IEEE Communications Magazine • May 2015
2
0163-6804/15/$25.00 © 2015 IEEE
Zhongshan Zhang,
Xiaomeng Chai, and
Keping Long are with the
University of Science and
Technology Beijing.
Athanasios V. Vasilakos
is with the University of
Western Macedonia.
Lajos Hanzo is with the
University of Southampton.
This work was supported
by the key project of the
National Natural Science
Foundation of China
(No. 61431001), the 863
project No.
2014AA01A701, Pro-
gram for New Century
Excellent Talents in Uni-
versity (NECT-12-0774),
the open research fund
of National Mobile
Communications
Research Laboratory
Southeast University
(No. 2013D12), Funda-
mental Research Funds
for the Central Universi-
ties, and the Foundation
of Beijing Engineering
and Technology
Research Center for
Convergence Networks.
INTRODUCTION
The spectral efficiency (SE) of networks has to
be further improved in order to deliver ever
increasing data rates. However, the operational
wireless communication systems usually rely on
half duplex (HD) operations, leading to erosion
of resource exploitation. The promise of radical
full duplex (FD) operation, on the other hand,
improves the achievable SE of wireless commu-
nication systems by always transmitting and
receiving in the entire bandwidth.
The main driving force behind the advances
in FD communications is the promise of nearly
doubled channel capacity compared to conven-
tional HD communications, thus offering the
potential to complement and sustain the evolu-
tion of the fifth generation (5G) technologies
toward denser heterogeneous networks with flex-
ible relaying modes [1]. Recently, a range of the-
oretical and practical aspects of FD
communications have been investigated by quan-
tifying the performance gains of FD modes
(FDMs) [2], which exhibits advantages over the
half-duplex mode (HDM) in terms of either hav-
ing increased throughput or reduced outage
probability (OP), albeit achieved at the cost of
increased complexity. Furthermore, recent
advances in FD communications have increased
both the attainable throughput and the diversity
orders of wireless communication systems. Once
increased hardware/software complexity is toler-
ated to facilitate more sophisticated signal pro-
cessing, it would be possible for an FD device to
reduce the bit error rate (BER). In addition, the
packet loss ratio (PLR) of FDM may also be
reduced, provided that a larger buffer size is
provided by FD devices.
However, as a downside, the FD gain is erod-
ed by self-interference (SI) due to the large
power difference between the power imposed by
a device’s own transmissions and the low-power
received signal arriving from a remote transmit
antenna. Excessive SI may even result in reduced
capacity for FD systems that falls below that of
HD systems. Consensus reached by both indus-
try and academia show that it is critical to per-
form efficient SI suppression/cancellation in
implementing radical FD communication sys-
tems. Apart from the aforementioned physical-
layer issues, the conception of FD medium
access control (MAC) protocols requires sub-
stantial further research. Experience indicates
that FD schemes may not always outperform
their HD counterparts, and hybrid schemes that
switch between HDM and FDM can also be
ABSTRACT
The wireless research community aspires to
conceive full duplex operation by supporting
concurrent transmission and reception in a sin-
gle time/frequency channel for the sake of
improving the attainable spectral efficiency by a
factor of two as compared to the family of con-
ventional half duplex wireless systems. The main
challenge encountered in implementing FD
wireless devices is that of finding techniques for
mitigating the performance degradation imposed
by self-interference. In this article, we investigate
the potential FD techniques, including passive
suppression, active analog cancellation, and
active digital cancellation, and highlight their
pros and cons. Furthermore, the troubles of FD
medium access control protocol design are dis-
cussed for addressing the problems such as the
resultant end-to-end delay and network conges-
tion. Additionally, an opportunistic decode-and-
forward-based relay selection scheme is analyzed
in underlay cognitive networks communicating
over independent and identically distributed
Rayleigh and Nakagami-m fading channels in
the context of FD relaying. We demonstrate that
the outage probability of multi-relay cooperative
communication links can be substantially
reduced. Finally, we discuss the challenges
imposed by the aforementioned techniques and
a range of critical issues associated with practical
FD implementations. It is shown that numerous
open challenges, such as efficient SI suppression,
high-performance FD MAC-layer protocol
design, low power consumption, and hybrid
FD/HD designs, have to be tackled before suc-
cessfully implementing FD-based systems.
