A Survey of Intelligent Reflecting Surfaces (IRSs):

Towards 6G Wireless Communication Networks

Jun Zhao Yang Liu
Assistant Professor Research Fellow
Nanyang Technological University, Singapore Nanyang Technological University, Singapore
[email protected] [email protected]
arXiv:1907.04789v3 [eess.SP] 2 Nov 2019

Abstract—Intelligent reflecting surfaces (IRSs) tune wireless Intelligent Reflecting Surface (IRS)

environments to increase spectrum and energy efficiencies. In

view of much recent attention to the IRS concept as a promising
technology for 6G wireless communications, we present a survey IRS Unit
of IRSs in this paper. Specifically, we categorize recent research
studies of IRSs as follows. For IRS-aided communications, the
summary includes capacity/data rate analyses, power/spectral
optimizations, channel estimation, deep learning-based design,
and reliability analysis. Then we review IRSs implementations
as well as the use of IRSs in secure communications, terminal- …
positioning, and other novel applications. We further identify
future research directions for IRSs, with an envision of the IRS Mobile User UAV Smart Vehicle
technology playing a critical role in 6G communication networks Base Station
similar to that of massive MIMO in 5G networks. As a timely
summary of IRSs, our work will be of interest to both researchers Fig. 1: Communications between a base station and a mobile
and practitioners working on IRSs for 6G networks. user/UAV/smart vehicle with the aid of an intelligent reflecting
Index Terms—Intelligent reflecting surface, 6G communica- surface (IRS).
tions, massive MIMO, wireless networks.

I. I NTRODUCTION
any other terminal, where a tree blocks the line-of-sight. In
As 5G communication networks are now putting into com- the rest of the paper, we mostly consider IRS-aided commu-
mercialization [1], technologies for the next-generation (i.e., nications of a BS and a MU or several MUs without loss of
6G) communications are also being explored to achieve faster generality.
and more reliable data transmissions [2]. Among these tech- Comparing IRSs with massive MIMO and other
nologies, intelligent reflecting surfaces have received much related technologies. The IRS concept can be con-
interest recently in the academia [3]–[6] and industry [7]. In sidered to be related with the technology of massive
November 2018, the Japanese mobile carrier NTT DoCoMo multiple-input multiple-output (MIMO) [4], where large ar-
and a smart radar startup MetaWave demonstrated the appli- rays of antennas are utilized to improve spectral and energy
cation of meta-structure technology to data communication in efficiencies. Hence, we envision IRS to play a crucial role in
28GHz band [7]. 6G communication networks similar to that of massive MIMO
Intelligent reflecting surfaces. An intelligent reflecting in 5G networks. Thus, IRS can be used to help achieve massive
surface (IRS) [3] comprises an array of IRS units, each of MIMO 2.0 [15]. What differentiates IRSs from massive MIMO
which can independently incur some change to the incident is that IRSs tune the wireless propagation environment for
signal [8]. The change in general may be about the phase, the communication [11], [16]. Hu et al. [10] present the first
amplitude, frequency, or even polarization [8]. To date, in most analysis on information-transfer capabilities of IRSs. They
studies [3], [4], [9]–[13], the change is considered as a phase prove that the capacity which can be harvested per square
shift only to the incident signal, so that an IRS consumes metre (m2 ) surface-area has a linear relationship with the
no transmit power. In essence, an IRS intelligently configures average transmit power, instead of being logarithmic in the
the wireless environment to help the transmissions between case of a massive-MIMO deployment. In addition to massive
the sender and receiver, when direct communications have MIMO, other existing technologies which have been compared
bad qualities. Example places to put IRSs are walls, building with IRSs in recent studies [17], [18] include backscatter com-
facades, and ceilings [14]. munication [19], millimeter (mm)-wave communication [20],
IRS-aided communications. Figure 1 illustrates IRS-aided and network densification [21]. However, these related tech-
communications between a base station (BS) and a mobile user nologies do not control the wireless environment and typically
(MU), an unmanned aerial vehicle (UAV), a smart vehicle, or consume much power [22].

1
 Comparing IRS with other recent notions. In addition to A. Capacity/data rate analyses of IRS-aided communications
IRS, other recent notions have also been given in the literature. Hu et al. [10] establish that the capacity achieved per
These names include the following: square metre (m2 ) surface-area is linearly proportional with
• large intelligent surface (LIS) [4], [5], [9], [10], [13], the average transmit power, instead of having a logarithmic
[23]–[27], which is preferred for asymptotic analysis relationship as the case of massive MIMO.
assuming infinite length(s) of the surface, or sufficiently Hu et al. [4] analyze capacities of single-antenna ter-
large number of antennas. It is worth noting that in [4], minals communicating to an IRS. They first consider the
[5], [9], [10], [13], [23]–[26], the surface indeed transmits entire surface as a receiving antenna array. In this setting,
signals actively, instead of passively reflecting signals for a sufficiently large surface-area, their result is that the
from base stations as in the case of IRS. received signal following a matched-filtering operation can
• large intelligent metasurface (LIM) [28] and pro- be well represented by a sinc-function-like intersymbol in-
grammable/reconfigurable metasurface [29]–[32], where terference channel. Afterwards, they derive the capacity per
the prefix “meta-” derives from a metallic pattern called square metre (m2 ) surface-area and show its convergence to
P
meta-atom, through which the surface is engineered to 2N0 [nats/s/Hz/volume-unit] when the wavelength λ tends to
have some property which is not found in naturally zero, where P is the transmit power per volume-unit (which
occurring surfaces can be m, m2 , or m3 ), and N0 denotes the additive white
• smart reflect-arrays [33]–[35], which emphasize the sur- Gaussian noise’s power spectral density.
face’s reflection function (in the same spirit as IRS), A recent work [23] by Hu et al. examines the degradations
instead of being used for transmission that is provided in capacity when IRSs are allowed to have hardware impair-
by amplify-and-forward/decode-and-forward relays such ments. They show that splitting an IRS into an array consisting
as MIMO technologies, of a number of small IRS units can mitigate the degradation.
