Trinh Van Chien, Hien Quoc Ngo, Symeon Chatzinotas,

and Björn Ottersten

Reconfigurable Intelligent Surface-Assisted Massive MIMO

Favorable propagation, channel hardening, and rank deficiency

M
assive multiple-input multiple- ware impairments by exploiting different will be exploited. Achieving such an
output (MIMO) and reconfigu- channels to different users [3]. Nonethe- understanding will directly enable us
rable intelligent surface (RIS) are less, many scenarios limit channel capac- to design a robust and energy-efficient
two promising technologies for 5G-and- ity despite deploying massive antenna RIS-assisted massive MIMO system.
beyond wireless networks, capable of arrays, especially under harsh conditions
providing large array gain and multiuser such as large blockages, which result in Prerequisites
spatial multiplexing. Without requiring ill-conditioned channels [4]. Readers require a basic knowledge of
additional frequency bands, those tech- To improve the performance of mas- random variables, linear algebra, signals
nologies offer significant improvements sive MIMO under harsh propagation con- and systems, RISs, and massive MIMO.
in both spectral and energy efficiency by ditions, the use of an RIS implemented
simultaneously serving many users. The through metasurfaces is very promising. Problem statement
performance analysis of an RIS-assisted An RIS is an emerging technology that We consider an RIS-aided communica-
massive MIMO system as a function of smartly controls the propagation envi- tion system where a BS is equipped with
channel statistics relies heavily on fun- ronments by a planar array with many M antennas to serve K single-antenna
damental properties, including favor- engineered scattering elements to form users, as shown in Figure 1. To enhance
able propagation, channel hardening, electromagnetic waves in the desired system performance, an RIS with N engi-
and rank deficiency. The coexistence of structure [5], [6]. Each engineered scat- neered scattering elements is deployed
both direct and indirect links results in tering element induces a phase shift on in the coverage area between the BS
aggregated channels, whose properties the incident signals and reflects them and the users. The phase-shift matrix
are the main concerns of this “Lecture passively without the requirement of is denoted as U = diag (e ji1, f, e ji N)
Notes” article. For practical systems radio-frequency chains. Thus, with an with - r # i n # r, 6n, representing the
with a finite number of antennas and RIS, high power consumption and expen- phase shift of the nth passive-engineered
engineered scattering elements of the sive hardware can be avoided. In the sce- scattering element, which is controllable.
RIS, we evaluate the corresponding narios of large-distance/heavy blockages For the indirect link from the BS to a user
deterministic metrics, with Rayleigh between the transmitter and the receiver, through the RIS (see Figure 1), let us denote
fading channels as a typical example. an RIS has been shown to enhance the H ! C M # N as the channel matrix between
received signal strength thanks to a the BS and the RIS, while g k ! C N repre-
Relevance phase-shift design that leads to a con- sents the channel vector between the RIS
Antenna arrays and propagation envi- structive combination of multiple arriv- and user k. For the direct link between the
ronments are key factors fundamentally ing waves at the receiver, and therefore BS and user k (see Figure 1), the propaga-
determining the performance of wireless yields better system performance tion channel is u k ! C M . We now con-
communication systems. Massive MIMO than the absence of an RIS [7]. sider the uplink transmission where
communications have demonstrated the As RIS-assisted massive MIMO all users simultaneously transmit signals
possibilities of increasing the communi- is a very new topic, there is no stan- to the BS. Note that, the same meth-
cation throughput by coherently process- dard reference presenting and provid- odology should be straightforwardly
ing many antenna signals at each base ing the effective properties when many extended to the down­link transmission,
station (BS) compared to the number of antennas and phase-shift elements are especially when the time-division duplex-
served users [1]. Each antenna element of installed in the system. This “Lecture ing (TDD) protocol is exploited. Let s k
a massive array can contribute an extra Notes” article fills this gap by present- with E {; s k ;2} = 1 be the symbol transmit-
degree of freedom to spatial processing, ing the channel property aspects of ted by user k. Then, the received signal at
which allows the system to obtain array RIS-assisted massive MIMO systems the BS, denoted by y ! C M, is
gain and suppress mutual interference when the numbers of BS antennas and K K
[2]. Massive MIMO thus offers unprec- engineered scattering elements grow y= p / uk sk + p HU / g k s k + w
k= 1 k= 1
edented improvements even under hard- large. Three important properties,
K
Digital Object Identifier 10.1109/MSP.2021.3128352
namely, favorable propagation, chan- = p / z k s k + w, (1)
Date of current version: 27 April 2022 nel hardening, and rank deficiency,  k= 1

