and the fading time-variation
will destroy the orthogonality among subcarriers and intro-
duce ICI. The ICI in systems with doubly-selective fading
is a direct consequence of Doppler spread, and it can be
utilized to provide Doppler diversity. Thus the design of high
mobility OFDM systems should be able to simultaneously
collect the benefits of Doppler diversity, and mitigate the
negative impacts of ICI.
6) HIGH PENETRATION LOSS
In HSR systems, the well sealed train carriages introduce
severe penetration loss of the wireless signals. The carriages
are made of special alloys and difficult to penetrate.
Furthermore, they form the Faraday cage effects that will
451

further reduce the received signal power in both uplink and
downlink. The penetration loss from train carriage is usually
in the range between 20 dB to 35 dB [6], [11]. The large
penetration loss will seriously degrade the signal-to-noise
ratio (SNR) at both the basestation (BS) and mobile sta-
tion (MS), and makes it even more difficult to achieve reliable
broadband communications in high mobility environments.
7) FAST AND FREQUENT HANDOVERS
Handover, or handoff, is the process by which a MS maintains
its connection active while moving from one point of attach-
ment to another. A high mobility MS may travel through the
coverage areas of multiple BSs or access points (APs) within
a short period of time. Traveling across the boundaries of
areas covered by two different BSs will trigger handover, and
high mobility systems will experience frequent handovers.
In addition, due to the fast movement of the MS, there is
only a very small time window to complete the handover
process, and handover failures will occur if the handover pro-
cess cannot be completed on time. Therefore, new handover
algorithms are needed to meet the requirements of handovers
in high mobility systems, such that we can reduce the number
of handovers, the handover latency, and the handover failure
probability. For HSR systems, the group handover is another
big problem when hundreds of in-train MS trying to handover
at the same time, generating the so-called signaling storm.
Fortunately, group handovers may be eliminated by
introducing on-train mobile relays.
8) NETWORK ARCHITECTURE
Fast movement of the communication devices also intro-
duces fast topology variation of the entire communication
network. Based on the unique properties and geometric layout
of high mobility systems, new network architectures can
be developed or adapted to meet the requirement of high
mobility communications. In 5G systems, there are a
plethora of new network architectures, such as heterogeneous
networks (HetNet) with macrocells and small cells, cloud
radio access network (C-RAN), control-plane and user-
plane (C/U-plane) decoupling, and mobile relays, etc., which
can be utilized to provide more effective and efficient
coverage to high mobility users.
B. OUTLINE
The objective of this paper is to provide a review and survey of
theories and practices that can cope with the challenges and
opportunities of high mobility communications. The survey
will focus primarily on the physical layer operations, with
some discussions on upper layer operations such as handover
and network architectures due to their importance for high
mobility communications.
The remainder of this paper is organized as follows.
Section II reviews the models of fast time-varying channels
for various high mobility systems, including both statisti-
cal models and geometry-based models for fixed-to-mobile,
mobile-to-mobile, and air-to-ground scenarios.
452
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
Section III discusses the estimations of fast time-varying
fading channels. Due to the fast time-varying nature of
the channel, blind estimation algorithms are in general not
suitable for high mobility systems. Thus we focus on pilot
assisted channel estimations, including the design of pilot
patterns and channel estimation algorithms.
Section IV presents the fundamental theory and implemen-
tations of Doppler diversity transmissions for systems with
or without perfect CSI. In the theoretical aspect, we review
the recent theoretical results on the fundamental performance
limits of Doppler diversity transmissions with imperfect CSI.
Then the practical implementations of Doppler diversity
transmissions are discussed.
Section V discusses multicarrier modulations in high
mobility systems. We first discuss the impacts and estimation
of CFO, then review the techniques for joint ICI cancellation
and Doppler diversity collection in doubly-selective fading.
The recently proposed concept of mobility adaptation for
OFDM is reviewed, where the system adapt the operation
parameters of OFDM systems based on the velocity of the
mobile stations.
Section VI reviews non-coherent detections for high
mobility systems. Non-coherent detection is specifically suit-
able for high mobility systems because it does not require the
knowledge of CSI, and can operate only with the knowledge
of the channel statistics. The information theoretical limits
of non-coherent detection are reviewed first, then several
practical non-coherent schemes are presented.
Section VII reviews handover operations in high mobility
systems. We first discuss the challenges faced by the
handover designs in high mobility systems, then present
several handover schemes in various system structures, such
as distributed antenna system (DAS), coordinate multipoint
transmission (CoMP), and HetNet.
In Section VIII, we present several network architectures
that are suitable for high mobility operations, including the
network architecture for railway systems, the C/U-plane
decoupling in HetNet, mobile relays, and C-RAN.
Finally Section IX concludes the paper, and briefly
discusses future research directions and challenges for high
mobility communications.
II. RAPIDLY TIME-VARYING CHANNEL MODELING
The high mobility wireless communication channels have
several unique properties, such as large Doppler spread,
large angular spread, and non-stationarity, etc. High mobility
channel models should be able to accurately represent such
properties.
The time-varying impulse response of a wide-band wire-
less channel can be represented by h(t, τ ), which is the
response of the channel at time t to an impulse input at
time t − τ . The variable t represents the time variation
of the channel, and τ represents the propagation delay.
The channel impulse response (CIR) usually incorporates
large-scale fading such as pathloss and shadowing, and small-
scale fading. Since large-scale fading changes slowly even in
VOLUME 4, 2016
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
high mobility systems, we will focus on small-scale fading,
which changes very fast in high mobility systems due to the
Doppler effects.
The CIR, h(t, τ ), of a high mobility system is a
non-stationary random process with time-varying statisti-
cal parameters. The objective of channel modeling is to
approximate the properties of h(t, τ ) through analysis and
simulations. High mobility channel models can be classified
into two categories, statistical models and geometry-based
models. Statistical models represent the wireless channel by
using existing statistical distributions. The behaviors of the
wireless channel are abstracted into analytical parameters,
such as density function of fading envelope, auto-correlation
function, level crossing rate, etc. The statistical models are
simple in that they can be represented with a few parame-
ters. They are usually used for wide-sense stationary (WSS)
channels, and they might not be able to capture the rich
interactions and time-varying behaviors in a high mobility
environment. The geometry-based models represent the chan-
nels by using ray tracing over a large number of scatters
placed on certain geometry shapes. Compared to the statisti-
cal models, geometry-based models provide a more realistic
and accurate representation of the high mobility environment,
at the cost of high modeling complexity.
A. STATISTICAL MODELS
There are three widely used statistical models for time-
varying wireless channels: the classical Jake’s model [12],
the basis expansion model (BEM) [13], and the Markovian
models [14]. All models are used to represent the time
evolution of the small scale fading.
1) JAKE’s MODEL
The Jake’s model is a simulation implementation of Clarke’s
two-dimensional (2-D) isotropic scattering model [15], which
assumes that the wireless signals propagate in a 2-D plane and
arrive at the MS from all directions with equal probability.
Under the assumption of wide sense stationary uncorrelated
scattering (WSSUS), the autocorrelation function of the
complex fading coefficients is
Rhh (t2 − t1 , τ1 , τ2 ) = E h(t1 , τ1 )h∗ (t2 , τ2 )
 
= Ch (τ1 )J0 (2π fD (t2 − t1 ))δ(τ2 − τ1 ),
(1)
where Ch (τ ) is the average power of h(t, τ ), J0 (x) is the
zero-order Bessel function of the first kind, fD is the maxi-
mum Doppler spread, and δ(x) is the Direc delta function.
The corresponding power spectral density (PSD) can be
obtained through the Fourier transform of Rhh (t, τ1 , τ2 ), and
it is [16]
 
Ch (τ1 ) f models of HSR channels are proposed in [14] and [26],
8hh (f , τ1 , τ2 ) = q rect δ(τ2 − τ1 ), (2)
π fD2 − f 2 2fD
where rect(x) = 1 for x ∈ [−0.5, 0.5] and 0 elsewhere. It is
clear that the fading spectrum is limited between [−fD , fD ].
VOLUME 4, 2016
The main advantage of the 2-D isotropic scattering model
is its simplicity and analytical tractability, thus it is the most
widely used model in the literature. Several improvements
of Jake’s model can be found in [17] and [18]. The Jake’s
model is extended to vehicle-to-vehicle (V2V) scenarios
in [19] and [20].
2) BEM
The BEM model is mainly used for the discrete-time CIR,
h(n, l) = h(nTs , lTs ), where Ts is the sampling period. The
discrete-time CIR in the time domain can be transformed into
a low-dimensional transform domain as
Q−1
X
h(n, l) = cq (l)bq (n), (3)
q=0
where {cq (l)}Q are a set of random BEM coefficients at delay
τ = lTs , {bq (n)}Q are deterministic basis functions, and
Q is the BEM dimension. Usually, BEM coefficients remain
invariant over a short period of time, e.g. several millisec-
onds (ms). In this time period, the time-varying channel is
represented as the weighted summation of deterministic basis
functions multiplied by time invariant BEM coefficients.
One of the most commonly used basis function is the com-
N −1
plex exponential (CE) basis, where {bq (n) = ej2πωq n }n=0 ,
Q
N is the block length, wq = (q−d 2 e)/N for uniform Doppler
sampling, and Q = 2dfD NTs e. The CE-based BEM converts
a block of N time-domain coefficients into Q transform-
domain coefficients. Usually Q  N , thus the model
complexity is reduced.
Other basis functions include polynomial basis [21], [22],
Karhunen-Lòeve basis [23], and discrete prolate spheroidal
sequences [24], [25]. All focus on the reduction of the channel
dimension in a transform domain.
3) MARKOVIAN MODEL
The Markovian model family represents the wireless channel
as finite-state Markovian chains, that is, the current channel
coefficient depends on the preceding ones. The most com-
monly used Markovian model is the auto regressive (AR)
model. The first-order AR model is
h(n, l) = ah(n − 1, l) + v(n, l), (4)
where the parameter a represents the model evolution, and
v(n, l) is the model error. The impact of the Doppler spread
is characterized by the parameter a. The AR model is sim-
ple and analytically tractable, but it cannot capture the rich
interactions in a high mobility environment.
A large number of Markovian models have been proposed
for different high mobility environments [14], [26]–[31]. The
where the birth-death of multi-path components are modeled
as Markov chains. A class of finite-state Markov channel
models are proposed in [31] to model different aeronautical
environment, such as parking and taxi environment,
453

TABLE 1. Goemetry-based channel models.
takeoff and landing situations, and en-route scenarios for
ground-air and air-air links.
All statistical models are developed for WSS channels.
In high mobility communications, the channel might be
non-WSS. In this case, we can divide the time duration into
small blocks such that the channel is WSS in each block, and
the model parameters change from block to block.
B. GEOMETRY-BASED MODELS
The geometry-based models try to represent the propaga-
tion environment by placing scatters according to certain
deterministic or stochastic geometry layouts. The geometry-
based models can be classified as [32]: geometry-based
deterministic model (GBDM) and geometry-based stochastic
model (GBSM).
GBDM describes physical channel parameters in a
completely deterministic way. Despite its high accuracy,
GBDM requires a detailed description of the propagation
environment and extensive computational resources.
In GBSM, the impulse responses of channel is
characterized by the law of wave propagation applied to
specific transmitter, receiver, and scatterer geometries that
are predefined in a stochastic fashion according to certain
probability distributions [33]. Different types of GBSM differ
mainly in the proposed scatterer distributions. Based on
the positions of the effective scatterers, GBSM can be fur-
ther classified as regular-shaped GBSM (RS-GBSM) and
irregular shaped GBSM (IS-GBSM). In RS-GBSM, all
effective scatterers are placed on regular shapes, e.g.,
one-ring, two-ring, and ellipses for 2-D models and
one-cylinder, two-cylinder, one-sphere, two-sphere, elliptic-
cylinders for three-dimensional (3-D) models. Unlike
RS-GBSM, IS-GBSM places the effective scatterers with pre-
defined properties at random locations with certain statistical
distributions.
We will provide a review of the geometry-based
models under three scenarios: fix-to-mobile (F2M),
mobile-to-mobile (M2M), and air-to-ground (A2G). The
geometry-based models discussed here are summarized
in Table 1.
454
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
1) F2M COMMUNICATION SYSTEMS
In F2M system, the mobility is on user’s end. The BS is
stationary, elevated, and relatively free of local scattering.
HSR channel model belongs to this scenario.
GBDM: Most GBDMs are constructed through
3-D ray-tracing, which tracks the wave propagation in
various environments. In [34], a 3-D ray-tracing model for
HSR tunnels is developed. The model is represented in the
form of polarimetric and complex CIRs and the Doppler dia-
gram, and the results agree well with measurements results.
A GBDM for multiple input single output (MISO) system is
proposed in [35], which analyzes the channel characteristics
such as the delay spread and Doppler spread. The model
shows that the Doppler spread can be reduced by using direc-
tional antennas. GBDM can also be developed by using prop-
agation graph, which predicts CIRs by thoroughly searching
for propagation paths. A GBDM based on random propaga-
tion graph is proposed in [36] to characterize time varying
HSR channels in the rural, urban, and suburban environments.
This modeling approach can be performed by considering the
geometry of the simulated environment, e.g., the distribution,
mobility, and visibility of the scatterers. To avoid the high
complexity of implementing GBDM while maintaining suf-
ficient accuracy, semi-deterministic models for viaduct and
cutting scenarios of HSR is proposed in [37].
GBSM: Due to its flexibility, GBSM has been widely
used for standard channel models including COST 259,
COST 273, Spatial Channel Model (SCM) and WINNER
channel models [38]. In [38], a WSSUS one-ring GBSM is
proposed to model HSR channels in urban scenarios. The
scatterers are assumed to be randomly distributed in a ring
around the MS and different distributions of the scatterers are
analyzed. An empirical geometry-based model is proposed
in [39] for HSR systems. A non-stationary ellipse channel
model is proposed in [40] for wideband MIMO HSR channels
in rural scenarios. Fig. 1 shows the geometry of non-
stationary ellipse channel model in [40]. All the previously
reported models for HSR assume that the fields incident on
the BS and MS antennas are composed of a number of waves
traveling only in the horizontal plane. This assumption is
VOLUME 4, 2016
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
FIGURE 1. The geometry of non-stationary ellipse channel model in [40].
valid only for plain or viaduct. In [41], a 3-D one-sphere
GBMS is proposed for the terrain cutting, suburb, and urban
scenarios.
2) M2M COMMUNICATION SYSTEMS
In M2M communications, both the transmitter and the
receiver are in motion and are equipped with low eleva-
tion antennas. Applications include mobile ad-hoc wireless
networks, intelligent transportation systems, and relay-based
cellular networks.
GBDM: A sum-of-sinusoid based simulation model for
M2M channels is proposed in [42], where the model is
obtained by summing over sinusoidal signals with random
amplitudes and phases propagated over deterministic
scatters placed on two-ring centering around the transmitter
and receiver, respectively.
GBSM: A RS-GBSM for narrowband M2M Rayleigh
fading channels is proposed in [43], where the model is
developed by using a 2-D two-ring geometry with double-
bounced rays and isotropic scattering. A RS-GBSM for
Ricean fading channel is proposed in [44]. The effects of non-
isotropic scattering are considered in [45]. A 3-D narrowband
two-cylinder RS-GBSM is proposed in [46], and it is
further extended to a 3-D wideband cocentric-cylinder
RS-GBSM in [47]. Fig. 2 shows the geometry of the
two cocentric-cylinders model in [47]. All aforementioned
FIGURE 2. Two concentric-cylinders model presented in [47].
VOLUME 4, 2016
RS-GBSMs cannot characterize the impact of the vehicular
traffic density (VTD) on channel statistics, which might
have significant impacts on channel statistics in vehicular
networks. The effects of VTD are incorporated in a 2-D
non-isotropic scattering wideband MIMO M2M RS-GBSM
in [48], and a 3-D RS-GBSM model in [49]. A 3-D polarized
MIMO channel model is proposed in [50], which takes into
account 3-D power angular spectrum and comprehensive
propagation characteristics of electromagnetic waves excited
by polarized antennas. A RS-GBSM model for wideband
M2M systems are proposed in [51].
3) A2G COMMUNICATION SYSTEMS
A2G channel is the communication link between an air-
craft aloft in the troposphere and a ground station (GS).
It is assumed that the aircraft is within the radio line of
sight (RLOS) of the GS, because curvature of the earth makes
link closure essentially impractical beyond the RLOS [52].
GBDM: In [53], a method of real-time A2G aeronautical
communications channel modeling is proposed with
ray-tracing. By using the ray-tracing method, the model
can directly produce the real-time simulation results of the
channel. The results can be further employed to build up the
statistical model.
GBSM: A 3-D geometric ellipsoidal channel model is
developed for simulating A2G communications where the
GS is in a cluttered environment [54].
C. CONCLUDING REMARKS
The high mobility channel models are reviewed in this section
by following two categories: statistical channel models
and geometry-based models. Statistical models abstract the
behaviors of high mobility channels into a group of param-
eters. They are relatively simple and analytically tractable.
Thus the statistical models are widely used in system analysis
and design, such as channel estimations and diversity trans-
missions. However, the simple structure of statistical models
might not capture the complicated behaviors of high mobility
channels. In addition, most statistical models are WSS, yet the
high mobility channel is usually non-WSS. In order to apply
statistical models to high mobility systems, we need to divide
the time axis into small intervals such that the channel in one
interval is WSS. The geometry-based models can provide a
more accurate representation of the high mobility propaga-
tion environments, at the cost of higher complexity. Different
geometry-based models can be used to approximate signal
propagations in a wide range of environments, with different
geometry layouts and scatter distributions. In addition, they
can easily capture the non-WSS behaviors of high mobility
channels. They are mainly used for simulations, but rarely
used for design and analysis, due to the lack of analytical
tractability and high complexity. Analysis and simulation
results with the geometry-based models can be used to con-
struct new or to improve existing statistical models.
The future research directions in high mobility chan-
nel modeling include the following aspects. First, statistical
455