• reconfigurable intelligent surface [8], [36], [37], where The conference paper [25] and its full version [9] by
“reconfigurable” means that the angle of reflection can Jung et al. present an asymptotic analysis of the uplink data
be reconfigured (via software) regardless of the angle of rate in an IRS-based communication system. Their analysis
incidence, considers channel estimation errors and model interference
• software-defined surface (SDS) [38] and software-defined channels to be spatially correlated Rician fading [45]. Fur-
metasurfaces (SDMs) [39], which are inspired by the thermore, channel hardening effects are also taken into con-
definition of software-defined radio [40] and consider the sideration. They show that the asymptotic capacity result is in
interaction between the surface and incoming waves to accordance with the exact mutual information as the numbers
have a software-defined fashion, of antennas and mobile devices increase. For uplink data rates
• passive intelligent surface (PIS) [41] and passive intelli- in IRSs, Jung et al. [25] present an asymptotic analysis where
gent mirrors [42], which underline the passive reflection channel estimation errors are taken into consideration.
without consuming transmit power. Guo et al. [11] maximize the weighted sum of downlink
For consistency, in the rest of the paper, we will use the name rates by finding the optimal active beamforming at the BS and
IRS instead of other terms listed above. In addition, it is worth passive beamforming at the IRS, where each weight represents
mentioning that frequency selective surfaces recently studied the priority of a mobile user. For practicality and simplicity
in [43], [44] are used to reduce the coupling effects in ultra- of optimization analysis, they consider the IRS phase shifts to
massive MIMO and are different from IRSs. take only discrete values.
Nadeem et al. [27] consider a single-cell multi-user system,
Organization of this paper. In Section II, we classify recent
where a base station (BS) with multiple antennas communi-
studies of IRSs into different categories. Section III highlights
cates many single-antenna users via an IRS. For the downlink,
the differences between our work and recent reviews of IRSs.
they investigate how to maximizes the SINR by optimizing the
In Section IV, we identify several directions for future research
linear precoder and power allocation at the BS, and the IRS
of IRSs. Finally, Section V concludes the paper.
phase matrix. Their analysis involves different rank structures
of the channel matrix between the BS and the IRS, and also
II. C ATEGORIZING R ECENT S TUDIES OF I NTELLIGENT the spatial correlations among the IRS elements.
R EFLECTING S URFACES The conference paper [17] and its full version [46] by
Wu et al. maximize the signal-to-interference-plus-noise ratio
We now classify recent research work on IRSs. For IRS- (SINR) received at mobile users by jointly optimizing the
aided communications, we discuss capacity/data rate anal- IRS phase matrix and the transmit beamforming of the active
yses, power/spectral optimizations, channel estimation, deep antenna array at the BS.
learning-based design, and reliability analysis. We also review
IRSs implementations as well as the use of IRSs in secure B. Power/spectral optimizations in IRS-aided communications
communications, terminal-positioning, and other novel appli- In this subsection, we summarize recent optimization studies
cations. of power/spectral efficiency in IRS-aided communications.

2
 The work [3] and its earlier version [37] by Huang et al. matrix factorization, using the information of the IRS state
maximize the bit-per-Joule energy efficiency of the downlinks matrix, which contains the ON/OFF information of each IRS
by finding the IRS phase matrix and the optimal power unit at each time. The third stage uses properties of the channel
allocation at the BS. To simplify the analysis, they consider matrices to recover the missing entries. The three stages are
that the BS employs a well-designed zero-forcing precoding solved by the algorithms of bilinear generalized approximate
matrix to achieve perfect interference suppression among message passing [51], greedy pursuit [52], and Riemannian
signals received by the mobile users. manifold gradient [53], respectively.
Fu et al. [47] solve the downlink transmit power minimiza- Mishra and Johansson [41] design a channel estimation
tion for an IRS-aided multiple access network by optimizing mechanism for IRS-aided energy transfer from a power beacon
both the transmit beamformers at the BS and the phase shift with multiple antennas to a single-antenna user. They further
matrix at the IRS. use the estimation results to design active and passive energy
Yu et al. [12] and Jung et al. [13] both consider max- beamforming at the power beacon and IRS, respectively.
imization problems of the spectral efficiency in IRS-aided Independent of [22], the work [41] also proposes the approach
communication systems. Specifically, the work [12] maximizes of dividing the total channel estimation time into several sub-
the spectral efficiency by optimizing the beamformer at the phases and allowing only one IRS unit to be ON in each
access point and the IRS phase shifts. The study [13] considers sub-phase. Moreover, the channel estimation protocol in [41]
the typical setting where pilot signaling is used to obtain also permits that ON/OFF modes of IRS units may not be
channel state information and hence pilot training structure implemented perfectly in practice.
impacts the achievable spectral efficiency. The authors of [13] As already discussed in Section II-A, Jung et al. [9], [25]
first derive an asymptotic value of the spectral efficiency and take channel estimation errors into account to analyze uplink
then use the result to find the optimal pilot training length data rates of IRS-aided communications.
which maximizes the asymptotic spectral efficiency.
