ArticlePDF Available

Abstract and Figures

Wireless industrial networks require reliable and deterministic communication. Determinism implies that there must be a guarantee that each data packet will be delivered within a bounded delay. Moreover, it must ensure that the potential congestion or interference will not impact the predictable properties of the network. In 2016, IEEE 802.15.4-Time-Slotted Channel Hopping (TSCH) emerged as an alternative Medium Access Control to the industrial standards such as WirelessHART and ISA100.11a. However, TSCH is based on traditional collision detection and retransmission, and can not guarantee reliable delivery within a given time. This article proposes LeapFrog Collaboration (LFC) to provide deterministic and reliable communication over an RPL-based network. LFC is a novel multi-path routing algorithm that takes advantage of route diversity by duplicating the data flow onto an alternate path. Simulations and analytical results demonstrate that LFC significantly outperforms the single-path retransmission-based approach of RPL+TSCH and the state-of-the-art LinkPeek solution.
Content may be subject to copyright.
From Best-Effort to Deterministic
Packet Delivery for Wireless Industrial IoT Networks
Remous-Aris Koutsiamanis, Georgios Z. Papadopoulos, Member, IEEE,
Xenofon Fafoutis, Member, IEEE, Juli´
an M. Del Fiore, Pascal Thubert, and Nicolas Montavont
AbstractWireless industrial networks require reliable and
deterministic communication. Determinism implies that there must
be a guarantee that each data packet will be delivered within a
bounded delay. Moreover, it must ensure that the potential congestion
or interference will not impact the predictable properties of the
network. In 2016, IEEE 802.15.4-Time-Slotted Channel Hopping
(TSCH) emerged as an alternative Medium Access Control to
the industrial standards such as WirelessHART and ISA100.11a.
However, TSCH is based on traditional collision detection and
retransmission, and can not guarantee reliable delivery within a
given time. This article proposes LeapFrog Collaboration (LFC) to
provide deterministic and reliable communication over an RPL-based
network. LFC is a novel multi-path routing algorithm that takes
advantage of route diversity by duplicating the data flow onto an
alternate path. Simulations and analytical results demonstrate that
LFC significantly outperforms the single-path retransmission-based
approach of RPL+TSCH and the state-of-the-art LinkPeek solution.
Index Terms—Deterministic Networks, Multi-path Routing
Algorithm, Route Diversity, Leapfrog Collaboration
I. INT RODUCTION
Industry 4.0 is an emerging domain of application for the Internet
of Things (IoT), with goals to reduce the management cost and to
contribute to the automation of the Operational Technology (OT)
found in production chains in factories [
1
], [
2
]. Cost reduction can
be achieved, in particular, by replacing the existing cables with
a wireless medium, as long as an appropriate level of service for
critical applications can still be guaranteed at all times. To that aim,
the network must exhibit deterministic performance in terms of
network reliability and timely delivery [
3
], [
2
], [
4
]. More precisely,
an industrial communication framework must provide several nines
of reliability in data delivery. For instance, several consecutive
losses in an industrial automation control loop are sufficient to
stop a production chain. Moreover, it should guarantee a worst
case latency for a data packet across the network. This latency
must be known in advance, and remain constant throughout the
lifetime of the associated path. In order to replace wires, a wireless
network should exhibit a high delivery ratio with an ultra-low
jitter, regardless of transient variations in the wireless medium and
of the network congestion. However, the currently deployed IoT
technologies focus on best-effort traffic, where the data packets
may incur delays due to retransmission, queuing and rerouting.
Time-Slotted Channel Hopping (TSCH) is one of the Medium
Access Control (MAC) protocols defined in IEEE 802.15.4-2015
standard [
5
]. TSCH is a scheduled MAC-layer technology that
is especially suited for industrial networks since it provides strict
X. Fafoutis is with DTU Compute, Technical University of Denmark (e-mail:
xefa@dtu.dk). This work was conducted when he was with the University of Bristol,
UK.
guarantees, i.e., low-power, low-delay and reliable channel access.
However, TSCH is prone to retransmissions when a data packet
transmission fails, due to low link quality (i.e., multipath fading,
distance) or external interferences which may decrease reliability [
6
].
Scenarios such as factory automationwhere the product can be for
instance cars, require Low-power and Lossy Networks (LLNs) that
consist of hundreds of sensors and actuators communicating LLN
Border Router (LBR) [
7
], [
8
]. In order to extend the network beyond
the radio coverage of one node, a mesh technology enables a node
to act as a relay for others, but, beyond one hop, it will require a pro-
tocol for routing packets throughout the network. The IPv6 Routing
Protocol for Low-Power and Lossy Networks (RPL) [
9
] is one of the
most adopted routing protocol for the IoT. RPL builds a Destination
Oriented Directed Acyclic Graph (DODAG) using a distance-vector
technique whereby each node selects one or more parent(s), acting
as a relay toward its root based on a common objective function.
