Available via license: CC BY 4.0
Content may be subject to copyright.
Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2024.0429000
Real-time performance evaluation for 5G
multi-link communication in industrial
application
DA LIU1, YANLING ZHAO2, GONGDAI LIU2, CONGBO WANG1, LINGYUN ZHOU1, AND YING
QIAN1
1College of Information Science and Technology, Dalian Maritime University, Dalian 116026, China (email: daliu@dlmu.edu.cn)
2Instrumentation Technology and Economy Institute, Beijing 100055, China (e-mail: zhaoyanling@tc124.com)
Corresponding author: Yanling Zhao (e-mail: zhaoyanling@tc124.com).
This work was supported in part by National Key Research and Development Program of China under Grant 2021YFF0600300
ABSTRACT The rapid advancement of network communication technologies has unlocked significant
potential for advanced manufacturing industries. This new wave of industrial transformation, known as the
Industry 4.0 era, is characterized by the integration of cutting-edge technologies such as Fifth Generation
mobile communication technology (5G). Experimental commercial 5G deployments in factories have been
reported, but these applications are generally limited to specific use cases. At present, there are no foun-
dational standards for 5G in production lines, and its industrial adoption remains in the exploratory phase.
This motivates an in-depth investigation into the potential of 5G for practical applications in production
workshops, with a particular focus on evaluating its real-time performance in the presence of concurrent
operations. This research focuses on the latency and packet loss performance of multi-link communication
between multiple programmable logic controllers to evaluate the system’s real-time performance and
reliability. By using different scenarios and parameter settings, and introducing other concurrently running
services as background interference, it conducts extensive measurements and comprehensively analyzes the
behavior of the resulting data. The empirical results indicate that the latency of the 5G system increases
with higher load, particularly when concurrent services are running. However, adding more links does not
lead to a significant increase in system latency, indicating the system’s robustness. Measured results also
demonstrate the system’s high reliability in multi-link setups. The abnormal latency behavior can be caused
by increased load and background interference, emphasizing the need for enhancing 5G performance to meet
the high real-time requirements of industrial applications.
INDEX TERMS 5G, Industry 4.0, latency, multi-link, reliability, real-time performance
I. INTRODUCTION
IN response to the demands of the new wave of global
technological revolution and industrial transformation, In-
dustry 4.0 is rapidly advancing, bringing greater digitization,
automation, and intelligence to the industrial sector. The dis-
tinctions of Industry 4.0 from traditional industries are man-
ufacturing process is networked as connected cyber–physical
systems so that whole supply chain is optimized via operation
and analysis, based on large amount of produced data [1].
More precisely, it refers to the deep integration and interac-
tion between Operational Technology (OT) and Information
Technology (IT). Although wired networks are in general
used as infrastructure for equipment in shop floor such as
sensors, machines and control sector, complexities of indus-
trial environment highlight introduction of new technologies
especially in mobile and untouchable scenarios [2]. In this
context, the requirement for stationary or mobile seamless
connectivity in industrial facilities, such as equipment, in-
struments, terminals, and machinery, makes various wireless
network access technologies potential candidates.
The 5th-Generation mobile communication technology
(5G), a well-known commercial use cellular network, is nat-
urally considered as potential and promising implementation
paradigm. 5G is typically related as three scenarios, that is,
VOLUME 11, 2023 1
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2025.3539679
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
enhanced mobile broadband (eMBB), and Massive machine
type communications (mMTC), ultra reliable low latency
communications (URLLC), which are all pivotal to the appli-
cation in Industry 4.0 [3]. The eMBB is key feature for virtual
reality that is crucial function in tactile internet. The mMTC
determines the link efficiency of various of devices. Most of
all, the URLLC, which is the most attractive characteristic of
5G for industrial community, describe a gorgeous blueprint
that ubiquitous connections in the factory with negligible
latency for real-time control comes true.
As a competitor, WiFi, an IEEE standard for enterprise
connection, also have potential capabilities for industrial
manufacture. It is easy to install, simple to maintain, and
cost-effective. WiFi has evolved to its seventh generation
(802.11be), though WiFi 6 (802.11ax) remains the main-
stream choice. Despite the advantages of WiFi in implement,
it is concluded that 5G is superior in condition that network
load is increasing and circumstances is getting deteriorated
[4]. As network structures in factory become more complex
and network performance demands for industrial application
continue to rise, 5G is definitely more suitable for application
in industrial scenarios.
Since 5G New Radio (NR) begins to focus on industrial
communication, some typically scenarios are defined and
presented in 3rd generation partnership project (3GPP) stan-
dardization TR 38.901 v16.0.0 [5]. As the requirement of
5G Alliance for Connected Industries and Automation (5G-
ACIA), scenarios referred as indoor factory have been studied
in Release 16 and 17, where four sub-scenarios are further
distinguished by clutter density and the relative height of
the base station (BS). Based on these, many works have
been conducted on various conditions to analyze and evaluate
system performance for apply 5G in factory [6]–[8]. However,
these studies are all simulations that did not consider the
complex layouts and harsh operating conditions of real-world
environments. Therefore, performance testing and evaluation
of 5G systems in actual industrial settings become particu-
larly important.
The practical application of 5G in industrial control is still
in the exploratory and validation stage. Various stakeholders,
including telecom operators, equipment manufacturers, and
industrial users, are working together to assess the value of
5G technology in scenarios such as automated production
line control, coordinated equipment operation, and remote
machine control. For example, Qualcomm and Siemens have
jointly conducted evaluations and tests on the integration of
5G with industrial protocols; U.S. carrier AT&T and General
Motors have utilized 5G technology to enable partial automa-
tion in workshop control; Ericsson, Volvo, and Swedish oper-
ators have tested 5G remote control of industrial equipment;
Yokogawa Electric and NTT Docomo successfully completed
a proof-of-concept test of 5G-based remote control technol-
ogy. Nevertheless, there is still no unified standard or speci-
fication for industrial 5G applications, and extensive testing
and validation across various scenarios remain necessary [9].
