Received May 26, 2021, accepted June 21, 2021, date of publication June 29, 2021, date of current

version July 8, 2021.

Digital Object Identifier 10.1109/ACCESS.2021.3093460

5G K-SimSys for Open/Modular/Flexible

System-Level Simulation: Overview and Its
Application to Evaluation of 5G Massive MIMO
JAEWON LEE 1 , MINSIG HAN1 , MINJOONG RIM 2,

AND CHUNG G. KANG 1 , (Senior Member, IEEE)

1 School of Electrical Engineering, Korea University, Seoul 02841, Republic of Korea
2 Information and Communication Engineering, Dongguk University, Seoul 04620, Republic of Korea
Corresponding author: Chung G. Kang ([email protected])
This work was supported in part by ‘the Cross-Ministry Giga KOREA Project’ grant funded by the Korea government (Ministry of
Science and ICT) (No. GK18S0400, Research and Development of Open 5G Reference Model) and in part by the Brain Korea
21 Plus Project in 2020.

ABSTRACT 5G K-SimSys is a system-level simulator which has been designed and implemented to
provide an open platform and a tractable testbed for evaluating the system-level performance of 5G standard.
In this paper, we present its design overview with the overall modular structure, including the functionality
of the flexible modules. Meanwhile, massive multi-input multi-output (MIMO) is a key technology for
improving the spectral efficiency of 5G systems. While massive MIMO has been standardized in 3GPP
Rel-13 (LTE-Advanced Pro), the next generation of massive MIMO standard is now available in Rel-16,
a.k.a New Radio (NR) interface, with further improvement by introducing more antenna elements over
the higher frequency band. In particular, the beam-based air interface for above 6 GHz band involves
various antenna configurations and feedback schemes, requiring a more complex testbed for system-level
performance evaluation. This paper examines the multi-antenna technologies in 3GPP NR specification
to develop its system-level model in 5G K-SimSys in order to test the factors affecting performance
in the corresponding environments through various embodiments. In the simulation results, the baseline
performance for massive MIMO in NR is evaluated by changing the number of antenna ports and the number
of spatial multiplexing layers in different experimental environments. Particularly, the effect of vertical
beamforming on interference is evaluated for Full Dimension MIMO (FD-MIMO). In order to demonstrate
that 5G K-SimSys is an easy-to-design simulator that facilitates modification and reconfiguration, owing to
its modularized and customized structure, we consider the proposed bandwise analog beamforming (BAB)
scheme. The modular and flexible structure immediately allows implementation of virtual modules for
bandwise analog beamforming by reusing the elementary modules as hierarchically designed simulator
objects.

INDEX TERMS 3GPP, 5G, K-SimSys, beamforming, massive MIMO, new radio, system-level simulation.

I. INTRODUCTION the communication system. Therefore, simulations such as

Performance evaluation of a cellular system involves simu- channel coding, MIMO processing, multicarrier modulation,
lations at different levels, e.g., link-level simulation (LLS) channel estimation, and equalization are performed using
for block error rate (BLER) and system-level simula- fixed-point point-to-point channels [6], e.g., cluster delay
tion (SLS) for average system throughput, under various line and tapped delay line channel models, without dealing
environments modeled by channel measurements [1]–[5]. with any high-level resource allocation for packet scheduling
Of these, LLS focuses only on the physical (PHY) layer of among the multiple user equipments (UEs). LLS could
also be implemented for multi-point communication evalu-
The associate editor coordinating the review of this manuscript and ation; however, its computational complexity becomes pro-
approving it for publication was Li Zhang. hibitively high as the considered number of points increases.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 9, 2021 94017
 J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

In contrast, SLS evaluates the system-level performance for to the additional objects. In other words, it would take
the complex multi-user communication scenarios, on both enormous efforts to integrate a new object with different
PHY and medium access control (MAC) layers. PHY technical features into the existing structure. Meanwhile,
layer performance is abstracted by link-to-system mapping the one in [13] is developed with open-source ns-3 simulator,
for reducing its computational complexity, which allows integrating a full-stack mmWave module that can assess the
prediction of BLER performance based upon the received end-to-end performance at the PHY layer and across all
signal-to-noise and interference ratio (SINR) in any channel higher layers of the communication protocol stack for the
condition. Thus, LLS and SLS play a complementary role 5G network. As it incurs inherent complexity of processing
in calculating the SINR for the system-level performance, MAC and TCP layers in SLS, only a limited number of
as specified by simulation methodology for evaluation of base and mobile stations can be considered. Therefore,
IMT-2020 specification in the International Telecommunica- dynamic conditions such as inter-cell interference cannot
tion Union [7]. be immediately captured for more practical system-level
5G mobile communication systems consider three service performance assessment of multi-antenna technologies in
scenarios, including eMBB, mMTC, and URLLC with cellular mobile systems.
eight key performance indicators (KPIs), besides various The earlier design principles of 5G K-SimSys and its
performance requirements for support services [7], [8]. embodiments have been presented with a brief overview
Also, the extension of the spectrum to the millimeter-wave in [10] along with its application to evaluation of
(mmWave) bands led to a new system design that supports ultra-reliable & low latency communication in [11] and of
analog beamforming to warrant sufficient coverage, while massive MIMO [12]. Our objective in the current paper
conforming to the performance requirement of lower fre- is to present how our SLS has been applied to evaluate
quency bands. As different types of network configurations the performance of 5G massive MIMO, while providing
and channel models in different frequency bands must be con- a comprehensive overview of 5G K-SimSys. This is a
sidered, tremendous efforts are required for an individual or complete and self-contained version of presentation in the
a development group to implement an SLS by its own efforts. sense that it integrates new features and concepts while
Therefore, there is an on-going demand for an open-source aggregating the key elements from recent developments.
system-level simulator that can develop and verify various To this end, we present a complete review of 3GPP NR
5G technologies, which creates a common platform to allows specification on massive MIMO, which was fully modeled
anyone to participate in the implementation and evaluation. for the current implementation. Furthermore, we propose a
However, the existing few open-source SLSs are not flexible new beamforming scheme, referred to as bandwise analog
enough to accommodate various operational features and beamforming (BAB), which is an innovative idea of handling
new evaluation scenarios without a significant amount of the frequency selectivity issue to improve the performance of
reprogramming. Therefore, 5G K-SimSys was developed as massive MIMO over the wideband operation. It will serve as
an SLS framework in a flexible, open, and modular (FOM) an exemplary case that demonstrates how 5G K-SimSys can
form [9]–[12]. be employed to evaluate performance.
To begin with, 5G K-SimSys is flexible as it can support Massive MIMO is widely recognized as a key technology
various types of 5G environments and service scenarios. for improving the spectral efficiency in 5G [8], [14]–[17].
As an open-source software, this flexibility is further enriched It is usually associated with a digital array of a large
by the participation of many researchers and developers. number (hundreds) of antenna elements serving much fewer
Second, 5G K-SimSys can be considered as one of the easy- UEs. With a large number of antennas, the channel vectors
to-design simulators due to its modularized and customiz- of multiple users tend to be nearly orthogonal with each
able structure. It can facilitate simulator modification and other and thus, the effect of multi-user interference and
reconfiguration while integrating various algorithms under thermal noise is largely reduced, allowing simple transmit
development in any evolving stages and unforeseen scenarios. or receive techniques. Three-dimensional (3D) massive
This modular feature has been realized by building the MIMO/Full Dimension MIMO (FD-MIMO) configurations
elementary modules that can be reproduced or reused as have been introduced in 3GPP by deploying active antenna
hierarchically designed simulator objects. elements in a two dimensional (2D) antenna array, which
5G K-SimSys is intended to be compatible with guidelines can dynamically control the beam direction toward a desired
for evaluating radio interface technologies for IMT-2020 [7], user [18]–[21]. In fact, digital arrays with a large number
which must be implemented for the evaluation of candidate of antenna elements are made feasible owing to advances
technology submissions and for deciding on the qualifying in active array antenna technology. Performance evaluation
RIT/SRIT technologies for 5G. It can be distinguished of massive MIMO systems requires intensive scheduling and
from the existing simulators [3], [13] by several factors. interference calculations which involves complicated system
The Vienna system-level simulator in [3] is developed in models, including accurate 3D channel model, antenna con-
MATLAB code. Even if it provides the key modules for figurations, feedback information, and multi-user scheduling
SLS as an open sources, its flexibility is rather limited. The algorithm. Meanwhile, massive MIMO also refers to analog
rigid modular structure is difficult to modify and extend beamforming with many antenna elements at mmWave band

