principles of ultra-reliable low latency
communications (URLLC)
Petar PopovskiAalborg University
Denmark
5G V2X Communications @ KCL, London, June 11, 2018
5G V2X Communications @ KCL, London, June 11, 2018
outline
▪ future connectivity landscape
▪ URLLC performance and statistics
▪ URLLC building blocks
▪ wireless network slicing in 5G
5G V2X Communications @ KCL, London, June 11, 2018
the future wireless connectivity landscape
▪ can be seen as eigenvalues for composing services,
e.g. in Virtual Reality,
rather than three isolated services.
5G V2X Communications @ KCL, London, June 11, 2018
future wireless connectivity landscape
5G but a lot of (great!) other wireless systems
▪ connectivity type not necessarily provided by the 5G radio interface
▪ LPWA, 802.11ah, etc.
5G V2X Communications @ KCL, London, June 11, 2018
distilled service requirements
eMBB
▪ acceleration of 4G, large payloads, active over longer periods
▪ maximize rate, moderate reliability (e.g. 10E-3)
mMTC
▪ fix low rate, unknown active subset from a massive device set
▪ maximize arrival rate, low reliability (e.g. 10E-1)
URLLC
▪ intermittent transmissions, but from a much smaller device set
▪ offer high reliability (e.g. 10E-5) while localized in time
5G V2X Communications @ KCL, London, June 11, 2018
the IoT modes: massive and ultra-reliable access
100 Mbps 95% of the time
or100 kbps 99.999% of the
time
error probability
da
ta r
ate
reliability limit
for control
information
1 Mbps from 100 devices
or10 kbps from 10000 devices
# devices
da
ta r
ate
access
protocol
limit
5G V2X Communications @ KCL, London, June 11, 2018
adoption of ultra-reliable communication
we need to divide the applications into two groups
▪ cable replacement
how would we design a system
if we could trust to the wireless
as much as to the wired?
▪ ”native” wireless applications
which new systems can we think of once
we are empowered with wireless connectivity?
5G V2X Communications @ KCL, London, June 11, 2018
MTC use cases
mMTC▪ environmental monitoring
of large areas
▪ large infrastructuresroads, ports, industrial plants
▪ available parking places
▪ management of object fleets
vehicles, bicycles
URLLC▪ commercial and public safety
▪ industrial control and automation
▪ smart energy and smart grid
▪ V2X and UAV control
▪ Augmented Reality (AR) and digital interaction with physical objects
5G V2X Communications @ KCL, London, June 11, 2018
a communication engineer models known unknowns
channel
state
noise interference
objective: find 𝛼 and, if 𝛼 = 1, find also 𝑥
a simple communication-theoretic model
𝑦 = ℎ ∙ 𝛼 ∙ 𝑥 + 𝑧 + 𝑖
user activity
5G V2X Communications @ KCL, London, June 11, 2018
ultra-reliability requires to
▪ model accurately the known unknowns
▪ bound the impact of the unknown unknowns
▪ the standard culprit 𝑧 seems easy
▪ interference can be arbitrarily varying
a simple communication-theoretic model
𝑦 = ℎ ∙ 𝛼 ∙ 𝑥 + 𝑧 + 𝑖
5G V2X Communications @ KCL, London, June 11, 2018
sources of uncertainty
▪ activity 𝛼 is the problem of a MAC protocol
▪ ℎ is the problem of channel estimation
and channel knowledge
▪ 𝑖 is a matter of interference management
and spectrum regulation
– spectrum license is paid to acquire the right to
control interference.
a simple communication-theoretic model
𝑦 = ℎ ∙ 𝛼 ∙ 𝑥 + 𝑧 + 𝑖
5G V2X Communications @ KCL, London, June 11, 2018
the worst case is when there is no prior informationabout the user activity
▪ random access
grant-free access means thatthe packet reception in the uplink
is not conditioned on a correct downlink reception
▪ can improve latency, even reliability
its uncertainty is removed by
▪ scheduling
▪ the receiver predicts the activity variable
the user activity 𝛼
5G V2X Communications @ KCL, London, June 11, 2018
URLLC performance and statistics
5G V2X Communications @ KCL, London, June 11, 2018
latency-reliability characterization
latency t
tR: time of data reception
reliability Pr(𝑡𝑅 ≤ 𝑡)
11-Pe
diversity:
time?
