1

Reconfigurable Intelligent Surfaces:

Three Myths and Two Critical Questions
Emil Björnson, Özgecan Özdogan, Erik G. Larsson

Abstract—The search for physical-layer technologies that can towards the user device, wherever it is in the room, the RIS
play a key role in beyond-5G systems has started. One option must be reconfigurable. By using an RIS in this setup, the
is reconfigurable intelligent surfaces (RIS), which can collect signal-to-noise ratio (SNR) can be improved.
wireless signals from a transmitter and passively beamform them
towards the receiver. The technology has exciting prospects and An RIS is a thin surface composed of 𝑁 elements, each
being a reconfigurable scatterer: a small antenna that receives
arXiv:2006.03377v2 [cs.IT] 1 Oct 2020

is quickly gaining traction in the communication community, but

in the current hype we have witnessed how several myths and and re-radiates without amplification, but with a configurable
overstatements are spreading in the literature. time-delay [3]. For narrowband signals, this delay corresponds
In this article, we take a neutral look at the RIS technology. We to a phase-shift. Assuming the phase-shifts are properly ad-
first review the fundamentals and then explain specific features
that can be easily misinterpreted. In particular, we debunk three justed, the 𝑁 scattered waves will add constructively at the
myths: 1) Current network technology can only control the receiver. This principle resembles traditional beamforming:
transmitter and receiver, not the environment in between; 2) A each element has a fixed radiation pattern but the collection of
better asymptotic array gain is achieved than with conventional phase-shifts determines where constructive interference among
beamforming; 3) The pathloss is the same as with anomalous the scattered waves occurs. The color pattern at the RIS in
mirrors.
To inspire further research, we conclude by identifying two Fig. 1 represents the phase-shifts necessary to steer a beam
critical questions that must be answered for RIS to become a towards the receiver. Each element is substantially smaller than
successful technology: 1) What is a convincing use case for RIS?; the wavelength (e.g., a fifth of the wavelength in each direction
2) How can we estimate channels and control an RIS in real time? [5]) so it scatters signals almost uniformly, giving the surface
the ability to form equally strong beams in any direction [6].
The propagation analysis of an RIS essentially entails (i)
I NTRODUCTION finding the Green’s function of the signal source (a sum
of spherical waves if close, or a plane wave if far away),
The electromagnetic waves that carry information in wire- (ii) computing the impinging field at each RIS element, (iii)
less communications interact with objects and surfaces on integrating this field over the surface of each element to find
their way from the transmitter to the receiver. Although the the current density, (iv) computing the radiated field from each
superposition of many propagation paths gives rise to random- element, and (v) applying the superposition principle to find
like fading phenomena, every propagation path has a constant the field at the receiver. Since the elements are small, one can
behavior. However, there exist engineered materials whose approximate the re-radiated field by pretending each element
interactions with electromagnetic waves are not constant but is a point source and then the received signal is a superposition
reconfigurable. These materials are not naturally occurring but of phase-shifted, amplitude-scaled source signals [6].
can be manufactured and deployed to shape the propagation There are many prospective use cases for RIS-aided wire-
environment. The prospects of including such reconfigurable less communications, in addition to improving the SNR as
intelligent surfaces (RIS) as a part of beyond-5G network in Fig. 1. The RIS can also mitigate interference between
architectures are attracting much attention [1], [2]. RIS have users that are spatially multiplexed or limit the signal-leakage
also been called software-controlled metasurfaces [3] and outside the intended coverage area, to mitigate eavesdropping
intelligent reflecting surfaces [4]. [2]–[4]. Support for wireless power transfer, backscattering,
A basic use case of RIS is illustrated in Fig. 1, where a and spatial modulation is also conceivable; most things that
rooftop-mounted base station (BS) is transmitting to an indoor can be implemented using antenna arrays can also be carried
user. Suppose the building materials are such that the direct out by an RIS [5].
path through the wall experiences massive penetration losses, The definition of an RIS is a surface with real-time re-
while the path through the window only experiences minor configurable scattering properties (e.g., amplitude, delay, and
losses. Inside the window, an RIS is deployed to capture signal polarization) that is controlled to improve the communication
energy proportional to its area and re-radiate it in the shape of performance. The concept is often connected with metasur-
a beam towards the receiver. To ensure the beam is focused faces, which are two-dimensional surfaces consisting of arrays
of reconfigurable elements of metamaterial [5]. However,
The paper was supported by ELLIIT and the Swedish Research Council there are other potential ways of implementing RIS [2]. One
(VR). example is using small patch antennas terminated with an
The authors are with the Department of Electrical Engineering
(ISY), Linköping University, SE-58183 Linköping, Sweden adjustable impedance. In any case, the reconfigurability will
{emil.bjornson,ozgecan.ozdogan,erik.g.larsson}@liu.se. likely be limited to a finite set of states per element (with
 2