94018 VOLUME 9, 2021

J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

which can fulfill the performance requirements for 5G, large number of antenna elements were reduced a smaller
providing sufficient coverage against higher transmission number of reference signals to consider only for an effective
loss. channel. Further, a transceiver unit (TXRU) has been
The aforementioned technical features imply that a sophis- introduced as a special interface for the active antenna
ticated means of evaluating the system-level performance is system (AAS) to control the gain and phase of the individual
required for the massive MIMO-based 5G system. In fact, antenna elements, by combining the features of a passive
enormous efforts must be made to implement a system-level antenna element array with an active TXRU array. In addition,
simulator for verifying the system efficiency or developing a it involves various types of antenna configurations, e.g., 2D
specific algorithm for massive MIMO, subject to the various planar, to reduce the hardware complexity associated with a
environments and system configurations. In this paper, large number of antenna elements.
we examine the 5G multi-antenna technologies in 3GPP Meanwhile, NR aims at further improvements to increase
NR specification, so as to develop the system-level model the capacity with a major enhancement in a new flexible,
for 5G K-SimSys. One of our objectives is to discuss modular, and scalable CSI framework [32]. It also includes a
how new modules and other software structures can be high-resolution CSI reporting mode to improve MU-MIMO
implemented with 5G K-SimSys. Furthermore, we evaluate operation. Furthermore, FD-MIMO/massive MIMO is con-
the performance of BAB, which requires a much more sidered an essential element of the systems operating in the
complex structure for system-level evaluation. Through these above 6 GHz frequency range, e.g., mmWave band, since the
various embodiments, we demonstrate the flexibility and high free-space path and penetration loss at those frequencies
usefulness of 5G K-SimSys and how it can be employed for require large array gains to warrant sufficient signal-to-noise
the advanced system design and optimization. ratio (SNR), i.e., for sufficient coverage [23]. However,
The remainder of this paper is organized as follows. large antenna arrays increase the system complexity with
Section II presents the massive MIMO system model, conventional full digital beamforming having each antenna
which includes analog beamforming and single/multi-panel element connected to a separate radio frequency (RF)
digital beamforming based on the 3GPP NR specification. chain [24], [25]. In order to reduce its complexity, there
Section III presents actual implementation of 5G K-SimSys have been efforts to find the optimal hybrid beamforming
as a system-level simulator for massive MIMO, along structure under various constraints for a given antenna
with its overall platform architecture that aims to meet configuration [26]–[30]. Furthermore, more flexible CSI
the open/modular/flexible design. Then, baseline simulation acquisition and high-resolution feedback are specified in
results are presented for various experiments. Section IV NR, along with a dual-stage codebook where a precoder is
introduces the notion of bandwise analog beamforming and factorized into two components, one to capture the wideband
then, presents its implementation result, which demonstrates channel properties and the other to capture the frequency
the modular and flexible features as a specific use case of 5G selective part. This section presents an overview of massive
K-SimSys. Finally, conclusions are made in the last section. MIMO specifications and their system models, focusing on
the NR-specific features and functionalities as a basis for
II. GUIDELINES FOR MASSIVE MIMO: SYSTEM MODEL implementing 5G K-SimSys.
FOR 5G K-SimSys
Since multi-antenna techniques have been successfully intro- A. ANTENNA CONFIGURATION
duced for improving spectral efficiency in LTE, they still play In a digital array architecture, the antenna elements are
an essential role in 5G system design. In particular, mmWave arranged into 1-D linear array or 2-D planar array. Each
band in 3GPP NR specification involves a beam-centric antenna element is equipped with its own RF chain and
design to overcome higher transmission losses while pro- analog-to-digital (AD) converter. Under a typical line-
viding sufficient coverage. To this end, analog beamforming of-sight (LoS) channel in the mmWave band, many degrees
is supported both at the gNB and UE, specifying beam of freedom may not be essential due to the fact that the
management procedures. Furthermore, beamforming is also benefits of frequency-selective digital precoding over analog
supported for initial access and broadcast signals in NR. beamforming are negligible. Therefore, a fully analog array
FD-MIMO/massive MIMO is also a beneficial technology architecture would be more acceptable for the LoS-dominant
at below 6 GHz frequencies for improving spectral efficiency. environments, e.g., in a small cell. In fact, to trade off the
The underlying multi-antenna features developed in the later advantages and disadvantages of the fully digital and analog
releases of LTE have been enhanced to further improve array architectures, a hybrid architecture has been considered
spectral efficiency for higher data rates and capacity over earlier in the LTE specification. Inheriting the key structure of
the limited spectrum. In [22], standardized FD-MIMO was MIMO transmission over digital antenna ports, the standard
reported for implementing elevation beamforming, and was has been extended to a hybrid beamforming structure with
later developed into an essential element of LTE-Advanced multi-panel arrays in NR. The structure establishes and
Pro (3GPP Rel-14), for 3D beamforming, i.e., beamforming maintains beam pairs for analog beamforming in both the BS
in both horizontal and vertical dimensions. Both uplink and the UE for the mmWave band. As illustrated in Fig. 1,
feedback and downlink pilot overheads associated with a a 2D planar array can be divided into the rectangular

VOLUME 9, 2021 94019

 J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

subarrays, referred to as panels, each of which is a uniform system under consideration, we set the cross-polarization
planar array. The panel arrays allow a partially connected angles to (0◦ , 90◦ ) and ±45◦ in the UE and gNB, respectively.
hybrid architecture, which facilitates a simpler approach
of designing the beamformers in an independent manner,
i.e., one for each panel. Then, this particular architecture B. HYBRID BEAMFORMING FOR 2D AAS
overcomes the difficulty of jointly designing the analog and A baseband precoder processes the output of the layer map-
digital beamformers, even if the channel at antenna element per, which will be distributed to antenna ports. An antenna
level cannot be independently considered due to the analog port is defined as a logical element to which a reference
beamforming [22], [32]. signal is assigned for channel state information, channel
estimation for demodulation, sounding signal, etc, in the
3GPP specification. A single physical antenna can belong
to multiple antenna ports, whose effective channels are
different, e.g., two signals on the same physical antenna
carrying different beamforming weights. In the specification,
there are various types of reference signals, e.g., a channel
state information reference signal (CSI-RS), demodulation
reference signal (DMRS), phase tracking reference signal
(PTRS), and sounding reference signal (SRS), that are the
transmitted reference signals assigned to antenna ports.
Hence, the antenna port can be defined by the transmitted
FIGURE 1. Configuration for panel antenna arrays [6].
reference signal. Meanwhile, a TXRU is a special interface
for AAS, having power amplifier (PA) and low noise
We first consider a single panel array antenna in Fig. 1. amplifier (LNA), so that gNB can control the gain and
It is an M × N uniform planar array with N antennas in phase of the individual antenna element. It establishes a
each row and a total number of M rows in the vertical hardware connection between the baseband signal path and
dimension with equal spacing. A larger planar array can be antenna array elements via a radio distribution network,
constructed with the multiple panels. Fig. 1 illustrates the which delivers the transmit signal from PA to antenna
antenna system that consist of Mg vertical and Ng horizontal elements, and receive signals from antenna elements to
antenna panels. Depending on vertical and horizontal panel LNA. As opposed to a fully digital array which has an
spacing, the panel structures can be of two different types: RF chain and a data converter for each antenna element,
a uniform antenna array and a non-uniform antenna array. it combines large-dimensional analog pre/postprocessing
Let dH and dV denote the distances between horizontal with lower-dimensional digital processing and thus, a much
and vertical antenna elements in a panel, respectively. lower number of active antenna elements are employed which
Furthermore, let dg,H and dg,V denote the distance between reduces the power consumption and hardware cost [20].
horizontal and vertical panels, respectively. In the uniform When implementing a 2D AAS in an FD-MIMO architecture,
antenna array, the antenna elements of the same polarization the radio resource comprises of antenna ports, TXRUs, and
are uniformly distributed across the whole panel array physical antenna elements. A data stream on an antenna
such that there is no explicit spacing to differentiate the port is precoded on TXRUs in the digital domain, which
adjacent antenna
elements from different antenna panels, is a process referred to as antenna port virtualization
e.g., dg,H , dg,V = ((M − 1)dH , (N − 1)dV ) in Fig. 1. (port-to-TXRUs mapping). Considering one-to-one mapping
On the contrary, in the non-uniform antenna array, the space between ports and TXRUs allows for both terms to be used
between adjacent antenna elements in adjacent panels
is interchangeably.
different from that within the panels, e.g., dg,H , dg,V  The signal on a TXRU is precoded by phase shifts or
((M − 1)dH , (N − 1)dV ) in Fig. 1 [33]. variable gain amplifiers on a group of physical antenna
The above antenna configuration arranges the transmit elements in the analog domain. This process, referred to as
signals as virtual beams to support the MU-MIMO in 3D, TXRU virtualization, defines a mapping of TXRUs signals
i.e., serving multiple users on the same bandwidth simultane- to those of the antenna elements. In other words, TXRU
ously in full dimension.

