Scalability Aspects of Cell-Free Massive MIMO

Giovanni Interdonato∗† , Pål Frenger∗ and Erik G. Larsson†

∗
Ericsson Research, 583 30 Linköping, Sweden
†
Department of Electrical Engineering (ISY), Linköping University, 581 83 Linköping, Sweden
{[Link], [Link]}@[Link], [Link]@[Link]

Abstract—Ubiquitous cell-free massive MIMO (multiple-input nology [13] and user-centric transmission. The TDD massive
multiple-output) combines massive MIMO technology and user- MIMO operation brings high spectral and energy efficiency,
centric transmission in a distributed architecture. All the access and enables substantial reductions of the estimation overhead
points (APs) in the network cooperate to jointly and coherently
serve a smaller number of users in the same time-frequency as the uplink estimates can be reused in the downlink for
resource. However, this coordination needs significant amounts precoding by exploiting (after calibration) channel reciprocity.
of control signalling which introduces additional overhead, while The conventional operation consisting in estimating the down-
data co-processing increases the back/front-haul requirements. link channel followed by the user feedback transmission can be
Hence, the notion that the “whole world” could constitute one conveniently avoided. Hence, the estimation overhead scales
network, and that all APs would act as a single base station, is
not scalable. In this study, we address some system scalability with the number of users rather than the number of APs.
aspects of cell-free massive MIMO that have been neglected in Moreover, the massive number of serving APs brings a two-
literature until now. In particular, we propose and evaluate a fold benefit: it introduces additional macro-diversity and it
solution related to data processing, network topology and power reduces the multi-user interference, thanks to the favorable
control. Results indicate that our proposed framework achieves propagation [13] phenomenon. The inter-cell interference is
full scalability at the cost of a modest performance loss compared
to the canonical form of cell-free massive MIMO. effectively suppressed by the user-centric data transmission
Index Terms—Cell-free Massive MIMO, distributed wireless design: each user equipment (UE) is surrounded by serving
system, power control, spectral efficiency, system scalability. APs, hence, it experiences no cell boundaries.
Although many aspects of a cell-free massive MIMO net-
I. I NTRODUCTION work have been deeply studied in the literature, system scala-
Coordinated distributed wireless systems [1] leverage signal bility has not been fully addressed until now. In the canonical
co-processing at multiple access points (APs) to guarantee case [2], a cell-free massive MIMO network comprises many
high connectivity, reduce inter-cell interference and improve distributed APs simultaneously serving a smaller number of
the user experience. Connectivity is enhanced thanks to the UEs. All the APs are connected to one central processing
shorter AP-to-user distance; the joint coherent transmission unit (CPU) which is responsible for coordination and data
from geographically distributed APs yields macro-diversity processing. Hence, from the UE perspective, the entire network
gain; and the coordination enables APs to select transmit acts as an infinitely large single cell, served by a single base
strategies jointly (by sharing channel state information) in station comprising all the APs and the CPU. Such a framework
order to reduce inter-cell interference. is certainly unrealistic in practice, and the notion that data from
Ubiquitous cell-free massive MIMO [2]–[4], where MIMO all the APs would be processed coherently is not scalable.
stands for multiple-input multiple-output, relies on these prin- Contributions: In this study, we propose a scalable cell-free
ciples. However, it differs from prior coordinated distributed massive MIMO framework, specifically considering data trans-
wireless systems such as network MIMO [5], [6], virtual mission strategies and power control. Unlike prior studies, we
MIMO [7], multi-cell MIMO cooperative networks [8] and co- consider a realistic scenario where multiple CPUs serve dis-
ordinated multipoint with joint transmission (CoMP-JT) [9]– joint clusters of APs. Finally, the spectral efficiency provided
[11]. The performance of all the aforementioned frameworks, by the proposed scheme is compared with the canonical cell-
in their canonical forms, are limited by two factors: (i) the free massive MIMO, and conventional (cell-centric) CoMP-JT.
inter-cell (out-of-cluster) interference inherent in the static
cell-centric transmission design [12]; (ii) the large amount II. T HE S CALABILITY P ROBLEM
of control signalling, channel state information (CSI) and
data exchange, required for the coordination, that increases Compared to prior coordinated distributed wireless systems,
the back/front-hauling requirements and the overhead. These canonical cell-free massive MIMO is a step forward in terms of
become more significant as the number of APs grows. system scalability. As mentioned earlier, the TDD operation
Cell-free massive MIMO combines the benefits derived allows to make the estimation overhead independent of the
from using time division duplex (TDD) massive MIMO tech- number of APs. In addition, the use of conjugate beamforming
(i.e., maximum-ratio transmission, MRT), which has been
This paper was supported by the European Union’s Horizon 2020 research
and innovation programme under grant agreement No 641985 (5Gwireless), especially advocated in the literature, enables a simple and
and ELLIIT. fully distributed processing.