1) Transmitter

3) RIS with elements

configured to shape a beam
towards the receiver
2) Signal reaching the RIS

4) Beam from the RIS

5) Receiver

Fig. 1. A typical use case of an RIS, where it receives a signal from the transmitter and re-radiates it focused towards the receiver. To focus the beam in the
right direction, the RIS must be configured properly.

given delays and amplitudes) and mutual coupling between Myth 1: Current network technology can only control the
adjacent elements is another limitation. We refer to [5] for a transmitter and receiver, not the environment in between.
recent survey on the implementation aspects for metasurfaces. Many wireless systems indeed consist of a transmitter that
There are decades of research on reflectarrays and array lenses communicates with a receiver without the involvement of other
[7], which are architectures for building transmitters consisting entities. The radio environment is then uncontrollable; the
of a feed antenna that sends the signal via a reconfigurable transmitter and receiver must conform to it by adaptive mod-
surface capable of electronically tunable beamforming. The ulation/coding, beamforming, and power control. However,
key difference is that an RIS is not co-located with the this is a choice made by the network provider because the
transmitter or receiver, but deployed in between to aid the technology for controlling the signal propagation between the
communication. end points has existed all along. The wireless repeater was
invented in 1899 and advanced relaying technology, capable
BASIC F EATURES AND R ELATED M YTHS of adaptively improving the channel between the transmitter
and receiver, has been supported by cellular standards since
We will now describe three fundamental features that the
3G [9]. Hence, the statement above is a myth.
RIS technology possess. Along the way, we will also debunk
three myths that are flourishing in the literature. We will now put the RIS technology into a historical con-
text. The term cooperative communications is broadly used to
refer to network architectures containing entities between the
Feature 1: Creating Controllable Radio Environments transmitter and receiver that enhance the physical channel, by
A key feature of RIS is the ability to alter how wireless exploiting diversity, beamforming, and/or multiplexing gains
signals propagate between the transmitter and receiver. It [9]. These entities are co-optimized with the transmitter and
is a technology for creating controllable/smart/programmable receiver, thus satisfying the definition of controllable radio
radio environments, which are defined as environments that environments. Two main categories are transparent relaying
can customize how signals propagate from the transmitter to and regenerative relaying. In the former category, each relay
the receiver [2]. This feature enables joint optimization of is an entity that receives a signal from the transmitter and
the transmitter/receiver and the controllable entities in the processes it in analog (or digitally) before re-radiating it
environment, using channel state information (CSI). When towards the receiver. Amplify-and-forward is a classic protocol
motivating the novelty of this feature, the following claim has for creating additional signal paths by re-radiating an amplified
been made repeatedly [3], [4], [8]. signal in a way that can be transparent to the receiver. No
 3

In summary, the RIS technology can control/optimize the

14
propagation environment between the transmitter and receiver,
12 just as previous relaying technologies. The unique feature of
RIS is that it reduces the hardware complexity at the price of
10 requiring a larger surface.
8

