Conference PaperPDF Available

A Neural-Network-Based Non-linear Interference Cancellation Scheme for Wireless IoT Backhaul with Dual-Connectivity

Authors:
A Neural-Network-Based Non-linear Interference
Cancellation Scheme for Wireless IoT Backhaul
with Dual-Connectivity
Huiliang Zhang*, Zhonglong Wang*, Fei Qin†, Meng Ma*, and Jianhua Zhang*
*Department of Electronics, Peking University, Beijing, 100871, China
†vivo Mobile Communication Technology Co., Ltd, Beijing, China
E-mail:mam@pku.edu.cn
Abstract—In this paper, we consider an Internet of Things
(IoT) wireless network using Long Term Evolution (LTE) cellular
system as backhaul. To provide high throughput, by using dual-
connectivity technique, the IoT gateway simultaneously connects
to two evolved Node Bs (eNBs) on two carriers, one for downlink
and the other for uplink. As a result, the receive link will
be severely interfered by the harmonic interference (HI) and
intermodulation (IM) components caused by the imperfections of
power amplifier (PA) and in-phase/quadrature (I/Q) modulator.
To solve this problem, in this paper, an neural-network (NN)-
based non-linear interference cancellation scheme is proposed for
dual-connectivity IoT gateway. In the proposed scheme, the non-
linear interference is first reconstructed by using the transmit
signal and the trained NN in baseband, and then subtracted from
the received signal in digital domain at receiver. The NN precisely
models the link behavior from the baseband transmitter to the
baseband receiver, including all the linear and non-linear effect.
Additionally, the NN can be used to reconstruct and cancel not
only the HI, but also the IM components of the mirror-frequency
interference (MFI) caused by I/Q imbalance, and direct current
(DC) bias caused by local oscillator (LO) leakage. To evaluate
the performance of the proposed scheme, a hardware prototype
is designed and implemented. Experimental results show that the
proposed scheme has a superior performance in dual-connectivity
system compared with the traditional non-linear interference
cancellation scheme using polynomial (PM) model.
Index Terms—Iot gateway, neural-network, dual-connectivity,
intermodulation, non-linear interference
I. INTRODUCTION
Internet of Things (IoT) is a set of technologies that can
interconnect anything, from daily life objects to more sophisti-
cated networked devices. As the number of devices connected
to the Internet is growing substantially, a massive number of
small packets are generated and transmitted via wireless access
networks, thus posing a great challenge for wireless networks to
sustain high throughput performance. One of the key problem
is that the throughput performance degrades dramatically for
small packets, since the headers of transport protocol and
packet scheduling are usually substantial, thus leading to more
transmission resources used for data that is not directly visible
for applications [1]. IoT wireless backhaul is an effective way
to solve this problem, in which the wireless IoT gateway first
collects data from IoT devices, and then forwards the data via
the backhaul link connected to end users or IoT clouds [2], [3].
As cellular networks have already be almost ubiquitous, they
can be utilized to provide IoT traffic access to the Internet
regardless of their location. In this paper, we consider to utilize
Long Term Evolution (LTE) network as wireless backhaul.
Though a single IoT device usually does not generate a sig-
nificant amount of traffic, but the number of IoT devices in an
IoT backhaul network can be massive, thus leading to a massive
amount of aggregated traffic data at the IoT gateway. To provide
high throughput for the backhaul link, carrier aggregation
technique can be adopted, where mobile stations are able to
consume radio resources of several serving cells simultane-
ously, and utilize the potentially aggregate bandwidth across
all of them [4], thus boosting capacity of the backhaul link.
Additionally, the emerging dual-connectivity [5] is regarded as
an attractive access mechanism in heterogeneous networks; in
LTE network, by using dual-connectivity technique, the mobile
stations are allowed to be connected simultaneously to two
evolved Node Bs (eNBs) over two carriers, thus significantly
improving the throughput and robustness.