D. Deep learning-based design for IRS-aided communications
C. Channel estimation for IRS-aided communications Liaskos et al. [39] use deep learning for configuring IRSs to
In a typical setting, IRSs are passive and do not have sensing aid wireless communications. Specifically, they regard wireless
capabilities, so downlinks are estimated at the base station via propagation as a deep neural network, where IRS units are
control signals. Then the channel information is reported by neurons and their cross-interactions as links. After training
the base station to the IRS controller, which sets the phase from the data, the wireless network learns the propagation
shifts accordingly [22]. basics of IRSs and configures them to the optimal setting.
Nadeem et al. [22] present a channel estimation protocol As already discussed in Section II-C, Taha et al. [48]
based on minimum mean squared error (MMSE). Specifically, utilize deep learning for channel estimation in IRS-aided
they divide the total channel estimation time into a number of communications. Specifically, qualities of wireless channels
sub-phases. In the first sub-phase, all IRS units are turned OFF at all the IRS units are learnt via a deep neural network using
and the base station estimates the direct channels for all users. channels seen only at those IRS units which are connected to
In each of the following sub-phases, each IRS element takes the baseband of the IRS controller. Furthermore, deep learning
turns to be ON while all other IRS units are OFF, to allow is used to guide the IRS to learn the optimal interaction with
estimations by the base station. At the end of the protocol, the incident signals. A short conference version of [48] appears
estimation results of all sub-phases are taken together using the as [54].
MMSE approach to obtain a comprehensive picture of channel
estimation. E. Reliability analysis of IRS-aided communications
Taha et al. [48] address the channel estimation problem Jung et al. [26] present a reliability analysis of IRS-aided
using compressive sensing [49] and deep learning [50]. In communications in terms of uplink rate distribution and outage
their setting, IRS units which are connected to the baseband of probability. The distribution of the data sum-rate is obtained
the IRS controller are referred to as being active, and the rest using the Lyapunov central limit theorem [55]. Then the outage
IRS units are considered as passive. In the proposed solutions, probability is given by the probability that the sum-rate is
they utilize compressive sensing and deep learning techniques below a desired value. Note that although the authors’ earlier
respectively to estimate the channels at all the IRS units from work [9] analyzes the mean and variance of the rate, the
the channels seen only at the active IRS units. probabilistic distribution of the rate is needed to compute the
He et al. [28] tackle channel estimation for IRS-aided outage probability.
MIMO systems using a three-stage mechanism. The three
stages include sparse matrix factorization, ambiguity elimina- F. Implementations of IRSs
tion, and matrix completion, respectively. The first stage takes Hu et al. [4] investigate IRS implementations as a grid of
the received signal and uses matrix factorization to derive the antenna elements. Subject to the constraint that every spent
channel matrix between the base station and the IRS, as well antenna can earn one signal space dimension, they show that
as the channel matrix between the IRS and the mobile user. the hexagonal lattice minimizes the IRS surface-area given
The second stage eliminates the ambiguity of the solutions to the desired number of independent signal dimensions. The

3
analysis of [4] leverages the classical lattice theory [56], which can be exactly derived and the latter is solved by letting the
has various applications beyond wireless communication, in- objective function be its approximation using results from [76].
cluding information theory [57], cryptography [58], machine Different from [6], [61], [63] above, Chen et al. [62]
learning [59], and knowledge representation [60]. examine the case of multiple legitimate receivers and multiple
Taha et al. [48], [54] propose IRS architectures consisting eavesdroppers. Specifically, in [62], the considered IRS-aided
of two types of IRS units: active and passive ones. An downlink broadcast system consists of a multi-antenna base
active IRS unit is connected to the baseband of the IRS station, multiple legitimate receivers with each having single
controller, whereas a passive one is not. Then the system antenna, and multiple eavesdroppers. Then [62] maximizes
optimization can be booted from channel information at the the minimum secrecy data rate among all legitimate receivers
active IRS units, which capture the environmental conditions by finding the optimal transmit beamforming and IRS phase
and sender/receiver locations. shifts via alternating optimization, where both cases of phase
shifts taking discrete and continuous values are considered.
G. IRSs for secure communications Moreover, [62] also studies a case where the IRS reflecting
A number of recent studies [6], [61]–[63] have leveraged amplitude is allowed to be less than 1. The optimization
IRSs to secure the physical layer of wireless communications. techniques used by [62] include path-following iterative al-
In the simplest wiretap channel introduced first by Wyner [64], gorithms [77] and heuristic projection methods [78].
