Physical Communication 46 (2021) 101341

Contents lists available at ScienceDirect

Physical Communication
journal homepage: www.elsevier.com/locate/phycom

Full length article

Performance comparisons of FDD MIMO and 2.6 GHz TDD massive

MIMO: An experimental analysis
∗
Engin Zeydan a , , Omer Dedeoglu b , Yekta Turk c
a
Centre Technologic de Telecomunicacions de Catalunya, Castelldefels, Barcelona, 08860, Spain
b
Türk Telekomünikasyon A.S. Istanbul, 34889, Turkey
c
Mobile Network Architect Istanbul, 34889, Turkey

article info a b s t r a c t

Article history: Massive multiple-input multiple-output (MIMO) is considered as a breakthrough technology in 5G and
Received 20 October 2020 beyond 5G systems. Some of its main advantages are providing high spectral efficiency to many users
Received in revised form 12 April 2021 simultaneously in the same time–frequency blocks, strong directive signals towards short-range areas
Accepted 12 April 2021
and little interference leaks. However, while massive MIMO exhibits interesting benefits, it is important
Available online 18 April 2021
to investigate its main gains through real deployment scenarios in an operator’s infrastructure. In
Keywords: this paper, we focus on performance comparisons of traditional frequency-division duplex (FDD)-
Massive MIMO based MIMO and 2.6 GHz Time-division duplex (TDD)-based massive MIMO deployments through
Experimental trials experimental analysis under different spectrum and bandwidth in a total of three separate sites and
TDD one co-site in an operational infrastructure of an operator in Turkey. We also provide design guidelines
FDD and requirements for massive MIMO network deployment and proper acceptance of Key Performance
Cellular networks Indicators (KPIs) collection and comparisons criteria. Our experimental results reveal up to 66%, 56%
and 23% performance benefits in terms of downlink (DL) cell throughput of 2.6 Ghz TDD-based massive
MIMO compared to FDD-based MIMO sites in 800 Mhz (site with approximately same number of User
Equipment (UEs) compared with TDD massive MIMO), 1800 Mhz (site with higher number of UEs
compared with TDD massive MIMO) and 2600 Mhz (site with lower number of UEs compared with
TDD massive MIMO) respectively each having 10 Mhz bandwidth. On the other hand, LTE 1800 Mhz
FDD MIMO at 20 Mhz can yield higher user throughput values in comparison to 2.6 GHz TDD-based
massive MIMO at 10 Mhz. We also observed that the maximum paired layer reached 14 layers in DL of
TDD-based massive MIMO. At the end of the paper, we address the main observations and takeaways
of TDD-based massive MIMO deployments.
© 2021 Elsevier B.V. All rights reserved.

1. Introduction of fact, higher frequencies can even increase the interference. On

the other hand, it is known theoretically that the SE and hence
Cellular networks are experiencing exponential traffic growth data rate can grow with the number of antennas and users [2].
recently. According to the Ericsson Mobility Report of June 2020, To achieve uniform coverage inside cellular networks, stronger
mobile traffic is expected to grow by 31 percent annually be- signals with same interference levels at the cell coverage area
tween 2019 and 2025 and most of it will be from video traf- are desired. This can be achieved with beamforming via multiple
fic [1]. Moreover, in the same report 5G networks are expected antennas. To achieve beamforming, the same signal needs to be
to carry nearly half of the world’s mobile data traffic in 2025. transmitted from all antennas by either varying phase/amplitude
Higher traffic brings together ultra dense cellular network de- per antenna or per user.
Massive Multiple Input Multiple Output (MIMO) ensures di-
ployments which also increase the total interference levels in
rected signaling by deploying beamforming antennas rather than
the network. Interference still remains to be the major limitation
traditional antennas to enhance energy efficiency as well as SE.
of improving Spectral Efficiency (SE) in cellular networks. There
The main characteristics of massive MIMO is that the number of
have been many interference management techniques studied
antennas (transceiver chains) in the Base Stations (BSs) is larger
in both academia and industry. To circumvent the interference, than the number of users in the coverage area [3]. Together with
higher frequencies are not the appropriate solution. As a matter massive MIMO, BSs can fully separate the users by providing
favorable propagation environment for separate users [4]. It uses
∗ Corresponding author. antenna arrays that are on a chip for easy beamforming and re-
E-mail address: [email protected] (E. Zeydan). sults in higher received power and spatial multiplexing. Antenna

https://doi.org/10.1016/j.phycom.2021.101341
1874-4907/© 2021 Elsevier B.V. All rights reserved.
E. Zeydan, O. Dedeoglu and Y. Turk Physical Communication 46 (2021) 101341