The antenna configuration is denoted virtualization allows for a group of antenna elements to
by M , N , P, Mg , Ng , where P is the number of polarization be configured for transmitting or receiving the same data.
dimensions, e.g., P = 2 for cross-polarization (X-pol). Then, From the end-to-end design point, in the transmitter, hybrid
the total number of antenna elements is given by MNPMg Ng . beamforming can be modeled as antenna port and TXRU
Even though a uniform linear antenna is more appropriate virtualizations to deal with analog and digital domain signals,
for MU-MIMO than P = 2, subject to the same number of respectively. In other words, it combines digital precoding of
antenna elements [34], X-pol allows for enhanced capacity the transmission layers into antenna ports and then analog
with more antenna elements within the limited physical precoding of the antenna ports into physical antennas. This
space, as typically practiced in 5G systems. In the current process requires the channel estimation only for each antenna

94020 VOLUME 9, 2021

J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

FIGURE 2. CSI-RS transmission type for hybrid beamforming: Illustrative structures (P = 1).

port, rather than channel estimation for all MNPMg Ng as follows:

physical antenna elements.
y = HW(A) W(D) x + n. (1)
The downlink transmit signals isPrepresented  x =
by
[x1 , x2 , · · · xK ]T with covariance of = E xxH , where

where W(A) and W(D) represent analog and digital precoding
xi denotes the transmit signal for UE i, for corresponding matrices, respectively. Meanwhile, n denotes the Gaussian
modulation symbol, e.g., QPSK, 16-QAM, or 64-QAM, noise, including inter-cell interference with an average power
depending on modulation and coding set (MCS). Let H of σn2 in a multi-cell scenario.
denote a channel matrix of a rank of K , with each element Depending on the configuration of the antenna ports
to represent a complex channel gain between each Tx and Rx of the physical antenna elements, there are two different
antennas. In the hybrid beamforming structure, the received TXRU structures for hybrid beamforming. One is an array
signals, denoted by K -dimensional vector y, for the K partitioning structure, in which the antenna elements are
scheduled UEs, each with a sinlge antenna element are given divided into multiple groups. Each group of antenna elements

VOLUME 9, 2021 94021

 J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

is connected to one TXRU while orthogonal CSI-RSs specification are presented in the following subsection. Each
are assigned to each group. As each TXRU transmits antenna port is allocated to the different antenna elements.
its CSI-RS, the UE can measure the channel from the As it does not require any precoding to differentiate one port
non-precoded CSI-RS of all TXRUs. The other is an array from the other on the same antenna, this particular type of
connected structure in which output signals of multiple hybrid beamforming is referred to as non-precoded CSI-RS
TXRUs are delivered to the single antenna element while transmission.
orthogonal CSI-RSs are assigned to each TXRU. As the
multiple TXRUs share a group of antennas, each CSI-RS 2) ARRAY CONNECTED STRUCTURE WITH BEAMFORMED
is precoded with different weights, allowing the UE to CSI-RS TRANSMISSION
measure the precoded channel from the beamformed CSI- A separate set of antenna ports can be configured for a
RS. Non-precoded CSI-RS and beamformed CSI-RS are group of UEs that employ the same analog beam while
the two possible implementation practices that have been fully connecting each antenna port with the physical antenna
introduced in LTE Rel-13. Even if non-precoded CSI-RS and elements within the partitioned set. Furthermore, different
beamformed CSI-RS are not explicitly referred to in NR, they user signals can be transmitted using the same physical
are considered as two possible implementation practices for antenna that corresponds to the different antenna ports if
system-level evaluation of massive MIMO. Fig. 2 illustrates they are beamformed with the different analog precoders.
(D)
two different TXRU structures for hybrid beamforming, Let {wi,j,q }N 1P
j=1 denote a digital precoder for UE i in group
one with nonprecoded CSI-RS transmission for the array q. The precoded symbols for each UE are mapped to the
partitioning structure and the other with beamformed CSI-RS set of antenna ports for each group. More specifically,
transmission for the array connected structure. In the sequel, Fig. 2(b) illustrates two UEs, namely UE 1 and UE 2, in the
we give details for these two TXRU structures, in which an same analog beam, each of which is digitally precoded by
M × N uniform planar array is considered. (D) CSI beams
{wi,j,q } for i = 1 and 2. A total number of NRES
are generated with the analog precoding matrix W(A) ∈
CSI
1) ARRAY PARTITIONING STRUCTURE WITH CN1 NRES ×MNP , allocating MNP/NRES CSI antenna ports to each
NON-PRECODED CSI-RS TRANSMISSION CSI
beam. The NRES CSI-RS resources, each associated with
As illustrated in Fig. 2(a), N horizontal and M vertical a single analog beam, are available over each time to
antenna elements are mapped into PN1 and N2 antenna interval (TTI). If we denote the precoding matrix for beam
ports by partitioning them into multiple groups, each with (A)
q as W(q) , the overall analog precoding matrix can given
L horizontal and U vertical elements. Each antenna port  T
(A) (A) (A)
is fully connected to all physical antenna elements within by W (A) = W(1) , W(2) , · · · , W CSI . Similar to (2)
(NRES )
the corresponding partitioned group, creating a partitioned (A)
antenna array, represented as an analog precoding matrix and (3), the analog precoding matrix W(q) in the current
W(A) ∈ CMNP×N1 N2 P . By denoting the vertical and horizontal implementation is set as
(A) (A) (A) (A) (A)
= [w1,V , w2,V · · · wU ,V ] and wH =
 
weights as wV (A) 1 2π
wq,m,l = √ exp −j (m − 1)dV cos θq,V
(A) (A) (A)
[w1,H , w2,H · · · wL,H ], respectively, the analog precoding ML λ
(A) (A)

matrix is given by the Kronecker product W(A) = wV ⊗wH . 2π
· exp −j (l − 1)dH sin θq,H ,
Even if it is not specifically provided in the NR specification, λ
the following setting is employed in our implementation: m ∈ (1, 2, · · · , M ) , l ∈ (1, 2, · · · , L) (4)
where θq,V and θq,H are the tilting angles for vertical and
 
(A) 1 2π
wu,V = √ exp −j (u − 1)dV cos θV ,
U λ horizontal antennas, respectively, for beam q [22]. Note that
u = 1, 2, · · · , U (2) (4) is a generalized expression for when analog beamforming
(A) 1

2π
 is applied to both vertical and horizontal planes. Because
wl,H = √ exp −j (l − 1)dH sin θH , more antenna elements are connected to the antenna port
L λ
l = 1, 2, · · · , L (3) associated with each beam, the beamformed CSI-RS type
can generate sharper beams than the non-precoded CSI-RS
where λ is the wavelength, and θV and θH are the tilting type (while scaling the beam power by NRES CSI ). For selecting

angles for vertical and horizontal antennas, respectively [22]. the best beam, each UE receives the reference signals
Note that beamwidth of the analog beam is governed by with different beams over NRESCSI CSI-RS resources and then

U and L, i.e., more antenna elements leading to a sharper measures the reference signal received power (RSRP). Then,
beam. In this configuration, channels are estimated for N1 N2 P the best RSRP is determined, and its beam index known
antenna ports using the CSI-RS, one for each port, in the as a CSI-RS Resource Indicator (CRI) is reported together
receiver. with RSRP and CSI sets that indicate the channel quality
(D) 1 N2 P
In Fig. 2(a), {wi,j }Nj=1 denotes the digital precoder and other channel properties in terms of Channel Quality
for UE i, selected using the feedback information from Indicator (CQI), Precoding Matrix Indicator (PMI), and Rank
all UEs. The codebooks for digital precoding in the 3GPP Indicator (RI).

94022 VOLUME 9, 2021

J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

The hybrid beamforming structure with all the above fea- TABLE 1. Antenna port configurations [32].
tures in the LTE specification, more specifically based upon
the later releases, i.e., 3GPP Rel-13 and Rel-14, is still main-
tained in the NR specification while a high-resolution CSI
reporting is included for improved MU-MIMO operation.
The following subsection presents the detailed enhancement
for digital beamforming with high-resolution CSI, based on
the dual-stage codebooks. Furthermore, beam management is
discussed in a separate subsection.
TABLE 2. ‘Type I single panel’ codebook [32].

