1

The Road to 6G: Ten Physical Layer Challenges for

Communications Engineers
Michail Matthaiou, Okan Yurduseven, Hien Quoc Ngo, David Morales-Jimenez,
Simon L. Cotton, and Vincent F. Fusco
arXiv:2004.07130v2 [eess.SP] 28 Sep 2020

Abstract—While the deployment of 5G cellular systems cell-free massive MIMO. Yet, transforming these
will continue well into the next decade, much interest speculative academic concepts into commercially
is already being generated towards technologies that will viable solutions is a very challenging exercise. Very
underlie its successor, 6G. Undeniably, 5G will have trans-
recently, some related articles have elaborated on:
formative impact on the way we live and communicate,
yet, it is still far away from supporting the Internet-of- 6G driving applications, metrics and new service
Everything (IoE), where upwards of a million devices per classes [1], the importance and challenges of the
km3 (both terrestrial and aerial) will require ubiquitous, THz-based communications [2], 6G requirements
reliable, low-latency connectivity. This article looks at some and overview of supporting technologies [3], and,
of the fundamental problems that pertain to key physical finally, potential use cases and 6G-enabling network
layer enablers for 6G. This includes highlighting challenges
architectures [4].
related to intelligent reflecting surfaces, cell-free massive
MIMO and THz communications. Our analysis covers On the other hand, our article moves away from
theoretical modeling challenges, hardware implementation the state-of-the-art and makes the following contri-
issues and scalability, among others. The article concludes butions: (i) we investigate the realizable potential of
by delineating the critical role of signal processing in the the three above mentioned scientific pillars; (ii) we
new era for wireless communications.
identify ten immediate engineering challenges that
I. I NTRODUCTION need to be addressed at the physical layer to boost
follow-up research in the 6G ecosystem. We point
After nearly eight years of intensive academic
out that the timescales of these different challenges
research and industrial testing on 5G, the lessons we
vary depending on advances in other fields (e.g.
have taken are the following: a) 5G can indeed sup-
electronics, CMOS design); (c) we finally articulate
port emerging data-hungry applications (e.g. ultra-
the critical role of fundamental signal processing
fast broadband, high-definition video streaming),
(SP) in the 6G era along with the associated chal-
mainly through advances in the massive MIMO
lenges. The vision of this article is to open up oppor-
(mMIMO) space; b) 5G is still falling short of sup-
tunities in electromagnetic theory, communication
porting the so-called Internet-of-Everything (IoE),
theory and transceiver design.
where myriads of devices in a geographic cube
require either low-latency, ultra-reliable connectivity II. I NTELLIGENT R EFLECTING S URFACES (IRS)
or wireless Gpbs Internet access by availing of the The antenna architecture in a wireless communi-
mm-wave/THz spectrum [1], [2]. In the eve of a new cation system can be considered as the key front-
decade, the idea of 6G has tentatively started to cir- end component, and will be the enabling physical
culate within the wireless community, and the con- technology for 6G networks. In the realization of
sensus is that 6G will try to address the shortcom- the physical layer to facilitate the 6G technology,
ings of 5G through three scientific pillars by boldly understanding the design challenges of the antenna
(1) pushing the communication to higher frequency structure is crucial to truly appreciate the potential
bands (mm-wave and THz), (2) creating smart radio of the next generation communication networks. In
environments through reconfigurable surfaces and view of this, in this section, we aim to explain
(3) by removing the conventional cell structures, aka some of the key design challenges of IRS and show
that the radiation characteristics of these apertures
The authors are with the Institute of Electronics, Communications
and Information Technology (ECIT), Queen’s University, Belfast, vary substantially as a result of these challenges.
Belfast BT3 9DT, U.K. (e-mail: [email protected]). A reflective surface is a planar aperture synthesized
 2