separations are also small. With few antennas, only broad beams Extensive measurements at 3.5 GHz using a sounding system that
can be achieved, whereas massive number of antennas (e.g. more captures the dynamic channels towards users are also used to
than 200) results in many narrow beams (hence fewer multipath compare different precoding schemes for TDD and FDD massive
components), less leakage in undesired directions (good for urban MIMO systems in [19].
neighborhoods) and main lobe to be focused on desired users. Various deployment options for massive MIMO are also pro-
Together with more antennas, transmit power can be reduced and posed in both academia and industry perspectives [20]. For indoor
more simultaneous users can be multiplexed. massive MIMO deployments, standalone massive MIMO at a sin-
Experimental validations of massive MIMO are important to gle location, distributed MIMO without cooperation, and network
understand the gains in real-world scenarios. In this paper, we MIMO with full cooperation are available in the literature [21].
study the performance comparisons of 2.6 GHz Time-division Traditional massive MIMO BSs does not cooperate. For outdoor,
duplex (TDD)-based massive MIMO scheme with different op- single, cluster or city deployments are possible in practice. In
erating frequency Frequency-division duplex (FDD)-based MIMO cluster deployments, typically a number of sites on the order of
schemes in an operational infrastructure of an operator in Turkey. tens is allocated to a cluster so that managing and tuning the
cluster becomes easier.
On the other hand, in Downlink (DL) MIMO side, The 3rd Gen-
2. Related work and motivation
eration Partnership Project (3GPP) has defined various Transmis-
sion Modes (TMs) to be used in LTE and in 5G with beamforming
The major benefits from massive MIMO deployments has been
options. The white paper in [22] provides a good overview of LTE
shown in capacity and throughput improvements in many of the
TMs and beamforming used in MIMO technologies. As defined in
previous works [3,5]. Thanks to its multiplexing and array gains,
3GPP Release 15 [23,24], TM7 to TM9 are used in 5G NR. Besides,
massive MIMO can also provide high SE (in bits/s/Hz). Massive TM3 to TM6 use codebook based precoding and TM7 to TM9 use
MIMO based 5G networks is becoming a trending area of study non-codebook based precoding in LTE.
in both academic and industry. Massive MIMO is converting into Some recent advances such as cell-free massive MIMO sys-
a practical technology to be used in operators’ network infras- tems have also been proposed as a core architecture of the
tructure after its introduction ten years ago. It has emerged in next-generation mobile communication networks [25–28]. Cell-
many of the products of major vendors such as Nokia’s AirScale, free massive MIMO aims to overcome the physical limitations
Ericsson’s AIR and Huawei’s AAU and is currently being deployed (e.g. inter-cell interference) of traditional cellular networks. It
widely today by various operators (e.g. Sprint, TMO in North is also shown to save more downlink transmission power than
America) across their LTE networks on a large scale. Massive co-located massive MIMO systems [28]. An aerial intelligent re-
MIMO is going to be a key enabler technology of 5G as well. At flecting surface assisted cell-free massive MIMO system to reduce
the same time, the basic principles of massive MIMO operation the so-called “shadow areas” limitations of cell-free massive
is going to be similar to Long Term Evolution (LTE). Additionally, MIMO systems is proposed in [27]. At the same time, there
5G New Radio (NR) is optimized for massive MIMO design with are investigations that aim to realize the implementation of
additional features to support beamforming, advanced high spa- cell-free massive MIMO systems via radio stripes [25]. How-
tial resolution codebooks for support up to 256 antennas, high ever, raising scalability issues due to increased network size
resolution channel state information (CSI)-Reference Signal (RS) brings higher computation complexity and demanding fronthaul
design and reporting mechanisms [6]. requirements. Hence, this problem needs to be handled appropri-
Massive MIMO is also foreseen as a promising enabler technol- ately in real-world implementations of cell-free massive MIMO
ogy for 6G under the name ultra-massive MIMO [7]. The authors systems [26].
in [8] investigate the potential of massive MIMO in green com- Theoretical and simulation results are some of the main works
munications. Large number of measurement campaigns have also that are studied in the literature of massive MIMO. However, as
been held for massive MIMO [9–15]. The authors in [9] provide far as the authors are aware of, no detailed works on real-world
a good overview of recent 5G experimental activities in massive experimental validations for comparisons between TDD-based
MIMO. Some of the massive MIMO experimental campaigns in- massive MIMO and traditional FDD-based MIMO scenarios with
clude measurements on spectrum efficiency and capacity [10], different bandwidth utilizations using an operator’s infrastructure
channel to model channel matrix [11] and its characteristics [12], has been attempted before in the literature. Similar to the ex-
frequency-domain channel sounding [13], spectrum sensing in periments performed in this paper, in our previous study of [5]
we investigated an experimental analysis of a massive MIMO trial
a cognitive radio-like network [14], channel hardening and user
focusing on TDD-based deployment using one of the telecommu-
orthogonality in [15].
nication operator’s infrastructure over a commercial site based in
New trials are beginning to emerge converging 5G with mas-
Turkey. However, the study in [5] did not investigate the effect
sive MIMO technology. In 2019, US carrier Sprint (now T-Mobile)
of different carrier/operating frequencies on the performance of
has completed 5G data usage via its 2.5 GHz massive MIMO
FDD based MIMO in comparison with TDD-based massive MIMO
BS working on 3GPP 5G NR [16]. In 5G trials, Ericsson and T-
and lacks design guidelines and requirements for successful de-
Mobile has demonstrated a 16-layer Multi User (MU)-MIMO trial ployment of massive MIMO. This paper extends the previous
to achieve peak cell throughput of more than 5.6 Gbps on one study by providing a comprehensive performance comparison
channel of 2.5 GHz spectrum and 50 bps/Hz SE with 100 MHz analysis under different operating frequencies under both co-site
of total 5G spectrum [17]. During this joint experiment, mas- and separate FDD based MIMO deployments with different load
sive MIMO with 64 antennas are used that transmit 16 layers characteristics.
(unique data streams each capable of transmitting more than 35 Main Contributions: In this paper, we study TDD-based mas-
Mbps). At the receiving end, eight smartphones that are using sive MIMO deployment trial inside the commercial infrastructure
the same radio resource simultaneously were used with each of a telecommunication operator in Turkey. More specifically,
capable of receiving two data streams, i.e. 2-layers (which totals our focus has been on comparisons of TDD-based massive MIMO
to 700 Mbps per device). In terms of comparisons of FDD and deployment with co-site and separate FDD-based MIMO deploy-
TDD massive MIMO, e.g. based on data obtained in measurement ments under different operating frequencies. Our comparisons
campaigns, the authors in [18] have compared TDD based massive are done via observing several operator specific Key Performance
MIMO with four flavors of FDD beamforming distinguished based Indicators (KPIs). The main contributions of the paper can be
on feedback of CSI using the channel measurements at 2.6 GHz. summarized as follows:
2
E. Zeydan, O. Dedeoglu and Y. Turk Physical Communication 46 (2021) 101341

Fig. 1. An illustration of a massive MIMO system with layer mapping and precoding operations.

• A real world experimental trial on a commercial site in K ) is passed into layer mapping where data is mapped to one
Turkey for monitoring the performance differences of 2.6 or multiple layers. Later, multiple data streams (data stream-1
GHz TDD-based massive MIMO with a total of four different up to data-stream K ) are precoded with CSI and transmitted to
sites FDD-based MIMO deployments that are under different intended UEs simultaneously by the M antennas massive MIMO
operating frequency and bandwidth. BS. At each time instant [t], the transmitter applies a pre-coder
• Revealing up to 66%, 56% and 23% performance benefits in to the KN × 1 data vector s[t ] that is intended for the K UEs.
terms of DL cell throughput of 2.6 GHz TDD-based massive The KN × 1 data vector s[t ] is the stacking of the N × 1 vectors
MIMO compared to FDD-based MIMO sites in 800 MHz sk [t ] = [sk,1 [t ], sk,2 [t ], . . . , sk,N [t ]]T of the K UEs users where each
(site with approximately same number of user equipments entry is the data point from the modulation constellation. During
(UEs) compared with TDD massive MIMO), 1800 MHz (site precoding, a precoding matrix W is applied and the transmitted
with higher number of UEs compared with TDD massive M × 1 data vector x[t ] can be described by
MIMO) and 2600 MHz (site with lower number of UEs
x[t ] = WH s[t ] (1)
compared with TDD massive MIMO) respectively for 10 MHz
bandwidth. On the other hand, LTE 1800 MHz FDD MIMO where the KN × M matrix W contains the parameters of all
at 20 MHz can also yield higher user throughput values channel elements, s[t ] represents KN × 1 data vector intended
in comparison to 2.6 GHz TDD-based massive MIMO at 10 for the K UEs.
MHz. The received signal at kth UE is collected in an N × 1 vector
• Providing design guidelines and requirements for massive rk [t ] = [rk,1 [t ], rk,2 [t ], . . . , rk,N [t ]]T and is given by
MIMO network deployment and proper acceptance of KPIs
rk [t ] = HTk xk [t ] + nk [t ] (2)
collection and comparisons criteria.
• Addressing main takeaways of the experimental results and where xk [t ] is the M × 1 pre-coded data vector, the elements hkm,n
the possible trade-offs of massive MIMO deployments. of the M × N channel matrices Hk are the complex channel gains
from the mth transmitting antenna to the nth receiving antenna
The rest of the paper is organized as follows: In Section 3, we
and N × 1 vector nk ∼ CN (0, σDL 2
I) is the independent additive
detail the system model, concepts, utilized metrics and practical
receiver noise with variance σDL and I is the N × N identity matrix
2
settings and requirements for massive MIMO deployment. In Sec-
for kth UE.
tion 4, we provide our experimental results and point out some
of the main outcomes of the conducted experimental scenarios.
3.1. TDD and FDD MIMO systems
Finally in Section 5 we give the conclusions and future work of
the paper. The transmission in MIMO systems can be done either in TDD
Notation: The mathematical notations X, x and x denote a or FDD mode. In FDD systems, Uplink (UL) and DL use different
generic matrix, vector and scalar respectively, C N ×M denotes the frequency bands but in same time, hence have different CSI in UL
set of complex valued N × M matrices, AH is the conjugate trans- and DL. TDD systems, UL and DL use the same frequency bands
pose of matrix A, CN (x, X ) is the complex Gaussian distribution but at different times. It is generally assumed that the UL and DL
with mean x and covariance matrix X . channels are reciprocal, i.e. they are the same in TDD systems.
In most of the classical wireless systems and deployments,
3. System model and concepts FDD based deployments are operating, whereas TDD-based de-
ployments is beginning to be in use as a candidate system with
We consider the DL of a cellular network as given in Fig. 1. the introduction of massive MIMO systems. FDD based MIMO and
We assume that the access network involves one massive MIMO TDD based massive MIMO systems have their own characteristics,
BSs and each equipped with M antennas. There are K UEs with benefits and bottlenecks. As the number of antennas in BS in-
N antennas where M is much larger than K to leverage the array creases, CSI estimation in FDD systems becomes near impossible,
gain in massive MIMO, which are spatially multiplexed onto the incurring a huge training overhead which reduces the SE of the
same time–frequency resource, i.e. each time–frequency Resource network. To solve for CSI acquisition problem in FDD MIMO
Block (RB) of BS serves the K UEs simultaneously. As the number systems, two main categories based on either coding (e.g. [30]) or
of antennas in the system increases, each UE will benefit from neural network approaches (e.g. [31]) are proposed. However, in
a separate wireless channel with high channel gain and low TDD systems channel estimation overhead is independent of the
interference [29]. For FDD based MIMO systems, we consider the number of BS antennas. In TDD mode, UE can transmit the pilots
DL of 3GPP Rel. 15 system. In our FDD systems, all BS sectors and at the same frequency in both UL and DL directions to estimate
UEs are equipped with MFDD antennas. the channel and later determine the best beam to be selected
Fig. 1 gives an illustration of a typical DL massive MIMO and increase the beam accuracy. Hence, TDD systems become
system in which each user data (user data 1 up to user data more tolerable due to channel reciprocity that requires only UL
3
E. Zeydan, O. Dedeoglu and Y. Turk Physical Communication 46 (2021) 101341