IEEE SIGNAL
Authorized licensed use limited to: University of Leeds. Downloaded PROCESSING MAGAZINE
on December | 10:32:03
05,2024 at | from IEEE Xplore. Restrictions apply.
May 2022 UTC 97
 where p is the transmit power allocated bers of BS a­ ntennas and e­ ngineered Definition 1
to each data symbol, w~CN (0, v 2 I M) scattering elements grow large, e.g., The aggregated channels offer the
is additive noise, and M, N " 3. favorable propagation property if the
inner product of two arbitrary channel
z k = u k + HUg k, (2) Solution vectors z k and z l, k ! l, satisfies
In general, aggregated channels do not
which is the aggregated channel of follow the same distributions as conven- z kH z l = 0. (3)
user k, comprising both the direct and tional massive MIMO channels due to
indirect links. Channels need to be the weighted product from the indirect Favorable propagation is very impor-
acquired for signal processing. One link in (2). Note that, with the absence tant. If the channels are favorable, then
conventional approach is that both of the RIS, aggregated channels reduce the signals transmitted from the K users
the BS-RIS and RIS-user channels to conventional massive MIMO chan- will belong to K orthogonal spaces. As a
are estimated separately. This is very nels, i.e., z k = u k, 6k. Consequently, result, the BS can decode the signal sent
challenging because the RIS lacks the results presented in this “Lecture by user k without interuser interference
a digital processing unit. Another Notes” article are a general version of by simply projecting the received signal
m e t h o d u s e d for conventional mas- (1) onto z kH as
channel acquisition is With TDD, the importance sive MI MO com-
when the BS needs of channel hardening is munications thanks st k = z kH y = p z kH z k s k
only to estimate the even more significant to the presence of K
aggregated channels. in the downlink data the RIS. In the fol- + p / z kH z l s l + z kH w
l = 1, l ! k
This technique yields lowing, the favorable 
the great benefit of
transmission. propagation, channel =
2
p z k s k + z kH w. (4)
reducing the system cost with the same hardening, and rank-deficiency proper-
pilot training overhead as communi- ties will be discussed. The aforementioned projection cor-
cation systems without the presence responds to the linear maximum-ratio
of the RIS. One example with fine, Favorable propagation combing technique. This implies that
step-by-step details on an aggregated This section presents the favorable the favorable propagation property
channel estimation can be found in [4]. propagation property in RIS-assisted yields optimal signal detection per-
Therefore, aggregated channel statis- massive MIMO systems. Specifical- formance with only a simple linear
tics are of particular interest for per- ly, it describes orthogonality among processing. Nonetheless, this property
formance evaluation purposes. The the aggregated channel vectors from will rarely hold in practice. The asymp-
problem at hand is to investigate the the BS to the K users, as shown in the totically favorable propagation given in
channel properties when the num- following definition. the “Definition 2” section is, therefore,
more interested.

Definition 2
As M, N " 3, the aggregated channels
Controller offer the asymptotically favorable prop-
RIS agation property if

z kH z l
" 0, (5)
E $ zk .E $ zl .
2 2
H

holds for any k ! l.

In contrast to the strict requirement
in the “Definition 1” section, in which
gk

the inner product z kH z l must be exactly

uk zero, the condition (5) gives practical
tolerability in the sense that z kH z l does
not need to be zero for finite values.
For real systems, the BS and the RIS
are equipped with a large (but finite)
number of BS antennas/scattering
Direct Link Indirect Link
elements. Conditioned on an aggre-
gated channel model that offers the
FIGURE 1. An RIS-assisted massive MIMO system. asymptotically favorable propagation,

98
Authorized IEEE SIGNAL
licensed use limited to: University of Leeds. Downloaded 05,2024 at| 10:32:03
PROCESSING MAGAZINE
on December | from IEEE Xplore. Restrictions apply.
May 2022 UTC
 we now evaluate how this property zk
2 Otherwise, the channel model does
" 1, (7)
E $ zk .
behaves with a finite-value M and N 2 not harden. Consequently, (9) can be
by the deterministic-favorable chan- utilized to determine whether a chan-
nel propagation. with the almost-sure convergence. nel model offers the channel-hardening
From (7), the effective channel gain property, as shown in the “Definition 5”
Definition 3 < z k < 2 can be replaced by its mean section.
For the given channel vectors z k and z l, value by virtue of many antennas at
the deterministic-favorable propagation the BS and many scattering elements at Definition 5
metric is defined as the RIS. If the channels harden, in the For each aggregated channel, the deter-
uplink, the BS can replace the instan- ministic channel-hardening metric is
Var {z kH z l} taneous channel gain by its mean value defined as
FPkl = . (6)
E $ zk .E $ zl .
2 2
for signal detection. This significantly
Var $ z k .
2
simplifies the signal processing as well
CH k = . (10)
` E $ z k 2 .j
2
By evaluating (6), we enable measure- as resource allocation designs at the
ment of how close to the favorable prop- BS because all designs can now be
agation the channel is when the number done on the large-scale fading time The deterministic channel-hard-
of antennas is finite. In other words, (6) scale. With TDD, the importance of ening metric in (10) can be com-
represents the speed of convergence in channel hardening puted for practical
(5). It is worth emphasizing that this is even more signifi- propagation chan-
deterministic metric is independent cant in the downlink A rich scattering nel models. In some
of small-scale fading coefficients and data transmission. propagation environment par ticular scena r-
therefore, by knowing the large-scale In the downlin k, may not often offer a i o s w e c a n eve n
fading coefficients and phase shifts, we thanks to the chan- channel profile with the derive this metric
can evaluate the favorable propagation nel-hardening prop- in the closed-form
rank deficiency.
property of the aggregated channels. erty, each user can expression.
treat the mean value
The first lesson learned of the effective channel gain as the true The second lesson learned
The favorable propagation means one to detect the desired signal. Thus, The channel-hardening property makes
that the aggregated channel vectors no downlink pilot overhead is required each effective channel gain approach its
from the BS to different users are pair- for the downlink channel estimation. mean value with a high probability. The
wisely orthogonal. Under the favorable Since z k is a random vector, it is deterministic metric (10) measures the
propagation, the optimal performance nontrivial to exploit “Definition 4” to hardening level of the channel vectors
can be achieved with simple linear pro- prove the channel-hardening property with a finite number of BS antennas and
cessing techniques. The deterministic of the channels. To seek a more trac- engineered scattering elements.
metric (6) measures how close to the table metric, we employ the Chebyshev
favorable propagation the aggregated inequality [8] such that Rank deficiency
channels offer with a finite number of This section presents the rank defi-
2 2