C. DIGITAL BEAMFORMING
As opposed to employing both codebook-based and non-
codebook-based schemes for MIMO transmission in LTE,
NR is mainly based on only non-codebook-based precoding
for MIMO while employing CSI-RS and DMRS for CSI
acquisition and coherent demodulation, respectively. As dis-
cussed in Section II.B, CSI must be estimated and reported for
each port, in the downlink feedback-based CSI acquisition at
the transmitter. In fact, it is the quantized CSI in the form of
CQI, PMI, RI, and CRI. For example, codebook-based CSI with 2D-DFT:
 
acquisition is used for each UE, to evaluate which precoding (1) 1 vl,m
W(D) = Ws i ,s i ,n = √ . (5)
matrix in the codebook would give the best performance if 1 1,1 2 1,2 N1 N2 P φn vl,m
used by the BS [32] and to feed back the corresponding
| {z } | {z }
l m
PMI. It is critical to reduce the feedback overhead and CSI
complexity for the codebook-based CSI, especially when where φn = ejπn/2 where n = 0, 1, 2, 3 is the cross-polarized
the number of antennas port increases. To this end, several phase difference and vl,m denotes the element of oversampled
codebooks in the LTE and NR are designed to exploit a DFT matrix with the oversampling factor of O1 and O2 ,
correlation between antenna ports for closely spaced antenna representing the channel direction for reference polarization,
elements with limited angular spread. given as follows:
As opposed to the LTE-Advanced Pro which specifies only h 2πl 2πl(N1 −1)
iT
a single-panel codebook, NR includes four different PMI vl,m = um ej N1 O1 um . . . ej N1 O1 um (6)
types, depending on the number of panels and codebook where
types, namely ‘Type I single-panel’, ‘Type I multi-panel’, (h 2πm(N −1)
i
‘Type II’, and ‘Type II port selection’. The 3GPP standard j N2πm j N O2 , N2 > 1
1 e 2 O2 · · · e 2 2
has specified a double codebook that supports the X-pol um = (7)
1, N2 = 1
antenna in Release 10 and beyond. Here, the DFT matrix and
phase difference between cross-polarizations are sufficient to Note that s1 and s2 in (5) are used to select the
specify the codebook with 2D-DFT. However, representing codebook subset. The antenna configurations for ‘Type I
the precise CSI might be challenging when the number of single panel’ codebook is given by Table 2. Each UE
generated DFT vectors is limited by the number of antenna selects the optimal precoding matrix given by (5). The
ports. Therefore, to improve the accuracy, DFT vectors are selected matrix will be reported with the index values, i1 =
oversampled with the factors Oi , i = 1, 2, in LTE-Advanced i1,1 , i1,2 and i2 , which determines the codebook as defined
 

Pro and NR specifications. in Table 2. ‘‘Type I’’ codebook has two codebook modes,
As in the codebook with 2D-DFT, the panel-based codebookMode = 1 and 2, depending on feedback overhead.
codebook can be specified by the phase difference from Note that codebookMode = 2 follows Table 2 considering
the reference panel. As opposed to Type I codebook that i2 ∈ (0, 15) while taking the more DFT subsets than
selects only one oversampled DFT vector, Type II codebook codebookMode = 1 with i2 = 0, 1, 2, 3. See [31] for
selects a set of beams, where the quantized amplitudes codebooks for other rank or codebookMode.
are controlled by the power per each path, involving more For multiple panels in NR, digital precoding is considered
feedback overhead than Type I codebook. Finally, ‘Type II only for one-dimensional antenna ports. Therefore, although
port selection’ codebook is a beamformed CSI-RS version of not expressed in the standard specification, the mentioned
Type II codebook, applied to a specific antenna port. consideration should be met for determining the analog
In this study, we consider only the rank 1 ‘Type I’ beam. Multi-panel digital precoding needs to be performed
codebook, which has the following panel precoding matrix on the effective channel projected onto the vertical panel.

VOLUME 9, 2021 94023

 J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

In addition, for Ng = 2 or 4, horizontal panels are supported. beam change, and P3 for UE receive beam change, have been
In ’Type I Multi Panel’ codebooks, codebookMode is defined in radio resource control connected state. For exam-
also used for inter-panel correction updates [32]. When ple, the BS performs a beam sweep over an angular sector
codebookMode = 1, it is similar to the concept of digital that covers the entire cell by transmitting a unique reference
precoding for ‘Type 1 single-panel’. The DFT coefficient signal in each beam (P1), which are relatively wide to limit
of digital precoding for ‘Type 1 multi-panel’ is determined the number of beams to sweep. Then, a beam can be refined
by phase relationship between the two antenna panels. More by performing a beam sweep over a narrower angular sector
(1)
specifically, referring to Wl,m,n in (5) as a reference single around the best beam reported by the UE. Narrowing the
panel, precoding matrices for the remaining panels are range allows the UE to detect the best one and report it to BS
determined by the inter-panel co-phasing factor ϕk , which (P2). UE refines its receiver beam by repeatedly using a fixed
reflects the panel correlation with the reference panel. The beam transmitted by the BS, where beam pairs are established
precoding matrix W(D) for multi-panel antenna is then given for analog beamforming in both the BS and UE (P3).
as Beam maintenance is required to deal with beam mis-
h iT alignment. More specifically, beam tracking and refinement
W(D) = W(1) (1)
l,m,n ϕk Wl,m,n is implemented so that a beam can be switched to another
T beam with better quality. Beam recovery is required to find
vl,m φn vl,m ϕk vl,m ϕk φn vl,m

= √ (8) a new beam if a communication link cannot be maintained
N1 N2 P
with the current beam. Beam measurement measures the
where ϕk = ejπk/2 with k = 0, 1, 2, 3. codebookMode = 1 quality of different beams on SSB or CSI-RS transmitted
is available when Ng = 2 or 4. In contrast, by the BS. There are two different types of beam reporting,
codebookMode = 2 is supported only when Ng = 2. both indicating the best beams along with their measured
By updating of the phase relationship between panels and qualities: grouping-based reporting and nongrouping-based
phase difference between cross-polarization, the matrix can reporting. In grouping-based reporting, a group of BS Tx
be expressed as beams can be received simultaneously by the UE, allowing

vl,m φn vl,m ap1 bn1 vl,m ap2 bn2 vl,m
T the establishment of links between multiple beam pairs. How-
(D)
W = √ (9) ever, in nongrouping-based reporting, the UE is configured to
N1 N2 P sweep the multiple beams individually and report up to four
where ap = ejπ(2p+1)/4 and bn = ejπ(2n−1)/4 . beams. Due to its robustness, the grouping-based reporting
has been shown to outperform the nongrouping-based
D. ANALOG BEAM MANAGEMENT reporting by approximately 15% [35]. However, this study
As channel conditions can be more vulnerable to mobility considers only the nongrouping-based reporting.
or blockage for much sharper and more beams in the
mmWave band, beam management has been introduced in III. 5G K-SimSys FOR MASSIVE MIMO:
NR for efficient analog beamforming. It is defined as a set IMPLEMENTATION & PERFORMANCE ANALYSIS
of procedures for layers 1 and 2 to acquire and maintain The 5G K-SimSys was designed and implemented to provide
a set of BS and UE beams, including features such as an open platform and testbed for evaluating the system-level
beam sweeping, beam measurement, beam reporting, beam performance of 5G standard [9]–[12]. One of the key
determination, beam maintenance, and beam recovery. Two features in 5G K-SimSys is its modular and flexible structure
different reference signals are used for downlink beam that allows the reuse of the same modules for various
management: synchronization signal block (SSB) and CSI- system configurations by modifying existing or including
RS. SSB consists of synchronization signals and physical additional modules for newly introduced algorithms and/or
broadcast channel (PBCH). As opposed to a block of synchro- functionalities. Furthermore, it will be open to public so that
nization signal and PBCH in LTE that has fixed time-domain all its modules can be redesigned and shared for research
and frequency-domain positions in every subframe, SSB and development in 5G system.1 It should be noted that
transmits signals in various patterns, depending on subcarrier implementation of massive MIMO, especially conforming to
spacing and some other parameters. Furthermore, a set of the current 5G standard specification, involves many existing
multiple SSBs is defined as an SS burst set, which is repeated key modules in 5G K-SimSys.
with a given periodicity, e.g., 20 ms. The multiple SSBs in Fig. 3 illustrates the overall modular structure of
a SS burst set are then transmitted repeatedly in different the 5G K-SimSys having five key functionality lay-
beams, allowing the UE to measure the different SSBs to ers: Simulation Top, Network Configuration,
determine the best BS transmit beam during an initialization Channel Model, Radio Resource Management,
phase. The maximum number of SSBs within a SSB set, and Link Performance [9]–[12]. Furthermore, the func-
denoted by Lmax , is specified in NR, i.e., Lmax = 4 or 8 in tionalities in an individual layer are executed by three
sub 6 GHz and Lmax = 64 in mmWave band.
Three different types of beam sweeping processes, P1 for 1 The current version of 5G K-SimSys is available at
initial transmit/receive beam selection, P2 for BS transmit https://github.com/5GKSimSys/5G-K-SimSys