a transmitter and a legitimate receiver have communications,
which are wiretapped by an eavesdropper. This simple model H. IRSs for terminal-positioning and other novel applications
has been extended to the broadcast wiretap channel [65], com- The conference paper [24] and its journal version [5] by
pound wiretap channel [66], Gaussian wiretap channel [67], Hu et al. utilize IRS for terminal-positioning. In particular,
and MIMO wiretap channel [68]. they derive the Cramér–Rao lower bounds (CRLBs) [79] for
The intuition of applying an IRS to secure communications all three Cartesian dimensions of a terminal. The result is
under a wiretap channel is that an IRS can be used to increase that in general the CRLB decreases quadratically with respect
the data rate at a legitimate receiver while decreasing the data to the IRS surface-area, except for the case of a terminal
rate at an eavesdropper. This improves the difference between locating perpendicular to the IRS center where the CRLB
the two rates (the former minus the latter), which is defined for the distance from the IRS decreases linearly in the IRS
as the secrecy data rate. We now elaborate the contributions surface-area. The analyses in [5], [24] also involve the Fisher
of [6], [61]–[63]. information, since it is no greater than the CRLB of an
Cui et al. [6], Shen et al. [63], and Yu et al. [61] study unbiased estimator [80].
an IRS-aided communication system with a multi-antenna Basar [38] uses IRS to aid index modulation (IM), which
transmitter communicating to a single-antenna legitimate re- manipulates the indices of the transmit entities to convey
ceiver in the presence of an eavesdropper. All of the three information [81]. In [81], with IRS-space shift keying and
papers [6], [61], [63] consider the optimal design of the IRS-spatial modulation, IM is realized by selecting a particular
base station’s transmit beamforming and the IRS’s reflect receive antenna index based on the information bits.
beamforming to maximize the legitimate communication link’s Jiang et al. [82] employ IRS to assist over-the-air compu-
secrecy rate subject to the base station’s transmit power tation (AirComp), where the base station aims to compute
constraint. The differences among [6], [61], [63] lie in the some function from data of all mobile users. The opti-
specific details of the approaches to solving the optimization mization problem formulated in [82] is find the IRS phase
problems. In particular, [6] uses alternating optimization [69] shifts and the base station’ decoding vectors to minimize
to design the transmit beamforming and IRS phase shifts the distortion after signal decoding, which is defined as the
alternately. More specifically, in each iteration of [6], the mean squared error among the decoding results. In view of
transmit beamforming can be exactly computed, but for de- the non-convexity of the problem, the authors of [82] use
riving the IRS phase shifts, the semidefinite relaxation [70] the majorization-minimization [76] technique to propose an
technique and the Charnes–Cooper transformation [71] are alternating difference-of-convex algorithm [83].
combined to convert a non-convex problem into a convex Basar [84] proposes the novel concept of using an IRS
semidefinite programming problem [72], which can be solved as an access point (AP). In the proposed design, an radio
by the interior-point method [73]. In each iteration of [61], frequency (RF) signal generator close to the IRS generates
the transmit beamforming is determined by a generalized an unmodulated carrier signal and sends it to the IRS with
eigenvalue problem [74], while each phase shift is solved by negligible fading. Then the adjustable IRS phase shifts are
an element-wise block coordinate descent method [75], which used to convey information bits.
can be seen as a generalization of alternating optimization Mishra and Johansson [41] leverage IRS to support wireless
and optimizes the objective function with respect to a block energy transfer from a power beacon with multiple antennas
of optimization variables in each iteration while fixing the to a single-antenna user. As discussed previously in Sec-
other blocks. Finally, the work [63] also adopts alternating tion II-C, [41] first presents a channel estimation protocol and
optimization to compute transmit beamforming and the IRS then uses the estimation results to set energy beamforming at
phase shifts alternately, where in each iteration the former the power beacon and IRS, respectively.

4
 III. R ECENT R EVIEWS OF I NTELLIGENT R EFLECTING R EFERENCES
S URFACES AND R ELATED T ECHNOLOGIES [1] M. Patzold, “It’s time to go big with 5G mobile radio,” IEEE Vehicular
Three recent studies [8], [14], [18] provide nice overviews Technology Magazine, vol. 13, no. 4, pp. 4–10, 2018.
[2] W. Saad, M. Bennis, and M. Chen, “A vision of 6G wireless systems:
for the IRS technology. Compared with our current paper, Applications, trends, technologies, and open research problems,” arXiv
Basar et al. [8] elaborate many mathematical details in IRS- preprint arXiv:1902.10265, 2019.
aided communications. Di Renzo et al. [18] highlight the flavor [3] C. Huang, A. Zappone, G. C. Alexandropoulos, M. Debbah, and
C. Yuen, “Reconfigurable intelligent surfaces for energy efficiency in
of artificial intelligence in IRSs for empowering smart radio wireless communication,” IEEE Transactions on Wireless Communica-
environments. Wu et al. [14] focus on the key challenges in tions, 2019.
the design and implementation of IRS-aided communications. [4] S. Hu, F. Rusek, and O. Edfors, “Beyond massive MIMO: The potential
of data transmission with large intelligent surfaces,” IEEE Transactions
Compared with our current paper, [8], [14], [18] do not on Signal Processing, vol. 66, no. 10, pp. 2746–2758, 2018.
thoroughly categorize IRS studies appearing in the literature. [5] ——, “Beyond massive MIMO: The potential of positioning with large
Moreover, recent papers which are not covered by [8], [14], intelligent surfaces,” IEEE Transactions on Signal Processing, vol. 66,
no. 7, pp. 1761–1774, 2018.
[18] but are discussed in our current work include [6], [11], [6] M. Cui, G. Zhang, and R. Zhang, “Secure wireless communication via
[12], [22]–[25], [47], [63]. intelligent reflecting surface,” IEEE Wireless Communications Letters,
Other recent reviews are listed as follows. Tasolam- 2019.