6 Feature 2: Passive Beamforming

Beamforming appears when delayed copies of the same
4 @
I
Equivalent to one signal are emitted from multiple antennas. This gives rise
2 isotropic antenna to constructive interference at spatial locations where the
copies are received synchronously and destructive interference
0
10 -3 10 -2 10 -1 10 0 10 1 10 2 10 3
elsewhere. If the time-delays at 𝑁 transmit antennas are tuned
to achieve constructive interference at the receiver, it will
receive 𝑁 times more power than if the same total power was
Fig. 2. This figure revisits the setup in Fig. 1 and compares the use of an
RIS with the use of a multi-antenna DF relay deployed at the same place. transmitted from a single antenna. This is the conventional
The direct path is assumed non-existing, while there are line-of-sight paths array gain that shows how the beamformed signal becomes
via the RIS/relay. The transmitter is 300 m from the RIS/relay, while the user more spatially focused as the array size grows.
is 10 m from it. The transmit power is equivalent to 10 W per 20 MHz at the
BS and 0.1 W at the relay. The antenna gain is 10 dBi at the BS and 0 dBi at An RIS is capable of passive beamforming. It receives signal
the relay and user. The glass penetration loss is −20 dB and the noise figure power from the transmitter proportional to its surface area,
is 10 dB. The figure shows the SE achieved by an RIS and a DF relay for which is proportional to the number of elements, 𝑁. When
different surface areas. For practical SEs (below 8 bit/s/Hz), the DF relay is
smaller, while the advantage of the RIS is the lack of power amplifiers and the RIS re-radiates the signal, with time-delays selected to
full-duplex mode. beamform at the receiver, an array gain of 𝑁 is obtained just
as with conventional beamforming. The combination of these
two effects, both being proportional to 𝑁, leads to an SNR
baseband processing is required, only amplification. In regen- at the receiver proportional to 𝑁 2 . This is called the “square
erative relaying, each relay is decoding the received signal law” [4].
and processes it in the digital baseband, before retransmitting Suppose we compare the setup in Fig. 1 with the case when
it in an optimized manner towards the receiver. Decode-and- the RIS is replaced with a transmitter array having the same
forward (DF) is a common example. Classical relays operate in size. The RIS setup will achieve an SNR that grows as 𝑁 2 ,
half-duplex, where reception and retransmission are separated while the SNR in the latter setup only grows as 𝑁. It has
in time, but regenerative full-duplex relays capable of receiving been claimed that these are asymptotic scaling laws [4], which
and transmitting simultaneously are emerging [10]. means that the SNR with the RIS grows unboundedly with 𝑁
The RIS technology is unique in that it fills an empty at the order of 𝑁 2 and eventually becomes larger than with
slot in the relaying taxonomy: it is a transparent relay with the transmitter array. This is incorrect.
a full-duplex protocol [1], thus it affects the propagation in Myth 2: A better asymptotic array gain is achieved than
real-time. The potential advantage over traditional relays is with conventional beamforming.
that large surfaces can be implemented with reduced energy The first issue with this statement is that array gains of the
consumption and cost since the use of printed metamaterial type described above only appear when the surface area (of the
requires no amplifiers, but only power dissipation in the RIS or transmitter array) is small compared to the propagation
hardware controlling the reconfigurability. The drawback is distances. The transmitter/receiver must be in the geometric
the reduced signal range due to the lack of amplification. far-field of the surface so that the pathloss is approximately
Fig. 2 illustrates this in a setup that resembles the one the same to all parts of the surface. Since the surface area
illustrated in Fig. 1, but where the RIS is possibly replaced by a grows with 𝑁, the far-field approximation eventually breaks
multi-antenna half-duplex repetition-coded DF relay (a simple down as 𝑁 increases. Neither linear nor quadratic asymptotic
but suboptimal relaying scheme). Perfect CSI is assumed and power scaling laws can exist since the law of conservation
each RIS element scatters all the incoming signal energy with of energy dictates that we cannot receive more energy than
a perfectly controlled phase. The figure shows the array’s was transmitted. Nevertheless, the SNR achieved with an RIS
surface area required to achieve a particular spectral efficiency actually grows quadratically with the number of elements for
(SE) when using either an RIS or a DF relay. The results are many practically-sized surfaces [11]. Hence, it might seem
frequency-independent but the number of elements that fits possible that a better SNR can be achieved in the RIS setup
into the surface area grows quadratically with the wavelength. when considering large, equal-sized arrays. The second issue
We observe that the DF relay can have a much smaller form with the statement is the premise that the quadratic power
factor than the RIS, except if very high SE is required. The scaling is advantageous. The pathloss from the transmitter to
reason is that the DF relay achieves a much higher SNR but each RIS element is huge in the far-field, thus it is more
it also needs a higher SNR to achieve the same SE since accurate to say that the power loss between the transmitter
it operates in half-duplex, whereas the RIS operates in full- and RIS reduces as 1/𝑁 [11]. The SNR achieved with an
duplex. RIS cannot surpass the SNR achieved when replacing it by an
 4