frequency
antennas
interfaces
5G V2X Communications @ KCL, London, June 11, 2018
design targets
latency
reliability
1
broadband rate-
oriented systems
latency
reliability
1
ultra-reliable low latency
communication URLLC
5G V2X Communications @ KCL, London, June 11, 2018
▪ in absence of interference,
we need to characterize the lower tail of 𝛾𝑆
▪ if 𝛾𝑆 is known,
we need to characterize the upper tail of 𝛾𝐼
a simple error model
Pr 𝐸 = Pr𝛾𝑠
1 + 𝛾𝐼< 𝛾𝑡ℎ
SINR
5G V2X Communications @ KCL, London, June 11, 2018
▪ assume that the interference is absent.
▪ we (somehow) know that the channel is Rayleigh.
▪ the target error rate is 𝜀𝑈, average SNR is ҧ𝛾𝑆
how do we choose the rate R?
channel uncertainty in URLLC
Pr 𝐸 = Pr log2 1 + 𝛾𝑠 < 𝑅
𝑅 = log2 1 + ҧ𝛾𝑆 ln1
1 − 𝜀𝑈where is the problem?
5G V2X Communications @ KCL, London, June 11, 2018
▪ the knowledge of average SNR
is based on n collected samples
▪ when n is low,
the rate should R be chosen
very conservatively
▪ online update of the estimate
and rate (or power) adaptation
channel uncertainty in URLLC
P. Popovski et al., ”Ultra-Reliable Low Latency Communication is Difficult: A Statistical
Assessment”, in preparation
5G V2X Communications @ KCL, London, June 11, 2018
building blocks for URLLC
5G V2X Communications @ KCL, London, June 11, 2018
▪ channel models
▪ transmission of short packets
▪ high diversity
▪ lean protocol design with respect to latency
– focus on control information
▪ network architecture
▪ wireless slicing and coexistence with other services
5G V2X Communications @ KCL, London, June 11, 2018
▪ channel models
▪ transmission of short packets
▪ high diversity
▪ lean protocol design with respect to latency
– focus on control information
▪ network architecture
▪ wireless slicing and coexistence with other services
5G V2X Communications @ KCL, London, June 11, 2018
currently there is lack of experimental evidence
for URC-relevant statistics of wireless channels
initial analysis of common wireless channel models in
URC regime
▪ block fading
▪ 𝑃𝑅 is the minimal SNR to decode data rate 𝑅
▪ the analysis reveals the URC-behavior:
Pr𝑃𝑅ത𝑃< 𝐿 ≈ 𝜀 ≈ 𝛼
𝑃𝑅ത𝑃
𝛽
wireless channel model behavior
in ultra-reliable regime
5G V2X Communications @ KCL, London, June 11, 2018
two-wave model with equal amplitudes
represents one of the worst cases
deterministic two-path model
HTX
HRX
VTW =Gλ
4⇡dLe− j 2⇡
λdL + Γ
Gλ
4⇡dRe− j 2⇡
λdR
where dL/ dR is the direct/ reflected path distance.
envelope
r = |VTW | = |⇢1 + ⇢2 exp jφ|
P. Popovski (Aalborg Uni) ult ra-reliable wireless ITA, San Diego, Feb 2017 6 / 17
outdoor physical setupoutdoor physical setup
5G V2X Communications @ KCL, London, June 11, 2018
indoor case dominated by
diffuse components,
good for high reliability
indoor physical setup
P. Eggers, M. Angjelichinoski, and P. Popovski, ”Wireless Channel Modeling Perspectives for
Ultra-Reliable Low Latency Communications”, available on Arxiv, 2018.
5G V2X Communications @ KCL, London, June 11, 2018
an example of a short packet format
UNB (ultra narrowband) system
reliability of the packet reception is a product of the
reliabilities of different parts
Pr 𝑠𝑢𝑐𝑐𝑒𝑠𝑠 = Pr 𝑃𝐴 Pr 𝑠𝑦𝑛𝑐 Pr 𝐼𝐷 Pr 𝑑𝑎𝑡𝑎 …
5G V2X Communications @ KCL, London, June 11, 2018Philippe Petit,
http://www.msnbc.com/msnbc/philippe-petit-twin-towers-balancing-act-remembered#slide1
▪ repetition coding for control information inefficient
▪ the proverbial 1-bit feedback becomes questionable
5G V2X Communications @ KCL, London, June 11, 2018
communication theory and protocol information
5G V2X Communications @ KCL, London, June 11, 2018
at short blocklengths there is a penalty
that keeps the rate away from capacity.