CSI to be estimated. Therefore, TDD-based massive MIMO is more success rate KPI simply measures successful attachment counts of
favorable than FDD-based massive MIMO [5]. UEs into the network during RRC connection request of UEs which
Conventional MIMO systems: In these systems, BSs serve can be formulated as
multiple users and average power per unit service area is dis- # of RRCSetUpSuccess
tributed homogeneously. In these systems both BSs and UEs RRCSetUpSR = × 100% (4)
# of RRCSetUpAttempt
have multiple antennas and antenna ports where multiple data
streams are transmitted simultaneously using the same time/ where RRCSetUpSuccess is RRC connection establishment’s success
frequency resources. For example, in case of 2 × 2 MIMO and count and RRCSetUpAttempt is RRC connection establishments
4 × 4 MIMO, the peak throughput of a single UE can be doubled attempt count. After successful RRC connection, the network goes
and quadrupled respectively. In MU MIMO case, the BS sends from RRC _idle mode to RRC _connected mode. Some possible prac-
multiple data streams for each UEs again utilizing the same tical reasons for observing low RRC setup success rates in a
time–frequency resources. call are related to resource allocation failure (due to UE admis-
TDD-based massive MIMO: Due to channel hardening (be- sion failures) or no response from UE (due to poor coverage or
cause of sufficient randomness and many antennas), almost no terminal problem) [36].
channel quality variations are observed in massive MIMO sys- An E-RAB carries the service data of UEs as an access layer
tems. Scheduling is also seldom needed in case of variations in bearer. E-RAB setup success rate is related to accessibility and E-
user load and is left as optional which means that advanced RAB counter KPI is utilized after successful RRC connection. The
scheduling algorithms may not provide the desired gains [32]. E-RAB success rate depends on successful connections to Core
Cell-edge performance is also improved by a factor of number Network (CN) which can be formulated as
of antennas. On the other hand, pilot contamination (degrading # of ERABSetUpSuccess
system performance), time reciprocity assumption (can poten- ERABSetUpSR = × 100% (5)
# of ERABSetUpAttempt
tially break), increased computational complexity with precoder
design, CSI estimation requirements (in case of large number where ERABSetUpSuccess is successful E-RAB establishments and
of transmit antennas), scheduler design (if the number of users ERABSetUpAttempt is received E-RAB establishment attempts [36].
becomes more than the number of antennas), signal detection
challenges and hardware cost (due to large number of antennas) 3.3. Practical settings and requirements for massive MIMO deploy-
are some examples of bottlenecks for TDD-based massive MIMO ment
implementation [33].
A good comparison of multi-antenna transmission modes us- In general, operators want to deploy massive MIMO for capac-
ing Single User (SU), MU and massive MIMO with FDD and TDD ity improvements over their current LTE infrastructure without
strategies is provided in our previous work [5]. major degradation over their other operating networks. In case
a problem is detected in some problematic sites, a lot of effort
3.2. Utilized metrics and involvement of network experts are required to identify the
reasons for degraded KPIs. Performance improvements of massive
In this subsection, we give the definitions of some of the uti- MIMO is site-dependent that depends on various factors such
lized metrics namely, Radio Resource Control (RRC) setup success as traffic loads and user distribution. In a typical massive MIMO
rate, E-UTRAN Radio Access Bearer (E-RAB) setup success rate, deployment scenario, the number of nodes can vary, e.g. from one
user throughput, cell throughput, Channel Quality Indicator (CQI) site (busy site) to one hundred (in city wide deployments). For
and cell capacity. this reason, enhancing capabilities of different cells based on cell
In closed loop MIMO systems, a UE periodically report back behavior and potential performance expectations are critical. For
large scale deployments, testing no performance degradation to
its dedicated CQI reports that contain information on CQI val-
existing network infrastructure after the introduction of massive
ues, Precoding Matrix Index (PMI) and Rank Indicator (RI). CQI
MIMO can be costly. Multiple tools and drive test results are
value represents the most spectrally efficient Modulation Coding
also necessary for this purpose. Therefore, an integrated approach
Scheme (MCS) that can be supported by the DL channel without
to correlate data from configuration management, performance
exceeding a target Block Error Rate (BLER). The capacity of a
management, fault management, performance management or
cellular network in a given area which is measured in bps/km2
counters is needed so that before and after performance can be
can be calculated as
provided to resolve any issues. For this reason, a clear set of
C = B [MHz] × D [cells/km2 ] × SE [bps/Hz/cell] (3) requirements and acceptance criterias should be set before and
after massive MIMO deployment. Some of those requirements
where B is the available spectrum, D is the average cell density,
that are needed for KPI performance acceptance are:
and SE is the per-cell SE which represents the amount of infor-
mation transferred per second over a unit bandwidth [34]. The 1. Selecting the set and number of performance KPIs to be col-
SE measures simply the bits per Physical Resource Block (PRB) lected, tracked before and after massive MIMO deployment
in Hz and is calculated as UL cell throughput in bits divided by at cell/cluster/network level
number of PRBs used by Physical Uplink Shared Channel (PUSCH) 2. Selecting the time period for KPI data collection before
dedicated radio bearer per msec, RB in Hz and number of UL and after massive MIMO deployments, i.e. for pre-check
antennas [5]. In practical LTE deployment scenarios, signal-to- and post-check respectively. Normally, two weeks for pre-
interference-plus-noise ratio (SINR) values are first mapped into check and ten days for pro-check would be sufficient.
CQI and later into SE using 3GPP specification tables (see Table Moreover, the time duration samples for each KPIs should
7.2.3-1 in [35]). Finally, it loops through MCS indexes to find the be large enough to have enough samples to be analyzed.
best Transport Block Size (TBS)- MCS pair that can approximate At least one hour of data collection per KPI would be
the obtained spectral efficiency and maps an MCS index into a sufficient.
TBS (see Table 7.1.7.1-1 in [35]) during one Transmission Time 3. Deciding on a well-defined schema of the collected KPI data
Interval (TTI). (for both pre and post operations) for persistent storage
DL average user throughput considers user throughput values which will help in easy access and analytics operations at
and DL average cell throughput measures cell capacity. RRC setup later stages.
4
E. Zeydan, O. Dedeoglu and Y. Turk Physical Communication 46 (2021) 101341