10
but can be synthesized by an engineered inhomogeneous
10 surface. A property of mirrors is that the receiver observes
the transmitting source as being behind the mirror. One can
analyze the wave propagation as if the transmitter is moved
10 5
6 to the location of the mirror image, as illustrated in Fig. 4.
The “square law” It has been stated that an RIS can generally be viewed as
tapers off an anomalous mirror if it has a width and length larger than
ten wavelengths [8]. If that is the case, the pathloss in Fig. 4
10 0 can be computed based on the sum of the distance from the
transmitter to the RIS and from the RIS to the receiver [8],
which is the distance from the mirror image to the receiver.
These are myths that are summarized as follows.
10 -5 -3
10 10 -2 10 -1 10 0 10 1 10 2 10 3 Myth 3: The pathloss is the same as with anomalous
mirrors.
An ideal mirror reflects a signal with zero beamwidth. If a
Fig. 3. This figure continues the example from Fig. 2 by comparing the end- plane wave is impinging on a finite-sized RIS that is configured
to-end SNR achieved by the RIS and the DF relay. The RIS benefits from the
“square law” by achieving a steeper slope for practical surface areas (below
to focus the signal towards a receiver located in the far-
100 m2 ). Nevertheless, the DF relay is consistently providing a better SNR field, then the radiated field will be strongest in the angular
and both curves converge to a finite number since there are no asymptotic direction of the receiver but it will not be a plane wave. Far-
scaling laws.
field focusing is called beamforming and the beamwidth is
the same as for beamforming from an equal-sized transmitter
equal-sized antenna array transmitting with the same power as array. Hence, the half-power beamwidth of the reflected signal
in the RIS case, but the difference reduces as 1/𝑁. is inversely proportional to the size of the RIS (measured
The reason that the DF relay outperformed the RIS for most in wavelengths) and becomes 6◦ for a surface that is ten
SE values in Fig. 2 is that the RIS suffers from the power wavelengths in each dimension [6].
loss inherent in the “square law”. To demonstrate this, Fig. 3 Mirrors and plane waves are theoretical idealizations that
revisits the example by showing the end-to-end SNR achieved only appear approximately in practice. They can be fairly
with the RIS and DF relay for different surface areas. Since accurate approximations in visible light and are, thus, used
we use logarithmic scales, the quadratic array gain is observed in geometrical optics to analyze imaging. The situation is
from the steeper slope of the RIS curve. However, this curve different in the radio spectrum used for communications.
begins at a much smaller value and when it approaches the DF A surface that our eyes perceive as a mirror might be far
relay curve, the steeper slope has tapered off. Both curves will from mirror-like for radio signals. Since the wavelength is
eventually converge to a finite number [11]. The reason that roughly 100000 times larger in radio spectrum than in visible
the RIS became preferable for very high SEs in Fig. 2 is that light (e.g., comparing green light at 600 THz with a radio
the SNR gap eventually becomes so small that the half-duplex signal at 6 GHz), a surface must be 100000 times larger in
operation of the DF relay becomes the bottleneck. each dimension to identically reflect signals. The transmitter
In summary, an RIS is capable of passively beamforming a must be 100000 times further away if its emitted spherical
signal towards the receiver. Due to the faster-than-linear SNR- waves should be approximated as planar, and the receiver
scaling, physically large surfaces are highly preferable. must be 100000 times further away to perceive the reflected
signals as plane waves. Since mirrors only exist in asymptotic
limits, there are no finite-sized surfaces that always can be
Feature 3: Synthesizing a Different Surface Shape approximated as mirrors. If the RIS is viewed from far enough
The RIS can not only form a beam, it can synthesize the away, its radiated field will have a beamwidth that is inversely
scattering behavior of an arbitrarily-shaped surface of the same proportional to its size.
size. For example, it can create a superposition of multiple Even if we limit the scope to setups where conventional
beams or act as a diffuse scatterer [5]. mirrors approximately exist, the statement above remains a
A common example is to synthesize an anomalous mir- myth since the pathloss achieved by an equal-sized RIS is
ror/reflector. A mirror is a surface that reflects an impinging widely different. An RIS can both affect the direction and
plane wave as an outgoing plane wave, also known as specular shape of the reflected signal [11], as illustrated by the red
reflection. A conventional mirror satisfies the law of reflection: ribbons in Fig. 4 where the signal is focused at the receiver.
the angles of the impinging and reflected waves to the surface For this reason, the SNR achieved by the RIS is proportional to
normal are the same but on opposite sides, as illustrated by the 𝑁 2 and is inversely proportional to the product of the squared
blue ribbons in Fig. 4. An anomalous mirror reflects impinging distances to the RIS [6], rather than inversely proportional to
plane waves as outgoing plane waves with a different “unnat- the squared sum of the distances as with a mirror.
ural” angle to the surface normal [2]. A conventional mirror To explain the fundamental differences, Fig. 5 continues
is an infinitely large homogeneous surface and approximations the example from Figs. 2 and 3 by showing how the end-to-
thereof appear naturally (e.g., a metal plate or water surface). end pathloss depends on how far the receiver is from the RIS
In contrast, an anomalous mirror is not naturally appearing (the distance between the transmitter and RIS is as before).
 5

