URLLC for 5G and Beyond: Physical, MAC, and Network Solutions

The University of Sydney
URLLC for 5G and Beyond:
Physical, MAC and Network
Solutions
2020 IEEE WCNC Tutorial
Yonghui Li
Mahyar Shirvanimoghaddam
Rana Abbas
Changyang She
Centre of IoT and Telecommunications
School of Electrical and Information Engineering
Faculty of Engineering and Information Technologies
*A modified version of this tutorial was presented at IEEE
Globecom, Kona, Hawaii, USA Dec 2019

About Us

University of Sydney
Australia's oldest university established in 1850, ranked 36th in World
University Rankings by the Times higher Education Supplement (UK)

University of Sydney
Breakthrough Innovations
Ultrasound scanner, George Kossof
Artificial Heart Pacemaker, Edgar Booth
Ventilator
Cochlear, Graeme Clark
Aircraft black box
WI-FI
...

– A world renown and largest Research Centre in
Australia in Telecommunications
– Mission: provide innovative solutions to
the telecommunication industry
– R&D team:
– 7 Academic staff (3 IEEE Fellows)
– 6 Research Fellow
– More than 40 PG students
– Areas: 5G, uRLLC, Networks, Industrial Internet
of Things, Signal processing, Wireless
Networked Control, AI in Wireless
Communications
– Activities: Research, product development,
system design and industry consultancy
– Publications: >200 IEEE journal papers in the
past five years

Over 40 industry projects, more than $20million
Design of last mile and home area networks for $500 million Australian Smart Grid Smart
City project
Participate in 7-years’ $100million SmartSat CRC project
Contributions to the IEEE Standard (802.X)
Soft output detection and decoding used in 2G, 3G and 4G base stations and terminals
Adaptive CDMA receivers used in 3G cellular systems
Soft frequency reuse has been adopted by 3GPP standard
Patents: 15 (wireless system architecture, solutions, protocols, core technology)

Presenters
Dr. Mahyar
Shirvanimoghaddam
Dr. Rana Abbas Dr. Changyang SheProf. Yonghui Li

Introduction to URLLC & Timeliness
Professor Yonghui Li

Introduction to URLLC& Timeliness
→ 5G: From H2H to M2M
What is URLLC?
Why do we need URLLC?
Latency and Reliability Components
How do we achieve URLLC?

G. Fettweis and S. Alamouti, “5G: Personal mobile internet beyond: What cellular did to telephony,” IEEE Communications Magazine, February 2014.
Kumar, Liu, Sengupta and Divya, “Evolution of Mobile Wireless communication Networks: 1G to 4G,” IJECT, December 2010. 4GAmericas.org, “Infographic:
Mobile Broadband Connected Future,” White, “Tablets trump smartphones in global website traffic” Adobe Digital Marketing blog, March 6, 2013. Meeker and
Wu, “Internet Trends 2013” kpcb.com. Ericsson Consumer Insight Summary Report, June 2013
Cellular Network Development
Wireless Moore’s
Law

5G Paradigm Shift:
From H2H to IoT M2M Communications

M2M is Everywhere

https://www.accenture.com/t00010101t000000z__w__/it-it/_acnmedia/pdf-5/accenture-industrial-internet-of-things-positioning-paper-report-2015.pdf
Industrial Internet of Things (IIoT)

What is M2M ?
M2M = Sensors + Connectivity + Intelligence
Machine – To – Machine (M2M) means no human
intervention whilst devices are communicating end-to-end.
• Support of a huge amount of nodes, sending small data each
• Mission-critical data provision
• Autonomous operation
• Self-organization
• Power efficiency
• Reliability
• etc.
This assumes some fundamental M2M system
characteristics:

✔ 5G: From H2H to M2M
→ What is URLLC?

IIoT Communication Network Requirements
Image Source: G Empowering vertical industries. White Paper, 2016, https://5g-ppp.eu/wp-content/uploads/2016/02/BROCHURE_5PPP_BAT2_PL.pdf
• High range of reliability from medium to
ultrahigh, PER 10-3 to 10-9
• High capacity
• High range of data rates, 1kbps to 10Gbps
• High range of latencies from 200ns to 1000s
• Low cost - $1 hardware, $1/year connectivity
• Flexible access
• High range of the number of nodes, from 100 to
1 million per cell
• Long battery life >10 years
• Message prioritisation
• Internet protocol (IP)
• Standard-based
• Ubiquitous coverage
• High security
• Fast configuration
• No current network meeting these
requirements!

5G Technical Requirements
Image Source: 5G Americas
– 1000 times higher mobile data capacity per cell
– 100 times higher typical user data rate
– 100 times higher number of connected devices
– 10 times longer battery life for low power M2MC
– 10 times reduced latency

URLLC for Mission Critical IoT
Ultralow latency <1ms air link; <10ms E2E
Ultra reliable networks
No wireless standard available
Image Source: LoRa Alliance

Industrial Network Requirements
– Ultra-low latency
– Link level latency <1ms vs 100ms in 4G
– Ultra-high reliability
– Packet error rate < 10-9 vs 10-2 in 4G
– Determinism
– Criticality
– Scalability
Image Source: https://www.roboticsbusinessreview.com/sme-building-a-smart-factory-with-ai-and-robotics/

Image Source: https://andersontech.com/three-critical-reasons-choose-wired-network-small-business/
Wireless vs Wired Networks for IoT
– Currently wired networks are
dominant
– Trends towards wireless due to
– lower installation cost
– lower maintenance
– easier redundancy
– higher flexibility
– enable mobile applications
– even higher long-term reliability

Introduction to URLLC & Timeliness
✔ What is URLLC?
→ Why do we need URLLC?

– Industry Automation URLLC is
one of the enabling
technologies in Industry 4.0
– Industrial control is automated
by deploying networks in
factories which are
traditionally wired.
– End-to-end latency lower than
0.5 ms and an exceedingly
high reliability with BLER of
10–9 should be supported.
Industrial Automation

IoT in Industrial Automation
– Current industrial robots form a closed
system of manipulators and controllers
– They are connected via wired networks
– Expensive set-up
– New architecture (fog and roof)
enables their separation
– High performance wireless networks
enable their connectivity
– Controllers located in general purpose
computers

Wireless Networked Control
– Enabling mobile robotics
– Enabling a pool of controllers
– Enabling High-Precision Robotic Arms

Why do we need URLLC ?
– Automated driving
– Road safety
– Traffic efficiency services
– The typical use cases of this
application are automated
overtake, cooperative collision
avoidance and high-density
platooning, which require an
end-to-end latency of 5–10 ms
and a BLER down to10–5 [S-2].Fully connected vehicles can be enabled with URLLC
Image Source: Wikipedia
Intelligent Transportation Systems

– Remote surgical consultations:
– Can occur during complex life-saving
procedures after serious accidents with the
patient having a health emergency that
cannot wait for transporting to a hospital.
– First-responders at an accident venue may
need to connect to surgeons in a hospital to
get advice and guidance to conduct complex
medical operations.
– Remote surgery:
– The entire treatment procedure of patients is
executed by a surgeon at a remote site,
where hands are replaced by robotic arms.
Example of remote surgery with robotic arms
Image Source: BMC Biomedical Engineering
Telesurgery

URLLC Applications, Requirements & Markets
Lema, Maria A., et al. "Business case and technology analysis for 5G low latency applications." IEEE Access 5 (2017):
5917-5935.

What levels of latency and reliability do we need?
[S-2] Chen, He, et al. "Ultra-reliable low latency cellular networks: Use cases, challenges and approaches." IEEE
Communications Magazine 56.12 (2018): 119-125. https://doi.org/10.1109/MCOM.2018.1701178

✔ What is URLLC?
✔ Why do we need URLLC?
→ Latency and Reliability Components

Reliability in URLLC
What kind of reliability are we after?
• Reliability is defined as the probability that the latency does not exceed a
predetermined deadline (1 – Pe)
• Pe is the probability of a packet drop/packet error = the probability that the
latency exceeds a predetermined deadline
• The exact deadline and the reliability level are application-dependent.
Popovski, P., Nielsen, J.J., Stefanovic, C., De Carvalho, E., Strom, E., Trillingsgaard, K.F., Bana, A.S., Kim, D.M.,
Kotaba, R., Park, J. and Sorensen, R.B., 2018. Wireless access for ultra-reliable low-latency communication: Principles
and building blocks. Ieee Network, 32(2), pp.16-23.

Latency in URLLC
Latency Components

Latency in URLLC
Latency in Wireless Networks

Latency in URLLC
– Fundamental trade-offs between capacity, coverage, latency,
reliability and spectral efficiency….
– One metric is optimized for improvement à degradation of another
metric
– High capacity needs large control overhead (e.g., cyclic prefix,
transmission mode, and pilot symbols); this makes the portion of
overhead unacceptably high in shorter TTI.
– In LTE, packet retransmission takes around 8ms, and removal of
retransmission will affect packet error significantly.
Constraints for achieving low latency

✔ What is URLLC?
✔ Why do we need URLLC?
✔ Latency and Reliability Components
→ How do we achieve URLLC?

