In-Band Full-Duplex Interference for Underwater
Acoustic Communication Systems
Bilal A. Jebur
College of Electronics Engineering
Ninevah University
Mosul, Iraq
bilal.jebur@uoninevah.edu.iq cornelius.healy@newcastle.ac.uk
Jeffrey Neasham
School of Engineering
Newcastle University
Newcastle upon Tyne, UK
jeff.neasham@ncl.ac.uk
Abstract—This work presents an adaptive self-interference
cancellation (SIC) method for in-band full-duplex underwater
acoustic (IBFD-UWA) systems along with a model for the self-
interference (SI) for shallow-water acoustic channels. The pro-
posed system utilizes orthogonal frequency division multiplexing
with quadrature phase shift keying modulation to exchange
information between two nodes operating in IBFD mode. The
proposed adaptive SIC scheme employs the normalized least-
mean-square (NLMS) algorithm to suppress the SI signal and
avoid saturating the local analog-to-digital (ADC) converter.
Unlike existing research works, we investigate the effect of
imperfect detection of the signal of interest on the ability of the
SIC to diminish the SI signal. We provide experimental results
to support the SI model developed and simulation results to
demonstrate the ability of the proposed adaptive SIC scheme to
mitigate the SI signal to approximately the level of the ambient
noise.
I. I NTRODUCTION
Underwater acoustic (UWA) communication systems face
several challenges, such as long multipath delay spread caus-
ing intersymbol interference over hundreds of information
symbols, motion-induced Doppler spread causing severe sym-
bol dilation, and extremely limited system bandwidth amongst
others [1], [2]. The limited bandwidth of the UWA systems has
motivated this research to focus on exploiting the full-duplex
(FD) communication mode to improve the attained capacity of
UWA communication systems. FD implies concurrent trans-
mission and reception over the same frequency band, hence,
it is also referred to as in-band full-duplex (IBFD) communi-
cations. However, IBFD mode of communication comes with
a formidable challenge, i.e. the strong self-interference (SI)
signal, which can swamp the signal of interest at the analog-
to-digital converter (ADC). Thus, it is essential to effectively
suppress the SI signal before the ADC to enable the FD mode.
Hence, in the last decade, the researchers have considered FD
in the RF communication systems and focused on the self-
interference cancellation (SIC) schemes [3], [5]–[11].
978-1-7281-1450-7/19/$31.00 ©2019
Authorized licensed use limited to: IEEE
IEEE Associate Editors. Downloaded on June 18,2021 at 10:07:50 UTC from IEEE Xplore. Restrictions apply.
Cornelius T. Healy Charalampos C. Tsimenidis
School of Engineering School of Engineering
Newcastle University Newcastle University
Newcastle upon Tyne, UK Newcastle upon Tyne, UK
charalampos.tsimenidis@ncl.ac.uk
Jonathon Chambers
Department of Engineering
University of Leicester
Leicester, UK
jonathon.chambers@le.ac.uk
Thus far, limited research works have considered the IBFD
mode in UWA communication systems. The feasibility of
utilizing frequency division multiple access (FDMA) and code
division multiple access (CDMA) schemes to achieve FD-
UWA communication was investigated in [12]. The results
of this work have shown that the CDMA techniques are
more bandwidth-efficient than the FDMA techniques. In [13],
a CDMA scheme along with a spread spectrum (SS) tech-
nique, are exploited to enable the FD mode in an orthogo-
nal frequency-division multiplexing (OFDM) based multi-user
acoustic communication system. Moreover, in [14], a model
of the path loss of the multi-path SI channel in the acoustic
FD systems was suggested along with a self-interference
cancellation (SIC) scheme for the FD-UWA nodes, which
comprises a combination of analog and digital SIC techniques.
Next, in [15], the authors presented a digital SIC scheme
that employs a time reversal technique in conjunction with
soft-decision feedback equalization (SDFE) to suppress the SI
signal and eliminate the intersymbol interference (ISI) of the
signal received from the remote node.
In this paper, we present the first experimental measure-
ments of the SI channel in IBFD-UWA systems along with
a model of this channel, which is fundamental in order
to investigate the challenges associated with the FD mode
in UWA communication systems. Moreover, we propose an
adaptive SIC scheme that exploits the normalized least-mean-
square (NLMS) algorithm to diminish effectively the SI signal
before the local ADC. Unlike the existing SIC techniques,
we use the detected signal to reconstruct the signal received
from the distant node and subtract it from the received signal
after the SIC to obtain the residual SI signal and feed it to
the NLMS algorithm. This improves the performance of the
proposed adaptive SIC during the communication between the
nodes.
The paper is organized as follows. Section II illustrates
the system model along with the SI channel model and the
SIC scheme. The results are shown in Section III. Finally, the
conclusions of this paper are presented in Section IV. 𝑑𝑅𝑥  