AWGN
SNR 0 dB
fundamental theory of finite blocklength transmission
5G V2X Communications @ KCL, London, June 11, 2018
▪ low SNR
▪ 10 bytes control information
▪ 10 bytes data
▪ same amount of channel uses
probability of error
10e-3
probability of error
10e-6
gain in reliability
5G V2X Communications @ KCL, London, June 11, 2018
mixing data and control information has energy cost
M
D
time
M
Dfrequency
time
fre
qu
en
cy
the notion of frame in cellular
systems should be revisited
5G V2X Communications @ KCL, London, June 11, 2018
separated data and metadata
useful for energy efficiencydata for Bob, Carol turns off her
receiver after the metadata
Alice
Bob Carol
joint data and metadata
better coding of the metadata
however, everybody decodes
everything
Alice
Bob Carol
connection between packetization and energy efficiency
5G V2X Communications @ KCL, London, June 11, 2018
some observations
▪ basic tradeoff between
energy efficiency and ultra-reliability
▪ departure from the common causal relationship
metadata -> data
▪ low latency usually means sending with
few channel uses (DoF)
– DoF can be increased in e.g. frequency or space
5G V2X Communications @ KCL, London, June 11, 2018
downlink communication to K users
▪ a user is active and there is a packet for her
with probability q.
▪ the message for each active user is drawn randomly
from a set of predefined message sizes.
▪ metadata should inform about
▪ who is active
▪ the message size.
▪ K. F. Trillingsgaard and P. Popovski, "Downlink Transmission of Short Packets:
Framing and Control Information Revisited," in IEEE Transactions on
Communications, vol. 65, no. 5, pp. 2048-2061, May 2017.
example:
a theoretical treatment of downlink framing
5G V2X Communications @ KCL, London, June 11, 2018
conventional framing
with pointers
example:
a theoretical treatment of downlink framing
ptr 2
ptr 3 UG 1 UG 2 UG 3
ptr 4 UG 4
m essages withm essage sizes messages wit h
alternative framing
5G V2X Communications @ KCL, London, June 11, 2018
▪ new tradeoff arises for short packets– latency is minimized when all packets are jointly encoded;
– power is minimized when each packet is encoded separately.
K=32 users
latency-energy tradeoff
5G V2X Communications @ KCL, London, June 11, 2018
massive MIMO and ultra-reliability
pros
▪ very high SNR links
▪ quasi-deterministic links, fading immunity
▪ extreme spatial multiplexing capability
cons
▪ expensive CSI acquisition procedure
▪ additional protocol steps
5G V2X Communications @ KCL, London, June 11, 2018
massive MIMO/URLLC: mitigating the CSI problem
downlink beamforming based on channel structure
non-coherent energy detection in the uplink
A. Sabin-Bana, M. Angjelichinoski, E. de Carvalho and P. Popovski, ”Massive MIMO for Ultra-
reliable Communications with Constellations for Dual Coherent-noncoherent Detection”, in
IEEE WSA, Bochum, 2018.
5G V2X Communications @ KCL, London, June 11, 2018
1
latency:x
reliability:P(X≤x)
latencydistribution
Pe
timeout
TransmissionerrorsInfrastructurefailures
cloning
2-out-of-3
transmission
strategies
multi-interface transmission
interface diversity
5G V2X Communications @ KCL, London, June 11, 2018
results based on lab measurements▪ 1 day, 100 ms interval
▪ Wi-Fi– achieves 10 ms for 90% of
packets
– but 99% requires almost 100 ms
▪ cellular: LTE and HSPA– also requires ~100 ms for 99%
▪ cloning (1 copy per IF)– 99% at 25 ms
– 99.999% at 60 ms latency1 2 5 10 20 50 100 200 500
l [ms]
0.99999
0.9999
0.999
0.99
0.9
0
RTT
LTE
HSPA
Wi-Fi
Cloning all
2-out-of-3
interface diversity
experimental results
J. J. Nielsen, R. Liu, and P. Popovski, “Ultra-Reliable Low Latency Communication
(URLLC) using Interface Diversity”, IEEE Transactions on Communications,
accepted, available at ArXiv, 2017.
