Delay Tradeoff and Adaptive Finite Blocklength

Framework for URLLC

Yixin Zhang, Wenchi Cheng, Jingqing Wang, and Wei Zhang

Abstract—With various time-sensitive tasks to be served, within the constraints of total energy and delay [4]. However,
ultra-reliable and low-latency communications (URLLC) has be- end-to-end delay not only includes transmission delay, but also
come one of the most important scenarios for the fifth generation contains queuing delay, processing delay, propagation delay,
(5G) wireless communications. The end-to-end delay from the
sub-millisecond-level to the second-level is first put forward for backhaul delay, routing delay, and the number of retransmis-
a wide range of delay-sensitive tasks in the future sixth generation sions caused by transmission error probability [5]. To further
(6G) communication networks, which imposes a strict require- reduce delay to meet the sub-millisecond level delay demand
arXiv:2407.12237v1 [cs.IT] 17 Jul 2024

ment on satisfying various real-time services and applications in 6G, it is necessary to comprehensively consider the impact
with different stringent quality of service (QoS) demands. Thus, of FBL on each delay component of the end-to-end delay.
we need to find out new delay reduction framework to satisfy
the more stringent delay requirements. In this article, a state- In addition, the short TTIs introduced by 5G NR are
of-the-art overview of end-to-end delay composition and delay based on a fixed blocklength structure, which impacts the
analysis combined with access protocols are elaborated. We reveal flexibility of URLLC [6]. Firstly, fixed blocklength is difficult
the tradeoff relationship among transmission delay, queuing to satisfy the diverse quality of service (QoS) needs of IoE,
delay, and retransmission times with the change of blocklength including video streaming, instant messaging, sensor data, etc.
in the finite blocklength (FBL) regime. Based on the tradeoff
and combining the grant-free (GF) random access (RA) scheme, Secondly, packet arrival is highly variable and bursty in IoE
we propose the adaptive blocklength framework and investigate scenarios, where fixed blocklength cannot be able to effec-
several effective algorithms for efficiently reducing the over-the- tively adapt to traffic changes, which leads to large delay and
air delay. Numerical results show that our proposed framework low transmission efficiency. Next, some blocks only contain a
and schemes can significantly reduce the over-the-air delay for small amount of data while the rest is idle, resulting in resource
URLLC.
waste and low resource utilization. Also, the transmission error
Index Terms—Ultra-reliable and low-latency communica- probability no longer approaches zero, thereby exacerbating
tions (URLLC), queuing delay, transmission delay, retransmis- the retransmission problem and increasing delay. Therefore, it
sion times, delay tradeoff, adaptive finite blocklength.
is necessary to consider a more flexible blocklength structure,
such as variable blocklength, to be more suitable for fast and
I. I NTRODUCTION reliable packet transmission in URLLC.
LTRA-RELIABLE and low-latency communications Considering the large number of low-latency devices
U (URLLC), the core business of the fifth-generation (5G)
communication networks, aims to provide the end-to-end delay
in IoE scenarios, we also need to consider multiple access
protocols, which lead to different delay values and retrans-
of less than 1 ms with a 99.999% success probability for mission conditions. Thus, we need to analyze the end-to-
32-bit packet transmission specified by the 3rd Generation end delay reduction problem in combination with the access
Partnership Project (3GPP) [1], enabling various low-latency protocol. The grant-based (GB) random access (RA) consists
real-time services. The upcoming sixth generation (6G) com- of four-step handshaking, which wastes time and is not feasible
munication networks are expected to support a wide range for URLLC. To improve the efficiency and simplify access
of Internet of Everything (IoE) services and applications, process, the grant request step is eliminated in the grant-free
such as autonomous driving, remote control, augmented/virtual (GF) RA, where the handshaking and waiting delay during
reality (AR/VR), factory automation, and Tactile Internet, thus the resource scheduling phase can be avoided for URLLC [7].
putting forward more stringent end-to-end delay requirement Also, since the channels are shared by multiple users, the
with sub-millisecond level for delay-sensitive tasks in the next collision introduced by GF RA also leads to the increase
generation wireless networks [2]. in retransmission times, which inevitably harms the delay
Most IoE traffics are with short packets, where the finite performance.
blocklength (FBL) theory has been proposed as one of the In general, the majority of current studies focus on the
effective methods for short packets to reduce the transmission optimization of a single delay component using a fixed block-
delay [3]. The FBL theory provides an accurate tool to describe length structure and do not consider specific access protocols,
the tradeoff among delay, reliability, and achievable rate for which leads to a lack of practical integration with specific
single short packet transmission. Introducing shorter transmis- RA protocols, joint optimization of multiple end-to-end delay
sion time intervals (TTIs) and using smaller blocklengths in components, and flexible design of blocklength structure. For-
5G New Radio (NR), the transmission delay of an individual tunately, a novel tradeoff relationship exists among transmis-
packet can be hence reduced. The joint blocklength and power sion delay, queuing delay, and retransmission times in terms
optimization method for downlink transmission schemes were of FBL, serving as a guide to the aforementioned problems.
proposed, aiming to minimize the decoding error probability To mitigate the combined impact of delay components on end-
 2