[7] “NTT DoCoMo and Metawave announce successful demonstration
prou et al. [85] present a detailed discussion about issues of 28GHz-band 5G using world’s first meta-structure technology,”
in physical implementations of IRSs for wireless millimeter https://www.marketwatch.com/press-release/ntt-docomo-and-metawave-announce-succ
(mm)-wave communications. Sanguinetti et al. [15] sum- accessed: July 7, 2019.
[8] E. Basar, M. Di Renzo, J. de Rosny, M. Debbah, M.-S. Alouini, and
marize techniques for improving massive MIMO to its 2.0 R. Zhang, “Wireless communications through reconfigurable intelligent
version, where the IRS technology is only briefly mentioned. surfaces,” arXiv preprint arXiv:1906.09490, 2019.
Björnson et al. [86] discuss different potential technologies re- [9] M. Jung, W. Saad, Y. Jang, G. Kong, and S. Choi, “Performance analysis
of large intelligence surfaces (LISs): Asymptotic data rate and channel
lated to massive MIMO, including intelligent massive MIMO, hardening effects,” arXiv preprint arXiv:1810.05667, 2018.
large-scale MIMO radar, and holographic massive MIMO. [10] S. Hu, F. Rusek, and O. Edfors, “The potential of using large antenna
arrays on intelligent surfaces,” in IEEE 85th Vehicular Technology
IV. F UTURE R ESEARCH D IRECTIONS FOR I NTELLIGENT Conference (VTC Spring), 2017, pp. 1–6.
[11] H. Guo, Y.-C. Liang, J. Chen, and E. G. Larsson, “Weighted sum-
R EFLECTING S URFACES rate optimization for intelligent reflecting surface enhanced wireless
networks,” arXiv preprint arXiv:1905.07920, 2019.
We envision that IRS will play a fundamental role in 6G [12] X. Yu, D. Xu, and R. Schober, “MISO wireless communication systems
wireless communication networks similar to that of massive via intelligent reflecting surfaces,” arXiv preprint arXiv:1904.12199,
MIMO in 5G networks. Moreover, IRS can be utilized to help 2019.
[13] M. Jung, W. Saad, and G. Kong, “Performance analysis of large
realize massive MIMO 2.0 [15]. Now we identify three future intelligent surfaces (LISs): Uplink spectral efficiency and pilot training,”
directions for IRS research. arXiv preprint arXiv:1904.00453, 2019.
First, most existing studies on IRSs and their applications [14] Q. Wu and R. Zhang, “Towards smart and reconfigurable environment:
Intelligent reflecting surface aided wireless network,” IEEE Communi-
to wireless communications are about theoretical analyses cations Magazine, to appear, 2019 (arXiv preprint arXiv:1905.00152).
with simulations as validations. Hence, an important research [15] L. Sanguinetti, E. Björnson, and J. Hoydis, “Towards massive MIMO
direction is to confirm the theoretical results with data from 2.0: Understanding spatial correlation, interference suppression, and
real-world system implementations and experiments. pilot contamination,” arXiv preprint arXiv:1904.03406, 2019.
[16] C. Liaskos, S. Nie, A. Tsioliaridou, A. Pitsillides, S. Ioannidis, and
Second, existing models on how IRSs change the incident I. Akyildiz, “A new wireless communication paradigm through software-
signals are simple. In contrast, an IRS’s behavior depends on controlled metasurfaces,” IEEE Communications Magazine, vol. 56, pp.
162–169, September 2018.
its physical materials and manufacturing processes [14]. Mod-
[17] Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wireless
els taking these issues into account can more accurately guide network: Joint active and passive beamforming design,” in IEEE Global
the optimization of IRSs for aiding wireless communications. Communications Conference (GLOBECOM), 2018, pp. 1–6.
Third, scaling laws need to be established for a funda- [18] M. Di Renzo, M. Debbah, D.-T. Phan-Huy, A. Zappone, M.-S.
Alouini, C. Yuen, V. Sciancalepore, G. C. Alexandropoulos, J. Hoydis,
mental understanding of the performance limits in IRS-aided H. Gacanin, J. de Rosny, A. Bounceu, G. Lerosey, and M. Fink,
communications. Answering this question requires a deep “Smart radio environments empowered by reconfigurable AI meta-
understanding of how IRSs impact traditional information- surfaces: An idea whose time has come,” EURASIP Journal on Wireless
Communications and Networking, 2019.
theoretic models. [19] A. N. Parks, A. Liu, S. Gollakota, and J. R. Smith, “Turbocharging
ambient backscatter communication,” ACM SIGCOMM Computer Com-
V. C ONCLUSION munication Review, vol. 44, no. 4, pp. 619–630, 2015.
[20] T. S. Rappaport, G. R. MacCartney, M. K. Samimi, and S. Sun, “Wide-
In this survey paper, we categorize recent studies of IRSs band millimeter-wave propagation measurements and channel models
and identify future research directions for IRSs. As a promis- for future wireless communication system design,” IEEE Transactions
on Communications, vol. 63, no. 9, pp. 3029–3056, 2015.
ing technology to facilitate 6G wireless communications, IRSs [21] J. G. Andrews, X. Zhang, G. D. Durgin, and A. K. Gupta, “Are we
induce smart radio environments to increase spectrum and approaching the fundamental limits of wireless network densification?”
energy efficiencies. We envision that the coming years will IEEE Communications Magazine, vol. 54, no. 10, pp. 184–190, 2016.