Pr * - 1 # e4
BS antennas and engineered scatter- zk ciency, which has been a fundamental
ing elements. E $ zk . issue in many wireless communication
2

aspects. A rich scattering propagation

= 1 - Pr * - 1 $ e4
2 2
Channel hardening zk environment may not often offer a chan-
This section represents the channel- E $ zk . nel profile with the rank deficiency.
2

Var $ z k .
hardening property, which is a phe- 2 Nonetheless, this may appear in a poor
nomenon where the norm square of the $ 1 - , (8) scattering environment, such as mil-
e ` E $ z k .j
 2 2
aggregated channel vectors from the 2 limeter-wave communications with a
BS to the users does not fluctuate much limited number of reflections and prop-
(even the small-scale fading channels for any e 2 0. Note that if x is a random agation paths. For RIS-assisted massive
randomly change). Mathematically, an variable with the mean xr and variance MIMO systems, we provide the rank-
2
aggregated channel offers channel hard- v x, the Chebyshev inequality gives deficiency definition of an aggregated
ening if the condition in the “Definition Pr (; x - xr ; $ e) # v 2x /e 2 . Observing the channel, as in “Definition 6.”
4” section holds. last inequality of (8), one can have the
channel hardening if Definition 6
Definition 4 The rank deficiency indicates fewer chan-
As M, N " 3, an aggregated channel Var $ z k . nel degrees of freedom than the upper
2

model offers the channel-hardening " 0. (9) limit imposed by the number of BS anten-
property if ` E $ z k .j
2 2
nas and engineered scattering ­elements.

IEEE SIGNAL
Authorized licensed use limited to: University of Leeds. Downloaded PROCESSING MAGAZINE
on December | 10:32:03
05,2024 at | from IEEE Xplore. Restrictions apply.
May 2022 UTC 99
 Mathematically, we can measure the product property. The observation such that the (m, n) -th element is
number of channel degrees of freedom from (11) is that an RIS can possi- given as
for an aggregated channel as rank (R t k), bly improve the received signal power,
where R t k = E {z k z kH} ! C M # M is the on average, as E " < z k < 2 , $ E " < u k < 2 ,,
covariance matrix of which the eq­­uality [R o] mn =
So v 2o
bo
the aggregated chan- The rank deficiency can holds when the RIS
So s = 1
/ e j (m - n) sin (
r } s)
e- 2
(r (m - n) cos (} s)) 2,