978-1-5386-8088-9/19/$31.00 ©2019 IEEE

 Let M , K the number of APs and UEs, respectively. The Simpler effective policies, proposed in [3], [14], enable
data signal sent by AP m to all UEs, using MRT, is given by distributed computation of the power control coefficients at the
√ XK √ cost of reduced performance. In the distributed power control
∗
xm = ρd ηmk ĝmk qk , (1) case, there is no power control coefficient exchange over the
k=1
front-haul network, and only the power control coefficients
where qk is the unit-power data symbol intended for the kth associated to the same AP are entangled each other (local
UE, ρd is the normalized transmit signal-to-noise ratio related “butterfly effect”) to satisfy the per-AP power constraint in (3).
to the data symbol (i.e., the radiated power over the power
∗
of the noise figure). The term ĝmk represents the precoding B. User-centric vs Cell-centric Clustering
factor, namely the conjugate of the channel estimate between The basic way to attempt to address the scalability problem
AP m and UE k. The power spent by AP m on the service of consists in deploying clusters of APs and to confine the
UE k is parametrized in terms of a power control coefficient signal co-processing within the cluster. In literature, there are
ηmk , where 0 ≤ ηmk ≤ 1 [3]. two suggested approaches: (i) cell-centric clustering, which
With MRT, the precoders are determined at each APs by consists in deploying fixed disjoint clusters of APs where the
using only local CSI, thus no CSI and information about APs in a cluster serve only the UEs residing in their joint
precoders are exchanged over the front-haul network between coverage area; (ii) user-centric clustering, which consists in
APs and CPU. Hence, precoding is also scalable. deploying dynamic (possibly partially overlapped) clusters of
However, there are at least three main scalability issues: APs based on the needs of each served UE.
• Data processing. Data destined for every UE in the Cell-centric clustering constitutes the canonical method to
network would have to be sent from the CPU to every group cooperating APs. Each cluster is served by one CPU,
AP. This would render the computational complexity at and the APs connected to a given CPU will form a cluster.
each AP unsustainable; These clusters will either mutually interfere, or they will have
• Network topology. The complexity of the interconnect to cooperate through coherent transmission, which brings back
at the CPU does not scale as the CPU will need one the scalability problem. In such a system there is no coherent
connection to each AP in the network; cooperation on data, or cooperation on power control, between
• Power control. The calculation of the power control the CPUs. Although this system is fully scalable, it suffers
coefficients does not scale, even taking computational from poor performance (it will be addressed later). Network
issues aside. This aspect is discussed in detail next. MIMO, CoMP-JT and multi-cell MIMO cooperative network
are conventionally implemented in a cell-centric fashion.
A. Is Power Control Really Scalable? Conversely, the idea behind the user-centric clustering is
Each AP has a transmission power constraint related to ρd , that, for each UE, only a small number of APs should
i.e., the per-AP power constraint is given by participate in the service of that UE. Effectively each UE
is served by a cluster of near-by APs. One can view user-
E |xm |2 ≤ ρd , ∀m,