The resulting acyclic topology is also used to route back to the node,
using source routing in the so-called Non-Storing Mode (NSM), in
which case the root is aware of the network topology. Phinney et al.
[
7
] describe traffic patterns and network topologies in the industrial
context and how RPL can provide the baseline protocol to address
some specific applications. RPL is commonly used with 6TiSCH
and is perceived as a solution for out-of-band industrial information
transfer. As such, it is positioned for routing information which is not
a part of the industrial process itself, but which is necessary auxiliary
information for enhancing the industrial process. For example,
diagnostics and asynchronous alerts are within the application field.
We propose extensions to enhance reliability and determinism in
such a context. The work presented in this article is using RPL
NSM as the baseline, and extends it with multi-paths redundancy.
This article investigates the forwarding mechanism and proposes
to duplicate the data flow on alternate path, where multiple copies of
the same data packet traverse on parallel paths through the wireless
network. The proposed scheme, named LeapFrog Collaboration
(LFC), allows combating potential link failures on a single path and
exploiting path diversity in a wireless network to avoid retransmis-
sions as much as possible. Since the first copy that arrives at the root
is the one that matters, LFC lowers end-to-end delay performance.
Furthermore, LFC comes with a topology-adapted scheduling algo-
rithm that guarantees data delivery from the source to the destination
within a slotframe (or not at all) and, thus, bounds jitter. Given the
introduced overhead, we target lower bandwidth applications, such
as critical monitoring or alerts. For these uses, LFC is implemented
over IEEE 802.15.4-TSCH and RPL to reach network reliability
above 99% while providing ultra-low <15ms jitter performance.
This article extends [
10
], making the following additional
contributions:
EB
A > D
C -> D
B -> D
D -> E
cell
Slotframe
Channel offset
0 1 2 3 4
0 1 2 3 4
A -> D
A ED
C
B
dedicated slots
shared slot
Fig. 1:
An example of TSCH scheduling for node D. A
D stands for ’A
transmits to D’, while EB cells are used for broadcast and advertisement
frames.
It comes with over-provisioning strategy, where one additional
timeslot is reserved per data transmission, to guarantee the
ultra-high Packet Delivery Ratio (PDR).
It introduces a scheduler that is adapted to the topology,
whereby each data packet is delivered to the destination within
one slotframe (101 timeslots in our examples).
It provides analytical expressions for the calculation of per-
formance metrics of LFC, including the delay-jitter trade-off,
the PDR, and upper limits on the end-to-end delay and jitter.
It evaluates the robustness of the proposed scheme under
various link qualities, on top of the COOJA simulator. In
addition, LFC is compared to single-path retransmission based
approach (i.e., RPL
+
TSCH configuration) as well as the
state-of-the-art solution LinkPeek [11].
II. BASELINE PROTOCOLS AND LFC OVERVIEW
A. IEEE 802.15.4-2015 TSCH
At its core, TSCH uses Time Division Multiple Access (TDMA)
and Frequency Hopping Spread Spectrum (FHSS) techniques to
achieve high network reliability, reduce energy consumption and
mitigate multi-path fading and the impact of external interference.
In a TSCH network, nodes are constantly synchronized. The time is
sliced into timeslots of equal length, sufficient enough to transmit a
data packet and to receive an acknowledgement. A set of timeslots
constructs a slotframe that repeats perpetually. Furthermore, an
Absolute Sequence Number (
ASN
) is assigned to each timeslot to
count the number of timeslots since the establishment of the TSCH
network. All nodes in the network are aware of the current ASN.
To define a TSCH scheduler, for each radio link a collection of
timeslots and channel offsets is assigned, called its cells. A channel
offset is a “virtual channel” that is translated into a physical radio
channel that is going to be employed for communication. The
translation is carried out by a FHSS algorithm:
frequency =F(ASN+channelOffset)%nFreq (1)
where
nFreq
is the number of available physical channels (e.g.,
16 when using IEEE802.15.4-compliant radios at
2.4
GHz with all
channels in use) [
12
].
F
is a look-up table function that translates
the result from the operation to actual radio channel (i.e., from 11th
to 26th in 2.4GHz band). In Fig. 1, a TSCH schedule is depicted.