Reradiated plane wave Infinite-sized mirror

Receiver ...
Normal
Mirror image
of transmitter
Reradiated Impinging
signal beam plane wave

...
Impinging plane wave

Finite-sized ... Transmitter

RIS

Fig. 4. A mirror reflects an impinging plane wave as a plane wave in an angular direction determined by the law of reflection, so the receiver perceives
the transmitter as being located at the mirror image location. An RIS can both configure the angle of the reflected beam and its shape, thus it should not be
interpreted as an anomalous mirror. The figure illustrates how the RIS focuses the signal at the receiver to maximize the SNR.

a mirror-mimicking RIS that is configured to delay the signals

-85
as a cutout from an infinitely large anomalous mirror would do.
-90 This curve is close to the optimized RIS when the receiver is
far from the surface, while it begins to oscillate in the vicinity
-95 of the mirror approximation at shorter distances. This indicates
-100 that the mirror analogy can be used for identifying suitable
Only point where the mirror time-delays when the transmitter and receiver are far away.
-105 approximation is accurate This is the reason why (approximate) anomalous reflection
is a canonical use case in the electromagnetic literature [5],
-110 ?
where pathloss modeling is not considered.
-115 In summary, the pathloss achieved with an RIS does not
coincide with that of an anomalous mirror. When the receiver
-120
0 10 20 30 40 50 is far from the surface, it is too small to behave like a mirror.
When the receiver is near the surface, the RIS can approximate
the mirror behavior but it would be suboptimal; a mirror
Fig. 5. This figure revisits the setup from the previous figures and considers beamforms to points infinitely far away, while a RIS can
an RIS that is 2 × 2 m, which represents 20 × 20 wavelengths at a 3 GHz
frequency and 200 × 200 wavelengths at a 30 GHz frequency. The figure
focus on the actual receiver location. One way to describe the
shows the end-to-end pathloss as a function of the distance between the RIS capabilities of an RIS is as a parabolic reflector with curvature
and the receiver. An optimally configured RIS is compared with an RIS that and direction that can be electronically steered, but that is
is configured to mimic a mirror and the pathloss obtained if it was an ideal
mirror. It is clear that an RIS can generally not been interpreted as a mirror.
also a simplification since an RIS is capable of mimicking the
scattering of arbitrarily-shaped objects having the same size.