URLLC
PHY
Theoretical Bounds
Channel code
candidates
Potential of
Fountain Codes
Advanced
Ordered Statistic
Decoding Scheme
MAC
Grant-Based
Orthogonal Access
Non-Orthogonal
Access (NOMA)
Grant-Free Access
Cross-Layer
Design
Network
Network
Availability
Multi-connectivity
in Terrestrial
Communications
Mobile-Edge
Computing
Systems
Deep Learning for
Network
Management

PHY
Theoretical
Bounds
Channel code
candidates
Potential of
Fountain Codes
Advanced
Ordered Statistic
Decoding Scheme
• Revisit the Shannon
Bound
• Normal Approximation
• Turbo Codes
• Convolutional Codes
• BCH Codes
• LDPC Codes
• Polar Codes
• Approach near MLD
Performance
• Reduced Complexity
• Reduced Decoding
Latency
How do we address URLLC at the Physical Layer (PHY)?
• Motivation for Rateless
Codes
• Rateless Code Design
for Short Block lengths
• Rateless Code
Performance

Fundamental Limit of Finite Block Length
– Existing systems are designed to efficiently transmit long data
packets
– Several thousands of channel uses
– Moderately low packet-error rates (around 0.01)
– Relevant for current mobile broadband services
– Shannon proved that a communication link can achieve zero
probability of error if the code is long enough, as long as the data
rate is below the channel capacity
– Most of the recent advances in the design of high-speed wireless
systems are based on the Shannon formula

– URLLC is usually characterised by short packet communications (~100s
bits in a packet) and stringent packet-error requirement (e.g., 10^-9)
– Errors cannot be avoided even the transmission rate is below the
Shannon Capacity. Shannon bound is no longer accurate!
– Polyanskiy-Verdu-Poor Bound:
– The existing analyses, designs, optimisations based on Shannon bound
should be re-visited or re-formulated.
– Develop fundamental trade-offs between capacity, latency and
reliability for single and multi-user scenarios.
– New trade-offs require different designs of many communication
algorithms and protocols
Fundamental Limit of Finite Block Length

– Low latency goal prohibits long Shannon capacity approaching
codes
– Fundamental capacity bound defined by Polyanskiy for limited
length codes
– Conventional capacity approaching codes, such as LDPC and
turbo codes, do not perform well in the short block length regime
– The conventional short codes, such as BCH and convolutional
codes are too complex for decoding
– The decoding is the dominant part of receiver processing delay
– Design of high performance and low complexity decoding is
required
– Each retransmissions introduces 8ms delay in LTE
– How to design the techniques which can avoid retransmissions
Short Code Design for reducing TTI

Codes with Diversity
– For short codes, achieving a high reliability requires very high
SNR, which is not feasible for IoT devices with only limited
transmission power.
– To reduce the SNR requirements, diversity techniques are needed.
– Time diversity is not appropriate as it increases latency.
– How to design space frequency codes to achieve the spatial and
frequency diversity?
– How to explore multi-connectivity diversity in wireless networks?

MAC
Grant-Based
Orthogonal
Access Techniques
Non-Orthogonal
Access (NOMA)
Grant-Free
Access
Cross-Layer
Design
• Why can’t 4G support
URLLC ?
• Why can’t 5G support
massive URLLC?
• Information theoretic
bounds
• Existing NOMA schemes
• Queueing Delay
• Effective Bandwidth
• Packet Dropping
How do we address URLLC at the Medium Access Control Layer
(MAC)?
• Motivation for Grant-
Free Access
• Performance analysis
• Different transmission
schemes for Grant-Free
NOMA
• Deep Learning
Approach for Grant-
Free NOMA

Resource reservation for uRLLC
Problem and challenge
In the request contention period, uRLLC needs to compete with other
services, leading to uncertain access delay;
In the joint scheduling among uRLLC and other services, uRLCC need to
be served immediately, which would interrupt ongoing transmission of
other services.
Solution
reserve resources for uRLLC to ensure its immediate transmission
– trade-off between latency and reserved resources

Grant-free access for reducing access delay
• Grant acquisition and random access procedures in current standards are two
major sources of delay
• BS needs to first identify the users through contention-based random access.
Key problems: severe collisions and high latencies when the number of users
increases
• In grant-free multiple access, users encode their IDs and data together and
transmit them directly without grant acquisition. This eliminates the contention
and random access phase, significantly reducing the latency, at the expense of
larger interference. Signals from multiple users are superimposed and
successive interference cancellation (SIC) is used at the receiver to decode the
messages

Network
Network
Availability
Multi-
connectivity
Mobile-Edge
Computing
Systems
Deep Learning
for Network
Management
• Definitions • Improving Network
Availability
• Terrestrial
Communications
• Air-to-Ground and
Ground-to-Air
Communications
• Summary of our recent
results
• Motivation of using
deep learning in URLLC
• Examples
How do we address URLLC at the Network Layer ?
• Analysis of the
computing delay using
short packets
• Optimisation of user
association, computing
offloading, and radio
Cross-Layer Design

Edge computing for reducing backhaul latency
• To reduce backhaul latency, computation and content resources should be
moved from cloud to the edge – VR, AR, Vehicular networks (driving, urban
sensing, content distribution, mobile advertising and intelligent transportation);
• Proximal users are allowed to communicate directly;
cloud
P. Mach and Z. Becvar, "Mobile Edge Computing: A Survey on Architecture and Computation Offloading," in IEEE Communications
Surveys & Tutorials, 19(3), 2017.

Channel Coding Techniques for URLLC
Dr. Mahyar Shirvanimoghaddam

→ Performance metric: Reliability and flexibility
Channel code candidates
Comparison of Channel Codes
Advanced Ordered Statistic Decoder
Short Analog Fountain Codes

– Reliability is defined as the success probability of transmitting K
information bits within the desired user plane latency at a
certain channel quality.
– Sources of failure from a higher layer perspective are when the
packet is lost, or it is received late, or it has residual errors.
– It is essential to maximize the reliability of every packet in order
to minimize the error rate, so as to minimize the number of
retransmissions.
– To provide a high level of reliability, a channel code with low
code rates is generally used.
Reliability
[S-3] M. Shirvanimoghaddam et al., "Short Block-Length Codes for Ultra-Reliable Low Latency Communications," in IEEE Communications
Magazine, vol. 57, no. 2, pp. 130-137, February 2019. https://doi.org/10.1109/MCOM.2018.1800181

– Bit-level granularity of the codeword size and code operating
rate is desired for URLLC.
– The actual coding rate used in transmission could not be
restricted and optimized for specified ranges.
– The channel codes therefore need to be flexible to enable
hybrid automatic repeat request (HARQ).
– The number of retransmissions, however, needs to be kept as low
as possible to minimize latency
Flexibility

– There are two effects that should be distinguished here to better
understand the code design problem for short blocks.
– The first one is the gap to Shannon’s limit
If we decrease the block length, the coding gain will be reduced and
the gap to Shannon’s limit will increase. This is not a problem of code
design but is mainly due to the reduction in channel observations that
comes with finite block lengths.
Performance Benchmark

We will use the normal approximation (NA) that incorporates the
reduction in channel observations, as the performance benchmark
for comparison.
C is the channel capacity (a function of SNR)
V is the channel distortion (a function of block length and SNR)
n is the block length
Pe is the error rate
Performance Benchmark: Normal Approximation
Y. Polyanskiy, H. V. Poor and S. Verdu, "Channel Coding Rate in the Finite Blocklength Regime," in IEEE Transactions on Information
Theory, vol. 56, no. 5, pp. 2307-2359, May 2010.

Performance Benchmark: Normal Approximation for BI-AWGN
T. Erseghe, "Coding in the Finite-Blocklength Regime: Bounds Based on Laplace Integrals and Their Asymptotic Approximations," in IEEE
Transactions on Information Theory, vol. 62, no. 12, pp. 6854-6883, Dec. 2016.

– The second effect is the gap to the finite length bounds, that is if
we decrease the block length, modern codes, such as LDPC or
Turbo codes, show a gap to finite length bounds.
– This is often due to the suboptimal decoding algorithms.
Performance Benchmark

✔ Performance metric: Reliability and flexibility
→ Channel code candidates
Comparison of Channel Codes

– Bose, Chaudhuri, and Hocquenghem (BCH) codes are a class of
powerful cyclic error-correcting codes that are constructed using
polynomials over finite fields.
– The main feature of BCH codes is that the number of guaranteed
correctable symbols, t, is defined during the code design process.
The minimum distance dmin of BCH codes is at least 2t + 1.
– The decoding of BCH codes is usually done using a bounded
distance decoder, like the Berlekamp-Massey algorithm, that can
correct any combination of up to t symbol errors.
– In order to increase the coding gain, in particular on noisy
channels, one may use a soft-input decoder, such as ordered
statistics decoder (OSD).
BCH
[S-3] M. Shirvanimoghaddam et al., "Short Block-Length Codes for Ultra-Reliable Low Latency Communications," in IEEE
Communications Magazine, vol. 57, no. 2, pp. 130-137, February 2019. https://doi.org/10.1109/MCOM.2018.1800181
S. Lin, D. Costello, Error Control Coding: Fundamentals and Applications, Pearson-Prentice Hall, 2004.

Convolutional Codes
• Convolutional codes (CC) were first
introduced by Elias in 1955.
• They differ from block codes as the
encoder contains memory.
S. Lin, D. Costello, Error Control Coding: Fundamentals and Applications, Pearson-Prentice Hall, 2004.
• Generally, a rate R = k/n convolutional encoder with memory
order m can be realized as a linear sequential circuit with input
memory m, k inputs, and n outputs, where inputs remain in the
encoder for m time units after entering.
• Large minimum distances and low error probabilities for
convolutional can be achieved by changing m.

Turbo Codes
– In 1993, Berrou, Glavieux, and Thitimajshima introduced Turbo coding,
which combines a parallel concatenation of two convolutional encoders
and iterative maximum a-posteriori probability (MAP) decoding.
– Turbo codes have been extensively used for the data channel in LTE.
– For large blocks, Turbo codes are capable of performing within a few
tenths of dB from Shannon’s limit.
– Turbo codes with iterative decoding in short and moderate block
lengths show a gap of more than 1 dB to the finite-length performance
benchmark.
– For Turbo codes, 1-bit granularity is feasible for all coding rates and
for a full range of block sizes, and the ability of Turbo codes to
support both Chase combining and incremental redundancy HARQ is
well known.

LDPC Codes
• Low-density parity-check (LDPC)
codes were originally proposed by
Gallager in the early 1960s and
later rediscovered in the 1990s,
when researchers began to
investigate codes-on-graph based.
• LDPC codes are now being used in WiFi and will be used for 5G eMBB.
• The main benefit of LDPC codes is the low-complexity decoding algorithm
which can be implemented using parallel processing.
• They also closely approach the Shannon limit in large block lengths using
iterative belief propagation decoder.

