ACCEPTED FROM OPEN CALL

Intelligent Radio for Next Generation

Wireless Communications: An Overview
Qi-Yue Yu, Hong-Chi Lin, and Hsiao-Hwa Chen

Abstract of the most challenging frontiers for futuristic

wireless communications [2–5]. It should be stat-
This article discusses various issues on intelligent ed that we are not going to give all the details
radio (IR) from a perspective of intelligent UE and about the IR designs. Instead, we summarize only
intelligent BS. Unlike a traditional UE, an intelligent the very basic ideas on what IR will be and which
UE can automatically access channel resources, functions IR should offer, as realistic implementa-
cooperate with other UEs, and manage interferenc- tion of a full-functional IR is still an open issue.
es. To implement such an intelligent UE, innovative In an IR cellular network, traditional UEs and
techniques should be used, such as smart resource BSs should be replaced by “intelligent UEs” and
sensing and grant-free transmission. The core of “intelligent BSs.” Unlike a traditional UE, an intel-
IR lies in the development of thinking hardware, ligent UE can offer a high degree of flexibility,
which can collect/process vital information of the and it can autonomously sense communication
environment and manage its resources optimally resources, access multi-user channels, work coop-
to support ultra-broadband communications. In eratively with other UEs, and set up its parameters
particular, an integrated radar-communication unit automatically. In this sense, a UE has evolved from
is one of the most critical environment sensors in a plant (without intelligence) to an animal (with
an intelligent UE, which helps it to acquire accurate intelligence), as shown in Fig. 1. To enable such
location and mobility information of its neighbors an important evolution of UEs, many innovative
for cooperation and interference avoidance. The techniques should be developed and utilized, from
traditional physical layer designs (such as beam- fundamental communication theories about IR to
forming and MIMO) should be revolutionized for extremely sophisticated hardware with thinking
the development of IR. ML and other AI algorithms capabilities. For example, an intelligent UE should
play an important role in both BS and UEs. The be able to learn environment efficiently once its
communication protocols should be re-designed power is turned on, and it will set up its working
as well. The IR can offer a very high spectrum effi- parameters (including carrier-frequency, band-
ciency and power efficiency and is one of the most width, modulation, channel coding, and so on)
important research topics for future-generation automatically to optimize its bandwidth efficiency
wireless communications. and power efficiency.
On the other hand, to implement an intelligent
Introduction BS, AI algorithms should be applied extensively in
The upcoming 5G wireless networks are expected a cell site as well as in the core network. To imple-
to support a massive number of user equipments ment AI based BSs and the core network, there
(UEs), including both human beings and machines. are many open issues remained to be investigat-
Traditional UEs rely on grant-based transmissions ed from physical layer communication theories to
(such as FDMA, TDMA, CDMA, and OFDMA), upper-layer protocol designs. For example, how to
where each UE transmits its data according to a configure the core network optimally in terms of
scheduling grant from a base station (BS). The BS overall bandwidth efficiency and power efficiency
allocates appropriate resources to the UEs, that in response to fast varying traffic loading is a very
is, frequency, time-slot, and power. Traditional challenging issue. With the help of ML and other
grant-based transmission regulates the use of the highly efficient AI algorithms, timely and yet opti-
resources. However, it suffers from a high over- mal network configuration (including BSs and UEs)
head and a low spectrum efficiency (SE). With the will become possible, based on the information
help of spectrum sensing, cognitive radio (CR) collected by various environment-sensing technol-
was proposed for wireless communications to ogies, such as the geo-locations of BSs and UEs,
improve the SE [1]. Many works have been done network-wise/cell-wise instant traffic loading, data
on CR, such as channel estimation/prediction and link management, and radio resource allocation.
dynamic spectrum management. Unfortunately, In general, the development of IR systems
the up-to-date research on CR still fails to offer depends mainly on three technological break-
a real thinking capability. In this article, we are throughs, which include a digitized ultra-wideband
motivated to discuss the issues on the design and radio frequency (RF) front-end, a powerful AI core
implementation of intelligent radio (IR), which for signal processing, and a high-speed wireless
works based on artificial intelligence (AI) as one backhaul. The digitized ultra-wideband RF front-

Digital Object Identifier: Qi-Yue Yu and Hong-Chi Lin are with the Harbin Institute of Technology;
10.1109/MWC.2019.1900001 Hsiao-Hwa Chen (corresponding author) is with National Cheng Kung University.