However, when the IoT gateway works in dual-connectivity
mode, the receive link will be severely interfered by harmonic
interference (HI) caused by the non-linear effect of its own
power amplifier (PA). For example, if the gateway transmitter
works at 1.8 GHz, and the receiver works at 3.6 GHz, the
even harmonic would directly leak to its own receiver. Addi-
tionally, the other concern is the transmitter imperfections of
in-phase/quadrature (I/Q) modulator, which leads to amplitude
and phase mismatching between the I/Q signal branches of the
modulator in direct-conversion radios [6]. These impairments
introduce mirror-frequency interference (MFI), direct current
(DC), and extra intermodulation (IM) components at the output
of PA, which will dramatically degrade the detection perfor-
mance [7] and deteriorate the bit error rate. In addition, the RF
chain from transmitter to receiver contains not only non-linear
devices, but also linear devices, such as pulse shaping filters
and micro strip lines. Hence, the system can be modeled as
a Hammerstein model [8], where the non-linear part is used
for modeling transmitter PA, and the linear part is used for
modeling channel between transmitter and receiver.
So far, most of studies on the non-linear interference can-
 444
2019 32nd IEEE International System-on-Chip Conference (SOCC) 978-1-7281-3483-3/20/$31.00 ©2020 IEEE 10.1109/SOCC46988.2019.1570559857
Authorized licensed use limited to: Peking University. Downloaded on June 02,2020 at 03:52:57 UTC from IEEE Xplore. Restrictions apply.
cellation have been oriented towards the digital predistortion
at transmitter to cancel the non-linear interference of PA with
modulator imperfections. The most general forms are described
by polynomial (PM) model and Volterra series [9], [10]; and
pilot symbol sequence in this models is usually utilized to
obtain parameter estimates. However, these models are found
to be effective for mildly non-linear, but not appropriate for
strongly non-linear. Additionally, some studies focus on the
joint cancellation for the non-linear effect of PA and I/Q
modulater. Contemporary solutions contain two-step processes
[11], [12], in which MFI and DC are considered first, followed
by the cancellation of HI for PA linearization. In [13], a real-
valued focused time-delay neural-network (NN) model has
been proposed to mitigate the distortions in one-step, which
simplifies the predistortion process, and has a strong adaptive
nature of behavioral modeling of PA. In [14], a NN-based
method is proposed to cancel the non-linear interference in
radio frequency (RF) chains, which leads to additional costs in
terms of power and hardware complexity.
In this paper, an NN-based non-linear interference cancel-
lation scheme is proposed for dual-connectivity IoT wireless
gateway system. In the proposed scheme, the non-linear can-
cellation interference can be split into two steps; the first is to
reconstruct the HI and IM components by using the transmit
signal and the trained NN in baseband, and the second is to
subtract them from the received signal in digital domain at
receiver. Additionally, the proposed scheme addresses both the
non-linear and linear effects over the channel from transmitter
to receiver. To evaluate the performance of the proposed
scheme, a hardware prototype is designed and implemented.
Experimental results show that outstanding performance can
be achieved by using the proposed NN-based cancellation,
compared with that of using traditional PM model.
The remainder of this paper is organized as follows. Section
II introduces the system model of IoT wireless gateway work-
ing in dual-connectivity mode with a non-linear interference
canceller. Section III describes the design of NN-based non-
linear interference cancellation scheme. In Section IV, the
hardware prototype, measurement setup and conditions are
introduced, and the performance of the proposed scheme is
also presented and discussed. Finally, conclusions are drawn
in Section V .
II. SYSTEM MODEL
In order to provide high throughput for backhaul link, an
IoT gateway is simultaneously connected to two eNBs on two
carries, one for uplink and the other for downlink by using
dual-connectivity as shown in Fig. 1. f1and f2are the carriers
frequency of uplink and downlink for backhaul, respectively.
When f22f1, the IoT gateway’s receiver would be severely
interfered by the HI caused by the non-linear effect of its
own PA, because the even harmonic interference would directly
leak from the transmit chain to the receive chain without any
propagation loss. Additionally, the MFI and LO caused by
I/Q imbalance also inevitably degrade the quality of desired
signal. A basic block diagram of an IoT gateway with dual-
Fig. 1. IoT gateway connected to two eNBs on two carries and IoT devices.
connectivity mode is shown in Fig. 2. The transmit signal in
digital domain can be expressed as
X(n)=I(n)+iQ(n),(1)
where i=1, and I(n)and Q(n)are the I/Q components
of the baseband digital signal. After digital modulation and
up frequency conversion, X(n)is converted into RF signal
˜
Xmod (t), and non-linear components are introduced due to
the imperfection of I/Q mixer which leads to DC offsets and
I/Q imbalance. Thus, the RF signal can be represented as
˜
Xmod (t)=(
˜
I(t)+Ci) cos(ω1t)α(˜
Q(t)+Cq)sin(ω1t+ϕ),
(2)
where ˜
I(t)and ˜
Q(t)are the I/Q components of the baseband
signal after pulse shaping, ω1represents the angular frequency
of the carrier, and Ciand Cqare the DC offsets in the
I and Q channels, respectively. αand ϕare the gain and
phase imbalances between the I and Q channels. At the end
of transmitter RF chain, PA also introduces non-linear effect,
which can be expressed as
˜
Xout(t)=Gup ×fup [˜
Xmod ],(3)
where Gup and function fup represent the gain and the non-
linear transfer function of PA, respectively.