Among various 5G industrial application scenarios,
URLLC is the most crucial and core one, determining whether
5G can be used, rather than how well it performs. It is
commonly with stringent key performance index (KPI) limit
for time-critical industrial use case, e.g., the controller-to-
controller latency bound is less than 100ms and service avail-
ability should satisfy more than 99.9% [10]. In spite of 5G’s
claims of 1ms latency and 99.99999% reliability, there is no
conclusive evidence demonstrating that it can achieve this
standard across various industrial scenarios and applications
[11], [12]. Moreover, end-to-end latency and reliability alone
cannot ensure consistency in multi-node coordination opera-
tion.
To bridge this gap, we aim to measure and analyze the KPIs
of 5G in actual industrial application scenarios through field
testing, in order to evaluate its practical value for deploy-
ment in factories. The experiments and measurements were
conducted on a 5G industrial testbed, focusing on multi-link
communication between Programmable Logic Controllers
(PLCs), which represent a common but critical form of real-
time communication in factories. We further investigated
the real-time performance of the 5G system under concur-
rent communication, concentrating specifically on the perfor-
mance of the studied links in the presence of other network
loads, to better assess the system’s capacity under actual
working conditions. Our goal is to obtain valuable results
through meticulously designed experiments and measure-
ment research, which will help researchers gain a deeper
understanding of the role of 5G in vertical sectors, promote
the application of 5G in the industrial field, and contribute to
the realization of Industry 4.0.
The remainder of this paper is structured as follows: Sec-
tion II provides a summary of related studies that evaluate net-
works in industrial environments. In Section III, the method-
ology is outlined, including the scenario, setup, and metrics
used to assess network performance. Section IV presents the
results and analysis, along with a discussion of the system’s
limitations. Lastly, Section V offers the conclusions.
II. RELATED WORKS
Although 5G has been commercially available for a few
years, there are few reports on its applications in vertical
domain, especially in industrial manufacture. The reasons
lie in the complexity and variability of industrial scenarios,
which make it difficult to establish unified standards. Addi-
tionally, the widespread use of traditional wired industrial net
raises concerns about compatibility with wireless networks,
and the strict requirements for real-time control also impose
challenges on the application of 5G in factory environments.
Some forerunners have already attempted to use 5G in
vertical industries to validate its potential for integration with
specific applications [13]–[15]. It is demonstrated that the
prospects for 5G applications in farming, the railway sector,
and teleoperated forestry exists. Despite these efforts showing
the availability of 5G in certain fields, there is still a lack of
detailed KPI testing specifically for real industrial scenarios
in factories.
2VOLUME 11, 2023
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2025.3539679
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
Due to the urgent demand of Industry 4.0 aligning with
5G performance, some organizations and researchers have
also investigated related issues. In addition to simulations,
there are instances in the literature discussing case studies on
integrating commercial equipment to form industrial 5G net-
works deployed in real factory environments. These studies
mainly focus on whether the KPIs of 5G meet the specific
needs of various industrial scenarios. In [16], A complete
5G network was set up in an indoor industrial scenario,
with the user equipment (UE) used as a measurement device
to test system performance of latency and packet loss. The
experiment empirically analyzed the changes in the measured
metrics under different conditions, including varying packet
sizes, whether the UE was stationary or moving, and with an
increasing number of interfering UEs. The results concluded
that 5G demonstrated low packet loss and stable latency
performance. In [17], A non-public 5G network (NPN) was
integrated with the public network (PN) as a unified system,
and tests were conducted based on this setup. The perfor-
mance metrics included throughput, latency, and energy con-
sumption. However, the system and network parameters were
fixed, and no performance analysis was conducted through
parameter variation. In factory, the traffic from various sen-
sors and monitoring systems is primarily uplink traffic, which
differs from typical 5G application scenarios. Therefore, it is
especially important to investigate the performance of uplink
traffic. 5G uplink performance in video streaming has been
evaluated in [18]. Latency and jitter of 5G uplink were tested
by varying throughput as a parameter. Moreover, KPIs were
evaluated by adjusting the data rate and introducing uplink
interference. Nevertheless, this study was not conducted in
an industrial scenario. Some literature have focused on the
impact of core network and terminal variations on network
performance. In [19], tests were conducted on different three
configurations of the core network, NPN 5G Standalone (SA),
Evolved Packet Core (EPC), and EPC+IPSec, with the same
access network being an independent radio access network
(RAN). User plane data was routed back to a cloud sever
in the laboratory. The conclusion was that the performance
of the PN 5G NSA was close to that of the NPN 5G SA.
While in the NPN 5G industrial scenario, the authors in [20]
measured the uplink and downlink throughput and round-trip
time of various commercial terminals, in the presence of other
communication interferences such as Wi-Fi. It is reported
that the terminals did not achieve the nominal performance
parameters, which was attributed to the mismatch between
the hardware and software of the terminals and the network.
Research on the use case of wireless connections for auto-
mated guided vehicle (AGV) in factory environments has also
been documented in the literature [1], [21], [22]. An overview
of the 5G-enabled mobile applications for industrial equip-
ment is presented in [1] and a lab testbed case is outlined.