II. S YSTEM M ODEL
In this section we present the architecture of the IBFD-UWA 𝑑𝑤   𝑑𝑇 𝑥  
node along with the SI channel model and measurement.
A. Transmitter Structure
We consider two UWA nodes utilizing orthogonal frequency 
division multiplexing (OFDM) with an uncoded quadrature
phase shift keying (QPSK) modulation scheme to exchange
information in IBFD mode. First, a 𝐾-points inverse discrete
Fourier transform (IDFT) operation is performed on a block of
𝐾 QPSK symbols, X𝑙 , to obtain the signal to be transmitted Fig. 1. Full-Duplex scenario for shallow-water acoustic communications.
from the local node in the time domain, which is given as
90
𝐾−1
1 ∑ 𝑗2𝑛𝑘
120 0 dB 60
𝑥𝑙 (𝑛) = √ 𝑋𝑙 (𝑘)𝑒 𝐾 , 𝑛 = 0, 1, . . . , 𝑁 − 1. (1)
𝐾 𝑘=0 150 -5 dB 30
-10 dB
To cope with multipath induced ISI, a guard interval is
inserted at the beginning of the sequence x𝑙 . This guard 180
-15 dB
0
interval, referred also to as the cyclic prefix (CP), which
is created by appending the last 𝐿𝐶𝑃 samples of x𝑙 to the
beginning of the original sequence. The CP length, 𝐿𝐶𝑃 , in 210 330
the proposed system is chosen such that 𝐿𝐶𝑃 ≥ 𝐿𝑓 , where, 240 300
𝐿𝑓 represents the excess delay spread of the impulse response, 270
h𝑓 , of the multipath channel between the nodes. Consequently,
Fig. 2. Toroidal beampattern in 𝑧𝑦 and 𝑧𝑥 directions.
the transmitted OFDM block is given as
[ ]𝑇
x𝑙𝐶𝑃 = 𝑥𝑙 (𝐾 − 𝐿𝐶𝑃 ), ...., 𝑥𝑙 (𝐾 − 1), 𝑥𝑙 (0), ...., 𝑥𝑙 (𝐾 − 1) ,
(2) since incident angles close to zero are considered. In practice,
which is then transmitted over the channel between the nodes 10 is selected for cylindrical and 20 for free-space spreading.
and the SI channel. The next subsection shows the model and In this paper, 𝜉 is chosen as 15 for practical spreading. The
empirical measurement of the SI channel. frequency-dependent, sea-water attenuation factor, 𝛼(𝑓 ), is
given as
B. SI Channel Model
𝑏𝑓0 𝑐𝑓1
In this subsection, we model the SI propagation channel. 𝛼(𝑓 ) = 𝑎𝑓 2 + ( )2 + ( )2 (4)

