Packet Duplication for URLLC in 5G Dual
Connectivity Architecture
Jaya Rao and Sophie Vrzic
Huawei Technologies Canada
Ottawa, Ontario, Canada K2K 3J1
E-mail:{jaya.rao,sophie.vrzic}@huawei.com
Abstract – This paper addresses the problem of satisfying the
extreme requirements related to Ultra-Reliable Low Latency
Communications (URLLC) in 5G Radio Access Network (RAN).
Complementary to the existing Physical (PHY) layer techniques,
this paper focuses primarily on higher layer solutions,
particularly, on Packet Duplication (PD) as a practical and low
complexity technique for URLLC. The theoretic framework
behind PD is investigated and the recent enhancements made in
the 5G Dual Connectivity (DC) architecture for supporting PD are
discussed. For improving the radio resource utilization and to
dynamically control the activation of PD, an optimization problem
subject to URLLC constraints is formulated and solved
heuristically to give the resource configuration in terms of MCS
and PRB allocation over multiple links. Following this, it is shown
numerically that performing PD in various deployment scenarios
results in better utilization of radio resources compared to using a
single highly reliable link while effectively satisfying the URLLC
requirements.
Keywords- URLLC, Dual Connectivity Architecture, 5G NR
I. INTRODUCTION
Ultra-reliable low-latency communications (URLLC) are
characterized by extreme requirements targeted for supporting use
cases requiring high criticality, resilience and robustness. For
URLLC, the reliability requirement is intertwined with latency and
both performance metrics have to be jointly considered in system
design. This is because transmitting packets with high reliability is
consequential only if the packets are received within the sub-
milisecond latency constraint. The corresponding use cases
evaluated by 3GPP for URLLC include industry automation,
mobile eHealth, interactive augmented reality, drone
communications and connected vehicles [1]. The requirements for
these use cases as defined by 3GPP are a maximum round trip time
of 1ms on the user-plane (UP) and transmission reliability of 1-10-5
for a packet size of 32 bytes [2]. For more advanced URLLC use
cases the latency and reliability requirements can range between
0.5ms to 10ms and 1-10-5 to 1-10-9, respectively for packet sizes of
up to 300 bytes.
The existing PHY layer techniques in LTE, designed primarily
for providing high spectral efficiency, cannot be straightforwardly
extended to support the typical URLLC requirements in 5G New
Radio (NR). As an illustration, to achieve a block error rate (BLER)
target of 1-10-5 on a single link, it is necessary to have highly
favorable channel conditions at all times, use a low and robust
modulation and coding scheme (MCS) index (e.g. QPSK with 1/3
coding rate) and a bandwidth allocation of at least 20MHz [3]. For
higher reliability requirements, it is necessary to use either high
diversity orders (up to 16) or significantly increase the signal-to-
noise (SNR) and bandwidth allocation [4]. The implementation of
other PHY layer solutions based on link adaptation [5] and link
combining [6] reveal certain improvement in terms of reliability but
not without trading off major increase in functional complexity.
Also, such schemes can be applied only in certain highly restrictive
scenarios (e.g. cell center, low load, very low mobility), thus
limiting their usage to support only a moderate number URLLC-
capable devices in most deployment scenarios.
In legacy LTE RAN, the reliability requirements are
conventionally satisfied at the Radio Link Control (RLC) and
Medium Access Control (MAC) layers using the Automatic Repeat
Request (ARQ) and Hybrid ARQ (HARQ) retransmission
techniques. Although these techniques enable certain reliability
level to be achieved, the resulting latency due to packet
retransmissions exceed the sub-milisecond latency requirements for
most of URLLC use cases. Moreover, incorporating enhancements
to these techniques come at the expense of architectural
modification, increase in resource usage and in standardization
effort. As such, it is necessary to consider a fundamentally new
technique that is not only effective for URLLC but can also be
practically implemented with low complexity.
Considering the different architectural options available in 5G,
a promising solution to address the extreme requirements is packet
duplication (PD). Particularly, the PD technique can be directly
applied in the DC architecture [7] without excessively increasing
the complexity in the RAN. The fundamental principle underlying
PD involves generating multiple instances of a packet at higher
layers and transmitting the packets simultaneously over different
uncorrelated channels or transmission links [8]. At the receiver, the
redundancy and diversity in the channel conditions is exploited
such that higher transmission reliability is achieved.
The PD technique has been recently adopted by 3GPP for
satisfying the reliability and latency requirements in 5G. The
corresponding standardization effort to incorporate PD in the NR
RAN protocol stack is currently underway [9]. In light of the recent
progress, this paper provides an overview of the URLLC related
standardization activities in NR RAN. Particularly, more focus is
given towards the solutions and the enhancements made at the
higher layers (i.e. above PHY) and in the RAN architecture for
supporting the PD technique. In this regard, the system model and
the DC architecture enhancements for PD are discussed in Sections
II and III. This is followed by techniques and system design
considering fast triggering of PD via dynamic control in Section IV.