using an array of sub-wavelength elements (or unit phase range of 0-2π radians is achieved, the number
cells). Due to their sub-wavelength unit cell sam- of quantization levels used to discretize this phase
pling, reflective surfaces can be considered a distinct range has a direct impact on the fidelity of the syn-
form of metasurfaces. There has been a substantial thesized wavefront upon reflection from the reflec-
amount of research conducted in EM wave control tive surface. To investigate the effect of these design
using metasurface apertures with applications rang- constraints, we consider a reflective metasurface and
ing from imaging to EM invisibility [5]. Despite study these cases individually. First, we assume that
the fact that these structures are well understood the unit cells can alter the phase response of the in-
within the applied EM community, their adoption in coming reference-wave across the full phase range,
wireless communication networks has been under- 0-2π radians. Note that many examples appear in
investigated to date. the literature for unit cell topologies that can achieve
Wireless systems conventionally rely on a mature phase shifts of up to 2π, and the approaches used
antenna technology to establish a communication are now relatively mature, e.g. [6]. For this scenario,
link, which is known as phased arrays. A phased in Fig. 1, we investigate two different quantization
array consists of individual antennas with dedicated levels, 1-bit and 4-bit. We consider that the reflective
phased shifting circuits and power amplifiers to metasurface is illuminated by an arbitrarily selected
synthesize the desired aperture wavefront. Phased plane-wave incident along optical axis (z-axis) of
arrays can be power hungry and exhibit a rather the surface. The objective function is defined to be a
complex hardware architecture. Different from the wavefront radiated in the broadside direction (θ=0◦ ,
phased array technology, metasurfaces rely on a φ=0◦ ) upon reflection from the surface.
holographic principle to achieve the desired phase
modulation. The incoming wave illuminating the
aperture surface acts as a reference-wave, which
is converted to a desired wavefront upon reflec-
tion from the reflective metasurface aperture. Their
major advantage is that they can synthesize any
arbitrary waveform using this simple, yet strong,
holographic principle without the need for expensive
and power hungry phase shifters. Designing an IRS
for wireless systems exhibits two main challenges:
In Challenge 1, going beyond the conventional
aperture level discussion, we consider the design
challenges of the IRS considering their building
blocks; i.e. the unit cells, and the effect of aberration
on a unit cell level on the overall performance of
the IRS. Satisfying the conditions of challenge 1, in
challenge 2, we cover the dynamic reconfiguration
aspect - a necessity for the ”intelligent” operation,
and review potential techniques to dynamically re-
configure the phase response of the unit cells.
Challenge 1: Unit cell phase range and phase
quantization levels Fig. 1: 3D Radiation pattern of the reflective surface
An important limitation in the design process of as a function of phase quantization level (a) 1-bit (b)
a reflective surface is the achievable phase range of 4-bit (c) comparison of the radiation patterns along
the unit cells synthesizing the aperture. Ideally, each the azimuth plane (H-plane) and elevation plane (E-
unit cell across the reflective surface should provide plane). Colorbar in dB scale.
a full phase control across a phase range of 0-2π
radians. Another important constraint in the design In Fig. 1, the H-plane is the azimuth plane (or
process of a reflective surface is the quantization xz-plane) with φ=0◦ . Similarly, the E-plane is the
of the phase range of the unit cells. Even if a full elevation plane (or yz-plane) with φ=90◦ . Although,
 3