properties of high mobility channels can be better estimated
and understood with experimental and simulation results, in
various communication scenarios. Second, with the help of
the statistical properties, more sophisticated statistical mod-
els can be constructed to better represent the propagation
properties of high mobility channels. Third, hybrid and adap-
tive channel models can be developed to represent more
propagation environments. Fourth, for more accurate channel
description, scenario-specific geometry-based models should
be investigated, such as viaducts, cuttings, tunnels and sound
insulation walls built along the railways and highways.
III. ESTIMATIONS OF FAST TIME-VARYING
FADING CHANNELS
High mobility communications dictate an operation
environment of doubly-selective (both time-selective and
frequency-selective) fading, which introduces channel
dispersions in both time and frequency domains. The fast
time-variation along with the long time-dispersion of doubly-
selective fading impose great challenges on the accurate
estimation and tracking of the doubly-selective channel.
The channel estimation methods are typically classified
as blind [56], [57], semi-blind [58], and pilot-aided channel
estimations [59], [60]. The blind channel estimations are
realized by using the statistics of the received data [56], the
cyclic prefix (CP) [61], the virtual carriers [62], and receiver
diversity [57]. Due to the low convergence speed, blind esti-
mations often require the channel to be constant for a large
number of blocks, therefore they are not suitable for high
mobility applications. Most communication standards such
as LTE and Digital Video Broadcasting (DVB) adopt pilot
assisted channel estimations. Thus in this survey, we focus
on the study of pilot-aided channel estimation algorithms.
Pilot-aided channel estimations include both pilot designs
and estimation algorithms. In high mobility channel estima-
tions, each aspect has unique features and challenges due
to the fast time-variation and long time-dispersion of the
channels.
A. OPTIMUM PILOT DESIGN
In pilot-aided channel estimations, pilots known to both
transmitters and receivers are embedded in the transmitted
information sequence. Based on the Nyquist sampling theo-
rem, the pilots should sample the channel at a rate no less than
twice the maximum Doppler spread, i.e., T1p ≥ 2fD , where
Tp is the time interval between two adjacent pilots [63]. Pilot
designs can be performed with respect to several parameters,
such as the pilot pattern, the number and percentage of pilots,
pilot power, and pilot constellations, etc., with respect to
various design metrics.
1) OPTIMUM PILOT DESIGN IN NARROW-BAND SYSTEMS
A commonly used pilot design metric is to minimize the
channel estimation MSE. Quasi-periodic placement of pilot
symbols is considered to be optimum in terms of minimizing
estimation MSE in fast time varying fading [64], [65].
456
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
Based on the equally-spaced pilots, the MSE of minimum
mean squared error (MMSE) channel estimation is stud-
ied through asymptotic analysis in [63], where the channel
estimation MSE is expressed as closed-form expressions of
parameters such as the Doppler spread and the percentage
of pilot symbols. As expected, the MSE is a monotonically
decreasing function in pilot percentage, which means more
pilots provide a more accurate channel estimation result.
Besides optimum pilot percentage, power allocation between
data and pilot symbols provides extra performance gains
to channel estimations [66], [67]. More energy to the pilot
symbols leads to a lower channel estimation MSE [63], [66].
Lower MSE does not necessarily mean a better overall
performance due to the energy and spectral overhead caused
by the pilot symbols. This motivates pilot designs with other
performance metrics, such as the error probability and the
spectral efficiency. In [66] and [68], the analytical symbol
error rate (SER) of high mobility systems operating in
the presence of channel estimation errors is derived. The
results reveal that under a total power constraint, allocating
excessive power to pilot symbols will degrade SER even
though the MSE is monotonically decreasing in pilot power.
The pilot designs are performed with respect to spectral
efficiency by considering both pilot percentage and power
allocation [65], [69], [70]. It is shown in [70] that the optimum
pilot percentage that maximizes spectral efficiency is the
Nyquist rate of the channel.
2) OPTIMUM PILOT DESIGN IN BROADBAND SYSTEMS
For doubly selective fading channels, intersymbol interfer-
ence (ISI) in the time domain or intercarrier interference (ICI)
in the frequency domain imposes additional challenges to
optimum pilot designs [71]. The key idea of pilot designs in
broadband systems is to maintain the orthogonality between
pilot and data symbols at the channel output [72]. The
orthogonality condition can be met with three pilot patterns:
time-domain Kronecker delta (TDKD) family, frequency-
domain Kronecker delta (FDKD) family, and linear chirp
sequences. In TDKD, pilots are clustered in the time domain
with guard symbols to avoid ISI. In FDKD, pilots are clus-
tered in the frequency domain with guard subcarriers on
adjacent bands to avoid ICI. The TDKD and FDKD schemes
are shown in Fig. 3.
With TDKD and FDKD, it is shown in [73] that equally
spaced pilot clusters are optimum in terms of channel
estimation MSE in doubly selective channels. The TDKD
pilot structure is used to maximize the average channel
capacity [74] or minimize MSE [75]. It is shown in [74] that
maximizing capacity or minimizing MSE results in the same
optimum training sequence.
In OFDM, comb-type pilot patterns can be placed on the
2-D time-frequency grid to cope with the fast channel varia-
tions. As shown in Fig. 4, one pilot symbol can be placed over
every Dx sub-carriers and Dy OFDM symbols. The values
of Dx and Dy depend on the coherence time and coherence
bandwidth of the channel. The location, number, and power
VOLUME 4, 2016
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
FIGURE 3. TDKD and FDKD pilot clusters.
FIGURE 4. Scattered pilots (a) Dx = 4, Dy = 1, (b) Dx = 2, Dy = 2.
of pilot symbols embedded in OFDM systems over rapidly
fading channels are optimized to minimize MSE or maximize
channel capacity in [76], where the channel is assumed to be
time-invariant in one OFDM symbol, and time varying from
symbol-to-symbol.
In high mobility OFDM systems, the channel varies within
one OFDM symbol and this will cause strong ICI, which is
the frequency-domain dual of ISI. Thus many time domain
designs can be mirrored into the frequency domain designs
in OFDM. Based on the orthogonality condition in [72],
the FDKD pilots are considered for frequency domain
pilot designs [59], [77], [78]. For FDKD pilots, the pilot
sub-carriers are grouped as clusters, with one active pilot
surrounded by guard sub-carriers to reduce or remove ICI.
The cluster size is determined by the maximum Doppler
spread [72], [78]. To increase the spectral efficiency, based
on the clustered comb pilots, pilot clusters consisting of more
than one non-zeros pilots are proposed in [79]. The optimum
length of the active pilots and the length of the guard pilots are
analyzed for different Doppler spreads with a BEM channel
model [59].
B. PILOT-AIDED CHANNEL ESTIMATION ALGORITHMS
Channel estimations can be performed either through the
direct estimation of the fading coefficients, or through dimen-
sion reduction techniques that transform fading coefficients
to a lower dimensional domain. In high mobility systems,
the number of channel coefficients to be estimated increases
dramatically due to fast channel variation, and this necessities
the development of dimension reduction techniques. In this
subsection, we will summarize various channel estimation
VOLUME 4, 2016
algorithms that utilize low dimensional channel representa-
tion techniques.
1) LEAST SQUARES ESTIMATION
The least squares (LS) channel estimation does not require
knowledge of priori channel statistics, and it is widely
adopted in practical systems for its simplicity. In high mobil-
ity channel estimations, LS combining BEM is adopted
to estimate the channel coefficients in a low dimensional
domain. A windowed LS (WLS) estimator with BEM is pro-
posed in [79] and it shows a good estimation accuracy at the
high SNR range. In OFDM systems, the channel coefficients
can be either estimated in the time domain, or in the frequency
domain, or the combination of the two. In [80], the frequency
domain channel coefficients of pilot sub-carriers are first
estimated, and they are then transformed to the time domain
to further reduce the noise by selecting the significant taps.
The combination of estimations in the time and frequency
domains renders a better resistant to noise.
2) LINEAR MMSE (LMMSE) ESTIMATION
The LMMSE (or Wiener filter) utilizes the prior statistics
of the channel fading during the estimation process, and it
usually results in a better performance than the LS algorithms,
at the cost of a higher complexity.
When the second-order statistics of the channel
coefficients are known perfectly, the exact MSE of LMMSE
channel estimation is expressed as a closed-form function of a
number of system parameters, such as the maximum Doppler
spread, the percentage of pilot symbols, and the power
allocation between pilot and data symbols [63], [66], [68].
It is discovered that, if the pilots sample the channel at a
rate no less than its Nyquist rate, then the estimation of the
channel coefficients of the data symbols through temporal
interpolation yields the same MSE as the direct estimation of
the channel coefficients of pilot symbols.
The complexity of LMMSE estimation is mainly caused by
the inverse of the channel covariance matrix, which is usually
large in a high mobility environment. The complexity can be
reduced by using a low rank approximation of the channel
covariance matrix. Specifically, the covariance matrix can
be approximated by identifying its dominate subspace with
the help of eigenvalue decomposition, and eigenvalues less
than a threshold can be ignored to achieve a low rank
approximation [81]–[84]. Since the channel statistics changes
much slower than the channel coefficients, the complexity of
LMMSE estimation can be reduced significantly.
An alternative way for dimension reduction is to apply
LMMSE with BEM to reduce the number of parameters
to be estimated. In [59], [73], [77], [85], and [86], the
LMMSE is performed over the BEM of the channel coef-
ficients in the time domain. For a doubly selective fading
channel with L taps, the number of BEM coefficients to
be estimated is 2dfD NTs eL, where N is the number of time
domain samples and Ts is the sample period. The number
of BEM coefficients is usually much less than the number
457

of time domain samples NL. The BEM can also be applied
to frequency domain channel coefficients, and it results in a
strictly banded frequency domain channel matrix, where the
width of the band is determined by the maximum Doppler
shift [59], [77], [85], [86].
LMMSE channel estimation requires the knowledge of
the second-order statistics of both channel and noise, and
this information is usually unavailable at the receiver. This
problem can be partly addressed by using the so-called robust
channel estimators, which use a generic reference model
for the channel statistics. A uniform power delay profile is
assumed in [83]. In [87], a rectangular power delay profile
is used, and the model mismatch errors between the
reference and true channel models are analyzed in [87].
A delay-Doppler scattering function prototype with flat spec-
trum in a 2-D region is used in [82]. The above reference
channel models are insensitive to variations in channel statis-
tics and are relatively simple to implement.
3) RECURSIVE ESTIMATION
Instead of using the priori information of channel statistics
explicitly, adaptive filters can follow the channel statistics and
adapt their coefficients to track the channel characteristics.
The statistic tracking capability of adaptive filters is espe-
cially important to high mobility channels, which are usually
non-stationary and with time-varying channel statistics.
Recursive least-squares (RLS) filters can be used to track
the time-varying statistics of high mobility systems [60], [88].
RLS-based filters employ three basic iterative steps:
initializing the filter parameters, computing estimation errors,
and updating the filter coefficients. A 2-D RLS filter
is proposed in [89] by tracking the channel statistics of
doubly-selective fading channels in both the time and fre-
quency domains.
Another commonly used recursive estimation algorithm is
the Kalman filter, which can provide accurate tracking of
the dynamic high mobility channels. A Kalman filter is a
recursive algorithm consisting of a predictor and a corrector
stage in each iteration. In the predictor stage, the current
state, i.e., the CSI and the error covariance are estimated
based on an evolution model, which is usually the first order
AR model as in (4). The output of the predictor stage is used
as priori inputs to the corrector stage. In the corrector stage,
the priori input is combined with a new measurement, i.e., the
observed pilot, to obtain an improved a posteriori estimate of
the channel. The measurement could be either from the time
domain, the frequency domain, or both.
In high mobility systems, the state model in (4) describes
a symbol-to-symbol variation of the channel coefficients.
In this case, the Kalman filter cannot be directly applied
because it will require a pilot symbol for each state update.
This problem can be solved by using the BEM model, where
the BEM coefficients are time-invariant in a block, but evolve
from block to block. Tracking the BEM coefficients rather
than the channel tap gains makes the symbol level variation
becomes block level.
458
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
In summary, the RLS algorithm is a special case of
the Kalman filter for a stationary state-space model. The
application of Kalman filter relies on the prior knowledge of
the channel covariance. Hence, the Kalman filter approach
can provide better tracking properties in fast-fading chan-
nels with the dynamic state-space model. However, both the
RLS algorithm and the Kalman based algorithms are compu-
tation intensive due to the slow recursive procedure.
C. CONCLUDING REMARKS
The estimation and tracking of high mobility channel are
discussed in this section by considering both optimum
pilot designs and various channel estimation algorithms.
The unique challenges of high mobility channel estimations
are dealt with from the following three aspects. First, low
dimensional channel representation, such as BEM or
eigenspace approximation, can significantly reduce the
channel estimation complexity. Second, adaptive channel
estimation algorithms can track the time-varying statistics
of high mobility channels, and provide accurate estimation
and tracking of non-stationary channels. Third, the effects
of ICI caused by fast time-varying fading can be efficiently
suppressed by employing pilot patterns with guard intervals.
The common channel estimation algorithms are summarized
in Table 2.
The future research on high mobility channel estimation
includes the following aspects. First, how to balance the
tradeoff between the pilot overhead and the estimation per-
formance in non-stationary high mobility systems remains
an important problem. Both theoretical results and practi-
cal designs are needed to achieve the optimum overhead-
performance tradeoff by taking into considerations of the
non-stationarity of the high mobility channels. Second, the-
oretical guidelines are needed regarding the optimality of
pilot placements and channel estimation in various domains.
Channel estimations can be performed in the time domain,
frequency domain, or other transform domains. In highly
dispersive high mobility channels, there is severe ISI in the
time domain and strong ICI in the frequency domain. Thus
the choice of the domain to perform channel estimation
should take into considerations of various parameters, such as
delay spread, Doppler spread, and the estimation complexity,
etc. Third, low complexity algorithms are needed to provide
accurate tracking of non-stationary high mobility channels.
Existing adaptive algorithms usually have high computa-
tional complexities due to their iterative nature.
IV. DOPPLER DIVERSITY TRANSMISSIONS
In high mobility systems, the fast time variation of the fading
channel can be utilized to provide time diversity or Doppler
diversity, given that signals are dispersed in the frequency
domain due to the Doppler effects. The Doppler effects
provide an additional degree-of-freedom that can provide sig-
nificant performance gains to high mobility communications.
In this section, we will provide an overview of
Doppler diversity transmissions by seeking answers to the
VOLUME 4, 2016
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
TABLE 2. Channel estimation algorithms.
following two questions: what is the maximum Doppler
diversity achievable in a high mobility system? And how
to design practical high mobility communications that can
achieve the maximum Doppler diversity transmissions? Both
questions will be answered by considering two scenarios:
systems with perfect CSI, and systems with imperfect CSI.
In high mobility systems, due to the fast time variation of
the fading channel, channel estimation errors are usually
non-negligible and might have significant impacts on system
performance and design. Therefore we will study the impacts
of imperfect CSI on the design and performance of Doppler
diversity systems.
A. FUNDAMENTAL PERFORMANCE LIMITS
In this section we study the fundamental performance limits
of Doppler diversity transmissions, for systems with or with-
out perfect CSI.
1) DOPPLER DIVERSITY WITH PERFECT CSI
The diversity order of a wireless communication system is
defined as
log P(E)
D = lim , (5)
γ →∞ log γ
where γ is the SNR and P(E) is the error probability. With
the above definition, it can be shown that the diversity order
of a system with correlated channel coefficients equals to
the rank of the channel covariance matrix. Specifically, the
Doppler diversity order is the rank of the covariance matrix
of the time-varying channel coefficients, and the covariance
matrix is determined by the maximum Doppler spread fD , the
sampling period Ts , and the number of samples N .
In [90], the time-varying frequency flat fading chan-
nel is represented by using a BEM model with order
Q = 2dNfD Ts e. The corresponding Doppler diversity order
is Q + 1. The above results are extended to doubly selective
fading in [91], where the maximum Diversity order (including
both Doppler diversity and multipath diversity) is shown to
VOLUME 4, 2016
be (Q+1)(L+1), where L is the order of basis in the multipath
domain.
In a MIMO system with Nt transmit and Nr receive anten-
nas, the maximum diversity order of the system with flat
fading is shown to be Nt Nr (Q + 1), the product of the space
diversity and Doppler diversity [92]. For a MIMO system that
experiences doubly selective fading, the maximum diversity
order is Nt Nr (Q + 1)(L + 1) [93]. That is, full diversity in
space, time, and multipath domains can be achieved simul-
taneously with properly designed precoders. It should be
noted that the Doppler diversity orders in all above works are
represented as integers due to the limit of the BEM model,
the order of which has to be an integer.
2) DOPPLER DIVERSITY WITH IMPERFECT CSI
The presence of channel estimation errors might degrade
the achievable Doppler diversity in high mobility systems.
There is a fundamental tradeoff between maximum achiev-
able Doppler diversity order and channel estimation errors in
high mobility systems. A larger fD means a higher potential
Doppler diversity, but also a larger channel estimation MSE,
which might degrade the achievable Doppler diversity. The
impacts of both channel estimation errors and Doppler diver-
sity for code division multiple access (CDMA) systems are
studied in [94], where an approximate of the error probability
is expressed as a function of the channel estimation MSE.
The fundamental tradeoff between Doppler diversity and
channel estimation errors is identified in [95] by performing
asymptotic analysis of a pilot assisted transmission system.
Since the Doppler diversity order is linearly proportional
to the time duration, it is normalized with respect to time.
Let γp and γd denote the SNR of pilot and data symbols,
ξ
respectively. If γp = O(γd ) and the pilot samples the channel
at its Nyquist rate, the maximum Doppler diversity order in
unit time is [95]
1 log P(E)
Dnorm = lim = 2fD min(ξ, 1/ξ ). (6)
γ →∞ NTs log γ
N →∞
459