Finally, the PD related performance evaluations as well as the
conclusions are provided in Sections V and VI, respectively.
II. SYSTEM MODEL FOR PACKET DUPLICATON
At the fundamental level, the reliability of a wireless system can be
increased by transmitting the same packet over multiple redundant
links, each experiencing a different channel condition [10]. To
2018 IEEE Wireless Communications and Networking Conference (WCNC)
978-1-5386-1734-2/18/$31.00 ©2018 IEEE
realize this in practical network implementations, consider a RAN
architecture consisting of ܰ transmission links, each utilizing a
non-overlapping frequency carrier for transmitting simultaneously
to the UE (User Equipment). In this case, the overall system
reliability R can be determined as:
ܴ = 1 −ෑ(1 − ܴ)
ே
ୀଵ
(1)
where ܴ is the success probability over link ݅ ∈ ܰ. Assuming ܮ is
the overall latency, ܴܵܰ is the SNR achievable on link ݅ and Φ is
the bandwidth allocation over link ݅, then ܴ is defined as:
ܴ = ܲݎ(ܮ ≤ ߣ்)ܲݎ൫ܴܵܰ > ்ܴܵܰ |݉൯ܲݎ(Φ > ݂(ܶܤܵ)|݉) (2)
where i) ߣ் is the latency requirement for URLLC, ii) ்ܴܵܰ =
݂(ܤܮܧ்ܴ ) is the SNR threshold for achieving a BLER target
ܤܮܧ்ܴ on link ݅ when using MCS index ݉ ∈ ࡹ (from the
available MCS set M) and iii) ܶܤܵ is the transport block size of
the URLLC packet when using MCS index ݉ . Note that ܮ
includes the latency due to processing, propagation and
transmission between the transmitting and receiving Packet Data
Convergence Protocol (PDCP) entities in the RAN (i.e. UE and
access node).
From Eq. (1) and (2), clearly to increase the overall reliability
ܴ it is necessary to increase either the reliability of each link ܴ or
the number of links ܰ carrying the same packet. The
straightforward approach to improve the robustness of each link
against channel effects, and consequently ܴ , is to increase the
transmit power and allocate more radio resources over each link.
Both of these techniques, however, may adversely affect the
spectral efficiency, power efficiency and interference, hence are
not applicable for NR system design. On the other hand, to satisfy
the latency requirements in URLLC, the packets transmitted in
each link have to be received within the latency deadline of less
than 1ms. To this end, the use of PD where the packets are
proactively transmitted simultaneously, addresses both the latency
and reliability requirements without having to rely on feedback
and retransmissions as done in ARQ and HARQ in LTE.
At first glance, duplication may imply potential loss in
throughput and spectral efficiency. However, at closer inspection it
becomes clear that exploiting the diversity from using multiple
links provides the means to achieve high reliability on a statistical
basis without actually expending more resources. More precisely,
targeting a lower BLER value on a particular link and,
correspondingly, using a higher MCS index on that link makes it
possible to minimize the radio resources usage on that link.
Consequently, when multiple links are configured to support PD, it
is possible to minimize the total amount of resources required over
all links to be less than that of using a single highly reliable link.
Such techniques can be applied in practical systems in NR RAN to
meet the URLLC requirements and to balance the reliability-
resource usage trade off.
III. ARCHITECTURE ENHAMCEMENTS IN 5G NR RAN
A. NR RAN Protocol Stack Enhancements for PD
The NR RAN protocol stack in the user plane (UP) is collectively
responsible for ensuring reliable over-the-air transmission of
protocol data units (PDUs) in both uplink (UL) and downlink (DL)
directions [9]. The RRC entity, which is the primary control plane
(CP) function in RAN, is responsible for configuring the protocol
layers in both the network and the UE. The RRC is also responsible
for establishing, maintaining and releasing of the radio bearers
between the network and the UE. The radio bearers in NR are
categorized into two types namely, the data radio bearers (DRBs)
and signaling radio bearers (SRBs), both of which are used for
transmission of UP and CP packets, respectively. The DRBs are
generally configured to satisfy a set of Quality of Service (QoS)
requirements which include a guaranteed bit rate and priority level.
In comparison to DRBs, SRBs are characterized by less frequent
transmissions, smaller PDU sizes and higher scheduling priority.
In NR, a new Service Data Adaptation (SDAP) layer is
introduced for performing the mapping between the QoS flows and
DRBs. This is to ensure that the QoS flows, which originate and
terminate in core network (CN) are handled appropriately in the
RAN with the right priority treatment and resource provisioning.