5G V2X Communications @ KCL, London, June 11, 2018
* Source: Ciscus Sarasota
final remarks: protocol challenge is immensely larger
9. RRC Conn. Reconf. Compl.
8. RRC Conn. Reconfiguration
7. RRC Security Mode Complete
6. RRC Security Mode Command
5. RRC Conn. Setup Complete
4.b. RRC Conn. Setup
4.a. Contention resolution
3. RRC Conn. Request
2. Random access response
1. Random Access preamble
UE eNB
10. Small Data Payload
access
Attempt
connection
establishment
5G V2X Communications @ KCL, London, June 11, 2018
summary and outlook
▪ ultra-reliable wireless has the potential to
profoundly change systems and devices
▪ essential:
▪ short packet transmission
▪ communication-theoretic attention to the control information
▪ every step in the protocol needs a careful reliability design
▪ careful use of diversity
▪ large number of steps in real protocols
impair reliability and latency
▪ lean protocol design
coexistence of URLLC
with the other 5G services
joint work with
Kasper F. Trillingsgaard, Aalborg University, Denmark
Osvaldo Simeone, King’s College London, UK
Giuseppe Durisi, Chalmers University, Sweden
5G V2X Communications @ KCL, London, June 11, 2018
5G V2X Communications @ KCL, London, June 11, 2018
reminder: distilled service requirements
eMBB
▪ acceleration of 4G, large payloads, active over longer periods
▪ maximize rate, moderate reliability (e.g. 10E-3)
mMTC
▪ fix low rate, unknown active subset from a massive device set
▪ maximize arrival rate, low reliability (e.g. 10E-1)
URLLC
▪ intermittent transmissions, but from a much smaller device set
▪ offer high reliability (e.g. 10E-1) while localized in time
5G V2X Communications @ KCL, London, June 11, 2018
the problem of slicing
slicing: share the resource while providing
heterogeneous guarantees to different services
uplink scenario
more challenging
due to lack of coordination
eMBB mMTCURLLC
5G V2X Communications @ KCL, London, June 11, 2018
two types of slicing
eMBB mMTC URLLC idle
fre
que
nc
y
t ime
fre
que
nc
y
t ime
fre
que
nc
ies
rese
rved
for
UR
LL
C
freq
ue
nc
ies
all
oc
ate
dfo
r U
RL
LC
non-orthogonal
eMBB mMTC URLLC idlefr
eq
ue
nc
y
t ime
fre
que
nc
y
t ime
fre
que
nc
ies
rese
rved
for
UR
LL
C
freq
ue
nc
ies
all
oc
ate
dfo
r U
RL
LC
eMBB mMTC URLLC idle
fre
que
nc
y
t imefr
eq
ue
nc
yt ime
fre
que
nc
ies
rese
rved
for
UR
LL
C
freq
ue
nc
ies
all
oc
ate
dfo
r U
RL
LC
orthogonal
5G V2X Communications @ KCL, London, June 11, 2018
system model
▪ 𝐹 frequency radio resources
▪ 𝑆 minislots
▪ eMBB transmission takes one
frequency resource
▪ 𝑎𝑈 is the probability of active
URLLC device in a minislot
▪ an URLLC transmission
spreads over 𝐹𝑈 frequencies
▪ 𝐴𝑀 is the Poisson-distributed
number of active mMTCs
radio resource at a specific frequency channelfr
equ
en
cy
ch
an
ne
l
t imeminislot
t ime slot
eMBB mMTC URLLC idle
1
2
3
4
5
6
7
Fig. 3. An example of time-frequency grid with F = 7 resources and nS = 6 minislots. A single resource (frequency channel)
is allocated for mMTC transmission. Each URLLC transmission is spread over FU = 4 frequency channels.
since this is the key transmission phase for this type of traffic, due to the massive population of
devices. Extensions of our model will be discussed in Section V-B.