94 1536-1284/19/$25.00 © 2019 IEEE IEEE Wireless Communications • August 2019

Authorized licensed use limited to: Tampere University. Downloaded on December 30,2022 at 23:19:06 UTC from IEEE Xplore. Restrictions apply.
 BS Intelligent
BS

RRM
Time RRM
Coding slot
scheme Frequency Security
spectrum Cognitive Cooperation
radio
(frequency)
Interferences
Transmit
power Grant-free
MA
Others
Others

Intelligent
User Equipment User Equipment

FIGURE 1. A diagram to show the differences between a traditional UE (a plant: without intelligence) and
an intelligent UE (an animal: with intelligence).

end is to digitize wideband RF signals, followed by Intelligent

the AI core, which is responsible for various local BS
AI computations. The high-speed wireless backhaul
is necessary for accessing big data centers for pat-
tern recognition and other more sophisticated AI
algorithms.
The main contributions of this article can be
summarized in two parts:
• This article introduces some of the core tech-
niques of intelligent UEs, especially related Intelligent Radar Intelligent
physical layer techniques. UE
target UE

• The issues on implementing an intelligent BS is

investigated, which will play a very important
role in an IR based cellular system. A sophis- Communication
signals
ticated AI core can help to achieve the opti- Radar target
mized performance in both air links and core Intelligent
UE
Intelligent
UE
networks. Communication
Radar target
signals
The remainder of this article can be outlined
as follows. The following section focuses on iden- Communication
tifying research topics on intelligent UEs. Then the signals
issues on intelligent BSs will be discussed, followed
by the concluding remarks in the final section. Intelligent
UE

IntellIgent user eQuIpMent

In a futuristic wireless communication system, an FIGURE 2. Integrated radar-communication sensors for intelligent radio.
intelligent UE can be anything, not limited to a
terminal like a smart-phone. For instance, it can to simultaneously support a massive number of
be an autonomous-driving vehicle, which requires intelligent UEs to access intelligent BSs is extremely
much more information for its operation than a important. Another important issue is security and
phone. Pressing research topics on the implemen- privacy provisioning. This section will discuss some
tation of an intelligent UE include its hardware of these research topics.
design, radar-communication integration, uplink
grant-free multiple access, intelligent physical MultIfunctIonAl rf:
layer, and security and privacy provisioning. Smart IntegrAted rAdAr-coMMunIcAtIon sensors
hardware design is the foundation of an intelligent Multifunctional RF is one of the most critical
UE. Without smart hardware, the implementation techniques to support intelligent UEs. Advanced
of intelligent UEs is impossible. Radar-communica- multifunctional RF is an integration of various RF
tion integration gives a powerful sensing capability functions, such as communications, radar, and
for intelligent UEs. With the help of radar sensors, other sensors [6]. Traditionally, radar and commu-
an intelligent UE can acquire accurate position nication are two independent research topics, and
and mobility information of its peers, such that a joint implementation of the two should consider
cooperative communications and interferences their mutual influences with each other, which
avoidance can be made possible, as shown in Fig. 2. involves antenna design, electromagnetic interfer-
The smart hardware and radar sensors can be ence mitigation, and radar sensing accuracy guar-
viewed as a “physical core” of an intelligent UE, antee (i.e., position and velocity). With the help
on which novel physical-layer techniques can be of multifunctional RF techniques, the differences
built. An intelligent UE should be able to adaptively between radar and communication hardwares will
sense frequency spectrum, coding scheme, modu- diminish [7]. They can share the same frequency
lation type, and transmit power of all other trans- bands and similar RF transceivers, especially with
mitting devices nearby. The investigation on how the help of large antenna arrays. With a broad-