Polar Codes
• Polar codes as introduced by Arikan are
binary linear codes that can provably
achieve the capacity of a binary-input
discrete memoryless channel using low-
complexity encoding and decoding as
the code length tends to infinity.
E. Arikan, "Channel Polarization: A Method for Constructing Capacity-Achieving Codes for Symmetric Binary-Input Memoryless
Channels", IEEE Trans. Inf. Theory, vol. 55, no. 7, pp. 3051-73, July 2009.
• Channel polarization is a central technique in the construction of these codes, in
which the block code translates N independent and identical binary-input
discrete memoryless channels into N synthesized channels with capacities either
(close to) zero or one.’
• The message is only sent over the set of near-perfect channels, and the
unreliable channels are unused; in practice, they are assigned constant inputs a
priori known for both the encoder and decoder (frozen symbols).

✔ Channel code candidates
→ Comparison of Channel Codes

Comparison
Code Rate = ½ and Block length =128
Maximum Likelihood Decoding

Comparison
Code Rates = 1/3 (blue), 1/6 (red), and 1/12 (black)
Information Block length = 40 and 128
Practical Decoder

Comparison
– Under maximum likelihood decoding, the BCH code
outperforms all other existing codes owing to its better distance
spectrum.
– Other codes are mainly designed to provide good
performance while maintaining the decoding complexity at a
reasonable order.
– Using practical decoders, Polar codes show very
good performance with no error floor.
– LPDC code suffer from error floor at short block lengths
and Turbo codes perform well only at large block lengths.

Comparison
Rate performance of different candidate codes at a BLER of 10–4 when the codeword
length is N = 128.

✔ Comparison of Channel Codes
→ Advanced Ordered Statistic Decoder

Segmentation-Discarding Algorithm
– Pre-processing
– Reprocessing
Preprocessing
𝐫 = 𝐬 + 𝐧
AWGN channel
output
abs
Reliabilities
𝛂 = 𝐫
Permutation 𝜋1
&
Permutation 𝜋2
෨𝐆 = 𝜋1(𝜋2(𝐆))
෤𝐫 = π1(π2(𝐫))
෥𝛂 = 𝜋1(𝜋2(𝛂))
Reprocessing
MRB
positions Calculate distance 𝒟 𝑒,
and update the minimum
distance 𝒟 𝑚𝑖𝑛. Output
corresponding estimation.
෤𝐲
෤𝐲 𝐵
Re-encoding
෤𝐜 𝑒 = ෤𝐲 𝐵 ⊕ 𝐞 ෨𝐆
TEP
s
෤𝐜 𝑒
෤𝐫, 𝒟 𝑚𝑖𝑛
Ƹ𝐜 𝑜𝑝𝑡
Estimated
codeword
OSD algorithm [6]
First K positions: Most Reliable Basis (MRB)
[P-7] C. Yue, M. Shirvanimoghaddam, Y. Li and B. Vucetic, . "Segmentation-Discarding Ordered-Statistic Decoding for Linear Block Codes." IEEE Global Communication
Conference, Kona, HI, 2019 https://arxiv.org/abs/1901.02603

Segmentation-Discarding Algorithm (SDA)
67
– All the weight-𝑙 TEPs are cut into several segments in 𝑙-th order decoding.
– Some least reliable segments are discarded according to a discarding rule.
– Complexity is significantly reduced.
𝐫 = 𝐬 + 𝐧
AWGN channel
output
Preprocessing
෨𝐆, ෤𝐫
0-reprocessing
(Hard-decision)
𝑙-reprocessing
𝑙 = 1: 𝑚
(stopping rule)
෤𝐲, ෤𝐫
Ƹ𝐜 𝑜𝑝𝑡
Estimated
codeword
Producing
𝑙-TEPs
TEPs
Segmentation Discarding
𝒟 𝑚𝑖𝑛
𝒟 𝑚𝑖𝑛...෥𝛂
...
𝑆𝑙1
𝑆𝑙 𝑄
𝑆𝑙1
𝑆𝑙 𝑖
𝑄: maximum segments number 𝑆𝑙 𝑖
: the 𝑖-th TEP segment in 𝑙-reprocessing
[P-7] C. Yue, M. Shirvanimoghaddam, Y. Li and B. Vucetic, . "Segmentation-Discarding Ordered-Statistic Decoding for Linear Block Codes." IEEE Global Communication
Conference, Kona, HI, 2019 https://arxiv.org/abs/1901.02603
for an order d OSD algorithm one need to generate O(kd) test error patterns.

Segmentation and Discarding Rules
Discarding Rule
68
– Decoding scheme combining segmentation and discarding
– Light-colored blocks represent the segments that are discarded
– Dark-colored segments are retained
[P-7] C. Yue, M. Shirvanimoghaddam, Y. Li and B. Vucetic, . "Segmentation-Discarding Ordered-Statistic Decoding for Linear Block
Codes." IEEE Global Communication Conference, Kona, HI, 2019 https://arxiv.org/abs/1901.02603

Performance and Complexity Comparison
69
Performance comparison in decoding (128,64,22) eBCH code (code rate 0.5)
M. P. C. Fossorier and S. Lin, “Soft-decision decoding of linear block codes based on ordered statistics,” IEEE Transactions on
Information Theory, vol. 41, no. 5, pp. 1379–1396, Sep 1995.
J. Van Wonterghem, A. Alloum, J. J Boutros, and M. Moeneclaey, “On performance and complexity of OSD for short error correcting
codes in 5G-NR,” 06 2017.

Performance and Complexity Comparison
Performance comparison in decoding (128,22,48) eBCH code (code rate 0.17)
M. P. C. Fossorier and S. Lin, “Soft-decision decoding of linear block codes based on ordered statistics,” IEEE Transactions on
Information Theory, vol. 41, no. 5, pp. 1379–1396, Sep 1995.
J. Van Wonterghem, A. Alloum, J. J Boutros, and M. Moeneclaey, “On performance and complexity of OSD for short error correcting
codes in 5G-NR,” 06 2017.

Reduced Complexity OSD algorithm
– The complexity of OSD can be further reduced by utilizing
efficient techniques.
– In our latest work recently submitted to IEEE Transaction
on Information Theory, we fully analyzed the OSD decoder
and devised several low complexity alternatives based on sufficient
and necessary conditions.
– The paper can be found on Arxiv
C. Yue, M. Shirvanimoghaddam, Y. Li and B. Vucetic, A Revisit
to Ordered Statistic Decoding: Distance Distribution and
Decoding Rules https://arxiv.org/abs/2004.04913

Comparison: Complexity vs. Performance

✔ Comparison of Channel Codes
✔ Advanced Ordered Statistic Decoder
→ Short Analog Fountain Codes

– In conventional wireless communication, the transmitter is fed back
some indicator of its channel state such that it can choose the best
modulation and coding scheme.
– For asymptotically long blocks, the overhead associated with this
control information is negligible in comparison to the block length.
This signaling is estimated to incur 5–8 ms latency which violates
the low latency requirement of URLLC
– The signaling overhead is estimated to be around 30–50% for
payloads of length 200 symbols with 7–10 users, which is very
costly and inefficient (even for mMTC).
Motivation

– We propose a class of self-adaptive channel codes, that can
transmit the required block length without any CSIT.
– The coded symbols are sequentially transmitted until the receiver
can successfully decode the information. A stop-feedback is sent to
the transmitter to terminate the transmission.
– Existing self-adaptive codes:
– Luby Transform codes
– Raptor codes
– Protograph-based raptor like codes
– Rate compatible modulation (RCM)
– Strider codes
– Analog fountain codes (AFC)
Motivation
Tailored for asymptotically
long block lengths

Encoding
Precoder BPSK
S-AFC
Encoder
AWGN
S-AFC
Decoder
Binary
Decoder
High-rate LDPC (~0.95), for long block lengths
High-rate BCH (~0.95) – short block lengths
Iterative belief propagation
Belief propagation for LDPC
Ordered Statistics Decoder for BCH
Variable block length

Performance of AFC in the Asymptotic Block-Length Regime
– AFC can achieve near capacity
performance for asymptotically
long blocks
– The design of AFC is based on a
degree d and a weight set W(set of
real numbers)
Near-Capacity Performance when k = 10,000 bits
𝑐 𝑛
𝑏1 𝑏 2 … 𝑏 𝑑
Set of chosen information symbols in each
encoding stage is random!
[P-6] M. Shirvanimoghaddam, Y. Li and B. Vucetic, "Near-Capacity Adaptive Analog Fountain Codes for Wireless Channels," in IEEE
Communications Letters, vol. 17, no. 12, pp. 2241-2244, December 2013. https://doi.org/10.1109/LCOMM.2013.101813.131972

Performance of AFC in the Short Block-Length Regime
Poor performance at low SNR in comparison to the Polyanskiy-Poor and Verdu
Bound (Normal Approximation), k = 192
[P-1] R. Abbas, M. Shirvanimoghaddam, T. Huang, Y. Li and B. Vucetic, "Novel Design for Short Analog Fountain Codes," in IEEE
Communications Letters, vol. 23, no. 8, pp. 1306-1309, Aug. 2019. https://doi.org/10.1109/LCOMM.2019.2910517

Design of Short AFC
We propose a new weight set design based on two
rules:
- Power constraint
- Distance constraint
We want to maximize the minimum Euclidean
distance of the constellation while ensuring
the power constraint is met!

Performance of Short AFC
Much better performance in comparison to original AFC

Great tail distribution!