We assume a typical shallow-water scenario with a water- Fresh water atten. 1 + 𝑓𝑓0 1 + 𝑓𝑓1
column depth 𝑑𝑤 as shown in Fig. 1. To maximize the direct      
MgSO4 relax. Boric acid relax.
path attenuation between the transmitter, Tx, and receiver, Rx,
they are positioned as far apart as possible at depths, 𝑑𝑇 𝑥 where
and 𝑑𝑅𝑥 , respectively. It is worth noting that this arrangement
𝑎 = 1.3 × 10−7 + 2.1 × 10−10 (𝑇 − 38)2 , (5)
may not result in the most convenient deployment scenario,
however, since the equations developed are generic the Tx and 𝑏 = 2𝑆 × 10−5 , (6)
Rx depths can be easily adjusted to obtain different scenarios.
In practice, higher direct path attenuation can be achieved by 𝑓0 = 50(𝑇 + 1), (7)
utilizing sound absorbing obstructions. Both Tx and Rx utilize
toroidal type of transducers with the beam pattern illustrated 𝑐 = 1.2 × 10−4 , (8)
in Fig. 2 exhibiting 𝛼𝑧 = −15 dB of attenuation in the 𝑧𝑦
𝑓1 = 10(𝑇 −4)/100 , (9)
and 𝑧𝑥 planes.
The transmission loss in dB is calculated using [16] where 𝑆 is the salinity in h and 𝑇 is the temperature in ∘ C.
For the calculation of 𝛼(𝑓 ), we assume 𝑆 = 35h, 𝑇 = 14 ∘ C,
𝑇 𝐿 = 2𝛼𝑧 + 𝜉 log10 (𝑟) + 𝛼(𝑓 )𝑟 (3)
and 𝑓 is assumed to be the carrier frequency chosen as 𝑓𝑐 =
where 𝑟 is the propagation distance in m, 12 kHz.
The multiplier 𝜉 can exhibit huge variations between 10 To compute the total transmission loss of the locally propa-
and 20 depending on the surface and bottom losses especially gating waves, we compute the associated propagation delay per

Authorized licensed use limited to: IEEE Associate Editors. Downloaded on June 18,2021 at 10:07:50 UTC from IEEE Xplore. Restrictions apply.
    
𝜌𝑤,𝑎 ≈ −1 80
𝜏1
ÊÜ 75

𝑑𝑤 =  70
𝜏0

T L (dB)
65
TL for direct path
ÌÜ 60 TL for τS1 (n)
𝜏2 TL for τB1 (n)
𝜌𝑤,𝑏 ≈ 1
55 TL for τS2 (n)
TL for τB1 (n)
Fig. 3. Local multipath propagation.
50
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
path and convert it to distance assuming an isothermal sound- τ (s)
speed propagation profile with 𝑐0 = 1500 m/s. In case 1, Fig. 4. Transmission loss for local multipath propagation.
shown in Fig. 3, we follow the delays of the wave propagating
upwards. The delays of the direct path, surface and bottom
reflections are computed as
𝑑𝑇 𝑥 − 𝑑𝑅𝑥 0.1
𝜏0 = , (10)
𝑐0
𝜏𝑆1 (𝑛) = 𝜏0 + 𝜏1 + (𝑛 − 1)𝜏𝑑𝑤 , (11) 0.05

𝜏𝐵1 (𝑛) = 𝜏0 + 𝜏1 + 𝑛𝜏𝑑𝑤 ,

hSI (τ )

(12)
0
where 𝑛 = 1, 2, . . . , 𝑁𝑟 , with 𝑁𝑟 being the number of
observed reflections, and
-0.05
2𝑑𝑅𝑥
𝜏1 = , (13)
𝑐0
-0.1
2𝑑𝑤
𝜏𝑑𝑤 = . (14) 0 0.1 0.2 0.3 0.4 0.5 0.6
𝑐0
τ (s)
In case 2, we follow the wave propagating initially downwards.
The delays of surface and bottom reflections are computed as Fig. 5. Computed channel impulse response for SI.