Backhaul delay
Core network
Active users Retransmission times

Over-the-air delay
Over-the-air delay M
=  DSingle access delay (m)
SigAcc ( m)
m =1
BS
BS
Terminal
Routing delay

Queuing delay Transmission delay Processing delay Propagation delay

Upper Fixed blocklength

layer Server 01
arrival
Adaptive blocklength
traffic Coding Modulation

Data-link layer Physical layer Wireless channel

Fig. 1: The end-to-end delay components.

to-end delay as blocklength varies, the joint delay optimiza- and the number of access attempts, i.e., retransmission times.
tion scheme and flexible blocklength framework are highly Specifically, the single access delay consists of transmission
demanded. Therefore, we need to comprehensively analyze all delay, queuing delay, processing delay, and propagation delay.
related delay components and access protocols to consider the In Table I, we provide the existing delay reduction technolo-
impact of blocklength for the new delay reduction framework. gies of over-the-air delay components.
The rest of the article is structured as follows. We
provide a comprehensive review of end-to-end delay with B. Access Protocol-Specified Delay
access protocol for low-latency communications. Then, we In real scenarios, since the retransmission times, specific
reveal the new tradeoff in the FBL regime. We also propose the delay components, and delay values are related to different
adaptive blocklength framework and the corresponding delay access protocols, it is necessary to analyze the access protocols
minimization problem. On this basis, we present a case study to obtain the specific delay of real-time services. In Table I, we
and show novel delay-related research directions. Finally, the provide the specific access delay composition and comparison
conclusion is given. of GB and GF RA protocols.
1) The Delay in GB RA: Table I shows the access delay
II. T HE D ELAY C OMPOSITION A ND ACCESS P ROTOCOL components in GB RA. The scheduling request (preamble)
A. The Components Of End-to-end Delay and contention resolution procedure in GB RA lead to an
Figure 1 shows the various end-to-end delay components additional access delay of at least several milliseconds within
in URLLC scenario. The user plane end-to-end delay con- the Long-Term Evolution (LTE) network configurations, which
sists of the over-the-air delay, the routing delay in the core is unsuitable for low-latency communications.
networks, and the backhaul delay [8]. The backhaul delay 2) The Delay in GF RA: Due to the characteristic of
refers to the time taken for a packet to be transmitted from intermittent and bursty traffic regarding delay-sensitive tasks,
the core network to the small subnetworks at the edge of GF RA is proposed as an efficient transmission technology
the network, and this duration is influenced by the physical to support low-latency connectivity. As shown in Table I, in
distance from the base station (BS) to the core network as GF RA, unlike GB RA, active users are not required to wait
well as the type of physical medium used. The utilization for a grant from the BS, which significantly reduces delay by
of fiber optics significantly reduces the backhaul delay to a eliminating the need for the transmission grant process. This
duration of less than 1 millisecond [9]. The routing delay means that as soon as a user becomes active, it immediately
denotes the duration required for a packet to travel through sends data along with a preamble and awaits acknowledgment
routing paths within a network or across multiple networks. (ACK) from the BS, thereby bypassing the grant step and
Software-defined routers (SDRs) have become an effective consequently reducing delay [11].
approach to achieve a relatively routing delay while offering
programmable capabilities [10]. The over-the-air delay, which C. Existing Problems
is a very important part as well as the bottleneck of end-to- 1) Separate Delay Components Optimization: For the
end delay for URLLC, is influenced by the single access delay conventional over-the-air delay optimization problem, each de-
 3