When f2is twice of f1, the receive link will be severely
interfered by the HI and IM interference from its own non-
linearity [15].
At the receiver, the received HI can be expressed as
˜rst(t)=He˜
Xout +n(t),(4)
where Hand ϕdenote linear channel gain and phase rotation
coefficient, respectively. n(t)denotes the additive Gaussian
noise. As we only focus on HI and IM interference cancellation
in this paper, the desired signal is omitted in (4). The output
of a complete receiver in downlink at f2is given as
˜r(t)=Gdown ×fdown rst(t)],(5)
where the function fdown and gdown are the downlink non-
linear function and gain of the PA at receiver, respectively;
then, the digital received signal r(n)can be obtain through
ADC. The goal of digital NN is to reconstruct an accurate
copy of r(n). The output of NN, denoted by ˜r(n), is then
subtracted from the received signal. Therefore, the residual
non-linear interference can be calculated by
ε(n)=r(n)˜r(n).(6)
445
Authorized licensed use limited to: Peking University. Downloaded on June 02,2020 at 03:52:57 UTC from IEEE Xplore. Restrictions apply.
Fig. 2. IoT gateway in dual-connectivity with one-step non-linear interference
cancellation scheme.
III. ANNN-BASED NON-LINEAR INTERFERENCE
CANCELLATION SCHEME
In this section, the design of NN-based non-linear inter-
ference cancellation scheme is detailed for dual-connectivity
IoT gateway system. Our aim is to cancel the HI and IM
components caused by the non-linear effect in PA, and the I/Q
imbalance from the received signal in digital domain.
Considering that NN is not primarily designed to operate
with complex signals while the transmit signal and received
HI are both in complex domain, a real-valued feed-forward
NN model, as shown in Fig. 3, is adopted in this paper. The
I/Q components of current samples are used as the inputs to
the NN, thus, each neuron of the hidden layer treats a linear
combination of I and Q components.
Fig. 3. Real-valued feed-forward NN model.
Usually, NN is composed of an input layer, hidden layers
and an output layers, and the number of hidden layers and
their corresponding neurons are commonly determined during
training of the network or by using empirical methods [16],[17].
According to the features of the digital signal used in IoT and
the approximation theorem that a feed-forward network with
a single hidden layer, containing a finite number of hidden
neurons, is a universal approximator for continuous functions
[18], we selected one hidden layer in this paper. The output of
each neuron is equal to the bias plus the sum of the products of
the input signals and corresponding weights, which is expressed
as [19]
dl
i(n)=
p
i=1
ωl
i,j xl1
j(n)+bl
i,(7)
where ωl
i,j represents the weight connecting the j-th neuron of
the (l1)-th layer to the i-th neuron at the l-th layer. xl1
jis
the input signal from the previous (l1)-th layer to the j-th
neuron, and bl
iis the bias of the i-th neuron at the l-th layer.
To model the non-linear effect and the linear channel from
PA to the receiver chain, we apply hyperbolic tangent sigmoid
[17] transfer function as activation function, which can be
written as
tansig(x)= 2
1+exp(2x)1.(8)
The output layer of the NN contains 2 neurons. The cost
function is calculated in batch mode during the forward pass
[17], and can be defined as
E=1
2N
N
n=1
[(Iout(n)ˆ
Iout(n))2+(Qout (n)ˆ
Qout(n))2],
(9)
where Iout(n)and Qout (n)are the I and Q components of
the desired output signals, while ˆ
Iout(n)and ˆ
Qout(n)are the
outputs of output layer.
The NN is trained by minimizing least square (LS) criterion
using Leven-Marquardt backpropagation algorithm [20], and
the training process stops when the NN satisfies the desired
modeling performance or reached the preset maximum epoch
or overtraining [16].
The reconstructed signals at the output of NN can be
expressed as
ˆr(n)=ˆ
Iout(n)+iˆ
Qout(n),(10)
which is then subtracted from the received signal in digital
domain to achieve non-linear interference cancellation. The
proposed scheme needs only the input and output data to adapt
to any imperfection in dual-connectivity system.