from a design perspective, using a 1-bit quantization the 0-π radians phase range can be achieved using
level would be rather simple, in comparison to the a single resonant element, going beyond this phase
4-bit quantization case, the radiation pattern exhibits range, such as the studied 0-2π radians scenario,
substantially higher sidelobes (8 dB higher) and less requires the combination of multiple independent
directivity (21 dB lower in the broadside direction). resonances. As a result, careful consideration should
Increasing sidelobes is a significant concern for be given to the achievable phase range from the
wireless communication networks due to the pos- unit cells to find the optimum trade-off between
sibility to introduce strong interference across other the complexity of the unit cell structure and the
channels. Similarly, reduced directivity adversely desired radiation characteristics of IRS apertures for
affects the link budget. As a result, from Fig. 1, it 6G networks. From Figs. 1 and 2, it is evident that
is evident that there is a direct relationship between to have a realistic estimate of the link budget in a
the phase quantization level and the fidelity of the communication channel and a better understanding
radiation pattern of the reflective surface. This is of interference characteristics, it is important that
an important outcome because whereas the 1-bit the design limitations of reflective surfaces are taken
quantization level can be achieved using a simple into account. Yet, it is often assumed that the
ON/OFF type binary modulation mechanism, phase reflective surface is ideal, suggesting that the link
quantization levels greater than that, such as the 4- budget calculations do not consider these unit cell
bit scenario, require more sophisticated techniques based aberrations.
to modulate the phase response of the unit cells,
such as gray scale phase modulation [5]. This can
Challenge 2: Dynamic reconfigurability and IRS
substantially increase the complexity of the unit
cell design, and hence, a careful trade-off study is
needed to achieve the desired hardware complexity Albeit producing highly desirable radiation char-
and radiation performance from such apertures for acteristics, conventional reflective surfaces are
6G networks. We now study the effect of the achiev- static, hence, beam characteristics are hard-coded
able unit cell phase range on the radiation pattern into the surface during the design process. Dif-
of the reflective surface. We select the quantization ferent from static metasurfaces, an IRS has the
level to be maximum (4-bit) and investigate two unit capability to dynamically tune the reflection re-
cell phase ranges, 0-π and 0-2π, respectively. The sponse of the aperture in an all-electronic manner.
radiation pattern of the reflectarray surface vs the This is particularly important as communication
achievable unit cell phase range is shown in Fig. 2. environments have dynamic characteristics, namely
variations in the number of connected users and
non-static location distribution over time. Thus, the
capability to intelligently change the characteristics
of the reflected wavefront to meet their dynamic
metrics plays a crucial role in future wireless com-
munication systems. The dynamic modulation of
the IRS can be achieved using several techniques,
such as leveraging materials with variable electrical
properties, i.e. liquid crystals, or loading the unit
cells with low-power semiconductor elements, e.g.,
PIN diodes and varactors [5], [7]. Challenges 1 and
2 are a direct consequence of an electromagnetic
Fig. 2: Radiation pattern of the reflective surface as engineering problem. While both challenges can
a function of unit cell phase range. be addressed using rather complex, multi-layer, re-
configurable unit cell architectures [5], engineering
From Fig. 2, limiting the unit cell phase range to such unit cells to exhibit low form-factor, loss,
0-π reduces the directivity of the surface by 7.3 dB cost, system complexity and power consumption is
while increasing the sidelobe levels by as much as 7 a prerequisite in order for this technology to be a
dB. This is an important outcome because whereas feasible option for future 6G networks.
 4

III. C ELL - FREE MASSIVE MIMO -1 4.5

Downlink Rate
Cell-free mMIMO has been proposed in [8] to 4
overcome the boundary effect of cellular networks. -0.5
3.5
In cell-free mMIMO, many access points (APs)

Position (km)
distributed in a geographic coverage area coherently 0
3

serve many users in the same time-frequency re- 2.5

sources. There are no cells, and hence, no boundary 2

0.5
effects. Key points of cell-free mMIMO:
1.5
• Cell-free mMIMO relies on mMIMO technol-
1
ogy. More precisely, using many APs, cell-free -1 -0.5 0 0.5 1
Position (km)
mMIMO offers many degrees of freedom, high
(a) Cell-free mMIMO
multiplexing gain, and high array gain. As a
result, it can provide huge energy efficiency -1
Downlink Rate
and spectral efficiency with simple SP. 6