TABLE I: Reduction technologies of over-the-air delay components and the corresponding access protocol-determined delay.
Over-the-air delay Access protocol-determined delay
Reduction technologies
components Description GB RA GF RA
Increase the subcarrier spacing (SCS): Number
2 1
Transmission to reduce the symbol duration. of times
Shortened TTIs
delay Mini-slot: to reducethe number of
1 1
symbols per TTI. Use Data
Based on FBC theory. Preamble Data
Average Queuing Markov queuing model Number
1 0 0
queuing theory Two-dimensional of times
Queuing
delay Markov chain
delay
Resource
Upper-bound Large Based on Queue Immediate data
allocation Queue is
of queuing deviation the concept Situation is not transmission
Scheduling empty.
delay principle of effective empty. without waiting.
scheme
(LDP) capacity.
Heterogeneous
QoS
provisioning
Number
Mobile edge computing (MEC) for intelligent and fast responses 3 1
Processing of times
delay Distributed caching/control: to reduce redundant data
2 1
traffic and user content access delay. Location At BS
Parallel-processing strategy: to improve the efficiency
At BS At user
of cache servers.
Number
Non-terrestrial networks for global coverage 4 2
of times
Propagation 1 1 1
Prediction and communication co-design for long-distance URLLC
delay Access
Use Preamble Data
response
Non-geostationary (NGSO) satellite constellations for 1 1 1
satellite communications Contention Access
Data
resolution response
Access strategy: to reduce RA conflict.
Packet duplication
Number
Expanded channel state information (CSI) Low High
Reliability of times
Retransmission feedback
times Slot aggregation
Multiple transmission and reception
points (TRPs) support
Flexible HARQ round triptimes (RTT)
Hybrid automatic
configuration
repeat request Resolution Contention-free Contention-based
Low-order modulation and coding scheme
(HARQ)
Adaptive redundancy matching with
enhancements
enriched feedback

lay component is separately optimized as an independent part shown tradeoff, new system delay performance can be derived
as shown in Table I. Optimizing a single component of the end- in the FBL regime and the new frame structure combined with
to-end delay separately lacks systematic and comprehensive specific access protocols can be accordingly redesigned.
consideration.
2) Fixed Blocklength Structure: The TTI design of LTE
A. The Structure And Impact of FBL
and 5G NR are both fixed, which limits the flexibility of
system design for transmission delay reduction. When hetero- The blocklength refers to the number of symbols trans-
geneous packets with different delay requirements and packet mitted within a block, and it is determined by multiplying
sizes arrive at the same time, fixed frame structure is difficult the time span (TTI) and the subchannel bandwidth resource.
to adapt to variable delay and traffic situations. Short packets in URLLC correspond to finite blocklengths,
3) Lack of Specific Access Protocol Combined Analysis: and the corresponding achievable rate in the FBL regime has
Different access protocols relate to different kinds of compo- been derived. It is a function of blocklength and decreases as
sition regarding transmission delay, queuing delay, processing the blocklength decreases [3]. Especially in areas where the
delay, and propagation delay, as well as different retransmis- blocklength is short, the achievable rate is severely impacted.
sion times. However, existing research lacks delay analysis Different from the traditional long-frame structure, in the case
combined with specific access protocols, which cannot demon- of short packets with short-frame structure, the influence of
strate the overall end-to-end delay performance of the system. FBL on system performance should be fully considered.

III. D ELAY T RADEOFF R ELATIONSHIP B. Tradeoff Relationship Among Transmission Delay, Queuing
IN T HE FBL R EGIME Delay, and Retransmission Times
To solve the above problems, we delve into the delay 1) IBL Regime: In the traditional IBL regime, the achiev-
tradeoff relationship in the FBL regime. Based on the newly able rate remains constant regardless of the blocklength. As
 4

Transmission delay
optimal blocklength. In order to achieve the optimal tradeoff
for minimizing over-the-air delay, the following aspects should