Each radio resource f isassumed to bewithin the time- and frequency-coherence interval of the
wireless channel, so that the wireless channel coefficients are constant within each radio resource.
Furthermore, we assume that the channel coefficients fade independently across the F radio
resources. The channel coefficients of the eMBB, URLLC, and the mMTC devices, which we
denote by HB ,f , HU,f , and Hm,f , m 2 { 0, . . . , AM } ,1 are independent and Rayleigh distributed,
i.e., HB ,f ⇠ CN (0,ΓM ), HU,f ⇠ CN (0,ΓU ), and Hm,f ⇠ CN (0,ΓM ) for m 2 { 0, . . . , AM }
across all radio resources f 2 1, ..., F . The channel gains for the three services in a radio resource
f are denoted by GB ,f = |HB ,f |2, GU,f = |HU,f |2, and Gm,f = |Hm,f |2 for m 2 1, ..., AM .
The average transmission power of all devices is normalized to one. The differences in the
actual transmission power across various users and in the path loss are accounted for through
the average channel gains ΓB , ΓU , and ΓM . Furthermore, the power of the noise at the BS is
1Throughout, we use the convention that the subscripts B , U , and M indicate a quantity referring to eMBB, URLLC, and
mMTC, respectively.
8
5G V2X Communications @ KCL, London, June 11, 2018
the received signal in a minislot
𝐘𝑠,𝑓 = 𝐻𝐵,𝑓𝐗𝐵,𝑓 + 𝐻𝑈,𝑓𝐗𝑈,𝑠,𝑓 +
𝑚=1
𝐴𝑀
𝐻 𝑚 ,𝑓𝐗 𝑚 ,𝑠,𝑓 + 𝐙𝑠,𝑓
not a classical multiple access channel
▪ different arrivals, different decoding criteria, etc.
independent Rayleigh-faded 𝐻𝑖,𝑓
if there is no transmission, 𝐗𝑖,𝑓 = 0
5G V2X Communications @ KCL, London, June 11, 2018
some more bits and pieces about the model
eMBB
▪ has a full CSI and transmits with channel inversion
▪ not transmitting w.p. 1 − 𝑎𝐵 results in outage
𝑟𝐵,𝑓 = log2 1 + 𝐺𝐵,𝑓tar
URLLC
▪ find maximal rate 𝑟𝑈 that satisfies 𝜀𝑈
▪ no CSIT and no power adaptation
Pr 𝐸𝑈 =Pr1
𝐹𝑈
𝑓=1
𝐹𝑈
log2 1 + 𝐺𝑈,𝑓 < 𝑟𝑈 ≤ 𝜀𝑈
5G V2X Communications @ KCL, London, June 11, 2018
some more bits and pieces about the model
mMTC: use of successive interference cancellation (SIC)
SNRs: 𝐺[1] ≥ 𝐺[2] ≥ ⋯ ≥ 𝐺[𝐴𝑀]
SINR: 𝜎 𝑚0=
𝐺[𝑚0]
1+σ𝑚=𝑚0+1𝐴𝑀 𝐺[𝑚]
decoding condition: log2 1 + 𝜎 𝑚0≥ 𝑟𝑀
5G V2X Communications @ KCL, London, June 11, 2018
the reliability diversity
𝜀𝑈 ≪ 𝜀𝐵 ≪ 𝜀𝑀design that benefit from heterogeneous reliability requirements
example of interfering eMBB and URLLC:
(1 − 𝜀𝑈)1 − 𝜀𝑈 1 − 𝜀𝐵
′
≥ (1 − 𝜀𝐵)
5G V2X Communications @ KCL, London, June 11, 2018
the reliability diversity
𝜀𝑈 ≪ 𝜀𝐵 ≪ 𝜀𝑀design that benefit from heterogeneous reliability requirements
example of interfering eMBB and URLLC:
(1 − 𝜀𝐵) always ≤ (1 − 𝜀𝑈)
5G V2X Communications @ KCL, London, June 11, 2018
slicing for eMBB and URLLC
orthogonal non-orthogonal
with SIC
Pr1
𝐹𝑈
𝑓=1
𝐹𝑈
log2 1 +𝐺𝑈,𝑓
1 + 𝐺𝐵,𝑓tar < 𝑟𝑈 ≤ 𝜀𝑈
5G V2X Communications @ KCL, London, June 11, 2018
slicing for eMBB and URLLC
non-orthogonal
with puncturing
perspective of the
eMBB
eMBB uses erasure code of rate 1 −𝑘
𝑆
and thus has a decreased rate