• In cell-free mMIMO, the service APs are dis- -0.5 5

tributed over the whole network, and hence,

Position (km)
4
we can obtain macro-diversity gains. Thus, 0
cell-free mMIMO can offer a very good net- 3

work connectivity. There is no dead zone. 0.5 2

Figures 3a–3b show the downlink achievable
1
rates displayed with scaled colors for cell-free
1
mMIMO and colocated mMIMO, respectively. -1 -0.5 0 0.5 1
Position (km)
Clearly, cell-free mMIMO can offer much more
uniform connectivity for all users. (b) Colocated mMIMO
• Different from colocated mMIMO, where the Fig. 3: The downlink achievable rates for different
base station is equipped with very large anten- user locations displayed with scaled colors (obtained
nas, in cell-free mMIMO each AP has a few using simulation approach described in [8]); (a): 100
antennas. Thus, cell-free mMIMO is expected APs are uniformly located at random in a 2km×2km
to be built by low-cost, low-power components area; (b): all 100 service antennas are located at the
and simple SP APs. original point (i.e. the center of the square area).
The above benefits (in particular the high network
connectivity) fulfil the main requirements of future
wireless networks. Therefore, cell-free mMIMO has that it is not implementable when the network size
become one of the promising technologies of be- (number of APs and/or number of users) grows
yond 5G and towards 6G wireless networks, and at- large. Designing a scalable structure is one of the
tracted a lot of research interest [9]. Designing a low main challenges of cell-free mMIMO. It is shown
cost and scalable system is the ultimate objective in [8] that, owing to the path loss, only 10-20%
of cell-free mMIMO research. To do this, we need of the total number of APs really participate in
scalable transmission protocols and power control serving a given user. Thus, each user should be
techniques, which are discussed in Challenges 3 and served by a subset (not all) of APs. There are two
4. In addition, it is important to have new SP designs ways to implement this: network-centric and user-
which can be implemented in a distribute manner centric approaches. In the network-centric approach,
to improve the system performance, scalability and the APs are divided into disjoint clusters. The APs
robustness. This is discussed in Challenge 5. in a cluster coherently serve the users in their
Challenge 3: Practical user-centric approaches joint coverage area. Network-centric-based systems
In canonical cell-free mMIMO [8], all APs par- still have boundaries, and hence, are not suitable
ticipate in serving all users through the backhaul for cell-free mMIMO. By contrast, in user-centric
connections with one or several central processing approach, each user is served by its selected subset
units (CPUs). This is not scalable in the sense of APs. There are no boundaries, and hence, user-
 5