(2) centric transmission as a special case of the common cell-
which can be expressed as, free massive MIMO setup where all power control coefficients
{ηmk } of a given UE k are constrained to be zero, except
XK
ηmk γmk ≤ 1, ∀m, (3) for those associated with the closest APs. The user-centric
k=1 approach enables to suppress the inter-cell interference thus it
where
 γmk is the mean-square of the channel estimate, i.e., performs better than the cell-centric approach. However, this
E |ĝmk |2 = γmk . It is proportional to the mean-square of concept fundamentally does not solve the network topology
the effective channel βmk [3]. problem described earlier, i.e., that all APs must be connected
For given power control coefficients, analytical capacity to a CPU. In addition, it requires more control signalling to
lower bounds (“achievable rates”) exist that quantify perfor- dynamically form user-specific clusters “on demand”. User-
mance given some pre-determined path-loss and fading model. centric approach has been subsequently introduced in multi-
The power control coefficients are functions only of the long- cell cooperative network [15], CoMP-JT [16], cooperative
term channel statistics and must be computed centrally. Hence, small cells, under the name of cover-shifts [17], and C-
they need to be sent by the CPU to all the APs. RAN1 [18], [19].
Algorithms for the optimal selection of {ηmk } are available.
Specifically, max-min fairness power control, that ensures that III. P ROPOSED S OLUTION
every UE in the network obtains the same quality of service In this section, we describe our proposed fully scalable and
(rate), is possible (though computationally very demanding) distributed cell-free massive MIMO framework. Distributed
through the use of convex optimization tools [3]. However, the resource allocation problems for cooperative beamforming
power control coefficient associated with some UE k and some
1 Cloud Radio Access Network is an architecture that moves the baseband
AP m depends on the channel statistics of UE-AP pairs very
processing from the APs to “the cloud”. Today, C-RAN is mainly used to
far away. This “butterfly effect” entangles the power control implement conventional cellular systems, but in the future it might be used
coefficients across the whole network. to implement a cell-free massive MIMO system.
systems have been considered in [20, Section 4.2]. However,
to our knowledge no scalable solutions for power control CPU2

and precoding are available that do not rely on perfect CSI AP11 AP12
AP10
assumptions. AP13
UE2 AP14
The APs are grouped into N pre-determined cell-centric AP7
AP9 UE3
clusters D1 , . . . , DN . Each cluster in turn is connected to one AP16
AP8 AP15
AP19
CPU. The CPUs are interconnected but operate autonomously. UE4
CPU1 AP5 AP17
It is assumed that a global phase reference is shared. Each UE AP18 AP20
AP6 UE1
is receiving service from one or a few cell-centric clusters. AP3
AP4
Let Bk be the number of cell-centric clusters that partici-
pate in the service of the kth UE (typically this will be a AP2 AP1
CPU3
small number). We denote these serving cell-centric clusters
by Dk1 , . . . , DkBk . To select said clusters, the user-centric Fig. 1. Example of cell-free massive MIMO system with user-centric
concept is applied. Specifically, for the kth UE, the Lk selected clustering and multiple interconnected CPUs. Different colors for CPUs and
APs, i.e., APk1 , . . . , APkLk , that form the user-centric cluster front-haul networks correspond to different cell-centric clusters.
are identified, as shown in the example in Fig. 1. These APs
are selected according to some metric, for example distance,
set as follows
or channel quality. The clusters involved by the selected APs 
define the serving cell-centric clusters. Hence, the data to UE k  P fγ(Gmk )
, if m ∈ {Dk1 ∪. . .∪DkBk },
mk0 f (Gmk0 )
is distributed only to the involved CPUs: CPUk1 , . . . , CPUkBk . ηmk = k0 ∈Tm (4)
0, otherwise,

As shown in Fig. 1, a user-centric cluster might include
APs belonging to different cell-centric clusters. Hence, UE k where f (·) is a pre-determined function that is calculated
is served by all the APs of the selected cell-centric clusters. locally at AP m, and Tm is the set of UEs served by APm , i.e.,
For instance, in Fig. 1, UE2 is served by all the APs managed given m, the set of k for which m ∈ {Dk1 ∪ . . . ∪ DkBk }. f (·)
by CPU1 and CPU2 , that is cell-centric cluster D1 and D2 . is function of Gmk0 , a set that comprises long-term statistical
Power control is applied independently in each APs. To knowledge of the channel, such as γmk0 , or βmk0 , k 0 ∈ Tm .
satisfy (3) with equality, the power control coefficients are The denominator in (4) constitutes the normalization term en-

𝐶𝑃𝑈1 𝐶𝑃𝑈2 𝐶𝑃𝑈3

𝑘 𝐴𝑃𝑠 ∈ 𝐷1 𝐴𝑃𝑠 ∈ 𝐷2 𝐴𝑃𝑠 ∈ 𝐷3
uplink pilot

Channel estimation to UE 𝑘 Channel estimation to UE 𝑘 Channel estimation to UE 𝑘

Performed per AP

𝑘 =𝑘+1
𝑘 =𝑘+1

𝑘 =𝑘+1

Is the channel Is the channel Is the channel

good enough to serve good enough to serve good enough to serve
N UE 𝑘? N UE 𝑘? N UE 𝑘?