Next, sequence numbering, header compression and ciphering
operations are performed in the NR PDCP to ensure in-order and
secure delivery of both the UP and CP packets. To enhance
transmission reliability, a new duplication function is incorporated
in PDCP whose role is to make duplicates of the PDUs associated
with a set of DRBs and SRBs configured by RRC. Also, each
instance of the duplicate PDU carries the same PDCP sequence
number in order to facilitate the receiving PDCP entity to detect and
remove the duplicates.
This is followed by the RLC layer which is responsible for PDU
segmentation and handling of different transmission modes which
include the Acknowledged Mode (AM) and Un-acknowledged
Mode (UM). When PD is configured the original and duplicated
PDUs are handled by two RLC entities, each correspondingly
assigned to a unique logical channel. While both RLC transmission
modes are supported with PD in NR, for URLLC the RLC operates
in the UM mode where the reception status of the RLC PDUs does
not require to be acknowledged to the transmitter to further
minimize the latency.
Subsequently, the MAC layer performs scheduling,
multiplexing and mapping of the PDUs originating from different
logical channels to transport channels. For ensuring reliable
transmission, each transport channel is assigned to a separate
HARQ process which enables transmission of ACK/NACK
feedback messages and the retransmission of PDUs. Also, for
URLLC a maximum of 1 HARQ retransmission can be
accommodated. When PD is supported, the RRC configures
mapping restrictions in the MAC to ensure that the PDUs in the two
logical channels do not end up in the same transport channel and
consequently, assigned to the same carrier in the PHY layer.
At the PHY layer the transport channels are mapped to physical
channels, after which the provisioning of the Physical Resource
Block (PRBs) and MCS selection is done in an assigned carrier. In
NR, the PHY layer is expected to support multiple numerologies,
each configured with different subcarrier spacing ranging from
15kHz to 120kHz as well as shorter transmission time interval
(TTI) of 0.125ms [11]. This is to enable greater flexibility in
supporting diverse set of use cases with varied requirements.
B. Enhancements in NR DC Architecture for PD
Dual-connectivity (DC) architecture in NR is primarily intended to
provide high throughput and high reliability by enabling the use of
radio resources from two access nodes with distinct schedulers of
the same or different radio access technologies (RATs) [7]. The
access nodes in DC consist of the Master Node (MN), which hosts
the full RAN protocol stack, and Secondary Node (SN) hosting the
lower layers (i.e. RLC, MAC and PHY). Both the MN and SN are
connected via a non-ideal backhaul over the Xn interface. While the
Xn interface supports data forwarding and flow control functions,
the fact that the packets may traverse through a non-ideal backhaul
may result in high latency, hence restricting the ability to perform
inter-node coordination. As such only semi-static coordination at
the RRC level is supported in DC. On the other hand, both MN and
2018 IEEE Wireless Communications and Networking Conference (WCNC)
SN have greater flexibility in independently scheduling resources
for the UEs.
The MN supports multiple carriers (i.e. frequency bands) which
collectively form the Macro Cell Group (MCG). The SN, in turn,
controls the Secondary Cell Group (SCG) consisting of its own set
of carriers. Also, both MN and SN are capable of supporting direct
bearers as well as split bearers. Specifically, in the case of direct
bearers the packet flow is routed directly from the CN through
either MN or SN while in split bearer case, the packet flow is split
between MN and SN using a splitting ratio configured by RRC. The
split bearer enables the UE to receive and send packets
simultaneously from two access nodes to realize higher throughput
performance.
In 5G, both access nodes host the NR RAN protocol stack and
are connected to the 5G Core Network (5GC) in standalone NR-NR
DC architecture [7] as shown in Fig. 1. In this case, the MN and SN
are referred to as Master Next-Gen Node B (MgNB) and Secondary
Next-Gen Node B (SgNB), respectively. Here, the MgNB supports
the MCG bearer and MCG split bearer while the SgNB supports
SCG bearer and SCG split bearer. In comparison, the existing LTE
DC does not support the SCG split bearer.
Fig 1: NR-NR DC Architecture. MN and SN use the NR RAN protocol stack
In the case when the MN and SN belong to different RATs in a
non-standalone architecture, where one of the nodes uses the LTE
RAN protocol stack while another node has NR protocol stack, the
generalized DC is referred to as the Multi-RAT DC (MR-DC).
Within the MR-DC architecture, there can be multiple options
depending on the type of CN available and the RAT supported at
the MN and SN. These options are listed as follows:
i) EN-DC: MN uses LTE protocol stack (eNB) and SN uses
NR protocol stack (gNB). MN is connected to (legacy)
Evolved Packet Core (EPC) and SN is connected to MN via
(legacy) X2 interface
ii) NGEN-DC: MN uses LTE protocol stack (eNB) and SN
uses NR protocol stack (gNB). MN and SN are connected to
5GC and SN is connected to MN via Xn interface
iii) NE-DC: MN uses NR protocol stack (gNB) and SN uses
LTE protocol stack (eNB). MN and SN are connected to
5GC and SN is connected to MN via Xn interface
In the regards to PD, since the existing LTE PDCP is not
enabled with the duplication function there will be certain
restrictions on the variants of DC architectures that can be used for
URLLC. Particularly, in the MCG split bearer case, PD can only be
configured in certain MR-DC architectures that host the NR PDCP.