IEEE Wireless Communications • August 2019 95

Authorized licensed use limited to: Tampere University. Downloaded on December 30,2022 at 23:19:06 UTC from IEEE Xplore. Restrictions apply.
 Frequency Frequency Frequency techniques, and it can be viewed as a type of
Power Power Power space division approach. Massive MIMO is an
active research topic in communication research.
With an increasing number of antenna elements,
various pre-coding techniques can be explored to
+ + increase channel capacity and to mitigate interfer-
ences. There are generally three types of pre-cod-
ing techniques, including digital pre-coding, analog
pre-coding, and hybrid pre-coding. These pre-cod-
Time Time Time
User 1 User 2 User 3
ing techniques have different features in different
application scenarios. For the implementation of a
Frequency Base station (BS) radar system, adaptive phase array radar (APAR)
Power has been widely used in various military platforms,
such as in navy vessels. Since an APAR does not
need mechanical parts, it is very convenient to
Empty CTU User1+User2 steer its beams for multiple objects detection and
User 1 User1+User3
tracking. Therefore, we can use the beam division
approach for both radar sensors and communica-
User 2 User2+User3 tions in an IR. A large antenna array can be divided
User 3 User1+User2+User3 into a number of sub-arrays to support multiple
Time
beams, which include analog beamformers, nar-
row-band digital beamformers, and wide-band dig-
FIGURE 3. A diagram of grant-free uplink multiple access system, where CTU ital beamformers [6]. Different sub-arrays can be
stands for “contention transmission unit.” configured by different strategies for different pur-
poses. Generally speaking, the same sub-array can-
not be used for both communications and radar
band antenna array, the multifunctional RF can sensing at the same time. In addition, the nearby
transmit and receive multiple independent beams two sub-arrays may not be able to transmit and
simultaneously, performing both communication receive at the same time because the side lobes of
(a massive MIMO) and radar (Aegis) functions the transmit beam may interfere with the main lobe
using an active phased antenna array. Followed of the receive beam, and vise versa. A number of
by the multifunctional RF unit is an advanced sig- antenna elements can also be used to implement a
nal processing core (an AI core) for high-speed synthetic-aperture radar (SAR) for two-dimensional
data processing with various AI algorithms as an or three-dimensional imaging, which works as a
effort to maximize the performances working in weather-proof supplementary to vision sensors in
different RF scenarios. an IR (such as in autonomous-driving vehicles).
There are many appealing features of multifunc- Mixed Signal Design: Different from the time
tional RF systems, which should offer, for example, and beam division approaches, mixed signal design
both high power efficiency and spectrum efficiency. constructs new signal waveforms to support radar
Thus, we can obtain location/mobility information sensing and communications simultaneously.
from radar, then process and deliver the information To this end, a communication system should be
via communication systems to achieve a very high designed with additional radar functions, and com-
quality of service (QoS) of the networks. To achieve munication data streams could be superimposed
this goal, the key issue is how to seamlessly combine on top of radar signals. With its two-dimensional
radar and communication hardware as well as the resource elements, orthogonal frequency division
follow-up signal processing in an AI core. To discuss multiplexing (OFDM) is a good platform for radar
these issues specifically, we could proceed in dif- communication integration. In addition to OFDM,
ferent aspects for radar and communication signals chirp multi-carrier modulation and spread spec-
processing [3, 4, 7], including time division, beams trum techniques are also appealing choices. Never-
division, and mixed signal design. theless, how to reduce the interferences between
Time Division: Time division approach is a most radar and communication signals is an interesting
straightforward way to multiplex radar and commu- research topic [8].
nication functions using a strobe switch. When the
strobe switch selects the communication mode, grAnt-free MultIple Access
the radar function is disabled. On the other hand, To efficiently support a massive number of intel-
when the radar mode is selected, the communi- ligent UEs, uplink grant-free multiple access is an
cation is stopped. This approach is easy to realize emerging technique [9, 10], which can reduce
and basically it does not require a big change in the latency of the transmissions from UEs by
the hardware and software in communication and avoiding complicated multi-user scheduling pro-
radar systems, and thus they could be designed cedures, which are required in all current multiple
and implemented independently. However, due access technologies. It is a well-known fact that
to the nature of time division multiplex, communi- the amount of overhead and control signal laten-
cation signals cannot occupy too long time, other- cy is an extremely serious issue for implementing
wise it will affect the radar system, and vise versa. massive connectivity, especially in small packets
This approach can be used in an intelligent UE, transmissions, such as in the Internet of Things
which does not require an extremely low latency, (IoT) applications and machine type communica-
such as in a terminal with a relatively low mobility. tions (MTCs).
Beam Division: The beam division approach One of the most attractive solutions for grant-
for radar communication integration was devel- free multiple access is a semi-persistent scheme
oped with the help of advanced antenna array with pre-configured radio resources, such as time,

96 IEEE Wireless Communications • August 2019

Authorized licensed use limited to: Tampere University. Downloaded on December 30,2022 at 23:19:06 UTC from IEEE Xplore. Restrictions apply.
 Input Hidden layer Output
CTU pattern set layer
layer …

Multiple users’
Offline
transmit symbols
training

Channel state
information
(CSI)

Online detection

Antenna Input Hidden layer Output

User i layer
layer … User i
User i’s CSI
Codebook i Transmitter

Signal

...
Receiver
...
...

processing

Antenna
User j User j
User j’s CSI
Codebook j Transmitter

FIGURE 4. AI-based multiple-user signal detection.