nel of user k. The rank is not active or is

deficiency implies
be explicitly observed by without contribu-
(14)
rank (R t k) 1 M. analyzing the eigenvalues tions. As the cova-
The rank of a of each covariance matrix. riance matrices are where o ! {s, k} and b o is the corre-
covariance matrix is positive semidefinite, sponding large-scale fading coefficient,
equal to the number of nonzero eigen- one possibly designs the phase shifts S o is the number of scattering clusters
values, so projecting the covariance to obtain around the BS, and } s is the nominal
matrices utilizing eigendecomposition angle of arrival (AoA) for cluster s.
and analyzing the eigenvalues is one rank (E {z k z kH}) $ rank (E {u k u kH}), The multipath components of a cluster
way to observe the rank deficiency. (12) is assumed to have AoAs following a
Each eigenvalue stands for the aver- Gaussian distribution that is distributed
age squared amplitude in a subset of which the equality holds, for exam- around the nominal AoA with the angu-
angular directions, the rank-deficiency ple, if the covariance matrix is full lar standard deviation v o . By assuming
results in zero eigenvalues or almost-zero rank, which happens in rich scatter- that the RIS is fabricated in a rectangu-
eigenvalues (compressible eigenvalues). ing environments. In contrast, the lar surface of size N H # N V where N H
Consequently, we can project a given RIS is beneficial to a propagation and N V are the numbers of elements in
covariance matrix onto the eigendo- environment suffering from the rank- each column and row, respectively, the
main with the eigendecomposition as deficiency issue. covariance matrices R si and R ik are
Rt k = Vk R k V kH, where Vk ! C M # M is an given as
unitary matrix including the eigenvec- The third lesson learned
tors of R t k, and the diagonal matrix The rank deficiency can be explicitly R si = R, R ik = b ik R, (15)
R k = diag ([d 1, f, d M] T ) ! C M # M i n­­ observed by analyzing the eigenvalues
cludes the eigenvalues d 1, f, d M on of each covariance matrix. Moreover, where b ik is the la rge-scale fad-
the diagonal. The rank deficiency the RIS can be potentially utilized to i n g c o ef f ic ient a sso c ia t e d w it h
appears if a covariance matrix has at compensate for the rank deficiency cascaded channel g k . The spatial
least one zero eigenvalue that reduces appearing in massive MIMO commu- correlation matrix R ! C N # N has the
the system capacity. nications. (m, n) - th element defined as [R] mn =
We now demonstrate the rank- d H d V sinc (2|| v m - v n ||/m), where v x =
improved benefits of installing an RIS Example: Rayleigh fading channels [0, mod (x -1, N H) d H, 6(x -1) /N H@ d V ] T
in the vicinity of a massive MIMO sys- This section considers a spatially with x ! {m, n} [9], [10]; m is the
tem by first observing from (4) that the correlated Rayleigh model, which wavelength of a plane wave; mod (·)
effective power from user k scales up is aligned with a BS radiating iso- and 6 · @ are the modulus operation
with < z k < 2. One can use the channel- tropically in an RIS. All the chan- and floor function, respectively; each
hardening property to approximate the nels between the BS and user k are phase-shift element is of the size
effective gain < z k < 2 by the mean value defined as d H # d V , where d V is the vertical
E " < z k < 2 , . We further process the mean height and d H is the horizontal width.
value as H = R 1s /2 HR u 1si/2, g k = R 1ik/2 gu k, As a special case, the spatially

and u k = R k1/2 uu k, (13) uncorrelated fading channel model gives
E $ z k . = E $ u k . + E " u kH HUg k ,
2 2
R o = b o I M and R = d H d V I N , whose
+ E " g kH U H H H u k , + E $ HUg k .
2 where H u ! C M # N , gu k ! C N , and uu k ! C M instantaneous channels are defined as
represent the small-scale fading matrix/
$ E $ u k . + E $ HUg k .
(a) 2 2
vectors whose elements are independent u , g k = b ik d H d V gu k,
H = bs dH dV H

= tr ^E " u k u k ,h
(b) and identically distributed CN (0, 1); u = b k u k,
u (16)
H
 k
H H H
R s ! C M # M, R si ! C N # N , R ik ! C N # N ,
 + tr (E{HUg k g k U H }), (11) and R k ! C M # M represent the cova- which ignores the spatial correlation
riance matrices. With a limited scat- among elements that is, therefore, less
where (a) is found by designing terer number at the BS, we can model practical. Even though the original
the phase-shift matrix U to obtain the corresponding covariance matri- channels follow circularly symmetric
E {u kH HUg k} + E {g kH U H H H u k} $ 0, ces by, for instance, the approximate Gaussian distributions, each aggregated
and (b) is obtained by the trace of a Gaussian local scattering model [3] channel follows a non-Gaussian due to

100
Authorized IEEE SIGNAL
licensed use limited to: University of Leeds. Downloaded 05,2024 at| 10:32:03
PROCESSING MAGAZINE
on December | from IEEE Xplore. Restrictions apply.
May 2022 UTC
 the weighted product of multiple Gauss- Meanwhile, one uses [11, Lemma 8] to the Hermitian matrices U H R si U and
ian random variables. For a given phase- compute the next three expectations in R ik sorted in descending order. We
shift matrix, we still can compute the the last equality of (20) as observe that
channel statistics as in “Lemma 1.”
E $ b . = E $ c . = tr (R s R k) tr (H ck # e / m n (U H R si U) oe / m n (R ik) o
2 2 (a) N N
u k),
Lemma 1
E " ad , = tr (R s) tr (R k) tr (H k)
u n=1 n=1
If the channel model (13) is utilized,  (=
b) (c)
 u k) .
= M 2 b s b k tr (H (22) tr (U H R si U) tr (R ik) = N 2 d 2H d 2V b ik,
the second and fourth moments of each (26)
aggregated channel vector in (2), respec-
tively, are given as We tackle the last expectation in the where (a) is obtained because the
last equality of (20) as eigenvalues are nonnegative, (b) is
E $ z k . = Mb k + Mb s tr (H
2 u k), (17) due to the trace and eigenvalues rela-
E {; d ;2} = E {< HUg k <4} tionship, and (c) is acquired by the
= E " g kH U H R ik1/2 H u ik1/2 Ug k 2 ,
E $ z k . = (Mb k + Mb s tr (H
4
u H R s HR covariance matrix structure in (15)
u k)) 2
 = E " R ik1/2 Ug k 4 t kH R s t k 2 ,, (23) and because the phase matrix is uni-
u k)
+ 2tr (R s R k) tr (H tary. From (25) and (26), we observe
2
u k) tr (R s )
+ tr (H 2
 u 1ik/2 Ug k / < R 1ik/2 Ug k <. As
where t k = HR that E {< z k <2} scales up with an array
u 2k ) (M 2 b s2 + tr (R s2))
+ tr (H t k + CN (0, I M) and noting that t k and gain provided by the RIS in the order
g k are independent circularly sym- from N to N 2 depending upon the
+ tr (R 2k ), (18)
 metric Gaussian vectors, we recast phase-shift design. Specifically, the
(23) as right hand of (25) is upper bounded by
u k = U H R si UR ik .
where U Mb k + M N 2 d 2H d V2 b s b ik, whose scaling
E {; d ;2} = E {< R ik1/2 Ug k <4} law is possibly obtained by the opti-
Proof # E {; t kH R s t k ;2} mal phase-shift design [5]. For the
The second moment of the aggregated u k) ;2 + tr (H u k2)) case of uncorrelated Rayleigh fading
= (; tr (H
channel of user k is first processed by  channels, the channel statistics are
# (; tr (R s) ;2 + tr (R s2))
the independence of the direct and indi- given in “Corollary 1.”
u k) ;2 + tr (H
= (; tr (H u k2))
rect links as
# (M 2 b s2 + tr (R s2)) . (24) Corollary 1
E$ z k . = E $ u k . + E $ HUg k . If the channel model (16) is utilized,
2 2 2