B. RPL
RPL [
9
] is a distance-vector routing protocol specifically
designed to manage hundreds of nodes in LLNs. At its core, RPL
constructs a DODAG, i.e., a directed graph with no cycles. To
this aim, RPL assigns to each device participating in the routing a
rank, i.e., a metric which denotes the virtual distance from the root.
Note that the rank is dynamically computed and constantly updated
throughout the network lifetime. Different Objective Function (OF)
can be defined to compute a node’s rank based on different types of
metrics (hop count, link quality, expected retransmission times, etc.).
RPL initiates the DODAG construction from the root. It
periodically broadcasts a DODAG Information Object (DIO), i.e.,
a control packet that includes certain configuration information as
well as root’s rank value. Nodes may discover their neighbors and
their respective ranks upon the reception of such messages. Note
that the frequency of DIO packets heavily depends on the network
stability, i.e., the more stable is the network the less frequently a
DIO packet is transmitted. When a node receives a DIO message,
it computes a new rank based on a given OF, and compares it with
its current rank. Then, if the newly calculated rank is smaller than
the current, it will add the transmitter’s address in its set of potential
parents. From this set of parents, a node chooses its preferred parent
as the node from which its rank will be minimal. If the new rank
is bigger than the current rank, it is ignored. Once the network
stabilizes, a node ends up with a preferred parent, a list of possible
parents, children and siblings. Then, a node may push its packets
toward the root through its preferred parent. Note that once a node
computed its own rank, based on the rank of its preferred parent
and link quality, it periodically transmits its own DIO packets.
C. Motivation: Toward Deterministic Industrial Networks
A deterministic network ensures that data packets traverse the
RPL network in a bounded window of time. It also guarantees that a
periodic process will be repeated identically throughout the network
lifetime. It particularly means that potential congestions or external
interferences must not affect the predictable and deterministic
behavior of the network.
As previously detailed, IEEE 802.15.4-2015 comes with resource
reservation i.e., transmissions are scheduled. However, wireless links
are heavily affected by external interference and noise. Therefore,
wireless communication comes with retransmission schemes, but
at a cost of energy consumption, delay (in best-effort traffic) and
bandwidth, since additional timeslots are required. In TSCH, if
a data packet transmission fails, the transmitter will retransmit in
the following slotframe, e.g., after 101 timeslots in our example,
which is more than a second. In the case of a node crash, failures or
over-the-air-programming, the link quality between two nodes will
drastically decrease (or eliminate the link) for some time, which
will essentially increase the losses in the network. In such a scenario,
retransmission-based schemes will not allow the packet to pass
through this link. To overcome this issue, RPL comes with a failure
solution wherein a child node will select another parent. However,
the time needed for failure detection and new parent selection is
large and during this time, all data packets will be discarded.
To address this limitation, this article presents a novel technique
which takes advantage of path diversity and data duplication to
combat the potential losses and to minimize the delay and jitter in
wireless networks. It demonstrates that determinism can be ensured
by using multiple parallel paths instead of retransmissions over the
default DODAG path.
D. System model and assumptions
The context for LFC is an IEEE 802.15.4-2015 TSCH network
running the RPL protocol for routing. The focus is on the transport
of information from internal network nodes to the root node of
the DODAG. The root node may be seen as a gateway to another
network, possibly wired, and incoming packets can be sent to
external destinations for further processing, but this is out of scope
for this work.
The information is in the form of UDP packet payloads.
Therefore, at the transport layer the sending node does not expect
or receive a delivery confirmation for the packets it sends, and thus
no feedback loop exists at this layer. We are agnostic to the actual
payload information transferred, which is considered an application
layer concern.
The intended use is networks with at least 2 hops, since in 1-hop
networks simple retransmission provides the same performance.
We consider only upwards traffic (i.e., towards the DODAG
root), which is typical of critical monitoring or alert information
transmission. Given this use case, we mainly focus on reliability
rather than throughput, so given the network overhead introduced
for achieving high reliability, very high bandwidth applications are
not an appropriate use-case. In our analysis and simulations, we
use symmetric wireless network links with a static (e.g.,
80%
) or
variable (uniform
70%100%
) error probability, and in both cases
packet-loss follows a uniform distribution with the given probability.
We rely on the default IEEE 802.15.4 TSCH retransmission scheme
when packet losses occur; each hop acknowledges a data packet,
and if the acknowledgement is not received, the sending node
performs a retransmission. In our analysis and experiments, we use
different maximum numbers of retransmissions. The network nodes
themselves are assumed not to malfunction or operate maliciously
and their internal clocks are assumed to be synchronized by IEEE
802.15.4-2015 TSCH (i.e., beacon packets and data packets with
timing information to help correct drift as well as guard time in
each time slot to compensate for uncorrected drift [13], [14], [15]).