In [21], mobile robots on actual production lines, which have
a wide variety of heterogeneous sensors, transmit data to a
cloud server via 5G. The test results for throughput and block
error rate are presented. In [22], for different use cases, the
one-way delay (OWD) of communication between robots and
the core network was measured, and statistical results and
analysis were provided. However, the above work did not con-
sider multi-link communication scenarios between multiple
PLCs, nor did it take into account other concurrent operations
that might be running simultaneously. Additionally, typical
result analysis usually only provides latency and packet loss
measurements, without analyzing the dynamic behavior of
the data. Therefore, this study focuses on addressing these
gaps.
III. METHODOLOGY
A. EXPERIMENT INFRASTRUCTURE DESCRI PTION
Measurement campaigns are carried out on the test bed which
is specially designed for evaluating 5G performance in indus-
trial verticals. An overview of the testbed is shown in Fig.
1. Various manufacturing equipment has been deployed in
an area with 200 m2, such as assembly unit, magnetic levi-
tation line milling machine, lathe machine, AGV and so on.
Some other key equipment that plays a supportive and com-
plementary role in industrial production are also included,
such as warehouse unit, inspection unit, robot cabinet and
so on, which contribute to forming a wireless transmission
environment with clusters in industrial scenarios.
FIGURE 1. Overview of the testbed
The lower layer of the production line equipment is con-
nected via standard wired industrial buses, such as EtherCAT,
while the transmission of control information at the upper
layer is supported by the 5G NPN. The 5G NPN is sup-
ported by China Mobile and Huawei in terms of equipment
and technology and it adopts an ’exclusive’ model, where
dedicated base stations are provided on the wireless side, as
shown in Fig. 2. The core network implements a user plane
function (UPF) edge computing and mobile edge computing
(MEC) solution to meet latency requirements. Control plane
modules and network elements utilize relevant functions from
the mobile to business network. Different priority services
VOLUME 11, 2023 3
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2025.3539679
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
are distinguished through slicing based on the 5G quality
of service indicator method, allowing for priority-based re-
source scheduling. The 5G industrial private network is built
on the 3GPP R16 version. It employs precise pre-scheduling,
traffic orchestration, accurate gate control, and data packet
redundancy scheduling, along with other network collabora-
tion technologies, to achieve end-to-end deterministic quality
of service assurance. The network operates in time division
duplex (TDD) mode in band N79, with an uplink speed of
382 Mbps and a downlink speed of 1.33 Gbps. The output
power of the deployed base station ranges from 2 dBm to 8
dBm.
FIGURE 2. Architecture of 5G NPN network
B. SCENARIO SETUP AND CONFIGURATION
In industrial environments, different applications have vary-
ing requirements for real-time performance, which are clas-
sified into two modes: soft real-time and hard real-time.
Soft real-time refers to applications with certain real-time
requirements, but these are less stringent, typically ranging
from 100 ms to 1 s, such as mobile robots, AGVs, and remote
control. On the other hand, hard real-time applies to industrial
applications with strict real-time demands, usually less than
100 ms or even lower, such as communication between PLCs
and motion control. The classification of these two types of
real-time performance is shown in Table I.
TABLE I. Real-time classification in industrial applications
Index Applications Requirement Type
1 Mobile robots and AGV 100ms-1s Soft real-time
2 Remote control 100ms-1s Soft real-time
3 Communication between
industrial controllers
10-100ms Hard real-time
4 Motion control <1ms Hard real-time
In this work it is focused on real-time control performance
of 5G in factories, and low-latency communication between
industrial facilities was evaluated, specifically centering on
the communication between PLCs. The challenges to real-
time communication between factory facilities arise from
several factors: the clustered environment of the factory floor,
which leads to non-line-of-sight conditions and time-varying
characteristics in the 5G channels; the high traffic gener-
ated by multi-link communication, which puts pressure on
medium access control (MAC) layer scheduling and causes
interference between links; and the operation of other indus-
trial equipment, which can also affect the performance of the
test devices.
In an actual factory environment, all facilities must work
together to achieve an optimal balance of low latency. In other
words, all devices should be able to establish the necessary
low-latency communication links without interfering with
each other, especially in multi-link communication scenarios.
Therefore, we set up dedicated test scenarios, where the PLCs
of the production facilities to be tested are connected to the
5G network via 5G deterministic deterministic gateways, and
the BSs operate in the 4.9 GHz frequency band are installed
on the ceiling to cover the entire experimental area. Multi-
link communication is primarily considered in testing and
evaluating the communication quality between production
facilities under various conditions, such as different transmis-
sion intervals and packet sizes. The PLC link configurations
are shown in Figure 1, where three PLCs are utilized, and
additional links between the PLCs are established to evaluate
the real-time performance under both unidirectional and bidi-
rectional communication scenarios. Specifically, additional
extensions, i.e., the interference caused by the operation of
other devices, are also incorporated into the test scenarios.
In terms of measurement, the interference from other devices
can be regarded as a type of ‘noise’, which affects real-time
performance through changes in the channel, traffic conges-
tion, and competition for network resources. It is a common
but non-negligible factor in industrial production, so each
scenario configuration is further divided into noise-free and
noise-affected states.
FIGURE 3. The setup of scenarios in the experimentk
4VOLUME 11, 2023
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2025.3539679
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
C. MEASUREMENT METHOD AND PERFORMANCE
METRICS
We mainly consider the following two metrics as evaluation
criteria for real-time system performance testing:
1) Latency
During the measurement process, the control computer con-
figures the communication parameters of each PLC via con-
figuration files and triggers the logic to start the measure-
ments. The PLCs are synchronized using network time pro-
tocol (NTP), and the latency for each link are measured
through the test logic, as shown in Figure 4. Specifically,
the control computer configures each PLC through a wired
router, and the NTP synchronization between the PLCs is
also achieved through wired connections, with PLC1 acting
as the synchronization server and PLC2 and PLC3 as clients.