Fig. 2. Locations of experimental sites in Turkey with 2 external FDD based MIMO sites (site-1 (800 MHz band), site-2 (1800 MHz band) and site-3 (2600 MHz
band)) all operating in 10 MHz bandwidth and co-sites with TDD-based massive MIMO and FDD based MIMO (800 MHz band and 10 MHz bandwidth, 1800 MHz
band and 20 MHz bandwidth).

4. Aggregating the KPI samples according to user needs, e.g. 4.2. Massive MIMO main KPIs
daily, weekdays only, busy hour, user defined customized,
etc. Fig. 3 shows some of the main KPIs that are observed dur-
5. Displaying the cell/cluster/network level on geographic ing the observation duration of massive MIMO deployment. The
maps and visualizing the collected KPIs as a daily trend investigated KPIs are
visualization so that in any poor performance results, op-
erators can decide the reasoning based on some transient • Access KPIs: RRC set-up success rate (%), E-RAB set-up suc-
or some local event. cess rate (%),
• User throughput KPIs: DL user throughput (Mbps), DL av-
After massive MIMO deployments, pre and post KPI data are
erage paired layers,
compared to observe the main gains of the deployment. Addition-
• Channel Quality: Rank I CQI, average CQI,
ally, post KPI data are compared with the pre-defined KPI targets.
For KPI comparisons, average KPI value of the baseline period can From access KPIs’ perspective, Fig. 3 shows that both RRC
be compared with the average KPI value after massive MIMO de- setup success rate and E-RAB setup success rate values are above
ployments including the standard deviations or confidence values 99.4% and 93% respectively. Hence, we can conclude that RRC and
(if number of samples or days to be included into the analysis for E-RAB setup success rates have been relatively stable after the
more fair comparisons as in [37]). activation of TDD-based massive MIMO. From Fig. 3, we can also
observe that channel quality values are in general good. Average
4. Experimental results
CQI values are between 12 and 14 and rank-1 CQI values are
slightly better and seldom drop below 13. Note that the values for
4.1. Details of the experimental setup
the rank-1 CQI indicate the single CQI feedback values reported
by UEs [38].
We compare the system performance of TDD-based Massive
MIMO scheme with conventional FDD-based SU-MIMO schemes
by means of extensive experimental trials over the commercially 4.3. Massive MIMO traffic
available sites given in Fig. 2 that consists of a total of four
sites with three sectors per site. Fig. 2 shows the locations of Fig. 4 shows the overall user and PS traffic distributions among
experimental three sites with FDD based MIMO (site-1, site-2 co-site and other FDD MIMO sites. First of all, we can observe
and site-3) together with co-sites having both TDD-based massive the period behavior of a number of users and throughput where
MIMO and FDD based MIMO in the middle of the map. TDD- the peak hours are reached around noon time and low periods of
based massive MIMO operates @2.6 GHz with 64T64R, M = traffic are observed in the early morning. The pre-check period.
64 (using 10 MHz bandwidth) and co-site’s FDD-based MIMO i.e. before massive MIMO deployment observation duration, is
operates @800 MHz (using 10 MHz bandwidth) and @1.8 GHz selected to be 5 days (from 14 September 2018 to 18 September
(over 20 MHz bandwidth) with 2T2R MFDD = 2. Site-1 in Fig. 2 op- 2018) and post-check period, i.e. after massive MIMO deployment
erates FDD @800 MHz, site-2 operates FDD @1800 MHz and site-3 observation duration, is selected to be 3 days (from 23 September
operates FDD 2600 Mhzh each with 2T2R and 10 MHz bandwidth. 2018 to 25 September 2018). Each measurement is collected
Table 1 shows the experimental parameters and their corre- and averaged over one hour time intervals. In Fig. 4, other FDD
sponding values used in our TDD massive MIMO experiments. We represents the total number of UEs and total Packet Switched (PS)
used TM9 capable UEs of LTE Release 10 during the experimental traffic in site-1, site-2 and site-3 of Fig. 2. LTE 800 MHz (having 10
tests with massive MIMO. In order to evaluate results better in MHz bandwidth) and LTE 1800 MHz (having 20 MHz bandwidth)
the considered LTE scenarios, we chose similar traffic and CQI cells represent the co-site’s number of UEs and PS traffic values.
condition FDD cells for fair performance comparisons. Note that the number of UEs and PS traffic does not change much
5
E. Zeydan, O. Dedeoglu and Y. Turk Physical Communication 46 (2021) 101341

Table 1
Experimental network parameters and their corresponding values used in TDD-based massive MIMO and FDD-based
MIMO trials.
Parameter Value Parameter Value
RAT LTE TDD (Massive MIMO UE category TM-9 capable UEs
site) and FDD (MIMO of LTE Rel. 10
sites)
Carrier frequency 2575–2615 MHz Receiver noise power −112 dBm
(2.6 GHz B38)
Number of subcarriers 600 Subcarrier bandwidth 15 kHz
Cyclic prefix overhead 6.67% Frame dimensions 10 ms
Tx power 40 W Occupied bandwidth 10 MHz
Antenna config. 2 × 20 MHz 64T64R Transmission scheme OFDM
(massive MIMO)
Antenna height 28 m Mechanical Tilt 0 degrees
Electrical Tilt 1 degrees Azimuth 310 degrees (D 220,
G 340)
Scheduling algorithm Proportional-fair Instantaneous 40 MHz
BandWidth (IBW)
Common Public Radio 2 CPRI Port rate 100 Gbps
Interface (CPRI) port
number
CPRI based topology Trunk Max. distance from BBU 10 km
Dimensions (massive 860 mm × 520 mm × Dimensions (FDD MIMO) 1509 mm ×
MIMO) (H × W × D) 170 mm (H × W × D) 469 mm × 206 mm
Polarization mode +45◦ and –45◦ Gain (massive MiMO) Center downtilt:
(FDD MIMO and TDD 16.3 dBi all
massive MIMO) downtilts: 16.1
±0.7 dBi
Number of supported 64 Capacity (massive max. 3 carriers
RF channels (massive MIMO)
MIMO)
Cross polar isolation ≥ 28 Antenna spacing λ/2
(dB)
Interband isolation ≥ 26 Impedance () 50
(dB)
Gain (FDD MiMO) at 14.0 dBi (800 MHz), 16.7 dBi (1800 MHz), and 17.6 dBi (2600 MHz)
mid Tilt