frequency, and codebooks, and users do not need including beamforming, channel estimation, Adap-
traditional grant/request procedures before trans- tive data Rate Coding and Modulation (ARCM),
missions. In general, the pre-configured resources and synchronization techniques, as shown in Fig. 5.
are called contention transmission units (CTUs) Beamforming: Beamforming techniques have
[9], which consist of time, frequency, codebooks, been widely used in communications, radar, and
power, and space. Each active UE can send its own physical layer security. Thus, IR should use it to
CTUs to a BS, and then the BS detects active users design beams jointly to support different applica-
and estimates their corresponding fading channels tions simultaneously. A traditional antenna array
without the knowledge of their CTUs. Figure 3 configures its beams according to the information
shows a diagram of a grant-free uplink transmission of direction of arrival (DoA) of the signals, whereas
system. MIMO is realized by space-time coding schemes.
As shown in Fig. 3, assume that there are three There are similarities and differences between
active users and different users occupy different beamforming and MIMO techniques, and the
CTUs. At the receiver of a BS, some CTUs are key issue is to make the best use of CSI. With the
still free, some CTUs are occupied by one active help of hybrid precoding and signal processing,
user, and some CTUs are utilized simultaneous- an intelligent UE is able to perform radar-commu-
ly by three users (as shown in different colors in nication functions at the same time. In addition to
the lower part of Fig. 3). Since the active users precoding, multi-user signals detection algorithms
may send their data randomly to a BS, it is not also play an important role in the implementation
easy for the BS to acquire the knowledge of the of intelligent UEs.
active users’ CTUs. Therefore, the BS has to detect Beamforming techniques carry many attrac-
active users in each CTU in a blind way. Generally tive features that are vital in intelligent UEs. For
speaking, a BS should detect active users jointly example, beamforming techniques can support
via estimating the channel state information (CSI) multi-user transmissions, which can be viewed as
of each user and separating multiple users’ data another type of orthogonal/non-orthogonal multi-
information, which works in a different way from a ple access technique. Thus, it is an interesting topic
traditional multiple access scheme. There are many to design beamforming based grant-free multiple
works left for more investigations on grant-free access techniques. Moreover, beamforming helps
multiple access, such as joint multi-user detection to achieve both diversity and multiplex gains, as
and intelligent resources (CTUs) allocation. Based well as a multi-user processing gain in some special
on AI technologies, it will be desirable to apply ML cases, that is, opportunistic beamforming (OBF).
to estimate channels and detect multi-user signals, Exploring the advantages of beamforming tech-
as shown in Fig. 4. niques, an intelligent UE can offer a superb BER
performance even in hostile fading channels.
IntellIgent phYsIcAl lAYer Channel Estimation and Prediction: Channel
The intelligent physical layer is extremely import- state information (CSI) discloses vital and funda-
ant, and there are many widely open research mental information in wireless communications.
topics. This article discusses only four of them, Thus, a highly efficient channel estimation scheme

IEEE Wireless Communications • August 2019 97

Authorized licensed use limited to: Tampere University. Downloaded on December 30,2022 at 23:19:06 UTC from IEEE Xplore. Restrictions apply.
 Communications Interference
user

Beamforming UE BS

estimation
Channel
technique
Cooperative

...
Radar users
UE UE

...
Interference
Physical layer users
security
Cooperative
(a) user
(b)

Carrier
Data rate
Adaptive date rate

frequency & phase

Synchronization
& coding &
modulation

Q
(ARCM)

Q
Adaptive Symbol
Modulation
Q
I selection rate & phase
I

I
Multiple users’
Coding
signals
(c) (d)

FIGURE 5. IIlustration of IR based physical layer techniques.

is important for any communication systems. In Synchronization: Synchronization is yet anoth-

the context of intelligent UEs, channel estimation er difficult but important issue for implementing IR
should deliver more content-rich information, systems. The difficulty can be summarized mainly
such as channel gains (in both frequency and in three aspects as follows:
time domains), signal phase, and DoA of arrival • There are propagation delays between transmit-
signals. Moreover, the transceiver in an intelligent ters and receivers.
UE should estimate not only the channels between • UEs may use arbitrarily different transmit data
UEs and BS but also the channels among UEs. rates and carriers.
There is a massive amount of channel information • UEs may access the system randomly, leading
to be estimated. Thus, it presents a big challenge to multiple access interferences (MAIs).
to today’s state-of-the-art hardware and signaling All of these facts make the synchronization in IR
designs (e.g., to design and assign pilot sequences systems much more difficult. Without synchroni-
for accurate channel estimation among UEs and zation among users, MAIs may degrade the sys-
between UEs and BS). Traditional designs of pilot tem performance significantly. Without accurate
signals are not good enough, in which uplink pilot synchronization to sample signals at a proper
sequences have to be multiplexed with data traffic. instance periodically, it is impossible to recover
The multiplexed pilot sequences and data streams data [11].
generate interferences, leading to channel estima- The synchronization consists of three parts,
tion errors. It is necessary to re-design pilot signals including carrier synchronization, symbol (data
for the intelligent UEs. rate) synchronization, and network synchroniza-
Moreover, there is another interesting issue on tion. Traditional synchronization techniques deal
the joint design of channel estimation and multi-us- with the three issues separately. With the help
er detection algorithms in grant-free transmission. of AI, a joint synchronization becomes possible,
Owing to the random access nature of uplink which however requires much more investigation.
channels, the user signals are aliasing and a receiv-
er has to detect user data and estimate channels securItY And prIVAcY proVIsIonIng
simultaneously. In this scenario, we should choose There will be a large number of intelligent UEs
appropriate AI algorithms for channel estimation in future networks, and they need to communi-
and multi-user detection, which is still a widely cate and exchange information with each other.
open research topic. Thus, security and privacy provisioning becomes
Adaptive Date Rate Coding and Modulation: a pressing issue. Traditional radio and CR net-
Unlike traditional cellular UEs, intelligent UEs should works use passive security strategies. With ML
be able to select a different data rate adaptively and other AI techniques, intelligent UEs work-
according to particular applications. Therefore, tra- ing in complex communication environments
ditional adaptive coding and modulation (ACM) can predict possible risks and threats and thus
schemes must be renovated to perform adap- take appropriate countermeasures to deal with
tive data rate, coding, and modulation (ARCM) them, through the feedback from physical and
selection. Based on different CSI and applications, network infrastructures (connection with big data
ARCM can automatically alter the related param- centers via high-speed links is a must). There are
eters, such as transmit data rate, code rate, cod- primarily two ways to deal with security issues.
ing scheme, and modulation. In short packets and One is the use of the distributed secret keys that
voice communications, intelligent UEs may reduce were designed using block chain technologies,
transmit data rate, while in video and multimedia and which have been widely used in bitcoin and
communications, the transmit data rate can be other virtual-money systems. Another is physical
increased correspondingly. The appropriate ARCM layer security, which exploits wireless communica-
parameter setup lies on the decisions made by the tion techniques as “finger prints” (such as MIMO
AI core of an intelligent UE. channel matrix) in public channels [12]. The effec-