𝜏𝑆2 (𝑛) = 𝜏0 + 𝜏1 + 𝜏2 + (𝑛 − 1)𝜏𝑑𝑤 (15)

respect to 𝑧 axis) will be additionally propagating resulting a
𝜏𝐵2 (𝑛) = 𝜏0 + 𝜏2 + (𝑛 − 1)𝜏𝑑𝑤 (16)
more diffuse impulse response.
where Next, to validate our model of the SI channel, we exper-
2(𝑑𝑤 − 𝑑𝑇 𝑥 ) imentally measured the SI channel as depicted in Fig. 6. A
𝜏2 = . (17)
𝑐0 closer inspection of Fig. 6 shows that the envelop of the
We assume that for normally incident waves at the sea surface SI signal reverberation closely match computed SI channel
boundary, the pressure reflection coefficient is 𝜌𝑤,𝑎 ≈ −1, impulse response in Fig. 5, which supports our model of the
while at the sea-bottom, 𝜌𝑤,𝑏 ≈ 1 [16]. This results in even SI signal.
surface and bottom reflections that add constructively at Rx; in
contrast, odd surface and bottom reflections add destructively. C. Receiver Structure
The resulting delays and corresponding transmission losses are The received signal at the local node can be expressed as
shown in Fig. 4 for 𝑁𝑟 = 10, while the associated channel
impulse response is depicted in Fig. 5. The envelope of the 𝑦𝑙 (𝑛) = 𝑠𝑓 (𝑛) + 𝑠𝐼 (𝑛) + 𝑤(𝑛), (19)
channel impulse response can be empirically approximated as
where 𝑠𝑓 (𝑛) represents the signal received from the far node
0.15
∣ℎ𝐼 (𝜏 )∣ = 𝜏 (18) and is given as
1 + 0.04 s
𝑀
∑ −1
It is worth noting that in practice it is expected that beyond the
𝑠𝑓 (𝑛) = ℎ𝑓 (𝑛 − 𝑚)𝑥𝑓 (𝑚), (20)
normally incident waves, signal waves with small angles (with 𝑚=0

Authorized licensed use limited to: IEEE Associate Editors. Downloaded on June 18,2021 at 10:07:50 UTC from IEEE Xplore. Restrictions apply.
 After removing the CP and performing the discrete Fourier
1.5
transform (DFT), the received signal in the frequency domain
SI
signal is given as
1
SI signal 𝑌𝑙 (𝑘) = 𝑆𝑓 (𝑘) + 𝑅𝐼 (𝑘) + 𝑊 (𝑘), (25)
SI signal reverberation
0.5 reverberation envelop where 𝑆𝑓 (𝑘), 𝑅𝐼 (𝑘) and 𝑊 (𝑘) denote the frequency domain
Amplitude

representation of the signal of interest, residual SI and ambient

0 noise, respectively, while 𝑘 represent the subcarrier index.
0.2 Next, a zero-forcing (ZF) equalizer followed by maximum-
-0.5 likelihood detection, which exploit the knowledge of the
0 frequency response of the channel between the nodes, is used
to detect the signal transmitted from the distant node, i.e.
-1
𝑋𝑓 (𝑘). The received signal after the ZF equalizer can be given
-0.2 as
0 0.2 0.4
-1.5 𝑆𝑓 (𝑘) 𝑅𝐼 (𝑘) 𝑤(𝑘)
0 1 2 3 4 5 𝑌𝑧𝑓 (𝑘) = + + , (26)
𝐻(𝑘) 𝐻(𝑘) 𝐻(𝑘)

where 𝑆𝑓 (𝑘) is given as 𝑆𝑓 (𝑘) = 𝐻(𝑘)𝑋𝑓 (𝑘). Then, 𝑋𝑓 (𝑘)