FULL DUPLEX COMMUNICATIONS
Zhongshan Zhang, Xiaomeng Chai, Keping Long, Athanasios V. Vasilakos, and Lajos Hanzo
Full Duplex Techniques for 5G Networks:
Self-Interference Cancellation,
Protocol Design, and Relay Selection
HANZO_LAYOUT_Author Layout 4/24/15 4:27 PM Page 2
IEEE Communications Magazine • May 2015
3
developed for adaptively exploiting the radio
resources, while at the same time maximizing the
SE [3]. Again, an FD scheme may not always
outperform its HD counterpart, requiring a
hybrid HD/FD scheme to be implemented to
gain an advantage over either of the individual
schemes. In this article, we survey/compare dif-
ferent FD techniques. Some existing SI cancella-
tion techniques such as passive suppression,
active analog, and digital cancellation are dis-
cussed. Furthermore, the critical issues associat-
ed with FD-based MAC-layer protocols are also
studied. Finally, the choice of the optimal relay
selection scheme conceived for FDM is elaborat-
ed on, followed by a variety of new directions
and open problems. The main contributions of
this article include
• Surveying the critical issues related to FD
transmissions from a physical-layer perspec-
tive relying on SI suppression
• Giving cognizance to the MAC-layer proto-
cols
• Proposing an FD-based opportunistic
decode-and-forward (DF)-based relay selec-
tion scheme in the context of underlay cog-
nitive networks and analyzing the OP of the
multi-relay cooperative communication
links
• Outlining several challenges associated with
FDM-based device/system realizations
• Discussing both the advantages and draw-
backs of various FD techniques, while iden-
tifying their challenges and new directions
The remainder of this article is organized as fol-
lows. The classification of both passive and
active SI suppression is detailed next. Typical
FD MAC-layer protocols, such as the FD-MAC
technique [4], are then discussed, followed by
several critical issues related to the associated
practical implementation and commercial real-
izations. We then propose an opportunistic
FDM relay selection scheme, followed by a
range of open challenges and the future direc-
tions of FD communications. Finally, our con-
clusions are provided.
SELF-INTERFERENCE CANCELLATION
Existing studies [5, 6] showed that it is critical to
accurately measure and suppress the SI in FD
communication. For instance, as revealed in [7],
the SI power as well as spatial reuse may sub-
stantially reduce the FD gain over the HDM in
terms of the network-level capacity, rendering it
well below 2 in common cases. However, if the
SI level at the input of the FD relay (i.e., after
performing SI suppression) can be at least 3 dB
lower than the noise level, the remaining SI may
Figure 1. Practical implementable SI suppression algorithms and their performance comparison.
Digital
cancellation
S
I power reduction due to path loss effect
FPGA
RSSI
Passive
suppression
Algorithm
Transmit
power
Center
frequency
Antenna
distances
Antenna
separations
Cancellation
capability
Full duplex
gain
Bandwidth
C
ompact/separated
a
ntennas
5
dBm
2
.6 GHz
2
0 mm
5 m
4
8 dB
70 dB
1
00 MHz
A
ntenna
c
ancellation
2
.4 GHz
λ/2 60 dB 1.84
100 MHz
ZigZag 2.4 GHz 1.25
Antenna
separation
–5 dBm
~15 dBm
2.4 GHz
20 cm
40 cm
39 dB
45 dB
6
25 kHz
Directional
diversity
12 dBm 2.4 GHz
10 m
15 m
≥45°
≥90°
1.6~1.9
≥1.4
20 MHz
x
signal
(
t)
z
noise
(
t)
x
Self-interference
y
residual
RX signal pathTX signal path
D
etector 3
D
etector 2
D
etector 1
C
oupler
C
oupler
C
oupler
I
nverter
Tx
Power
splitter
RF reference
A
GC
C
ombiner
Spectrum
analyzer
β
Analog
cancellation
ADC
I
nterference
c
ancellation
L
oop channel
c
ancellation
D
b
Relay
h
II
h
II
h
II
X
(i)
R
x
Rx Tx
Amplitude and phase
E
qual amplitude Inverse phase
HANZO_LAYOUT_Author Layout 4/24/15 4:27 PM Page 3
IEEE Communications Magazine • May 2015
4
not seriously degrade the end-to-end throughput
[8]. SI cancellation techniques are usually classi-
fied into passive and active suppressions, as
shown in Fig. 1.