98 IEEE Wireless Communications • August 2019

Authorized licensed use limited to: Tampere University. Downloaded on December 30,2022 at 23:19:06 UTC from IEEE Xplore. Restrictions apply.
 The brain of an intel-
ligent BS is in charge
Configuration Computational offloading of BS system man-
Intelligent control module
MAC Radio access technique artificial intelligence (AI) Caching and distributed Edge agement, from radio
layer coordination storage computing
functions resource management
Traffic prediction
Scheduling Fog elements’ collaboration (RRM), network man-
Service awareness agement, to macro/
Radio resource management small cells manage-
Requirement cognition Seamless switching
(RRM) Intelligent ment. Moreover, an
PHY network
layer Radio bearer control (RBC) CSI estimation and prediction Interference coordination
management intelligent BS also offers
License authentication Others... Loading balancing
the functions of edge
computing, caching
and distributed
RF model
storages.

Wireless channels

FIGURE 6. A conceptual block diagram of an intelligent BS.

tiveness of physical layer security lies in the intelli- The brain of an intelligent BS is in charge of BS
gence provided by both local AI core and remote system management, from radio resource man-
big data centers for precise estimation of CSI agement (RRM), network management, to macro/
among UEs and between UEs and BSs. small cells management. Moreover, an intelligent
In an IR system, intelligent UEs actively and yet BS also offers the functions of edge computing,
timely acquire secret keys of different wireless chan- caching and distributed storages. A conceptual
nels. By predicting the CSI of eavesdroppers in a block diagram of an intelligent BS is shown in Fig.
real-time manner, an intelligent UE can construct 6. The implementation of intelligent BSs will pres-
antenna null space by beamforming or transmit arti- ent new challenges to communication systems and
ficial noise to the eavesdroppers. An intelligent UE network designs.
can also seek collaboration from other intelligent
UEs to maximize secrecy capacity. Moreover, via network cApAcItY
traffic and congestion situations prediction, intelli- In an IR based cellular network, there are a large
gent UEs can know whether the communication number of intelligent UEs with different traffic
network is secure or not at the moment, and thus requirements, and they are located at different
adaptively select secure routes to avoid possible places with different mobility requirements. Thus,
attacks or threats. In addition to the prediction of the maximization of network capacity is a chal-
security risks/threats, the security level of applica- lenging information-theoretical issue. We can
tions can be protected by data trace analysis in the use AI to predict or simulate different UEs’ loca-
AI cores in both local and remote processors. Intelli- tions and moving speeds on a real-time basis to
gent UEs can significantly reduce the threats of virus maximize the network capacity before a cell is
and malware attacks by partitioning different data established. After that, the locations of BSs can be
and software with the help the AI algorithms. determined with the help of the data provided by
The purpose of privacy provisioning is to hide AI algorithms aiming to maximize network-wise
private information away from any possible intrud- or cell-wise capacity. This is called smart cell-plan-
ers, eavesdroppers, or hackers. Sometimes, extra ning.
virtual information could be added to real data
through encryption, cryptographic or watermarking IntellIgent rrM
techniques for user data privacy protection. For The purpose of radio resource management
example, intelligent UEs can utilize source cod- (RRM) is to enhance system coverage, improve
ing and/or joint source-channel coding. If it is not network capacity, and satisfy users QoS require-
uniquely decodable, the eavesdroppers cannot ments. There are the relationships among cover-
obtain the real information sent through channels, age, capacity, and QoS requirements, while RRM
and thus the user’s privacy is protected. helps to strike a good balance between these
IR technologies make it possible to use more performance metrics. On one hand, UEs want to
diverse biometrics for privacy protection, such as achieve a good user experience and a high com-
electrocardiogram (ECG) and electroencephalog- munication quality. On the other hand, the BSs
raphy (EEG) based biometrics. The AI core at an co-ordinate RRM to ensure a satisfactory network
intelligent UE can use powerful pattern recognition performance.
algorithms to authenticate the identities of users In the traditional RRM, a BS manages the entire
quickly and reliably. resources and allocates resources to each user,
and UEs transmit signals with the assigned resourc-
IntellIgent bAse stAtIons es. In the IR systems, grant-free multiple access is
Likewise, an intelligent BS is a much more power- used to govern the communication processes, and
ful platform than a traditional BS. One of the most each intelligent UE makes smart use of the chan-
important features is that an intelligent BS has a nel resources. Thus, an intelligent BS acts like a
“brain” realized by the AI core available either steward to assist numerous intelligent UEs in a cell.
locally or remotely (in a cloud data center). The intelligent BS predicts the amount of resources