Fig. 6. The signal transmitted by local node along with the signal received
from the far node.
while 𝑠𝐼 (𝑛) denotes the SI signal and can be presented as
𝑃
∑ −1
𝑠𝑙 (𝑛) = ℎ𝑆𝐼 (𝑛 − 𝑝)𝑥𝑙 (𝑝),
𝑝=0
and 𝑤(𝑛) represents the ambient noise. It is worth pointing
out that the power of the SI signal is much higher than
that of the signal received from the distant node. Hence, to
protect the signal of interest, 𝑠𝑓 (𝑛), from being swamped by
the SI signal at the ADC and enable the receiver to detect
it, the SI signal should be suppressed effectively before the
ADC. In this paper, an adaptive self-interference cancellation
(SIC) scheme is exploited and integrated with the architecture
of the proposed node to effectively suppress the SI signal.
As depicted in Fig. 7, the IBFD-UWA communication node
exploits its perfect knowledge of its own transmitted signal to
generate a replica of the SI signal, i.e. 𝑐𝑠(𝑛), with the aid of
an adaptive filter, ĥSI , such that
𝑐𝑠(𝑛) = ĥ𝐻
𝑆𝐼 x𝑙 (𝑛),
where (.)𝐻 denotes the Hermitian transpose. The length of the
adaptive filter is set to be equal to the length of the SI channel,
𝐿𝑆𝐼 . The adaptive filter output, 𝑐𝑠(𝑛), is then subtracted from
the received signal to mitigate effectively the effect of the SI
signal before the ADC. The received signal after subtracting
cs(n) from the total received signal can expressed as
𝑦𝑙 (𝑛) = 𝑠𝑓 (𝑛) + 𝑟𝐼 (𝑛) + 𝑤(𝑛),
can be obtained by utilizing the ML detection such that
ˆ 𝑓 (𝑘) = argmin ∣ 𝑌𝑧𝑓 (𝑘) − 𝐻(𝑘)𝒞(𝑞) ∣2 ,
𝑋 (27)
(𝑞)∈ℤ𝑀
where 𝒞 represents the QPSK constellation vector.
In order to enable the adaptive SIC scheme to track the
(21) changes in the SI channel, the residual SI should be extracted
from the received signal after the SIC, i.e. (23), and fed to
the NLMS algorithm. To attain this the output of the ML
ˆ 𝑓 (𝑘) is exploited to reconstruct the signal that
detection, i.e. 𝑋
is received from the remote node, 𝑠𝑓 , using the knowledge of
the frequency response of the channel between the nodes such
that
𝐾−1
1 ∑ ˆ 𝑓 (𝑘)𝑒 𝑗2𝑛𝑘
𝑠ˆ𝑓 (𝑛) = √ 𝐻(𝑘)𝑋 𝐾 . (28)
𝐾 𝑘=0
Subsequently, this signal is subtracted from the received signal
after the SIC, (23), to obtain the residual SI signal such that
𝑅𝑆(𝑛) = 𝑠𝑓 (𝑛) − 𝑠ˆ𝑓 (𝑛) + 𝑟𝐼 (𝑛) + 𝑤(𝑛), (29)
where 𝑠ˆ𝑓 (𝑛) is the estimated version of the signal received
from the distant node, 𝑠𝑓 (𝑛), and is given as
(22)
𝑠ˆ𝑓 (𝑛) = 𝑠𝑓 (𝑛) + 𝜖𝑠𝑓 (𝑛), (30)
where 𝜖𝑠𝑓 (𝑛) denotes the uncorrelated estimation error ex-
hibiting a complex Gaussian distribution, 𝒞𝒩 (0, 𝜎𝜖2𝑠 ). This
𝑓
residual SI signal, (29), is then fed to the adaptive SIC scheme
to track the changes in the SI channel. The NLMS algorithm
in the proposed adaptive SIC exploits the residual SI signal,
(29), to update the tap-weight vector, ĥSI (𝑛 + 1), that is used
(23) to generate the replica of the signal, 𝑐𝑠(𝑛), such that [17]

where 𝑟𝐼 (𝑛) is the residual SI after the SIC, which is given 𝜇

˜
ĥ𝑆𝐼 (𝑛 + 1) = ĥ𝑆𝐼 (𝑛) + x𝑙 (𝑛)𝑅𝑆 ∗ (𝑛), (31)
as 𝛿 + ∣∣x𝑙 (𝑛)∣∣2
𝑟𝐼 (𝑛) = 𝑠𝐼 (𝑛) − 𝑐𝑠(𝑛). (24) where 0 < 𝛿 ≤ 1 and 𝜇
˜ is a positive adaptation constant.