C RITICAL QUESTION 1: W HAT IS A CONVINCING USE CASE

The solid curve is for an RIS that is optimized to achieve the
FOR RIS?
highest SNR, while the dashed curve represents an anomalous
mirror. We notice that a mirror is a poor approximation of an An immense amount of time and resources are required to
RIS at most distances. When the receiver is far from the RIS, bring a new technology concept, such as RIS, from theory
the pathloss is worse than with a mirror since the RIS is too to practice. Very convincing benefits compared to existing
small to emit approximately plane waves. When the receiver technologies must be established to motivate such an invest-
is close to the RIS, the pathloss is instead much better than ment; we essentially need to demonstrate 10 times improve-
with a mirror. This is like when you look into a large mirror ments with respect to a practically important performance
and your reflection only appears in a small part of it; the rest metric, not just 20% gains that might disappear in an imper-
of the mirror is not needed. A well-configured RIS makes use fect implementation. Massive MIMO (multiple-input multiple-
of the entire surface by focusing the signal at the receiver as output) and mmWave communications passed this test in the
illustrated in Fig. 4. The dash-dotted curve in Fig. 5 represents 5G development since the former can increase the number
 6

simultaneously served users by ten times while the latter can RIS
increase the data rate per user by ten times using much wider
bandwidths. Several other “5G-branded” technologies failed RIS controller
the test because the gains were too limited.
RIS technology has many technical features beyond current
mainstream technology [2], [5]. However, to motivate the 3) Feedback
practical development of RIS technology, the critical question of preferred
is: what is a convincing use case? The question is open; RIS configuration
is a hammer looking for a nail. There is no shortage of visions
on what an RIS can be used for (some ideas were listed
in the introduction) but will it excel at anything? Coverage
extension is one option but Fig. 2 showed that conventional
half-duplex relaying is a competitive solution, and full-duplex
2) Switching between
regenerative relays are emerging [10]. Since each RIS element different configurations
must be identically configured over the entire frequency band, 1) Repeated pilot
transmission
the RIS technology has a further competitive disadvantage
over wideband channels. Improved spatial multiplexing and
interference mitigation is another potential use case, but then it Fig. 6. One approach to configure the RIS is to transmit pilots that the RIS
needs to beat Cell-free Massive MIMO, which is the emerging scatters using different configurations. The receiver feeds back a preferred
deployment of distributed jointly-operating antennas. Perhaps configuration to the RIS.
it is in terahertz bands, where the implementation of coherent
transceivers is truly challenging and the sparse channels make
additional propagation paths useful even if they are weak, that Another approach is to alter the passive nature of the RIS
the RIS technology will be most beneficial. These are just by having a few elements with receiver chains [13], which
speculations since there is no hard evidence yet. enables sensing and channel estimation directly at the RIS.
The ability to extrapolate a few measurements to estimate
C RITICAL QUESTION 2: H OW CAN WE ESTIMATE the entire wideband channel requires spatially sparse channels
CHANNELS AND CONTROL AN RIS IN REAL TIME ? with a known parametrization. This might be reasonable in