Fig. 3. TDD-enabled massive MIMO BS KPI values for access, user throughput and channel quality.

over the course of pre-check and post-check periods over other co-site, more LTE users are switched into TDD-based massive
FDD sites. MIMO network. Therefore, we can observe that TDD users mainly
After changing neighbor cell re-selection priority at FDD co- migrated from co-site LTE 1800 MHz cells (namely cell-1 and cell-
site to migrate UEs from FDD to TDD massive MIMO enabled 2) after changing the cell re-selection priority. Fig. 5(a) and (b)
6
E. Zeydan, O. Dedeoglu and Y. Turk Physical Communication 46 (2021) 101341

Fig. 4. Overall performance analysis among all test sites (co-site and site-1, site-2 and site-3) (a) User distribution (b) PS traffic distribution.

show the FDD DL PRB utilization values after re-selection priority. Fig. 7 shows the scatter plot of cell throughput performance
We can observe that FDD DL PRB used ratio has decreased in comparisons between FDD 1800 MHz with 10 MHz bandwidth
both co-sites. Hence, the massive MIMO TDD site has absorbed and 2600 MHz TDD Massive MIMO with 10 MHz bandwidth
the traffic from its FDD LTE 1800 MHz co-site. Additionally, we as the number of UEs is increasing in X -axis of site-2. In this
can also infer that since higher signal strength and coverage are figure, the yellow colored line represents the fitted regression line
provided in the serving area with the TDD-based massive MIMO for TDD-based massive MIMO case whereas the red colored line
network, more UEs previously on cell-edge has more probability represents the fitted regression line for FDD-based MIMO case. In
of being in the coverage of the site now. this case, the number of UEs is close to 300 UEs in FDD scenario
and it is the same distribution as in Fig. 6 for TDD-based massive
4.4. Performance comparisons for different number of UEs MIMO. Fig. 7 shows that TDD-based massive MIMO with 10 MHz
bandwidth yields approximately 56% improvement when number
In this subsection, we compare TDD-based massive MIMO of UEs is around K = 100 compared to FDD-based MIMO in 10
with FDD at 10 MHz bandwidth with the different number of MHz bandwidth. This again validates the clear advantage of TDD-
based massive MIMO compared to FDD-based MIMO under same
UEs in the system in Figs. 6–8. First of all, we can observe from
bandwidth. In comparison with the improvements in Fig. 6, Fig. 7
those figures that in terms of distribution of number of UEs
shows that increased frequency for FDD system has degraded the
in the coverage area, site-1 exhibits almost the same behavior
system performance.
with TDD-based massive MIMO, site-2 has higher number of UEs
Fig. 8 shows the scatter plot results of DL cell throughput per-
and site-3 has less number of UEs in comparisons to TDD-based
formance comparisons FDD 2600 MHz with 10 MHz bandwidth
massive MIMO.
versus 2600 MHz TDD Massive MIMO with 10 MHz bandwidth
Fig. 6 shows the scatter plot for MIMO experimental test re-
in Y -axis and the number of UEs on the site-3’s commercial user
sults comparisons of DL cell throughput of FDD 800 MHz with 10 traffic in X -axis. In this figure, the yellow colored line repre-
MHz bandwidth and 2600 MHz TDD Massive MIMO with 10 MHz sents the fitted regression line for TDD-based massive MIMO case
bandwidth versus the number of UEs on the site-1’s commercial whereas the red colored line represents the fitted regression line
user traffic on X -axis. In this figure, the yellow colored line for FDD-based MIMO case. The number of UEs in FDD scenario
represents the fitted regression line for TDD-based massive MIMO reaches up to 80 UEs whereas it goes up to 120 UEs in TDD cases.
case whereas the red colored line represents the fitted regression We can observe from Fig. 8 that TDD-based massive MIMO with
line for FDD-based MIMO case. The number of UEs in TDD case 10 MHz bandwidth yields approximately 66% improvement when
has exceeded little above 120 UEs whereas it is a little below 120 number of UEs is around K = 60 compared to FDD-based MIMO
UEs in FDD scenario. We can observe from Fig. 6 that TDD-based in 10 MHz bandwidth. This shows again the clear advantage of
massive MIMO with 10 MHz bandwidth yields approximately 23% TDD-based massive MIMO compared to FDD-based MIMO under
improvements when number of UEs is around K = 80 compared the same bandwidth and degradation effect of high frequency in
to FDD-based MIMO in 10 MHz bandwidth. This shows the clear FDD system.
advantage of TDD-based massive MIMO compared to FDD-based Note that when comparing Figs. 6, 7, and 8, we can ob-
MIMO under the same bandwidth. serve that as the number of users increase, sometimes FDD-based
7
E. Zeydan, O. Dedeoglu and Y. Turk Physical Communication 46 (2021) 101341

Fig. 5. PRB utilization percentages of LTE 1800 FDD co-site after massive MIMO activation on (a) Cell-1 (b) Cell-2.

Fig. 6. Cell throughput performance comparisons FDD 800 MHz with 10 MHz bandwidth vs. 2600 MHz TDD Massive MIMO with 10 MHz bandwidth at site-1.

MIMO achieves better cell throughput than TDD Massive MIMO, of UEs are now much closer to BS with LOS and as also pointed
and sometimes it does not. As a matter of fact, cell throughput de- out by the authors in [18], FDD performance depends critically
pends on various factors such as bandwidth, channel quality (Line on the existence of advantageous propagation conditions, namely,
of Sight (LOS), Non-line of Sight (NLOS)). The radio conditions in LOS with high Ricean factors.
those cells differ due to factors such as of the considered UEs with For FDD MIMO systems, in Fig. 6 the cell throughput has an
respect to BSs (good, bad or excellent coverage, cell edge users, increasing trend, whereas in both Figs. 7 and 8 cell through-
users closer to the cell center, etc.) and mobility of users inside put values tend to decrease as the number of users increase.
the cell. User mobility can be a limitation especially for TDD We can also observe that Fig. 8 with 1800 MHz FDD system
Massive MIMO performance due to a limit on the time interval site has a higher number of users compared to others. On the
during which the channel response must be acquired to update other hand, TDD massive MIMO systems tend to exhibit best cell
the precoding matrix. We can also observe that as a general rule, throughput results at an optimal number of users (around 60
lower frequencies (e.g. 800 MHz plot of Fig. 6) can accommodate users) according to fitted regression lines. The results in Figs. 6–
more users than higher order frequencies (e.g. 2600 MHz plot of 8 have also indicated that increased frequency for FDD system
Fig. 8) (although 1800 MHz FDD site of Fig. 7 is observed to be has degraded the system performance. One of the main reasons
deployed in a more densely populated region). This is basically can be due to the real user behavior which is different from a
due to higher coverage areas which increase the possibility of the controlled test environment setup. Under normal circumstances
number of users that are served. We can also observe that the in a controlled experimental environment, average cell through-
number of occurrences of better cell throughput observations of put of high and low operating frequencies are expected to be
FDD-based MIMO compared to TDD Massive MIMO is higher in equal under the same bandwidth constraints and number of
Fig. 8, i.e. at high frequencies. The reason is that higher percentage users. Moreover, in this experimental controlled environment,
8
E. Zeydan, O. Dedeoglu and Y. Turk Physical Communication 46 (2021) 101341

Fig. 7. Cell throughput performance comparisons FDD 1800 MHz with 10 MHz bandwidth vs. 2600 MHz TDD Massive MIMO with 10 MHz bandwidth at site-2.