No error floors exhibited up to 10-7
[P-1] R. Abbas, M. Shirvanimoghaddam, T. Huang, Y. Li and B. Vucetic, "Novel Design for Short Analog Fountain Codes," in IEEE Communications Letters,
vol. 23, no. 8, pp. 1306-1309, Aug. 2019. https://doi.org/10.1109/LCOMM.2019.2910517

Orders of magnitude improvement for very short blocks

Analog Fountain Codes
Potentials:
– Low complexity encoder and decoder
– 1-bit granularity can be achieved for code rate and block length
– A very suitable candidate for URLLC because of its excellent
performance across SNRs at short block lengths
– Eliminate the need for CSI feedback
Challenges
The precoder design
Narrowing down the CDF for guaranteed reliability

Conclusion and future work
85
– Channel code for the URLLC has not been standardized yet.
– Polar codes suffer from latency due to the SIC.
– LDPC codes show error floor in low-to-moderate block lengths and
gap to the NA in short blocks.
– BCH codes suffer from high complexity OSD decoder
Moving Forward:
1. Reducing complexity of OSD to enable real time BCH decoding to
achieve the best error rate performance
2. Using analog fountain codes for autmatic adaptation to the
channel, therefore reducing the latency.

Medium Access Control Layer for URLLC
Dr. Rana Abbas

Medium Access Control Layer (MAC)
Link Request/Establishment
& Retransmissions

→ Why 4G MAC techniques cannot support massive URLLC
5G Approaches to Reducing Latency in Channel Access
Motivation for Grant-Free NOMA & Key Challenges
Performance Analysis of Massive Grant-Free NOMA
Grant-Free NOMA with Rateless Codes
Multi-Layer Grant-Free NOMA (Power and Code Domain)
Deep Learning Approach for Grant-Free NOMA
Conclusions and Future Directions

Why 4G cannot support Massive URLLC ?
Existing systems are designed to efficiently
transmit long data packets and few human
users:
– Several thousands of channel uses
– Moderately low packet-error rates (around
0.01)
– Control overhead is negligible to the size
of the payload
Current uplink transmissions based on multi-stage
and heavy signaling have a tremendous impact on
latency and reliability:
– For asymptotically long blocks, the
overhead associated with this control
information is negligible in comparison to
the block length. This signaling is
estimated to incur 5–8 ms latency
which violates the low latency
requirement of URLLC
– The signaling overhead is estimated to
be around 30–50% for payloads of
length 200 symbols with 7–10 users,
which is very costly and inefficient (even
for mMTC).
Current LTE end-to-end latency is
not guaranteed and ranges from
100 ms to a few seconds.

✔ Why 4G MAC techniques cannot support massive URLLC
→ 5G Approaches to Reducing Latency in Channel Access
Motivation for Grant-Free NOMA & Key Challenges

1. Cat-M1 uses 6 Resource Blocks (RBs) with 12 tones per RB at 15 kHZ SCS;
2. Cat-NB1 uses 1 Resource Block (RB) with 12 tones with 12 tones per RB at 15 kHz SCS, single-tone option also available
5G New Radio Rel-15,16
Pre-configured/semi-persistent
scheduling.
– This is only suitable
for predictable traffic patterns,
but otherwise exhibits low
efficiency.
– It is also only suitable for the
initial transmission (not for
HARQ)
For retransmissions (HARQ),
– K-repetition
– Proactive repetition
Spectrum Flexibility
(Subcarrier Spacing 15 kHz)
Low Latency
(TTI 0.125 ms)
Instant Uplink Access

5G New Radio Rel-15,16 : 2-Step Random Access Procedure
https://www.5gamericas.org/wp-content/uploads/2020/01/5G-Evolution-3GPP-R16-R17-FINAL.pdf

✔ 5G Approaches to Reducing Latency in Channel Access
→ Motivation for Grant-Free NOMA & Key Challenges

Motivation for NOMA: Multiple Access Channel Capacity
The traditional capacity region is valid when the block length T is close to
10,000 symbols and more!

Motivation for NOMA
95
The sub-optimality of orthogonal
multiple access schemes increases
with the decrease in block length
and with the increase in the
number of users!
E. MolavianJazi and J. N. Laneman, "A Second-Order Achievable Rate Region for Gaussian Multi-Access Channels via a
Central Limit Theorem for Functions," in IEEE Transactions on Information Theory, vol. 61, no. 12, pp. 6719-6733, Dec. 2015,
doi: 10.1109/TIT.2015.2492547.

Study on NOMA for 5G-NR (Rel-16)*
MA
Signatures
Bit-Level Processing Scrambling LCRS
NCMA
Interleaving IDMA
IGMA
Symbol-Level Processing Welch-Bound RSMA
WCMA
Complex-Valued MUSA
Sparse Spread SCMA
PDMA
*3GPP TR 38.812 V16.0.0 (2018-12)
[S-4] M.B. Shahab, R. Abbas, M. Shirvanimoghaddam, and S. J. Johnson. "Grant-free Non-orthogonal Multiple Access for IoT: A Survey." arXiv
preprint arXiv:1910.06529 (2019). https://arxiv.org/abs/1910.06529, accepted to appear in IEEE Communications Surveys and Tutorials

Key Challenges for Grant-Free NOMA
User detection/identification
Collision avoidance/detection/resolution
Low-complexity decoders for joint user decoding
Load estimation for parameter tuning e.g. code rate
Synchronisation

✔ Motivation for Grant-Free NOMA & Key Challenges
→ Performance Analysis of Massive Grant-Free NOMA

Grant-Free NOMA: Framework

Grant-Free NOMA: Transmission Scheme
We have a set of 𝐿 orthogonal pilot sequences of length 𝑞 symbols, e.g.,
Zadoffchu, Golden codes, m-sequences, etc.
We have a set of 𝐿 codebooks.
Each device chooses a pilot uniformly at random and submits
simultaneously with the remaining devices.
The device that chose pilot sequence 𝑖 will use codebook 𝐶𝑖.
A collision event is defined as the event of two or more devices choosing
the same pilot.

Grant-Free NOMA: Detection & Decoding Scheme
The AP performs cross-correlations to detect the transmitted pilot
sequences and estimate the channel parameters for each codebook used.
Let us assume for now that the AP knows which “layers” are in collision:
In the literature, interference is
treated as binary!

Aggregate Interference Power: PPP Approximation
The number of collided devices is
a truncated Poisson random
variable.
However, their aggregate power,
for a given Ls, can be well-
approximated by a PPP.
For a PPP, the aggregate power
follows a skewed truncated stable
distribution.
102
Aggregate interference power for Lc = 200

Grant-Free NOMA: Case Studies
Performance
Metric
Coding Scheme Decoding Scheme
Outage
Probability
The probability that a
device, which has not
collided, is not
successfully decoded in a
given time slot.
Fixed Rate
Rate is fixed a priori and
does not change (Fixed
codeword lengths)
Successive Joint
Decoding
Successively jointly
decoding a subset of the
users (strongest) while
regarding the rest as
interference. This persists
until decoding is
successful.
Throughput
The ratio of the number of
successfully decoded
information bits to the
length of the transmitted
codeword (M −q),
averaged over all slots
Rateless
Rate is not fixed a priori
and can adapt to network
conditions e.g. load,
channel quality, etc.
Successive
Interference
Cancellation
Successively decoding the
strongest user while
regarding the rest as
interference

Case of Successive Joint Decoding

Comparison between Successive Joint Decoding &
Successive Interference Cancellation

- Open loop power control performs poorly with GF-NOMA
- The collision probability is the dominating factor of performance
- Open loop power control requires prohibitively long preamble sequences
(large number of contention transmit units)
- Joint decoding is far more superior than SIC in throughput
- We cannot hope to recover more than 3 users at a time with SIC (in practical
scenarios) due to error propagation
Key Take-aways

✔ Performance Analysis of Massive Grant-Free NOMA
→ Grant-Free NOMA with Rateless Codes

Overview of Rateless Codes
In conventional wireless communication, the transmitter is fed back some
indicator of its channel state such that it can choose the best modulation
and coding scheme.
For asymptotically long blocks, the overhead associated with this control
information is negligible in comparison to the block length. This signaling is
estimated to incur 5–8 ms latency which violates the low latency
requirement of URLLC
The signaling overhead is estimated to be around 30–50% for payloads of
length 200 symbols with 7–10 users, which is very costly and inefficient
(even for mMTC).
Motivation

M. Shirvanimoghaddam, M. Dohler and S. J. Johnson, "Massive Non-Orthogonal Multiple Access for Cellular IoT: Potentials and Limitations,"
in IEEE Communications Magazine, vol. 55, no. 9, pp. 55-61, Sept. 2017.

The maximum arrival rate versus the initial backlog obtained from the weak and
strong stability conditions for different Ms, when Ns = 20, W = 1 MHz, L = 1000
M. Shirvanimoghaddam, M. Condoluci, M. Dohler and S. J. Johnson, "On the Fundamental Limits of Random Non-Orthogonal Multiple Access in
Cellular Massive IoT," in IEEE Journal on Selected Areas in Communications, vol. 35, no. 10, pp. 2238-2252, Oct. 2017.

The maximum packet arrival rate versus the delay constraint under weak and
strong stability conditions, when W = 1 MHz, L = 1000, and collision
probability is set to pc = 0.01.
M. Shirvanimoghaddam, M. Condoluci, M. Dohler and S. J. Johnson, "On the Fundamental Limits of Random Non-Orthogonal Multiple Access in
Cellular Massive IoT," in IEEE Journal on Selected Areas in Communications, vol. 35, no. 10, pp. 2238-2252, Oct. 2017.

✔ Grant-Free NOMA with Rateless Codes
→ Multi-Layer Grant-Free NOMA (Power and Code Domain)

Grant-Free NOMA with Multi-Layer Design
Transmission Scheme
Layer 2 (P2)
Layer 1 (P1)

Example of Superposition of Signals

Mixed-Linear Integer Programming (MILP)

✔ Multi-Layer Grant-Free NOMA (Power and Code Domain)
→ Deep Learning Approach for Grant-Free NOMA

Deep Learning Approach for GF-NOMA
[S-6] R. Abbas, T. Huang, M.B. Shahab, M. Shirvanimoghaddam, Y. Li and B. Vucetic. "Grant-Free Non-Orthogonal Multiple Access: A Key Enabler
for 6G-IoT.” arXiv preprint arXiv:2003.10257 (2020). https://arxiv.org/abs/2003.10257

Fully Connected Neural Network Architecture
• We adopt the structure of the auto-
encoder: 4 hidden layers, where each
layer consists of 2048 neurons.
• The activation function used for each
hidden layer in the encoder is ReLU
and the activation function used for its
output layer is a Sigmoid function.
• For the decoder, we also used 4
hidden layers and each layer consists
of 2048 neurons. The activation
function for each hidden layer is ReLU.
The output layer of the decoder
contains 2k-1 nodes, and the loss
function used is a binary cross-entropy
function.