？
The overall
Blocklength Queuing delay over-the-air be taken into account.
delay
Retransmission times C. Factors for Achieving Optimal Tradeoff
In order to obtain the optimal tradeoff relationship to
Optimal ✔The minimum
Optimal tradeoff reduce the over-the-air delay, the following three aspects
blocklength over-the-air delay
should be considered.
1) Adaptive Blocklength Framework: Even the scalable
Adaptive blocklength slot based scheduling can simultaneously support a variety of
applications, the TTI is still fixed, which prevents further re-
duction of the over-the-air delay. Based on the above analysis,
Adaptive bandwidth

Maximum time period

the change of blocklength per frame impacts the over-the-
User K
air delay. Under different packet arrival rates, packet sizes,

Total bandwidth
Block 1 Block 2 Block L
and traffic loads, the optimal tradeoff among transmission
≈

≈
User 2 delay, queuing delay, and retransmission times cannot be
Block 1 Block 2 Block L
achieved with a fixed blocklength-based frame structure. This
≈

User 1 Block 1 Block 2 Block L

situation results in an increase in the overall over-the-air delay.
≈

Adaptive TTI
Therefore, it is necessary to study a new frame framework with
adaptive blocklength to find out the optimal blocklength.
Packet 2) Joint Optimization of Different Delay Components:
Queuing buffer Transmission
(Variable TTI)
Based on the tradeoff analysis, the optimization of a single
Variable TTI-based queuing

delay component may lead to the increase of other delay

Packets arrival

Block 2 Block 1
process of user 1

components. Thus, each delay part should be combined with

other delay components to form a comprehensive delay anal-
Block 4 Block 3
ysis and joint optimization method. To achieve the lowest
over-the-air delay, it is necessary to jointly optimize different
delay components to find the optimal global solution instead
of treating them as independent parts.
Time
D. Applicability for Short And Long Packets Coexistence
Fig. 2: Tradeoff relationship, adaptive blocklength framework, In different types of low-latency tasks, real-time inter-
and variable TTI-based dynamic queuing process. actions are operated using text, speech, images, and aug-
mented/virtual reality, which relate to different packet types,
packet sizes and delay requirements. Based on real-time packet
a result, the service rate of the queuing process in the link arrival rates and task loads, the proposed adaptive blocklength
layer does not change, leading to an unchanged queuing framework can flexibly adjust the blocklength for both short
delay. Furthermore, since the transmission error probability and long packets to accommodate the diverse delay require-
approaches zero in the IBL regime, the variation in blocklength ments of various tasks. The specific implementation strategy
does not affect retransmission times. Only the transmission is in the following.
delay changes with the blocklength, decreasing as the block- 1) Traffic Monitoring and Analysis: Deploy traffic mon-
length decreases. Consequently, the over-the-air delay also itoring systems to analyze the characteristics of real-time
monotonously decreases with the reduction in blocklength. traffic, including the size of packets, transmission frequency,
2) FBL Regime: Fig. 2 shows the change of transmission and service type. Based on the results of traffic analysis,
delay, queuing delay, and retransmission times with the finite dynamically adjust the blocklengths of packets.
blocklength. In the FBL regime, a reduction in blocklength 2) Performance Testing and Feedback Mechanism: Es-
leads to a decrease in transmission delay. However, this tablish a real-time feedback mechanism to adjust the strategy
reduction in blocklength also leads to a decrease in the according to changes in network conditions, ensuring that
achievable rate, which in turn causes an increase in queuing the network operates optimally and enabling assessment and
delay. Besides, unlike the assumption in the IBL regime, the further optimization of the improvements based on test results.
FBL leads to additional error probability, which no longer By adaptively adjusting the blocklengths of packets,
approaches zero. As the blocklength decreases, the error prob- better QoS can be provided according to the needs of different
ability increases, leading to more frequent retransmissions. applications, especially for applications sensitive to delays.
The transmission delay, queuing delay, and retransmission
times do not change in the same direction with the blocklength, IV. A C ASE S TUDY: A DAPTIVE B LOCKLENGTH S CHEME
which reveals a tradeoff relationship among them. The min- FOR URLLC
imum over-the-air delay can be achieved when the optimal In this section, we propose a dynamic queuing process
tradeoff is reached, and the corresponding blocklength is the based on variable TTI and the adaptive blocklength frame-
 5