Y Y Y

Destined to the CPU the AP belongs to | Convey the message: “I serve UE 𝑘”

Destined to all the APs belonging to the cluster | Convey the message: “You serve UE 𝑘” + DATA for UE 𝑘

Precoder, and Power Precoder, and Power

per AP

Control Coefficient Control Coefficient

DETERMINATION to UE 𝑘 DETERMINATION to UE 𝑘

Data transmission
Data transmission

Fig. 2. Signalling diagram depicting the operation of the proposed cell-free massive MIMO framework.
suring that (3) is satisfied. Importantly, there is no inter/intra- 1.5
cluster interaction between APs in the selection of the power
control coefficients. 1
UE k
The signalling diagram in Fig. 2 depicts the operation of the
proposed framework from the UE k perspective. In the figure, 0.5
we follow the same example as in Fig. 1: UE2 , is served by

y [km]
all APs in the cluster D1 ∪ D2 , since the user-centric cluster 0
of UE2 involves APs of the cell-centric cluster D1 and D2 .
CPU3 does not participate in serving UE2 . We recall that the -0.5
data transmission does not regard only UE k. A given AP m
serve coherently all the UEs belonging to Tm , as follows -1
√ X √ ∗
xm = ρd ηmk ĝmk qk , (5) -1.5
k∈Tm
-1.5 -1 -0.5 0 0.5 1 1.5
which differs from (3) as not all the UEs are served when a x [km]
user-centric transmission is designed. In the example in Fig. 2, (a) Example of conventional CoMP-JT.
it is assumed that the APs selected to form the user-centric
cluster are chosen according to the channel quality they offer 1.5

to UE k, e.g., the largest large-scale-fading-based AP selection

criterion proposed in [21]. 1
UE k

IV. N UMERICAL R ESULTS 0.5

Our simulations aim to compare the downlink spectral effi-

y [km]
ciency (SE) provided by the proposed framework for cell-free 0
massive MIMO with the canonical cell-free massive MIMO,
and a conventional multi-cell cooperative MIMO system (e.g., -0.5
CoMP-JT). Everything else being equal, the only difference
between these three setups consists in the set of APs that serve -1
each single UE.
-1.5
A. Simulation Scenario -1.5 -1 -0.5 0 0.5 1 1.5
x [km]
Instead of implementing a classical wrap-around technique,
(b) Example of the proposed cell-free framework with Bk = 2.
we use a simple embedding technique to substantially elimi-
Fig. 3. Cross markers indicate APs while the circle marker indicates an
nate border effects. We consider square A of 2.5 km × 2.5 arbitrary UE k.
km. In the middle of A, a focus square B of 1 km × 1 km is
defined. For transmission, all the elements in A are considered,
but only the elements inside of B are taken into account for Dk1 , Dk2 which in Fig. 3(b) have been “merged” (they can
the performance evaluation. The effect is substantially that be identified by visual inspection from Fig. 3(a)).
elements inside of B are samples of a “stationary” distribution, Lastly, in canonical cell-free massive MIMO, all the APs
not affected by edge effects; and that edge effects affect only in the network serve every single UE (i.e., Fig. 3(a) with the
UEs at the boundary of A. We consider 625 single-antenna cluster containing all the APs).
APs and 125 single-antenna UEs uniformly at random placed
in A, such that M = 100 APs and K = 20 UEs fall into B, B. Spectral Efficiency Evaluation
and the remaining 525+105 “dummy” elements fall into the To evaluate the per-user downlink SE, we use the closed-
area between A and B. form expression given in [3], which assumes single-antenna
Fig. 3(a) shows the deployment of a CoMP-JT network APs, conjugate beamforming, channel estimation errors and
in the considered area. Polygons of different colors contain non-orthogonal uplink pilots. The channel is modeled as in [3],
APs (cross markers) of the same cell-centric cluster, i.e., APs and includes: three-slope path-loss with Hata-COST231 prop-
served by the same CPU. In conventional CoMP-JT, a given agation model; uncorrelated shadow fading; and independent
UE k (circle marker) is served by all the APs belonging to the Rayleigh fading. The simulation settings also resemble those
cell-centric cluster that guarantees the highest quality service in [3]: the maximum radiated power is 200 mW per AP
(the cluster containing UE k depicted in Fig. 3(a)). and 100 mW per UE; the carrier frequency is 1.9 GHz, the
The same APs deployment but for the proposed cell-free transmission bandwidth 20 MHz, and the coherence bandwidth
massive MIMO framework is illustrated in Fig. 3(b). In this 200 kHz; the noise figure is 9 dB; the antenna height of the
example, it is assumed that, to serve UE k, the APs selected APs and the UEs are 15 m and 1.65 m, respectively.
by the user-centric approach belong to two cell-centric clusters We assume that each AP has a set of 10 orthogonal pilots.
(Bk = 2). Hence, UE k is served by the cell-centric clusters The uplink pilots are randomly assigned to the UEs and there
 1
3
95%-likely
0.8 Median M = 200
2.5