When the measurement is triggered, NTP synchronization is
initiated, and during the subsequent measurement process,
periodic NTP calibration is performed to ensure the accuracy
of time synchronization.
FIGURE 4. Measurement principle diagram
Considering the most complex scenario, where all PLCs
operate in full-duplex mode and each pair of PLCs sends and
receives messages from each other, each PLC has two trans-
mission links and two receiving links. Each PLC can record
the timestamps of sent and received packets and generate logs,
which are transmitted to the computer via a wired connection.
The OWD can then be calculated as follows:
OWD (n) = Tn−Tn−1(1)
where Tnis the arrival time, and Tn−1is the transmission time.
2) Packet loss
When the reception time recorded by the PLC exceeds the
threshold or no valid packet is received, the reception time for
that packet is discarded. During log processing by the control
computer, this will be treated as a packet loss. After the
measurement for a specific scenario with predefined settings
is completed, the total packet loss can be calculated, and
packet loss rate can then be expressed as
Packet loss =Ntime_out
Ntotal
(2)
where Ntime_out is the number of packets that experienced a
timeout during reception, and Ntotal is the total number of sent
packets in a measurement campaign. Latency and packet loss
are important performance metrics for assessing whether a
system can operate stably over an extended period. Therefore,
in this study, we will evaluate the system’s performance under
various conditions, focusing on these metrics.
IV. RESULTS AND ANALYSIS
In this section, the experimental results are summarized and
analyzed to provide insights into the practical application
of 5G in industrial scenarios. Measurement campaigns are
conducted under various conditions, such as different packet
sizes and different transmission intervals, based on the setup
of six scenarios mentioned above. In order to evaluate the soft
real-time and hard real-time capabilities of 5G, the packet
intervals are set to 50 ms and 500 ms, with packet sizes
of 64 bytes and 1024 bytes, respectively, and the number
of packets sent in each experiment is 10,000. Additionally,
background noise, referring to the simultaneous 5G commu-
nication by other industrial equipment, is also considered in
the comparison of performance results. For convenience, the
’+’ symbol is used in the annotations to represent this scenario
configuration. The key performance metrics for evaluation
include latency and reliability, which are used to assess the
real-time performance of the system. In addition, the statistics
and analysis of abnormal situations help to further understand
the limitations of 5G applications in industrial environments.
A. SYSTEM LATENCY PERFORMANCE
1) Scenario 1
In this setup, only two PLCs are involved, with PLC1 trans-
mitting and PLC2 receiving, which is a commonly used sce-
nario configuration in the majority of 5G performance testing
studies. The statistical latency data for Scenario 1, represented
by CCDF curves, are shown in Figure 5. It can be seen that all
the curves have nearly the same starting point, indicating that
their minimum values, approximately 7 ms, are very close.
However, different parameter settings exhibit varying latency
distributions. For instance, in the absence of background
noise, a 50 ms interval compared to a 500 ms interval and 64
bytes compared to 1024 bytes demonstrate a more determinis-
tic latency distribution. This observation is further supported
by statistical data, where the average latencies for 50 ms/64
bytes, 50 ms/1024 bytes, 500 ms/64 bytes, and 500 ms/1024
bytes are 11.9 ms, 17.5 ms, 28.6 ms, and 39.4 ms, respectively.
The reason is that larger packet sizes require more resources
for 5G scheduling, which may lead to resource competition
and subsequently higher latency. Moreover, a smaller packet
transmission interval results in lower latency, which might
seem counterintuitive. This is because, as the transmission
VOLUME 11, 2023 5
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2025.3539679
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
interval increases, the 5G network may allocate fewer re-
sources to the connection, lowering its transmission priority.
Consequently, after resource reallocation, packets experience
longer scheduling delays. This suggests that the 5G scheduler
dynamically adapts to traffic variations to maximize its ability
to meet application requirements.
0 50 100 150 200 250 300 350 400 450 500
Lantency(ms)
10-4
10-3
10-2
10-1
100
CCDF
Scenario 1
S1_50ms_64b
S1_50ms_1024b
S1_500ms_64b
S1_500ms_1024b
S1+_50ms_64b
S1+_50ms_1024b
S1+_500ms_64b
S1+_500ms_1024b
FIGURE 5. Latency CCDF for Scenario 1
When background noise is present, all conditions exhibit
an increase in latency, particularly in the case of 500ms/1024
bytes, indicating that the operation of other devices does
indeed lead to resource competition between devices, thereby
having a non-negligible impact on the system’s real-time per-
formance. For the maximum latency, under the 50 ms setting,
it stays below 100 ms in 99.99% of the cases (1 −10−4),
which satisfies the hard real-time requirements of Type 3.
However, under the 500 ms setting, the maximum latency
reaches 262 ms, which only meets the soft real-time require-
ments of Type 2.
2) Scenario 2
This scenario setup is slightly more complex than Scenario
1, as PLC1 sends messages simultaneously to both PLC2
and PLC3. Figure 6 shows the latency CCDF, where it can
be observed that the overall latency distribution for each
setting follows the same pattern as summarized in Scenario
1. In other words, latency increases with the packet size and
transmission interval, and latency in the presence of noise
is higher than that without noise. The latency distributions
for the two links, PLC1 to PLC2 and PLC1 to PLC3, are
similar, and the statistical latency values, such as the max-
imum, minimum, and average, show no significant increase
compared to Scenario 1. This indicates that simply increasing
the number of unidirectional links has little impact on the real-
time performance of the system, which can be attributed to the
multi-connectivity capability of 5G.
3) Scenario 3
In this scenario, PLC2 and PLC3 are the transmitters, while
PLC1 is the receiver, which is the opposite of Scenario 2.