III. LEAPFROG COLLABORATION DE SIGN
In this Section, the LeapFrog Collaboration (LFC), a cross-layer
(MAC and Routing) scheme that minimizes the delay and the jitter
metrics is detailed.
In a nutshell, LFC computes two parallel or interleaved paths
for one track, with promiscuous listening between them, to allow
the nodes on one path to overhear transmissions along the other
path. Each node participating in a track selects a preferred or default
parent and a so-called alternative parent. For every data packet
transmission within this track, the data packet is sent twice, one copy
to the default parent and another copy to the alternative parent. For
instance, in Fig. 2 a typical ladder-based topology of two parallel
paths,
86421
and
87531
, is depicted.
In such a scenario, using packet replication and elimination would
allow transferring a copy of the packet along one or both of these
paths, in a ship-by-night fashion.
1
2
3
4
8
Replication
Elimination
Fig. 2:
LFC: Red arrows represent the RPL DODAG tree whereas blue
ones the alternative paths. Packet Replication: 8 transmits twice the same
packet, to its DP 6 and to its AP 7. Packet Elimination: 5 discards the
packet from 7, since it received it earlier from 6.
The LFC algorithm can be designed and implemented in either a
centralized or a distributed scheduling fashion. Under the centralized
design, a central entity (e.g., the root) computes the routes and sched-
ules the communication among the nodes in the network, similar
to a label-switched path. Alternatively, in the distributed approach,
each node constructs its path to the root, typically by employing a
source routing header. In this article, the later case is investigated.
A. Alternative Parent Selection
As previously mentioned, when running the RPL protocol, each
node maintains a list of potential parents. LFC defines the preferred
parent through the RPL DODAG as the Default Parent (DP), as
shown in Fig. 2. To construct an alternative path toward the root, in
addition to the DP, each node in the network registers an Alternative
Parent (AP) as well. There are multiple potential AP selection meth-
ods, but this article presents a method which allows the two paths
to remain correlated: a node will select an alternative parent close
to its default parent in order to allow overhearing between parents.
Thus, to choose an AP, a node will select another parent from its list
of parents that has a common parent with the node DP. So the node
will check if its Default Grand Parent (DGP), the DP of its DP, is in
the set of parents of a potential AP. If several potential APs follow
this condition, the AP with the lowest rank will be selected.
B. LFC Operations
1) Packet Replication: To provide determinism in a wireless
industrial network, LFC guarantees predictability in every level of
the forwarding path. To that aim, under LFC, each node transmits
(i.e., replicates) each data packet to its DP and its AP, respectively,
in unicast transmission mode. Note that this procedure takes place
at each level of the DODAG, between all relay nodes. In Fig. 2, the
replication operation is illustrated, where node 6 is transmitting the
data packet to both parents, nodes 4 (i.e., DP) and 5 (i.e., AP), in
two different timeslots within the same TSCH slotframe. As a result,
given the ladder-based topology, each data packet may traverse the
wireless network over parallel paths. However, if multiple parents are
not available, and thus no AP as well, LFC falls back to the normal
RPL+TSCH operation, i.e., forwarding data to the preferred parent
only, at the specific node where only one parent is available. There-
fore, performance will be identical to the default RPL+TSCH case
for that subset of the DODAG where only one parent is available.
2) Packet Elimination: By using the replication operation, it
follows that a node may receive several copies of the same data
packet, which increases the traffic load and, thus, may impact the
network congestion and the impact on battery lifetime [
16
]. In
order to avoid such a scenario, the Packet Elimination operation
is introduced. Once a node receives the first copy of a data packet,
it will discard the following received copies, as shown in Fig. 2. To
do so, a sequence number is attached in each data packet to identify
the duplicates. Note that the packet elimination operation is applied
at each DODAG level.
C. LFC Features
1) Promiscuous Overhearing: LFC exploits the shared
properties of the wireless medium to compensate for the potential
loss that is incurred with radio communication. Considering that the
wireless medium is broadcast by nature, then any neighboring node
may listen to (i.e., overhear) a transmission if it is in the range of the
sender, often called Promiscuous Overhearing. Thus, a given relay
node may have more opportunities to receive a given data packet by
listening to unicast transmissions towards other relaying nodes. This
is a frequent occurrence in LFC since packet replication is also used.