Plugging (21), (22), and (24) into (20), the second and fourth moments of each
= tr (R k) + E" g kH U H H H HUg k , we obtain the result as shown in the aggregated channel vector in (2), respec-
= Mb k + Mb s tr (H u k), (19)
 lemma. tively, are given as
Note that the even moments in
where the second moment of the indirect “Lemma 1” are multivariate func- E {< z k <2} = Mb k + M N b s b ik d 2H d 2V ,
link is computed by using the channel tions of phase shifts and covariance (27)
model (13) together with [11, Lemma 8]. matrices. Consequently, this allows E {< z k <4} = M 2 b 2k + 4M Nb k p k
Next, the fourth moment of the aggre- us to further investigate other utility
+ (M 2 + M ) (N 2 + N ) p k2
gated channel of user k is computed as metrics, such as spectral efficiency
 + Mb k2 (28)
and outage probability in closed form,
E $ z k . = E $ u k + HUg k . which is independent of small-scale
4 4

fading coefficients. By means of [12, with p k = b s b ik d 2H d 2V .

= E$ a + b + c + d .
2
Th. II.1], we observe the power-scal-
= E$ a . + E$ b . Proof
2 2
ing law for a single-user system (user
+ E $ c . + 2E " ad ,
2 k only) over the spatially correlated By substituting the covariance matri-
channels as ces on the def in ition into (17)
+ E $ d .,
2
 (20) and (18) with tr (H u ) = Nb ik d 2H d 2V and
N
Mb k + Mb s / m n (U H R si U) m N - n + 1 (R ik) u ) = Nb ik d H d V , we arrive at the
tr (H 2 2 4 4

where a = < u k < ,2

b = u kH HUg k,
c= n=1 desired results in “Corollary 1.” 
H H H
g k U H u k, and d = < HUg k <2 .
The # E {< z k <2} # Mb k “Lemma 2” concludes the favorable
first expectation in the last equality  N
propagation of the spatially correlated
of (20) is computed by deploying [11,
+ Mb s / m n (U H R si U) m n (R ik), (25)
Rayleigh fading model.
=1
1n44444424444443
Lemma 9] as _ ck

Lemma 2
E $ u k . = tr (R k) + tr ^R k2 h
4 2
w h e r e {m n (U H R si U)} nN= 1 a n d B y a s s u m i n g t h a t liminfN " 3 tr

=M 2 2
bk + tr ^R 2k h . (21) {m n (R ik)} nN= 1 are the eigenvalues of (U H RUR) / N 2 0 a n d limsup N " 3