Figure 7 depicts the CCDF of measured latency. From the
figure, it can be observed that the latency distribution still
follows the previously summarized pattern. However, the
statistical mean and maximum values are slightly higher com-
pared to Scenario 2, indicating that the increase in receiving
links has minimal impact on the system’s performance. The
rise in latency originates from the processing capability of
the receiver, highlighting the need to consider not only the
network but also the performance parameters of the receiver
in 5G industrial applications. It is noticeable that the link from
PLC2 to PLC1, under the setting of 500 ms/64 bytes, exhibits
an abnormal value that even exceeds 1s. This suggests that
the 5G system may experience occasional spikes in latency,
which reduces its real-time performance. This problem will
be further discussed and analyzed in subsequent sections.
4) Scenario 4
A bidirectional link setup is added to this scenario, and the
latency CCDF is illustrated in Figure 8. It is clear that the
overall distribution follows the expected pattern. Compared
to Scenario 3, the statistical average latency does not in-
crease notably, indicating that adding transmission links has
little impact on the system. However, the latency for the
500ms/1024bytes setting increases significantly due to back-
ground noise, suggesting that when the transmission interval
is larger, the 5G system is more likely to allocate resources to
other communication tasks. It is worth noting that there are
more outliers in this scenario compared to Scenario 3, which
suggests that the addition of more links may lead to a decrease
in the system’s real-time performance.
5) Scenario 5
An additional bidirectional link is added in Scenario 5, mean-
ing that only the link from PLC3 to PLC2 is unidirectional,
whereas all other connections are bidirectional. It is observed
form CCDF illustrated in Figure 9 that the overall pattern still
holds. It is also found in the results that the link from PLC1 to
PLC3 achieves the smallest average latency compared to the
other links. This is because PLC3 has only one receiving link,
which further validates the inference that an increase in the
number of receiving links leads to higher latency. Moreover,
the number of outliers has also increased compared to the
previous scenarios.
6) Scenario 6
All the links in this scenario are bidirectional, and the CCDF
of latency is shown in Figure 10. It is similar to Scenario 5,
and the distribution pattern also follows the expected trend. In
terms of latency statistics, the mean, minimum, and maximum
values show no significant differences compared to previous
scenarios, indicating that, statistically, the 5G system can sup-
port multiple bidirectional links without a noticeable decline
in real-time performance.
6VOLUME 11, 2023
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2025.3539679
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
FIGURE 6. Latency CCDF for Scenario 2
FIGURE 7. Latency CCDF for Scenario 3
B. SYSTEM RELIABILITY PERFORMANCE
The packet loss for each scenario has been statistically ana-
lyzed, and the results are shown in Figure 11. It is evident that
packet loss occurs only in Scenario 3 and Scenario 4, with
a higher packet loss rate observed in the presence of back-
ground noise. The overall system packet loss rate is calculated
to be 0.0006875%. These results lead to three conclusions:
first, packet loss is somewhat random and does not necessar-
ily increase with network load; second, the introduction of
background noise may cause an increase in the packet loss
rate; third, the 5G system demonstrates a low packet loss rate
and high reliability in industrial applications.
C. ABNORMAL BEHAVIOR ANALYSIS
Although the distribution characteristics can represent the
overall pattern of latency, transient analysis is also essential
for investigating the system’s dynamic behavior and its im-
pact on performance. In general, the normal latency behavior,
as shown in Figure 12 (a), displays no obvious outliers and
exhibits stable characteristics, derived from the measured link
with the 50ms/64bytes setting in Scenario 2. However, an
undesirable condition may occasionally occur, as shown in
Figure 12 (b), which is derived from the 50ms/64bytes setting
in Scenario 1. It is evident that there are two outliers at the
initial stage of the measurement data. The other two abnormal
cases are shown in Figure 12 (c) and (d), derived from the
50ms/1024bytes and 500ms/64bytes settings in Scenario 1+,
respectively. The former exhibits a generally stable trend,
with occasional sudden spikes at certain points, while the
latter demonstrates a larger deviation from the overall la-
tency distribution relative to the mean, indicating a higher
standard deviation (std). To distinguish these three abnormal
VOLUME 11, 2023 7
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2025.3539679
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
FIGURE 8. Latency CCDF for Scenario 4
behaviors, we refer to them as initial spike, anomalous spike,
and large deviation. We will also discuss and analyze their
characteristics and patterns in the context of 5G industrial
applications.
1) Initial spike
It is observed that in some measurement data, there is a
significant jump at the initial stage, after which the values
transition to a normal range. To summarize the pattern of this
behavior, a threshold of 5×std above the average is set, and
the occurrences of this event under various settings in each
scenario are statistically analyzed. The results are presented
in Table II, where ‘✓’ indicates the occurrence of an initial
spike. For clarity, cases with multiple links are grouped under
the same parameter settings. In other words, if any single link
within a setting exhibits an initial spike, that parameter setting
is marked with a ‘✓’.
TABLE II. Statistics of the initial spike
S1 S1+ S2 S2+ S3 S3+ S4 S4+ S5 S5+ S6 S6+
50ms/64b ✓✓ ✓✓✓✓✓✓✓✓
50ms/1024b ✓✓✓✓✓✓✓✓✓✓✓✓
500ms/64b ✓ ✓
500ms/1024b ✓✓✓✓✓✓✓✓✓✓
It can be seen from the results that when the transmission
interval is 50 ms, an initial spike tends to occur, regardless of
the presence of background noise. However, for a transmis-
sion interval of 500 ms, initial spikes occur more frequently
when the number of links is increased, the packet size is
larger, and background noise is present. The fundamental
cause of this phenomenon is that when the transceiver initiates
uplink transmission, the 5G system must allocate resources
for it, which leads to some delay when the system load is
high. Therefore, situations that increase the system load are
more likely to produce initial spikes. It is clear that this is
detrimental to the system’s real-time performance, and as
a result, the 5G system needs to implement more efficient
scheduling to alleviate this issue.