IV. EXPERIMENTAL MEASUREMENT SETUP,RESULTS AND
DISCUSSION
In order to verify the performance of the proposed interfer-
ence cancellation scheme, we implement a hardware prototype
for non-linear interference generation and cancellation.
We use quadrature phase shift keying (QPSK) modulated
orthogonal frequency division multiplexing (OFDM) signal as
the transmit signal, which is generated by using MATLAB,
and then downloaded into YunSDR. The YunSDR is a radio
platform consisting of FPGA (Xilinx Zynq SoC XC7Z020) and
AD9361, and can perform digital modulation, digital-to-analog
conversion, frequency up-conversion and PA. At the transmitter,
the transmit RF signal at the output of YunSDR is sent to
the PA (ZRL-2400LN+); then it is fed to YunSDR receiver
by a coaxial cable. The received signal is down-converted and
sampled by ADC, and stored to computer for offline analysis.
Additionally, the transmit and receive signal used as the input
and output data of NN, are obtained when the IoT backhaul
works in transmitting state and can capture the interference data
at f2without receiving the desired signal from receive link.
The received baseband data (Iout(n)and Qout(n)) are then
analyzed by using MATLAB. By using the transmit and receive
446
Authorized licensed use limited to: Peking University. Downloaded on June 02,2020 at 03:52:57 UTC from IEEE Xplore. Restrictions apply.
TABLE I
NMSE PERFORMANCE COMPARISON FOR DIFFERENT TRANSMIT POWER
IN NN-BASED AND PM MODELS.
Transmitted
signal
power(dBm)
NMSE(dB)
NN-based
Optimal
Num.of neurons
NMSE(dB)
PM
9.50 -30.0792 200 -7.094206
7.36 -31.8119 200 -10.31424
6.14 -32.1645 200 -15.80482
5.12 -31.0917 130 -20.69239
3.16 -32.2692 150 -18.21896
signal as the input and output data of NN, the NN is trained
to approximate the non-linear behavior of the PA. The desired
modeling performance in terms of MSE and maximum epochs
are set to be 106and 100, respectively. Once the training
process stopped, the transmit signal can be applied to NN for
non-linear interference reconstruction.
The interference signal is continuously reconstructed using
the NN and then subtracted from received signal to eliminate
the non-linear interference. The NN-based model is retrained
and the parameters are updated periodically using a new set of
data to adapt time variant.
In this paper, 2048×5×2 independent samples of the input
and output signal are used for training, validation and testing
purpose. During each experiment, 65% data is used for training,
25% for the purpose of the validation process, and the rest 10%
is used for testing. The performance of the normalized-mean-
square error (NMSE) is then validated, and NMSE is given
by
NMSE = 10 log
N
n=1 |(r(n)ˆr(n))|2
N
n=1 |r(n)|2
.(11)
Table I shows the NMSE performance of the NN-based
model compared with PM model for different transmit power.
The PM model which is used in non-linear interference formu-
lates the non-linearity in dual-connectivity as
r(n)=He
p=1
apCp1
2p|X(n)|2p2·X2(n).(12)
where 2pis the non-linearity order, and apis the coefficient
of PM model. In the PM scheme, the pilot symbol sequence
is required to obtain estimates of H,ϕand ap. The coefficient
apis calculated using LS algorithm.
The OFDM signal is modulated by a 2048-point inverse fast
fourier transform, and only 12 sub-carriers are used for trans-
mitting QPSK symbols. Thus, the bandwidth is approximate to
200 kHz, which is the same as that in NB-IoT system. Table
I shows that the proposed scheme keeps a better cancellation
capability and robustness than PM model for different transmit
power. This is because that the PM model is not appropriate for
TABLE II
MAIN NON-LINEAR COMPONENTS.
Non-linear
components
Frequency
(MHz) Main components
HI 6MHz Even harmonic
IM2(a) 0MHz IM component from
MFI, DC and odd harmonic of signal
IM2(b) -6MHz IM component from
MFI and even harmonic of MFI
IM2(c) 3MHz IM component from
DC and odd harmonic of signal
IM2(d) -3MHz IM component from
DC and odd harmonic of signal
strong non-linear behavior modeling, while NN has advantage
for modeling strongly non-linear system. In order to analyze
the non-linear components, power spectrum density (PSD) of
received OFDM signal is plotted in Fig. 4. It is shown that the
received non-linear components at 3.6 GHz contains not only
the HI component, but also various IM components, which
were analyzed in Table II.