5G V2X Communications @ KCL, London, June 11, 2018
slicing for eMBB and URLLC: results
Γ𝑈 > Γ𝐵
0 2 4 6 8 10 12
r B ,sum
0.0
0.5
1.0
1.5
2.0
2.5
3.0
r U
orthogonal
orthogonal (LB)
SIC
SIC (LB)
puncturing
puncturing (LB)
5G V2X Communications @ KCL, London, June 11, 2018
slicing for eMBB and URLLC: results
Γ𝑈 < Γ𝐵
0 5 10 15 20 25 30 35 40
r B ,sum
0.0
0.2
0.4
0.6
0.8
1.0
1.2
r Uorthogonal
orthogonal (LB)
SIC
SIC (LB)
puncturing
puncturing (LB)
5G V2X Communications @ KCL, London, June 11, 2018
slicing for eMBB and mMTC
we look into a single radio frequency resource
▪ orthogonal slicing achieved by time-sharing
▪ non-orthogonal slicing achieved by SIC
SINR: 𝜎 𝑚0=
𝐺[𝑚0]
1+𝐺𝐵,𝑓tar+σ𝑚=𝑚0+1
𝐴𝑀 𝐺[𝑚]
5G V2X Communications @ KCL, London, June 11, 2018
slicing for eMBB and mMTC: results
three regimes
(1) small 𝑟𝐵 (2) intermediate 𝑟𝐵 (3) large 𝑟𝐵
0 1 2 3 4 5 6 7
r B
0
20
40
60
80λ
M
SIC, ΓB = 10 dB
SIC, ΓB = 20 dB
SIC, ΓB = 30 dB
orthogonal, ΓB = 10 dB
orthogonal ΓB = 20 dB
orthogonal, ΓB = 30 dB
5G V2X Communications @ KCL, London, June 11, 2018
summary and outlook
▪ simple model that captures the features of the 3 services
▪ non-orthogonal slicing can be beneficial, but not always
▪ reliability diversity plays a role in setting design guidelines
many extensions possible
▪ use the model to analyze the slicing for all thee services
▪ multiple URLLC devices
▪ multiple mMTC channels and hopping
▪ mMTC repetitions
5G V2X Communications @ KCL, London, June 11, 2018
additional references
▪ P. Popovski, J. J. Nielsen, C. Stefanovic, E. de Carvalho, E. Strom, K. F. Trillingsgaard, A.-S. Bana, D. M. Kim, R. Kotaba, J. Park, R. B. Sørensen, "Ultra-Reliable Low-Latency Communication (URLLC): Principles and Building Blocks", IEEE Network Magazine, Special issue on 5G for Ultra-Reliable Low Latency Communications, in revision, 2017.
▪ K. F. Trillingsgaard, and P. Popovski, “Generalized HARQ Protocols with Delayed Channel State Information and Average Latency Constraints”, accepted for publication in IEEE Transactions on Information Theory, 2017.
▪ A. Kalør, R. Guillaume, J. Nielsen, A. Mueller and P. Popovski, “Network Slicing in Industry 4.0 Applications: Abstraction Methods and End-to-End Analysis”, IEEE Transactions on Industrial Informatics (SS on From Industrial Wireless Sensor Networks to Industrial Internet-of-Things), 2017.
▪ G. Durisi, T. Koch and P. Popovski, "Toward Massive, Ultrareliable, and Low-Latency Wireless Communication With Short Packets," Proceedings of the IEEE, vol. 104, no. 9, pp. 1711-1726, Sept. 2016.
▪ P. Popovski, ”Ultra-Reliable Communication in 5G Wireless Systems”, 1st International Conference on 5G for Ubiquitous Connectivity, Levi, Finland, November 2014.
▪ P. Popovski, K. F. Trillingsgaard, O. Simeone, and G. Durisi, ”5G Wireless Network Slicing for eMBB, URLLC, and mMTC: A Communication-Theoretic View”, available on Arxiv, 2018.