[22] Q.-U.-A. Nadeem, A. Kammoun, A. Chaaban, M. Debbah, and M.-
see much research and development for the IRS technology to S. Alouini, “Intelligent reflecting surface assisted multi-user MISO
build 6G communications. communication,” arXiv preprint arXiv:1906.02360, 2019.

5
[23] S. Hu, F. Rusek, and O. Edfors, “Capacity degradation with modeling on Nanoscale Computing and Communication (ACM NanoCom), 2018,
hardware impairment in large intelligent surface,” in IEEE Global pp. 27:1–27:6.
Communications Conference (GLOBECOM), 2018, pp. 1–6. [44] ——, “Mutual coupling reduction for ultra-dense multi-band plasmonic
[24] ——, “Cramér–Rao lower bounds for positioning with large intelligent nano-antenna arrays using graphene-based frequency selective surface,”
surfaces,” in IEEE Vehicular Technology Conference (VTC-Fall), 2017, IEEE Access, vol. 7, pp. 33 214–33 225, 2019.
pp. 1–6. [45] K. A. Hamdi, “Capacity of MRC on correlated Rician fading channels,”
[25] M. Jung, W. Saad, Y. Jang, G. Kong, and S. Choi, IEEE Transactions on Communications, vol. 56, pp. 708–711, 2008.
“Uplink data rate in large intelligent surfaces: Asymptotic [46] Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wireless
analysis under channel estimation errors,” 2018. [Online]. Available: network via joint active and passive beamforming,” IEEE Transac-
https://www.researchgate.net/publication/328827179 Uplink Data Rate in LargetionsIntelligent Surfaces
on Wireless Asymptotic Analysis
Communications, under
to appear, Channel
2019 (arXivEstimation
preprint Errors
[26] ——, “Reliability analysis of large intelligent surfaces (LISs): Rate arXiv:1810.03961).
distribution and outage probability,” arXiv preprint arXiv:1903.11456, [47] M. Fu, Y. Zhou, and Y. Shi, “Intelligent reflecting surface for
2019. downlink non-orthogonal multiple access networks,” arXiv preprint
[27] Q.-U.-A. Nadeem, A. Kammoun, A. Chaaban, M. Debbah, and M.-S. arXiv:1906.09434, 2019.
Alouini, “Asymptotic analysis of large intelligent surface assisted MIMO [48] A. Taha, M. Alrabeiah, and A. Alkhateeb, “Enabling large intelligent
communication,” arXiv preprint arXiv:1903.08127, 2019. surfaces with compressive sensing and deep learning,” arXiv preprint
[28] Z.-Q. He and X. Yuan, “Cascaded channel estimation for large intelligent arXiv:1904.10136, 2019.
metasurface assisted massive MIMO,” arXiv preprint arXiv:1905.07948, [49] Z. Gao, L. Dai, S. Han, I. Chih-Lin, Z. Wang, and L. H. Hanzo,
2019. “Compressive sensing techniques for next-generation wireless commu-
[29] M. Di Renzo and J. Song, “Reflection probability in wireless networks nications,” IEEE Wireless Communications, vol. 25, pp. 144–153, 2018.
with metasurface-coated environmental objects: An approach based on [50] C. Zhang, P. Patras, and H. Haddadi, “Deep learning in mobile and
random spatial processes,” arXiv preprint arXiv:1901.01046, 2019. wireless networking: A survey,” IEEE Communications Surveys and
[30] L. Zhang, X. Q. Chen, S. Liu, Q. Zhang, J. Zhao, J. Y. Dai, G. D. Tutorials, March 2019.
Bai, X. Wan, Q. Cheng, G. Castaldi et al., “Space-time-coding digital [51] J. Parker, P. Schniter, and V. Cevher, “Bilinear generalized approximate
metasurfaces,” Nature communications, vol. 9, no. 1, p. 4334, 2018. message passing-Part I: Derivation,” IEEE Transactions on Signal Pro-
[31] J. Zhao, X. Yang, J. Y. Dai, Q. Cheng, X. Li, N. H. Qi, J. C. Ke, G. D. cessing, vol. 62, pp. 5839–5853, 2014.
Bai, S. Liu, S. Jin et al., “Programmable time-domain digital-coding [52] P. Cheng, L. Gui, Y. Rui, Y. Guo, X. Huang, and W. Zhang, “Compressed
metasurface for non-linear harmonic manipulation and new wireless sensing based channel estimation for two-way relay networks,” IEEE
communication systems,” National Science Review, vol. 6, no. 2, pp. Wireless Communications Letters, vol. 1, pp. 201–204, June 2012.
231–238, 2018. [53] J. Knuth and P. Barooah, “Collaborative localization with heterogeneous
[32] S. Foo, “Liquid-crystal reconfigurable metasurface reflectors,” in 2017 inter-robot measurements by Riemannian optimization,” IEEE Interna-
IEEE International Symposium on Antennas and Propagation & US- tional Conference on Robotics and Automation, pp. 1534–1539, 2013.