94024 VOLUME 9, 2021

J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

FIGURE 4. Overall simulator structure: Sub-modules.

embodied by submodules, which are the lower functional

blocks that implement the detailed simulation process. The
submodules can be added or reset according to the simulation
scenario. In this section, we discuss how the modular
and flexible feature of 5G K-SimSys can immediately
FIGURE 3. Main modules in 5G K-SimSys.
extend the current baseline simulator to implement 5G
massive MIMO for NR (NR-MIMO), which is analyzed
in Section II.
different entities, each referred to as an ‘‘object.’’ The
object SIM is a main entity that defines and triggers overall
A. IMPLEMENTATION
simulation processes for all modules in the different layers.
The submodules with black-line box are the basic mod-
Every object inherits the same functional layer structures.
ules that do not change in the different entity object.
A specific functionality in each layer is triggered by the
Meanwhile, SLS for NR-MIMO involves modification
object SIM, and the actual execution is completed by the
of only several submodules, indicated by the red-line
other two entities, object MS and object BS. Depending
boxes in Fig. 5. They include submodules of Feedback
on the functional requirements and performance indicators,
and ReceivedSINRCalculation in Scheduling
additional classes of objects can be defined and integrated.
module and BeamInformation in Performance
For example, two different classes of object MS can be
module.
defined when eMBB and uRLLC devices coexist with the
different performance indicators, e.g., object MSeMBB
for eMBB devices and object MSuRLLC for uRLLC
devices [10]. The current modular and flexible structure
immediately facilitates extension of the baseline simulator
to various scenarios and functional requirements.
Individual functionality in each layer of the different object
is embodied by a module, which implements the detailed
procedures and stores data generated through the simulation.
Its function and data are identified by the corresponding
object which facilitates the retrieval [9]–[12]. Simulation
Top is the highest layer that integrates the overall simulation
processes and manages the entire flow of simulation by
triggering individual functionalities in the lower layers of FIGURE 5. 5G K-SimSys: Feedback module for NR-MIMO.
each object. However, only three modules, SystemSim,
SystemBS, and SystemMS are transparent in this layer, The module Scheduling involves a timely interaction
abstracting the specific functionalities in the lower layers. between gNB and UE, achieved by invoking the submodule
Therefore, the readability of the entire simulator can be Feedback for feedback information. A feedback set (CQI,
improved. PMI, RI, and CRI) is determined upon each module call as
The module SystemSim executes the simulation process the output of submodule Feedback. Fig. 5 presents detailed
by calling up the modules in the lower layers (vertical implementation of the submodule Feedback that interacts
processing), which subsequently calls up the corresponding with module Scheduling. Note that all submodules
modules in the same layer of module SystemBS and/or in each module are controlled by a main object SIM
module SystemMS (horizontal processing). Furthermore, that merely triggers the corresponding modules in objects
Simulation Top layer directly interacts with all modules in MS and BS. For example, the submodule Feedback
the lower layers for initialization. in the module Scheduling triggers a Feedback sub-
Fig. 4 illustrates an embodiment of module SystemSim, module that will be executed by object SchedulingBS
which interacts with all other modules in the lower layers or SchedulingMS, depending on the entity indicated
of the baseline 5G K-SimSys. Furthermore, each module is by the module call input. The function beamforming

VOLUME 9, 2021 94025

 J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

is a key element of submodule Feedback in which a the CQI/PMI/RI/CRIs and RSRPs for the reported CRIs by
link performance is mainly governed by the beamforming all UEs, the Scheduling module determines the set of UEs
gain. The antenna configuration of (N1 , N2 , P), channel for to be served simultaneously for MU-MIMO. It considers a
each UE, and beam information are initialized as input possible spatial multiplexing gain in terms of the number
to the function beamforming. Note that two types of of layers for each UE. Different algorithms, including the
CSI-RS transmissions are supported for hybrid beamforming, proportional fair (PF) algorithm, can be introduced for
as analyzed in Section II. The configurations with N2 = 1 multi-user scheduling [33]. Towards this end, an active set of
and N2 > 1 correspond to beamformed CSI-RS and UEs that require HARQ must be considered, along with the
nonprecoded CSI-RS types, respectively. As shown in Fig. 4, scheduled UEs. Because the specific scheduling algorithm is
the submodule BeamInformation triggers either function not a rigid part of 5G K-SimSys, existing algorithms can be
beamMaintenance or function beamRecovery with modified or replaced in a flexible and modular manner.
two-bit indicator. The function analogBeamforming is Similarly, the submodule ReceivedSINRCalcula-
required if one of the two bits is indicated as ‘1’ during tion in the Module Scheduling triggers the peer
performance measurement. In the case of beam maintenance, submodule ReceivedSINRCalculation in module
the first bit of beam information becomes ‘1’ when RSRP SchedulingBS or SchedulingMS, for calculating the
of the selected beam drops more than twice. In addition, received SINR of the UEs that are scheduled over TTI,
the current implementation limits the maximum number of based on their beamforming information. This particular
HARQ retransmissions to four, after which a link failure is submodule reuses sINRCalculation and l2SMapping
declared. Beam refinement is executed until two consecutive functions in the submodule Feedback. The effective
HARQ retransmissions fail. During beam recovery, if more received SINR is computed by (11) and returned to
than two consecutive HARQ retransmissions fail, the second submodule ReceivedSINRCalculation in the module
beam information is set to ’1’. Beam sweeping searches Scheduling using the effective estimated channel via
for an analog beam pair, which is an essential part of the DMRS. Let S` denote a set of UEs that are scheduled for
initial access procedure. Toward this end, as the function MU-MIMO transmission in BS `, ` ∈ C, where C denotes
analogBeamforming must be called, two-bit indicator a set of BS’s. Without loss of generality, we consider a
of beam information is initialized as ‘11’. If at least one single data stream per user. Let H`,k denote a MIMO channel
(A)
of two bits is set to ‘1’, it is updated as ‘00’ after beam matrix between BS ` and UE k. Furthermore, let W` =
determination. If beam information indicates ‘00’ at the time h iT h iT
(A) (A) (D) (D) (D)
W`,1 , . . . , W`,|S` | and W` = W`,1 , . . . , W`,|S` |
of calling the submodule Feedback, it implies that the
previously selected analog beam works, i.e., only a function represent the analog and digital precoding matrices in BS
digitalBeamforming is executed without invoking the `, respectively. Assuming equal power allocation to all
function analogBeamforming. Each UE then stores the users served
P
by the given gNB with transmit power of P,
best analog beam sets that have been selected for the given i.e., tr = |SP` | Ik , the received signal for UE is given
CSI-RS transmission type. as
In the function digitalBeamforming, PMI that
maximizes the signal-to-leakage and noise ratio (SLNR) is y`,k
(A) (D)
!
selected for ‘Type I single-panel’ codebook in Table 1 or p H`,k W`,u W`,k x`,k
= P/|S` | · (A) (D)
‘Type I multi-panel’ codebook for multi-panel case. For
P
+ j∈S` \k H`,k W`,j W`,j x`,j
each UE, a maximum of 8 layers can be supported for SU- X p
MIMO. Once PMI is determined, SINR of each subcarrier + P/|Sc | · Hc,k W(A) (D)
c Wc xc + n`,k (10)
c∈C\`
is calculated using the function sINRCalculation. Then,
the effective SINR is calculated by the l2SMapping func- where x`,k and n`,k denote the transmitted symbol and the
tion, in which the frequency-domain channel and interference additive white Gaussian noise for UE k in BS `, respectively.
are taken into account for measuring the received signal Assuming that a minimum mean square error-interference
strength. rejection combining (MMSE-IRC) filter g`,k is employed in
The link-to-system (L2S) mapping, e.g., mutual the receiver for UE k in BS `, the SINR for UE k in BS ` is
information-based exponential SNR mapping (MIESM), represented as
is also an essential part of calculating the effective SINR, (A) (D) 2
and can be replaced with other L2S mapping methods to g`,k H`,k W`,k W`,k
capture the specific characteristics. An effective SINR is γ`,k = ! 2
(A) (D)
H`,k W`,−k W`,−k
computed to identify CQI from BLER curve, which is |S` | σ`,k
2 + gk P (A) (D)
obtained by LLS over the additive white Gaussian channel, + c∈C\` Hc,k Wc Wc
F
as described in the SLS methodology [1]–[5]. Afterwards, (11)
CQI, PMI, and RI are fed back by UE. For the case of
2 (A) (D)
N2 = 1, i.e., beamformed CSI-RS type, CRI stored in the where σ`,k
2 = g
`,k σn /P while W−k and W−k denote
2
function analogBeamforming is also reported. Based on precoding matrices for analog and digital beamforming that
94026 VOLUME 9, 2021
J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

TABLE 3. Simulation models and parameters. NR for the given number of antenna ports. Fig. 6 shows
the average spectral efficiency for both types of CSI-RS
transmissions on varying the number of CSI-RS ports. For
N1 N2 P = 64, the performance is obtained by employing
a fully digital beamforming structure in which all physical
antenna elements are mapped into an individual port,
i.e., one-to-one mapping between antenna element and port.
However, the efficiency decreases with the smaller number
of CSI-RS ports, as the degree of freedom that can be
controlled by feedback is reduced. Because more feedback
bits are employed for the nonprecoded CSI-RS transmission,
its performance can be improved over the beamformed
CSI-RS transmission as N1 N2 P increases. In contrast,
the performance of the beamformed CSI-RS transmission
can be improved by reducing N1 N2 P as digital beam-
forming becomes less effective within the sharper analog
beam.

exclude the kth precoding matrix in BS `, respectively. Note

that the MMSE-IRC filter in (11) is given as
 H  P −1
g`,k = H̃`,k H̃`,k H̃H
`,k + σ 2
I
n Nr (12)
|S` |
(A) (D)
where H̃`,k = H`,k W`,k W`,k .