IEEE Wireless Communications • August 2019 99

Authorized licensed use limited to: Tampere University. Downloaded on December 30,2022 at 23:19:06 UTC from IEEE Xplore. Restrictions apply.
The estimated and required by UEs and makes a dynamic allocation Edge computing refers to the techniques to move
predicted CSI is used plan with the help of the AI core, and each intel- some computing and storage resources from data
to perform RRM and ligent UE may select its preferred plan to optimize centers to network edges close to the UEs for
to provide optimized its performance metrics under the guidance of the latency reduction.
intelligent BS. An intelligent BS should be able to take advan-
services to UEs. In The estimated and predicted CSI is used to per- tage of edge computing, caching, and distributed
addition to service form RRM and to provide optimized services to storage to perform its functions. The use of edge
perception, intelligent UEs. In addition to service perception, intelligent computing in IR networks can shorten network
RRM can also provide RRM can also provide environment cognition, traf- response time, reduce energy consumption, and
environment cognition, fic prediction, and other advanced functions to alleviate the pressure on cloud computing back-
maximize cell-wise energy efficiency (EE), which is hauls to support many low latency services [14].
traffic prediction, and defined as a ratio between cell-wise capacity divid- The major functions of edge computing include
other advanced func- ed by cell-wise power consumption. computation offloading, distributed storage, and
tions to maximize cell- fog-computing collaboration. We will discuss them
wise energy efficiency Intelligent Network Management as follows.
(EE), which is defined Intelligent network management is important to Computation Offloading: Computation off-
achieve seamless handover or switching, traffic loading in mobile edge computing (MEC) can in
as a ratio between cell- load balance, intelligent power control, fault diag- general be divided into full and partial computation
wise capacity divided nosis, and network reliability prediction. offloading schemes. The main challenge is to find
by cell-wise power When UEs move across different cells, the BSs a proper trade-off between energy consumption
consumption. involved should help the UEs for their seamless and execution delay in different application scenar-
handovers. An intelligent BS can reserve wireless ios. An intelligent BS can determine whether a UE
resources in advance and manage the inter-cell needs full cloud computing, cloud-fog cooperative
interference coordination (ICIC) function accord- computing, or only fog-computing based on the
ing to the prediction of UEs’ mobility. In this way, applications.
the probability of handover breaks can be mini- Caching and Distributed Storage: The appli-
mized, and UEs can always meet their communica- cations of caching and distributed storage help to
tion QoS requirements in the handover processes. reduce the pressure on the cloud backhaul links
Load balance is an important requirement that and improve caching efficiency. An intelligent BS
measures network congestion status and resource can predict and manage the selection of caching
utilization efficiency, and it helps to guarantee UEs contents, in order to enhance system performanc-
QoS requirements and to minimize the latency es and the users experience. Because of the phys-
of a cellular network. Therefore, an intelligent BS ical limitation in UE devices, its storage capacity
should be able to add some cellular-collaboration can never be comparable to cloud storage. Thus,
functions to balance the loading and improve net- to find a way to strike a good balance on edge
work-wise stability. computing and storage efficiency is an important
With an increasing number of UEs in both mac- research topic. We should choose the most appro-
ro-cell and small cells in 5G and beyond systems, priate source coding scheme for particular storage
the energy consumption in a cellular network may contents to optimize the storage efficiency. There
increase dramatically. Owing to the energy costs is also an open issue to find the best strategy to
of the cellular networks, a high EE in network oper- use caching and distributed storage versus cloud
ation is critically important, and green communi- storage, such that an intelligent BS can be robust
cations were proposed to improve the EE [13]. enough to recover lost and/or damaged data.
Moreover, the electromagnetic pollution becomes Fog Computing: Fog computing was proposed to
severe. The operation of intelligent BSs promis- ease the pressure on the cloud computing backhauls.
es energy saving, because it can predict system Owing to its distributed cooperative nature, a single
behaviors and perform dynamic and adaptive ener- fog element may limit the overall performance of the
gy allocation for UEs. With the help of intelligent whole communication system. Thus, one attractive
network management, the energy cost of a cellular issue is to find a way to cooperate several fog ele-
network can be reduced significantly. ments with the help of some highly efficient sched-
Fault diagnosis is another useful function uling schemes. And these fog elements may form
embedded in an intelligent BS. With AI based big a smaller resource pool similar to that in the cloud
data signal processing, intelligent network manage- computing. With joint signal processing among these
ment can ensure reliable QoS for all UEs with an fog elements, the QoS at UEs can be improved sig-
extremely high reliability and low operation cost. nificantly due to the nature of fog computing that has
a closer proximity to end-users.
Edge Computing
Recently, there has been a trend to integrate Conclusion
computing, control, and data storage into a This article introduced the basic concepts of IR sys-
cloud network. In the communication systems, tems, which was discussed from the perspective
resource utilization efficiency can be enhanced of a cellular system consisting of intelligent UEs
by aggregating baseband processing unit (BBU) and intelligent BSs. The fundamental issues on
pools. However, with an exponential growth in the designs of intelligent UEs were discussed first
the number of UEs and small cells, it creates great based on physical layer techniques. An intelligent
pressure on the backhaul links of the cloud net- UE requires a broadband RF unit to support many
works. It is not wise to increase the number of advanced functions, such as integrated radar-com-
backhaul links to satisfy the needs of UEs. Edge munication and grant-free uplink transmission.
computing emerges as a solution to tackle this Radar-communication integration plays an import-
problem, which can also be evolved to fog com- ant role in IRs to enable precision location and
puting to differentiate it from cloud computing. vision sensing in addition to self-configurable high