Fig. 8. Cell throughput performance comparisons FDD 2600 MHz with 10 MHz bandwidth vs. 2600 MHz TDD Massive MIMO with 10 MHz bandwidth at site-3.

each user is assigned to download similar sized files, videos, etc. and 1800 MHz FDD with 20 MHz bandwidth and Fig. 9(b) shows
in a continuous manner, which keeps the total cell throughput the the same comparisons between 2600 GHz TDD massive MIMO
same. However in real network environments hence uncontrolled and 800 MHz FDD 10 MHz bandwidth over download data traffic
environments, this is not the case. User behavior is different for values in X -axis. In both of these figures, green colored line repre-
each UE and it is not guaranteed that each user will exhibit the sents the fitted regression line for TDD-based massive MIMO case
same download/upload behavior compared to controlled envi- whereas yellow colored line represents the fitted regression line
ronments depending on their choice of service/application usage. for FDD-based MIMO case. We can observe that when download
Even under the same number of users in the cell, cell throughput data traffic value reaches 8 GB, average user throughput decline of
can be different and vary as the users can do different access 2600 GHz TDD massive MIMO in comparison with 1800 MHz FDD
activities (e.g. one user with video and another user can be text is approximately 41% for Fig. 9(a) and when download data traffic
messaging) which can also be observed from Fig. 9. The above value reaches to 4 GB, average user throughput improvement
results signify the validity and importance of performing real- of 2600 GHz TDD massive MIMO in comparison with 800 MHz
world experiments as the results may depend on several other
FDD is approximately 315% for Fig. 9(b). In Fig. 9, due to higher
factors and need to be validated via experimental observations
bandwidth utilization of FDD systems (20 MHz), higher DL user
before and after deployments.
throughput values are obtained in Fig. 9(a) compared to Fig. 9(b)
In summary, TDD-based massive MIMO has better average cell
as expected. Moreover, in this higher bandwidth case, FDD-based
throughput than FDD based MIMO scheme in all the considered
MIMO performed better than TDD-based massive MIMO in all
cases above when the user number varies. On the other hand,
download data traffic values.
at high frequencies the performance of FDD MIMO observed
to degrade gradually compared to TDD-based massive MIMO
scheme. 4.6. Mobile pair monitoring

4.5. TDD massive MIMO and FDD MIMO user throughput perfor- Pairing of UEs means to identify UEs that are separate in
mance space from each other such that each one of them is served by
a different grid of beams. In general, the correlation function of
Fig. 9(a) shows the scatter plot results of DL user throughput beamforming weights of different UEs are checked by the sched-
values in Y -axis of co-sites 2.6 GHz TDD-based massive MIMO uler. If the correlation function is close to zero, those UEs are
9
E. Zeydan, O. Dedeoglu and Y. Turk Physical Communication 46 (2021) 101341

Fig. 9. Co-Site’s user throughput versus download data traffic performance comparisons between 2600 TDD massive MIMO and (a) 1800 FDD with 20 MHz bandwidth
(b) 800 FDD 10 MHz bandwidth.

paired together in a set. Later at each TTI, those UEs in the same Figs. 10 and 11 provide values for average number of PRBs
set are scheduled together. During experiments, 16 unique data successfully paired for MU beamforming measured at each layer
streams or layers are transmitted from the massive MIMO capable (from layer-2 to layer-15) with TDD-based massive MIMO. In
BS for geographically scattered UEs to be scheduled together. Fig. 10 the most number of PRB utilization are concentrated at
Therefore at each TTI, 8 paired UEs (each with 2 layers) can be layers 2, 3 and 4. Depending on the time of the observation, the
scheduled together. average number of PRBs that are successfully paired can be up to
During our experiments in the DL side, we have observed 17 in the first layer, followed by 11 in layer 2 and 7 in the third
above described mobile pairing opportunities. In DL, the maxi- layer. We can also observe that the peak values for the number of
mum paired layer reached 14 layers. At each layer, it is observed pairings have occurred around noon over the observation periods
that different average number PRBs are shared among layers
(from 23 September 2020 to 25 September 2020). The remaining
(in which multiple UEs are simultaneously scheduled together)
layers for the pairings of PRBs given in Fig. 11 are observed to be
during experiments. Layer-1 is particularly used to measure the
lower than 3. These results indicate that the number of shared
number of PRBs that can be paired for MU beamforming and
PRBs depends on the geographic dispersion of UEs over the cell
the remaining ones are used to determine PRB pairing for MU
beamforming. The remaining layers are used to determine the area of massive MIMO.
average number of PRBs successfully paired for MU beamform-
ing. The samples of the measurements for both the number of 4.7. Main observations and takeaways
PRBs that can be paired for MU beamforming at layer 1 and the
average number of PRBs successfully paired for MU beamforming As the network becomes larger and more complicated (with
at other layers are obtained per second in DL. Later, the average of many users, antennas and dense cells), better network planning
those samples are used to calculate the corresponding percentage (i.e. selections of neighbors, locations, etc.), coverage, resource,
values. traffic (or capacity) and interference management schemes are
10
E. Zeydan, O. Dedeoglu and Y. Turk Physical Communication 46 (2021) 101341

Fig. 10. DL average pair layer PRB success rates (a) Pair layers 2 to 4 (b) Pair layers 5 to 8.

Fig. 11. DL average pair layer PRB success rates (a) Pair layers 9 to 12 (db) Pair layers 13 to 15.

required. For this reason, tracking the best optimization parame- improving RRC success rate, flexible and dynamic switching op-
ters and configuration details before and after activation 2.6 GHz tions (hard or easy handover) between TMs (e.g. SU-MIMO or
massive MIMO) are all other important key performance param-
TDD-based massive MIMO is important. Deciding on the percent- eters that need to be optimized depending on the considered use
age of UEs to migrate from LTE FDD MIMO to TDD massive MIMO, case or scenario (e.g. big event or urban deployments, etc.).
11
E. Zeydan, O. Dedeoglu and Y. Turk Physical Communication 46 (2021) 101341