Per-User SE [bit/s/Hz]
0.6
2 M = 100
0.4
1.5

0.2 M = 50
1

0 0.5
10-2 10-1 100 101

0
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Fig. 4. The CDF of the downlink per-user minimum spectral efficiency. α, f(γmk , α) = γmk
α

Fig. 5. Per-user SE versus channel-dependent power control strategy: α

is no coordination among the APs in assigning the pilots. adjusts the relationship between ηmk and γmk .
Fig. 4 shows the cumulative distribution function (CDF) of
the downlink minimum spectral efficiency per UE provided
by the proposed scalable cell-free massive MIMO framework, of the power control coefficients is low (i.e., large-scale fading
canonical cell-free massive MIMO, and CoMP-JT. In these time scale, which spans over multiple coherence intervals).
√
simulations, we set Gmk = γmk , f (γmk ) = 1/ γmk . Hence, Fig. 5 shows the per-user spectral efficiency achieved by
the coefficient that adjusts the downlink transmitted power the scalable cell-free massive MIMO framework, varying the
from AP m to UE k is given by exponent α in the function f (γmk , α) = γmk α
. In these
 1/√γ simulations, we use the same settings as in Fig. 4, and
 P √mk γmk0 , if m ∈ {Dk1 ∪. . .∪DkBk },
ηmk = k0 ∈Tm (6) the largest large-scale-fading-based AP selection method, de-