IEEE SIGNAL
Authorized licensed use limited to: University of Leeds. Downloaded PROCESSING MAGAZINE
on December | 10:32:03
05,2024 at | from IEEE Xplore. Restrictions apply.
May 2022 UTC 101
 E $ z k z lH .
< U H RUR <2 1 3, the spatially cor- 2 at the bottom of the page as M, N " 3.
related Rayleigh channel model offers In (37), the numerator and denomina-
u u
the asymptotically favorable propa-  = tr ((R k + tr (H k) R s) (R l + tr (H l) R s)). tor, respectively, scale up in the order
(33)
gation property, and the deterministic- of N 2 p k p l and MN 2 p k p l, thus the
favorable propagation metric can be Substituting (17) and (33) into (6), we convergence is proportional to 1/M. By
computed in the closed-form expres- obtain the result as shown in the lemma only considering the spatially corre-
sion as and conclude the proof.  lated model, it explicitly indicates that
We now exploit the identity the deterministic-favorable propaga-
FPkl = tr (X + Y) = tr (X) + tr (Y) for the two tion metric converges to zero quickly as
u k) R s) (R l + tr (H
tr ((R k + tr (H u l) R s)) matrices, X and Y; the numerator of (29), more antenna and phase-shift elements
u
(Mb k + Mb s tr (H k)) (Mb l + Mb s tr (H u l)) . denoted by Num kl, is reformulated as are installed. Figure 2(a) plots the deter-
(29) ministic-favorable propagation metric in
Num kl = tr (R k R l) + tr (H u k) tr (R s R l)
(29) for a system with an equal number
Proof + tr (H u l) tr (R k R s) of antennas and phase-shift elements.
By utilizing the aggregated channel on  + tr (H u k) tr (H
u l) tr (R 2s ). (34) The covariance matrices are given in
the definition in (2) together with its (14) and (15) with S o = 3, v o = 3c, and
statistic information in “Lemma 1,” we Meanwhile, denoting by Den kl the d H = d V = m/4, respectively. The
first observe (30) denominator of (29), and it is reformu- BS is located at the origin of a Carte-
lated as sian system coordinate, while the RIS
z kH z l is at the (x, y)- coordinate (125, 125)
# 1 Den kl = tr (R k) tr (R l) + tr (Hu k) tr (R s) tr (R l)
E {< z k <2} E {< z l <2} a kl m and two users are at (250, 12.5)
u
+ tr (H l) tr (R k) tr (R s)
H H H H
u kH u l u k HUg l g k U H u l m and (125, - 250) m, respectively.
# + +  + tr (H u k) tr (H
u l) (tr (R s)) 2 . (35) We observe that the correlated Ray-
MN MN MN
 leigh fading channels offer the favor-
g kH U H H H HUg l
+ " 0, as M, N " 3. As the covariance and phase-shift matri- able propagation, with the lowest value
MN
(30) ces are all positive semidefinite, we can (the best performance) belonging to the
use the identity tr (XY) # tr (X) tr (Y) conventional massive MIMO (denoted
I n (30), a kl = b s b ik b il mu, where for the positive semidefinite matrices as “Direct Link” in Figure 2). The worst
mu = liminfN " 3 tr (U RUR) /N. The ob­­­­­­
H
to obtain Num kl # Den kl . Dividing the case is for a system with only the indi-
tained result in (30) implies that the numerator and denominator of (29) by rect links (indicated as “Indirect Link”
asymptotically favorable propagation N 2 M 2, one obtains in Figure 2). The aggregated channels
in (5) holds. (designated as “Both Links” in Figure 2)
For a random variable X, its va­­ lim FPkl = fall in between.
M, N " 3
r ia nce is comput e d a s Var {X} = u k) tr (H
u l) tr (R 2s ) / (N 2 M 2) We now compute the deterministic
tr (H
E {| X | 2} - |E {X}| 2 . Thus, Var {z kH z l} = lim
M, N " 3 tr (H k) tr (H l) (Mb s) 2 / (N 2 M 2)
u u " 0, channel-hardening metric for the spa-
E {|z kH z l | 2} d u e t o t h e zero mean (36) tially correlated Rayleigh fading chan-
o f the aggregated channels. Next nel model (13), as shown in “Lemma 3.”
we have thanks to the fact that tr (R 2s ) /M 2 " 0
as M " 3. As a consequence of (36), Lemma 3
E $ z kH z l . = tr ^E " z l z lH , E " z k z kH ,h .
2
under the spatially correlated Rayleigh If the propagation channels are spatially
(31) fading propagation, the convergence of correlated Rayleigh fading channels by
(29) is very fast as M increases. For the (13), the deterministic channel-hard-
From (31), we now compute E {z k z kH}. uncorrelated Rayleigh fading model, ening metric is computed in the closed
We have the deterministic-favorable propaga- form as in (38) shown at the bottom of
tion metric is computed in (37) shown the page.
E {z k z kH}
= E {u k u kH} + E " HUg k g kH U H H H , 2
b k b l + Nb k p l + Nb l p k + N p k p l
FPkl = " 0, (37)
= R k + E " R s1/2 HR u H R 1s /2 ,
u 1si/2 UR ik U H R si1/2 H Mb k b l + MNb k p l + MNb l p k + MN 2 p k p l
(32) u k ) R s,
= R k + tr (H
2tr (R s R k) tr (H u k) 2 tr (R 2s ) + tr (H
u k) + tr (H u 2k ) (M 2 b 2s + tr (R 2s )) + tr (R 2k )
CH k = u .
where the last equality in (32) is (Mb k + Mb s tr (H k)) 2
(38)
obtained by utilizing [11, Lemma 8]. In
a similar manner, we obtain E {z l z lH} = 2Nb k p k + N 2 p 2k + (M + 1) Np 2k + b 2k
u l) R s . Hence, we obtain (31) CH k = " 0, (39)
R l + tr (H M (b k + Np k) 2
in the closed form as