PASSIVE SI SUPPRESSION
P
assive SI suppression is defined as the signal-
power attenuation imposed by the path loss due
to the physical separation between the transmit
and receive antennas of the same device. Typical
passive SI suppression techniques include:
Directional SI suppression: In this technique,
the main radiation lobes of the transmit/receive
antennas of an FD device have minimal intersec-
tion, enabling the SI to be partially suppressed
p
rior to the receiver’s RF front-end.
Antenna separation and SI cancellation:
Increasing the path loss between the transmit/
receive antennas constitutes an effective
approach to attenuate the SI power, in which
method a higher antenna separation implies bet-
ter SI suppression performance. When relying
on antenna separation, the natural isolation may
also exploit the surrounding buildings or the
beneficial inclusion of a shielding plate, provided
that strict restrictions imposed on the device size
can be satisfied.
ACTIVE SELF-INTERFERENCE SUPPRESSION
In [9], active SI suppression methods were shown
experimentally to be capable of facilitating FD
communication at ranges up to 6 m and at trans-
mit powers typical of WiFi devices, revealing
that the interference level can be reduced by 50
dB and 40 dB under static and dynamically fad-
ing interference channel scenarios, respectively,
if an RF SI canceller is combined with a base-
band canceller. The family of active suppression
techniques can be subdivided into analog cancel-
lation, digital cancellation, and combined ana-
log/digital cancellation, as discussed below.
Analog Cancellation — In analog cancellation,
the family of time-domain (TD) cancellation
algorithms such as training-based methods can
be employed by both single-input single-output
(SISO) and multiple-input multiple-output
(MIMO) based techniques, where the latter may
perform SI suppression by exploiting the spatial
diversity achieved by the associated multiple
transmit and/or receive antennas.
•Classic TD training-based methods can be
beneficially utilized for estimating the SI leak-
age, while facilitating reliable SI cancellation.
Asymmetric complex signals, in which the inputs
are chosen to be complex but not circularly sym-
metric, can also be utilized for mitigating the SI
in single-antenna-aided FDM relays under DF
relaying. The optimum SI cancellation weight
vectors can be exploited by increasing the signal-
to-noise ratio (SNR) of the source relay and
relay destination links, thus beneficially
improving the attainable throughput of FD
relaying channels.
•The increased degree of freedom (DoF)
offered by the spatial domain (SD) antenna
arrays of MIMO systems may be utilized to pro-
vide a range of new solutions for SI cancellation.
In MIMO aided FD systems, relays are capable
of operating in either the antenna-partitioning-
based mode (i.e., all antennas operating in the
FDM but partitioned into transmit and receive
antenna sets) or antenna-sharing-based mode
(i.e., allowing antennas to be utilized more effi-
ciently by exploiting the increased dimensions of
MIMO channels and/or by relying on time-divi-
s
ion duplexing, TDD, aided channel reciprocity).
Digital Cancellation — Since analog SI cancel-
lation methods are never perfect, the residual SI
after analog cancellation should be further
reduced with the aid of digital cancellation. Of
the existing digital cancellation protocols, ZigZag
[10] exhibits a significant advantage in terms of
the achievable FD gains. Note that ZigZag
i
mposes no change on the conventional IEEE
802.11 MAC protocols when there is no colli-
sion, thus maintaining the same throughput as if
the colliding packets were scheduled a priori in
separate time slots in the presence of transmis-
sion collisions. It has been observed that 10 per-
cent of the transmitter-receiver pairs of a
wireless network often experience severe packet
loss due to packet collisions imposed by statisti-
cal channel multiplexing. The asynchronous
nature of successive collisions can be successfully
exploited in ZigZag to address the problem of
high packet loss rate (PLR). By using ZigZag,
the average PLR at hidden terminals was shown
to be reduced from 72.6 to about 0.7 percent,
while improving the average throughput by 25.2
percent compared to the conventional IEEE
802.11 standards.
Performance Comparison — The SI suppres-
sion capabilities of some typical algorithms are
characterized in Fig. 1. Although numerous
sophisticated techniques have been proposed for
performing SI cancellation in FD devices, both
advantages and disadvantages are exhibited in
the context of each approach, as shown in Table 1.
OPEN RESEARCH ISSUES IN SI SUPPRESSION
Although passive SI suppression techniques are
capable of attenuating the SI power in propor-
tion to the path loss, enabling a higher antenna
separation usually requires a larger or even
infeasible device size. More detrimentally,
increasing the antenna separation implies a
degradation of the SI channel estimation. Fur-
thermore, numerous additional challenges have
to be addressed in the context of the existing
active SI suppression techniques. For instance,
the achievable SI cancellation capability may be
limited by relying on standalone analog or digital
cancellation. It is thus rather critical to effective-
ly balance the roles of the analog- and digital-
domain functions in the overall SI cancellation,
carefully revealing the overall benefits of com-
bined analog/digital cancellation. In the follow-
ing, a number of possible solutions to the
above-mentioned challenges should be proposed.