Furthermore, in the case where only one parent is available for a
node, as with Packet Replication, the behaviour will fall back to the
normal TSCH+RPL case of performing no overhearing and thus
being no worse than the default case. This fall back will be limited
to the subset of the DODAG where only single parents are available.
Since parents with a common ancestor are selected, a grandparent
will have several opportunities to receive a given data packet by
promiscuous overhearing. For instance, as shown in Fig. 3a, when
the intermediate node 6 is transmitting to its DP (i.e., node 4),
the AP (i.e., node 5) may receive this data packet as well, and
vice versa in Fig. 3b. Thus, each parent (the DP and the AP) has
twice the chances to receive a data packet for each transmission
from each child: the original transmission directed to it and an
overhearing from the child’s transmission to its other parent. Finally,
the probability of successful transmission from one DODAG level
to its upper one can be enhanced, by considering the overhearing
feature of the sibling nodes, i.e., the nodes which have the same
parent as the transmitting node. For example, the transmission from
node 6 to its DP 4, can be overheard not only by its AP, but also
by its sibling as well, node 7. As a result, promiscuous overhearing
not only improves network reliability, but may also decreases delay
and jitter as long as the transmission opportunities from one level
to another are grouped in the slotframe.
2) Over-provisioning: Using the previously described
mechanisms (packet replication to an alternative parent and
promiscuous overhearing) together results in multiple opportunities
for a packet to be transmitted from one layer to the next. For instance,
in Fig. 3, node 4 has up to four opportunities to receive the data
packet: twice from node 6 (i.e., i) through direct unicast transmission:
4 is the DP of 6, ii) through overhearing the transmission from 6
(a) Unicast to DP (b) Unicast to AP
Fig. 3:
By employing the Overhearing operation, the DP, the AP and the
sibling nodes have more opportunities to receive the same data packet.
TABLE I: Reception opportunities
Nodes
/ Level
Parents /
Node
Overhearing Over-provisioning Total
2 2 2 2 24=16
1 2 2 2 23=8
2 1 2 2 23=8
to its AP node 5) and, similarly, it will receive twice from node 7.
As a result, considering the ladder-based topology in Fig. 3, there
are up to eight opportunities for a data packet to be received by the
upper level of DODAG, i.e., four opportunities per node.
To further improve reliability, the concept of Over-provisioning is
introduced, which can be summarized as conditional retransmission.
More specifically, each transmitter after sending a data packet will
wait for an acknowledgement. In case of unsuccessful transmission,
thanks to the over-provisioning algorithm, the transmitter will
have another opportunity to transmit its data packet within the
same slotframe, as shown in Fig. 4. It is worth mentioning that the
over-provisioning timeslots are scheduled right after the original
transmissions to bound the delay and, consequently, the jitter
performance. For example, in timeslot 0 node 8 transmits unicast
to node 7 while node 6 overhears, and in timeslot 1 these events are
repeated in case a retransmission is needed. Thus, by introducing
over-provisioning timeslots in the schedule, opportunities of
successful packet reception is doubled.
3) Summary: Taking all the mechanisms used into account and
used concurrently, the opportunities for reception belong to two
groups: i) if both the sending and the receiving levels have two
nodes, then there are a total of
16
reception opportunities, otherwise,
ii) for the levels which have either only one parent or only one
sending node, there are a total of
8
reception opportunities. These
cases are summarised in Table I.
1
2
3
4
5
6
7
8
0 1 2 3 4 5 ... ... 54 55 56
... 72 73 74 75 76 77 78 ... 100 57 58 59 ...
...
Timeslot
Unicast Tx
DP/AP Overhearing
S Overhearing
Unicast Rx
LFB/DIO Tx
EB Tx
Idle
LFB/DIO/EB Rx
Fig. 4:
An example of the adaptive LFC scheduler for eight nodes according to the topology in Fig. 3. It illustrates the following operations: Replication,
Promiscuous Overhearing, Over-provisioning, as well as the Enhanced Beacons, LeapFrog Beacon and DIO control packets.
IV. LFC SCHEDULER
A. Topology-adapted Scheduler
Next, LFC comes with a scheduler adapted to the topology
to achieve minimum and constant delay performance. To do so,
the transmissions from the nodes far from the DODAG root are
configured to be transmitted first. More precisely, the leaf nodes are
configured to transmit first, and then relay nodes according to their
distance from the root. As a result, all nodes access the wireless
medium sequentially one after the other one and, therefore, the data
packet may arrive at the root within one single slotframe, according
to the requirements of deterministic networks in Industry 4.0 of
ultra-low jitter performance. The schedule is statically defined and
precalculated offline for the given topology to be able to guarantee
that the TSCH schedule will not externally affect the measurement
of the metrics of the examined methods. It is also noteworthy
that although in our description we use a schedule which contains
all the cells which are required for the transmission of the data
from the source to the root, nevertheless it is possible to use a
smaller schedule if multi-slotframe delays can be tolerated by the
application requirements. This is also a solution for networks which
are so large so as to preclude the use of only one slotframe.