One of the main reasons for major improvement of massive In this paper, we investigated the details of a massive MIMO
MIMO is also due to using TM9 capable devices. These devices deployment scenario. First, we provided design guidelines and
are capable of communicating with both SU and massive MIMO requirements for massive MIMO network deployment, proper
transmission modes. TM9-capable devices allow BSs to direct acceptance of KPIs collection and comparisons criteria and gave
user dedicated beams towards UE with more accurate CSI mea- an overview of the experimental set-up. Second, we compared
surements. This has improved the cell-edge conditions as beam- TDD massive MIMO with FDD MIMO systems in various oper-
forming allowed to increase the UEs’ received signal. Massive ating frequencies and different bandwidth for both co-site and
MIMO TDD system used in the experiment can also dynamically separate site scenarios using an operator’s network infrastructure
switch between TM9 and TM4 modes (supports closed loop mul- in Turkey. We have observed that TDD massive MIMO site’s user
tiplexing) without special signaling by higher layers. Based on throughput is observed to be higher than co-site LTE 800M site
our experimental observations, we also observed that massive and the capacity to be higher than same bandwidth other FDD
MIMO’s horizontal beamforming may need to be adjusted to be sites. It is also observed that most of the user traffic is transferred
same with the FDD sector to increase the cell throughput for from LTE 1800 MHz co-site after changing the neighbor cell
co-site TDD massive MIMO and FDD MIMO deployments. re-selection priority to activate massive MIMO. The maximum
During our experiments, the maximum paired layer reached paired layer is observed to reach 14 layers in DL. Finally, we
14 layers in DL using massive MIMO. Under normal circum- provided some important discussions on the outcome of the
stances, the number of UEs and the cell throughput have a di- experimental setup and the involved trade-offs.
rect impact on the mobile pairing opportunities. Pairing can free As a future direction of our work, more research is needed
up more PRB resources and Physical Downlink Shared Channel to introduce new FDD systems into massive MIMO systems and
(PDSCH) layers. Therefore, any UEs can benefit from maximum observe their experimental benefits in terms of channel gains,
resources from the scheduler. From Fig. 10, we observed that capacity or received power as well as the core KPIs that have
most of the time only the first three layers had high pairing been investigated within this paper. Cell-free massive MIMO,
opportunities with low PRB pairings in the remaining layers. which is proposed to overcome the physical limits of cellular net-
This signifies that the UE traffic was somehow observed to be works, is another future direction that demands further research
bursty and with a low payload profile. Therefore based on the implementation and real-world experimentation. Additionally,
low number of layers, we can conclude that few occasions for future experiments can be designed to compare the performances
pairings have emerged during the observation period. In case of brought by array gain via increasing the number of BS antennas
existence of high traffic volume, scheduling of UEs will be done versus the one brought by using more spectrum bandwidth.
by the TDD-based massive MIMO deployment system considering
them MU candidates. Hence, massive MIMO is very good for cell CRediT authorship contribution statement
throughput and capacity if the cell has high traffic. However, if the
traffic volume is low, scheduling will be done as SU candidates in Engin Zeydan: Supervision, Conceptualization, Methodology,
a given TTI cycle. Based on the low observed pairing opportunities Writing - original draft, Writing - review & editing. Omer
over time during our experiments, we can conclude that the total Dedeoglu: Data curation, Methodology, Validation. Yekta Turk:
amount of traffic generated by UEs and observed through cell Investigation, Visualization, Writing - original draft.
throughput were also not high. Additionally, as the pairing oppor-
tunities increase, more UEs can be scheduled together at a given Declaration of competing interest
TTI using the same PRBs. This means that the amount of user
throughput obtained by each user can be smaller in comparison The authors declare that they have no known competing finan-
to full TTI usage by a single UE that can utilize all PRB. During cial interests or personal relationships that could have appeared
experiments, Fig. 3 has demonstrated the inverse correlation to influence the work reported in this paper.
between user throughout and number of DL average paired layers
over the observation duration validating this expected behavior. Acknowledgments
At the same time, based on the user throughput scatter plot
of Fig. 9, we have observed that LTE 1800 MHz FDD MIMO at This work was partially funded by Spanish MINECO, Spain
20 MHz can yield higher user throughput values in comparison grant TEC2017-88373-R (5G-REFINE) and by Generalitat de
to 2.6 GHz TDD-based massive MIMO at 10 MHz. This result may Catalunya, Spain grant 2017 SGR 1195.
indicate two main take-away conclusions: The first one is that the
effect of higher bandwidth becomes a more dominant factor in References
increasing the user throughput than massive MIMO. The second is
that the user throughput has also its own limitations due to poor [1] Ericsson, Ericsson mobility report (june 2020) (white paper), 2019, [Online;
RF conditions even though the advantages brought by massive accessed August-2020], https://bit.ly/34MyE9y.
MIMO features can be in-place (e.g. high number of free PRBs [2] T.L. Marzetta, Fundamentals of massive MIMO, Cambridge University Press,
2016.
reserved for UEs) and may not help much under those poor RF
[3] T.L. Marzetta, Noncooperative cellular wireless with unlimited numbers
circumstances. Pilot contamination phenomenon exists in multi- of base station antennas, IEEE Trans. Wirel. Commun. 9 (11) (2010)
cellular system and is a major challenge for the massive MIMO 3590–3600.
system implementation. During our experiments, we have not [4] J. Hoydis, C. Hoek, T. Wild, S. ten Brink, Channel measurements for
been able to directly observe the effect of pilot contamination in large antenna arrays, in: 2012 International Symposium on Wireless
Communication Systems (ISWCS), IEEE, 2012, pp. 811–815.
dense user scenarios due to presence of single cell massive MIMO
[5] E. Zeydan, O. Dedeoglu, Y. Turk, Experimental evaluations of TDD-based
system. massive MIMO deployment for mobile network operators, IEEE Access 8
(2020) 33202–33214.
5. Conclusions and future work [6] Qualcomm, Designing 5G NR, 2018, [Online; accessed October-2020],
https://www.qualcomm.com/media/documents/files/the-3gpp-release-15-
5g-nr-design.pdf.
Massive MIMO is a novel and promising technology that is ex- [7] N. Rajatheva, I. Atzeni, E. Bjornson, A. Bourdoux, S. Buzzi, J.-B. Dore, S.
pected to be a key enabler for 5G and beyond 5G wireless cellular Erkucuk, M. Fuentes, K. Guan, Y. Hu, et al., White paper on broadband
networks by providing high throughput and reliable transmission. connectivity in 6G, 2020, arXiv preprint arXiv:2004.14247.