Apparently, using PM model in (12) can only cancel the HI
component, but not the other IM components. However, for the
NN-based interference cancellation scheme, it has the potential
advantage to cancel all components in Table II. One can find
from Fig. 4 that the NN-based scheme makes a good match
not only on the frequency of HI, but also on the frequencies
of other IM components.
-10-8-6-4-20246810
Frequency (MHz)
-70
-60
-50
-40
-30
-20
-10
0
10
20
30
Power Spectrum Density (dB/Hz)
received signal
reconstructed signal by NN-based model
reconstructed signal by PM model
IM2(b)
IM2(d)
IM2(a)
IM2(c)
HI
Fig. 4. Power spectrum density (PSD) of received OFDM signal, reconstructed
signal from NN-based and PM models.
Finally, the waveforms of the received and reconstructed
non-linear interference are plotted in Fig. 5. It is shown that a
good match is achieved between the received and reconstructed
non-linear interference.
447
Authorized licensed use limited to: Peking University. Downloaded on June 02,2020 at 03:52:57 UTC from IEEE Xplore. Restrictions apply.
2300 2310 2320 2330 2340 2350 2360 2370 2380 2390 2400
Samples
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
In-phase Components
received signal
reconstructed signal by NN-based model
2362 2364 2366 2368 2370
0.2
0.3
0.4
(a)
2300 2310 2320 2330 2340 2350 2360 2370 2380 2390 2400
Samples
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
Quadrature Components
received signal
reconstructed signal by NN-based model
2358 2360 2362 2364 2366
0.35
0.4
0.45
(b)
Fig. 5. Validation of NN model for OFDM signal output I and Q components
in time domain, (a) In-phase components, (b) Quadrature components.
V. C ONCLUSION
In this paper, we proposed an NN-based non-linear in-
terference cancellation scheme for IoT gateway with dual-
connectivity in baseband. The hardware prototype implemen-
tation and experimental results are also introduced. In the
proposed scheme, both the non-linear and linear effects in
system are precisely modeled by NN on baseband at the re-
ceiver. The non-linear interference is reconstructed by using the
transmit signal and the trained NN, and then subtracted from
the received signal in digital domain. The experimental results
have shown that the proposed scheme can effectively suppress
not only the HI components, but also the IM components from
MFI and DC bias.
ACKNOWLEDGMENT
This work was supported by the Program for Core tech-
nology tackling key problems of Dongguan City under grant
number 2019622109010, and the National Natural Science
Foundation of China under grant number 61671024.
REFERENCES
[1] E. Piri and J. Pinola, “Performance of LTE uplink for iot backhaul,
in Proc. IEEE 13th Annual Consumer Communications & Networking
Conference (CCNC), NV, USA, Jan. 2016, pp. 6–11.
[2] H. Yu, B. Ng, and W. K. Seah, “On-demand probabilistic polling for
nanonetworks under dynamic iot backhaul network conditions,” IEEE
Internet of Things Journal, vol. 4, no. 6, pp. 2217–2227, Dec. 2017.
[3] L. Hou, S. Zhao, X. Xiong, K. Zheng, P. Chatzimisios, M. S. Hossain, and
W. Xiang, “Internet of things cloud: Architecture and implementation,
IEEE Commun. Mag., vol. 54, no. 12, pp. 32–39, Dec. 2016.
[4] M. A. Lema, E. Pardo, O. Galinina, S. Andreev, and M. Dohler,
“Flexible dual-connectivity spectrum aggregation for decoupled uplink
and downlink access in 5G heterogeneous systems,” IEEE J. Sel. Areas
Commun., vol. 34, no. 11, pp. 2851–2865, May 2016.
[5] “Study on small cell enhancements for E-UTRA and E-UTRAN; Higher
layer aspects,” document 3GPP TR 36.842, V12.0.0, Dec. 2013.
[6] G. Fettweis, M. L¨
ohning, D. Petrovic, M. Windisch, P. Zillmann, and
W. Rave, “Dirty RF: A new paradigm,” in Proc. IEEE 16th. International
Symposium on Personal, Indoor and Mobile Radio Communications
(PIMRC), Berlin, Germany, Sep. 2005, pp. 2347–2355.