The envisioned use cases of RIS critically depend on a mmWave or terahertz bands but further work on channel and
proper configuration of the elements based on CSI. There are hardware modeling is required. The sparseness can also make
two reasons why channel acquisition is particularly challeng- the channels flat over rather wide bandwidths. Learning-based
ing with RIS. Firstly, unlike conventional transceiver archi- and sparsity-based estimation algorithms were considered in
tectures, an RIS is not inherently equipped with transceiver [13], [14]. Even if the RIS has sensing capabilities, a control
chains. It lacks sensing capabilities but simply “reflects” the loop is needed to jointly select the RIS configuration and the
impinging signals. Therefore, conventional channel estimation beamforming at the transmitter/receiver.
methods cannot be utilized. Secondly, introducing an RIS Estimation algorithms can leverage special channel char-
into an existing setup will increase the number of channel acteristics to reduce the pilot overhead. For instance, the
coefficients proportionally to the number of elements, 𝑁. As channel between the BS and RIS is semi-static and common
shown earlier, a large 𝑁 is needed for RIS to be competitive, for all users, which makes the end-to-end channels correlated
thus the estimation overhead might be huge. A key question is: between users. An estimation algorithm exploiting this correla-
can an RIS be real-time reconfigured to manage user mobility? tion was proposed in [15]. The BS-to-RIS channel can contain
The literature contains initial approaches to tackle the prob- many coefficients if the BS has many antennas but since
lem. One approach is to transmit a pilot sequence repeatedly this channel is semi-static, it can be estimated less frequently
and measure the received signal when using different RIS than the RIS-to-user channel, which typically contains fewer
configurations. For example, the elements can be turned on/off coefficients since users have fewer antennas.
according to a pattern or the array geometry can be used There is no doubt that RIS can be used for fixed communi-
to sweep through changes of the main reflection angle. At cation links, but mobile operation requires real-time channel
least 𝑁 reconfigurations must be tested in different time slots estimation and reconfiguration, even in indoor use cases. A
to excite all the channel dimensions. Only a concatenation few millimeters of movement will change the channels in
of the channels to/from the RIS are observed and mutual mmWave bands and above. It remains to be demonstrated if
coupling between RIS elements complicates the estimation. any estimation protocol can enable real-time reconfigurability
This approach is illustrated in Fig. 6 and requires a wireless and under what mobility conditions. Since the array is passive,
control loop between the receiver and the RIS controller circuit the RIS technology is potentially more energy-efficient than
with a capacity proportional to 𝑁. Even when CSI has been alternative technologies [8] but this remains to be demon-
acquired, it is computationally complex to select appropriate strated quantitatively. The RIS will require a power source
time-delays, particularly in wideband channels [12]. To reduce for reconfigurability and wireless control channels. It is likely
complexity, adjacent RIS elements can be grouped to have the that the control interface will consume most of the power at
same configuration [12], at the cost of a performance loss. the RIS, so one cannot predict the total power consumption
 7