Performance Evaluation : Code Rate = 1/2

Performance Evaluation : Code Rate = 1

Performance Evaluation : Different Loss Functions

✔ Multi-Layer Grant-Free NOMA (Power and Code Domain)
✔ Deep Learning Approach for Grant-Free NOMA
→ Conclusions and Future Directions

– Existing grant-based access techniques are ill-suited for Massive
URLLC due to overhead and access delays
– Sporadic traffic in Massive URLLC is especially challenging to
solve with Grant-Free NOMA being a promising solution
– The bottleneck of performance are collisions, with collision
detection/resolution schemes are lacking in the literature.
– Joint decoding significantly outperforms SIC
– Rateless codes present a promising solution to improve reliability
– Multi-layer Grant Free NOMA presents a promising solution for
power improvements and reducing collisions
MAC: Summary

– Develop Multi-Layer Grant free NOMA schemes with rateless
codes
– Develop low complexity decoders for joint user decoding
– Develop deep learning techniques that aim at joint detection
and decoding of users in Grant-Free NOMA
MAC: Future Research Directions

Cross Layer Solutions for URLLC

X. Wu, X. Zhu, G. Q. Wu, and W. Ding, “Data mining with big data,” IEEE Trans. Knowl. Data Eng., vol. 26, no. 1, pp. 97–107, Jan. 2014.
Cross-Layer Design: Major Problems
– Some fragmental ideas for related
problems:
– Cross-layer design is missing
– Our understanding:
– E2E delay
– Overall reliability
– Network availability
We need the whole picture!

– Transmission delay (How to design frame structure? Is
retransmission helpful? )
– Queueing delay (Will a packet wait in a queue? How to design
queueing policy?)
–random arrival rate that could be higher than service rate
– Computing delay (How to characterize such delay?)
– Backhaul delay (Which kind of backhaul shall we use?)
–one-hop fiber backhaul or mmWave backhaul with LOS path
– Propagation delay (When is propagation delay dominated?)
–light travels 300 km in 1 ms
E2E delay

What are the factors leading to packet loss?
– Transmission error: non-zero in the short blocklength regime!
– Queueing delay violation: a packet becomes useless if it is
delayed
– Packet dropping in deep fading channel: wireless channel is
never reliable
How to reduce these packet loss probabilities?
What is the dominating factor?

How to characterize network availability?
– Can we ignore shadowing as what we did for traditional real-
time service?
– How long is the communication distance of URLLC?
• propagation delay in core network/backhaul
• distance attenuation in radio access network
– How to handle interference?
• This is also an issue for traditional real-time service!
How to improve network availability?

Y. Polyanskiy, H. V. Poor, and S. Verdu, “Channel coding rate in the finite blocklength regime,” IEEE Trans. Inf. Theory, pp. 2307–2359, May 2010.
Cross-Layer Design: Useful Tools
– Achievable rate in finite blocklength regime
– Transmission delay and transmission error probability
2
0
log 1t gP
C D W
WN
 
= + 
 
( )1
0
1
ln 1
ln2
t
t
Qt
D W gP
s f
WN D W
−
  
= + −  
  
( )t t
D ，

W. Yang, G. Durisi, T. Koch, and Y. Polyanskiy, “Quasi-static multiple-antenna fading channels at finite blocklength,” IEEE Trans. Inf. Theory, pp.
4232–4264, Jul. 2014.
– Achievable rate in MIMO system
are the eigenvalues of channel matrix
– Approximation in high SNR regime
( )
( )
t r
1
1
min ,
0
( ) ln 1
ln2
Nt
tk k
N
t
k
Qk
D W g p V
s n f
N W D W
−
=
   
 + −  
   

( )
( )
( )
rtmin ,
r rt t2
1
0
min ,
1
,
1
min
N
k
k k
N
V N N
p g
N W
N N
=
= − 
 
+ 
 

kg

S. Schiessl, J. Gross, and H. Al-Zubaidy, “Delay analysis for wireless fading channels with finite blocklength channel coding,” in Proc. ACM MSWiM,
2015.
– Ensuring queueing delay with Stochastic network
calculus
–From bit domain to SNR domain (given bandwidth)
–A useful tool to analyzing delay with given resource allocation
– Not applicable when one needs to optimize bandwidth
– Not applicable with unbounded arrival process, e.g., Poisson
– No closed-form result, inconvenient in cross-layer optimization
( ),q q
D 

Cross-Layer Design: System Model
Local Communication Scenarios:
Negligible propagation delay and short backhaul latency

– Each user requests packets from multiple users
– Arrival process:
– Departure process:
Queueing model
( )ki ia n A
( ) ( ) ( ) min ,k k kb n Q n s n=
( ) ( ) ( )1 ( )
k
k k i k
i
Q n Q n a n b n

+ − = −A

– Frame structure
– Channel model
– Multiple transmit antennas; single receive antenna
– Flat fading channel
– Quasi-static: delay bound < channel coherence time
137

Bursty arrivals: Interrupt Poisson process
Auto-correlated arrivals: Switched Poisson process
Cross-Layer Design: Arrival Procceses
138

– Achievable rate in short blocklength regime
– Queueing delay violation (Poisson process)
Cross-Layer Design: Finite Blocklength Regime
139
ln(1/ )
(packets/frame)
ln(1/ )
ln 1
q
f kB
k q
f kq
q
k
T
E
T
D
D



=
 
+ 
 
( )
( ) ( )1
0
( ) ln 1 packets/frame
ln2
t
tk k kt k
k q k
k t k
g p nDW V
s n f
u N W DW

−
   
 + −  
   
t fD T=

C. Chang and J. A. Thomas, “Effective bandwidth in high-speed digital networks,” IEEE J. Sel. Areas Commun., pp. 1091–1100, Aug. 1995.
[M-7] She, Changyang, et al. "Cross-layer design for mission-critical iot in mobile edge computing systems." IEEE Internet of Things Journal (2019).
https://doi.org/10.1109/JIOT.2019.2930983
Cross-Layer Design: Effective Bandwidth
– Ensuring queueing delay with effective bandwidth
(e.g. Poisson)
( ),q q
D 
ln(1/ )
(packets/s)
ln(1/ )
ln 1
q
B
q
fq
q
E
T
D
D



=
 
+ 
 

[1] G. L. Choudhury, D. M. Lucantoni, and W. Whitt, “Squeezing the most out of ATM,” IEEE Trans. Commun., vol. 44, no. 2, pp. 203–217, Feb. 1996.
[[M-7] She, Changyang, et al. "Cross-layer design for mission-critical iot in mobile edge computing systems." IEEE Internet of Things Journal (2019).
https://doi.org/10.1109/JIOT.2019.2930983
Cross-Layer Design: Effective Bandwidth
– In general, effective bandwidth is only applicable with large
queue length or delay
– Applicable for arrival processes that are more bursty than Poisson
– We validated that effective bandwidth can be used in short
delay region for bursty arrival processes

– Constraint on and
– Power control scheme: channel inverse
– Finite transmit power
– Proactive packet dropping
– Overall reliability requirement
Cross-Layer Design: Overall Reliability
( ) B
k ks n E
max B
k ks Eth
k kg g
( )max
max 0, B
k kE s−
( ),q q
D  ( )t t
D ，
( )( )( ) max1 1 1 1c q h c q h
k k k k k k      − − − −  + + 

Cross-Layer Design: Optimisation
143
Optimising the packet loss probabilities
tot th
, , ,
1
1,2,...,
minq c h
k k kk
K
k
W
k
k K
P P
  
=
=

( )
( )
0
th
th
0
g
0
ln 1
s.t. 1
ln 1
k k
k k
k k
N W
kP
k
h
k
P g
N W
f g dg





  
+  
   −
 +
 
 

( ) ( ) ( )1
Q
ln2
,ln 1 q t
q k k
t t
B
k k
k k
D
D
u V
E f
W WD
  −
+ = +
max
c q h
k k k   + + 
max
0, , 1,...,
K
k k
k
W W W W k K  =

Cross-Layer Design: Solutions
– Flat-fading → frequency-selective channel
– Channel coefficient on subchannel j of user k
– Apply the achievable rate over MIMO channel in:
144
Extension II: Different Channel Models
1 1
kjh 

sc
1
2
0 ... 0
0 ... 0
...
0 0 ...
kkN
k
k
h
h
h
 
 
 =
 
 
  
H
( )
sc
1t
1 0 t
( ) ln 1
ln2
k
kj kj t
k k
j
N
Q
g pDW V
s n f
N W DW
−
=
   
 + −  
   

W. Yang, G. Durisi, T. Koch, and Y. Polyanskiy, “Quasi-static multiple-antenna fading channels at finite blocklength,” IEEE Trans. Inf.
Theory, pp. 4232–4264, Jul. 2014.

Cross-Layer Design: Numerical Results
145
M//D/1 queue: Poisson arrival process & Constant service rate
D. Gross and C. Harris, Fundamentals of Queueing Theory. Wiley, 1985.