Likewise, direct bearers can be configured for PD over multiple
carriers from any node hosting the NR PDCP, similar to the carrier
aggregation (CA) technique in LTE. While this can be applied to all
direct bearer DRBs, there are certain limitations for performing PD
in the SRB case. However, enhancements have been made in the
LTE Release 14 to support NR PDCP at the LTE node and to
remove the duplicates at the receiving LTE PDCP entity for the
SRBs when PD is performed at the transmitting PDCP entity.
In the MR-DC case, both the MN and SN have their own RRC
entities which can communicate with the UE either via direct SRBs
or split SRBs. For example the channel measurement reports
intended for the RRC entity in SN can be sent directly to the SN
using the direct SCG SRB. When configured for split SRBs, the
RRC PDUs from either MN or SN are sent via the Xn interface.
For supporting PD in the UL, the DC-capable UE uses a
common PDCP and a pair of lower layer stacks consisting of RLC,
MAC and PHY, mirroring that of the DC network. Additionally, at
the hardware level, it is assumed that the DC-capable UE utilizes
separate RF chains with different power amplifiers in order to
transmit the duplicate packets simultaneously to the MN and SN
using power control and beamforming techniques.
IV. DYNAMIC CONTROL OF PACKET DUPLICATION
As detailed in Section III, the RRC signaling is used for configuring
the radio bearers for PD in the DC architecture. While the semi-
static configuration using RRC may be adequate for duplication of
SRBs, for DRBs, which carry the bulk of the traffic in the network,
enabling PD at all times may be resource-wise wasteful. For this
reason it is necessary to incorporate a faster mechanism linked to
dynamic scheduling that allows more dynamic control of PD.
In NR RAN, it is the responsibility of the scheduler in network
to dynamically allocate physical resources in both UL and DL
transmissions for all UEs. The resource allocation can be performed
at the granularity of a TTI over a duration ranging from 1 slot to
multiple slots. Once the resource allocation decision is made, the
scheduler provides the resource assignments in the DL control
information (DCI) message in order for the UE to appropriately
decode and process the received packets. For supporting UL
transmissions, the scheduler provides the UE with UL grants,
indicating the PRBs and the MCS indices to be used in UL.
While the scheduling related messages are generally handled at
the PHY layer, certain control information related to dynamic
scheduling can be exchanged using the MAC control element
(MAC CE). In LTE, the MAC CEs sent in DL are used to carry
information pertaining to activation/deactivation of carriers, while
those in UL carry the Buffer Status Report (BSR) and Power
Headroom Report (PHR). In NR a new packet duplication
command (PDC) MAC CE (sent in DL) is introduced to facilitate
dynamic activation and deactivation of PD [9]. The following
describe the different triggering mechanisms for enabling dynamic
control of PD.
A. Network Triggered Packet Duplication
For making the dynamic PD decision, it is necessary for the
network to have the up-to-date channel and load (i.e. data in buffer)
information over all configured transmission links. The MN and SN
determine the channel quality in UL and DL based on the sounding
reference signals (SRS) and channel quality information (CQI)
reports transmitted by the UE. In UL, if the data buffers at PDCP
and RLC are non-empty, the UE may also transmit the BSR in the
UL MAC CE to each access nodes. Here, the BSR is triggered on
the basis of a logical channel group (LCG), composed of a set of
logical channels with similar QoS requirements. In the case of PD,
both the original and duplicated PDUs are mapped to two logical
channels which are associated with different LCGs to enable
independent handling of BSR generation and UL grant allocation.
Apart from BSR, the UE can also provide the PHR, indicating
the difference between the maximum allowable transmit power
level and the power currently used in UL to both MN and SN. This
Xn
MAC
RLCRLCRLC
PDCPPDCP
SDAP SDAP
PHYPHYPHY
MAC
RLCRLCRLC
PDCPPDCP
SDAP SDAP
PHYPHYPHY
MCG
Bearer
MCG
Split Bearer
SCG
Split Bearer
SCG
Bearer
MgNB SgNB
2018 IEEE Wireless Communications and Networking Conference (WCNC)
information enables the scheduler to determine the transmit power
and, consequently, the MCS index to be used in UL on the allocated
PRBs. Based on the provided buffer and channel measurements, an
optimization problem formulation, applicable at the scheduler in
each TTI for determining the optimum resource allocation over the
N available links, is given as follows:
min{௫,ெௌ,}ݔ
ே
ୀଵ
ܾ (3.0)
ݏ. ݐ. ܾ ≤ ܤ, ∀݅ (3.1)
݂(ܯܥܵ, ݔܾ) ≥ ܶܤܵ, ∀݅ (3.2)
ෑݔܤܮܧ்ܴ (ܯܥܵ, ܴܵܰ)
ே
ୀଵ
≤ 1 − ܴ (3.3)
where ݔ ∈ {0,1} determines the activation/deactivation of link ݅
and ܾ denotes the PRBs allocated on link ݅ from a total of ܤ
available PRBs. The function ݂(ܯܥܵ, ܾ) in Eq. (3.2) can be
modeled as a linear function that maps from the MCS index and
number of PRBs to a transport block size (TBS) [12]. The
reliability constraint in Eq. (3.3) follows from Eq. (1) where the
function ܤܮܧ்ܴ(ܯܥܵ, ܴܵܰ) is determined from link-level LTE
simulations with Rayleigh fading. The following algorithm
provides a low complexity technique that can be implemented at the
radio resource management (RRM) function in the network to solve
the optimization formulation in Eq. (3) heuristically. This based on
the greedy approach to determine the dynamic PD decision and the
corresponding resource allocation (MCS and PRB) to be applied on
the activated links.