centric approach is a suitable way to implement cell- good performance and can be implemented in a
free mMIMO. There are several simple methods distributed manner. Otherwise, the system will not
to implement user-centric approach, such as each be scalable. In canonical cell-free mMIMO [8], con-
user chooses some of its closest APs or chooses a jugate beamforming is normally considered since
subset of APs which contribute most of the total it can be implemented in distributed manner and
received power of the desired signal [10]. Yet, performs well. Yet, compared to other linear pro-
existing methods are not optimal, still require huge cessing schemes, such as zero-forcing and minimum
connections from all APs to the CPUs, and are mean-square error, the performance of conjugate
still fully network-controlled. In addition, the cluster beamforming is far below. To cover the gap between
formed by each user changes quickly depending conjugate beamforming and ZF/MMSE, we need
on the user locations. This requires more control to have additionally very large number of service
signaling. Hence, designing a practical user-centric antennas. Cell-free mMIMO with local ZF was
approach is a challenging research exercise. proposed in [12]. However, this scheme requires that
Challenge 4: Scalable power control each AP has a large number of antennas. This is
Power control is central in cell-free mMIMO more challenging for the uplink design. Currently,
since it controls the near-far effects and the interuser there are no distributed SP schemes available for the
interference to optimize the objectives (e.g. the max- uplink. Even with the simple matched filtering, we
min fairness or the total energy efficiency) we want need to send the (processed) signals from each AP
to achieve. Ideally, power control is done at the to the CPUs for signal detection.
CPU under the assumption that the CPU perfectly
IV. M OVING TO HIGHER FREQUENCY BANDS
knows all large-scale fading coefficients. Then, the
optimal power control coefficients will be sent to 6G wireless systems will rely on: i) Millimeter-
the APs (for the downlink transmission) and to the wave technologies (30 to 300 GHz); ii) THz tech-
users (for the uplink transmission). This requires nologies (300 GHz to 3 THz); and iii) Free space
huge front/back-hauling overhead. Yet, it is very optics (FSO) [2]. It is known that the exacerbated
difficult for the CPU to have perfect knowledge atmospheric attenuation and path loss at higher
of large-scale fading coefficients associated with a frequencies, can be compensated by miniaturized
potentially unprecedented number of APs and users. massive antenna arrays that can support super sharp
Thus, besides the unscalability of the current canon- beamforming. Also, there are natural synergies be-
ical transmission protocol (discussed in Challenge tween IRS/cell-free mMIMO and higher frequency
3), the above power control method also creates systems to extend the operating range and com-
another issue which makes the system unscalable. bat the fundamental distance problem. We now
Thus, power control should be done distributed at overview three challenges towards this objective.
the APs with local knowledge of the channel condi- Challenge 6: Packaging/interconnect techniques
tions. This is again problematic because it is hard to In addition to providing for a physical enclosure,
control the near-far effects and interuser interference packaging must provide a reliable interconnection
without full channel knowledge of all links from between interior and exterior operating environ-
all APs and users. Some heuristic power control ments. Some of the key driving factors include
schemes have been proposed, which, however, are integration of high-speed semiconductor integrated
developed based on a specific assumption of the circuits with advanced antenna systems and inte-
propagation environment, and hence, it is hard to gration with optoelectronics. At higher frequencies,
evaluate how well these schemes work in practice bond wires cause considerable signal degradation.
[9]. Promising approaches based on machine learn- The effects of bond wires are difficult to charac-
ing (ML) and deep learning (DL) have recently terize for large signal applications, such as power
been proposed [11]. A key question is whether these amplifiers and also for phase critical applications,
approaches are also scalable, in order to meet the such as beamformers for phased arrays, where these
foreseeable decentralization of cell-free mMIMO. can introduce side lobe levels. Traditional metallic
Challenge 5: Advanced distributed SP split-block packages provide excellent performance
One of the ultimate aims of cell-free mMIMO but are bulky and heavy. Noticeable progress has
research is designing a SP scheme which offers been made in interconnect and packaging technolo-
 6