NC/URSI National Radio Science Meeting. IEEE, 2017, pp. 2069–
[54] A. Taha, M. Alrabeiah, and A. Alkhateeb, “Deep learning
2070.
for large intelligent surfaces in millimeter wave and
[33] X. Tan, Z. Sun, D. Koutsonikolas, and J. M. Jornet, “Enabling indoor massive MIMO systems,” May 2019. [Online]. Available:
mobile millimeter-wave networks based on smart reflect-arrays,” in IEEE https://www.researchgate.net/publication/333246370 Deep Learning for Large Intelli
Conference on Computer Communications (INFOCOM), 2018, pp. 270–
[55] M. Chafii, J. Palicot, R. Gribonval, and F. Bader, “A necessary condition
278.
for waveforms with better PAPR than OFDM,” IEEE Transactions on
[34] X. Tan, Z. Sun, J. M. Jornet, and D. Pados, “Increasing indoor spec-
Communications, vol. 64, no. 8, pp. 3395–3405, 2016.
trum sharing capacity using smart reflect-array,” in IEEE International
[56] B. Ganter and R. Wille, “Applied lattice theory: Formal concept analy-
Conference on Communications (ICC), 2016, pp. 1–6.
sis,” in In General Lattice Theory, G. Grätzer editor, Birkhäuser, 1997.
[35] S. Nie, J. M. Jornet, and I. F. Akyildiz, “Intelligent environments
based on ultra-massive MIMO platforms for wireless communication in [57] M. O. Damen, H. El Gamal, and G. Caire, “On maximum-likelihood
millimeter wave and terahertz bands,” in IEEE International Conference detection and the search for the closest lattice point,” IEEE Transactions
on Acoustics, Speech and Signal Processing (ICASSP), 2019, pp. 7849– on information theory, vol. 49, no. 10, pp. 2389–2402, 2003.
7853. [58] K. Sotiraki, M. Zampetakis, and G. Zirdelis, “PPP-completeness with
[36] C. Huang, A. Zappone, G. C. Alexandropoulos, M. Debbah, and connections to cryptography,” in IEEE 59th Annual Symposium on
C. Yuen, “Reconfigurable intelligent surfaces for energy efficiency in Foundations of Computer Science (FOCS), 2018, pp. 148–158.
wireless communication,” IEEE Transactions on Wireless Communica- [59] M. Liquiere and J. Sallantin, “Structural machine learning with Galois
tions, 2019. lattice and graphs,” in International Conference on Machine Learning
[37] ——, “Large intelligent surfaces for energy efficiency in wireless (ICML), vol. 98, 1998, pp. 305–313.
communication,” arXiv preprint arXiv:1810.06934v1, 2018. [60] W.-Z. Wu, Y. Leung, and J.-S. Mi, “Granular computing and knowledge
[38] E. Basar, “Large intelligent surface-based index modulation: A new reduction in formal contexts,” IEEE Transactions on Knowledge and
beyond MIMO paradigm for 6G,” arXiv preprint arXiv:1904.06704, Data Engineering (TKDE), vol. 21, no. 10, pp. 1461–1474, 2008.
2019. [61] X. Yu, D. Xu, and R. Schober, “Enabling secure wireless
[39] C. Liaskos, A. Tsioliaridou, S. Nie, A. Pitsillides, S. Ioannidis, and communications via intelligent reflecting surfaces,” arXiv preprint
I. Akyildiz, “An interpretable neural network for configuring pro- arXiv:1904.09573, 2019.
grammable wireless environments,” IEEE International Workshop on [62] J. Chen, Y.-C. Liang, Y. Pei, and H. Guo, “Intelligent reflecting surface:
Signal Processing Advances in Wireless Communications (SPAWC), A programmable wireless environment for physical layer security,” arXiv
2019. preprint arXiv:1905.03689, 2019.
[40] T. Ulversoy, “Software defined radio: Challenges and opportunities,” [63] H. Shen, W. Xu, S. Gong, Z. He, and C. Zhao, “Secrecy rate
IEEE Communications Surveys & Tutorials, vol. 12, no. 4, pp. 531– maximization for intelligent reflecting surface assisted multi-antenna
550, 2010. communications,” IEEE Communications Letters, 2019.
[41] D. Mishra and H. Johansson, “Channel estimation and low-complexity [64] A. D. Wyner, “The wire-tap channel,” Bell System Technical Journal,
beamforming design for passive intelligent surface assisted MISO wire- vol. 54, no. 8, pp. 1355–1387, 1975.
less energy transfer,” in IEEE International Conference on Acoustics, [65] I. Csiszár and J. Korner, “Broadcast channels with confidential mes-
Speech and Signal Processing (ICASSP), 2019, pp. 4659–4663. sages,” IEEE Transactions on Information Theory, vol. 24, no. 3, pp.
[42] C. Huang, A. Zappone, M. Debbah, and C. Yuen, “Achievable rate 339–348, 1978.
maximization by passive intelligent mirrors,” in IEEE International [66] Y. Liang, G. Kramer, H. V. Poor, and S. Shamai, “Compound wiretap
Conference on Acoustics, Speech and Signal Processing (ICASSP), 2018, channels,” EURASIP Journal on Wireless Communications and Network-
pp. 3714–3718. ing, vol. 2009, p. 5, 2009.
[43] B. Zhang, J. M. Jornet, I. F. Akyildiz, and Z. P. Wu, “Graphene-based [67] S. Leung-Yan-Cheong and M. Hellman, “The Gaussian wire-tap chan-
frequency selective surface decoupling structure for ultra-dense multi- nel,” IEEE Transactions on Information Theory, vol. 24, no. 4, pp. 451–
band plasmonic nano-antenna arrays,” in ACM International Conference 456, 1978.