The result indicates that, if the energy of the pilot and
data symbols scales linearly with each other, that is, ξ = 1,
then the normalized Doppler diversity order per unit time
is 2fD , which is the same as that of systems with perfect
CSI. Therefore, systems with imperfect CSI can achieve
the same Doppler diversity order with systems with perfect
CSI, and the presence of channel estimation errors will not
degrade Doppler diversity. On the other hand, when ξ 6 = 1,
the Doppler diversity order of system with imperfect CSI is
strictly worse than that of system with perfect CSI.
Specifically, when ξ < 1, there is not enough energy to
pilot symbols, and the performance is dominated by channel
estimation errors. When ξ > 1, too much energy is allocated
to pilot symbols, and the performance degradation is due to
energy overhead from pilot symbols.
Even though systems with imperfect CSI can achieve the
same diversity order as systems with perfect CSI, there is
always a non-diminishing gap between the two systems in
terms of error probability, and this gap can be quantified by
using the SNR difference between the two systems under
the same error probability. Consider two systems, one with
perfect CSI and the other with imperfect CSI. If the two
systems have the same Doppler diversity order and the same
error probability, then the minimum SNR gap between the
two systems is [95]
s !
2fD Ts
20 log10 1 + (dB), (7)
1 − 2fD Ts
where Ts is the symbolqperiod. This minimum SNR gap is
achieved when γp = γd 2f 1Ts − 1. The gap is an increasing
D
function in fD , which means a higher Doppler spread results
in a larger SNR loss for systems with imperfect CSI. This is
due to the fact that more pilot overhead is needed for systems
with larger Doppler spread. The SNR gap will be 0 only in
quasi-static channels with fD = 0.
Fig. 5 shows the diversity order and SNR gap for various
system configurations. The results are obtained with simple
FIGURE 5. SER as a function of γ0 for various values of fD Ts in the high
SNR regime [95].
460
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
repetition code. The bit error rate (BER) curve obtained with
ξ = 1 has the same slope as the one with perfect CSI,
thus the same diversity order. The BER curves obtained with
ξ = 2 or ξ = 0.5 have the same Doppler diversity order,
which is about half of that with ξ = 1 as predicted by (6).
In addition, even though the BER with ξ = 1 has the same
slope (thus diversity order) as the perfect CSI case, there is
always a gap between the two curves, and the gap increases as
fD Ts increases.
The results in (6) and (7) reveal the fundamental perfor-
mance limits of Doppler diversity systems operating in the
presence of imperfect CSI, and they can be used to guide the
design of practical systems.
B. PRACTICAL DOPPLER DIVERSITY TRANSMISSIONS
This section reviews the designs of practical Doppler
diversity transmission schemes with or without perfect CSI.
The basic idea of Doppler diversity transmission is to
spread one information symbol over a time duration longer
than the channel coherence time, such that the symbol
experiences independent fading in the time domain. This
can be achieved by performing precoding at the transmitter.
In addition, the symbol period in multi-carrier systems is
in general much longer than single-carrier systems with the
same symbol rate. If the symbol period is longer than the
channel coherence time, Doppler diversity can be achieved
by multi-carrier system without precoding. We will focus on
Doppler diversity transmissions in single-carrier system in
this section. The design and performance of multi-carrier high
mobility systems will be discussed in Section V.
1) PRACTICAL DOPPLER DIVERSITY TRANSMISSIONS
WITH PERFECT CSI
The maximum Doppler diversity order can be achieved
by performing linear block precoding. The design criteria
for linear precoders that can achieve the maximum Doppler
diversity order is presented in [90]. It is shown that under
a complex exponential BEM model, a linear precoder
2 ∈ C N ×K can achieve the maximum Doppler diversity
if and only if there exist at least Q + 1 non-zero entries in
2e, ∀e 6 = 0, where Q = 2dfD NTs e is the order of the BEM
model, and K and N are the block length before and after
precoding, respectively.
Based on the above design criteria, several precoders are
proposed in [90]. The tall Vandermonde precoder can achieve
the maximum Doppler diversity for any modulation schemes,
at the cost of some spectral efficiency loss because the cod-
K
ing rate is N < 1. A class of rate-1 square Vandermonde
precoders with K = N can achieve the maximum Doppler
diversity for QAM constellations. A rate-1 Doppler domain
multiplexing (DDM) scheme proposed in [96] can approach
the performance of a maximum Diversity system, and it is
constellation independent.
The above precoders are developed for SISO time-varying
flat fading channel. The precoder designs for system with
doubly selective fading are discussed in [91] and [97],
VOLUME 4, 2016
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
where the Doppler and frequency diversity can be
obtained simultaneously. Space-time-Doppler block coding
are proposed for MIMO flat fading channel to achieve both
space diversity and time diversity [93].
2) PRACTICAL DOPPLER DIVERSITY TRANSMISSIONS
WITH IMPERFECT CSI
There are limited works devoted to the design of Doppler
diversity transmissions with imperfect CSI. The performance
of Doppler diversity transmission with imperfect CSI is stud-
ied in [94], where the Doppler diversity is achieved by means
of spreading spectrum operations. In [98], the maximum
Doppler diversity transmission in doubly selective fading
is discussed with a semiblind scheme, where the receiver
does not have the knowledge of the CSI. The signal at the
receiver is decoded with a decision feedback equalizer (DFE),
and the coefficients of the DFE and the information symbol
estimates are refined recursively by using normalized least-
mean-squares (NLMS) adaptation.
It has been shown in [99] that the presence of channel
estimation errors might destroy the optimality of transceivers
designed for systems with perfect CSI. Therefore new
transceivers need to be developed by considering channel
estimation errors. The statistical properties of channel estima-
tion errors are explicitly considered in the optimum designs
of several Doppler diversity systems [100]–[102]. In [100],
the maximum Doppler diversity is achieved with a simple
repetition code at the transmitter and an optimum combin-
ing receiver that considers the effects of channel estimation
errors. The spectral efficiency of the repetition code is
low. A spectral efficient Doppler diversity system with
imperfect CSI is developed in [101] and [102]. Optimum
and low complexity sub-optimum receivers are developed
in [101] and [102] by considering channel estimation errors.
The error probability of the system are expressed as explicit
functions of channel estimation errors and Doppler shift,
and they reveal the fundamental tradeoffs between Doppler
diversity and channel estimation errors.
C. CONCLUDING REMARKS
Doppler diversity transmissions in high mobility systems
are reviewed in this section from two aspects: fundamental
performance limits and practical designs of Doppler diver-
sity systems. It has been shown that there is a fundamental
tradeoff between channel estimation errors and Doppler
diversity order, given that fast time-varying fading means
a higher Doppler diversity order, but it is more difficult to
estimate and track the channel in high mobility scenarios.
With properly designed pilot symbols, systems with imper-
fect CSI can achieve the same Doppler diversity order as
those with perfect CSI, but there is always a non-diminishing
SNR gap between systems with and without perfect CSI.
The maximum Doppler diversity can be obtained in practical
systems by designing various linear precoders and optimum
receivers that take into considerations of channel estimation
errors.
VOLUME 4, 2016
One important future research direction is the
development of Doppler diversity transmissions for multi-
user multi-carrier multi-antenna systems. In multi-carrier
systems, fast time-varying fading introduces both Doppler
diversity and ICI. In multi-user multi-antenna systems, the
precoder designs usually require the knowledge of the CSI
at the transmitter (CSIT), which is a daunting task for
high mobility systems. In addition, there will be multi-
user interference (MUI) and multiplexing interference (MI)
among signals from different antennas. The Doppler diversity
transmissions need to be developed in the presence of ICI,
MUI, MI, and taking into consideration of channel estimation
errors at the receiver and channel aging at the transmitter.
The Doppler diversity transmission schemes should be devel-
oped to balance the tradeoff among a number of factors,
such as energy efficiency, spectral efficiency, imperfect CSI,
interference, and system complexity, etc.
V. MULTI-CARRIER COMMUNICATIONS
IN HIGH MOBILITY SYSTEMS
Multi-carrier modulation (MCM) has been used by a
large number of current and next generation wireless
communication standards due to its robustness against fre-
quency selective fading. The most popular MCM scheme
is OFDM, which modulates multiple narrow-band data
streams onto orthogonal subcarriers. With the insertion of CP,
OFDM system can completely remove the effects of ISI.
Conventional OFDM systems are originally designed for
quasi-static block fading channel, where the channel remains
constant over an entire OFDM symbol period.
In high mobility systems, the quasi-static fading assump-
tion is no longer valid, and the fading coefficients might
change inside one OFDM symbol. The large Doppler shift
and fast fading variation will destroy the orthogonality among
the sub-carriers, thus introduce ICI. In high mobility OFDM
systems, ICI can be introduced by two possible sources. The
first source is the carrier frequency offset (CFO) caused by
time-varying Doppler shifts and oscillator instability. The
second source is the time-variation of the fading coefficients
inside one OFDM symbol. We will discuss the effects of
both CFO and fast fading variation, and the corresponding
countermeasures.
A. CARRIER FREQUENCY OFFSETS
In high mobility systems, the carrier frequency of the received
signal will be shifted from its original value due to the
effects of Doppler shifts, which usually change with respect
to time. If the receiver cannot accurately estimate the time-
varying Doppler shifts, then there will be a mismatch between
the carrier frequency of the received signal and the carrier
frequency generated by the local oscillator, and this causes
CFO. CFO can also be caused by the instability of the local
oscillators at the transmitter and/or receiver. CFO will destroy
the orthogonality in OFDM systems, and introduce ICI. Thus
it is important for the receiver to accurately estimate and
compensate the effects of CFO.
461

Consider an OFDM system with N subcarriers. Given
frequency offset ν, the signal received on the k-th subcarrier
can be written as [103]–[106]
Yk = Hkν Xk + Ik + Wk , (8)
where Yk , Xk , and Wk are the received sample, transmit-
ted symbol, and AWGN in the k-th subcarrier, respectively,
Hkν = NHsin(πν/N
k sin(πν)
)e
jπν(N −1)/N with H being the frequency-
k coarse estimates to counteract the residual CFOs or short-
domain channel coefficient at the k-th subcarrier. The
ICI term Ik can be written as
X sin(π ν)ejπν(N −1)/N
Ik = Hu X u e−jπ(u−k)/N . (9)
N sin (π (u − k + ν)/N )
u6=k
The power of ICI increases as the CFO increases, and it is
clear that Ik = 0 if ν = 0. The CFO can be estimated and
compensated at the receiver, thus ICI caused by CFO can
be removed with accurate CFO estimation. Below we review
several CFO estimation methods.
1) BLIND CFO ESTIMATIONS
Blind CFO estimations are performed by utilizing the special
structure of the transmitted signals [107]. They do not need
redundancies such as training sequences, guard intervals, or
pilots. A minimum output variance (MOV) CFO estimator
is proposed in [108], where the CFO is estimated and com-
pensated based on the fact that the variance of the signal
amplitudes across different subcarriers should be very small
when the CFO is properly compensated. A similar approach
is used in [109], where the CFO is estimated and compensated
by using the variance of the signal power on different subcar-
riers. In [110], the CFO is estimated by minimizing the power
difference between all subcarriers in two consecutive OFDM
symbols. A decision-directed (DD) CFO estimation method
is proposed in [111], and it is shown that the DD scheme
outperforms data-aided schemes in terms of fine CFO traking.
Both [110] and [111] are developed under the assumption that
the channel remains constant in two consecutive OFDM sym-
bols. In high mobility systems, the CFO might change from
symbol to symbol. A subspace-based blind CFO estimation
algorithm is proposed in [112]. The subspace-based method
needs only one OFDM symbol to achieve CFO estimation,
thus it can efficiently track the fast time variation of the CFO.
2) REDUNDANCY-BASED CFO ESTIMATION
The CFO can be estimated by using redundant information
embedded in OFDM symbols. The redundant information can
be preambles in communication standards, specially designed
CP [113]–[115], pilot symbols and training sequences, or null
subcarriers that do not transmit any information [116]–[118].
Most CFO estimations are performed in two stages:
a coarse acquisition stage performed before the fast Fourier
transform (FFT) at the receiver, and a fine tracking stage
performed in the frequency domain after the FFT. Preambles
and CP are usually used by the pre-FFT CFO acquisition
stage, while pilots and null subcarriers are usually used by
the post-FFT CFO tracking stage [103], [119]–[124].
462
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
The two-stage acquisition and tracking procedure
can achieve rapid and accurate CFO estimation and
compensation [119]. During the acquisition stage, a coarse
estimate of the CFO is usually obtained by using ML esti-
mation [119], [125]. The CFO may initially be large. After
the acquisition stage, only a small residual CFO remains.
During the tracking stage, pilots are used to refine the
term CFO variations due to time-varying Doppler shifts. The
tracking stage is usually performed in the frequency domain,
where the residual CFO is estimated by measuring the phase
difference between two consecutive OFDM symbols. The
phase difference based approach is simple and effective
and widely implemented in practical systems [119]. The
estimation accuracy can be further improved by computing
the weighted average of the estimated CFO over all the pilot
subcarriers [126]. All above frequency domain CFO tracking
methods require the channel remain constant during two
consecutive OFDM symbols.
The CFO is often jointly estimated with other parameters,
such as phase noise and CSI [118], [127]–[131]. In [118],
a joint CFO, phase noise, and channel transfer function (CTF)
estimation algorithm is presented. The integer part of the
CFO is estimated by using frequency domain samples, and
the fractional part of the CFO is estimated by using the phase
difference of the time domain training symbols. In [128], an
optimum maximum a posteriori (MAP) estimator is proposed
to jointly estimate the CSI, CFO, and phase noise, by utilizing
the prior statistical knowledge of the phase noise. An iterative
method is proposed in [129] to jointly estimate the doubly
selective channel and the CFO within the framework of
expectation-maximization (EM). The approach is developed
with the BEM channel model, and it is designed specifically
for high mobility systems. Recently, recursive algorithms,
such as RLS filter [132], recursive ML estimator [133], and
Kalman filter [134], are employed to achieve more accurate
CFO estimations in fast time-varying fading channels.
B. INTERCARRIER INTERFERENCE
ICI is one of the main performance limiting factors for
OFDM systems operating in high mobility scenarios.
ICI caused by CFO can be compensated by means of CFO
estimation as described in the previous subsection. Another
major source of ICI in high mobility systems is the time
variation of the fading inside one OFDM symbol.
1) ICI DUE TO HIGH MOBILITY
The received OFDM signal in the frequency domain can be
represented as
y = Hx + w, (10)
where y = [y0 , · · · , yN −1 x = [x0 , · · · , xN −1
]T , and ]T ,
w = [w0 , · · · , wN −1 ]T are the received sample vector, data
vector, and AWGN vector, respectively, all in the frequency
domain, and H is the frequency domain channel matrix. For
doubly selective fading, the (i, k)-th element of H is the
VOLUME 4, 2016
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
 
N , N
i−k k−1
Doppler-frequency transfer function, g
as [135]
N −1 L−1
1 XX
g(d, f ) = h(n, l)e−j2πlf e−j2πnd
N large N . Cai and Giannakis [136] develop a matched-filter
n=0 l=0
where h(n, l) is the discrete-time CIR of a doubly selective
fading, d is the frequency variable in the Doppler domain,
and f is the variable in the frequency domain.
It is clear that for quasi-static block fading
 h(0, l) = · · · =
N , N
i−k k−1
h(N − 1, l), the off-diagonal elements, g
∀i 6 = k. On the other hand, for doubly selective fading, the
off-diagonal elements are not zero, and they introduce ICI.
From (11), the off-diagonal elements of H are contributed
by the Doppler shifts, and the main diagonal elements cor-
respond to a Doppler shift of 0. Due to the limit of the
maximum Doppler shift, the matrix H is a banded matrix,
with the indices of the non-zero diagonals between −dNfD Ts e
and dNfD Ts e. The structure of H with N = 128 under
different values of dNfD Ts e are shown in Fig. 6, where the
amplitudes of the elements in H are mapped to gray scales.
FIGURE 6. Banded structure of the frequency domain channel matrix H
(N = 128).
The off-diagonal elements of H have two opposite effects
on system performance. On one hand, they introduce ICI,
which degrades system performance. On the other hand, the
symbol xk can be considered as equivalently transmitted over
2dNfD Ts e + 1 subcarriers with channel vector hk , which is
the k-th column of H, and this introduces Doppler diver-
sity [135], [136]. Consequently, Doppler diversity is inherent
in multicarrier communications. Therefore, we need to design
effective algorithms that can remove the negative impacts
of ICI, and simultaneously collect the benefits of Doppler
diversity.
2) JOINT ICI CANCELLATION AND DOPPLER
DIVERSITY COLLECTION
The optimum receiver for the OFDM system in (10) is
to exhaustively search all possible vectors x ∈ S N that
VOLUME 4, 2016
, defined minimizes the cost function ky−Hxk2 . The optimum receiver
can fully collect the Doppler diversity and remove the neg-
ative impacts of ICI, but at the cost of a prohibitive com-
plexity for systems with larger constellation size and/or
(11)
bound (MFB) for OFDM transmissions over doubly selec-
tive Rayleigh fading channels. The MFB benchmarks the
best possible performance if ICI is completely canceled
without noise enhancement. A low-complexity MMSE-DFE
(decision-feedback equalizer) receiver is also given in [136]
= 0, to collect the diversity by ICI suppression. Other types of
ICI cancellation schemes, such as Hybrid interference can-
cellation (IC), multi-stage successive interference cancella-
tion (SIC), and fractionally sampling based IC, are developed
to improve the performance of OFDM systems in doubly
selective fadings [137]–[139].
The complexity of ICI cancellation can be reduced by
exploiting the banded structure of H. One of the main
complexity sources of IC is the matrix inversion. In [140],
the matrix inversion during IC is performed with the
Newton’s iteration method by using the special structure of H.
A sequential detection method with a parallel-filtering archi-
tecture is proposed in [141] to avoid matrix inversion. A pre-
processing scheme is introduced in [142] to further squeeze
ICI into a few coefficients to reduce complexity. In [143],
a low complexity soft MMSE equalizer is designed by using
the banded structure of the frequency domain channel matrix.
Even though Doppler diversity is inherent in
OFDM systems, uncoded OFDM systems usually cannot
collect frequency diversity, because independent data streams
are transmitted over different subcarriers. An oversampled
OFDM (OOFDM) structure is proposed in [135] to collect
both Doppler diversity and frequency diversity in doubly
selective channels. Oversampling in the time domain renders
an equivalent system model where each data stream is trans-
mitted over multiple subcarriers. In [144], a bit-interleaved
time-frequency coded modulation (BITFCM) is used to col-
lect diversity in both time and frequency domains. It is shown
that ICI caused by high Doppler spread might seriously
degrade system performance, thus ICI cancellation is neces-
sary to achieve a good performance in high mobility systems.
The analytical diversity order of BITFCM-OFDM system is
studied in [145]. All above works assume perfect CSI that
might not be available in high mobility systems.
A large number of works are devoted to the joint channel
estimation and data detection in OFDM systems [146]–[154].
In [147], a general framework is proposed for controlled
ICI removal and channel acquisition. The scheme allows
tradeoff between the residual ICI, the training overhead, and
complexity. In [151], the ICI coefficients are directly esti-
mated by using the correlation properties of the channel, and
it is shown to be robust over different channel conditions.
Joint ICI mitigation with pilot-aided OFDM can also be
performed by using iterative methods [148], [153], [155]
or MAP equalizers [152], all achieve good performance in
doubly selective fading.
463