In Fig. 4, a scheduler with a single channel offset is illustrated,
and it is adapted to eight nodes according to the topology in
Fig. 3. As it can be observed, the leaf node 8 transmits first its data
packet toward its DP, its RPL parent, while an additional timeslot
is assigned for retransmission to the same destination in case of
failed transmission. Then, another two timeslots are assigned to the
leaf node to transmit to its AP (i.e., timeslots
2
and
3
). Note that
both parents are overhearing the others transmissions (see yellow
boxes). Furthermore, it is worth mentioning that timeslots
4
to
11
are assigned to forward the data packet from level 3 to level 2 of
DODAG tree, which is equal to
8
transmissions and
16
opportunities
for a single data packet to reach the upper level of DODAG.
B. LeapFrog Beacon Control Packet
During the network initialization phase, nodes receive all
necessary RPL information through DIO messages. Thus, the
nodes know their set of parents and, consequently they decide
their preferred parent. Similarly, LFC uses control packets called
LeapFrog Beacons (LFB). These LFB are periodically transmitted
over the network in broadcast every
30s
. In the beginning, the LFB
contains RPL-related information i.e., a list of possible parents and
the preferred parent. Based on this information, nodes eventually
learn their DGP and, thus, they can define their AP, siblings and other
overheard neighborhood nodes. Later, this additional information
is included in the LFB and consequently propagated. Hence, in
case a node changes its DP, AP or possible parents, the other nodes
can learn about this change thanks to the LFBs and, consequently,
recalculate their own AP. Finally, if a change was necessary, nodes
will end up informing this to their neighbors with the upcoming
LFBs. Thus, the update process travels across the network and the
alternative paths can be kept dynamically available all the time.
V. ANALYSIS
The first part of this section presents the delay-jitter performance
trade-off of TSCH. The goal is to provide insight to the reader
on particular design decisions of LFC. The second part of this
section provides analytical expressions for the calculation of several
performance metrics of LFC, including the PDR, and upper limits
on the end-to-end delay and jitter.
A. The delay-jitter trade-off
In addition to promiscuous listening and exploiting multiple paths,
LFC incorporates an extra timeslot for one retransmission in each
link. Combining all these mechanisms, a particular packet has at
least
8
unique opportunities to progress to the next rank of the
network within the same slotframe. Fundamentally, the number of
unique opportunities a packets has to progress to the next rank is con-
trolled by a performance trade-off. Indeed, increasing this number
is beneficial for the end-to-end jitter, as it increases the probabilities
that a packet will reach the destination within a single frame. How-
ever, the more the allocated slots for retransmissions, the longer the
frame is required to be. As a result, the allocation of additional times-
lots for retransmissions within the same slotframe has a negative
impact on the delay. This is the delay-jitter performance trade-off.
Let us consider a TSCH neighborhood with one receiver (
r
) and
N
senders:
n1,n2, ..., nN
. The number of consecutive timeslots
per node per slotframe is denoted as
k
. For example,
k= 1
corresponds to a frame size of
N
timeslots with one timeslot
per node (
n1r,n2r,...
). Similarly,
k= 2
corresponds to a
frame size of
2N
slots with two consecutive timeslots per node
(
n1r,n1r,n2r,n2r,...
). Each transmission is modeled as
a independent Bernoulli trial with the same probability of success,
p
(i.e. PRR), for each trial. For simplicity, let us assume infinite
retransmissions. The probability that a packet will be successfully
delivered after ifailed attempts is given by:
qi=(1p)ip , (2)
Without loss of generality, let us focus on the
N
-th node and let
us assume that the packet is generated at beginning of the frame.
This assumption constitutes a worst case scenario for the delay. If
the transmission occurs on the
i
-th attempt, the delay is given by
the following equation:
di=kNji
kk+imod k+k(N1) .(3)
where
k(N1)
corresponds to the delay due to timeslots allocated
to the other
N1
nodes in the first slotframe,
kNji
kk
corresponds
to the delay of whole slotframes (when
ik
), and
imod k
corresponds to the delay due to failed attempts in the last slotframe.