6
[68] F. Oggier and B. Hassibi, “The secrecy capacity of the MIMO wiretap
channel,” IEEE Transactions on Information Theory, vol. 8, no. 57, pp.
4961–4972, 2011.
[69] L. Liu, R. Zhang, and K.-C. Chua, “Multi-antenna wireless powered
communication with energy beamforming,” IEEE Transactions on Com-
munications, vol. 62, no. 12, pp. 4349–4361, 2014.
[70] Z.-Q. Luo, W.-K. Ma, A. So, Y. Ye, and S. Zhang, “Semidefinite
relaxation of quadratic optimization problems,” IEEE Signal Processing
Magazine, vol. 27, pp. 20–34, 2010.
[71] M. Naeem, K. Illanko, A. Karmokar, A. Anpalagan, and M. Jaseemud-
din, “Energy-efficient cognitive radio sensor networks: Parametric and
convex transformations,” Sensors, vol. 13, no. 8, pp. 32–50, 2013.
[72] R. W. Ouyang, A. K.-S. Wong, and C.-T. Lea, “Received signal strength-
based wireless localization via semidefinite programming: Noncoop-
erative and cooperative schemes,” IEEE Transactions on Vehicular
Technology, vol. 59, no. 3, pp. 1307–1318, 2010.
[73] E. Xu, Z. Ding, and S. Dasgupta, “Source localization in wireless sensor
networks from signal time-of-arrival measurements,” IEEE Transactions
on Signal Processing, vol. 59, no. 6, pp. 2887–2897, 2011.
[74] R. R. Nadakuditi and J. W. Silverstein, “Fundamental limit of sample
generalized eigenvalue based detection of signals in noise using rel-
atively few signal-bearing and noise-only samples,” IEEE Journal of
Selected Topics in Signal Processing, vol. 4, no. 3, pp. 468–480, 2010.
[75] Q. Wu, Y. Zeng, and R. Zhang, “Joint trajectory and communication
design for multi-UAV enabled wireless networks,” IEEE Transactions
on Wireless Communications, vol. 17, no. 3, pp. 2109–2121, 2018.
[76] Y. Sun, P. Babu, and D. P. Palomar, “Majorization-minimization algo-
rithms in signal processing, communications, and machine learning,”
IEEE Transactions on Signal Processing, vol. 65, no. 3, pp. 794–816,
2016.
[77] A. A. Nasir, H. D. Tuan, D. T. Ngo, S. Durrani, and D. I. Kim, “Path-
following algorithms for beamforming and signal splitting in RF energy
harvesting networks,” IEEE Communications Letters, vol. 20, no. 8, pp.
1687–1690, 2016.
[78] O. P. Ferreira and S. Z. Németh, “Projection onto simplicial cones by a
semi-smooth Newton method,” Optimization Letters, vol. 9, no. 4, pp.
731–741, 2015.
[79] M. Angjelichinoski, D. Denkovski, V. Atanasovski, and L. Gavrilovska,
“Cramér–Rao lower bounds of RSS-based localization with anchor
position uncertainty,” IEEE Transactions on Information Theory, vol. 61,
no. 5, pp. 2807–2834, 2015.
[80] Y. Zhao, X. Fan, C.-Z. Xu, and X. Li, “ER-CRLB: An extended recursive
Cramér–Rao lower bound fundamental analysis method for indoor local-
ization systems,” IEEE Transactions on Vehicular Technology, vol. 66,
no. 2, pp. 1605–1618, 2016.
[81] E. Basar, M. Wen, R. Mesleh, M. Di Renzo, Y. Xiao, and H. Haas,
“Index modulation techniques for next-generation wireless networks,”
IEEE Access, vol. 5, pp. 16 693–16 746, 2017.
[82] T. Jiang and Y. Shi, “Over-the-air computation via intelligent reflecting
surfaces,” arXiv preprint arXiv:1904.12475, 2019.
[83] S. Zhang and J. Xin, “Minimization of transformed l1 penalty: theory,
difference of convex function algorithm, and robust application in
compressed sensing,” Mathematical Programming, vol. 169, no. 1, pp.
307–336, 2018.
[84] E. Basar, “Transmission through large intelligent surfaces: A new fron-
tier in wireless communications,” in European Conference on Networks
and Communications (EuCNC), 2019.
[85] A. C. Tasolamprou, A. Pitilakis, S. Abadal, O. Tsilipakos, X. Tim-
oneda, H. Taghvaee, M. S. Mirmoosa, F. Liu, C. Liaskos, A. Tsio-
liaridou, N. V. Kantartzis, D. Mannessis, J. Georgiouk, A. Cabellos-
Aparicio, E. Alarcón, A. Pitsillides, I. Akyildiz, S. A. Tretyakov, E. N.
Economouand, M. Kafesaki, and C. M. Soukoulis, “Exploration of
intercell wireless millimeter-wave communication in the landscape of
intelligent metasurfaces,” arXiv preprint arXiv:1907.02395, July 2019.
[86] E. Björnson, L. Sanguinetti, H. Wymeersch, J. Hoydis, and T. L.
Marzetta, “Massive MIMO is a reality—What is next? Five
promising research directions for antenna arrays,” arXiv preprint
arXiv:1902.07678, 2019.