100 IEEE Wireless Communications • August 2019

Authorized licensed use limited to: Tampere University. Downloaded on December 30,2022 at 23:19:06 UTC from IEEE Xplore. Restrictions apply.
data rate communications. On the other hand, an [11] J. G. Proakis, Digital Communications, Publishing House of An intelligent BS
Electronics Industry, Fifth Edition, 2013.
intelligent BS can do much more than an intelligent [12] A. Mukherjee et al., “Principles of Physical Layer Security should be able to take
UE due to its resource-rich nature, where a sophis- in Multiuser Wireless Networks: A Survey, ” IEEE Commun. advantage of edge
ticated AI core helps to optimize its performance Surveys & Tutorials, vol. 16, no. 3, 2014, pp. 1550–73.
computing, caching,
in both air-links and core networks. Basically, three [13] Z. Hasan, H. Boostanimehr, and V. K. Bhargava, “Green
Cellular Networks: A Survey, Some Research Issues and and distributed stor-
fundamental blocks are needed to implement an Challenges, ” IEEE Commun. Surveys & Tutorials, vol. 13, no.
IR system, which includes a digitized wideband RF 4, 2011, pp. 524–40. age to perform its
front end, a high-speed computation core, and a [14] Y. Mao et al., “A Survey on Mobile Edge Computing: The functions. The use of
high speed backhaul link. In conclusion, IR is an Communication Perspective,” IEEE Commun. Surveys &
Tutorials, vol. 19, no. 4, 2017, pp. 2322–58. edge computing in IR
appealing research topic, but many open issues are networks can shorten
still left for our further investigations.
Additional Reading network response time,
[1] H. Hartenstein and L.P. Laberteaux, “A Tutorial Survey on
Acknowledgment Vehicular Ad Hoc Networks,” IEEE Commun. Mag., vol. 46, reduce energy con-
The work presented in this article was support- no. 6, 2008, pp. 164–71. sumption, and alleviate
ed in part by the National Natural Science Foun- the pressure on cloud
dation of China (Nos. 61671171, 61671186 Biographies computing backhauls
and U1764263); the Natural Science Foun- Qi-Yue Yu (M’08-SM’17) received her B.Eng., M.Eng., and Ph.D.
degrees from Harbin Institute of Technology (HIT), China, in to support many low
dation of Heilongjiang Province of China (No. 2004, 2006, and 2010, respectively. She is currently a full pro-
YQ2019F009); and Taiwan Ministry of Science & fessor at the School of Electronics and Information Engineering, latency services.
Technology (Nos. 106-2221-E-006-028-MY3 and HIT. From April 2007 to March 2008 she studied at the Adachi
106-2221-E-006-021-MY3). Lab, Tohoku University, Japan, and was a research assistant in
the Tohoku University Global COE program, Japan. In 2010 she
was invited to City University of Hong Kong to do research on
References multi-user MIMO technology. She was invited to the Universi-
[1] S. Haykin, “Cognitive Radio: Brain-Empowered Wireless ty of Southern Queensland to do research on the distributed
Communications,” IEEE JSAC, vol. 23, no. 2, Feb. 2005, pp. antenna system from January 2014 to March 2014. From Sep-
201–20. tember 2015 to September 2016 she was a visiting scholar at
[2] M. G. Kibria et al., “Big Data Analytics, Machine Learning, the University of California, Davis, USA. Her research interests
and Artificial Intelligence in Next-Generation Wireless Net- include modulation and coding, information theory, multi-access