until the channel estimation and reconfigurability have been and Massive MIMO Networks: Spectral, Energy, and Hard-
solved and validated. ware Efficiency (2017). He received the 2018 IEEE Marconi
Prize Paper Award in Wireless Communications and the 2019
S UMMARY IEEE ComSoc Fred W. Ellersick Prize.
Özgecan Özdogan is a Ph.D. student at Linköping Uni-
An RIS is a full-duplex transparent relay that synthesizes
versity, Sweden. She received her M.Sc. degree from Izmir
the scattering behavior of an arbitrarily shaped object. Since
Institute of Technology, Turkey, in 2017.
the RIS is not amplifying the signal, a larger surface area is
Erik G. Larsson is Professor at Linköping University,
required to achieve a given SNR than using conventional relays
Sweden, and Fellow of the IEEE. He co-authored Fundamen-
or multi-antenna transceivers. RIS-aided communication is an
tals of Massive MIMO (Cambridge University Press, 2016).
emerging research topic where the main open problems are to
He received, among others, the 2015 IEEE ComSoc Stephen
identify convincing use cases and designing practical protocols
O. Rice Prize in Communications Theory, the 2017 IEEE
for reconfigurability.
ComSoc Leonard G. Abraham Prize, the 2018 IEEE ComSoc
Best Tutorial Paper Award, and the 2019 IEEE ComSoc Fred
R EFERENCES
W. Ellersick Prize.
[1] C. Huang, A. Zappone, G. C. Alexandropoulos, M. Debbah, and
C. Yuen, “Reconfigurable intelligent surfaces for energy efficiency in
wireless communication,” IEEE Trans. Wireless Commun., vol. 18, no. 8,
pp. 4157–4170, 2019.
[2] M. D. Renzo, A. Zappone, M. Debbah, M.-S. Alouini, C. Yuen,
J. de Rosny, and S. Tretyakov, “Smart radio environments empowered
by reconfigurable intelligent surfaces: How it works, state of research,
and road ahead,” arXiv, abs/2004.09352v1, 2020.
[3] C. Liaskos, S. Nie, A. Tsioliaridou, A. Pitsillides, S. Ioannidis, and
I. Akyildiz, “A new wireless communication paradigm through software-
controlled metasurfaces,” IEEE Commun. Mag., vol. 56, no. 9, pp. 162–
169, 2018.
[4] Q. Wu and R. Zhang, “Towards smart and reconfigurable environment:
Intelligent reflecting surface aided wireless network,” IEEE Commun.
Mag., vol. 58, no. 1, pp. 106–112, 2020.
[5] O. Tsilipakos et al., “Toward intelligent metasurfaces: The progress from
globally tunable metasurfaces to software-defined metasurfaces with an
embedded network of controllers,” Advanced Optical Materials, no.
2000783, 2020.
[6] Ö. Özdogan, E. Björnson, and E. G. Larsson, “Intelligent reflecting
surfaces: Physics, propagation, and pathloss modeling,” IEEE Wireless
Commun. Lett., vol. 9, no. 5, pp. 581–585, 2020.
[7] S. V. Hum and J. Perruisseau-Carrier, “Reconfigurable reflectarrays and
array lenses for dynamic antenna beam control: A review,” IEEE Trans.
Antennas Propag., vol. 62, no. 1, pp. 183–198, 2014.
[8] E. Basar, M. Di Renzo, J. De Rosny, M. Debbah, M. Alouini, and
R. Zhang, “Wireless communications through reconfigurable intelligent
surfaces,” IEEE Access, vol. 7, pp. 116 753–116 773, 2019.
[9] M. Dohler and Y. Li, Cooperative Communications: Hardware, Channel
and PHY. John Wiley & Sons, 2010.
[10] M. Heino, D. Korpi, T. Huusari, E. Antonio-Rodriguez, S. Venkatasub-
ramanian, T. Riihonen, L. Anttila, C. Icheln, K. Haneda, R. Wichman,
and M. Valkama, “Recent advances in antenna design and interference
cancellation algorithms for in-band full duplex relays,” IEEE Commun.
Mag., vol. 53, no. 5, pp. 91–101, 2015.
[11] E. Björnson and L. Sanguinetti, “Power scaling laws and near-field
behaviors of massive MIMO and intelligent reflecting surfaces,” IEEE
Open J. Commun. Soc., 2020.
[12] B. Zheng and R. Zhang, “Intelligent reflecting surface-enhanced OFDM:
Channel estimation and reflection optimization,” IEEE Wireless Com-
mun. Lett., vol. 9, no. 4, pp. 518–522, 2020.
[13] A. Taha, M. Alrabeiah, and A. Alkhateeb, “Deep learning for large
intelligent surfaces in millimeter wave and massive MIMO systems,” in
IEEE GLOBECOM, 2019, pp. 1–6.
[14] G. C. Alexandropoulos and E. Vlachos, “A hardware architecture for
reconfigurable intelligent surfaces with minimal active elements for
explicit channel estimation,” in IEEE ICASSP, 2020, pp. 9175–9179.
[15] Z. Wang, L. Liu, and S. Cui, “Channel estimation for intelligent reflect-
ing surface assisted multiuser communications: Framework, algorithms,
and analysis,” IEEE Trans. Wireless Commun., 2020, to appear.

B IOGRAPHIES
Emil Björnson is Associate Professor at Linköping Uni-
versity, Sweden. He has co-authored the textbooks Optimal
Resource Allocation in Coordinated Multi-Cell Systems (2013)