B. BASELINE PERFORMANCE EVALUATION FOR NR

This section evaluates the average spectral efficiency of 5G FIGURE 6. Average spectral efficiency as varying the number of CSI-RS
K-SimSys, as the antenna configurations are varied for ports [12].

massive MIMO. In particular, we first attempt to compare

the baseline performance of two different types of CSI-RS
transmission schemes with ‘Type I single-panel’ feedback.
The results for multi-panels are covered in Section IV.
Table 3 summarizes the simulation parameters for NR eMBB
service in the urban macro-cell (UMa) environment [37].
A 3D channel model is essential for evaluating the perfor-
mance of massive MIMO [21], [22], [36]. 5G K-SimSys
provides its source code that has been calibrated for channel
verification with the results in [21], [22], [36]. The 3D-SCM
generates a long-term channel, one for each link, along
with a short-term channel that varies with mobile speed.
Furthermore, it considers the mmWave band at 30 GHz
with the number of antennas MNP = 64 for each
panel.
For MNP = 64, the average cellular spectral efficiency
depends on the number of CSI-RS ports, i.e., N1 N2 P, for
both CSI-RS transmission types. Recall that a specific
pattern for antenna ports, i.e., N1 and N2 , is specified in FIGURE 7. CDF for varying the number of layers: N1 N2 P = 8 [12].

VOLUME 9, 2021 94027

 J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

Fig. 7 shows the distribution of the number of MU-MIMO

layers (the number of users that has been selected for
simultaneous transmissions assuming a single layer per
UE) for N1 N2 P = 8. It is notable that the beam-
formed CSI-RS transmission allows for more layers,
which results in selecting different users for each analog
beam.
Fig. 8 illustrates the snapshot for beam maintenance while
assuming that a specific UE is scheduled for certain slots.
We observe that a reference UE is served by selecting
a reception beam index 5, which has the highest RSRP
starting from a slot number of 30. When the normalized
RSRP 5 gradually decreases with time, two consecutive
NACK messages are reported at slot numbers 42 and 45,
respectively, eventually evoking beam maintenance. Then,
P2 is applied to select beam index 3 at slot number 48.
Subsequently, CSI feedback sets for the selected beam FIGURE 9. Average ICI/(ICI + IUI) ratio as varying the BS height in the
index 3 are determined. Fig. 8 also illustrates how the spectral different environments: N1 N2 P = 8.

efficiency of UE varies during beam maintenance. A seamless

beam selection for beam maintenance is found to maintain
spectral efficiency. converges to that of fully digital beamforming, as hBS
increases.

IV. DEMONSTRATION FOR USE CASE: BAND-WISE

ANALOG BEAMFORMING (BAB)
A. BAND-WISE ANALOG BEAMFORMING (BAB)
In this section, we introduce a notion of BAB, which
countermeasures the mismatch effect of independent analog
and digital beam selection for multi-user MIMO. Then, 5G
K-SimSys is employed to implement a system-level simulator
for evaluating a spectral efficiency gain achieved by BAB.
Recognizing that a typical module in 5G K-SimSys could be
reused to design the virtual modules for BAB, this section
intends to present one particular useful case of exploiting the
modular and flexible structure of 5G K-SimSys. The case will
be used to demonstrate how powerful the underlying modular
architecture is to structure the numerous new entities with
FIGURE 8. A snapshot for beam maintenance.
individual objects of the same functionality layers.
As a wideband mmWave band suffers from frequency
selectivity, its effect must be carefully taken into account
Next, we investigate the effect of vertical beamforming on in the course of implementing the hybrid beamforming
inter-user interference (IUI) and inter-cell interference (ICI) structures [29]. Each UE is allocated with multiple bandwidth
in the FD-MIMO environment. Note that the line-of-sight parts (BWPs), each of which is a group of contiguous physical
zenith angle of departure (LoS-ZoD) depends on the height resource blocks (PRBs). In the course of scheduling, however,
of base station, hBS , in 3D-UMi and 3D-UMa scenarios. only one BWP is activated, and its CSI feedback set is
In the current model, LoS-ZoD is mainly centered around determined by each UE. Meanwhile, analog precoding is
90◦ in 3D-UMa environment. As hBS increases, LoS-ZoD only applied to wideband; therefore, each BS must select
is shown to increase gradually, implying that most of (A)
the optimal wideband analog  beamforming matrix, W =
T
the UEs fall under a limited angle span in the vertical

(A) (A) (A)
dimension as compared to 3D-UMi. Therefore, it is nec- W(1) , W(2) , · · · , W CSI ) , which is then used as the set
(NRES
essary to make sharper vertical beams using more antenna of the scheduled UEs. Furthermore, digital beamforming
elements to reduce IUI. Fig. 9 shows the average ICI- is determined only on the scheduled PRB, along with the
to-ICI plus IUI (ICI/(ICI+IUI)) as hBS increases while wideband analog beam. Due to the limited number of analog
fixing hUT . As LoS-ZoD increases with hBS , ICI is reduced. beams that can be shared among all UEs in each TTI, e.g.,
Furthermore, vertical sectorization with analog beamforming CSI = 4, some UEs might not be scheduled for their own
NRES
becomes less effective, i.e., its performance eventually reported beams, limiting the overall system performance.

94028 VOLUME 9, 2021

J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

Such a mismatch in beam allocation might result in further implementation scenario, some modules must be carefully
performance degradation due to the frequency selectiv- redesigned and/or modified while reusing the existing func-
ity in wideband analog beamforming. In other words, tional blocks efficiently. Once UEs are connected, subbands
the current standards do not allow for joint analog and are selected on a much longer period than the periodic
digital beam selection. In this section, we introduce the reference signal. Selection of the individual analog beam
notion of BAB as a means to offset the effect of such a for each subband implies that independent scheduler and
mismatch by exploiting both frequency and beam diversities hybrid ARQ processes can be employed for the different
simultaneously. subbands. Therefore, each subband is subject to SINR
calculation, requiring multiple parallel submodules to be
redesigned. In fact, modules for Channel, Scheduling, and
Performance can be redesigned or reused so that each module
is performed independently on a band-wise basis. Due to the
highly modular structure of functionalities for all objects,
the existing simulator can be immediately modified by
systematic module calls.

FIGURE 11. Virtual modules for BAB in 5G K-SimSys.

Fig. 11 shows the modular structure of the 5G K-SimSys

for BAB, modified from that of the baseline simulator
FIGURE 10. Multipanel-based bandwise analog beamforming (BAB): in Fig. 5. The same Channel submodule is reused for
Illustration. generating the channel matrices for all different panels. More
specifically, let H̃mn (t) denote a time-domain channel matrix
BAB divides a whole band into multiple narrow bands, for (m,n)th panel, m = 1, 2, · · · , Mg ; n = 1, 2, · · · , Ng ,
each for independent analog beamforming. Then, different which is given by a submatrix of the channel matrix for all
antenna ports are assigned to different subbands. For antenna elements. Once a channel submatrix H̃11 (t) for a
example, the overall bandwidth is equally divided into B reference panel is generated by the baseline channel module,
CSI /B antenna ports to each
subbands, while allocating NRES new submodules for all other channels {H̃m,n (t)}m,n6=1 can
band. In NR specification, each BWP can be consid- be generated by considering the phase change due to the
ered as a subband for BAB. As array gain is reduced relative distance from the reference panel. Similarly, the same
with decreasing number of antenna elements per antenna Scheduling and Performance modules are employed
port, analog beamforming becomes more sensitive towards now for each subband. In other words, a simulator can be
inter-panel interference. However, multi-user diversity gain redesigned at the individual module and submodule levels
can be obtained if more beams are selected for the while maintaining a baseline structure. As a feedback process
different subbands. Fig. 10 illustrates subband allocation is executed for each subband independently, the existing
for a multi-panel-based BAB, in which each panel supports Feedback submodule can be also reused to create a virtual
different subband by dividing a whole band into Mg Ng module for each panel. The overall performance is configured
subbands, one for each panel. Then, each UE reports by SystemSim module, which collects the results from all
the feedback set independently for each subband, which virtual modules.
allows for frequency and beam diversity-aware flexible
scheduling. The performance evaluation by 5G K-SimSys C. PERFORMANCE ANALYSIS
is shown in the following sections after discussing its Assuming that Mg = Ng = 2, performance of BAB-based
implementation. sharing is compared with that of the wideband analog beam
(WAB)-based sharing that employs the ‘Type I multi-panel’
B. IMPLEMENTATION OF VIRTUAL MODULES FOR BAB codebook. Referring to the band configuration in Fig. 10,
From an implementation viewpoint, the most important a whole band is equally divided into four subbands for
feature of 5G K-SimSys is reusability of the baseline BAB. Fig. 12 shows the average cell spectral efficiency
modules centered around the main object SIM. For any new of both sharing schemes for the different number of UEs