gies for THz applications. Cutting-edge techniques thing connectivity, will produce high-dimensional
in micro-machining and LTCC technology yield and increasingly complex signals, which are subject
compact and low-cost solutions. Additive manufac- to increased interference and other impairments
turing techniques, such as metal coated 3D print- (e.g., related to synchronization, temporal corre-
ing of plastic devices, can realize low-cost, light lation etc.) that have so far been largely over-
weight and compact devices. Ceramic packages for looked. Current SP methods, generally based on
both low- and high-power devices are commercially low-dimensional signals and classic stationarity as-
available for applications up to, 50 GHz. sumptions, will need to be rethought. We now dis-
Challenge 7: Transceiver design cuss two of these methods: channel estimation and
The compact physical size and power efficiency adaptive filtering, and their associated challenges.
requirements become more challenging at higher Challenge 9: Channel estimation
frequencies. Hybrid beamforming will be best suited Extensive research in the context of 5G has
to implement large number of antenna elements been on how to reduce training overheads in pilot-
along with high efficiency amplifiers. Performance based channel estimation. While such need will be
parameters, such as the noise figure, output power most critical in 6G, viable solutions become ex-
and power efficiency degrade significantly at high tremely challenging due to the massive scale-up and
frequencies. Demodulation of higher order mod- connectivity demands, in particular: (i) supporting
ulated signals also becomes more challenging as high data rates (Gbps) in high-mobility scenarios—
phase noise increases at higher frequencies. Ad- a prime concern of operators—will require dealing
vanced array SP techniques must complement the with much shorter channel coherence times; (ii)
transceiver design to address these challenges. ultra-low latency requirements will see transmis-
Novel techniques, such as spatially oversampled sion intervals substantially shortened, and (iii) the
antennas and new phased array architectures can be number of parameters to estimate will be massively
leveraged to provide a solution to size, weight and large as a consequence of the scaling (not only
power consumption of large mm-wave/THz antenna of antennas/APs, but also of users/devices). The
arrays. MMIC beamformers and compact millimeter high-dimensional channels will need to be esti-
wave wireless communications modules incorporat- mated in severe under-sampling constraints, which
ing high gain phased antenna arrays that are capable might render pilot-based (coherent) estimation un-
of achieving high-speed data transfer at frequencies feasible, particularly under high-mobility or low-
up to 60 GHz are presently commercially available. latency requirements. Blind (non-coherent) estima-
Challenge 8: Measurements & standardization tion approaches—which do not require dedicated pi-
Two popular approaches for phase-sensitive mea- lot signals—stand as promising alternatives. Though
surements at THz use either vector network ana- significant efforts have been made in this direction,
lyzers (VNAs) or time domain spectrometers [13]. existing approaches require knowledge of the (high-
dimensional) signal covariance matrix [14], which
In both cases, calibration, verification, and measure-
ment traceability at THz frequency bands remains will again need to be acquired from a limited num-
a major challenge. For time domain systems, a ber of samples. To that end, fundamental research in
major challenge is the establishment of standardized the fields of random matrix theory (RMT) and high-
measurement and calibration, whereas for VNA dimensional statistics will be crucial; in particular,
systems, solutions are being sought for high pre- to develop accurate estimators of large covariance
cision waveguides and interconnects. Electro-optic matrices (and their eigen-spectrum) under limited
sampling is promising as a complementary approach sampling. More recent approaches based on ML
to THz measurements, though it is yet to extend the might also play a relevant role [11].
bandwidth to 1.5 THz and to improve resolution. Challenge 10: Adaptive filtering
In beamforming, transmitted signals are dynam-
V. T HE ROLE OF SP IN THE 6G E RA ically adapted (via digital precoding) to the propa-
The journey towards 6G will inevitably be hur- gation conditions, effectively mitigating interference
dled by significant challenges in the SP arena. Mas- and noise. Adaptive beamforming can be seen as
sively populated and decentralized (cell-free) net- a linear filter with a particular design objective,
works, supporting unprecedented Internet of Every- e.g., to maximize the signal-to-noise ratio (SNR).
 7