Authorized licensed use limited to: IEEE Associate Editors. Downloaded on June 18,2021 at 10:07:50 UTC from IEEE Xplore. Restrictions apply.
 TX
IFFT xl
Xl + DAC PA
CP

ĥSI (n) DAC

hSI
Adaptive
Filter
csl
Reconstruct ŝf
sf −
RX
Yl Remove CP yl Preamplifier sf
Xf Decision + ADC − +
FFT BPF

Fig. 7. Architecture of the proposed IBFD-UWA node.

III. R ESULTS
140
In this section, we present the results of the proposed
OFDM based IBFD-UWA communication system utilizing Residual SI power
Noise floor
a QPSK modulation scheme, 12 kHz center frequency and 130
Residual SI power (dB re Pa)

6 kHz bandwidth. First, the performance of the proposed

adaptive SIC was evaluated using the SI channel model that
was presented in Section II. Fig. 8 illustrates the power of 120
the residual SI after the SIC. A closer inspection of Fig. 8
shows that the proposed SIC needs approximately 6 seconds to
suppress effectively the SI signal to the level of the noise floor. 110
It is worth pointing out that the results in Fig. 8 were obtained
under the assumption that the signal of interest was perfectly
100
detected and reconstructed, i.e. 𝜖𝑠𝑓 = 0, and removed from
the received signal after the SIC before feeding the residual
SI to the NLMS algorithm. 90
Next, the impact of the estimation error, i.e. 𝜖𝑠𝑓 , which
results from the imperfect detection and decoding of the 0 2 4 6 8 10 12
signal of interest, on the performance of the proposed SIC is Time (s)
investigated. It is noteworthy that the majority of the existing
research work has inspected the performance of the SIC Fig. 8. The power of the residual SI signal after the adaptive SIC.
schemes during the training period, in which the distant node
does not transmit and the received signal contains the SI signal
only, which does not imitate pragmatic scenarios. In this work, the ability to suppress the SI signal to approximately the level
we examined the effect of 𝜖𝑠𝑓 on the performance of the of the ambient noise. Finally, we studied the effect of detecting
proposed SIC scheme as depicted Fig. 9. A closer inspection of the signal of interest with an error on the performance of
Fig. 9 reveals that the ability to detect the signal of interest and the proposed SIC technique. The obtained results have shown
correctly decode it has a significant impact on the performance that the signal detection error is a significant performance
of the SIC. Hence, the signal of interest should be considered limitation of the adaptive SIC schemes and can not be ignored.
in the performance investigation of the SIC algorithms.

IV. C ONCLUSIONS ACKNOWLEDGMENT

In this work, we introduced an IBFD-UWA node that
utilizes an NLMS adaptive SIC scheme to diminish effectively
the SI signal before the local ADC. Furthermore, we presented
a model for the SI channel along with experimental measure-
ment of this channel. The proposed adaptive SIC has shown
This work was supported by EPSRC under project
EP/R002665/1, Full-Duplex For Underwater Acoustic Com-
munications. The authors would like to thank the Research
Council for this funding.