[7] J. K. Cavers, “The effect of quadrature modulator and demodulator errors
on adaptive digital predistorters for amplifier linearization,IEEE Trans.
Veh. Technol., vol. 46, no. 2, pp. 456–466, May 1997.
[8] F. Chang and R. Luus, “A noniterative method for identification using
hammerstein model,” IEEE Trans. Autom. Control, vol. 16, no. 5, pp.
464–468, Oct. 1971.
[9] L. Ding, G. T. Zhou, D. R. Morgan, Z. Ma, J. S. Kenney, J. Kim, and
C. R. Giardina, “A robust digital baseband predistorter constructed using
memory polynomials,” IEEE Trans. Commun., vol. 52, no. 1, pp. 159–
165, Mar. 2004.
[10] B. Fehri and S. Boumaiza, “Baseband equivalent volterra series for
digital predistortion of dual-band power amplifiers,IEEE Trans. Microw.
Theory Tech., vol. 62, no. 3, pp. 700–714, Feb. 2014.
[11] J. K. Cavers, “New methods for adaptation of quadrature modulators
and demodulators in amplifier linearization circuits,” IEEE Trans. Veh.
Technol., vol. 46, no. 3, pp. 707–716, Aug. 1997.
[12] Y.-D. Kim, E.-R. Jeong, and Y. H. Lee, “Adaptive compensation for
power amplifier nonlinearity in the presence of quadrature modula-
tion/demodulation errors,” IEEE Trans. Signal Process., vol. 55, no. 9,
pp. 4717–4721, Aug. 2007.
[13] M. Rawat and F. M. Ghannouchi, “A mutual distortion and impairment
compensator for wideband direct-conversion transmitters using neural
networks,” IEEE Trans. Broadcast., vol. 58, no. 2, pp. 168–177, Apr.
2012.
[14] B. Mulliez, E. Moutaye, H. Tap, L. Gatet, and F. Gizard, “Predistortion
system implementation based on analog neural networks for linearizing
high power amplifiers transfer characteristics,” in Proc. IEEE 36th
International Conference on Telecommunications and Signal Processing
(TSP), Rome, Italy, Jul. 2013, pp. 412–416.
[15] J. C. Pedro and S. A. Maas, “A comparative overview of microwave and
wireless power-amplifier behavioral modeling approaches,IEEE Trans.
Microw. Theory Tech., vol. 53, no. 4, pp. 1150–1163, Apr. 2005.
[16] F. Mkadem, “Behavioural modeling and linearization of rf power am-
plifier using artificial neural networks,” Master’s thesis, University of
Waterloo, 2010.
[17] S. Haykin, Neural networks: a comprehensive foundation. Prentice Hall
PTR, 1994.
[18] G. Cybenko, “Approximation by superpositions of a sigmoidal function,
Mathematics of control, signals and systems, vol. 2, no. 4, pp. 303–314,
Dec. 1989.
[19] M. Rawat, K. Rawat, and F. M. Ghannouchi, “Adaptive digital predistor-
tion of wireless power amplifiers/transmitters using dynamic real-valued
focused time-delay line neural networks,” IEEE Trans. Microw. Theory
Tec h ., vol. 58, no. 1, pp. 95–104, Nov. 2009.
[20] J. Shawash and D. R. Selviah, “Real-time nonlinear parameter estimation
using the levenberg–marquardt algorithm on field programmable gate
arrays,” IEEE Trans. Ind. Electron., vol. 60, no. 1, pp. 170–176, Jan.
2012.
448
Authorized licensed use limited to: Peking University. Downloaded on June 02,2020 at 03:52:57 UTC from IEEE Xplore. Restrictions apply.
... Different research projects have combined the ideas of spectrum and interference management and multi-connectivity: [116][117][118][119][120][121][122]. ...
... For many of these studies, the objective was to determine how interference could be minimized to enable effective multi-connectivity in 5G cellular networks and beyond [116,117,119,120]. To achieve this, the authors of [116] introduced a user-centric approach called back-haul state-based distributed transmission. ...
... In addition, communication between one UE and two base stations was the only case considered. In the same way, the authors of [117] proposed an interesting nonlinear interference cancellation mechanism designed to enable multi-connectivity and higher throughput for IoT gateways. This solution was divided into two main phases. ...