2) Anomalous spike
Different from the initial spike, the anomalous spike occurs
at a later stage of the measurement data, and the reason for
distinguishing between the two behaviors is that their causes
are different. The cause of this anomaly may stem from
inherent flaws in the 5G system or external interference, and
it can have a significant impact on the real-time performance
of industrial systems. Similarly, a threshold of 8×std above
average is set to differentiate this anomaly, and the results
are shown in Table III. The results show that the occurrence
of the anomalous spike increases with the complexity of
the scenario, i.e., the increase in the number of links. Addi-
tionally, it is sensitive to background noise, which increases
8VOLUME 11, 2023
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2025.3539679
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
FIGURE 9. Latency CCDF for Scenario 5
the probability of its occurrence. Furthermore, when the
transmission interval decreases and the packet size increases,
the anomaly also becomes more frequent. For example, the
50ms/1024byte setting exhibits the highest occurrence of this
anomaly compared to other settings. This aligns with intuitive
expectations and proves that more complex and challenging
operating conditions degrade the real-time performance of the
5G system.
TABLE III. Statistics of anomalous spike
S1 S1+ S2 S2+ S3 S3+ S4 S4+ S5 S5+ S6 S6+
50ms/64b ✓✓ ✓✓✓✓✓
50ms/1024b ✓ ✓ ✓ ✓ ✓ ✓ ✓
500ms/64b ✓ ✓ ✓
500ms/1024b ✓ ✓
3) Large deviation
This anomalous situation is the most severe, as it not only
causes a significant increase in the mean latency but also
results in substantial jitter, which severely impacts the sys-
tem’s real-time performance. The severity of deviations can
be estimated by the standard deviation (std), and the statistical
results are presented in Table IV. It can be observed that the
std of latency follows the general pattern, increasing with
the reduction in transmission interval, the increase in packet
size, the addition of links, and the introduction of background
noise. Most of the std values for various settings in each sce-
nario align with expectations, but there are some cases where
larger values occur, such as in Scenario 1+ with the 500ms
setting, Scenario 3+ with the 500ms/1024 bytes setting, and
Scenario 4+ with the 500ms setting. These anomalies occur
in the presence of background noise, indicating that other
concurrent tasks increase the load on the 5G system, which
may impact the system’s real-time performance. Therefore,
5G should enhance network resource optimization and traffic
management, as well as improve device performance to meet
the high real-time requirements of industrial scenarios.
TABLE IV. Average of std under different parameter settings for each
scenario
S1 S1+ S2 S2+ S3 S3+ S4 S4+ S5 S5+ S6 S6+
50ms/64b 3.6 4.6 3.5 2.3 3.8 6.6 5.5 3.5 9.3 4.7 9.3 8.1
50ms/1024b 3.3 6.0 5.0 5.6 6.3 6.8 8.6 5.7 9.5 6.0 5.3 5.6
500ms/64b 12.2 48.9 18.8 13.0 19.8 12.2 14.2 27.1 14.8 18.2 11.9 18.0
500ms/1024b 13.6 47.4 18.6 47.9 13.5 51.7 15.1 33.4 16.4 18.5 13.8 17.8
D. SYSTEM LIMITATIONS
There are several limitations in this study:
Firstly, although we considered the presence of other ongo-
ing services during the operation of the 5G system, we did not
quantify their impact. Different communication applications
VOLUME 11, 2023 9
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2025.3539679
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
FIGURE 10. Latency CCDF for Scenario 6
FIGURE 11. Packet loss for each scenario
in various production processes may have different effects
on the performance of the test devices. These issues will be
addressed in future work. In this study, to simplify the experi-
ment and obtain clearer conclusions, The scenario setup con-
sists solely of the communication links between the three de-
vices. In actual factories, there are significantly more complex
link relationships. Although these links are expected to adhere
to the fundamental patterns identified in this study, their
specific performance characteristics are also worth further
exploration and investigation. The 5G deployed in the testbed
adopts a PNI-NPN architecture, where the RAN and UPF are
private, while the CPF is shared with the core network and
sliced to ensure exclusive service. Despite meeting general
latency requirements and ensuring data security and isolation,
the PNI-NPN architecture compromises performance com-
pared to the Standalone NPN (SNPN), making it difficult to
meet stringent real-time requirements such as motion control.
Therefore, further optimization of the network architecture
and the adoption of a SNPN may better fulfill the real-time
demands of industrial scenarios. Finally, all testing activities
were based on the fixed commercial equipment selected in
the testbed, and all measurement results are only applicable
to these devices. A broader set of device measurements will
be part of future work.
V. CONCLUSION
In this study, a series of measurement campaigns were con-
ducted on a 5G NPN Rel.16-based industrial testbed to eval-
uate the potential of 5G applications in industrial scenarios.
The focus was primarily on testing the latency and packet
loss performance of multi-link communication between PLCs
under various parameter settings, leading to the following
conclusions:
1) For all scenarios, latency increases as the transmission
10 VOLUME 11, 2023
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2025.3539679
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
0 2000 4000 6000 8000 10000
0
100
200
300
400
500
Lantency(ms)
(a) normal latency
0 2000 4000 6000 8000 10000
0
100
200
300
400
500
Lantency(ms)
(b) initial spike
0 2000 4000 6000 8000 10000
0
100
200
300
400
500
Lantency(ms)
(c) anomalous spike
0 2000 4000 6000 8000 10000
0
100
200
300
400
500
Lantency(ms)
(d) large deviation
FIGURE 12. Latency behavior plot
interval decreases, the packet size grows, and the number of
links increases. Latency also increases due to the influence of
background noise, especially when the packet size is larger,
which degrades the system’s real-time performance.