Authorized licensed use limited to: IEEE Associate Editors. Downloaded on June 18,2021 at 10:07:50 UTC from IEEE Xplore. Restrictions apply.
 [13] G. Qiao, S. Liu, Z. Sun, and F. Zhou, “Full-duplex, multi-user and
150 parameter reconfigurable underwater acoustic communication modem,”
in OCEANS - San Diego, Sept 2013, pp. 1–8.
[14] L. Li, A. Song, L. J. Cimini, X.-G. Xia, and C.-C. Shen, “Interference
cancellation in in-band full-duplex underwater acoustic systems,” in
140 OCEANS 2015 - MTS/IEEE Washington, Oct 2016, pp. 1–6.
Residual SI power (dB re Pa)

[15] J. Tian, S. Yan, L. Xu, and J. Xi, “A time-reversal based digital cance-
lation scheme for in-band full-duplex underwater acoustic systems,” in
130 OCEANS 2016 - Shanghai, April 2016, pp. 1–4.
[16] R. F. W. Coates, Underwater Acoustic Systems. MacMillan, 1990.
[17] S. S. Haykin, Adaptive filter theory. Pearson Education India, 2008.
120

110
SI power after the SIC
Noise floor
100

90
0 0.02 0.04 0.06 0.08 0.1

Fig. 9. The power of the residual SI signal after the adaptive SIC vs. 𝜖𝑠𝑓 .

R EFERENCES

[1] F. Qu, H. Yang, G. Yu, and L. Yang, “In-band full-duplex communica-

tions for underwater acoustic networks,” IEEE Netw., vol. 31, no. 5, pp.
59–65, 2017.
[2] J. Zhang, X. Ma, G. Qiao, and C. Wang, “A full-duplex based protocol
for underwater acoustic communication networks,” in 2013 OCEANS -
San Diego, Sept 2013, pp. 1–6.
[3] M. Duarte, C. Dick, and A. Sabharwal, “Experiment-driven characteri-
zation of full-duplex wireless systems,” IEEE Trans. Wireless Commun.,
vol. 11, no. 12, pp. 4296–4307, December 2012.
[4] E. Everett, A. Sahai, and A. Sabharwal, “Passive self-interference
suppression for full-duplex infrastructure nodes,” IEEE Trans. Wireless
Commun., vol. 13, no. 2, pp. 680–694, February 2014.
[5] B. Jebur and C. Tsimenidis, “Performance analysis of OFDM-Based
denoise-and-forward full-duplex PLNC with imperfect CSI,” in IEEE
International Conference on Communication Workshop (ICCW), June
2015, pp. 997–1002.
[6] E. Ahmed and A. Eltawil, “All-digital self-interference cancellation
technique for full-duplex systems,” IEEE Trans. Wireless Commun.,
vol. 14, no. 7, pp. 3519–3532, July 2015.
[7] B. A. Jebur, C. C. Tsimenidis, M. Johnston, and J. Chambers, “Outage
probability of an AF full-duplex physical-layer network coding system,”
in 2016 24th European Signal Processing Conference (EUSIPCO), Aug
2016, pp. 1828–1832.
[8] B. A. Jebur, C. C. Tsimenidis, and J. A. Chambers, “Tight upper
bound ergodic capacity of an AF full-duplex physical-layer network
coding system,” in 2016 IEEE 27th Annual International Symposium
on Personal, Indoor, and Mobile Radio Communications (PIMRC), Sep.
2016, pp. 1–6.
[9] D. Korpi, J. Tamminen, M. Turunen, T. Huusari, Y. S. Choi, L. Anttila,
S. Talwar, and M. Valkama, “Full-duplex mobile device: pushing the
limits,” IEEE Commun. Mag., vol. 54, no. 9, pp. 80–87, September
2016.
[10] B. A. Jebur, “Full duplex-transceivers: architectures and performance
analysis,” 2017.
[11] B. A. Jebur and C. C. Tsimenidis, “Impact of self and co-channel
interference on an DNF full-duplex one-way relay system,” in 2018
IEEE Global Communications Conference (GLOBECOM), Dec 2018,
pp. 206–212.
[12] G. G. Xie, J. H. Gibson, and K. Bektas, “Evaluating the feasibility of
establishing full-duplex underwater acoustic channels,” in Proc. of the
Third Annual Mediterranean Ad Hoc Networking Workshop, 2004.

Authorized licensed use limited to: IEEE Associate Editors. Downloaded on June 18,2021 at 10:07:50 UTC from IEEE Xplore. Restrictions apply.