Algorithm 1: Dynamic Control of Packet Duplication
Input: TBS of URLLC packet, ܴܵܰ over all links ݅ ∈ ܰ, overall
reliability target ்ܴ
Output: Number of activated links |݇|, MCS index on each link, PRB
allocation on each link ܾ
1: Sort links with index ݇ in descending order with decreasing SNR
2: Sort MCS with index ݉ in descending order
3: Initialize: Assign maximum MCS ݉ ← ݉௫ to all links
4: do
5: Set ݇ ← 1 (Selection of best link with max SNR)
6: do
7: Selected link ݈∗ ← link ݇ and assign ܯܥܵ∗ ← ݉
8: Compute ܤܮܧ்ܴ = ݂(ܯܥܵ, ܴܵܰ) and R using Eq. (1)
9: if (ܴ ≥ ்ܴ)
10: Compute ܾ = ݂(ܯܥܵ, ܶܤܵ) using Eq. (3.2)
11: Return |݇|, ܯܥܵ and ܾ, ∀݇
12: else
13: ݇ = ݇ + 1 (Select next best link)
14: while (݇ ≤ ݇௫)
15: ݉ = ݉ + 1 (Select next highest MCS index)
16: while (݉ ≤ ݉௫)
Note that Algorithm 1 determines the best transmission mode for
the UE with URLLC requirements by identifying the number of
links, |݇| to be used and the corresponding resource allocation. In
this case, when |݇| > 1 PD is activated and alternatively, when
|݇| = 1 PD is deactivated and a single best link is selected.
In the DL, the decision to activate PD results in both MN and
SN providing the resource assignments in the DCIs. It may be
possible that the decision made is to deactivate PD and to fall back
to the split bearer configuration using a traffic splitting ratio. In
either case, the PD decision will be transparent to the UE. The UE
will receive and process the received DL packets based on the DCIs
using the existing procedure in LTE.
In the UL, the decision for PD is made in the network and
conveyed to the UE via the PDC MAC CE as shown in Fig. 2.
Here, the PDC MAC CE contains a bitmap to identify the
command for activation and deactivation for each DRB configured
for PD. Note that in DC it is possible for the MN and SN to send
the MAC CEs independently from the individual schedulers. In this
case, the MN and SN are assumed to operate without coordination
and each will send the PD command based on the observation of its
own channel and loading conditions. Although, this may not result
in the optimal decision for activating and deactivating PD due to the
unavailability of complete information of the other links, it has the
advantage of avoiding the latency over the backhaul, hence suitable
for URLLC. The UE can either combine the MAC CEs received
from both access nodes or act based on the last received MAC CE.
Alternatively, in the coordinated case in DC, the node which hosts
the PDCP will send one PDC MAC CE containing the joint
decision. In both of these cases, the MN and SN should provide the
UL grants in the DCI to the UE to carry out the UL transmissions.
The MN may also activate semi-persistent scheduling (SPS)
configuration via the DCI. Based on the received grants, the UE
transmits the data on Physical UL shared channel (PUSCH) while
continuing to report the channel conditions on all activated links.
Fig 2: Signaling flow showing the dynamic control of Packet Duplication in
NR DC Architecture for UL transmission
B. UE Triggered Packet Duplication
In the UE triggered case the MN initially configures a set of DRBs
via RRC signaling for PD. However, in contrast, both the PD
decision and enforcement are made in the UE based on the
monitoring of the channel conditions on the configured links. Note
that the UE triggered case applies only for the UL transmissions
and used primarily to support URLLC.
In addition, the UE triggered approach also applies to the grant-
free or configured grant technique, introduced in NR as an
enhancements from SPS, for enabling fast UL transmission without
an UL grant. The grant-free technique relies on pre-allocated
resources which are provided by the gNB on a per-UE basis for
transmitting short URLLC packets up to K times without having to
go through the conventional dynamic scheduling procedure.