C. MOBILITY ADAPTATION
A new concept of mobility adaptation is proposed for
high mobility OFDM systems in [156], where various
system parameters, such as the number of subcarriers,
data rate, and transmission power, can be dynamically
adjusted and optimized based on the movement speeds. It is
well known that adaptive communications can effectively
adapt to the variations of the fading channel by adjust-
ing various system parameters according to the channel
conditions [84], [157]–[160]. However, adaptive communi-
cations require the precise knowledge of CSI at the trans-
mitter, which is difficult to obtain for high mobility systems
due to fast fading variations and feedback delays. Mobility
adaptation, on the other hand, adjusts system parameters
based on the knowledge of speed or Doppler spread, which
indirectly and approximately reflects the rapidly time-varying
channel characteristics, and can be easily estimated at both
the transmitter and receiver. Since the performance of conven-
tional adaptive communication relies heavily on the accuracy
of the CSI at the transmitter, it has been shown in [156]
that mobility adaptation outperforms conventional adaptive
communications with imperfect CSI when the speed is high.
The basic idea of mobility adaptation is to adjust various
system parameters to balance the tradeoff between perfor-
mance gains and the detrimental impacts of ICI. In [156],
the number of subcarriers is dynamically adjusted based on
the Doppler spread, such that less subcarriers are used for
systems with higher Doppler spread. At a given subcarrier
bandwidth, reducing the number of subcarriers will reduce
the power of ICI, at the cost of a larger percentage of overhead
due to CP. In Fig. 7, the average spectral efficiency (ASE)
is shown as a function of the number of subcarriers under
different mobile speed by using mobility adaptation [156].
The figure shows that the ASE is quasiconcave with respect
to the number of subcarriers. The ASE of a system with
higher speed is less than that of a system with lower speed.
Similarly, the transmission power can also be adjusted based
on speed [161]. At a given speed, a higher transmission power
means high power for both signal and ICI, and the ICI power
increases with speed. Thus the optimum power for a given
FIGURE 7. [156] ASE vs. number of subcarriers with different velocities
and SNRs.
464
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
speed can be obtained by considering the tradeoff between
signal power and ICI power. It is discovered in [161] that the
optimum power is a decreasing function in movement speed.
D. CONCLUDING REMARKS
This section reviews multicarrier communications in high
mobility scenarios. Three aspects are considered, CFO, ICI,
and mobility adaptation. CFO is caused by the time-varying
Doppler spreads and the local oscillator instability, and it is
present in both high mobility and conventional systems. If not
properly compensated, CFO will cause ICI that seriously
degrades system performance. CFO can be estimated and
compensated at the receiver, and various CFO estimation
algorithms are reviewed in this section. Another source of
ICI is the time variation of the fading coefficients inside
one OFDM symbol, and it is unavoidable in doubly selective
fading. On the other hand, ICI can also be exploited to collect
the Doppler diversity inherent in high mobility channels.
We have reviewed a large number of joint ICI suppression
and Doppler diversity collection schemes. Finally, mobility
adaptation can adjust various system parameters, such as the
number of subcarriers, data rate, and power, based on the
movement speed, without the knowledge of CSI. It can bal-
ance the tradeoff between various performance gains and ICI,
thus achieve significant performance gains over conventional
schemes.
Since many future wireless communication standards rely
on multicarrier communications and OFDM, it is impor-
tant to develop new multicarrier systems that can effectively
cope with the challenges and opportunities provided by high
mobility scenarios. One important research direction is the
development of joint ICI suppression and Doppler diversity
system that can operate with imperfect CSI. Imperfect or out-
dated CSI might have significant impact on the performance
of OFDM systems, and it will also affect the transceiver
structure. Therefore it is important to develop new multicar-
rier techniques that can operate effectively in doubly selec-
tive fading and are robust against imperfect CSI. Another
important direction is the development of mobility adaptation
schemes that can collect Doppler diversity. Existing mobility
adaptation schemes do not consider the gains that can be
provided by Doppler diversity, which can provide an extra
degree-of-freedom in mobility adaptations.
VI. NON-COHERENT DETECTIONS
Non-coherent detection is an attractive technique for high
mobility communications because CSI is not needed at either
transmitters or receivers. They only need to know the chan-
nel statistics, which change much slower compared to the
fading coefficients [162]–[165]. We will review non-coherent
detection in high mobility systems from the aspects of both
information theory and practical implementations.
A. NON-COHERENT CAPACITY IN
HIGH MOBILITY SYSTEMS
It is well known that, if the receiver has perfect CSI, the capac-
ity of a coherent MIMO system with M transmit antennas and
VOLUME 4, 2016
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
N receive antennas grows linearly in min(M , N ) [166]. The
coherent capacity at the high SNR regime can be expressed as
Ccoh (ρ) = min{M , N } log(ρ) + O(1), ρ → ∞, (12)
where ρ is the SNR, and O(1) indicates a bounded function
of ρ when the SNR is sufficiently large.
Lack of knowledge of CSI will in general degrade the
channel capacity. The non-coherent capacity of systems with
block fading is studied in [164] and [167]–[169], where
the channel remains constant over a block of T symbols
and changes independently from block to block. Under the
block fading assumption, Marzetta and Hochwald [167]
conclude that increasing the number of transmit antennas
beyond T cannot further increase the capacity, and a unitary
space-time modulation (USTM) is proposed for systems with
M < T [168]. Zheng and Tse [164] show that the non-
coherent capacity behaves in the high SNR regime as

M∗
 M -ary differential phase shift keying (M -DPSK). The infor-
∗
C(ρ) = M 1 − log(ρ) + O(1), ρ → ∞, (13)
T
where M ∗ = min{M , N , bT /2c}. The asymptotic capacity
pre-log is maximized by using M = min{N , T /2}. Compar-
ing (12) with (13) shows that the capacity loss due to the lack
of channel knowledge becomes larger as the coherence time T
becomes smaller. In high mobility and/or large scale MIMO
systems with T < M + N , it is shown by Yang et al. [169]
that USTM is not capacity achieving in the high SNR regime,
and the capacity achieving signal is a Beta-variate space-time
modulation (BSTM) scheme.
The non-coherent capacity of more general
MIMO flat fading channels is studied in [170]–[173].
Lapidoth and Moser [170] prove that the non-coherent capac-
ity grows double-logarithmically with SNR for general fading
with memory. The non-coherent capacity of a time-varying
block fading model are studied in [171] and [173], where the
channel changes correlatively inside a block of length T , and
independently across blocks. In the high SNR regime and
with symbol-spaced sampling, Liang and Veeravalli [171]
prove that the capacity pre-log is T −Q for Q < T , where
T on the assumption that the channel keeps constant over two
Q is the rank of the correlation matrix of the channel gains
inside a block of length T . Dorpinghaus et al. [173] point out
that the symbol-spaced sampling used in [171] is not capacity
achieving. It is shown in [173] that the capacity pre-log is
at least T T−1 if an oversampling factor of two is used at the
receiver. Comparing this result to (13) indicates that the non-
coherent capacity of time-varying block fading can be at least
as good as that of quasi-static block fading.
The bounds of the non-coherent capacity of WSSUS
channels are given in [165], [172], and [174], where the
channel is assumed to be underspread, that is, the product
of the channel’s delay spread and Doppler spread is small.
The capacity bounds are shown as explicit functions of the
channel’s scattering function. It is shown that, if the input
signals are under a peak constraint in both time and frequency,
then the non-coherent capacity bounds are maximized at a
large but finite bandwidth. Beyond this critical bandwidth,
VOLUME 4, 2016
additional bandwidth is detrimental and the capacity
approaches zero as bandwidth increases further [174].
B. NON-COHERENT DETECTIONS
The practical implementations of non-coherent detection
schemes are discussed in this subsection. With non-coherent
detection, the receivers do not have knowledge of the values
of the fading coefficients or the CSI, but they know the
statistics of the fading channel.
1) DIFFERENTIAL MODULATION AND
DIFFERENTIAL DETECTION
Differential modulation and differential detection is the most
popular technique for non-coherent detections. In differential
modulation, the information symbol is encoded in the
difference between two consecutively transmitted symbols.
Most differential modulation uses phase modulation, such as
mation symbol is encoded as the phase difference between
two consecutively transmitted symbols as ck = ck−1 sk ,
where sk ∈ S = {ej2πm/M |m = 0, · · · , M − 1} is the
information symbol, and ck is the differentially modulated
signal. In flat fading channel, the k-th received sample can
be represented as
yk = hk ck + wk , (14)
where hk is the fading coefficient, and wk is additive white
Gaussian noise (AWGN) with variance σw2 . The differen-
tially modulated symbols can be detected either coherently or
non-coherently [175].
In two-symbol differential detection for M -DPSK, The
information symbol is detected by identifying the phase
difference between two consecutively received symbols as
ŝk = argmin |yk y∗k−1 − sk |, (15)
sk ∈S
where ŝk is the detected symbol and S is the modulation
constellation. The two-symbol differential detection is based
consecutive symbol periods, that is, hk ≈ hk−1 , such that
the two consecutive symbols have the same phase distortions
from the channel. When this assumption is true, it is well
known that two-symbol differential detection suffers a 3 dB
loss compared to coherent detection, because the two-symbol
detection process introduces twice as much as noise as the
coherent detection process [175]. When the CSI changes from
symbol to symbol in fast fading channel, the performance loss
is much worse and error floor occurs.
The error floor of two-symbol differential detection
can be removed by using multi-symbol differential detec-
tion (MSDD), where N consecutively received samples are
used to jointly detect N −1 information symbols [176]–[178].
Denote a sequence of N consecutively received samples as
y = [y1 , · · · , yN ]T , which can be expressed as
y = C · h + w, (16)
465

where C = Diag{c} is a diagonal matrix with the ele-
ments of c on its main diagonal, and c = [c1 , · · · , ck ]T ,
h = [h1 , · · · , hk ]T , and w = [w1 , · · · , wk ]T are
N × 1 vectors of differentially encoded symbols, channel
coefficients, and AWGN, respectively. The ML detection rule
for MSDD is [177]
ŝ = argmin{yH CR−1 C H y}, (17)
s∈S N −1
where R = E[hhH ] + σw2 IN , and IN is a size N identity
matrix. The ML detection can be achieved by exhaustively
searching all possible M N −1 combinations of vector s corre-
sponding to all possible vectors of c.
The MSDD scheme requires only the second order statis-
tics of the fading channel. It does not require the assump-
tion that adjacent channel coefficients are approximately the
same, and it can remove the error floor of two-symbol dif-
ferential detection even with N = 3. It has been shown
in [177] that the performance of ML detection of MSDD
approaches that of coherent detection as N becomes large.
In addition, MSDD performs very well even if the receiver
has imperfect knowledge of the second order statistics of the
channel. In the mean time, the complexity of ML detection for
MSDD grows exponentially with M and N . The complexity
can be reduced by using a multi-symbol differential sphere
decoding (MSDSD) scheme [178], the complexity of which
is orders of magnitude below that of exhaustive search, with
almost no performance loss.
For multiple-antenna systems, differential space-time
modulation (DSTM) have been investigated in [179].
In DSTM, the information is encoded in the form of a M × T
matrix, where M is the number of transmit antennas and T is
the block length. In order to perform two-block differential
detection, the channel is assumed to be constant for at least
2 blocks, or 2T symbols, and there is a 3 dB performance
degradation similar to two-symbol differential detection. The
performance degradation is much larger for time-varying
channels [180]. In order to improve the performance under
fast time-varying channels, the MSDD detection has been
investigated for DSTM in MIMO channels in [181].
2) USTM AND BSTM
Unitary space-time modulation (USTM) [167], [168] is appli-
cable to systems with coherence time T larger than the
number of transmit antennas M . The signals are transmitted
in a block of length T over M < T transmit antennas,
and the channel keeps constant in one block and changes
independently from block to block. √The information symbol
is modulated as a unitary matrix T 8 ∈ C T ×M , where
8H 8 = IM , and the (t, m)-th element of 8 is transmitted
by the m-th transmit antenna at the t-th symbol index. The
unitary matrix 8 is chosen from a constellation set containing
L = 2RT unitary matrices, where R is the data rate in bits
per channel use. The columns of 8 spans a M dimensional
subspace of a T dimensional space. It is argued in [168] that
when M  T , or when the SNR is large and M < T ,
466
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
the probability that the received signal falls in the inter-
section of two subspaces spanned by two different unitary
matrices 8l and 8k is zero. Therefore, independent of the
channel coefficients, we can perfectly distinguish 8l from 8k
as long as their columns do not span the same subspace.
It is shown in [168] that USTM is capacity achieving for
M  T , or when SNR → ∞ and M < T . Various
constellations for USTM are designed in [168], and their
performance are evaluated via simulations.
Yang et al. [169] point out that USTM is no longer capacity
achieving when T < M + N , a situation that is relevant
to high mobility systems and/or system with large antenna
arrays. The capacity achieving input signal is BSTM. The
constellation matrix of BSTM is the product of a unitary
isotropically distributed random matrix and a diagonal matrix
whose nonzero entries are distributed as the square-root of the
eigenvalues of a Beta-distributed random matrix.
C. CONCLUDING REMARKS
Non-coherent detections for high mobility systems are
reviewed in this section. From the information theoretic
perspective, non-coherent detection in general suffers a
performance loss due to the lack of CSI at the receiver,
and the loss is related to the coherence time of the
channel. Several non-coherent detection techniques, such
as two-symbol non-coherent detection, multi-symbol non-
coherent detection, USTM, and BSTM, are reviewed.
In addition to the non-coherent detection methods
reviewed in this section, there are also a large family of
blind/semi-blind joint estimation and detection techniques,
where the receiver can utilize the special structure of the chan-
nel and transmitted signal to perform joint channel estimation
and signal detection [182]–[184]. The joint estimation and
detection is usually performed with iterative algorithms, such
as message passing (MP) [182], [183] or EM [184]–[188],
where the results of channel estimation is used to help detec-
tion, and vice versa. However, the blind/semi-blind approach
usually assumes that the channel changes slowly so they can
gradually converge to the true CSI through iterations, and
they might not be able to adapt to fast fading variations. Thus
the joint estimation and detection scheme is in general not
suitable to high mobility scenarios.
The future research directions for non-coherent detections
in high mobility systems are summarized as follows. First,
even though there are a number of asymptotic bounds for non-
coherent detections, the capacity of non-coherent system is
still unknown, especially when the coherence time is small,
and/or the Doppler spread and delay spread is large. Second,
new non-coherent techniques that can adapt to fast fading and
large scale of antenna arrays are needed. In large scale MIMO
systems, it is very difficult, if not impossible, to accurately
estimate and track the large amount of fading coefficients that
change rapidly over time. How to design transmission and
non-coherent detection schemes that do not require the CSI at
either transmitter or receiver remains an open problem.
VOLUME 4, 2016
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
VII. HANDOVER IN HIGH MOBILITY SYSTEMS
Handover is the process by which a user equipment (UE)
maintains its connection active while moving from one point
of attachment (BS or AP) to another. Handover management
is a key aspect for high mobility system designs, and it is
much more challenging than conventional systems with low
or medium mobility. The performance of handover can be
measured by several metrics, such as handover delay,
handover rate, and handover failure probability.
Handover delay is defined as the duration between the
initiation and completion of the handover process. Handover
delay is related to the complexity of the handover man-
agement process, and reduction of the handover delay is
especially important for delay-sensitive voice or multimedia
sessions.
Handover rate is defined as the number of handovers in a
unit time [189]. Reducing the handover rate is usually pre-
ferred as frequent handovers would cause waste of network
resources. A handover is considered to be superfluous when
the UE moves back to the original point of attachment within
a short time duration (ping-pong handover) and it should be
minimized.
A handover failure occurs when the handover is initiated
but the target network does not have sufficient resources to
complete it, or when the mobile terminal moves out of the
coverage of the target network before the process is finalized.
In the former case, the handover failure probability is related
to the resource availability of the target network, while in the
latter case it is related to the mobility of the user.
The main challenges for handover in high mobility systems
are summarized as follows. First, high mobility UEs usually
have a higher handover rate. Second, since the UEs can move
through the overlapping coverage area of two BSs in a very
short period of time, there might not be sufficient time for the
system to complete the handover procedure. Therefore a more
strict delay control is needed for handover in high mobility
systems. Third, the fast moving UE might miss the optimum
handover position, and this will increase the handover fail-
ure probability. Fourth, in HSR systems, handover might be
needed for a large number of users onboard the train at the
same time, and this creates the problem of group handover
that consumes a large amount of network resources.
In order to cope with the challenges of handover in high
mobility systems, many new handover technologies have
been developed in the context of new communication archi-
tectures, such as distributed antenna system (DAS), C-RAN,
coordinate multiple point transmission (CoMP), and HetNet,
etc. We will first review the general operations of handover,
and then present high mobility handover techniques in differ-
ent system architectures.
A. GENERAL HANDOVER OPERATIONS
There are three different types of handovers, including hard
handover, soft handover, and softer handover. In a hard
handover, the existing connection will be released when
a new connection is being established with the new BS.
VOLUME 4, 2016
In a soft handover, a new connection will be established
before the UE breaks from the old one. That is a UE maintains
a radio connection with two or more BSs in an overlapping
region and does not release any of the signals until it drops
below a specified threshold value. A softer handover is very
similar to a soft handover, except that the UE switches con-
nections over radio links that belong to the same access point.
Soft and softer handovers are more popular in high mobility
systems due to their flexibility.
Most of the handover algorithms are based on received
signal strength (RSS). If the differential RSS between the
serving BS and the target BS exceeds a certain handover
hysteresis value H , then the handover process is initiated.
Fig. 8 shows the triggering process of the handover, where
the handover is triggered when the RSS of the target BS or
e-NodeB (eNB) is H dBW higher than that of the serv-
ing eNB. In high mobility systems, small scale fading changes
very fast due to the fast moving of the UE. Therefore the RSS
used for handover is averaged over the effects of small scale
fading. Only the effects of large scale fading, such as pathloss
and shadowing, is considered
FIGURE 8. Hard handover triggering in LTE.
The hysteresis value H plays a critical role in handover
performance. A larger H can reduce handover rate, but will
increase the handover delay or handover failure probability.
On the other hand, a smaller H might lead to many unnec-
essary handovers and increase the network burden. In order
to improve handover performance, various adaptive handover
algorithms are proposed [190]–[192], where dynamic hys-
teresis values based on different factors, such as the UE-eNB
distance and UE speed, can be used to improve the handover
performance.
New adaptive handover techniques are developed and
analyzed for various communication systems [193]–[197].
In [193], the hysteresis value is dynamically adjusted based
on the RSS value of the serving eNB, given that a higher
RSS value from the serving eNB means that the UE can
still get a good QoS from the current serving BS and do not
need to perform handover. An adaptive handover algorithm
is proposed in [194] by estimating the pathloss between the
UE and various eNBs.
B. HANDOVER IN SYSTEMS WITH DAS
In DAS systems, multiple antennas are distributed at different
locations and are connected to a central unit with wired links
467