0, otherwise. scribed in [21], to form the user-centric clusters. Results
demonstrate that, both in terms of 95%-likely SE and me-
The AP selection method2 to form the user-centric clusters dian SE, very good performance can be achieved by setting
√
is, in this example, based on the AP-to-UE distance: the 5 f (γmk ) = 1/ γmk . Increasing the number of APs in the
closest APs are selected and determines the serving cell-centric system does not affect the choice of the power control strategy
clusters. Results show that the proposed scalable cell-free as the user-centric approach selects in any case very few
massive MIMO framework substantially outperforms CoMP- serving cell-centric clusters. This particular choice of the
JT. This gap derives from the user-centric transmission that power control coefficients ensures that the effective power
suppresses the inter-cell interference. Conversely, the perfor- allocated to the service of UE k by AP m is proportional to
√
mance loss compared to canonical cell-free massive MIMO γmk ηmk ∝ γmk ; i.e., the better channel between an AP and
is modest. Hence, very few serving cell-centric clusters are a UE, the more power is allocated by the AP to the service of
sufficient to guarantee good performance. Involving more that UE. This conclusion contrasts with that in [14] since the
APs in the user-centric cluster by adjusting the AP selection heuristic uniform power allocation (HUPA), therein proposed,
criterion increases the number of serving cell-centric clusters works well only for large number of serving APs.
per UE. This results in better performance (i.e., the dash- The latter is also confirmed by Fig. 6 which shows the CDF
dotted curve approaches the solid one) but also increases the of the per-user minimum SE achieved by the uniform power
complexity of the system as more APs and CPUs are involved allocation in [3], the HUPA proposed in [14], and the proposed
in the coherent transmission and requires coordination. √
channel-dependent power allocation with f (γmk ) = 1/ γmk .
C. Distributed Power Control Strategies The proposed power allocation provides substantial SE gain
over both the alternative schemes which perform almost iden-
The power control strategy chosen in (6) requires additional tically. This gain derives from the fact that, when user-centric
motivation. Firstly, this choice allows to make the power transmission is implemented, very few APs are effectively
control fully distributed as (6) involves only local CSI. This serving a given UE. Under this condition, the exponential
yields benefits to the scalability of the system as no power behavior of the power scatter plot proposed in [14] does
control coefficient needs to be exchanged over the front- √ not
hold. A similar gain is obtained by setting f (βmk ) = 1/ βmk ,
haul network. Secondly, it has low complexity as there is since βmk = γmk + mk [3], where mk is the variance of
no optimization problem to solve. In addition, only long-term the channel estimation error, and βmk = γmk if estimation
channel statistics are used in (6) thus the updating frequency is perfect. In general, we can conclude that it is better to
2 AP selection methods are nothing more than a power control strategy. An “customize” the transmitted powers by setting the power
AP m is not involved in the transmission to a UE k by setting ηmk = 0. control coefficients according to the specific user channel
 [2] E. Nayebi, A. Ashikhmin, T. L. Marzetta, and H. Yang, “Cell-free Mas-
1 sive MIMO systems,” in Proc. Asilomar Conf. Signals, Syst., Comput.,
Nov. 2015, pp. 695–699.
[3] H. Q. Ngo, A. Ashikhmin, H. Yang, E. G. Larsson, and T. L. Marzetta,
0.8 “Cell-free Massive MIMO versus small cells,” IEEE Trans. Wireless
Commun., vol. 16, no. 3, pp. 1834–1850, Mar. 2017.
[4] G. Interdonato, E. Björnson, H. Q. Ngo, P. Frenger,
0.6
and E. G. Larsson, “Ubiquitous cell-free Massive MIMO
communications,” CoRR, vol. abs/1804.03421, 2018. [Online].
0.4 Available: [Link]
[5] S. Venkatesan, A. Lozano, and R. Valenzuela, “Network MIMO: Over-
coming intercell interference in indoor wireless systems,” in Proc. IEEE
0.2 ACSSC, 2007, pp. 83–87.
[6] G. Caire, S. Ramprashad, and H. Papadopoulos, “Rethinking network
MIMO: Cost of CSIT, performance analysis, and architecture compar-
0 isons,” in Proc. Information Theory and Applications Workshop (ITA),
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 Jan. 2010, pp. 1–10.
[7] W. Feng, Y. Wang, N. Ge, J. Lu, and J. Zhang, “Virtual MIMO in
multi-cell distributed antenna systems: Coordinated transmissions with
Fig. 6. CDF of the per-user min SE provided by uniform power control large-scale CSIT,” IEEE J. Sel. Areas Commun., vol. 31, no. 10, pp.
allocation [3], HUPA [14], and proposed channel-dependent
√ power allocation 2067–2081, Oct. 2013.
√
with f (γmk ) = 1/ γmk , and f (βmk ) = 1/ βmk . [8] D. Gesbert, S. Hanly, H. Huang, S. Shamai, O. Simeone, and W. Yu,