102
Authorized IEEE SIGNAL
licensed use limited to: University of Leeds. Downloaded 05,2024 at| 10:32:03
PROCESSING MAGAZINE
on December | from IEEE Xplore. Restrictions apply.
May 2022 UTC
 Proof better channel-hardening effects. E {h k h kH} thanks to the assistance of
2 4
Note that Var {|| z k || } = E {|| z k || } - However, the closed-form expression the RIS with 100 antennas equipped
(E {|| z k || 2}) 2, then plugging (17) and (39) is independent of the phase-shift at the BS.
(18) into (10), we obtain the result as coefficients, and we do not have a
shown in the lemma.  smart environment control based on What we have learned
The numerator and denomina- the channel statistic with the uncor- This “Lecture Notes” article described
tor of (38) scale up with the order related Rayleigh fading channels. three fundamental properties of RIS-
of N 2 M + M 2 N and M 2 N 2, thus the Figure 2(b) displays the deterministic assisted massive MIMO channels com-
deterministic channel-hardening met- cha n nel-ha rden- p r isi ng favorable
ric converges to zero as the number ing metric in (38), The obtained result in propagation, chan-
of antennas and phase shifts grows. which demonstrates (30) implies that the nel hardening, and
For the uncorrelated Rayleigh fading that the channels asymptotically favorable rank deficiency. The
channels, one can simplify (38) as (39), are less hardened propagation in (5) holds. definitions of those
as M, N " 3 and the convergence is when including the properties were pre-
proportional to (M + N) / (MN) . The extra channel coefficients from the sented, along with sever a l me a-
metric (39) depends on the number indirect link under the presence of the s u r e s that are deterministic metrics
of phase-shift elements and anten- RIS. Furthermore, Figure 2(c) shows a expressed by channel statistics on ly.
nas that could be optimized to have higher rank of the covariance matrix The closed-form expressions of the

100 0.25
Deterministic Favorable Propagation

Deterministic Channel Hardening

10–1 0.2

10–2 0.15

10–3 0.1

10–4 0.05

10–5 0
100 150 200 250 300 350 400 100 150 200 250 300 350 400
Number of Engineered Scattering Elements Number of Engineered Scattering Elements

Both Links, M = N Direct Link, M = N Indirect Link, M = N

Both Links, M = N/2 Direct Link, M = N/2 Indirect Link, M = N/2

(a) (b)

100

10–5
Eigenvalue

10–10
Both Links, N = 400
Indirect Link, N = 400
10–15 Direct Link
Both Links, N = 100
Indirect Link, N = 100
10–20
0 20 40 60 80 100
Sorted Eigenvalue Index
(c)

FIGURE 2. The fundamentals of an RIS-assisted massive MIMO system with a finite number of BS antennas and engineered scattering elements of the
RIS. (a) The deterministic-favorable propagation metric, (b) deterministic channel-hardening metric, and (c) sorted eigenvalues with M = 100.