such as optical fibers, the so-called radio-over-fiber (RoF)
structure. Compared with the traditional system with
co-located antennas, DAS has many advantages in terms of
higher outage capacity, lower bit error rate, larger coverage
area, simpler network topology, and better handover perfor-
mance, etc. [198], [199].
Most DAS systems are installed on the
BS side [200]–[202]. In conventional systems with co-located
BS antennas, two adjacent cells need to have a large over-
lapped area to ensure sufficient handover time and to reduce
handover drop probability. However, a large overlapped
area usually leads to stronger co-channel interference (CCI),
which might degrade the overall system capacity. For systems
with DAS, the CCI can be significantly reduced because each
antenna only covers a much smaller service area. In addition,
softer handovers can be performed inside a cell to ensure the
QoS of the users.
One prominent DAS system is the C-RAN architecture
as shown in Fig. 9, where a large number of remote radio
heads (RRHs) are connected to a centralized baseband
unit (BBU) pool. Any BBU can cooperate with other BBUs
within the same BBU pool with very high bandwidth and low
latency. The UE can perform softer handover among RRHs
connecting to the same BBU, or soft handover between RRHs
connecting to different BBUs. The architecture of central-
ized BBUs and distributed RRHs can reduce handover delay
and handover failure probability, and meets the requirement
of handover in high mobility systems. In [203], a C-RAN
architecture is proposed for cloud high speed train commu-
nications (C-HSTC), where a virtualized single cell design is
used to mitigate handover failures and guarantee continuous
communication services in high mobility systems.
FIGURE 9. Cloud based communication system architecture.
The DAS systems can also be installed on the user side,
especially for HSR systems, where the DAS mounted on top
of a train can serve as mobile relays (MRs) between the
BS and UEs [195], [196]. Since all user data are congre-
gated at the MRs, handover will only be performed between
468
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
MRs and BS, and this can effectively eliminate the group
handover problem in HSR systems. A dual-link handover
scheme is proposed in [195] for HSR systems. In the dual-link
scheme, two relaying antennas are mounted at the front and
rear of the train, respectively. The front antenna executes the
handover while the rear one communicates with the current
serving BS. In [196], the handover is performed with the
assistance of antenna selection and power allocation among
the antennas. The results in both [195] and [196] show that
employing DAS can significantly reduce the handover failure
probability.
C. HANDOVER IN SYSTEMS WITH CoMP
CoMP transmission and reception allow geographically sep-
arated BSs to cooperate with each other during the trans-
mission and reception, such that each UE can be served by
multiple BS simultaneously. The CoMP scheme can reduce
CCI among different cells, and improve the system spectral
efficiency.
Various handover strategies have been proposed by using
the CoMP structure. In [204], a CoMP-based soft handover
strategy is proposed for HSR systems. The handover strategy
allows the train to be served by two adjacent BSs with CoMP
when the train travels through the overlapping area. In [205],
the handover is performed by sequentially selecting a subset
of CoMP BSs that can maximize the system capacity.
D. HANDOVER IN HETEROGENEOUS NETWORKS
HetNet uses multiple radio access technologies to pro-
vide services to the users. A HetNet can use conventional
macrocells, small cells including microcells, picocells, and/or
femtocells to offer wireless coverage in an environment with
a wide variety of wireless coverage zones. The convergence
of heterogeneous wireless access technologies has been envi-
sioned as a key component for the next generation wireless
networks. There are complex interoperations between macro-
cells and small cells, and among different radio access tech-
nologies. Handover capability between the diverse network
elements is an essential element in HetNet. Various handover
and mobility management algorithms in HetNet have been
summarized in [206] and [207].
In HetNet, there are two different types of handovers,
horizontal handover and vertical handover.
Horizontal handover occurs between APs supporting the
same network technology. Horizontal handover is mainly
triggered by the UE movement, which changes the RSS
from different APs. It shares many properties of handover in
homogeneous networks.
Vertical handover occurs between APs supporting different
network technologies, such as handovers from a small cell to
a macro-cell, or from a WiFi AP to a cellular BS. Decision for
vertical handover might be based on different criteria [206].
Network-related criteria include the link quality, bandwidth,
delay, security, and coverage, etc. Terminal-related criteria
include the mobility, battery, location, etc. User-related crite-
ria include user profile and preferences, etc. Service-related
VOLUME 4, 2016
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
criteria include service capability and QoS, etc. The network
and UE use one or more criteria to make a decision whether to
initiate the handover process. For example, it is more suitable
to serve a high mobility user with a macrocell than a microcell
to reduce the number of handovers, given that the user can
quickly move out the coverage area of small cells.
E. CONCLUDING REMARKS
Handover management is essential for high mobility commu-
nication systems, given that the fast movement of the UEs
demands frequent change of the serving BS or APs. A number
of handover techniques have been reviewed in this section,
and they can improve the handover performance by utilizing
various new networks architectures and technologies, such as
DAS, MRs, C-RAN, CoMP, and HetNet.
Future research on handover in high mobility systems
should rely on the unique architecture of next generation
wireless standards. It is expected that heterogeneous and
distributed systems will dominate future wireless communi-
cations. The seamless and efficient handover between differ-
ent APs and access technologies still remain a challenging
problem for high mobility communications.
VIII. ARCHITECTURE OF HIGH MOBILITY NETWORKS
Conventional cellular network architectures are mainly
designed for users with low and medium mobility. High
mobility communications demand new network architectures
that are tailored specifically for the needs of high mobility
users, and can address the challenges imposed by high user
mobility.
A. NETWORK ARCHITECTURE FOR HSR SYSTEMS
One of the most widely used network architectures for
HSR systems is GSM-R, which is built based on the popular
GSM technologies. GSM-R is mainly designed for voice
traffic and narrow-band data traffic for railway operation
controls, and it can operate reliably at train speed up
to 500 km/hr. GSM-R has been widely deployed in HSR
systems in Europe, Asia, and South Africa, etc. The
architecture for GSM-R is shown in Fig. 10. Dedicated
GSM-R Base Transceiver Stations (BTS) are deployed with
linear coverage. BS controller (BSC) and mobile switching
center (MSC) fulfill the functions of the wireless link
management and mobility management, respectively. Even
though GSM-R is a reliable communication architecture
for HSR systems, it can only be used for narrowband
control information. With the increasing popularity of HSR
FIGURE 10. System model.
VOLUME 4, 2016
systems and the growing demands for broadband commu-
nications on trains, it is necessary to develop new network
architecture that can support broadband traffic for general
users.
A natural evolution for GSM-R is LTE for railway, which
also has a linear coverage structure. LTE for railway system
is designed to provide broadband all-IP communications to
both railway operators and users. One of the main propo-
nent for LTE for railway is Alcatel-Lucent. Some of the
important features of LTE for railway solutions proposed by
Alcatel-Lucent can be summarized as follows [208]. First, it
utilizes distributed eNB architecture, where the multiple spa-
tially distributed RRHs are connected to BBU pool through
fiber links. The distributed eNB architecture can provide bet-
ter coverage, enable flexible resource allocations, and reduce
the number of equipment to be installed and maintained.
Second, small cells are deployed to reduce the transmission
power and provide better spectrum efficiency. Third, it pro-
vides mission critical service enhancement entities to ensure
that critical control information can be reliably delivered with
low latency. In addition, LTE uses an all-IP network with
sophisticated QoS built into the system implementation, thus
it can provide reliable broadband services to high mobility
users on board the train.
The performance of HSR communication systems can
also be improved by performing architecture or service opti-
mizations in several aspects, such as the locations of eNBs
along the rail track [209], the efficiency and fairness of the
utilizations of the eNBs [210], and the eNB transmission
power [211], etc.
B. MOBILE RELAYS
Employing MRs installed on the train roof can solve a num-
ber of unique issues related to HSR systems, such as high
penetration loss and group handover. The MRs are usually
installed on top of the train roof outside of the train. They are
connected to APs installed inside the train through wired links
such as optical fiber [212]–[214]. In the uplink, the traffic
from all UEs are first congregated at the MRs through the
APs, then the MRs forward the congregated user traffic to
the road side eNBs. In the downlink, the eNB transmits all
user traffic to the MRs, which then deliver the traffic to their
intended UEs through the APs inside the train carriage.
The MR structure has two airlinks, the AP-UE link and
the MR-eNB link. The AP-UE link is a stationary link with
short transmission distance, thus it can support high data rate
without any of the challenges of high mobility communica-
tions. The MR-eNB link is the bottleneck of the two-hop
relay system [215]. Since the MRs are installed outside of
the train, it eliminates the penetration loss. In addition, the
MRs usually have a line-of-sight with the eNBs, and this can
further improve the spectral efficiency of the system.
All MRs on top of the train can either be co-located at the
same location, or evenly spread out the entire train length.
Both scenarios are considered in [212], where the optimum
number of MRs are identified by maximizing the average
469

uplink capacity. It is discovered that systems with co-located
MRs slightly outperform those with distributed MRs, but the
performance difference is very small. An optimum power
allocation scheme is proposed in [214] for HSR systems with
MRs to balance the tradeoff between power consumption and
delays.
Employing MRs can also solve the group handover
problem, because the handover is performed between the
MRs and eNB. This can improve the overall handover
performance.
C. CONTROL PLANE/USER PLANE DECOUPLING
The logical network architecture can be divided into the
control plane (C-plane) and user plane (U-plane), which are
used to deliver control information and user information,
respectively. In HSR systems, the C-plane includes control
information for both communications and train operations.
In conventional network architecture, data in the C-plane
and U-plan are usually transmitted with the same physical
channel. In order to support broadband communications, the
user data are usually transmitted at high frequency bands
with wider bandwidth. However, high frequency bands might
suffer from high propagation loss and high penetration loss
induced by the train carriage. In addition, in HetNet, the
frequent handover among different network elements might
seriously affect the reliability of the mission critical C-plane
information.
To address these problems, a C/U-plane decoupled
architecture is proposed for upcoming 5G wireless communi-
cations [216]–[220], where the relatively important C-plane
is kept at lower frequency bands in macro cells, and the
U-plane is moved to small cells using higher frequency bands
to gain broader bandwidth. The detailed network architec-
ture of C/U-plane decoupling is depicted in Fig. 11. Since
small cells are only responsible for user traffic, the small
cell eNBs are only connected to serving gateway (SGW),
which supports U-plane session management functions. The
macro-cell eNBs can also provide low rate high reliability
services to U-plane, thus they are connected to both SGW
and mobility management entity (MME), which supports
LTE C-plane mobility management and connection
FIGURE 11. The C/U-plane decoupled heterogeneous railway wireless
network.
470
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
management functions. The control signaling for managing
small cells is handled by macro cells via interface X3.
D. C-RAN ARCHITECTURE FOR HSR
In HSR systems, the movement of users shows a very strong
time-geometry pattern. There is usually a large space between
two trains for safety consideration. Therefore, the number of
simultaneously running trains on the same route is very small.
Such a movement pattern creates the so-called ‘‘tidal effect’’,
where a large number of users pass along a small portion of
the railway track. Consequently, the BSs close to the train
are crowded, while most of the BSs along the track do not
serve any user. Thus the utilization efficiency of conventional
BS or eNB in HSR systems is very low, which results in a
huge waste of resources.
This problem can be effectively solved by using the
C-RAN architecture, where multiple RRHs can be installed
along the railway track, and they are connected to a pool of
BBUs through high capacity fiber links [221]. The BBU pool
enables centralized scheduling of all the RRHs and resources
in the network. As a train moves along the railway track, the
BBU can dynamically selects a subset of RRHs that can best
serve the users on the train, and flexibly allocates the process-
ing resources at the BBU to the users based on their respective
QoS needs. Such a scheme enables a global optimization of
resource allocation through virtualization technology, and it
effectively address the ‘‘tidal effect’’ for HSR communication
systems.
In addition, the distributed RRH or DAS structure enables
spatial diversity and frequency reuse among the RRHs,
and this can further improve the system capacity per area.
Spatially distributed RRHs can reuse the same spectrum
while maintaining a low CCI among themselves, and this can
significantly improve the area spectral efficiency of the entire
system. The spatial diversity of C-RAN or DAS systems
can be achieved by RRH selection/activation, where only a
small subset of RRHs are activated at a given time while
all other RRHs remain in a sleeping state to save power.
In [222] and [223], a joint RRH activation and beamforming
scheme is proposed for C-RAN, and the results show that
sparse RRH activation yields significant performance gains
over systems with all RRHs activated.
E. CONCLUDING REMARKS
Network architecture can be designed and optimized by using
the unique features of high mobility systems. The first widely
used high mobility network architecture is GSM-R, which
only provides narrow band services to railway operators.
New network architectures and operations, such as LTE
for railway, MRs, heterogeneous networks with C/U-plane
decoupling, and C-RAN, can effectively improve system per-
formance by exploiting the various features of high mobility
systems.
High mobility communication is an important component
for 5G communication systems. New high mobility net-
work architectures are needed to take advantage of the new
features and technologies enabled by 5G communications.
VOLUME 4, 2016
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
Research on network architecture for high mobility
communications in 5G is still at its infancy. Unlike the
3G and 4G standards that treats the high mobility issue as an
after thought, the 5G communication systems will treat high
mobility as an integral part of the communication standards,
and take the high mobility issue into consideration during the
initial design of the network architecture. How to design an
effective and efficient high mobility network architecture for
5G systems still remain an open problem.
IX. CONCLUSIONS
In this paper, we have provided a systematic review on the
challenges, modeling, design, and analysis of high mobility
communication systems. Due to the high mobility of the
communication devices, there is fast channel variation in the
physical layer, fast link variations in the link layer, and fast
topology variation in the network layer. The fast time-varying
operation environment imposes formidable challenges on
system designs, and it also provides Doppler diversity that
can be exploited to improve system performance. Consider-
ing the unique properties of high mobility communications,
we first summarized a list of key challenges and opportuni-
ties provided by the high mobility environment. To address
these challenges and take advantage of the opportunities, we
then provided reviews in various aspects of high mobility
system designs and analysis, including channel modeling,
channel estimation, Doppler diversity transmissions, multi-
carrier communications, non-coherent detections, handover,
and network architectures. The review focused primarily on
the physical layer operations, with some discussions on upper
layer operations such as handover and network architectures.
There are still a large number of issues that need to be
addressed for the development of future high mobility com-
munication systems. The key challenge remains to be how to
provide reliable broadband services to a large number of high
mobility users experiencing non-stationary fast time-varying
fading. This challenge can only be tackled by comprehen-
sively overhaul the designs and operations of current com-
munication systems, from signaling, transceiver designs, to
network architectures. We summarize the important research
topics for high mobility communications as follows: the
accurate modeling and estimation of the non-stationary high
mobility propagation channel, the design of new transceiver
structures that can effectively collect the benefits of Doppler
diversity, and operate with imperfect CSI or no CSI at all, new
signal processing techniques that can mitigate the impair-
ments (such as CFO and ICI) caused by high mobility chan-
nels, new communication techniques that can ensure QoS by
automatically adapt to the mobile speed, and new mobility
management and network architectures that are designed by
taking advantage of the unique properties of high mobility
systems, etc.
5G communication systems have incorporated high mobil-
ity communications as an integral part of the communication
standards. 5G systems are expected to provide broadband
services to a large number of simultaneous users (1000+)
VOLUME 4, 2016
traveling at a speed up to 500km/hr or even higher. The
realization of this vision requires the development of innova-
tive communication techniques that can cope with the unique
challenges and opportunities of high mobility communica-
tions. The new techniques should also be designed by taking
advantage of the new features of 5G communications,
such as cell densification, massive multiple-input multiple-
output (MIMO), full-duplex transmission, non-orthogonal
multiple access (NOMA), HetNet, and C-RAN, etc.
ACKNOWLEDGEMENT
The authors are grateful to Weixi Zhou, Ning Sun,
Zhicheng Dong, Liu Feng, Hao Song, and Ziyue Liu for their
helpful inputs and assistance during the preparation of this
paper.
REFERENCES
[1] P. Fan, ‘‘Advances in broadband wireless communications under high-
mobility scenarios,’’ Chin. Sci. Bull., vol. 59, no. 35, pp. 4974–4975,
Dec. 2014.
[2] P. Fan, E. Panayirci, H. V. Poor, and P. T. Mathiopoulos, ‘‘Special issue
on broadband mobile communications at very high speeds,’’ EURASIP J.
Wireless Commun. Netw., vol. 2012, no. 1, p. 279, Aug. 2012. [Online].
Available: http://jwcn.eurasipjournals.com/content/2012/1/279
[3] P. Fan, E. Panayirci, P. Li, C. Wang, and V. Tarokh, ‘‘Guest editorial:
Special issue on high mobility wireless communications,’’ J. Modern
Transp., vol. 20, no. 4, pp. 197–198, Dec. 2012.
[4] F. Hlawatsch and G. Matz, Eds., Wireless Communications Over Rapidly
Time-Varying Channels. San Francisco, CA, USA: Academic, 2011.
[5] Y. Zhou, F. Adachi, X. Wang, A. Manikas, X. Zhang, and W. Zhu, ‘‘Guest
editorial broadband wireless communications for high speed vehicles,’’
IEEE J. Sel. Areas Commun., vol. 30, no. 4, pp. 673–674, May 2012.
[6] X. Zhu, S. Chen, H. Hu, X. Su, and Y. Shi, ‘‘TDD-based mobile com-
munication solutions for high-speed railway scenarios,’’ IEEE Wireless
Commun., vol. 20, no. 6, pp. 22–29, Dec. 2013.
[7] Z. Ma, Z. Zhang, Z. Ding, P. Fan, and H. Li, ‘‘Key techniques for
5G wireless communications: Network architecture, physical layer, and
MAC layer perspectives,’’ Sci. China Inf. Sci., vol. 58, no. 4, pp. 1–20,
Apr. 2015.
[8] IMT-2020 (5G) Promotion Group. (May 2015). 5G Network Tech-
nology Architecture White Paper. [Online]. Available: http://www.
imt-2020.org.cn/en
[9] 5G Vision, Samsung Electronics Co. Ltd., Suwon, Korea, Feb. 2015.
[10] Requirements for Further Advancements for Evolved Universal
Terrestrial Radio Access (EUTRA) (LTE-Advanced), document
3GPP TR 36.913, 2014. [Online]. Available: http://www.3gpp.org/
DynaReport/36913.htm
[11] X. Zhang and X. Zhou, Eds., LTE-Advanced Air Interface Technology.
Boca Raton, FL, USA: CRC Press, 2012.
[12] W. C. Jakes and D. C. Cox, Eds., Microwave Mobile Communications.
New York, NY, USA: Wiley, 1994.
[13] G. B. Giannakis and C. Tepedelenlioglu, ‘‘Basis expansion models and
diversity techniques for blind identification and equalization of time-
varying channels,’’ Proc. IEEE, vol. 86, no. 10, pp. 1969–1986, Oct. 1998.
[14] S. Lin, Z. Zhong, L. Cai, and Y. Luo, ‘‘Finite state Markov modelling
for high speed railway wireless communication channel,’’ in Proc. IEEE
Global Telecommun. Conf. (GLOBECOM), Dec. 2012, pp. 5421–5426.
[15] R. H. Clarke, ‘‘A statistical theory of mobile-radio reception,’’ Bell Syst.
Tech. J., vol. 47, no. 6, pp. 957–1000, Jul./Aug. 1968.
[16] H. O. Beca, ‘‘An orthogonal set based on Bessel functions of the
first kind,’’ Univ. Beograd. Public. Elektrotehnicki. Fak. Ser. Mat.,
pp. 85–90, Mar. 1980.
[17] C. Xiao and Y. R. Zheng, ‘‘A generalized simulation model for Rayleigh
fading channels with accurate second-order statistics,’’ in Proc. 55th Veh.
Technol. Conf. (VTC-Spring), vol. 1. 2002, pp. 170–174.
[18] Y. R. Zheng and C. Xiao, ‘‘Simulation models with correct statistical
properties for Rayleigh fading channels,’’ IEEE Trans. Commun., vol. 51,
no. 6, pp. 920–928, Jun. 2003.
471