work. Then, we carry out a case study to solve the problem D. Multi-User Scenario with OMA Scenarios
of minimizing the over-the-air delay. 1) Joint Adaptive Bandwidth and TTI-Based Average
Over-the-Air Delay Minimization Problem: Different from the
A. Variable TTI-based Dynamic Queuing Process single-user or NOMA scenarios, multiple users share multiple
In contrast to the fixed TTI structure in 5G NR, we subchannels. According to the relationship between block-
set a variable TTI for each packet, as illustrated in Fig. 2. length and time-bandwidth resource, optimizing blocklength is
Depending on the packet arrival rates and traffic load con- equivalent to optimizing time-bandwidth resource allocation.
ditions, different TTIs are set when the packets are in the Therefore, in the multi-user with OMA scenario, an additional
queuing buffer, and the packet transmission stage also features constraint needs to be added that the bandwidth resource of
different TTIs. The TTIs are considered to be changed, where each subchannel can only be allocated to one user.
the corresponding blocklengths are also changed. Besides, the corresponding problem in the multi-user with
OMA scenario is more complicated due to the discreteness
B. Adaptive Blocklength Framework in The FBL Regime for of bandwidth allocation brought by subchannels. Thus, this
URLLC average over-the-air delay minimization problem includes con-
tinuous non-convex and discrete constraints, which cannot be
To achieve the optimal tradeoff relationship among trans-
solved using the traditional mathematical methods. A series
mission delay, queuing delay, and retransmission times, we
of deep reinforcement learning (DRL)-based algorithms can
consider optimizing the blocklength of each frame. Given that
be used to tackle this non-convex problem. Since the off-
the blocklength is the product of TTI and bandwidth, optimiz-
line training process is done before being implemented into
ing blocklength is equivalent to optimizing time-bandwidth
the system, it can be applied to low-latency communication
resource allocation. In this case, it is necessary to jointly
systems.
adjust these two factors to create the corresponding adaptive
2) DDPG: Based on the algorithm idea of deterministic
blocklength. That is, in addition to the above variable TTI,
policy gradient (DPG) and deep learning, deep deterministic
we also allocate bandwidth, that is, we assign subchannels to
policy gradient (DDPG), with dual network and experience
different users.
replay, can solve the problem that the DRL network is difficult
Fig. 2 illustrates the framework for adaptive blocklength,
to converge [12]. DDPG can be used to solve the adaptive
where each packet has different TTI and bandwidth configura-
bandwidth and TTI-based delay minimization problem in the
tion. In the frequency domain, the total bandwidth is allocated
multi-user scenario.
to all users, thus forming adaptive bandwidth. Similarly, in 3) Multi-DQN: Two subproblems need to be solved here,
the time domain, the maximum time period is divided into one is the bandwidth allocation problem, and the other is
several parts, allocated to packets of each user, thereby creating the TTI design problem. Even a small number of users,
adaptive TTIs. Through the adaptive TTI and bandwidth, the subchannels and frames can lead to a large action space.
blocklength of each frame can be adaptively changed, where The curse of dimensionality in the action space is a serious
the adaptive blocklength framework is formed. In addition to problem in the deep Q-network (DQN) since the exponentially
the above framework, which contributes to the improvement of increasing action space makes DQN quite complex and sample
delay performance, we set a fixed threshold for transmission inefficient, where the corresponding convergence speed will
error to simultaneously ensure high reliability in URLLC. be slowed down. Thus, Multi-DQN is used to decompose the
large action space. Using two DQNs to allocate bandwidth and
C. Singer-User Or NOMA Scenarios design TTI respectively, Multi-DQN can avoid large action
1) Adaptive TTI-Based Over-the-Air Delay Minimization space and low convergence speed.
Problem: To minimize the over-the-air delay, based on the 4) Performance Comparison of DDPG And Multi-DQN:
revealed tradeoff relationship, we consider optimizing the Figure 3 shows the convergence comparison of DDPG, Multi-
blocklength of each frame to balance these three delay compo- DQN, and Single-DQN for multi-user adaptive blocklength
nents. In single-user or NOMA scenarios, a fixed bandwidth scheme. As illustrated in Fig. 3, the reward of Single-DQN
is adopted without loss of generality, where the bandwidth does not reach those of Multi-DQN and DDPG. Multi-DQN
resources have been normalized. This means that adaptive is capable of matching the reward of DDPG, indicating that
blocklength is equal to adaptive TTI. Multi-DQN can achieve the performance limit set by DDPG.
We set the average over-the-air delay as the objective Additionally, Multi-DQN converges faster than DDPG, with
function of the problem. Then, we set up three constraints. the convergence time for Multi-DQN being approximately
First, the total number of bits contained in serviced packets 1/10 of that for DDPG. This means Multi-DQN can degrade
needs to be transmitted within the current period to ensure the dimension of action space and fast converge to optimal
reliability. Second, the sum of blocklength for all frames solution compared with Single-DQN and DDPG, implying
cannot exceed the period duration to avoid packet loss. Third, reduced energy consumption for calculation is required due
the blocklength of each frame should be set larger than or to fewer iterations in Multi-DQN. For future URLLC and IoT
equal to zero. The variable we optimize is the blocklength of scenarios, the low power consumption characteristic of Multi-
each frame. Using the adaptive blocklength framework, we can DQN can reduce the energy consumption caused by a large
dynamically adjust the blocklength to reduce the over-the-air number of sensors and connectivities. Therefore, we choose
delay. Multi-DQN to achieve the adaptive blocklength scheme.
 6