– Poisson arrival process
– IPP (bursty)
– SPP (auto-correlated)
– Effective bandwidth is applicable
for Poisson process and processes
that are more bursty!
146

Three packet loss probabilities are in the same level, none of them
can be ignored
147

Setting the three packet loss probabilities equal will cause minor power
loss

Cross-Layer Design: Summary
– A framework is proposed for cross-layer optimization for URLLC
– Effective bandwidth can be used to design resource allocation
policy for Poisson processes and the processes that are more bursty
than Poisson
– Proposed a proactive packet dropping mechanism to ensure the
strict QoS with finite maximal transmit power
– Transmission error, queueing delay violation and packet dropping
are in the same order of magnitude, and setting the three packet
loss probabilities equal will cause minor power loss
149

Network-Layer Solutions for URLLC
Dr. Changyang She

→ Performance metric: network availability
Multi-connectivity in terrestrial communications
Antenna deployment for ground-to-air communications
Mobile edge computing systems for mission-critical IoT
Deep learning for network management

– E2E delay: 1 to 2 ms [1]
– Overall reliability (packet loss probability: 10-5 to 10-7)
– Network Availability (99.9% to 99.999% ) [2,3]
– One user: percentage of service time
– Multiple users: percentage of the total number of users
– Unavailable: delay and reliability requirements cannot be
satisfied
Note: the requirements on QoS and network availability depend on specific applications
[1] 3GPP, “Study on scenarios and requirements for next generation access technologies.” TSG RAN
TR38.913 R14, Jun. 2017.
[2] D. Ohmann, A. Awada, I. Viering, et al., “Modeling and Analysis of Intra-Frequency Multi-Connectivity for
High Availability in 5G”, VTC-Spring 2018
[3] P. Popovski, et al., Deliverable D6.3 Intermediate system evaluation results, 2014
QoS requirements in URLLC

— Spectral efficiency
— For URLLC and other services in 5G
— Energy efficiency
— Battery lifetime of mobile devices
— Cost for operating BSs
Resource Utilization Efficiency

— How to guarantee QoS requirements of URLLC?
— How to maximize resource utilization efficiency subject to
diverse QoS requirements of eMBB, mMTC, and URLLC?
— How to design practical solutions for real-time
implementation?
Research Problems

✔ Performance metric: network availability
→ Multi-connectivity in terrestrial communications

Improving Network Availability
Network availability & QoS requirement [1]
—Definition: Ratio of users with QoS guarantee in a wireless network
—For one mobile user: fraction of service time with QoS guarantee
How to characterize the network availability?
How to achieve high available range with stringent QoS?
—Distance attenuation and shadowing lead to short commun. Range
—Given locations of users, increase availability (existing studies)
—Given availability requirement (99.999%), increase available range
Motivation
[1] P. Popovski, et al., “Deliverable d6.3 intermediate system evaluation results.” ICT-317669-
METIS/D6.3, 2014.
[N-4] C. She, Z. Chen, C. Yang, T. Q. S. Quek, Y. Li and B. Vucetic, “Improving Available Range of
Ultra-reliable and Low-latency Communications with Different Transmission Modes ”, IEEE Trans. on
Commun., Nov. 2018. https://doi.org/10.1109/TCOMM.2018.2791598

– Different users broadcast
packets on different
subchannels
– Available range: maximal
communication distance with
QoS guaranteeing
Illustration of system model

Two-phase transmission protocol:
– Cellular mode (UL+DL)
– Decode at BS
(with processing delay)
– Amplify-and-forward at BS
(without processing delay)
– D2D mode
Retransmission with frequency-
hopping

— Multi-Connectivity (MC) Modes
— AF MC mode (Amplify, without processing delay at BS)
— DF MC mode (decoded, with processing delay at BS)

– Shadowing
– Shadowing: lognormal
– Correlation model [1]
– Large-scale channel gains
( )sb br sr 0( , ) exp /r r   = −
[1] S. S. Szyszkowicz, H. Yanikomeroglu, and J. S. Thompson, “On the feasibility of
wireless shadowing correlation models,” IEEE Trans. Commun., Nov. 2010.

– Decoding error probability (UL) [1]
– Packet loss probabilities
– Cellular mode (decode)
– D2D mode
– MC mode (decode)
– Constraint on network availability
u u m
u u1 t
Q
0 1
ln2
ln 1 |g
TW g P b
f
V N W TW

 
     
 + −          
E
C u d u d
lP    + −=
D m m
1 2lP  =
( )  H m u m ud m C D
1 1 21 min ,l l lP P P     = − + 
 max 1 2 p b max APr ,lP T T D D D P + + +  
[1] W. Yang, G. Durisi, T. Koch, and Y. Polyanskiy, “Quasi-static multiple-antenna
fading channels at finite blocklength,” IEEE Trans. Inf. Theory, Jul. 2014.

– Problem formulation (DF MC mode)
– Optimize the transmission durations in the two phases, and
adjust the deployment of BSs
 
1 2
d
A
,
10 c 10 cell c 0
c
10 sr 10 A
max 1 2 p b ma
sr 0
1 2 t 1 1 f 2 2 f
x APr ,
max
s.t. 10log 10 log ( )
10log 10 log ( )
, .,
l
T T
P T T D D D P
r
R
r
T T D T k T T k T
   
   
 +
= − + +
= − + +
+  = =
+ +  

– We established a framework for improving available ranges
– Decoding-based MC mode can achieve large available ranges
than other modes when processing delay is negligible
– Amplifying-based MC mode can approach the performance of the
decoding-based MC mode when the UL SNR is high
– There is a tradeoff between available range of D2D links and
cellular links with MC modes
– It is better to use decoding-based mode in macro cells and use
amplifying-based mode in micro cells
Conclusions

✔ Multi-connectivity in terrestrial communications
→ Antenna deployment for ground-to-air communications

Ground-to-Air Communications
– Motivation
– Integrating unmanned aerial vehicles (UAVs) into cellular networks
has recently been recommended by the 3GPP [1]
– The latency and reliability of control & non-payload
communication (CNPC) links are crucial for UAVs
– How to support URLLC in CNPC links of UAV communications?[N-7]
– How to characterize the performance of CNPC links?
– How to improve the available range of CNPC links?
[1] 3GPP TR 36.777, “Study on enhanced LTE support for aerial vehicles.” TSG RAN,
v15.0.0, Dec. 2017.
[N-7] C. She, C. Liu, T. Q. S. Quek, C. Yang, and Y. Li, "Ultra-reliable and Low-latency
Communications in Unmanned Aerial Vehicle Communication Systems", IEEE Trans. on
Commun., May 2019. https://doi.org/10.1109/TCOMM.2019.2896184

– Illustration of system model
– High array gain in centralized multi-antenna systems
– High macro-diversity gain in distributed multi-antenna systems
– Both edge cloud and central cloud

– Delay components
– E2E delay (UAV to central cloud)
– UL transmission delay
– Processing delay
– Backhaul delay
– Round-trip delay (UAV to edge cloud, edge cloud to UAV)
– UL transmission delay
– Processing delay
– DL transmission delay

– Probability with Line-of-Sight (LoS) path
– A single link
– Multiple correlated links

Problem formulation

– Conclusions
– We proposed a framework of maximizing the available range of
the ground control station for ultra-reliable and low-latency UAV
communications
– We characterized the correlation of whether co-located APs have
LoS path towards to a UAV using a two-state Markov Chain, and
derived the probability that at least one of the APs has a LoS
path towards a UAV
– To solve the non-convex problem of maximizing the available
range, we proposed an algorithm that converges to the optimal
solution in two asymptotic scenarios
– We then generalized the algorithm to the general scenario of the
arbitrary number of antennas at each AP.

Performance metric: network availability
Multi-connectivity in terrestrial communications
→ Mobile edge computing systems for mission-critical IoT

Mission-critical IoT in MEC
– Background
– URLLC and eMBB services co-exist in future wireless networks
– Mission-critical IoT in smart factory, virtual/augmented reality, and
autonomous vehicles
– URLLC: short packets, short processing time
– eMBB: long packets (3D video), long processing time
– Motivation
– Design service order of MEC with both URLLC and eMBB
– Analyzing the distribution of delay experienced by URLLC
– Optimizing user association, offloading probability, and radio
resource allocation for URLLC

– MC-IoT and eMBB (background
service) in a MEC system
– Multiple APs equipped with edge
servers
– One partially centralized control
plane of user association
Illustration of system model

– Queueing system
– FCFS order at the local server of each device
– Processor sharing server at the MEC
– Service ability is equally allocated to all the tasks in the server
– Short packets do not need to wait for the processing of long packets

– Delay analysis in M/G/1/PS
– Service time of short packets is much shorter than long packets
– The number of tasks in the buffer does not change significantly
during the short service time of a short packet
– The service rate of a short packet can be approximated by
– The service delay can be approximated by
– According to the distribution of q,

– Cross-layer optimization framework
– User association: the AP a device will offload its’ packets to
– Offloading rate: the probability that a packet is offloaded to
MEC
– Bandwidth allocation: UL and DL bandwidth allocation for data
transmission between a user and an MEC
– Objective: minimizing the maximal packet loss probability

Problem formulation

– Two asymptotic scenarios: communication or computing is
the bottleneck of reliability
– Communication is the bottleneck: highest large-scale
channel gain
– Computing is the bottleneck: MEC with the lowest
workload
– Extend the algorithm into the general scenario
Finding the solution of this non-convex problem:

Distribution of delay experienced by short packets

Overall packet loss probability

– We analyzed the processing delay of short packets in the
M/G/1/PS server and derived the closed-form expression
of the CCDF of the processing delay
– We minimized the overall packet loss probability under the
constraint on E2E delay by optimizing association scheme,
packet offloading rates, and bandwidth allocation.
– We derived the optimal solutions of the problem in two
asymptotic cases: communication or computing is the
bottleneck of reliability
– We then generalized the algorithm to the general scenario.
Conclusions

✔ Multi-connectivity in terrestrial communications
✔ Antenna deployment for ground-to-air communications
✔ Mobile edge computing systems for mission-critical IoT
→ Deep learning for network management

Deep learning for URLLC
Learning at three levels (User-, Cell-, Network-levels)