[19] X. Cheng, C.-X. Wang, D. I. Laurenson, S. Salous, and A. V. Vasilakos,
‘‘New deterministic and stochastic simulation models for non-
isotropic scattering mobile-to-mobile Rayleigh fading channels,’’
Wireless Commun. Mobile Comput., vol. 11, no. 7, pp. 829–842,
Jul. 2011.
[20] X. Cheng et al., ‘‘An improved parameter computation method
for a MIMO V2V Rayleigh fading channel simulator under non-
isotropic scattering environments,’’ IEEE Commun. Lett., vol. 17, no. 2,
pp. 265–268, Feb. 2013.
[21] K. A. D. Teo, S. Ohno, and T. Hinamoto, ‘‘Kalman channel estima-
tion based on oversampled polynomial model for OFDM over doubly-
selective channels,’’ in Proc. IEEE 6th Workshop Signal Process. Adv.
Wireless Commun., Jun. 2005, pp. 116–120.
[22] H. Hijazi and L. Ros, ‘‘Joint data QR-detection and Kalman estimation
for OFDM time-varying Rayleigh channel complex gains,’’ IEEE Trans.
Commun., vol. 58, no. 1, pp. 170–178, Jan. 2010.
[23] M. Visintin, ‘‘Karhunen–Loève expansion of a fast Rayleigh fading pro-
cess,’’ Electron. Lett., vol. 32, no. 18, pp. 1712–1713, Aug. 1996.
[24] T. Zemen and C. F. Mecklenbrauker, ‘‘Time-variant channel estimation
using discrete prolate spheroidal sequences,’’ IEEE Trans. Signal Pro-
cess., vol. 53, no. 9, pp. 3597–3607, Sep. 2005.
[25] T. Zemen, L. Bernado, N. Czink, and A. F. Molisch, ‘‘Iterative time-
variant channel estimation for 802.11p using generalized discrete pro-
late spheroidal sequences,’’ IEEE Trans. Veh. Technol., vol. 61, no. 3,
pp. 1222–1233, Mar. 2012.
[26] L. Liu, C. Tao, J. Qiu, T. Zhou, R. Sun, and H. Chen, ‘‘The
dynamic evolution of multipath components in high-speed railway in
viaduct scenarios: From the birth-death process point of view,’’ in
Proc. IEEE Pers. Indoor Mobile Radio Commun. (PIMRC), Sep. 2012,
pp. 1774–1778.
[27] G. Acosta-Marum and M.-A. Ingram, ‘‘Six time- and frequency-selective
empirical channel models for vehicular wireless LANs,’’ IEEE Veh. Tech-
nol. Mag., vol. 2, no. 4, pp. 4–11, Dec. 2007.
[28] I. Sen and D. W. Matolak, ‘‘VehicleŰvehicle channel models for
the 5-GHz band,’’ IEEE Trans. Intell. Transp. Syst., vol. 9, no. 2,
pp. 235–245, Jun. 2008.
[29] I. Sen and D. W. Matolak, ‘‘The 5-GHz airport surface area channel—
Part II: Measurement and modeling results for small airports,’’ IEEE
Trans. Veh. Technol., vol. 57, no. 4, pp. 2027–2035, Jul. 2008.
[30] S. Elnoubi, ‘‘A simplified stochastic model for the aeronautical mobile
radio channel,’’ in Proc. IEEE Veh. Technol. Conf. (VTC), vol. 2.
May 1992, pp. 960–963.
[31] Q. Pan, S. Liu, M. Xu, and C. Jia, ‘‘Finite-state Markov model for
the aeronautical channel,’’ in Proc. Int. Conf. Wireless Commun. Netw.
Mobile Comput. (WiCom), Sep. 2009, pp. 1–4.
[32] C. Wang, X. Cheng, and D. I. Laurenson, ‘‘Vehicle-to-vehicle channel
modeling and measurements: Recent advances and future challenges,’’
IEEE Commun. Mag., vol. 47, no. 11, pp. 96–103, Nov. 2009.
[33] C. Wang, A. Ghazal, B. Ai, Y. Liu, and P. Fan, ‘‘Channel measurements
and models for high-speed train communication systems: A survey,’’
IEEE Commun. Surveys Tuts., to be published.
[34] D. J. Cichon, T. Zwick, and W. Wiesbeck, ‘‘Ray optical modeling of
wireless communications in high-speed railway tunnels,’’ in Proc. IEEE
Veh. Technol. Conf. (VTC), vol. 1. Apr. 1996, pp. 546–550.
[35] S. Knorzer, M. A. Baldauf, T. Fugen, and W. Wiesbeck, ‘‘Channel anal-
ysis for an OFDM-MISO train communications system using different
antennas,’’ in Proc. IEEE Veh. Technol. Conf. (VTC-Fall), Sep./Oct. 2007,
pp. 809–813.
[36] L. Tian, X. Yin, Q. Zuo, J. Zhou, Z. Zhong, and S. X. Lu, ‘‘Channel
modeling based on random propagation graphs for high speed railway
scenarios,’’ in Proc. IEEE 23rd Int. Symp. Pers. Indoor Mobile Radio
Commun. (PIMRC), Sep. 2012, pp. 1746–1750.
[37] K. Guan, Z. Zhong, B. Ai, and T. Kurner, ‘‘Semi-deterministic path-loss
modeling for viaduct and cutting scenarios of high-speed railway,’’
IEEE Antennas Wireless Propag. Lett., vol. 12, pp. 789–792,
Jun. 2013.
[38] Z. Qianqian, X. Chunxiu, and W. Muqing, ‘‘A novel MIMO channel
model for high speed railway system,’’ in Proc. IEEE Int. Conf. Commun.
Technol. (ICCT), Nov. 2012, pp. 31–35.
[39] X. Yin, X. Cai, X. Cheng, J. Chen, and M. Tian, ‘‘Empirical
geometry-based random-cluster model for high-speed-train channels in
UMTS networks,’’ IEEE Trans. Intell. Transp. Syst., vol. 16, no. 5,
pp. 2850–2861, Oct. 2015.
472
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
[40] A. Ghazal, C.-X. Wang, B. Ai, D. Yuan, and H. Haas, ‘‘A nonsta-
tionary wideband MIMO channel model for high-mobility intelligent
transportation systems,’’ IEEE Trans. Intell. Transp. Syst., vol. 16, no. 2,
pp. 885–897, Apr. 2015.
[41] B. Chen and Z. Zhong, ‘‘Geometry-based stochastic modeling for MIMO
channel in high-speed mobile scenario,’’ Int. J. Antennas Propag.,
vol. 2012, Sep. 2012, Art. ID 184682.
[42] A. G. Zajic and G. L. Stuber, ‘‘Simulation models for MIMO mobile-to-
mobile channels,’’ in Proc. IEEE Military Commun. Conf. (MILCOM),
Oct. 2006, pp. 1–7.
[43] A. S. Akki and F. Haber, ‘‘A statistical model of mobile-to-mobile land
communication channel,’’ IEEE Trans. Veh. Technol., vol. 35, no. 1,
pp. 2–7, Feb. 1986.
[44] X. Cheng, C.-X. Wang, B. Ai, and H. Aggoune, ‘‘Envelope level cross-
ing rate and average fade duration of nonisotropic vehicle-to-vehicle
Ricean fading channels,’’ IEEE Trans. Intell. Transp. Syst., vol. 15, no. 1,
pp. 62–72, Feb. 2014.
[45] X. Cheng, C.-X. Wang, D. I. Laurenson, S. Salous, and A. V. Vasilakos,
‘‘An adaptive geometry-based stochastic model for non-isotropic MIMO
mobile-to-mobile channels,’’ IEEE Trans. Wireless Commun., vol. 8,
no. 9, pp. 4824–4835, Sep. 2009.
[46] A. G. Zajic and G. L. Stuber, ‘‘Space-time correlated mobile-to-mobile
channels: Modelling and simulation,’’ IEEE Trans. Veh. Technol., vol. 57,
no. 2, pp. 715–726, Mar. 2008.
[47] A. G. Zajic and G. L. Stuber, ‘‘Three-dimensional modeling and sim-
ulation of wideband MIMO mobile-to-mobile channels,’’ IEEE Trans.
Wireless Commun., vol. 8, no. 3, pp. 1260–1275, Mar. 2009.
[48] X. Cheng, C. Wang, and D. I. Laurenson, ‘‘A geometry-based stochas-
tic model for wideband MIMO mobile-to-mobile channels,’’ in Proc.
IEEE Global Telecommun. Conf. (GLOBECOM), Honolulu, HI, USA,
Nov. 2009, pp. 1–6.
[49] Y. Yuan, C.-X. Wang, X. Cheng, B. Ai, and D. I. Laurenson, ‘‘Novel
3D geometry-based stochastic models for non-isotropic MIMO vehicle-
to-vehicle channels,’’ IEEE Trans. Wireless Commun., vol. 13, no. 1,
pp. 298–309, Jan. 2014.
[50] M.-T. Tao, V.-A. Nguyen, Y.-T. Im, S.-O. Park, and G. Yoon, ‘‘3D
polarized channel modeling and performance comparison of MIMO
antenna configurations with different polarizations,’’ IEEE Trans.
Antennas Propag., vol. 59, no. 7, pp. 2672–2682, Jul. 2011.
[51] X. Cheng, Q. Yao, M. Wen, C.-X. Wang, L.-Y. Song, and B.-L. Jiao,
‘‘Wideband channel modeling and intercarrier interference cancella-
tion for vehicle-to-vehicle communication systems,’’ IEEE J. Sel. Areas
Commun., vol. 31, no. 9, pp. 434–448, Sep. 2013.
[52] D. W. Matolak, ‘‘Air-ground channels & models: Comprehensive review
and considerations for unmanned aircraft systems,’’ in Proc. IEEE
Aerosp. Conf. (AERO), Mar. 2012, pp. 1–17.
[53] C. Zhang and J. Yu, ‘‘Real-time aeronautical channel simulator,’’
in Proc. IEEE/AIAA Digit. Avionics Syst. Conf. (DASC), Oct. 2013,
pp. 3B1-1–3B1-9.
[54] W. G. Newhall and J. H. Reed, ‘‘A geometric air-to-ground radio channel
model,’’ in Proc. IEEE Military Commun. Conf. (MILCOM), vol. 1.
Oct. 2002, pp. 632–636.
[55] D. J. Cichon, T. C. Becker, and W. Wiesbeck, ‘‘Determination of time-
variant radio links in high-speed train tunnels by ray optical modeling,’’
in Antennas Propag. Soc. Int. Symp. (AP-S), Dig., vol. 1. Jun. 1995,
pp. 508–511.
[56] J. K. Tugnait and W. Luo, ‘‘Linear prediction error method for blind
identification of periodically time-varying channels,’’ IEEE Trans. Signal
Process., vol. 50, no. 12, pp. 3070–3082, Dec. 2002.
[57] H. Wang, Y. Lin, and B. Chen, ‘‘Data-efficient blind OFDM channel
estimation using receiver diversity,’’ IEEE Trans. Signal Process., vol. 51,
no. 10, pp. 2613–2623, Oct. 2003.
[58] M. Khalighi and J. J. Boutros, ‘‘Semi-blind channel estimation using the
EM algorithm in iterative MIMO APP detectors,’’ IEEE Trans. Wireless
Commun., vol. 5, no. 11, pp. 3165–3173, Nov. 2006.
[59] X. Li and P. Fan, ‘‘Modified clustered comb pilot-aided fast time-varying
channel estimation for OFDM system,’’ J. Modern Transp., vol. 20,
pp. 220–226, Dec. 2012.
[60] R. C. Cannizzaro, P. Banelli, and G. Leus, ‘‘Adaptive channel estimation
for OFDM systems with Doppler spread,’’ in Proc. IEEE Workshop Signal
Process. Adv. Wireless Commun. (SPAWC), Jul. 2006, pp. 1–5.
[61] B. Muquet, M. de Courville, and P. Duhamel, ‘‘Subspace-based blind and
semi-blind channel estimation for OFDM systems,’’ IEEE Trans. Signal
Process., vol. 50, no. 7, pp. 1699–1712, Jul. 2002.
VOLUME 4, 2016
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
[62] C. Li and S. Roy, ‘‘Subspace-based blind channel estimation for OFDM
by exploiting virtual carriers,’’ IEEE Trans. Wireless Commun., vol. 2,
no. 1, pp. 141–150, Jan. 2003.
[63] N. Sun and J. Wu, ‘‘Maximizing spectral efficiency for high mobility
systems with imperfect channel state information,’’ IEEE Trans. Wireless
Commun., vol. 13, no. 3, pp. 1462–1470, Mar. 2014.
[64] M. Dong and L. Tong, ‘‘Optimal design and placement of pilot symbols
for channel estimation,’’ IEEE Trans. Signal Process., vol. 50, no. 12,
pp. 3055–3069, Dec. 2002.
[65] L. Tong, B. M. Sadler, and M. Dong, ‘‘Pilot-assisted wireless trans-
missions: General model, design criteria, and signal processing,’’ IEEE
Signal Process. Mag., vol. 21, no. 6, pp. 12–25, Nov. 2004.
[66] N. Sun and J. Wu, ‘‘Optimum designs for high mobility systems with
channel estimation errors,’’ in Proc. IEEE Int. Conf. Commun. (ICC),
Jun. 2014, pp. 5072–5077.
[67] N. Sun, J. Wu, and P. Fan, ‘‘Optimum designs of high mobility wireless
systems with channel estimation errors,’’ IET Commun., vol. 9, no. 13,
pp. 1677–1682, Sep. 2015.
[68] N. Sun and J. Wu, ‘‘Minimum error transmissions with imperfect channel
information in high mobility systems,’’ in Proc. IEEE Military Commun.
Conf. (MILCOM), Nov. 2013, pp. 922–927.
[69] S. Ohno and G. B. Giannakis, ‘‘Average-rate optimal PSAM trans-
missions over time-selective fading channels,’’ IEEE Trans. Wireless
Commun., vol. 1, no. 4, pp. 712–720, Oct. 2002.
[70] N. Sun, J. Wu, and P. Fan, ‘‘Optimum pilot designs for high mobil-
ity systems with channel estimation errors,’’ IET Commun., vol. 9,
pp. 1677–1682, Sep. 2015.
[71] X. Ren, W. Chen, and M. Tao, ‘‘Position-based compressed channel
estimation and pilot design for high-mobility OFDM systems,’’ IEEE
Trans. Veh. Technol., vol. 64, no. 5, pp. 1918–1929, May 2015.
[72] A. P. Kannu and P. Schniter, ‘‘MSE-optimal training for linear time-
varying channels,’’ in Proc. IEEE Int. Conf. Acoust., Speech, Signal
Process., vol. 3. Mar. 2005, pp. iii/789–iii/792.
[73] A. P. Kannu and P. Schniter, ‘‘Design and analysis of MMSE pilot-aided
cyclic-prefixed block transmissions for doubly selective channels,’’ IEEE
Trans. Signal Process., vol. 56, no. 3, pp. 1148–1160, Mar. 2008.
[74] X. Ma, G. B. Giannakis, and S. Ohno, ‘‘Optimal training for block trans-
missions over doubly selective wireless fading channels,’’ IEEE Trans.
Signal Process., vol. 51, no. 5, pp. 1351–1366, May 2003.
[75] N. Sun and J. Wu, ‘‘On the performance of doubly-selective fading
estimations in high mobility systems,’’ in Proc. IEEE 79th Veh. Technol.
Conf. (VTC Spring), May 2014, pp. 1–5.
[76] S. Ohno and G. B. Giannakis, ‘‘Capacity maximizing MMSE-optimal
pilots for wireless OFDM over frequency-selective block Rayleigh-fading
channels,’’ IEEE Trans. Inf. Theory, vol. 50, no. 9, pp. 2138–2145,
Sep. 2004.
[77] Z. Tang, R. C. Cannizzaro, G. Leus, and P. Banelli, ‘‘Pilot-assisted time-
varying channel estimation for OFDM systems,’’ IEEE Trans. Signal
Process., vol. 55, no. 5, pp. 2226–2238, May 2007.
[78] K. M. Z. Islam, T. Y. Al-Naffouri, and N. Al-Dhahir, ‘‘On optimum
pilot design for comb-type OFDM transmission over doubly-selective
channels,’’ IEEE Trans. Commun., vol. 59, no. 4, pp. 930–935, Apr. 2011.
[79] F. Qu and L. Yang, ‘‘On the estimation of doubly-selective fading chan-
nels,’’ IEEE Trans. Wireless Commun., vol. 9, no. 4, pp. 1261–1265,
Apr. 2010.
[80] C. Pirak, J. Wang, K. J. R. Liu, and S. Jitapunkul, ‘‘Adaptive channel esti-
mation using pilot-embedded data-bearing approach for MIMO-OFDM
systems,’’ IEEE Trans. Signal Process., vol. 54, no. 12, pp. 4706–4716,
Dec. 2006.
[81] O. Edfors, M. Sandell, J.-J. van de Beek, S. K. Wilson, and
P. O. Borjesson, ‘‘OFDM channel estimation by singular value decom-
position,’’ IEEE Trans. Commun., vol. 46, no. 7, pp. 931–939, Jul. 1998.
[82] T. Zemen and A. F. Molisch, ‘‘Adaptive reduced-rank estimation of non-
stationary time-variant channels using subspace selection,’’ IEEE Trans.
Veh. Technol., vol. 61, no. 9, pp. 4042–4056, Nov. 2012.
[83] T. Y. Al-Naffouri, K. M. Z. Islam, N. Al-Dhahir, and S. Lu, ‘‘A model
reduction approach for OFDM channel estimation under high mobility
conditions,’’ IEEE Trans. Signal Process., vol. 58, no. 4, pp. 2181–2193,
Apr. 2010.
[84] Y.-S. Choi, P. J. Voltz, and F. A. Cassara, ‘‘On channel estimation and
detection for multicarrier signals in fast and selective Rayleigh fad-
ing channels,’’ IEEE Trans. Commun., vol. 49, no. 8, pp. 1375–1387,
Aug. 2001.
VOLUME 4, 2016
[85] K. Zhong, X. Lei, B. Dong, and S. Li, ‘‘Channel estimation in OFDM
systems operating under high mobility using Wiener filter combined basis
expansion model,’’ EURASIP J. Wireless Commun. Netw., vol. 2012,
no. 1, pp. 1–15, 2012.
[86] T. Cui, C. Tellambura, and Y. Wu, ‘‘Low-complexity pilot-aided channel
estimation for OFDM systems over doubly-selective channels,’’ in Proc.
IEEE Int. Conf. Commun. (ICC), vol. 3. May 2005, pp. 1980–1984.
[87] Y. Li, L. J. Cimini, and N. R. Sollenberger, ‘‘Robust channel estimation
for OFDM systems with rapid dispersive fading channels,’’ IEEE Trans.
Commun., vol. 46, no. 7, pp. 902–915, Jul. 1998.
[88] J. K. Tugnait, S. He, and H. Kim, ‘‘Doubly selective channel estimation
using exponential basis models and subblock tracking,’’ IEEE Trans.
Signal Process., vol. 58, no. 3, pp. 1275–1289, Mar. 2010.
[89] X. Hou, S. Li, C. Yin, and G. Yue, ‘‘Two-dimensional recursive least
square adaptive channel estimation for OFDM systems,’’ in Proc. Int.
Conf. Wireless Commun., Netw. Mobile Comput. (WCNM), Sep. 2005,
pp. 232–236.
[90] X. Ma and G. B. Giannakis, ‘‘Maximum-diversity transmissions over
time-selective wireless channels,’’ in Proc. IEEE Wireless Commun. Netw.
Conf. (WCNC), vol. 1. Mar. 2002, pp. 497–501.
[91] X. Ma and G. B. Giannakis, ‘‘Maximum-diversity transmissions over
doubly selective wireless channels,’’ IEEE Trans. Inf. Theory, vol. 49,
no. 7, pp. 1832–1840, Jul. 2003.
[92] X. Ma, G. Leus, and G. B. Giannakis, ‘‘Space-time-Doppler block cod-
ing for correlated time-selective fading channels,’’ IEEE Trans. Signal
Process., vol. 53, no. 6, pp. 2167–2181, Jun. 2005.
[93] K. Fang and G. Leus, ‘‘Space–time block coding for doubly-selective
channels,’’ IEEE Trans. Signal Process., vol. 58, no. 3, pp. 1934–1940,
Mar. 2010.
[94] M.-A. R. Baissas and A. M. Sayeed, ‘‘Pilot-based estimation of time-
varying multipath channels for coherent CDMA receivers,’’ IEEE Trans.
Signal Process., vol. 50, no. 8, pp. 2037–2049, Aug. 2002.
[95] W. Zhou, J. Wu, and P. Fan, ‘‘High mobility wireless communications
with Doppler diversity: Fundamental performance limits,’’ IEEE Trans.
Wireless Commun., vol. 14, no. 12, pp. 6981–6992, Dec. 2015.
[96] J. Wu, ‘‘WLCp1-04: Exploring maximum Doppler diversity by
Doppler domain multiplexing,’’ in Proc. IEEE Global Telecommun.
Conf. (GLOBECOM), Nov./Dec. 2006, pp. 1–5.
[97] X. Luo, Z. Zhang, C. Zhong, and X. Jin, ‘‘Exploiting cooperative
multipath–Doppler diversity in relay-assisted high-speed train commu-
nications,’’ Chin. Sci. Bull., vol. 59, no. 35, pp. 5019–5028, Dec. 2014.
[98] Z. Tang and G. Leus, ‘‘Low-complexity equalization of time-varying
channels with precoding,’’ IEEE Trans. Signal Process., vol. 54, no. 9,
pp. 3642–3648, Sep. 2006.
[99] J. Wu and C. Xiao, ‘‘Optimal diversity combining based on linear estima-
tion of Rician fading channels,’’ IEEE Trans. Commun., vol. 56, no. 10,
pp. 1612–1615, Oct. 2008.
[100] W. Zhou, J. Wu, and P. Fan, ‘‘Maximizing Doppler diversity trans-
missions for high mobility systems with imperfect channel state
information,’’ in Proc. IEEE Int. Conf. Commun. (ICC), Jun. 2014,
pp. 5920–5925.
[101] W. Zhou, J. Wu, and P. Fan, ‘‘Spectral efficient Doppler diversity trans-
missions in high mobility systems with channel estimation errors,’’ in
Proc. IEEE 81st Veh. Technol. Conf. (VTC Spring), May 2015, pp. 1–6.
[102] W. Zhou, J. Wu, and P. Fan, ‘‘Energy and spectral efficient Doppler
diversity transmissions in high-mobility systems with imperfect channel
estimation,’’ EURASIP J. Wireless Commun. Netw., vol 2015, no. 1,
pp. 1–12, May 2015.
[103] M. Morelli, C.-C. J. Kuo, and M.-O. Pun, ‘‘Synchronization techniques
for orthogonal frequency division multiple access (OFDMA): A tutorial
review,’’ Proc. IEEE, vol. 95, no. 7, pp. 1394–1427, Jul. 2007.
[104] H. Cheon and D. Hong, ‘‘Effect of channel estimation error in OFDM-
based WLAN,’’ IEEE Commun. Lett., vol. 6, no. 5, pp. 190–192,
May 2002.
[105] K. Nehra and M. Shikh-Bahaei, ‘‘Spectral efficiency of adaptive
MQAM/OFDM systems with CFO over fading channels,’’ IEEE Trans.
Veh. Technol., vol. 60, no. 3, pp. 1240–1247, Mar. 2011.
[106] Z. Dong, P. Fan, E. Panayirci, and X. Lei, ‘‘Conditional power and
rate adaptation for MQAM/OFDM systems under CFO with perfect and
imperfect channel estimation errors,’’ IEEE Trans. Veh. Technol., vol. 64,
no. 11, pp. 5042–5055, Nov. 2015.
[107] W. Zhang, F. Gao, Q. Yin, and A. Nallanathan, ‘‘Blind carrier frequency
offset estimation for interleaved OFDMA Uplink,’’ IEEE Trans. Signal
Process., vol. 60, no. 7, pp. 3616–3627, Jul. 2012.
473