The average delay is calculated as the probability-based weighted
sum of di, and it is given by:
D=
X
i=0
qidi,(4)
and the jitter, i.e., the deviation from perfect periodicity in packet
delivery, be modeled as the standard deviation of the delay:
J=v
u
u
t
X
i=0
qi(diD)2.(5)
Fig. 5 demonstrates the delay-jitter trade-off for
N= 4
and for
various link qualities. It can be observed that as the number of
timeslots per node per slotframe increases, the jitter decreases,
yet with diminishing improvements. However, the delay increases
rapidly in an almost linear fashion.
LFC balances the delay-jitter trade-off by allocating up to
n2×m
timeslots for transmitting a given packet in the rank above, where
n=2
corresponds to the number of parents and
m=2
corresponds
to the maximum number of transmission to a single parent.
B. LFC performance in the worst case scenario
This section models the performance of LFC in the worst case
scenario, providing analytical expressions for calculating the end-
to-end PDR, as well as the maximum end-to-end delay and jitter.
Let us consider that a packet is generated in the source node
at the beginning of the frame in a TSCH network of
R
ranks. The
number of parents (including any alternative parents) is denoted
by
n
. The maximum number of transmissions to a single parent
within the same slotframe is denoted by
m
. From end to end, three
cases can be identified. The first type corresponds to the source
node,
i
, transmitting to its parents, which are located in rank
R1
.
Allocated timeslots per slotframe
1 2 3 4 5 6 7 8
Delay (slots)
0
10
20
30
p = 0.5
p = 0.7
p = 0.9
Allocated timeslots per slotframe
1 2 3 4 5 6 7 8
Jitter (slots)
0
2
4
6
Fig. 5:
An illustration of the delay-jitter trade-off in a star TSCH
neighborhood with
4
nodes. Increasing the allocated timeslots per slotframe
decreases the jitter, yet increases the delay.
The probability for the packet to fail to reach a parent,
j
, in rank
R1, denoted by qR1, is given by:
q(j)
R1=pmn
ij ,(6)
whereby
pij
is the link-layer packet error probability for node
i
transmitting to parent
j
. It can be observed that Eq. 6 captures the
fact that a packet has
n×m
opportunities to reach the next rank,
as dictated by the LFC schedule, due to the feature of alternative
parents and retransmissions.
The probability for the packet to fail to progress to an intermediate
rank
h[1,R2]
depends on if it managed to reach the previous
rank:
q(j)
h=
n
Y
i=1q(i)
h+1+(1q(i)
h+1)·pmn
ij .(7)
In the intermediate ranks, a packet has additional opportunities to
reach the next rank due to the feature of promiscuous overhearing,
captured in Eq. 7 by the product.
Lastly, the third case corresponds to the last hop to the root. The
probability for the packet to fail to reach the root, of rank
0
, is given
by:
q0=
n
Y
i=1q(i)
1+(1q(i)
1)·pm
ij .(8)
In this case, promiscuous overhearing is still possible, yet there are
no alternative parents.
Assuming that the packet is not retransmitted in a future frame,
the end-to-end PDR is calculated recursively by:
1q0
. Note that
the Eq. 7 and Eq. 8 does not take into account the probability of
sibling overhearing; thus, the model provides a lower bound of the
end-to-end PDR.
2 8 14 20
Network depth, ranks
0
200
400
600
Delay (slots)
2 parents
3 parents
4 parents
2 8 14 20
Network depth ranks
2
4
6
8
10
Jitter (slots)
Fig. 6:
The worst-case scenario performance of LFC in networks of various
depths (
R
) and widths in terms of parents (
n
). The delay increases linearly
with the network depth and quadratically with the number of alternative
parents. The jitter is independent to the network depth and increases
linearly with the number of parents.
Given the packet is successful at reaching the root within a
slotframe, the end-to-end delay in the worst case scenario is
provided by the following equation:
Dmax= 2·n·m+(R2)·n2·m , (9)
capturing the
n×m
timeslots reserved for the first (source to first
intermediate rank) and final (last intermediate rank to destination)
hops, as well as the
n×n×m
timeslots reserved for the
R2
intermediate hops. Similarly, the end-to-end jitter in the worst case
scenario is estimated by:
Jmax=n·m1.(10)
Dmax
and
Jmax
are given in timeslots and they must be multiplied
by the duration of the timeslot in order to be converted in time.
Fig. 6 plots
Dmax
and
Jmax
for networks of various depths (in
terms of ranks,
R
), and widths (in terms of number of parents,
n
).