Optimal solutions for the beamformer (and associ- [1] W. Saad, M. Bennis, and M. Chen, “A vision of 6G wireless
ated receiver filters) require the covariance matrix systems: Applications, trends, technologies, and open research
problems,” IEEE Netw., vol. 34, no. 3, pp. 134–142, 2020.
of the aggregated interference and noise; unknown [2] T. S. Rappaport et al., “Wireless communications and applica-
in practice, this needs to be estimated from ob- tions above 100 GHz: Opportunities and challenges for 6G and
served samples. Current solutions rely on classical beyond,” IEEE Access, vol. 7, pp. 78729–78757, 2019.
[3] P. Yang et al., “6G wireless communications: Vision and po-
estimators, such as the sample covariance matrix tential techniques,” IEEE Netw., vol. 33, no. 4, pp. 70–75, Jul.
(SCM), which will return a poor estimate in high- 2019.
dimensional 6G scenarios, due to: [4] M. Giordani et al., “Toward 6G networks: Use cases and
technologies,” IEEE Commun. Mag., vol. 58, no. 3, Mar. 2020.
• Scarcity of samples: While the numbers of an-
[5] H. T. Chen, A. J. Taylor, and N. Yu, “A review of metasur-
tennas and user devices will scale up massively, faces: Physics and applications,” Reports on Progress in Physics,
strict low-latency and high-mobility require- vol. 79, No. 7, 076401, 2016.
[6] F. Venneri et al, “An improved synthesis algorithm for reflectar-
ments will impose a rather limited number of rays design,” IEEE Antennas Wireless Propag. Lett, vol. 4. pp.
training (observed) samples. 258–261, Aug. 2005.
• Temporal correlation: The ultra-dense and [7] O. Yurduseven et al., “Dynamically reconfigurable holographic
metasurface aperture for a Mills-Cross monochromatic mi-
highly decentralized deployments, e.g., with crowave camera,” Optics Express, vol. 26, no. 5, 2018.
thousands of distributed APs in cell-free net- [8] H. Q. Ngo et al., “Cell-free massive MIMO versus small cells,”
works, will be subject to non-perfect synchro- IEEE Trans. Wireless Commun., vol. 16, no. 3, Mar. 2017.
[9] G. Interdonato et al., “Ubiquitous cell-free massive MIMO
nization (e.g., between interference and desired communications,” EURASIP J. Wireless Commun. Netw., vol.
signals) and non-stationarity effects. 2019, no. 1, p. 197, 2019.
• Outlying samples: With millions of intercon- [10] H. Q. Ngo et al., “On the total energy efficiency of cell-free
massive MIMO,” IEEE Trans. Green Commun. Netw., vol. 2,
nected devices (from electrical/smart appli- no. 1, pp. 25–39, 2017.
ances to connected vehicles), we expect multi- [11] M. Chen et al., “Artificial neural networks-based machine learn-
ple sources of impulsive noise and, in security- ing for wireless networks: A tutorial,” IEEE Commun. Surveys
sensitive applications, eventual sources of in- Tuts., vol. 21, no. 4, pp. 3039–3071, 2019.
[12] G. Interdonato et al., “Local partial zero-forcing precoding for
tentional interference (jamming). cell-free massive MIMO,” IEEE Trans. Wireless Commun., vol.
Traditional estimators (e.g., SCM) rely on the suffi- 19, no. 7, pp. 4758–4774, July 2020.
cient availability of samples (the number of samples [13] M. Naftaly et al., “Metrology state-of-the-art and challenges in
broadband phase-sensitive terahertz measurements,” Proc. IEEE,
should be far greater than the number of signals) vol. 105, no. 6, pp. 1151–1165, Jun. 2017.
and on the assumption that these samples are i.i.d. [14] H. Q. Ngo and E. G. Larsson, “EVD-based channel estimation
Under the conditions above, however, the mismatch in multicell multiuser MIMO systems with very large antenna
arrays,” in Proc. IEEE ICASSP, pp. 3249–3252, 2012.
between true and estimated covariance leads to [15] N. Auguin et al., “Large-dimensional behavior of regularized
highly inaccurate filters with severe performance Maronnas M-estimators of covariance matrices,” IEEE Trans.
losses, in terms of connectivity, reliability, and data Signal Process., vol. 66, no. 13, pp. 3529–3542, Jul. 2018.
rates. A fundamental challenge is then to develop AUTHORS
filtering solutions which are robust to the effects Michail Matthaiou is a Professor at Queen’s University
mentioned above. While research to that end has so Belfast (QUB), UK. His research spans signal processing,
far been very scarce, promising tools and directions massive MIMO, and mm-wave systems.
can be leveraged from the fields of robust statistics, Okan Yurduseven is currently a Senior Lecturer at QUB.
RMT, and high-dimensional covariance estimation His research interests include microwave and mm-wave imag-
(see, e.g. [15] and references therein). ing, MIMO radar, antennas and metamaterials.
Hien Quoc Ngo is a Lecturer at QUB. His main research
VI. C ONCLUSION interests include massive MIMO, cell-free massive MIMO,
Although the 6G era is a decade away, it is physical layer security, and cooperative communications.
extremely timely to understand what are the main David Morales-Jimenez is a Lecturer at QUB. His research
interests span statistical signal processing, random matrix
challenges for communications engineers. We iden-
theory, and high-dimensional statistics.
tified ten immediate challenges whose investigation Simon L. Cotton is a Professor of Wireless Communica-
will cross-leverage expertise in SP, information the- tions at QUB. Among his research interests are propagation
ory, electromagnetics and physical implementation. measurements and statistical channel modeling.
Vincent F. Fusco is the Chair of High Frequency Electronic
R EFERENCES Engineering at QUB. His research interests include active
antenna front-end techniques and self-tracking antenna arrays.