2.25
LTE
2 5G NR (0.5ms)
100 5G NR (0.25ms)
1.75 5G NR (0.125ms)

Average over-the-air delay (ms)

5G NR (0.0625ms)
80 Adaptive blocklength scheme
1.5

60 1.25
Reward

1
40
Multi-DQN 0.75

20 Single-DQN
Without adaptive bandwidth
DDPG 0.5

With adaptive bandwidth

0 0.25

0 100 200 300 400 500

Training steps 0
3 3.5 4 4.5 5 5.5 6 6.5 7
Total time (ms)
Fig. 3: The convergence comparison of DDPG, Multi-DQN,
(a) Delay performance comparison versus total time.
and Single-DQN.
30
GF-based adaptive blocklength scheme
GF-based 5G NR (0.5ms)
E. Delay Performance Evaluation 25
GF-based LTE
GB-based adaptive blocklength scheme
We consider the URLLC scenario, where packets arrive

Average over-the-air delay (ms)

with a size of 32 bytes. The maximum transmit power per
user is set to 100 mW and the total bandwidth is set to 20

1 MHz. The channel gain for each user is set in the range
of [−20dB, 20dB]. In order to satisfy the high-reliability 15
requirements in URLLC, the transmission error probability is
set to 10−7 . In addition, the duration of one block in 5G NR is
set as 0.5 ms, 0.25 ms, 0.125 ms, and 0.0625 ms, respectively. 10

The duration of one block in LTE is set as 1 ms.

Figure 4(a) depicts the delay performance of the proposed
5
adaptive blocklength framework and the fixed blocklength 5 10 15 20 25 30 35 40 45 50