Alternatively, the gNB can configure the grant-free resources in a
common pool accessible by multiple UEs on a contention basis.
The network may also provide resource configuration (e.g.
PRBs, range of potential MCS) on the grant-free resources via RRC
signaling. In addition, the grant-free resources may contain a
validity timer, indicating the duration in which the resources in each
UE
Measure UL
Channel
Measure UL
Channel
Enforce PD decision
UL SRS
UL SRS
BSR + PHR
PDC MAC CE
URLLC Packets
URLLC Packets
MN SN
PD Decision
BSR + PHR
UL Grant
UL Grant
Channel Info on Xn
Forward Packets on Xn
RRC Configuration for PD
2018 IEEE Wireless Communications and Networking Conference (WCNC)
of the corresponding links are valid and reserved for the UEs. In
regards to PD, the network may pre-allocate the resources on
multiple configured links without explicitly activating PD. Once the
PD decision is made at the UE, it would be possible to send the
duplicate PDUs in UL on multiple grant-free resource pools. While
this may increase the probability of collision, the UE can use higher
MCS on a minimum number of resources in different pools without
compromising on the overall achievable reliability. This approach
will not only improve the grant-free resource utilization but also
lowers the collision probability.
C. Push and Pull Buffer Management Mechanisms
In the split bearer case, the BSRs are used to notify the network on
the data buffered in both the PDCP and RLC entities at UE for UL
transmissions. The BSR is triggered when a certain data threshold
level is crossed at the PDCP in UE, based on which the network
allocates the resources and provides the UE with the UL grants. In
the case of PD, the PDCP data threshold is ignored and two BSRs,
containing both original and duplicated data, are sent by UE to the
schedulers in MN and SN as shown in Fig. 3. The network may
also consider other factors which include the overall traffic load of
all UEs and channel conditions when providing the UL grants.
Fig 3: Buffer Management Mechanism when performing PD
In general, the buffer management with PD can be performed
based on the Push and Pull mechanisms. The push mechanism
refers to the case where once PD is activated the data associated
with a DRB is pre-duplicated and pushed to each of the logical
channels in the corresponding buffers in RLC. In subsequent steps,
normal BSR triggering is used to request for UL grants in order to
clear the data buffers in the RLC entities. Since the resource
allocation is done individually for two BSRs by two schedulers, it is
possible that one of the links has much faster rate in transmitting
and clearing the buffer due to favorable channel and loading
conditions at the gNB. If the rate of duplication at PDCP is
governed by the rate at which the fast link clears the buffer then the
duplicated PDUs will be pushed to each of the RLC entities. In the
other slower link, the new and existing PDUs may result in buffer
pile-up due to the unavailability in UL grants, leading to further
delay. To address this issue, the feedback from lower layers (e.g.
HARQ in MAC on the fast link) can be used to discard the un-
transmitted packets in the slow link if the packets in the fast link are
received successfully.
Alternatively, the pull mechanism involves first buffering the
data in PDCP followed by “pulling” into the RLC only if the UL
grants have been provided by the gNBs. While the pull mechanism
enables to address the slow link problem because it is not governed
by the performance of either one of the RLC entities, it requires a
sizeable buffer size in the PDCP for storing the PDUs. Also,
another adaptation that can be implemented is by linking the
duplication function at the PDCP with the availability of UL grants.
Here, the BSRs are initially generated based on a virtual amount of
data in PDCP prior to duplication and then the PDUs are sent to
lower layers after duplication once sufficient UL grants are
available. Although this approach may reduce the buffer size
requirements in PDCP and RLC, it may result in higher latency
since duplication is not performed until the UL grants are available.
When PD is deactivated, implying only a single best link is used
and duplication is not applied for new PDUs, there may still be
certain un-transmitted PDUs in the RLC entities. If the UL grants
are available for both links, then the existing packets in the RLC
can be cleared regardless of the deactivation command.
Alternatively, the UE may suspend the transmission on the
deactivated link and discard the packets in the buffer. If the
deactivation command need not be applied immediately, a timer
can be used to clear the existing buffer after which the remaining
packets can be discarded.
V. PERFORMANCE ANALYSIS
This section presents the numerical results obtained from
evaluations performed to investigate the impact of packet size,
transmission link SNR and number of links on the effectiveness of
applying PD for URLLC. The evaluations are based on link-level
simulations adapted to consider a multi-connectivity scenario in the
RAN consisting of an MN node (primary link) and multiple SN
(secondary links) nodes as described in [7][12]. In the simulations,
the network nodes perform DL transmissions with variable packet
sizes of 32, 100 and 200 bytes over a carrier frequency of 2GHz.
Each link/carrier dedicated for the UE is assumed to be allocated
with a bandwidth of 20MHz (100 PRBs per link). Additionally, the
Extended Vehicular A (EVA-70Hz) channel model is applied and
adapted to ensure that the channel conditions on different links are
independent and uncorrelated.