Antenna configuration for practical size-lim-
ited FD devices: In passive SI suppression
schemes, the best antenna configuration in terms
of the attainable SI suppression can be achieved
upon installing the transmit and receive anten-
nas at the opposite sides of the device to create
sufficient separation, requiring the device size to
be large enough.
S
ince the analog SI
cancellation methods
are never perfect,
the residual SI after
analog cancellation
should be further
reduced with the aid
of digital cancella-
tion. In the existing
d
igital-cancellation
protocols, ZigZag
exhibits a significant
advantage in terms
of the achievable
FD gains.
HANZO_LAYOUT_Author Layout 4/24/15 4:27 PM Page 4
IEEE Communications Magazine • May 2015
5
Combination of active and passive SI sup-
pressions: Since none of the individual cancella-
tion techniques is capable of satisfying the system
requirements in terms of the attainable SI can-
cellation capability, a high-capability cancellation
scheme by combining the active and passive
methods is necessarily developed.
Low-complexity spatial-domain suppression
approaches: Many of the existing spatial domain
SI suppression methods relying on complex
matrix computations may significantly erode the
FD gains owing to their infeasibility. Therefore,
low-complexity algorithms conceived for high-
dimensional MIMO channels are capable of dra-
Table 1. Performance comparison among variant SI suppression algorithms.
Category Algorithm
T
x
× R
x
Advantage Disadvantage
Passive
suppression
Directional
diversity
Antenna
separation
1) SI attenuated due to path loss
2) Decreases inter-device interference
3) Improves power efficiency
4) More separation implies a better attenuation
of SI signal
1) Performance depends highly on AS and
beam pattern
2) AS is restricted by variant factors such
as device size and interference channel
estimation accuracy
3) Restriction applications to SISO
Active suppression
Analog cancellation
Antenna
Cancellation
2 × 1
1) Easy to implement
2) High cancellation capability
3) Robust in narrowband systems
1) Broadband-induced loss
2) Degrades the received signal
3) Limited transmit power
4) Requires fixed AS
Pre-Nulling M × 1
1) Simple to implement
2) No influence on receiver BER
3) Stringent requirements on antenna isolation
are required
1) SI channel estimation is required
2) Designed specifically for flat-fading
channels
AFC 1 × M
1) Low complexity
2) Needs no training sequence
3) No delay insertion in the relay
4) Compensates for multipath propagation
The second-order statistical information
of the source signal is required to be
exploited by the filter design
Pre-Coding/
Decoding
M × M
1) Better than pre-nulling
2) Enables advanced optimization
3) Capacity optimization
1) Requires SI estimation
2) Requires SVD of SI channel matrix
Block
Diagonization
M × M
1) Outpeforms ZF beamforming
2) Precoding with adaptive power allocation to
optimize the sum rate
1) CSI is required by the base station
2) SVD is required
3) Power allocation satisfies KKT conditions
ZF Filters M × M
1) High capacity for a high SNR
2) Multiple spatial streams are supported in the
MIMO relay
1) Perfomance loss in low-SNR regions
2) SVD is required
Optimal
Eigenbeam-
forming
M × M Power of the residual SI is minimized
1) Beam selection matrices are calculated
2) SVD is required
Maximum SIR M × M
1) Improves the useful signal
2) Suppresses both SI and noise
1) High complexity in deriving the opti-
mum matrices
2) Channel attenuation highly impacts the
performace
MMSE
Filtering
M × M
1) Improves the useful signal
2) Suppresses both SI and noise
High complexity
TAS M × M
1) Has a low complexity
2) Avoids losses in low-SNR regions
3) Adaptivity to varying SIRs
1) High-dimensional MIMO complicates
the best subset selection
2) Unique solution for the best subset
selection is not always achievable
Digital
cancellation
1) Residual SI after analog cancellation can be
eliminated in digital domain
2) Modulation independence
3) Addresses hidden terminal problem
4) High collision-combating capability
1) Quantization noise cannot be reduced
2) Becomes unneccessary if preceded by a
powerful analog cancellation
3) Limited cancellation capability
HANZO_LAYOUT_Author Layout 4/24/15 4:27 PM Page 5
IEEE Communications Magazine • May 2015
6
matically improving the SI cancellation capabili-
ty at a reasonable hardware/software cost.