IEEE SIGNAL
Authorized licensed use limited to: University of Leeds. Downloaded PROCESSING MAGAZINE
on December | 10:32:03
05,2024 at | from IEEE Xplore. Restrictions apply.
May 2022 UTC 103
 deterministic metrics were computed Stephen O. Rice Prize in 2015, IEEE processing with the Royal Institute of
for a system with arbitrary (but finite) ComSoc Leonard G. Abraham Prize in Technology, Stockholm, Sweden. He is
numbers of antennas and phase shifts. 2017, and Best Ph.D. Award from EUR- currently the director for the Interdisci-
The numerical results demonstrate that ASIP in 2018. He serves as the editor of plinary Centre for Security, Reliability
by exploiting the Rayleigh fading model, IEEE Transactions on Wireless Commu- and Trust, the University of Luxem-
the aggregated channels still offer favor- nications and IEEE Wireless Communi- bourg, Esch-sur-Alzette, L-1855, Lux-
able propagation and channel hardening cations Letters. His main research embourg. He is a recipient of the IEEE
but are worse than conventional massive interests include massive multiple-input, Signal Processing Society Technical
MIMO ­systems. We further provided multiple-output (MIMO), cell-free mas- Achievement Award, EURASIP Group
insightful observations on the rank sive MIMO, physical-layer security, and Technical Achievement Award, and
d ef ic ie ncy t hat occurs when there cooperative communications. European Research Council advanced
are a limited number of scat terers Symeon Chatzinotas (symeon. research grant (twice). He is a Fellow
in the propagation environments. We [email protected]) received his M.Eng. of IEEE.
approved that this issue can be handled degree in telecommunications from
by the assistance of an RIS. Aristotle University of Thessaloniki, References
[1] T. L. Marzetta, E. G. Larsson, H. Yang, and H. Q.
Greece and his M.Sc. and Ph.D. de- Ngo, Fundamentals of Massive MIMO. Cambridge,
Acknowledgments grees in electronic engineering from U.K.: Cambridge Univ. Press, 2016.
The work of Trinh Van Chien, Symeon the University of Surrey, U.K., in [2] F. Rusek, D. Persson, B. Lau, E. Larsson, T.
Marzetta, O. Edfors, and F. Tufvesson, “Scaling up
Chatzinotas, and Björn Ottersten was 2003, 2006, and 2009 respectively. He MIMO: Opportunities and challenges with very large
supported by Reconfigurable Intelligent is a full professor/chief scientist I and arrays,” IEEE Signal Process. Mag., vol. 30, no. 1,
pp. 40–60, 2013, doi: 10.1109/MSP.2011.2178495.
Surface for Smart Cities (RISOTTI) head of the SIGCOM Research Group
[3] E. Björnson, J. Hoydis, and L. Sanguinetti, “Massive
under project FNR/C20/IS/14773976/ at the Interdisciplinary Centre for MIMO networks: Spectral, energy, and hardware effi-
RISOTTI. The work of Hien Quoc Ngo Security, Reliability and Trust, the ciency,” Found. Trends Signal Process., vol. 11, nos.
3–4, pp. 154–655, 2017, doi: 10.1561/2000000093.
was supported by the U.K. Research and University of Luxembourg, Esch-
[4] T. V. Chien, H. Q. Ngo, S. Chatzinotas, M. D.
Innovation Future Leaders Fellowships sur-Alzette, Luxembourg. He is co- Renzo, and B. Otternsten, “Reconfigurable intelligent
under grant MR/S017666/1. ordinating the research activities on surface-assisted cell-free massive MIMO systems over
spatially-correlated channels,” submitted for publication.
communications and networking, act- [5] Q. Wu and R. Zhang, “Intelligent reflecting sur-
Authors ing as a PI for more than 20 projects face enhanced wireless network via joint active and
passive beamforming,” IEEE Trans. Wireless
Trinh Van Chien (trinhchien.dt3@ and main representative for the 3rd Commun., vol. 18, no. 11, pp. 5394–5409, 2019, doi:
gmail.com) received his Ph.D. degree in Generation Partnership Project, Euro- 10.1109/TWC.2019.2936025.
electrical engineering with specialization pean Telecommunications Standards [6] Q. Wu, S. Zhang, B. Zheng, C. You, and R.
Zhang, “Intelligent reflecting surface aided wireless
in communication systems from Institute, and Digital Video Broadcast- communications: A tutorial,” IEEE Trans. Commun.,
Linköping University, Sweden, in 2020. ing. He was the co-recipient of the vol. 69, no. 5, pp. 3313– 3351, 2021, doi: 10.1109/
TCOMM.2021.3051897.
He is currently a research associate 2014 IEEE Distinguished Contribu-
[7] C. Huang, A. Zappone, G. C. Alexandropoulos, M.
at University of Luxembourg, L-1855, tions to Satellite Communications Debbah, and C. Yuen, “Reconfigurable intelligent sur-
Luxembourg. His research interests fo- Award and Best Paper Awards at the faces for energy efficiency in wireless communication,”
IEEE Trans. Wireless Commun., vol. 18, no. 8, pp.
cus on convex optimization problems, EURASIP Journal on Wireless Com- 4157–4170, 2019, doi: 10.1109/TWC.2019.2922609.
theoretical analysis, and machine learn- munications and Networking, EAI In- [8] H. Q. Ngo and E. G. Larsson, “No downlink
ing applications for wireless communi- ternational Conference on Cognitive pilots are needed in TDD massive MIMO,” IEEE
Trans. Wireless Commun., vol. 16, no. 5, pp. 2921–
cations and image & video processing. Radio Oriented Wireless Networks, 2935, 2017, doi: 10.1109/TWC.2017.2672540.
He is an IEEE member. and International Communications [9] E. Björnson and L. Sanguinetti, “Rayleigh fading
modeling and channel hardening for reconfigurable
Hien Quoc Ngo ([email protected]. Satellite Systems Conference. He is intelligent surfaces,” IEEE Wireless Commun. Lett.,
uk) received his Ph.D. degree in com- currently on the editorial boards of vol. 10, no. 4, pp. 830–834, 2021, doi: 10.1109/
LWC.2020.3046107.
munication systems from Linköping IEEE Transactions on Communica-
[10] T. Van Chien, A. K. Papazafeiropoulos, L. T.
University, Sweden, in 2015. He is cur- tions, IEEE Open Journal of Vehicu- Tu, R. Chopra, S. Chatzinotas, and B. Ottersten,
rently a reader with Queen’s University lar Technology, and International “Outage probability analysis of IRS-assisted systems
under spatially correlated channels,” IEEE Wireless
Belfast, BT7 1NN, U.K., and a UKRI Journal of Satellite Communications Commun. Lett., vol. 10, no. 8, pp. 1815–1819, 2021,
Future Leaders Fellow. He has coau- and Networking. doi: 10.1109/LWC.2021.3082409.
thored many research articles in wire- Björn Ottersten (bjorn.ottersten@ [11] T. V. Chien and H. Q. Ngo, Massive MIMO
Channels. London: IET Publishers, 2020, ch. 11.
less communications and coauthored the uni.lu) received his Ph.D. degree in
[12] J. B. Lasserre, “A trace inequality for matrix
Cambridge University Press textbook electrical engineering from Stanford product,” IEEE Trans. Autom. Control, vol. 40, no. 8,
“Fundamentals of Massive MIMO” University, California, in 1990. In 1991, pp. 1500–1501, 1995, doi: 10.1109/9.402252.

(2016). He received the IEEE ComSoc he was appointed professor of signal SP

104
Authorized IEEE SIGNAL
licensed use limited to: University of Leeds. Downloaded 05,2024 at| 10:32:03
PROCESSING MAGAZINE
on December | from IEEE Xplore. Restrictions apply.
May 2022 UTC