12
E. Zeydan, O. Dedeoglu and Y. Turk Physical Communication 46 (2021) 101341

[8] S.N. Sur, Massive MIMO for green communication, in: Advances in Greener [31] C. Qing, B. Cai, Q. Yang, J. Wang, C. Huang, Deep learning for CSI feedback
Energy Technologies, Springer, 2020, pp. 833–842. based on superimposed coding, IEEE Access 7 (2019) 93723–93733.
[9] Y. Okumura, S. Suyama, J. Mashino, K. Muraoka, Recent activities of [32] S. Dierks, W. Zirwas, M. Jäger, B. Panzner, G. Kramer, Mimo and massive
5G experimental trials on massive mimo technologies and 5G system MIMO—Analysis for a local area scenario, in: 2015 23rd European Signal
trials toward new services creation, IEICE Trans. Commun. 102 (8) (2019) Processing Conference (EUSIPCO), IEEE, 2015, pp. 2451–2455.
1352–1362. [33] D.C. Araújo, T. Maksymyuk, A.L. de Almeida, T. Maciel, J.C. Mota, M. Jo,
[10] P. Harris, S. Zhang, M. Beach, E. Mellios, A. Nix, S. Armour, A. Doufexi, K. Massive MIMO: survey and future research topics, Iet Commun. 10 (15)
Nieman, N. Kundargi, LOS Throughput measurements in real-time with a (2016) 1938–1946.
128-antenna massive MIMO testbed, in: 2016 IEEE Global Communications [34] E. Björnson, J. Hoydis, L. Sanguinetti, et al., Massive MIMO networks:
Conference (GLOBECOM), IEEE, 2016, pp. 1–7. Spectral, energy, and hardware efficiency, Found. Trends Signal Process.
[11] J. She, W.-J. Lu, Y. Liu, P.-F. Cui, H.-B. Zhu, An experimental massive 11 (3–4) (2017) 154–655.
MIMO channel matrix model for hand-held scenarios, IEEE Access 7 (2019) [35] 3GPP TS 36.213 version 12.3.0. 3rd Generation Partnership Project Tech-
33881–33887. nical Specification, LTE evolved universal terrestrial radio access (e-UTRA)
[12] L. Valle, J.R. Pérez, R.P. Torres, Characterisation of indoor massive MIMO physical layer procedures (release 12), 2014.
channels using ray-tracing: A case study in the 3.2–4.0 GHz 5G band, [36] Y. Türk, E. Zeydan, On performance analysis of multi-operator RAN sharing
Electronics 9 (8) (2020) 1250. for mobile network operators, Turkish J. Electr. Eng. Comput. Sci. 29 (2)
[13] E.P. Simon, J. Farah, P. Laly, Performance evaluation of massive MIMO (2021).
with beamforming and nonorthogonal multiple access based on practical [37] A. Yildirim, E. Zeydan, I.O. Yigit, A statistical comparative performance
channel measurements, IEEE Antennas Wirel. Propag. Lett. 18 (6) (2019) analysis of mobile network operators, Wirel. Netw. 26 (2) (2020)
1263–1267. 1105–1124.
[14] I. Dey, P.S. Rossi, M.M. Butt, N. Marchetti, Experimental analysis of [38] 3G.P.P. T.S. 38.214 V16.4.0, NR; physical layer procedures for data (release
wideband spectrum sensing networks using massive MIMO testbed, IEEE 16, 2020, [Online; accessed 7-Feb.-2020], https://www.3gpp.org/ftp//Specs/
Trans. Commun. (2020). archive/38_series/38.214/.
[15] A.O. Martinez, J.Ø. Nielsen, E. De Carvalho, P. Popovski, An experimental
study of massive MIMO properties in 5G scenarios, IEEE Trans. Antennas
and Propagation 66 (12) (2018) 7206–7215.
[16] R. Chataut, R. Akl, Massive MIMO systems for 5G and beyond networks—
Engin Zeydan received his Ph.D. degree in February
Overview, recent trends, challenges, and future research direction, Sensors
2011 from the Department of Electrical and Computer
20 (10) (2020) 2753.
Engineering at Stevens Institute of Technology, Hobo-
[17] Inc. T-Mobile US, Buckle up! T-mobile achieves mind-blowing 5G speeds
ken, NJ, USA. Previously, he received his M.Sc. and B.Sc.
with MU-MIMO, 2020, [Online; accessed Sept-2020], https://t-mo.co/
degrees from the Department of Electrical and Elec-
3kgG6OV.
tronics Engineering at Middle East Technical University,
[18] J. Flordelis, et al., Massive MIMO performance—TDD versus FDD: What
Ankara, Turkey, in 2006 and 2004, respectively. Dr.
do measurements say?, IEEE Trans. Wireless Commun. 17 (4) (2018)
Zeydan has worked as an R&D engineer for Avea, a
2247–2261.
mobile operator in Turkey, between 2011 and 2016. He
[19] J.Ø. Nielsen, A. Karstensen, P.C. Eggers, E. De Carvalho, G. Steinböck, M.
was with Turk Telekom Labs working as a Senior R&D
Alm, Precoding for TDD and FDD in measured massive MIMO channels,
Engineer between 2016 and 2018. He was also a part-
IEEE Access 8 (2020) 193644–193654.
time instructor at Electrical and Electronics Engineering department of Ozyegin
[20] E. Björnson, L. Sanguinetti, H. Wymeersch, J. Hoydis, T.L. Marzetta, Massive
University between 2015 to 2018. He is currently with the Communication
MIMO is a reality—What is next?: Five promising research directions for
Networks Division of the Centre Tecnològic de Telecomunicacions de Catalunya
antenna arrays, Digit. Signal Process. 94 (2019) 3–20.
(CTTC) working as a Senior Researcher. He received the Best Paper Award from
[21] B. Panzner, W. Zirwas, S. Dierks, M. Lauridsen, P. Mogensen, K. Pajukoski,
the Network of Future Conference in 2017. His research interests are in the
D. Miao, Deployment and implementation strategies for massive MIMO
areas of telecommunications and big data networking.
in 5g, in: 2014 IEEE Globecom Workshops (GC Wkshps), IEEE, 2014, pp.
346–351.
[22] Rohde, Schwarz, LTE transmission modes and beamforming (white paper),
2015, [Online; accessed October-2020], https://www.rohde-schwarz.com/
ae/file/1MA186_2e_LTE_TMs_and_beamforming.pdf. Omer Dedeoglu received his B.S. degree in Electri-
[23] T.S. 38.101-1 v17.0.0, 5G NR; user equipment (UE) radio transmission cal&Electronics Engineering from Bilkent University in
and reception; part 1: Range 1 standalone, 2020, [Online; accessed 2001 and M.S. degree in Electrical&Computer Engineer-
5-Feb.-2020], https://portal.3gpp.org/desktopmodules/Specifications/ ing from New Mexico University in 2003. He worked
SpecificationDetails.aspx?specificationId=3283. for R&D projects and made Radio NW investment
[24] T.S. 38.101-2 v17.0.0, 5G NR; user equipment (UE) radio trans- plans at Turkcell for about 6 years. Since 2011, Omer
mission and reception; part 2: Range 2 standalone, 2020, [Online; Dedeoglu has been working at Turk Telekom as radio
accessed 7-Feb.-2020], https://www.3gpp.org/ftp//Specs/archive/38_series/ network planning expert& manager.
38.101-2/.
[25] E. Ernfors, Radio stripes: re-thinking mobile networks, 2019, [Online;
accessed February-2021], https://www.ericsson.com/en/blog/2019/2/radio-
stripes.
[26] E. Björnson, L. Sanguinetti, Scalable cell-free massive MIMO systems, IEEE Yekta Turk received his Ph.D. degree in 2018 from
Trans. Commun. 68 (7) (2020) 4247–4261. the Department of Computer Engineering at Maltepe
[27] T. Zhou, K. Xu, X. Xia, W. Xie, J. Xu, Achievable rate optimization for aerial University, Istanbul, Turkey Previously, he received
intelligent reflecting surface-aided cell-free massive MIMO system, IEEE his M.Sc. degree from in Telecommunications and
Access (2020). Computer Networks from the George Washington Uni-
[28] J. Qiu, K. Xu, X. Xia, Z. Shen, W. Xie, Downlink power optimization for versity, DC, USA, in 2007 and B.Sc. degree in Electrics
cell-free massive MIMO over spatially correlated Rayleigh fading channels, and Electronics Engineering from Anadolu University,
IEEE Access 8 (2020) 56214–56227. Turkey, in 2005. He has worked in fixed and mobile
[29] H.H. Yang, Massive MIMO meets small cell, Springer, 2017. network operators for 11 years. He is a mobile network
[30] J. Choi, D.J. Love, T. Kim, Trellis-extended codebooks and successive phase architect based in Istanbul, Turkey. His research inter-
adjustment: A path from LTE-advanced to FDD massive MIMO systems, ests are in the area of mobile radio telecommunications
IEEE Trans. Wireless Commun. 14 (4) (2014) 2007–2016. and computer networks.

13