[108] F. Yang, K. H. Li, and K. C. Teh, ‘‘A carrier frequency offset estimator
with minimum output variance for OFDM systems,’’ IEEE Commun.
Lett., vol. 8, no. 11, pp. 677–679, Nov. 2004.
[109] A. J. Al-Dweik, ‘‘Robust non data-aided frequency offset estimation
technique,’’ in Proc. IEEE Int. Symp. Pers. Indoor Mobile Radio
Commun. (PIMRC), vol. 2. Sep. 2004, pp. 1365–1369.
[110] A. Al-Dweik, A. Hazmi, S. Younis, B. Sharif, and C. Tsimenidis, ‘‘Carrier
frequency offset estimation for OFDM systems over mobile radio
channels,’’ IEEE Trans. Veh. Technol., vol. 59, no. 2, pp. 974–979,
Feb. 2010.
[111] K. Shi, E. Serpedin, and P. Ciblat, ‘‘Decision-directed fine synchro-
nization in OFDM systems,’’ IEEE Trans. Commun., vol. 53, no. 3,
pp. 408–412, Mar. 2005.
[112] T. Roman and V. Koivunen, ‘‘Subspace method for blind CFO estimation
for OFDM systems with constant modulus constellations,’’ in Proc. IEEE
Veh. Technol. Conf. (VTC), vol. 2. Jun. 2005, pp. 1253–1257.
[113] J.-J. Van de Beek, M. Sandell, and P. O. Borjesson, ‘‘ML estimation
of time and frequency offset in OFDM systems,’’ IEEE Trans. Signal
Process., vol. 45, no. 7, pp. 1800–1805, Jul. 1997.
[114] T. L. Lv, H. Li, and J. Chen, ‘‘Joint estimation of symbol timing and car-
rier frequency offset of OFDM signals over fast time-varying multipath
channels,’’ IEEE Trans. Signal Process., vol. 53, no. 12, pp. 4526–4535,
Dec. 2005.
[115] N. Lashkarian and S. Kiaei, ‘‘Class of cyclic-based estimators for
frequency-offset estimation of OFDM systems,’’ IEEE Trans. Commun.,
vol. 48, no. 12, pp. 2139–2149, Dec. 2000.
[116] D. Huang and K. B. Letaief, ‘‘Carrier frequency offset estimation for
OFDM systems using null subcarriers,’’ IEEE Trans. Commun., vol. 54,
no. 5, pp. 813–923, May 2006.
[117] J. Zhu and W. Lee, ‘‘Carrier frequency offset estimation for OFDM
systems with null subcarriers,’’ IEEE Trans. Veh. Technol., vol. 55, no. 5,
pp. 1677–1690, Sep. 2006.
[118] J. Tao, J. Wu, and C. Xiao, ‘‘Estimation of channel transfer function
and carrier frequency offset for OFDM systems with phase noise,’’ IEEE
Trans. Veh. Technol., vol. 58, no. 8, pp. 4380–4387, Oct. 2009.
[119] M. Speth, S. Fechtel, G. Fock, and H. Meyr, ‘‘Optimum receiver design
for OFDM-based broadband transmission—Part II: A case study,’’ IEEE
Trans. Commun., vol. 49, no. 4, pp. 571–578, Apr. 2001.
[120] A. J. Coulson, ‘‘Maximum likelihood synchronization for OFDM using a
pilot symbol: Algorithms,’’ IEEE J. Sel. Areas Commun., vol. 19, no. 12,
pp. 2486–2494, Dec. 2001.
[121] A. J. Coulson, ‘‘Maximum likelihood synchronization for OFDM using
a pilot symbol: Analysis,’’ IEEE J. Sel. Areas Commun., vol. 19, no. 12,
pp. 2495–2503, Dec. 2001.
[122] T. Keller, L. Piazzo, P. Mandarini, and L. Hanzo, ‘‘Orthogonal fre-
quency division multiplex synchronization techniques for frequency-
selective fading channels,’’ IEEE J. Sel. Areas Commun., vol. 19, no. 6,
pp. 999–1008, Jun. 2001.
[123] H. Minn, V. K. Bhargava, and K. B. Letaief, ‘‘A robust timing and
frequency synchronization for OFDM systems,’’ IEEE Trans. Wireless
Commun., vol. 2, no. 4, pp. 822–839, Jul. 2003.
[124] J. Lei and T.-S. Ng, ‘‘A consistent OFDM carrier frequency offset esti-
mator based on distinctively spaced pilot tones,’’ IEEE Trans. Wireless
Commun., vol. 3, no. 2, pp. 588–599, Mar. 2004.
[125] H.-T. Hsieh and W.-R. Wu, ‘‘Maximum likelihood timing and carrier
frequency offset estimation for OFDM systems with periodic
preambles,’’ IEEE Trans. Veh. Technol., vol. 58, no. 8, pp. 4224–4237,
Oct. 2009.
[126] P.-Y. Tsai, H.-Y. Kang, and T.-D. Chiueh, ‘‘Joint weighted least-squares
estimation of carrier-frequency offset and timing offset for OFDM sys-
tems over multipath fading channels,’’ IEEE Trans. Veh. Technol., vol. 54,
no. 1, pp. 211–223, Jan. 2005.
[127] T. Cui and C. Tellambura, ‘‘Joint frequency offset and channel estimation
for OFDM systems using pilot symbols and virtual carriers,’’ IEEE Trans.
Wireless Commun., vol. 6, no. 4, pp. 1193–1202, Apr. 2007.
[128] D. D. Lin, R. A. Pacheco, T. J. Lim, and D. Hatzinakos, ‘‘Joint estimation
of channel response, frequency offset, and phase noise in OFDM,’’ IEEE
Trans. Signal Process., vol. 54, no. 9, pp. 3542–3554, Sep. 2006.
[129] E. P. Simon, L. Ros, H. Hijazi, and M. Ghogho, ‘‘Joint carrier frequency
offset and channel estimation for OFDM systems via the EM algorithm in
the presence of very high mobility,’’ IEEE Trans. Signal Process., vol. 60,
no. 2, pp. 754–765, Feb. 2012.
474
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
[130] W. Zhang, Q. Yin, and F. Gao, ‘‘Joint CFO and DOA estimation for mul-
tiuser OFDMA uplink,’’ EURASIP J. Wireless Commun. Netw., vol. 2012,
no. 1, p. 126, Jun. 2012. [Online]. Available: http://asp.eurasipjournals.
com/content/2012/1/126
[131] Y. Yang and P. Fan, ‘‘Doppler frequency offset estimation and diversity
reception scheme of high-speed railway with multiple antennas on sepa-
rated carriage,’’ J. Modern Transp., vol. 20, no. 4, pp. 227–233, Dec. 2012.
[132] M. Jamalabdollahi and S. Salari, ‘‘RLS-based estimation and tracking
of frequency offset and channel coefficients in MIMO-OFDM systems,’’
Wireless Pers. Commun., vol. 71, no. 2, pp. 1159–1174, Jul. 2013.
[133] X. Dong, X. Li, and D. Wu, ‘‘Recursive maximum likelihood estima-
tion of time-varying carrier frequency offset for orthogonal frequency-
division multiplexing systems,’’ Wireless Commun. Mobile Comput.,
vol. 13, no. 11, pp. 1014–1026, Aug. 2013.
[134] M. Ashour, A. El-Keyi, and A. Sultan, ‘‘Low-complexity Kalman
filter-based carrier frequency offset estimation and tracking for OFDM
systems,’’ in Proc. IEEE Acoust., Speech Signal Process. (ICASSP),
May 2013, pp. 4923–4927.
[135] J. Wu and Y. R. Zheng, ‘‘Oversampled orthogonal frequency divi-
sion multiplexing in doubly selective fading channels,’’ IEEE Trans.
Commun., vol. 59, no. 3, pp. 815–822, Mar. 2011.
[136] X. Cai and G. B. Giannakis, ‘‘Bounding performance and suppress-
ing intercarrier interference in wireless mobile OFDM,’’ IEEE Trans.
Commun., vol. 51, no. 12, pp. 2047–2056, Dec. 2003.
[137] H. Song, J. Kim, S. Nam, T. Yu, and D. Hong, ‘‘Joint Doppler-frequency
diversity for OFDM systems using hybrid interference cancellation in
time-varying multipath fading channels,’’ IEEE Trans. Veh. Technol.,
vol. 57, no. 1, pp. 635–641, Jan. 2008.
[138] T. X. Lai, T. A. Tran, and A. B. Sesay, ‘‘Single-carrier and multi-carrier
space-time block coded transmissions over doubly selective fading chan-
nels,’’ IEE Proc. Commun., vol. 153, no. 5, pp. 746–754, Oct. 2006.
[139] S. Chen, G. Dai, and W. Rao, ‘‘ICI mitigation and diversity gain for
OFDM systems in time-varying multipath fading channels,’’ Eur. Trans.
Telecommun., vol. 22, no. 2, pp. 61–67, 2011.
[140] C.-Y. Hsu and W.-R. Wu, ‘‘Low-complexity ICI mitigation methods for
high-mobility SISO/MIMO-OFDM systems,’’ IEEE Trans. Veh. Technol.,
vol. 58, no. 6, pp. 2755–2768, Jul. 2009.
[141] G. Liu, S. Zhidkov, H. Li, L. Zeng, and Z. Wang, ‘‘Low-complexity
iterative equalization for symbol-reconstruction-based OFDM receivers
over doubly selective channels,’’ IEEE Trans. Broadcast., vol. 58, no. 3,
pp. 390–400, Sep. 2012.
[142] P. Schniter, ‘‘Low-complexity equalization of OFDM in doubly selective
channels,’’ IEEE Trans. Signal Process., vol. 52, no. 4, pp. 1002–1011,
Apr. 2004.
[143] K. Fang, L. Rugini, and G. Leus, ‘‘Low-complexity block turbo equaliza-
tion for OFDM systems in time-varying channels,’’ IEEE Trans. Signal
Process., vol. 56, no. 11, pp. 5555–5566, Nov. 2008.
[144] D. D. Huang, K. Letaief, and J. Lu, ‘‘Bit-interleaved time-frequency
coded modulation for OFDM systems over time-varying channels,’’ IEEE
Trans. Commun., vol. 53, no. 7, pp. 1191–1199, Jul. 2005.
[145] H.-D. Lin and T.-H. San, ‘‘On the diversity order of BICM-OFDM in
doubly selective fading channels,’’ IEEE Trans. Veh. Technol., vol. 62,
no. 4, pp. 1866–1871, May 2013.
[146] A.E.-S. El-Mahdy, ‘‘Adaptive channel estimation and equalization for
rapidly mobile communication channels,’’ IEEE Trans. Commun., vol. 52,
no. 7, pp. 1126–1135, Jul. 2004.
[147] A. Gorokhov and J.-P. Linnartz, ‘‘Robust OFDM receivers for disper-
sive time-varying channels: Equalization and channel acquisition,’’ IEEE
Trans. Commun., vol. 52, no. 4, pp. 572–583, Apr. 2004.
[148] S. Tomasin, A. Gorokhov, H. Yang, and J.-P. Linnartz, ‘‘Iterative inter-
ference cancellation and channel estimation for mobile OFDM,’’ IEEE
Trans. Wireless Commun., vol. 4, no. 1, pp. 238–245, Jan. 2005.
[149] Y. Mostofi and D. Cox, ‘‘ICI mitigation for pilot-aided OFDM mobile
systems,’’ IEEE Trans. Wireless Commun., vol. 4, no. 2, pp. 765–774,
Mar. 2005.
[150] L. Rugini, P. Banelli, and G. Leus, ‘‘Low-complexity banded equalizers
for OFDM systems in doppler spread channels,’’ EURASIP J. Appl.
Signal Process., vol. 2006, no. 1, p. 248, Aug. 2006. [Online]. Available:
http://asp.eurasipjournals.com/content/pdf/1687-6180-2006-067404.pdf
[151] X. Huang and H.-C. Wu, ‘‘Robust and efficient intercarrier interference
mitigation for OFDM systems in time-varying fading channels,’’ IEEE
Trans. Veh. Technol., vol. 56, no. 5, pp. 2517–2528, Sep. 2007.
VOLUME 4, 2016
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
[152] I. Barhumi and M. Moonen, ‘‘MLSE and MAP equalization for transmis-
sion over doubly selective channels,’’ IEEE Trans. Veh. Technol., vol. 58,
no. 8, pp. 4120–4128, Oct. 2009.
[153] M.-L. Ku, W.-C. Chen, and C.-C. Huang, ‘‘EM-based iterative receivers
for OFDM and BICM/OFDM systems in doubly selective channels,’’
IEEE Trans. Wireless Commun., vol. 10, no. 5, pp. 1405–1415,
May 2011.
[154] E. Panayirci, H. Dogan, and H. V. Poor, ‘‘Low-complexity MAP-
based successive data detection for coded OFDM systems over highly
mobile wireless channels,’’ IEEE Trans. Veh. Technol., vol. 60, no. 6,
pp. 2849–2857, Jul. 2011.
[155] H. Liu and P. Schniter, ‘‘Iterative frequency-domain channel estimation
and equalization for single-carrier transmissions without cyclic-prefix,’’
IEEE Trans. Wireless Commun., vol. 7, no. 10, pp. 3686–3691, Oct. 2008.
[156] Z. Dong, P. Fan, and X. Lei, ‘‘Mobility adaptation in OFDM systems over
rapidly time-varying fading channels,’’ in Proc. IEEE Int. Conf. Commun.
Syst. (ICCS), Nov. 2014, pp. 318–322.
[157] A. J. Goldsmith and S.-G. Chua, ‘‘Variable-rate variable-power MQAM
for fading channels,’’ IEEE Trans. Commun., vol. 45, no. 10,
pp. 1218–1230, Oct. 1997.
[158] Z. Dong, P. Fan, E. Panayirci, and P. Mathiopoulos, ‘‘Effect of power
and rate adaptation on the spectral efficiency of MQAM/OFDM sys-
tem under very fast fading channels,’’ EURASIP J. Wireless Commun.
Network., vol. 2012, no. 1, p. 208, Jul. 2012. [Online]. Available:
http://jwcn.eurasipjournals.com/content/2012/1/208
[159] Z. Dong, P. Fan, W. Zhou, and E. Panayirci, ‘‘Power and rate adaptation
for MQAM/OFDM systems under fast fading channels,’’ in Proc. IEEE
75th Veh. Technol. Conf. (VTC-Spring), May 2012, pp. 1–5.
[160] Z. Dong, P. Fan, R. Q. Hu, J. Gunther, and X. Lei, ‘‘On the spectral
efficiency of rate and subcarrier bandwidth adaptive OFDM systems over
very fast fading channels,’’ IEEE Trans. Veh. Technol., to be published.
[161] Z. Dong, P. Fan, and X. Lei, ‘‘Power adaptation in OFDM systems
based on velocity variation under rapidly time-varying channels,’’ IEEE
Commun. Lett., vol. 19, no. 4, pp. 689–692, Apr. 2015.
[162] V. Sethuraman, B. Hajek, and K. Narayanan, ‘‘Capacity bounds for
noncoherent fading channels with a peak constraint,’’ in Proc. Int. Symp.
Inf. Theory (ISIT), Sep. 2005, pp. 515–519.
[163] V. Sethuraman, L. Wang, B. Hajek, and A. Lapidoth, ‘‘Low-SNR capacity
of noncoherent fading channels,’’ IEEE Trans. Inf. Theory, vol. 55, no. 4,
pp. 1555–1574, Apr. 2009.
[164] L. Zheng and D. N. C. Tse, ‘‘Communication on the Grassmann manifold:
A geometric approach to the noncoherent multiple-antenna channel,’’
IEEE Trans. Inf. Theory, vol. 48, no. 2, pp. 359–383, Feb. 2002.
[165] F. Hlawatsch and G. Matz, Eds., Wireless Communications Over Rapidly
Time-Varying Channels. San Diego, CA, USA: Academic, 2011.
[166] E. Telatar, ‘‘Capacity of multi-antenna Gaussian channels,’’ Eur. Trans.
Telecommun., vol. 10, no. 6, pp. 585–595, 1999.
[167] T. L. Marzetta and B. M. Hochwald, ‘‘Capacity of a mobile multiple-
antenna communication link in Rayleigh flat fading,’’ IEEE Trans. Inf.
Theory, vol. 45, no. 1, pp. 139–157, Jan. 1999.
[168] B. M. Hochwald and T. L. Marzetta, ‘‘Unitary space-time modulation for
multiple-antenna communications in Rayleigh flat fading,’’ IEEE Trans.
Inf. Theory, vol. 46, no. 2, pp. 543–564, Feb. 2000.
[169] W. Yang, G. Durisi, and E. Riegler, ‘‘On the capacity of large-MIMO
block-fading channels,’’ IEEE J. Sel. Areas Commun., vol. 31, no. 2,
pp. 117–132, Feb. 2013.
[170] A. Lapidoth and S. M. Moser, ‘‘Capacity bounds via duality with applica-
tions to multiple-antenna systems on flat-fading channels,’’ IEEE Trans.
Inf. Theory, vol. 49, no. 10, pp. 2426–2467, Oct. 2003.
[171] Y. Liang and V. V. Veeravalli, ‘‘Capacity of noncoherent time-selective
Rayleigh-fading channels,’’ IEEE Trans. Inf. Theory, vol. 50, no. 12,
pp. 3095–3110, Dec. 2004.
[172] G. Durisi, V. I. Morgenshtern, and H. Bolcskei, ‘‘On the sensitivity of
continuous-time noncoherent fading channel capacity,’’ IEEE Trans. Inf.
Theory, vol. 58, no. 10, pp. 6372–6391, Oct. 2012.
[173] M. Dorpinghaus, G. Koliander, G. Durisi, E. Riegler, and H. Meyr, ‘‘Over-
sampling increases the pre-log of noncoherent Rayleigh fading channels,’’
IEEE Trans. Inf. Theory, vol. 60, no. 9, pp. 5673–5681, Sep. 2014.
[174] G. Durisi, U. G. Schuster, H. Bolcskei, and S. Shamai (Shitz), ‘‘Noncoher-
ent capacity of underspread fading channels,’’ IEEE Trans. Inf. Theory,
vol. 56, no. 1, pp. 367–395, Jan. 2010.
[175] M. R. Avendi, ‘‘Differential modulation and non-coherent detection
in wireless relay networks,’’ M.S. thesis, Dept. Elect. Comput. Eng.,
Univ. Saskatchewan, Saskatoon, SK, Canada, 2014.
VOLUME 4, 2016
[176] D. Divsalar and M. K. Simon, ‘‘Maximum-likelihood differential detec-
tion of uncoded and trellis coded amplitude phase modulation over
AWGN and fading channels-metrics and performance,’’ IEEE Trans.
Commun., vol. 42, no. 1, pp. 76–89, Jan. 1994.
[177] P. Ho and D. Fung, ‘‘Error performance of multiple-symbol differential
detection of PSK signals transmitted over correlated Rayleigh fading
channels,’’ IEEE Trans. Commun., vol. 40, no. 10, pp. 1566–1569,
Oct. 1992.
[178] L. Lampe, R. Schober, V. Pauli, and C. Windpassinger, ‘‘Multiple-symbol
differential sphere decoding,’’ IEEE Trans. Commun., vol. 53, no. 12,
pp. 1981–1985, Dec. 2005.
[179] B. L. Hughes, ‘‘Differential space-time modulation,’’ IEEE Trans. Inf.
Theory, vol. 46, no. 7, pp. 2567–2578, Nov. 2000.
[180] C. B. Peel and A. L. Swindlehurst, ‘‘Performance of space-time modula-
tion for a generalized time-varying Rician channel model,’’ IEEE Trans.
Wireless Commun., vol. 3, no. 3, pp. 1003–1012, May 2004.
[181] V. Pauli and L. Lampe, ‘‘Tree-search multiple-symbol differential decod-
ing for unitary space-time modulation,’’ IEEE Trans. Commun., vol. 55,
no. 8, pp. 1567–1576, Aug. 2007.
[182] A. Anastasopoulos, K. M. Chugg, G. Colavolpe, G. Ferrari, and R. Raheli,
‘‘Iterative detection for channels with memory,’’ Proc. IEEE, vol. 95,
no. 6, pp. 1272–1294, Jun. 2007.
[183] I. Motedayen-Aval, A. Krishnamoorthy, and A. Anastasopoulos,
‘‘Optimal joint detection/estimation in fading channels with polynomial
complexity,’’ IEEE Trans. Inf. Theory, vol. 53, no. 1, pp. 209–223,
Jan. 2007.
[184] E. Soltanmohammadi and M. Naraghi-Pour, ‘‘Semi-blind data detection
for unitary space-time modulation in MIMO communications systems,’’
IEEE Trans. Commun., vol. 61, no. 12, pp. 5006–5015, Dec. 2013.
[185] E. Chiavaccini and G. M. Vitetta, ‘‘MAP symbol estimation on frequency-
flat Rayleigh fading channels via a bayesian EM algorithm,’’ IEEE Trans.
Commun., vol. 49, no. 11, pp. 1869–1872, Nov. 2001.
[186] H. A. Cirpan and M. K. Tsatsanis, ‘‘Maximum-likelihood estimation of
FIR channels excited by convolutionally encoded inputs,’’ IEEE Trans.
Commun., vol. 49, no. 7, pp. 1125–1128, Jul. 2001.
[187] H. Nguyen and B. C. Levy, ‘‘Blind equalization of dispersive fast fading
Ricean channels via the EMV algorithm,’’ IEEE Trans. Veh. Technol.,
vol. 54, no. 5, pp. 1793–1801, Sep. 2005.
[188] M. Nissila and S. Pasupathy, ‘‘Adaptive Bayesian and EM-based detectors
for frequency-selective fading channels,’’ IEEE Trans. Commun., vol. 51,
no. 8, pp. 1325–1336, Aug. 2003.
[189] J.-Y. Zhang, Z.-H. Tan, Z.-D. Zhong, and Y. Kong, ‘‘A multi-mode
multi-band and multi-system-based access architecture for high-speed
railways,’’ in Proc. IEEE Veh. Technol. Conf. (VTC-Fall), Sep. 2010,
pp. 1–5.
[190] L. Kuang, Z. Wang, M. Xu, and Y. Chen, ‘‘An adaptive handoff triggering
mechanism for vehicular networks,’’ IEICE Trans. Fundam. Electron.
Commun. Comput. Sci., vol. 95, no. 1, pp. 278–285, Jan. 2012.
[191] H. Zhu and K.-S. Kwak, ‘‘Performance analysis of an adaptive handoff
algorithm based on distance information,’’ Comput. Commun., vol. 30,
no. 6, pp. 1278–1288, Mar. 2007.
[192] S. S. Wang, M. Green, and M. Malkawi, ‘‘Adaptive handoff method
using mobile location information,’’ in Proc. IEEE Emerg. Technol.
Symp. (ETS), Sep. 2001, pp. 97–101.
[193] H. Zhu and K. S. Kwak, ‘‘An adaptive hard handoff algorithm for
mobile cellular communication systems,’’ ETRI J., vol. 28, no. 5,
pp. 676–679, Oct. 2006. [Online]. Available: http://dx.doi.org/10.4218/
etrij.06.0206.0101
[194] N. Benvenuto and F. Santucci, ‘‘A least squares path-loss estimation
approach to handover algorithms,’’ IEEE Trans. Veh. Technol, vol. 48,
no. 2, pp. 437–447, Mar. 1999.
[195] L. Tian, J. Li, Y. Huang, J. Shi, and J. Zhou, ‘‘Seamless dual-link handover
scheme in broadband wireless communication systems for high-speed
rail,’’ IEEE J. Sel. Areas Commun., vol. 30, no. 4, pp. 708–718, May 2012.
[196] Z. Liu and P. Fan, ‘‘An effective handover scheme based on antenna
selection in ground–train distributed antenna systems,’’ IEEE Trans. Veh.
Technol, vol. 63, no. 7, pp. 3342–3350, Sep. 2014.
[197] M. Aguado, E. Jacob, J. Astorga, N. Toledo, and M. Berbineau, ‘‘The
cross layer RMPA handover: A reliable mobility pattern aware han-
dover strategy for broadband wireless communication in a high-speed
railway domain,’’ EURASIP J. Wireless Commun. Netw., vol. 2012,
no. 1, p. 298, 2012. [Online]. Available: http://jis.eurasipjournals.com/
content/2012/1/298
475