Transmit power control for improving SI
suppression: A higher transmit power will defi-
nitely lead to a lower SI channel estimation
error, but the absolute level of the residual SI
power may still increase for a high SI power;
however, the ratio between the residual error
and the overall SI might be reduced.
MAC LAYER PROTOCOL DESIGN
FOR FULL DUPLEX SYSTEMS
Apart from the aforementioned physical-layer
solutions, FD research opportunities have also
been explored in the context of efficient MAC
protocols for addressing the challenges of long
end-to-end delays of network congestion and the
hidden terminal problems. For instance, in [4], a
new MAC protocol referred to as FD-MAC was
developed and implemented for infrastructure-
based WiFi-like networks to provide opportuni-
ties for all the accessed nodes while trying to
maximize the overall network throughput and
maintaining fairness to all users simultaneously.
In order to satisfy the above-mentioned require-
ments, three mechanisms, shared random back-
off (SRB), snooping, and virtual contention
resolution, can be employed, as illustrated in
Fig. 2. FD-MAC is capable of guaranteeing
seamless wireless access while maximizing the
FD gains. Experimental results showed that FD-
MAC achieves a throughput gain of up to 70
percent over its comparable HD counterpart [4].
FD REALIZATIONS IN
PRACTICAL SYSTEMS
Although very few FD realizations have been
implemented in commercial systems due to the
technical and/or economic challenges, a substan-
tial amount of related research has already been
undertaken by addressing several challenges in
this context, discussed below.
HARDWARE LIMITATIONS
In [11], the performance of co-channel FDM-
based MIMO nodes was analyzed in the context
of modeling their realistic hardware characteris-
tics. Theoretically, an FD system having an infi-
nite dynamic range and perfect channel
estimation can perfectly eliminate the SI signal.
However, the hardware limitations, including
transmit/receive signal quantization, nonlineari-
ties, in-phase and quadrature (I/Q) mismatch,
and so on, all might erode the practical imple-
mentations of FD systems.
RECEIVER COMBINING
Apart from the impairments imposed by SI sig-
nals and the above-mentioned hardware limita-
tions, another challenge comes from the fact
that FD-based systems might not be capable of
invoking some sophisticated combining schemes
such as maximum ratio combining (MRC) unless
the source node and the FD-based relay are per-
fectly phase-synchronized. In order to address
this challenge, a co-phasing scheme can be
employed in the direct and relay links, facilitat-
Figure 2. Mechanisms of FD-MAC protocol [4] — shared random backoff (SRB), snooping, and virtu-
al contention resolution — can be employed for addressing the problem of the hidden terminal.
N
ode
A
N
ode
B
N
ode
C
H
idden terminal problem
F
D-MAC solutions
S
RB
•
Once a pair of nodes have
more packets for each other,
the SRB field in the FD-MAC
header can be used to share a
backoff counter with each
other;
•
Both nodes will then perform a
coordinated backoff for a
common duration;
•
The channel is free so as to be
contended for and captured by
the other nodes.
•
Nodes A and C are out of each other’s range;
•
When sending packets simultaneously to node
B, nodes A and C cannot detect a collision
while transmitting;
• CSMA/CD does not work, and collision occurs.
C
ollision!
S
nooping
Current buffer
E
nables one more
r
ound of FDM
S
witches
t
o HDM
Recordering
maximizing
throughput
Packet
for M
1
Packet
for M
2
•
The nodes observe the GFD-MAC
headers of all ongoing
transmissions within the radio
coverage;
•
If the ongoing transmissions
between the AP and other
nodes form a “clique” or
hidden node with themselves,
the Snooping mechanism
allows nodes to discover and
estimate their local topology.
• When the AP is capable of observing
multiple packets in its buffer, it decides
which packet should be served first.
V
irtual contention resolution
Theoretically, an FD
system having an
infinite dynamic
range and perfect
channel estimation
can perfectly elimi-
nate the SI signal.
H
owever, the hard-
ware limitations,
including
transmit/receive sig-
nal quantization,
non-linearities, in-
phase and quadra-
ture mismatch, etc.
all might erode the
practical implementa-
tions of FD systems.
HANZO_LAYOUT_Author Layout 4/24/15 4:27 PM Page 6