Article
Full-text available
To manage a growing number of users and an ever-increasing demand for bandwidth, future 5th Generation (5G) cellular networks will combine different radio access technologies (cellular, satellite, and WiFi, among others) and different types of equipment (pico-cells, femto-cells, small-cells, macro-cells, etc.). Multi-connectivity is an emerging paradigm aiming to leverage this heterogeneous architecture. To achieve this, multi-connectivity proposes to enable UE to simultaneously use component carriers from different and heterogeneous network nodes: base stations, WiFi access points, etc. This could offer many benefits in terms of quality of service, energy efficiency, fairness, mobility, and spectrum and interference management. Therefore, this survey aims to present an overview of multi-connectivity in 5G networks and beyond. To do so, a comprehensive review of existing standards and enabling technologies is proposed. Then, a taxonomy is defined to classify the different elements characterizing multi-connectivity in 5G and future networks. Thereafter, existing research works using multi-connectivity to improve the quality of service, energy efficiency, fairness, mobility management, and spectrum and interference management are analyzed and compared. In addition, lessons common to these different contexts are presented. Finally, open challenges for multi-connectivity in 5G networks and beyond are discussed.
... Recently, neural networks (NNs) have drawn much attention from researchers due to their strength in pattern recognition and complex system modeling. Neural networks have been investigated as a promising behavioral model for nonlinear distortions [8]- [11]. In [9], the feed-forward NN canceler was proposed to mitigate the nonlinear interference in fullduplex transceivers. ...
... In [10], a low-complexity NN structure with two neurons in the first hidden layer is proposed to cancel self-interference with nonlinearities and memory effect. In [11], a two-step nonlinearity cancellation scheme is proposed, where the nonlinearity distortion is firstly reconstructed using the transmit signal and a trained neural network and then subtracted from the received signal in the digital domain. ...
Article
Full-text available
In order to correct non-linearities due to High Power Amplifiers (HPA) operating near saturation in telecommunication transceivers, a new adaptive predistortion system based on analog Neural Networks (NNs) was developed. Based on size, consumption and bandwidth considerations, Multi-Layer Perceptron (MLP) type NNs were implemented in a 0.6 μm CMOS ASIC. The NNs parameters are digitally updated with a computer, depending on simulation conditions (temperature drifts, ageing variations). The interface between the analog part and the software updating system is integrated in an analog-digital PCB including a FPGA, 6 analog-to-digital converters and 62 digital-to-analog converters. This paper describes the realization of each part of the breadboard system and presents experimental validation results of the whole predistortion module.
Article
Full-text available
The Internet of Things (IoT), which enables common objects to be intelligent and interactive, is considered the next evolution of the Internet. Its pervasiveness and abilities to collect and analyze data which can be converted into information have motivated a plethora of IoT applications. For the successful deployment and management of these applications, cloud computing techniques are indispensable since they provide high computational capabilities as well as large storage capacity. This paper aims at providing insights about the architecture, implementation and performance of the IoT cloud. Several potential application scenarios of IoT cloud are studied, and an architecture is discussed regarding the functionality of each component. Moreover, the implementation details of the IoT cloud are presented along with the services that it offers. The main contributions of this paper lie in the combination of the Hypertext Transfer Protocol (HTTP) and Message Queuing Telemetry Transport (MQTT) servers to offer IoT services in the architecture of the IoT cloud with various techniques to guarantee high performance. Finally, experimental results are given in order to demonstrate the service capabilities of the IoT cloud under certain conditions.
Conference Paper
Full-text available
In order to correct non-linearities due to High Power Amplifiers (HPA) operating near saturation in telecommunication transceivers, a new adaptive predistortion system based on analog Neural Networks (NNs) was developed. Based on size, consumption and bandwidth considerations, Multi-Layer Perceptron (MLP) type NNs were implemented in a 0.6 μm CMOS ASIC. The NNs parameters are digitally updated with a computer, depending on simulation conditions (temperature drifts, ageing variations). The interface between the analog part and the software updating system is integrated in an analogdigital PCB including a FPGA, 6 analog-to-digital converters and 62 digital-to-analog converters. This paper describes the realization of each part of the breadboard system and presents experimental validation results of the whole predistortion module.