Fig 4: Number of PRBs vs. Packet Size for different number of links. PD is
applied only when more than 1 link is used
Fig. 4 shows the number of PRBs required for satisfying the
reliability and latency requirements of 1-10-5 and 1ms, respectively
for different URLLC packet sizes. Also, different number of links
are considered where each link is assumed to have fixed DL SNR
of 5dB. Note that, in comparison to LTE which targets lower
reliability of 1-10-3, higher number of PRBs are needed to satisfy a
more stringent reliability requirement of 1-10-5 in 5G. A key
observation from Fig. 4 is that there are scenarios where performing
PD over multiple links reduces the overall radio resource usage
while satisfying the URLLC requirements. The reason is with
multiple links, it is possible to use a higher MCS on each link
targeting a less stringent BLER value and collectively reduce the
total number of PRBs required over all links. The gain in terms of
resource reduction, however, is marginal when the packet sizes are
relatively small (i.e. less than 100 bytes). Also, the optimal number
of links with PD that minimizes the resource usage increases with
RLC RLC
MAC MAC
PHY PHY
PDCP:
Duplicate Data
RRC RRC
RLC RLC
MAC MAC
PHY PHY
ACK/NACK
ACK/NACK
Data SN:1
Data SN:1
Data Data
PDCP:
Remove Duplicate
UE Network
BSR
UL Grant
MN SN
32 Bytes 100 Bytes 200 Bytes
0
10
20
30
40
50
60
70
80
90
Nu
m
be
r
of
Re
so
ur
c
e
Bl
o
ck
s
Packet Size
1 Link
2 Links
3 Links
4 Links
2018 IEEE Wireless Communications and Networking Conference (WCNC)
the packet size. In conclusion, consideration of the packet size is
vital when performing PD over multiple links in order to realize the
benefits from lower resource usage.
The results in Fig. 5 and Fig. 6 show the number of PRBs
required for satisfying the reliability requirements of 1-10-5 and
1-10-7, respectively with varying number of links and the SNR of
both the primary link and secondary link(s). Note that when the
number of links used is greater than 2, a multi-connectivity PD
configuration is applied. The average DL SNR on the primary link
from MN is assumed to be fixed at 0dB or 5dB while the SNR on
the secondary link(s) from the SN(s) are varied by ±Δ2dB relative
to the MN SNR. Also, indications of either SNRMN ≥ SNRSN or
SNRMN < SNRSN apply in scenarios when more than 1 link is used,
otherwise only the SNR of the MN, SNRMN is applicable. In all
scenarios a fixed packet size of 32 bytes is assumed.
Fig 5: Number of PRBs vs. number of active links for fixed MN SNR and
varied SN SNR values when reliability requirement is set at R=1-10-5
Fig 6: Number of PRBs vs. number of active links for fixed MN SNR and
varied SN SNR values when reliability requirement is set at R=1-10-7
From the results in Figs. 5 and 6, it is observed that the optimum
number of links with PD that minimizes the radio resource usage is
strongly dependent on the SNR of the primary link as well as the
difference in the SNR of the secondary links with respect to the
primary link (ΔSNR). For both reliability requirements, PD with 2
links utilizes the PRBs more efficiently when the SNR of the
primary link is typically low (e.g. cell edge scenario) compared to
the case of using only a single link. Also, the optimal number of
links required to minimize the total PRBs increases with increasing
ΔSNR. This is because with the availability of secondary links with
higher SNR, it would be possible to use higher MCS on each link
such that using more links results in greater reduction in the number
of PRBs required. This also explains the observation in the shift in
the optimal links from 2 to 4 as the stringency in reliability R is
increased from 1-10-5 to 1-10-7 for 0dB SNR on the primary link.
On the other hand, as the SNR increases and ΔSNR decreases, PD
can be deactivated to avoid overspending the PRBs and the
transmissions can be performed using a single best link.
The results for the number of PRBs and optimal links are
determined assuming HARQ is not used due to excessive
retransmission delay. However, when using a short TTI in the PHY
layer, it would be feasible to support 1 retransmission and still
satisfy the URLLC latency bound. Since HARQ exploits another
dimension for diversity, the number of links required can be further
reduced in order to minimize the total PRBs. In conclusion, the PD
technique provides a viable and resource efficient option,
complementary to the techniques in the PHY layer, to satisfy the
stringent URLLC requirements.
VI. CONCLUSION
This paper studies the effectiveness of the PD technique to address
the URLLC requirements in NR DC architecture. The numerical
results affirm that optimizing the use of PD based on channel and
load conditions reduces the overall radio resource usage, although
only marginal savings is observed for small packet sizes and high
SNR conditions. Nevertheless, performing PD over higher number
of links is necessary to satisfy more stringent reliability
requirements and, at the same time, improve resource utilization.