VOLUME 9, 2021 94029

 J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

per sector. When there are four UEs per sector, BAB achieves number of UEs, i.e., one or two UEs. Furthermore, some
a performance gain of about 47% over WAB. UEs are not allocated to any beam, e.g., UEs 2, 4, 6, and 8,
due to frequency selectivity. Meanwhile, two different UEs
in the same beam can be supported by multi-user MIMO
with digital beamforming. In contrast, Fig. 13(b) shows
beam allocation for BAB with four subbands. We observe
that different beams are selected in each subband. For
example, UE 2 has two possible beams to use, beam 1 in
subband 2 and beam 2 in subband 4. Thus, all UEs are
served simultaneously over different subbands, indicating
that both multi-user and multi-subband diversity can be
achieved by BAB.

V. CONCLUSION
We discussed the implementation of 5G K-SimSys to evaluate
the massive MIMO performance, including the related 3GPP
New Radio specification details. Then, its actual performance
has been demonstrated for the precoding matrices, e.g.,
‘Type I Single/Multi-Panel’ codebook. As demonstrated
in this study, system-level evaluation of massive MIMO
FIGURE 12. Average spectral efficiency as varying the number of users
per sector: Wideband beam sharing vs. BAB-based beam is a good example to fully explore the detailed features
sharing. in 5G K-SimSys. We evaluated the baseline performance of
NR-MIMO using 5G K-SimSys. Furthermore, we compared
the performance of the two types of CSI-RS transmissions
by varying the number of antenna ports, and analyzed the
effect of vertical beamforming on the interference accord-
ing to the various FD-MIMO environments. In addition,
the effect of vertical beamforming under the given UE
distribution scenario was carefully evaluated for various
system aspects of massive MIMO. In fact, it has been
demonstrated that 5G K-SimSys is as a useful tool for a
comprehensive system-level evaluation of massive MIMO,
which involves the detailed system models in various
environments.
Finally, we identified that implementing the same analog
beam over the wideband limits the flexibility in scheduling.
Therefore, we proposed a bandwise analog beamform-
ing (BAB) scheme to improve a UE scheduling gain, allowing
for both multi-user and multi-band diversity, despite the
less antenna gain per panel. As demonstrated in this study,
the BAB yields approximately 47% better performance than
the NR with flexible scheduling. For its performance analysis,
we showed how a typical module in 5G K-SimSys could
FIGURE 13. A snapshot for beam allocation. be reused to design the virtual modules specialized for
BAB. In fact, its modular and flexible structure turns out
Additional performance gain is observed with more to be useful for introducing new features in the system
numbers of UEs in each sector. This can be confirmed by level, without having to design and implement the whole
investigating a snapshot of beam allocation for individual processes from the scratch. As 5G K-SimSys is currently
UEs for both schemes. In particular, Figs. 13(a) and (b) show available for public use, it can serve as an open platform and
specific snapshots for beam allocation with WAB and testbed for evaluating the system-level performance of 5G
BAB-based sharing schemes, respectively. We assume that systems.
CSI = 4 and 2 for WAB and BAB, respectively. Fig. 13(a)
NRES
demonstrates the allocation of beams to individual UEs, REFERENCES
where beam index 0 indicates no beam allocation. While
[1] J. C. Ikuno, M. Wrulich, and M. Rupp, ‘‘System level simulation of
generating rather sharper beams using all antenna elements LTE networks,’’ in Proc. IEEE 71st Veh. Technol. Conf., Taipei, Taiwan,
with a single wideband, each beam tends to serve a limited May 2010, pp. 1–5.