Our Results on AI for URLLC
User-level
• Burstiness aware bandwidth reservation (AI for traffic state
classification in tactile internet) [N-5]
• Prediction & communication co-design (AI for mobility prediction in
remote control applications) [N-8]
Cell-level
• Deep learning for resource management [N-9]
• Deep reinforcement learning for downlink scheduler design
Network-level
• Deep learning for user association [N-11]
• Deep reinforcement learning in software-defined networks

User-level Results
– Classify the packet arrival process into high and low traffic states
– Design bandwidth reservation according to traffic states
– Can save up 43.2% bandwidth subject to QoS constraints
User-level: Generating packet arrivals from a tactile device [N-5]
[N-5] Z. Hou, C. She, Y. Li, T. Q. S. Quek, and B. Vucetic, “Burstiness Aware Bandwidth Reservation for Ultra-reliable and Low-latency
Communications in Tactile Internet,” IEEE J. Sel. Areas Commun., Nov. 2018. https://doi.org/10.1109/JSAC.2018.2874113

User-level Results
– Predict the trajectory of the
device and transmit
predicted trajectory in
advance
– Can save user experienced
delay and achieve a better
reliability-delay tradeoff
User-level: Trajectory of a tactile device [N-8]
[N-8] Z. Hou, C. She, Y. Li, Z. Li, and B. Vucetic, “Prediction and Communication Co-design for Ultra-Reliable and Low-Latency Communications”, IEEE
TWC, 2019. https://doi.org./10.1109/TWC.2019.2951660

Cell-level Results
– Find labelled training samples
from optimization algorithms
– Train deep neural network
(DNN) offline for resource
allocation
– Input channel state
information and average
packet arrival rates of users
– Output number of subcarriers
and transmit power allocated
to users
– The pre-trained DNN can find
near-optimal solution in real time
Cell-level: Base station optimizes resource allocation [N-9]
[N-9] R. Dong, C. She, W. Hardjawana, Y. Li, and B. Vucetic, “Deep Learning for Radio Resource Allocation with Diverse Quality-of-Service
Requirements in 5G” IEEE TWC, submitted. (Conference version accepted by IEEE Globecom 2019) https://arxiv.org/pdf/2004.00507.pdf

Network-level Results
– Find labelled training samples
in a simulation platform
– Train a DNN offline (From
channel and traffic states to
association scheme)
– Outperforms benchmarks
– Nearest access point
– Highest SNR
– Game theory approach
– Close to the optimal scheme
Network-level: Mobility management entity optimizes user association [N-11]
[N-11] R. Dong, C. She, W. Hardjawana, Y. Li, and B. Vucetic, “Deep Learning for Hybrid 5G Services in Mobile Edge Computing Systems: Learn
from a Digital Twin” IEEE Trans. on Wireless Commun., Oct., 2019. https://doi.org/10.1109/TWC.2019.2927312

URLLC for Beyond 5G

URLLC for Beyond 5G
5G Road map
URLLC: Key enabler of 6G Applications
Machine-Learning Based Communication Systems
URLLC and Time Sensitive Networks
URLLC and Satellite Communications
URLLC and Quantum Computing
URLLC and Wireless Networked Control

5G Road map

Joint URLLC and Time Sensitive Networks (TSN)
URLLC Time Sensitive Networks (TSN) Flexible Spectrum
Sub ms delay
10−5 BLER
Microsec time synchronization Dedicated,
licensed or
unlicensed/shared
spectrum

https://www.smartcitiesworld.net/
6G Applications
The sixth-generation (6G) system,
with the full support of artificial
intelligence is expected to be
deployed between 2027 and 2030
– Artificial Intelligence
– Terahertz Communications
– Optical wireless technology
– Free space optic network
– Block chains
– 3D networking
– Quantum computing and
communications
– Big data analytics
– …

Satellite Communications
Existing communication network design has mainly focused
on terrestrial communication networks
LEO Satellite network has a key role to play here
Elon Musk's SpaceX is going to launch 42000 LEOs
to provide Global internet communications
Altitude of several hundreds kilometres, latency of few ms
Employ optical feeder links/ inter-satellite links and phased
array beam forming and digital processing technologies in
the Ku- and Ka-band
The system will be able to provide high speed (up to 1 Gbps
per user, which is 200 times faster than current average
internet speed), low latency broadband services for
consumers and businesses
Reliability due to rain attenuation, inter-beam-interference,
real-time data services, low latency, channel estimation due
to fast moving satellite
Visualisation of Elon Musk’s space internet
https://hipertextual.com/
Joint design of Satellite and Terrestrial communication networks

Deep Learning
Artificial Intelligence: The most important and newly introduced technology for 6G
communication systems
The upcoming 5G will support partial or very limited AI.
6G will be fully Draft supported by AI for automatization.
Advancements in machine learning will create more intelligent networks for real-
time communications in 6G.
AI will increase the efficiency and reduce the processing delay of the
communication steps. Time-consuming tasks, such as handover and network
selection, can be performed promptly by using AI.
AI is the key enabling technology for 6G for network automation and
intelligent management
https://arxiv.org/ftp/arxiv/papers/1909/1909.11315.pdf

N. Farsad, A. Goldsmith, Detection Algorithms for Communication Systems Using Deep Learning;
Machine Learning-Based Communication Systems
Existing communication system design is divided the whole system
into a chain of multiple processing blocks, each responsible for a
specific sub-task, e.g., source coding, channel coding, modulation,
channel estimation and equalization – though it is efficient, but
suboptimal – do not achieve the optimal end-to-end performance.
DL end-to-end learning of communications systems jointly
optimizes transmitter and receiver in a single process and does not
have such a rigid modular structure;
DL can be executed in parallel processing architectures with GPUs
and specialized machine learning chips – high processing speed,
low processing latency and low energy cost

S. Dörner, S. Cammerer, J. Hoydis, S. ten Brink, “Deep Learning-Based Communication Over the Air.
Deep Learning
Existing design relies on the prior mathematical modelling and
analysis. Algorithms are optimized for tractable mathematical
models, which are usually linear, stationary and Gaussian
distributed – practical systems have many imperfections and non-
linearity, which cannot be captured by such models
Deep learning (DL) based communication system (or processing
blocks) does not require a mathematically trackatble model. It treats
the system as a black box.
The recent significant advances in DL libraries and readily available
specialized ML chips can escalate the development of DL
communication systems
Machine learning communication systems

S. Dörner, S. Cammerer, J. Hoydis, S. ten Brink, “Deep Learning-Based Communication Over the Air.
Deep Learning
Current progress:
DL based channel decoding, modulation classifications, MIMO detection,
compressive sensing, compression, encryption/decryption of an
eavesdropper channel
An DL based SDR prototype of a complete uncoded end-to-end
communication systems with open source DL software libraries
Open challenges:
Biggest challenge is to scale to large message size – possible approach is
to embed the code structure and modulation information in the DL
Training SNRs – the learned system should operate at any SNR, regardless
at which SNR it was trained
Optimal choice of loss function
Channel state information learning ….
Leaning for random access …..
Machine learning communication systems

Quantum Computing
Google recently announced new record-breaking
72-qubit quantum processor, achieving quantum
supremacy, the point at which quantum computers
can perform calculations that are beyond
the capabilities of even the most advanced
supercomputers.
Time for decryption can be reduced from years to
minutes – new challenge for security protocols;
Google is also developing Quantum machine
learning chip, which can significantly speed up the
machine learning process – complexity may not be
a big issue with quantum processing
How to develop parallel communication
architecture tailored for Quantum
processor? Machine learning based
communication systems may be one possible
approach.
Google's Sycamore chip is kept cool inside their quantum
cryostat.
(Image: © Eric Lucero/Google, Inc.)
What’s the key takeaway?

Wireless Network Control
One key major application of uRLLC is in wireless networked control
Existing studies on how wireless communication can affect the control
performance, such as the effect of delay and packet dropouts on the stability
of the networked control system
The effects of practical wireless communication techniques on the
performance (i.e., stability + cost) of wireless control systems
– MIMO (diversity-multiplexing tradeoff),
– Short packet communication (latency-reliability tradeoff)
Resource allocation in future wireless control networks
– Joint downlink uplink scheduling/resource allocation
– Broadcast channel
– Multiple access channel
– Interference channel

References

[S-1] C. She, C. Yang, and T. Q. S. Quek, “Radio Resource Management for Ultra-reliable
and Low-latency Communications,” IEEE Commun. Mag., Jun. 2017.
https://doi.org/10.1109/MCOM.2017.1601092
[S-2] H. Chen, R. Abbas, M. Shirvanimoghaddam, W. Hardjawana, W. Bao, Y. Li and B.
Vucetic. "Ultra-reliable low latency cellular networks: Use cases, challenges and
approaches." IEEE Communications Magazine 56.12 (2018): 119-125.
[S-3] M. Shirvanimoghaddam et al., Short Block-Length Codes for Ultra-Reliable Low-
Latency Communications, in IEEE Communications Magazine, vol. 57, no. 2, 2019.
[S-4] M.B. Shahab, R. Abbas, M. Shirvanimoghaddam, and S. J. Johnson. "Grant-free
Non-orthogonal Multiple Access for IoT: A Survey." arXiv preprint arXiv:1910.06529 (2019).
https://arxiv.org/abs/1910.06529, accepted to appear in IEEE Communications Surveys
and Tutorials
[S-5] D. Feng, C. She, K. Ying, et al., "Towards Ultra-Reliable Low-Latency
Communications: Typical Scenarios, Possible Solutions, and Open Issues", IEEE Veh.
Tech. Mag., Jun. 2019. https://doi.org/10.1109/MVT.2019.2903657
[S-6] R. Abbas, T. Huang, M.B. Shahab, M. Shirvanimoghaddam, Y. Li and B. Vucetic.
"Grant-Free Non-Orthogonal Multiple Access: A Key Enabler for 6G-IoT.” arXiv preprint
arXiv:2003.10257 (2020). https://arxiv.org/abs/2003.10257
References – Survey Papers