[198] B. Chow, M.-L. Yee, M. Sauer, A. Ng’Oma, M.-C. Tseng, and C.-H. Yeh,
‘‘Radio-over-fiber distributed antenna system for WiMAX bullet train
field trial,’’ in Proc. IEEE Mobile WiMAX Symp. (MWS), Jul. 2009,
pp. 98–101.
[199] B. Lannoo, D. Colle, M. Pickavet, and P. Demeester, ‘‘Radio-over-fiber-
based solution to provide broadband Internet access to train passengers
[Topics in Optical Communications],’’ IEEE Commun. Mag., vol. 45,
no. 2, pp. 56–62, Feb. 2007.
[200] W. Choi and J. G. Andrews, ‘‘Downlink performance and capacity of
distributed antenna systems in a multicell environment,’’ IEEE Trans.
Wireless Commun., vol. 6, no. 1, pp. 69–73, Jan. 2007.
[201] J. Wang, H. Zhu, and N. J. Gomes, ‘‘Distributed antenna systems for
mobile communications in high speed trains,’’ IEEE J. Sel. Areas Com-
mun., vol. 30, no. 4, pp. 675–683, May 2012.
[202] Y. Zhou, Z. Pan, J. Hu, J. Shi, and X. Mo, ‘‘Broadband wireless communi-
cations on high speed trains,’’ in Proc. 20th Annu. Wireless Opt. Commun.
Conf. (WOCC), Apr. 2011, pp. 1–6.
[203] Q. Luo, W. Fang, J. Wu, and Q. Chen, ‘‘Reliable broadband wireless
communication for high speed trains using baseband cloud,’’
EURASIP J. Wireless Commun. Netw., vol. 2012, no. 285, pp. 1–12,
Sep. 2012. [Online]. Available: http://jwcn.eurasipjournals.com/
content/2012/1/285
[204] W. Luo, R. Zhang, and X. Fang, ‘‘A CoMP soft handover scheme for
LTE systems in high speed railway,’’ EURASIP J. Wireless Commun.
Netw., vol. 2012, no. 196, pp. 1–9, Jun. 2012. [Online]. Available:
http://jwcn.eurasipjournals.com/content/2012/1/196
[205] C.-C. Lin, K. Sandrasegaran, X. Zhu, and Z. Xu, ‘‘On the performance
of capacity integrated CoMP handover algorithm in LTE-advanced,’’
in Proc. IEEE 18th Asia-Pacific Conf. Commun. (APCC), Oct. 2012,
pp. 871–876.
[206] M. Kassar, B. Kervella, and G. Pujolle, ‘‘An overview of vertical han-
dover decision strategies in heterogeneous wireless networks,’’ Comput.
Commun., vol. 31, no. 10, pp. 2607–2620, Jun. 2008.
[207] M. Zekri, B. Jouaber, and D. Zeghlache, ‘‘A review on mobility man-
agement and vertical handover solutions over heterogeneous wireless
networks,’’ Comput. Commun., vol. 35, no. 17, pp. 2055–2068, Oct. 2012.
[208] Alcatel-Lucent, ‘‘LTE for metro railway operations: Unlocking new
opportunities to optimize railway operations with LTE-based train-to-
wayside communications technology,’’ White Paper, 2013. [Online].
Available: http://www.tmcnet.com/tmc/whitepapers/documents/
whitepapers/2013/8272alcatellucentltemetrorailwayoperations.pdf
[209] C. Zhang, P. Fan, Y. Dong, and K. Xiong, ‘‘Service-based high-speed
railway base station arrangement,’’ Wireless Commun. Mobile Comput.,
vol. 15, no. 13, pp. 1681–1694, Sep. 2015.
[210] Y. Dong, P. Fan, and K. B. Letaief, ‘‘High-speed railway wireless commu-
nications: Efficiency versus fairness,’’ IEEE Trans. Veh. Technol., vol. 63,
no. 2, pp. 925–930, Feb. 2014.
[211] Y. Dong, C. Zhang, P. Fan, and P. Fan, ‘‘Power-space functions in high
speed railway wireless communications,’’ J. Commun. Netw., vol. 17,
no. 3, pp. 231–240, Jun. 2015.
[212] Z. Liu, J. Wu, and P. Fan, ‘‘On the uplink capacity of high speed railway
communications with massive MIMO systems,’’ in Proc. IEEE 79th Veh.
Technol. Conf. (VTC Spring), May 2014, pp. 1–5.
[213] Y. Lu, W. Wang, L. Chen, Z. Zhang, and A. Huang, ‘‘Distance-based
energy-efficient opportunistic broadcast forwarding in mobile delay tol-
erant networks,’’ IEEE Trans. Veh. Technol., to be published.
[214] C. Zhang, P. Fan, K. Xiong, and P. Fan, ‘‘Optimal power allocation with
delay constraint for signal transmission from a moving train to base
stations in high-speed railway scenarios,’’ IEEE Trans. Veh. Technol.,
vol. 64, no. 12, pp. 5775–5788, Dec. 2015.
[215] T. Chen, H. Shan, and X. Wang, ‘‘Optimal scheduling for wireless on-
demand data packet delivery to high-speed trains,’’ IEEE Trans. Veh.
Technol., vol. 64, no. 9, pp. 4101–4112, Sep. 2015.
[216] H. Ishii, Y. Kishiyama, and H. Takahashi, ‘‘A novel architecture for
LTE-B: C-plane/U-plane split and phantom cell concept,’’ in Proc. IEEE
Globecom Workshops (GC Wkshps), Dec. 2012, pp. 624–630.
[217] L. Yan and X. Fang, ‘‘Decoupled wireless network architecture for high-
speed railway,’’ in Proc. Int. Workshop High Mobility Wireless Com-
mun. (HMWC), Nov. 2013, pp. 96–100.
[218] L. Yan and X. Fang, ‘‘Reliability evaluation of 5G C/U-plane
decoupled architecture for high-speed railway,’’ EURASIP J. Wireless
Commun. Netw., vol. 2014, no. 127, Aug. 2014. [Online]. Available:
http://jwcn.eurasipjournals.com/content/2014/1/127
476
J. Wu, P. Fan: Survey on High Mobility Wireless Communications
[219] L. Yan, X. Fang, and Y. Fang, ‘‘Control and data signaling decoupled
architecture for railway wireless networks,’’ IEEE Wireless Commun.,
vol. 22, no. 1, pp. 103–111, Feb. 2015.
[220] H. Song, X. Fang, and L. Yan, ‘‘Handover scheme for 5G C/U plane
split heterogeneous network in high-speed railway,’’ IEEE Trans. Veh.
Technol., vol. 63, no. 9, pp. 4633–4646, Nov. 2015.
[221] C. Liu, L. Zhang, M. Zhu, J. Wang, L. Cheng, and G.-K. Chang,
‘‘A novel multi-service small-cell cloud radio access network for
mobile backhaul and computing based on radio-over-fiber technologies,’’
J. Lightw. Technol., vol. 31, no. 17, pp. 2869–2875, Sep. 1, 2013.
[222] J. Li, J. Wu, M. Peng, W. Wang, and V. Lau, ‘‘Queue-aware joint remote
radio beamforming for green cloud radio access networks,’’ in Proc. IEEE
Global Telecommun. Conf. (Globecom), Dec. 2015, pp. 1–6.
[223] J. Li, J. Wu, M. Peng, W. Wang, and V. K. N. Lau, ‘‘Queue-aware joint
remote radio head activation and beamforming for green cloud radio
access networks,’’ in Proc. IEEE Globecom, Dec. 2015.
JINGXIAN WU (S’02–M’06–SM’15) received
the B.S. degree from the Beijing University of
Aeronautics and Astronautics, Beijing, China, in
1998, the M.S. degree from Tsinghua University,
Beijing, China, in 2001, and the Ph.D. degree
from the University of Missouri at Columbia,
Missouri, USA, in 2005, all in electrical engineer-
ing. He is currently an Associate Professor with the
Department of Electrical Engineering, University
of Arkansas, Fayetteville. His research interests
mainly focus on signal processing for large scale networks and wireless com-
munications, including energy efficient information sensing and processing,
green communications, and statistical data analytics. He served as a Co-Chair
of the 2012 Wireless Communication Symposium of the IEEE International
Conference on Communication, and the 2009 and 2015 Wireless Commu-
nication Symposium of the IEEE Global Telecommunications Conference.
He served as an Associate Editor of the IEEE TRANSACTIONS ON VEHICULAR
TECHNOLOGY from 2007 to 2011, and is an Editor of the IEEE TRANSACTIONS
ON WIRELESS COMMUNICATIONS, and an Associate Editor of the IEEE ACCESS.
PINGZHI FAN (M’93–SM’99–F’15) received
the Ph.D. degree in electronics engineering
from Hull University, U.K. He is currently a
Professor and the Director of the Institute of
Mobile Communications with Southwest Jiaotong
University, China. He has over 200 research papers
published in various academic English journals
(the IEEE/IEE/IEICE), and eight books (including
edited) published by John Wiley & Sons Ltd/RSP
(1996), the IEEE Press (2011), Springer (2004),
and Nova Science (2007). He holds 20 granted PCT and Chinese patents. His
current research interests include high mobility wireless communications,
5G technologies, wireless networks for big data, and signal design and
coding. He is a recipient of the UK ORS Award, the Outstanding Young
Scientist Award by NSFC, and the Chief Scientist of a National 973 Major
Research Project. He served as the General Chair or a TPC Chair of a number
of international conferences, and is the Guest Editor-in-Chief and a Guest
Editor or an Editorial Member of several international journals. He is the
Founding Chair of the IEEE VTS BJ Chapter, the IEEE ComSoc CD Chapter,
and the IEEE Chengdu Section. He also served as a Board Member of the
IEEE Region 10, the IET (IEE) Council, and the IET Asia-Pacific Region.
VOLUME 4, 2016