94030 VOLUME 9, 2021

J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

[2] L. Chen, W. Chen, B. Wang, X. Zhang, H. Chen, and D. Yang, [22] Study on Elevation Beamforming/Full-Dimension (FD) Multiple Input
‘‘System-level simulation methodology and platform for mobile cel- Multiple Output (MIMO) for LTE, document 3GPP TR 36.897 V13.0.0,
lular systems,’’ IEEE Commun. Mag., vol. 49, no. 7, pp. 148–155, Rel-13, Jun. 2015.
Jul. 2011. [23] A. F. Molisch, V. V. Ratnam, S. Han, Z. Li, S. L. H. Nguyen, L. Li, and
[3] M. K. Müller, F. Ademaj, T. Dittrich, A. Fastenbauer, B. R. Elbal, K. Haneda, ‘‘Hybrid beamforming for massive MIMO: A survey,’’ IEEE
A. Nabavi, L. Nagel, S. Schwarz, and M. Rupp, ‘‘Flexible multi-node Commun. Mag., vol. 55, no. 9, pp. 134–141, Sep. 2017.
simulation of cellular mobile communications: The Vienna 5G system [24] A. Goldsmith, Wireless Communications. Cambridge, U.K.:
level simulator,’’ EURASIP J. Wireless Commun. Netw., vol. 2018, no. 1, Cambridge Univ. Press, Aug. 2005.
pp. 1–17, Dec. 2018. [25] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.
[4] C. Mehlführer, J. Colom Ikuno, M. Šimko, S. Schwarz, M. Wrulich, Cambridge, U.K.: Cambridge Univ. Press, May 2005.
and M. Rupp, ‘‘The Vienna LTE simulators—Enabling reproducibility in [26] S. Han, C.-L. I, Z. Xu, and C. Rowell, ‘‘Large-scale antenna
wireless communications research,’’ EURASIP J. Adv. Signal Process., systems with hybrid analog and digital beamforming for millime-
vol. 2011, no. 1, pp. 1–14, Jul. 2011. ter wave 5G,’’ IEEE Commun. Mag., vol. 53, no. 1, pp. 186–194,
[5] M. Rupp, S. Schwarz, and M. Taranetz, The Vienna LTE-Advanced Jan. 2015.
Simulators: Up and Downlink, Link and System Level Simulation (Signals [27] S. Park, A. Alkhateeb, and R. W. Heath, Jr., ‘‘Dynamic subarrays for hybrid
and Communication Technology), 1st ed. Singapore: Springer, 2016. precoding in wideband mmWave MIMO systems,’’ IEEE Trans. Wireless
Commun., vol. 16, no. 5, pp. 2907–2920, May 2017.
[6] Study on Channel Model for Frequencies From 0.5 to 100 GHz, document
[28] O. E. Ayach, S. Rajagopal, S. Abu-Surra, Z. Pi, and R. W. Heath,
3GPP TR 38.901 V14.5.0, Rel-14, Jan. 2018.
Jr., ‘‘Spatially sparse precoding in millimeter wave MIMO systems,’’
[7] Guidelines for Evaluation of Radio Interface Technologies for IMT-2020, IEEE Trans. Wireless Commun., vol. 13, no. 3, pp. 1499–1513,
document Rep. ITU-R M.2412, Oct. 2017. Mar. 2014.
[8] M. Shafi, A. F. Molisch, P. J. Smith, T. Haustein, P. Zhu, P. De Silva, [29] A. Alkhateeb, J. Mo, N. Gonzalez-Prelcic, and R. W. Heath, Jr.,
F. Tufvesson, A. Benjebbour, and G. Wunder, ‘‘5G: A tutorial overview ‘‘MIMO precoding and combining solutions for millimeter-wave
of standards, trials, challenges, deployment, and practice,’’ IEEE J. Sel. systems,’’ IEEE Commun. Mag., vol. 52, no. 12, pp. 122–131,
Areas Commun., vol. 35, no. 6, pp. 1201–1221, Jun. 2017. Dec. 2014.
[9] S. Cho, S. Chae, M. Rim, and C. G. Kang, ‘‘System level sim- [30] F. Sohrabi and W. Yu, ‘‘Hybrid digital and analog beamforming design for
ulation for 5G cellular communication systems,’’ in Proc. 9th Int. large-scale antenna arrays,’’ IEEE J. Sel. Topics Signal Process., vol. 10,
Conf. Ubiquitous Future Netw. (ICUFN), Milan, Italy, Jul. 2017, no. 3, pp. 501–513, Apr. 2016.
pp. 296–299. [31] A. Zaidi, F. Athley, J. Medbo, U. Gustavsson, G. Durisi, and X. Chen,
[10] M. Han, J. W. Lee, C. G. Kang, and M. J. Rim, ‘‘5G K-SimSys: 5G Physical Layer: Principles, Models and Technology Components.
Open/modular/flexible system level simulator for 5G system,’’ in Proc. New York, NY, USA: Academic, 2018.
IEEE Int. Symp. Dyn. Spectr. Access Netw. (DySPAN), 2018, pp. 1–2, [32] Physical Layer Procedures for Data, document 3GPP TS 38.214 V15.3.0,
doi: 10.1109/DySPAN.2018.8610417. Rel-15, Oct. 2018.
[11] M. S. Han, J. W. Lee, C. G. Kang, and M. J. Rim, ‘‘System-level [33] Y. Huang, Y. Li, H. Ren, J. Lu, and W. Zhang, ‘‘Multi-panel
performance evaluation with 5G K-SimSys for 5G URLLC system,’’ in MIMO in 5G,’’ IEEE Commun. Mag., vol. 56, no. 3, pp. 56–61,
Proc. 16th IEEE Annu. Consum. Commun. Netw. Conf. (CCNC), 2019, Mar. 2018.
pp. 1–5, doi: 10.1109/CCNC.2019.8651885. [34] Y. Choi, J. Lee, M. Rim, C. G. Kang, J. Nam, and Y.-J. Ko, ‘‘System-level
[12] J. W. Lee, M. S. Han, C. G. Kang, and M. J. Rim, ‘‘5G K- performance of limited feedback schemes for massive MIMO,’’ ETRI J.,
SimSys for system-level evaluation of massive MIMO,’’ in Proc. 16th vol. 38, no. 2, pp. 280–290, Apr. 2016.
IEEE Annu. Consum. Commun. Netw. Conf. (CCNC), 2019, pp. 1–6, [35] Y.-N.-R. Li, B. Gao, X. Zhang, and K. Huang, ‘‘Beam management
doi: 10.1109/CCNC.2019.8651802. in millimeter-wave communications for 5G and beyond,’’ IEEE Access,
[13] M. Mezzavilla, M. Zhang, M. Polese, R. Ford, S. Dutta, S. Rangan, vol. 8, pp. 13282–13293, 2020.
and M. Zorzi, ‘‘End-to-end simulation of 5G mmWave networks,’’ IEEE [36] Study on 3D Channel Model for LTE, document 3GPP TR 36.873 V12.0.0,
Commun. Surveys Tuts., vol. 20, no. 3, pp. 2237–2263, 3rd Quart., 2018. Rel-12, Sep. 2014.
[37] Study on New Radio Access Technology Physical Layer Aspects, document
[14] T. L. Marzetta, ‘‘Noncooperative cellular wireless with unlimited numbers
3GPP TR 38.802 V14.2.0, Rel-14, Sep. 2017.
of base station antennas,’’ IEEE Trans. Wireless Commun., vol. 9, no. 11,
pp. 3590–3600, Nov. 2010.
[15] F. Rusek, D. Persson, B. Kiong Lau, E. G. Larsson, T. L. Marzetta,
and F. Tufvesson, ‘‘Scaling up MIMO: Opportunities and challenges with
very large arrays,’’ IEEE Signal Process. Mag., vol. 30, no. 1, pp. 40–60,
Jan. 2013.
[16] E. G. Larsson, O. Edfors, F. Tufvesson, and T. L. Marzetta, ‘‘Massive
MIMO for next generation wireless systems,’’ IEEE Commun. Mag.,
vol. 52, no. 2, pp. 186–195, Feb. 2014.
[17] L. Lu, G. Y. Li, A. L. Swindlehurst, A. Ashikhmin, and R. Zhang,
‘‘An overview of massive MIMO: Benefits and challenges,’’
IEEE J. Sel. Topics Signal Process., vol. 8, no. 5, pp. 742–758,
Oct. 2014.
[18] Y. Kim, H. Ji, J. Lee, Y.-H. Nam, B. L. Ng, I. Tzanidis, Y. Li, and J. Zhang,
‘‘Full dimension MIMO (FD-MIMO): The next evolution of MIMO
in LTE systems,’’ IEEE Wireless Commun., vol. 21, no. 2, pp. 26–33,
Apr. 2014.
[19] Y.-H. Nam, B. Ng, K. Sayana, Y. Li, J. Zhang, Y. Kim, and J.
Lee, ‘‘Full-dimension MIMO (FD-MIMO) for next generation cellu- JAEWON LEE received the B.S. degree from the
lar technology,’’ IEEE Commun. Mag., vol. 51, no. 6, pp. 172–179, School of Information and Communication Engi-
Jun. 2013. neering, Chungbuk National University, in 2014.
[20] H. Ji, Y. Kim, J. Lee, E. Onggosanusi, Y. Nam, J. Zhang, B. Lee, and He is currently pursuing the unified M.S. and
B. Shim, ‘‘Overview of full-dimension MIMO in LTE-advanced pro,’’ Ph.D. degrees with the Department of Telecom-
IEEE Commun. Mag., vol. 55, no. 2, pp. 176–184, Feb. 2017. munication System Technology, Korea University
[21] B. Mondal, T. Thomas, E. Visotsky, F. Vook, A. Ghosh, Y.-H. Nam, (KU). His research interests include massive
Y. Li, J. Zhang, M. Zhang, Q. Luo, Y. Kakishima, and K. Kitao, ‘‘3D MIMO and waveforms.
channel model in 3GPP,’’ IEEE Commun. Mag., vol. 53, no. 3, pp. 16–23,
Mar. 2015.

VOLUME 9, 2021 94031

 J. Lee et al.: 5G K-SimSys for Open/Modular/Flexible System-Level Simulation

MINSIG HAN received the B.S. degree in elec- CHUNG G. KANG (Senior Member, IEEE)
trical engineering from Korea University (KU), received the B.S. degree in electrical engineering
in 2016, where he is currently pursuing the unified from the University of California, San Diego,
M.S. and Ph.D. degrees in electrical engineering. in 1987, and the M.S. and Ph.D. degrees in
He is honored with the 2017 Global Ph.D. electrical and computer engineering from the
Fellowship (GPF) and funded by the National University of California, Irvine, in 1989 and 1993,
Research Foundation (NRF) for his research. His respectively. Since March 1994, he has been with
research interests include (machine learning- the Department of Radio Communication and
based) cross-layer design and optimization for Engineering, and later with the Department of
5G/6G wireless communication systems. Electrical Engineering, Korea University, Seoul,
Republic of Korea, where he is currently a Full Professor. He has
over 200 refereed publications in international journals and conference
proceedings in the fields of mobile communication and wireless networks,
while holding over 100 patents. He also coauthored a reference textbook
on the MIMO-OFDM wireless system, entitled MIMO-OFDM Wireless
Communication with MATLAB (Wiley, 2010). His research interests include
next generation mobile radio communication systems and broadband
wireless networks with an emphasis on cross-layer design and optimization.
His recent research has been focused on massive MIMO, grant-free
non-orthogonal multiple access, dynamic spectrum access, mobile caching
networks, and AI-based communication system design. He is a Senior
MINJOONG RIM received the B.S. degree in Member of IT and VT, a KICS Fellow, and a member of the National
electronics engineering from Seoul National Uni- Academy of Engineering of Korea. He has served as the Chair for the KICS
versity, Seoul, South Korea, in 1987, and the Ph.D. Mobile Communication Technical Activity Group, the President for KICS,
degree in electrical and computer engineering and the Chair for the Wireless Technology Committee in 5G Forum of
from the University of Wisconsin–Madison, USA, Korea. He served as the Chair for the 2.3GHz IMT-WiBro Project Group
in 1993. (PG702), Telecommunications Technology Association (TTA), which is a
From 1993 to 2000, he worked with Samsung Standard Development Organization (SDO), South Korea. As a recipient of
Electronics. He is currently a Professor with the Dr. Irwin Jacobs Award sponsored by Qualcomm Inc. for his contribution
Dongguk University, Seoul. His research interests to researches on mobile communication technologies.
include mobile and wireless communications.

94032 VOLUME 9, 2021