2) It is demonstrated that simply increasing the number of
unidirectional links has little impact on the real-time perfor-
mance of the system. For bidirectional links, an increase in re-
ceiving links has a greater impact on latency than an increase
in sending links. Nevertheless, multiple bidirectional links do
not lead to a significant increase in system latency, indicating
that the 5G system exhibits a certain level of robustness to
multi-link configurations.
3) The packet loss in the 5G system does not increase
with higher load, showing a certain level of randomness.
Background noise can cause a slight increase in packet loss.
Experimental results demonstrate that the reliability of the 5G
system is very high in multi-link setups.
4) The 5G system exhibits some abnormal behaviors, such
as initial spike, anomalous spike, and large deviation. The
occurrence of initial spikes and anomalous spikes increases
with higher system load and the introduction of background
noise, while large deviation is more sensitive to background
noise. This indicates that the 5G system should further en-
hance its performance to meet the high real-time requirements
in industrial scenarios.
The above results indicate that 5G can meet general soft
real-time requirements, and under low network load condi-
tions, it can also fulfill hard real-time requirements. In other
words, it can satisfy the general real-time communication
requirements of industrial applications with high reliability
under multi-link scenarios. However, It also has some perfor-
mance limitations, and may even experience abnormal delays,
especially when other concurrent tasks are running. More
attention should be given to these issues during the iterative
development of 5G, with targeted improvements to be made,
in order to achieve widespread industrial applications of 5G as
soon as possible. This paper aims to evaluate the performance
of 5G in multi-link industrial scenarios and serves as the
baseline for future comprehensive research work.
REFERENCES
[1] D. S. Fowler, Y. K. Mo, A. Evans, S. Dinh-Van, B. Ahmad, M. D. Higgins,
and C. Maple, ‘‘A 5G Automated-Guided Vehicle SME Testbed for Re-
silient Future Factories,’’ IEEE Open Journal of the Industrial Electronics
Society, vol. 4, pp. 242–258, 2023.
[2] S. Vitturi, C. Zunino, and T. Sauter, ‘‘Industrial communication systems
and their future challenges: Next-generation Ethernet, IIoT, and 5G,’’
Proceedings of the IEEE, vol. 107, no. 6, pp. 944–961, 2019.
[3] C. Xu, H. Yu, P. Zeng, and Y. Li, ‘‘Towards critical industrial wireless con-
trol: Prototype implementation and experimental evaluation on URLLC,’’
IEEE Communications Magazine, vol. 61, no. 9, pp. 193–199, 2023.
[4] E. A. Oyekanlu, A. C. Smith, W. P. Thomas, G. Mulroy, D. Hitesh,
M. Ramsey, D. J. Kuhn, J. D. Mcghinnis, S. C. Buonavita, N. A. Looper
et al., ‘‘A review of recent advances in automated guided vehicle tech-
nologies: Integration challenges and research areas for 5G-based smart
manufacturing applications,’’ IEEE Access, vol. 8, pp. 202 312–202 353,
2020.
[5] 3GPP Specifications, ‘‘3GPP Releases,’’ [Online].
[6] M. Cantero, S. Inca, A. Ramos, M. Fuentes, D. Martín-Sacristán, and
J. F. Monserrat, ‘‘System-level performance evaluation of 5g use cases for
industrial scenarios,’’ IEEE Access, vol. 11, pp. 37 778–37 789, 2023.
[7] R. Maldonado, A. Karstensen, G. Pocovi, A. A. Esswie, C. Rosa, O. Ala-
nen, M. Kasslin, and T. Kolding, ‘‘Comparing Wi-Fi 6 and 5G downlink
performance for industrial IoT,’’ IEEE Access, vol. 9, pp. 86 928–86 937,
2021.
[8] M. Mokhtari, G. Pocovi, R. Maldonado, and K. I. Pedersen, ‘‘Modeling
and System-Level Performance Evaluation of Sub-Band Full Duplexing
for 5G-Advanced,’’ IEEE Access, vol. 11, pp. 71 503–71516, 2023.
VOLUME 11, 2023 11
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2025.3539679
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
[9] I. Behnke and H. Austad, ‘‘Real-Time Performance of Industrial IoT Com-
munication Technologies: A Review,’’ IEEE Internet of Things Journal,
vol. 11, no. 5, pp. 7399–7410, 2024.
[10] J. Ansari, C. Andersson, P. de Bruin, J. Farkas, L. Grosjean, J. Sachs,
J. Torsner, B. Varga, D. Harutyunyan, N. König et al., ‘‘Performance of 5G
trials for industrial automation,’’ Electronics, vol. 11, no. 3, p. 412, 2022.
[11] G. Jo, J. Shin, and S. Oh, ‘‘5G URLLC evolving towards 6G: research
directions and vision,’’ in 2023 Fourteenth International Conference on
Ubiquitous and Future Networks (ICUFN). IEEE, 2023, pp. 853–855.
[12] S. Gangakhedkar, H. Cao, A. R. Ali, K. Ganesan, M. Gharba, and
J. Eichinger, ‘‘Use cases, requirements and challenges of 5G communi-
cation for industrial automation,’’ in 2018 ieee international conference on
communications workshops (icc workshops). IEEE, 2018, pp. 1–6.
[13] S. B. Damsgaard, N. J. H. Marcano, M. Nørremark, R. H. Jacobsen,
I. Rodriguez, and P. Mogensen, ‘‘Wireless communications for internet
of farming: An early 5G measurement study,’’ IEEE Access, vol. 10, pp.