Article
Full-text available
The Levenberg-Marquardt (LM) algorithm is a nonlinear parameter learning algorithm that converges accurately and quickly. This paper demonstrates for the first time to our knowledge, a real-time implementation of the LM algorithm on field programmable gate arrays (FPGAs). It was used to train neural networks to solve the eXclusive Or function (XOR), and for 3D-to-2D camera calibration parameter estimation. A Xilinx Virtex-5 ML506 was used to implement the LMA as a hardware-in-the-loop system. The XOR function was approximated in only 13 iterations from zero initial conditions, usually the same function is approximated in thousands of iterations using the error backpropagation algorithm. Also, this type of training not only reduced the number of iterations but also achieved a speed up in excess of 3 ×10 6 when compared to the software implementation. A real-time camera calibration and parameter estimation was performed successfully on FPGAs. Compared to the software implementation the FPGA implementation led to an increase in the mean squared error and standard deviation by only 17.94% and 8.04% respectively. The FPGA increased the calibration speed by a factor of 1.41 × 10 6 . There are a wide range of systems problems solved via nonlinear parameter optimization, this study demonstrated that a hardware solution for systems such as automated optical inspection systems or systems dealing with projective geometry estimation and motion compensation systems in robotic vision systems is possible in real time.
Article
Driven by the high sensing resolution provided by advanced nano technologies, Electromagnetic-based Wireless Nano Sensor Networks (EM-WNSNs) operating in the terahertz (THz) band is becoming an integral part of the Internet of Things (IoT). Data acquisition for EM-WNSNs faces two challenges: the dynamic IoT backhaul bandwidth and THz channel conditions that jointly impact resource utilization efficiency. Specifically, the mismatch between the demands of the EM-WNSNs and the available bandwidth of the backhaul link reduces the bandwidth efficiency of IoT backhaul or the energy efficiency of EM-WNSNs. To address the new constraints that emerge from the EM-WNSN, an On-demand Probabilistic (OP) polling scheme is proposed for the backhaul tier of EM-WNSNs which is composed of nano sinks that aggregate and transport data from nano sensors to the IoT gateway. OP polling adjusts the packet aggregation process of nano sinks according to the up-to-date network conditions. From performance evaluation, OP polling achieves an 18% higher bandwidth efficiency compared to the previously proposed On- demand Efficient (OE) polling and 69% less energy consumption compared to the greedy polling.
Article
Maintaining multiple wireless connections is a promising solution to boost capacity in fifth-generation (5G) networks, where user equipment is able to consume radio resources of several serving cells simultaneously and potentially aggregate bandwidth across all of them. The emerging dual connectivity paradigm can be regarded as an attractive access mechanism in dense heterogeneous 5G networks, where bandwidth sharing and cooperative techniques are evolving to meet the increased capacity requirements. Dual connectivity in the uplink remained highly controversial, since the user device has a limited power budget to share between two different access points, especially when located close to the cell edge. On the other hand, in an attempt to enhance the uplink communications performance, the concept of uplink and downlink decoupling has recently been introduced. Leveraging these latest developments, our work significantly advances prior art by proposing and investigating the concept of flexible cell association in dual connectivity scenarios, where users are able to aggregate resources from more than one serving cell. In this setup, the preferred association policies for the uplink may differ from those for the downlink, thereby allowing for a truly decoupled access. With the use of stochastic geometry, the dual connectivity association regions for decoupled access are derived and the resultant performance is evaluated in terms of capacity gains over the conventional downlink received power access policies.
Article
This paper presents a one-step solution for transmitter nonlinearity estimation and linearization control in the presence of I/Q modulator imperfections for wideband direct-conversion transmitters. These transmitters include power amplifiers with frequency-dependent nonlinearities and modulator imperfections. With the proposed two-hidden-layer feedforward neural network, traditional two-step characterization and specially designed training signals are not required in the parameter estimation stage; and, estimation can be done without interrupting the operation of the transmitter. The measurement results and comparisons of the proposed neural network with the existing state-of-the-art methods show the superior performance in the presence of extreme RF impairments.
Article
This paper proposes a new dual-band baseband equivalent (BBE) Volterra model devised to predict the behavior of dual-band power amplifiers (PAs) and/or to linearize their response. A series of derivations were applied to the original continuous passband Volterra-series expression to uncover a simple formulation that related the BBE envelopes of the output signals in each band to those at the input in the discrete domain. This yielded an inherently low-complexity dual-band Volterra-series model that does not require the empirical pruning commonly applied to the low-pass equivalent Volterra series. The proposed model was successfully applied to digitally predistort and linearize a dual-band 45-W class-AB GaN PA driven with different dual-band dual-standard test signals. For each band, the model necessitated less than 25 coefficients to reduce the adjacent channel leakage power ratio ACLR by up to 25 dB.