References – (PHY)
[P-1] R. Abbas, M. Shirvanimoghaddam, T. Huang, Y. Li and B. Vucetic, Novel Design for
Short Analog Fountain Codes, in IEEE COMM Letter, vol. 23, no. 8, 2019.
https://doi.org/10.1109/LCOMM.2019.2910517
[P-2] M. Shirvanimoghaddam et al., Short Block-Length Codes for Ultra-Reliable Low-
Latency Communications, in IEEE Communications Magazine, vol. 57, no. 2, 2019.
[P-3] S. Jayasooriya, M. Shirvanimoghaddam, L. Ong and S. J. Johnson, Analysis and
design of Raptor codes using a multi-edge framework, in IEEE TCOM, vol. 65, no. 12, Oct.
2017. https://doi.org/10.1109/TCOMM.2017.2750179
[P-4] S. Jayasooriya, M. Shirvanimoghaddam, L. Ong, G. Lechner and S. J. Johnson, A New
Density Evolution Approximation for LDPC and Multi-Edge Type LDPC Codes, in IEEE
TCOM, vol. 64, no. 10, 2016. 13. M. Shirvanimoghaddam and S. J. Johnson, Raptor Codes
in the Low SNR Regime, in IEEE TCOM, vol. 64, no. 11, 2016.
https://doi.org/10.1109/TCOMM.2016.2600660
[P-5] M. Shirvanimoghaddam, Y. Li, B. Vucetic, J. Yuan, P. Zhang, Binary Compressive
Sensing via Analog Fountain Coding, IEEE TSP, vol. 63, no. 24, 2015.
https://doi.org/10.1109/TSP.2015.2472362
[P-6] M. Shirvanimoghaddam, Y. Li, B. Vucetic, Near-Capacity Adaptive Analog Fountain
Codes for Wireless Channels, IEEE Communications Letters, vol. 17, no. 12, 2013.
https://doi.org/10.1109/LCOMM.2013.101813.131972

References – (PHY)
[P-7] C. Yue, M. Shirvanimoghaddam, et al, Segmentation-Discarding Ordered-Statistic
Decoding for Linear Block Codes, IEEE Globcom, Kona, HI, Dec. 2019.
https://arxiv.org/abs/1901.02603
[P-8] R. Abbas, M. Shirvanimoghaddam, et al, Performance Analysis of Short Analog
Fountain Codes, IEEE Globecom, Kona, HI, Dec. 2019.
[P-9] W. Lim, M. Shirvanimoghaddam, et al, On the Design of Analog Fountain Codes for
Short Packet Communications in 5G URLLC, IEEE VTC-Spring, Honolulu, HI, Sep. 2019.
https://doi.org/10.1109/VTCFall.2019.8891550
[P-10] C. Yue, M. Shirvanimoghaddam, Y. Li and B. Vucetic, Hamming Distance Distribution
of the 0-reprocessing Estimate of the Ordered Statistic Decoder, IEEE ISIT, Paris, France,
July 2019. https://doi.org/10.1109/ISIT.2019.8849229

References – (MAC)
[M-1] M. Shirvanimoghaddam, Y. Li, M. Dohler, B. Vucetic, S. Feng, Probabilistic Rateless
Multiple Access for Machine-to-Machine Communication, IEEE TWC, vol. 14, no. 12, 2015.
https://doi.org/10.1109/TWC.2015.2460254
[M-2] R. Abbas, M. Shirvanimoghaddam, Y. Li and B. Vucetic. "A Novel Analytical Framework
for Massive Grant-Free NOMA." IEEE Transactions on Communications (2018).
https://doi.org/10.1109/TCOMM.2018.2881120
[M-3] R. Abbas, M. Shirvanimoghaddam, Y. Li and B. Vucetic, "A Multi-Layer Grant-Free
NOMA Scheme for Short Packet Transmissions." 2018 IEEE Global Communications
Conference (GLOBECOM). IEEE, 2018. https://doi.org/10.1109/GLOCOM.2018.8647968
[M-4] M. Shirvanimoghaddam, M. Condoluci, M. Dohler and S. Johnson. "On the
fundamental limits of random non-orthogonal multiple access in cellular massive IoT." IEEE
Journal on Selected Areas in Communications 35.10 (2017): 2238-2252.
https://doi.org/10.1109/JSAC.2017.2724442
[M-5] M. B. Shahab, R. Abbas, M. Shirvanimoghaddam, and S. J. Johnson. "Grant-free Non-
orthogonal Multiple Access for IoT: A Survey." arXiv preprint arXiv:1910.06529 (2019). ).

References – (Cross-Layer)
[C-6] C. She, C. Yang, and TQS Quek. "Cross-layer optimization for ultra-reliable and low-
latency radio access networks." IEEE Transactions on Wireless Communications 17.1
(2017): 127-141. https://doi.org/10.1109/TWC.2017.2762684
[C-7] C. She, et al. "Cross-layer design for mission-critical iot in mobile edge computing
systems." IEEE Internet of Things Journal (2019). https://doi.org/10.1109/JIOT.2019.2930983

References – (Network)
[N-1] C. She, C. Yang, and T. Q. S. Quek, “Radio Resource Management for Ultra-reliable and Low-
latency Communications,” IEEE Commun. Mag., Jun. 2017.
[N-2] C. She, C. Yang, and T. Q. S. Quek, “Cross-layer Optimization for Ultra-reliable and Low-latency
Radio Access Networks,” IEEE Trans. Wireless Commun., Jan. 2018.
https://doi.org./10.1109/TWC.2017.2762684
[N-3] C. She, C. Yang, and T. Q. S. Quek, “Joint Uplink and Downlink Resource Configuration for
Ultra-reliable and Low-latency Communications,” IEEE Trans. Commun., Jan. 2018.
https://doi.org./10.1109/TCOMM.2018.2791598
[N-4] C. She, Z. Chen, C. Yang, T. Q. S. Quek, Y. Li and B. Vucetic, “Improving Available Range of
Ultra-reliable and Low-latency Communications with Different Transmission Modes ”, IEEE Trans. on
Commun., Nov. 2018. https://doi.org/10.1109/TCOMM.2018.2791598
[N-5] Z. Hou, C. She, Y. Li, T. Q. S. Quek, and B. Vucetic, “Burstiness Aware Bandwidth Reservation
for Ultra-reliable and Low-latency Communications in Tactile Internet,” IEEE J. Sel. Areas Commun.,
Nov. 2018. https://doi.org/10.1109/JSAC.2018.2874113
[N-6] C. Sun, C. She, C. Yang, T. Q. S. Quek, Y. Li, and B. Vucetic, "Optimizing Resource Allocation in
Short Blocklength Regime for Ultra-reliable and Low-latency Communications", IEEE Trans. on
Wireless Commun., Jan. 2019. https://doi.org/10.1109/TWC.2018.2880907
[N-7] C. She, C. Liu, T. Q. S. Quek, C. Yang, and Y. Li, "Ultra-reliable and Low-latency
Communications in Unmanned Aerial Vehicle Communication Systems", IEEE Trans. on Commun.,
May 2019. https://doi.org/10.1109/TCOMM.2019.2896184

[N-8] Z. Hou, C. She, Y. Li, Z. Li, and B. Vucetic, “Prediction and Communication Co-design for
Ultra-Reliable and Low-Latency Communications”, IEEE TWC, 2019.
https://doi.org./10.1109/TWC.2019.2951660
[N-9] R. Dong, C. She, W. Hardjawana, Y. Li, and B. Vucetic, “Deep Learning for Radio Resource
Allocation with Diverse Quality-of-Service Requirements in 5G” IEEE TWC, submitted.
(Conference version accepted by IEEE Globecom 2019) https://arxiv.org/pdf/2004.00507.pdf
[N-10] D. Feng, C. She, K. Ying, et al., "Towards Ultra-Reliable Low-Latency Communications:
Typical Scenarios, Possible Solutions, and Open Issues", IEEE Veh. Tech. Mag., Jun. 2019.
https://doi.org/10.1109/MVT.2019.2903657
[N-11] R. Dong, C. She, W. Hardjawana, Y. Li, and B. Vucetic, “Deep Learning for Hybrid 5G
Services in Mobile Edge Computing Systems: Learn from a Digital Twin” IEEE Trans. on Wireless
Commun., Oct., 2019. https://doi.org/10.1109/TWC.2019.2927312
[N-12] C. She, Y. Duan, G. Zhao, T. Q. S. Quek, Yonghui Li, and Branka Vucetic, "Cross-Layer
Design for Mission-Critical IoT in Mobile Edge Computing Systems", IEEE Internet-of-Things J.,
early access, 2019. https://doi.org/10.1109/JIOT.2019.2930983
[N-13] C. Pradhan, A. Li, C. She, Y. Li, B. Vucetic, “Computation Offloading for IoT in C-RAN:
Optimization and Deep Learning”, IEEE Trans. on Commun., submitted.
[N-14] C. She, R. Dong, Z. Gu, et al., "Deep learning for Ultra-Reliable and Low-Latency
Communications in 6G networks", IEEE Network, submitted. https://arxiv.org/abs/2002.11045
References – (Network)

Presenter Biographies

Yonghui Li is now a Professor in School
of Electrical and Information
Engineering, University of Sydney. He is
the recipient of the Australian Queen
Elizabeth II Fellowship in 2008 and the
Australian Future Fellowship in
2012. His current research interests are
in the area of wireless communications,
with a particular focus on MIMO,
millimeter wave communications,
machine to machine communications,
coding techniques and cooperative
communications. He is Fellow of IEEE.
https://sydney.edu.au/engineering/about/our-people/academic-
staff/yonghui-li.html

URLLC for 5G and Beyond: Physical, MAC, and Network Solutions

URLLC for 5G and Beyond: Physical, MAC, and Network Solutions

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (20)

Similaire à URLLC for 5G and Beyond: Physical, MAC, and Network Solutions

Similaire à URLLC for 5G and Beyond: Physical, MAC, and Network Solutions (20)

Dernier

Dernier (20)

URLLC for 5G and Beyond: Physical, MAC, and Network Solutions