105 263–105 277, 2022.
[14] D. Corujo, J. Quevedo, V. Cunha, A. Perdigão, R. Silva, D. Santos, R. L.
Aguiar, P. Paixão, P. E. Silva, R. Antunes et al., ‘‘An empirical assessment
of the contribution of 5G in vertical industries: A case for the transportation
sector,’’ IEEE Access, vol. 11, pp. 15 348–15 363, 2023.
[15] A. Mahmood, S. F. Abedin, M. O’Nils, M. Bergman, and M. Gidlund,
‘‘Remote-timber: an outlook for teleoperated forestry with first 5g mea-
surements,’’ IEEE Industrial Electronics Magazine, vol. 17, no. 3, pp. 42–
53, 2023.
[16] D. Segura, S. B. Damsgaard, A. Kabaci, P. Mogensen, E. J. Khatib, and
R. Barco, ‘‘An Empirical Study of 5G, Wi-Fi 6 and Multi-Connectivity
Scalability in an Indoor Industrial Scenario,’’ IEEE Access, 2024.
[17] J. Meira, G. Matos, A. Perdigão, J. Cação, C. Resende, W. Moreira,
M. Antunes, J. Quevedo, R. Moutinho, J. Oliveiraet al., ‘‘Industrial Internet
of things over 5G: A practical implementation,’’ Sensors, vol. 23, no. 11,
p. 5199, 2023.
[18] M. Uitto and A. Heikkinen, ‘‘Evaluating 5G uplink performance in low
latency video streaming,’’ in 2022 Joint European Conference on Networks
and Communications & 6G Summit (EuCNC/6G Summit). IEEE, 2022,
pp. 393–398.
[19] S. B. Damsgaard, D. Segura, M. F. Andersen, S. A. Markussen, S. Barbera,
I. Rodríguez, and P. Mogensen, ‘‘Commercial 5g npn and pn deployment
options for industrial manufacturing: An empirical study of performance
and complexity tradeoffs,’’ in 2023 IEEE 34th Annual International Sym-
posium on Personal, Indoor and Mobile Radio Communications (PIMRC).
IEEE, 2023, pp. 1–7.
[20] T. Lackner, J. Hermann, F. Dietrich, C. Kuhn, M. Angos, J. L. Jooste, and
D. Palm, ‘‘Measurement and comparison of data rate and time delay of
end-devices in licensed sub-6 GHz 5G standalone non-public networks,’’
Procedia CIRP, vol. 107, pp. 1132–1137, 2022.
[21] P. Sossalla, J. Rischke, F. Baier, S. Itting, G. T. Nguyen, and F. H. Fitzek,
‘‘Private 5G solutions for mobile industrial robots: A feasibility study,’’
in 2022 IEEE Symposium on Computers and Communications (ISCC).
IEEE, 2022, pp. 1–6.
[22] F. Voigtländer, A. Ramadan, J. Eichinger, J. Grotepass, K. Ganesan, F. D.
Canseco, D. Pensky, and A. Knoll, ‘‘5G for the factory of the future:
Wireless communication in an industrial environment,’’ arXiv preprint
arXiv:1904.01476, 2019.
DA LIU received the B.Eng. degree in 2007 and
the Ph.D. degree in 2013 from Dalian University of
Technology, China. He is currently a lecturer with
the College of Information Science and technol-
ogy, Dalian Maritime University, China. His cur-
rent research interests include intelligent wireless
communication technologies and systems, com-
munication signal processing, and machine learn-
ing.
YANLING ZHAO received the B.S. and M.S. de-
grees in Control Science and Engineering from
Harbin Institute of Technology, Harbin, China, in
2009. He is currently a Senior Engineer at the
Instrumentation Technology and Economy Insti-
tute in Beijing, China, where he also serves as the
Deputy Director of the Network Control Center.
His research interests include industrial control
systems, embedded software, and OPC UA.
GONGDAI LIU Gongdai Liu received the B.S. in
mechanical engineering from the Harbin Institute
of Technology, in 2014, the M.S. degree in me-
chanical engineering and M.S. degree in software
engineering from Northeastern University, Boston,
MA, in 2016 and 2019. From 2019 to 2020, he
was an integration and automation engineer at Net-
Brain Technology, Inc., Boston. Since 2021, he has
been a network development engineer in research
and development team in NetBrain. From 2024, he
is a research and development engineer in the Instrumentation Technology
and Economy Institute, China. He is the first author of two papers about
composite material during his first master degree. His research interests
include quality assurance needs of 5G industrial applications, 5G network
performance testing and data analysis and Industrial Artificial Intelligence.
CONGBO WANG received the B.S. communica-
tion engineering from Hangzhou Dianzi Univer-
sity, Hangzhou, China, in 2022. He is currently
pursuing the M.S. degree in electronic information
at Dalian Maritime University, Dalian, China. His
research interests include next-generation commu-
nication technologies and signal detection.
LINGYUN ZHOU received the B.S. electronic in-
formation from Harbin University of Commerce,
Harbin, China, in 2022. She is currently pursuing
the M.S. degree in communication engineering at
Dalian Maritime University, Dalian, China. Her
research interests include multi-agent systems and
resource allocation.
YING QIAN is an undergraduate student from
the 2022 class. Since 2024, she has been work-
ing as a research assistant in the laboratory. In
2024, she received the H Prize in the American
College Students’ Mathematical Modeling Com-
petition and won the provincial first prize in the
6th Global Campus AI Algorithm Elite Compe-
tition. Her research interests include RIS-assisted
communication systems, resource allocation, and
reinforcement learning.
12 VOLUME 11, 2023
This article has been accepted for publication in IEEE Access. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2025.3539679
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