structures in LTE and 5G NR. Two performance indicators Number of users

are evaluated, namely the average over-the-air delay and the (b) Delay performance comparison versus the number of users.
total time, where the total time is the time consumption to
transmit all packets. The average over-the-air delay and the Fig. 4: Delay performance comparison.
total time of the proposed adaptive blocklength framework
are both less than those in LTE and 5G NR. Also, Fig. 4(a)
demonstrates that if only the blocklength adaptively changes protocol, we obtained the tradeoff relationship among trans-
with fixed bandwidth, the corresponding average over-the-air mission delay, queuing delay, and retransmission times. On
delay is longer than that of LTE and 5G NR with adaptive this basis, we developed the adaptive blocklength framework
bandwidth allocation, indicating the importance of adaptive and corresponding case study, which can help to achieve the
bandwidth allocation in our proposed framework. optimal balance of the tradeoff. Compared to the fixed block-
Figure 4(b) shows the delay performance comparison in length structures in LTE and 5G NR, our proposed framework
GF RA and GB RA under different numbers of users. The can reduce the over-the-air delay and increase flexibility in
over-the-air delay of GB-based scheme is significantly higher system design. We hope this article guides the researchers and
than that of GF-based scheme, which means GF RA is more practitioners interested in applying the adaptive blocklength
suitable for URLLC. With the same number of users, GF-based framework to related issues for URLLC.
adaptive blocklength schemes can attain the lowest average
over-the-air delay compared to LTE and 5G NR. And when the R EFERENCES
number of users changes, the blocklength can be dynamically [1] 3GPP Tech. Rep. 38.913 v16.0.0, “Study on Scenarios and Requirements
adjusted to accommodate real-time tasks. for Next Generation Access Technologies,” July 2020.
[2] W. Saad, M. Bennis, and M. Chen, “A Vision of 6G Wireless Systems:
Applications, Trends, Technologies, and Open Research Problems,”
V. C ONCLUSIONS IEEE Network, vol. 34, no. 3, pp. 134–142, 2020.
[3] Y. Polyanskiy, H. V. Poor, and S. Verdu, “Channel Coding Rate in the
In this article, a new delay reduction framework, namely Finite Blocklength Regime,” IEEE Transactions on Information Theory,
the adaptive blocklength framework, was proposed for URLLC vol. 56, no. 5, pp. 2307–2359, Apr. 2010.
to support delay-sensitive services. By comprehensively ana- [4] H. Ren, C. Pan, Y. Deng, M. Elkashlan, and A. Nallanathan, “Joint
Power and Blocklength Optimization for URLLC in a Factory Automa-
lyzing related components of the end-to-end delay, the FBL tion Scenario,” IEEE Transactions on Wireless Communications, vol. 19,
transmission theory, and delay combined with the access no. 3, pp. 1786–1801, 2020.
 7

[5] J. Cao, X. Zhu, S. Sun, Z. Wei, Y. Jiang, J. Wang, and V. K. Lau, “To-
ward Industrial Metaverse: Age of Information, Latency and Reliability
of Short-Packet Transmission in 6G,” IEEE Wireless Communications,
vol. 30, no. 2, pp. 40–47, 2023.
[6] F. Hamidi-Sepehr, M. Sajadieh, S. Panteleev, T. Islam, I. Karls, D. Chat-
terjee, and J. Ansari, “5G URLLC: Evolution of High-Performance
Wireless Networking for Industrial Automation,” IEEE Communications
Standards Magazine, vol. 5, no. 2, pp. 132–140, 2021.
[7] A. C. Cirik, N. M. Balasubramanya, L. Lampe, G. Vos, and S. Bennett,
“Toward the Standardization of Grant-Free Operation and the Associated
NOMA Strategies in 3GPP,” IEEE Communications Standards Maga-
zine, vol. 3, no. 4, pp. 60–66, 2019.
[8] X. Jiang, H. Shokri-Ghadikolaei, G. Fodor, E. Modiano, Z. Pang,
M. Zorzi, and C. Fischione, “Low-Latency Networking: Where Latency
Lurks and How to Tame It,” Proceedings of the IEEE, vol. 107, no. 2,
pp. 280–306, Aug. 2019.
[9] C. She, C. Yang, and T. Quek, “Radio Resource Management for Ultra-
Reliable and Low-Latency Communications,” IEEE Communications
Magazine, vol. 55, no. 6, pp. 72–78, June 2017.
[10] H. Fatemidokht, M. K. Rafsanjani, B. B. Gupta, and C.-H. Hsu,
“Efficient and Secure Routing Protocol Based on Artificial Intelligence
Algorithms With UAV-Assisted for Vehicular Ad Hoc Networks in
Intelligent Transportation Systems,” IEEE Transactions on Intelligent
Transportation Systems, vol. 22, no. 7, pp. 4757–4769, 2021.
[11] Y. Liu, Y. Deng, M. Elkashlan, A. Nallanathan, and G. K. Karagiannidis,
“Analyzing Grant-Free Access for URLLC Service,” IEEE Journal on
Selected Areas in Communications, vol. 39, no. 3, pp. 741–755, Aug.
2021.
[12] Y. Liu, J. Yan, and X. Zhao, “Deep Reinforcement Learning Based
Latency Minimization for Mobile Edge Computing With Virtualization
in Maritime UAV Communication Network,” IEEE Transactions on
Vehicular Technology, vol. 71, no. 4, pp. 4225–4236, 2022.