Based on the numerical results, it is concluded that to achieve a
balanced tradeoff between reliability and resource usage under all
conditions, it is necessary to dynamically control the activation of
PD to certain scenarios when channel conditions are typically
unfavorable (e.g. low SNR, high mobility). For use cases where
the stringent URLLC requirements should also be satisfied in high
mobility scenarios, it would be necessary to enhance the DC
architecture with multi-connectivity capability to support PD over
more than two links. Here, PD will also improve the connection
robustness for V2X use cases while ensuring that packets are
reliably transmitted with low interruption time.
REFERENCES
[1] 3GPP Technical Report 22.862, “Feasibility study on new services and
markets technology enablers for critical communications; Stage 1”, v14.1.0,
Oct. 2016.
[2] 3GPP Technical Report 38.913, “Study on Scenarios and Requirements for
Next Generation Access Technologies”, v14.3.0, Jun. 2017.
[3] Samsung, “R2-170371: Packet Duplication Operations”, 3GPP TSG RAN-
WG2 Meeting #97bis, Spokane, WA, USA, Apr. 2017.
[4] N. A. Johansson, Y.-P. E. Wang, E. Eriksson, and M. Hessler, “Radio access
for ultra-reliable and low-latency 5G communications,” in 2015 IEEE
International Conference on Communication Workshop (ICCW), pp. 1184–
1189, Jun. 2015.
[5] H. Shariatmadari, Z. Li, M. A. Uusitalo, S. Iraji and R. Jäntti, "Link adaptation
design for ultra-reliable communications," 2016 IEEE International
Conference on Communications (ICC), Kuala Lumpur, pp. 1-52, May 2016.
[6] M. Soszka, M. Simsek and G. Fettweis, "On link combining methods for
highly reliable future wireless communication," 2016 ISWCS, Poznan, pp. 247-
252, Sept. 2016.
[7] 3GPP Technical Specification 37.340, “NR; Multi-connectivity; Overall
description; Stage-2”, Aug. 2017.
[8] Huawei, HiSilicon, “R2-1700172: Evaluation on Packet Duplication in Multi-
Connectivity”, 3GPP TSG RAN-WG2 NR Ad-hoc Meeting, Spokane, WA,
USA, Jan. 2017.
[9] 3GPP Technical Specification 38.300, “NR; Overall description; Stage-2”,
v0.6.0, Aug. 2017.
[10] D. Ohmann, M. Simsek, and G. P. Fettweis, “Achieving high availability
in wireless networks by an optimal number of Rayleigh-Fading links,” in
Globecom Workshop 2014, pp. 1402–1407, Dec. 2014.
[11] 3GPP Technical Specification 38.211, “NR; Physical channels and
modulation”, v0.1.0, Jul. 2017.
[12] 3GPP Technical Specification 36.212,“ Evolved Universal Terrestrial Radio
Access (E-UTRA): Multiplexing and channel coding”, v14.3.0, Jun. 2017.
1 2 3 4 5
0
10
20
30
40
50
60
70
80
90
100
Number of Links
Nu
m
be
r
of
Ph
ys
ic
al
Re
so
u
rc
e
Bl
oc
ks
1 2 3 4 5
0
10
20
30
40
50
60
70
80
90
100
Number of Links
Nu
m
be
r
of
Ph
ys
ic
al
Re
so
u
rc
e
Bl
oc
ks
SNR: 0dB-Δ2dB
SNR: 0dB-Δ1dB
SNR: 0dB
SNR: 0dB+Δ1dB
SNR: 0dB+Δ2dB
SNR: 5dB-Δ2dB
SNR: 5dB-Δ1dB
SNR: 5dB
SNR: 5dB+Δ1dB
SNR: 5dB+Δ2dB
SNRMN < SNRSN
SNRMN ≥
SNRSN
SNRMN ≥
SNRSN
SNRMN < SNRSN
1 2 3 4 5
0
10
20
30
40
50
60
70
80
90
100
Number of Links
Nu
m
be
r
of
Ph
ys
ic
al
Re
so
u
rc
e
Bl
oc
ks
1 2 3 4 5
0
10
20
30
40
50
60
70
80
90
100
Number of Links
Nu
m
be
r
of
Ph
ys
ic
al
Re
so
u
rc
e
Bl
oc
ks
SNR: 0dB-Δ2dB
SNR: 0dB-Δ1dB
SNR: 0dB
SNR: 0dB+Δ1dB
SNR: 0dB+Δ2dB
SNR: 5dB-Δ2dB
SNR: 5dB-Δ1dB
SNR: 5dB
SNR: 5dB+Δ1dB
SNR: 5dB+Δ2dB
SNRMN ≥
SNRSN
SNRMN ≥
SNRSN
SNRMN < SNRSN
SNRMN < SNRSN
2018 IEEE Wireless Communications and Networking Conference (WCNC)
学霸联盟