“Multi-cell MIMO cooperative networks: A new look at interference,”
IEEE J. Sel. Areas Commun., vol. 28, no. 9, pp. 1380–1408, Dec. 2010.
[9] M. Boldi, A. Tölli, M. Olsson, E. Hardouin, T. Svensson, F. Boccardi,
conditions rather than uniforming the power allocation. L. Thiele, and V. Jungnickel, “Coordinated multipoint (CoMP) systems,”
in Mobile and Wireless Communications for IMT-Advanced and Beyond,
V. C ONCLUSION A. Osseiran, J. Monserrat, and W. Mohr, Eds. Wiley, 2011, pp. 121–
155.
In cell-free massive MIMO, it is assumed that each UE is [10] P. Marsch, S. Brück, A. Garavaglia, M. Schulist, R. Weber, and A. Deko-
being served by all the APs, which are in turn managed by a rsy, “Clustering,” in Coordinated multi-point in mobile communications:
single CPU. Such a system is unrealistic and unscalable. From theory to practice, P. Marsch and G. Fettweis, Eds. Cambridge,
2011, ch. 7, pp. 139–159.
In this study, we proposed a fully distributed and scalable [11] R. Irmer, H. Droste, P. Marsch, M. Grieger, G. Fettweis, S. Brueck, H.-P.
user-centric architecture for cell-free massive MIMO. The APs Mayer, L. Thiele, and V. Jungnickel, “Coordinated multipoint: Concepts,
are grouped in cell-centric clusters. Each cluster is managed by performance, and field trial results,” IEEE Commun. Mag., vol. 49, no. 2,
pp. 102–111, 2011.
a CPU and operates autonomously. The presence of multiple [12] A. Lozano, R. W. Heath, and J. G. Andrews, “Fundamental limits of
CPUs, each one managing a cell-centric cluster, allows for cooperation,” IEEE Trans. Inf. Theory, vol. 59, no. 9, pp. 5213–5226,
reduced deployment complexity. The UE is then served by Sep. 2013.
[13] T. L. Marzetta, E. G. Larsson, H. Yang, and H. Q. Ngo, Fundamentals
all cell-centric clusters involved in the user-centric cluster. of Massive MIMO. Cambridge University Press, 2016.
This enables a limited distribution of the data payload as [14] E. Nayebi, A. Ashikhmin, T. L. Marzetta, H. Yang, and B. D. Rao,
the data destined to a given UE is only distributed among “Precoding and power optimization in cell-free Massive MIMO sys-
tems,” IEEE Trans. Wireless Commun., vol. 16, no. 7, pp. 4445–4459,
the CPUs of the cell-centric clusters selected. From the AP Jul. 2017.
perspective, computational complexity is reduced when it [15] A. Papadogiannis, D. Gesbert, and E. Hardouin, “A dynamic clustering
comes to channel estimation, calculation of power control co- approach in wireless networks with multi-cell cooperative processing,”
in Proc. IEEE International Conference on Communications (ICC), May
efficients, and precoding/decoding. From the UE perspective, 2008, pp. 4033–4037.
the design increases the SE by leveraging the coordination [16] P. Baracca, F. Boccardi, and V. Braun, “A dynamic joint clustering
among multiple cell-centric clusters. Very few clusters need to scheduling algorithm for downlink CoMP systems with limited CSI,”
in Proc. IEEE International Symposium on Wireless Commun. Systems
cooperate to achieve performance comparable to the canonical (ISWCS), Aug. 2012, pp. 830–834.
(and unscalable) form of cell-free massive MIMO. [17] V. Jungnickel, K. Manolakis, W. Zirwas, B. Panzner, V. Braun, M. Los-
A distributed channel-dependent power control scheme was sow, M. Sternad, R. Apelfrojd, and T. Svensson, “The role of small cells,
coordinated multipoint, and massive MIMO in 5G,” IEEE Commun.
proposed, where the power control coefficients scale pro- Mag., vol. 52, no. 5, pp. 44–51, May 2014.
portionally (at an appropriate rate) with the mean-square of [18] C. Pan, M. Elkashlan, J. Wang, J. Yuan, and L. Hanzo, “User-centric C-
the effective/estimated channel. This policy facilitates fully RAN architecture for ultra-dense 5G networks: Challenges and method-
ologies,” IEEE Commun. Mag., vol. 56, no. 6, pp. 14–20, Jun. 2018.
distributed computation of the power control coefficients. [19] J. Yuan, S. Jin, W. Xu, W. Tan, M. Matthaiou, and K. Wong, “User-
Future research directions include: derivation of a SE centric networking for dense C-RANs: High-SNR capacity analysis and
expression that considers the front/back-hauling overhead; antenna selection,” IEEE Trans. Commun., vol. 65, no. 11, pp. 5067–
5080, Nov. 2017.
performance evaluation under a more realistic channel model [20] E. Björnson and E. Jorswieck, “Optimal resource allocation in coordi-
that includes channel correlations; and extension to the multi- nated multi-cell systems,” Foundations and Trends in Communications
antenna UE case. and Information Theory, vol. 9, no. 2-3, pp. 113–381, 2013.
[21] H. Q. Ngo, L. N. Tran, T. Q. Duong, M. Matthaiou, and E. G. Larsson,
“On the total energy efficiency of cell-free Massive MIMO,” IEEE Trans.
R EFERENCES Green Commun. Netw.,, vol. 2, no. 1, pp. 25–39, Mar. 2018.
[1] S. Zhou, M. Zhao, X. Xu, J. Wang, and Y. Yao, “Distributed wireless
communication system: A new architecture for future public wireless
access,” IEEE Commun. Mag., vol. 41, no. 3, pp. 108–113, Mar. 2003.