works,” IEEE Access, vol. 6, 2018, pp. 32328–38. techniques and MIMO for broadband wireless communications.
[3] R. Li et al., “Intelligent 5G: When Cellular Networks Meet
Artificial Intelligence,” IEEE Wireless Commun., vol. 24, no. H ong -C hi L in (S’18) received his B.Eng. degree and M.Eng.
5, 2017, pp. 175–83. degeree in communications engineering from Harbin Institute
[4] N. Kato et al., “The Deep Learning Vision for Heterogeneous of Technology (HIT), China, in 2016 and 2019, respectively.
Network Traffic Control: Proposal, Challenges, and Future Currently he is a doctor student majoring in information and
Perspective,” IEEE Wireless Commun., vol. 24, no. 3, 2017, communication engineering at HIT. He became a student mem-
pp. 146–53. ber of the IEEE Communications Society in 2018. His research
[5] Y. Bengio, A. Courville, and P. Vincent, “Representation interests include power domain NOMA and UDN.
Learning: A Review and New Perspectives,” IEEE Trans. Pat-
tern Analysis and Machine Intelligence, vol. 35, no. 8, 2013, Hsiao-Hwa Chen (S’89, M’91, SM’00, F’10) is currently a Dis-
pp. 1798–1828. tinguished Professor in the Department of Engineering Science,
[6] G. C. Tavik et al., “The Advanced Multifunction RF Con- National Cheng Kung University, Taiwan. He obtained his B.Sc.
cept,” IEEE Trans. Microwave Theory and Techniques, vol. 53, and M.Sc. degrees from Zhejiang University, China, and a Ph.D.
no. 3, 2005, pp. 1009–20. degree from the University of Oulu, Finland, in 1982, 1985,
[7] S. Quan et al., “Radar-Communication Integration: An Over- and 1991, respectively. He has authored or co-authored over
view,” Proc. 7th IEEE/Int’l. Conf. Advanced Infocomm Tech- 400 technical papers in major international journals and confer-
nology, 2014, pp. 98–103. ences, six books, and more than 10 book chapters in the areas
[8] H. Deng and B. Himed, “Interference Mitigation Processing of communications. He has served as the general chair, TPC
for Spectrum-Sharing Between Radar and Wireless Com- chair, and symposium chair for many international conferences.
munications Systems,” IEEE Trans. Aerospace and Electronic He has served or is serving as an editor or guest editor for
Systems, vol. 49, no. 3, 2013, pp. 1911–19. numerous technical journals. He is the founding Editor-in-Chief
[9] A. Bayesteh et al., “Blind Detection of SCMA for Uplink of Wiley’s Security and Communication Networks Journal. He is
Grant-Free Multiple Access,” Proc. 2014 11th Int’l. Sympo- the recipient of the best paper award at IEEE WCNC 2008 and
sium on Wireless Commun. Systems (ISWCS), Barcelona, the recipient of the IEEE 2016 Jack Neubauer Memorial Award.
Spain, 26–29 Aug. 2014, pp. 1–6. He served as the Editor-in-Chief for IEEE Wireless Communica-
[10] B. Wang et al., “Dynamic Compressive Sensing-Based tions from 2012 to 2015. He was an elected Member-at-Large
Multi-User Detection for Uplink Grant-Free NOMA, ” IEEE of IEEE ComSoc from 2015 to 2016. He is a Fellow of IEEE, and
Commun. Lett., vol. 20, no. 11, 2016, pp. 2320–23. a Fellow of IET.

IEEE Wireless Communications • August 2019 101

Authorized licensed use limited to: Tampere University. Downloaded on December 30,2022 at 23:19:06 UTC from IEEE Xplore. Restrictions apply.