The figure demonstrates how LFC scales with the network size. In
particular, the delay increases linearly with the network depth and
quadratically with the network width. The jitter is independent from
the network depth and increases linearly with the network width.
C. The Capacity Loss Trade-off
The TSCH schedule constitutes a limiting factor for the
maximum traffic that can traverse through the network. For
example, if one timeslot per second is allocated to a particular
link, the maximum supported traffic is one packet per second. This
limiting factor is generally not considered an issue, as long as the
overlaying application generates traffic at a low rate; yet, it can be
a challenge when scheduling high-rate traffic [17].
More specifically, a TSCH schedule must over-allocate timeslots
to account for link-layer retransmissions [
17
]. For example, to
Network depth, ranks
2345678910
Timeslots per slotframe
0
20
40
60
80
LFC
TSCH, ETX=1
TSCH, ETX=2
TSCH, ETX=3
Fig. 7:
The number of timeslots per slotframe required for a packet to
traverse a network of
R
ranks (
n= 2
,
m= 2
). LFC sacrifices throughput
for high reliability, low delay and low jitter.
Network depth, ranks
2345678910
Bandwidth, kbps
0
10
20
30
40
50
LFC
TSCH, ETX=1
TSCH, ETX=2
TSCH, ETX=3
Network depth, ranks
2345678910
Overhead, %
0
20
40
60
80
100
LFC vs TSCH, ETX=1
LFC vs TSCH, ETX=2
LFC vs TSCH, ETX=3
Fig. 8:
The available end-to-end bandwidth for LFC and standard TSCH
(top). The bandwidth overhead of TSCH compared to standard TSCH (bot-
tom). LFC sacrifices throughput for high reliability, low delay and low jitter.
support a link with a worst-case
PRR = 0.5
, the TSCH schedule
should over-allocate timeslots by a factor of
ETX = 1/PRR = 2
.
In general, for a packet to traverse through a network of
R
ranks,
a total number of
R×ETX
timeslots are required. Note that,
in practice, over-allocation should be higher than the expected
transmission count due to error busts and finite queues.
LFC provides additional redundancy. Indeed, for a packet to
traverse through
R
, the total number of required timeslots is given
by the worst-case end-to-end delay, Eq. (9). Fig. 7 illustrates the
number of timeslots required by LFC in contrast to standard TSCH.
As demonstrated in the figure, LFC requires additional timeslots,
imposing a limitation to the maximum achievable throughput.
Therefore, LFC is suitable for industrial applications that can afford
to trade bandwidth for high reliability, low delay and low jitter.
This bandwidth overhead is better illustrated in Fig. 8. TSCH
supports up to
100
timeslots of
10
ms per second; thus, up to
100
packets can be transmitted over a single TSCH link per second.
This corresponds to an available bandwidth of
Bl=100 L, (11)
where
L
is the maximum packet size (
L=127
bytes or
1016
bits in
TSCH). Assuming for simplicity no downlink traffic, the available
end-to-end bandwidth, B, is given by
B=Bl
S,(12)
where
S
is the timeslots per slotframe (see y-axis of Fig. 7). The
available end-to-end bandwidth of LFC and standard TSCH for
various network depths is plotted in Fig. 8 (top), while the overhead
of LFC as compared to standard TSCH is shown in Fig. 8 (bottom).
The figure shows that, under ideal conditions (no link-layer packet
loss,
ETX=1
), LFC introduces a considerable bandwidth overhead
of more than
80%
. In more realistic scenarios (i.e.
ETX = 2
and
ETX = 3
), the bandwidth overhead of LFC is under
40%
and
20%
respectively. Overall, LFC is able to support applications that
require
2
kbps end-to-end bandwidth in TSCH networks with depth
of 7ranks or less.
VI. PE RFOR MANC E EVALUATION
A. Simulation Setup
To evaluate the performance of the LFC algorithm, COOJA,
the network simulator distributed as part of the Contiki OS
1
, was
employed.
The wireless network consists of 8 nodes that are distributed
in a typical ladder-based topology in an area of
130 m×40 m
, as
depicted in Fig. 3. In particular, node 8 is the source, node 1 is the
root, while nodes 2, 3, 4, 5, 6 and 7 are relays. The source node
transmits
1
data packet every
15
seconds, using a
50 m
transmission
radius so that given the ladder topology, exactly a node’s direct
children, parents and its sibling are reachable. The radios are
configured to communicate at
250kbits/s
. While this topology
is relatively simple, it strikes a balance between complexity and
simplicity which has the advantages that:
it can be studied analytically as well (see Section V),
it has enough complexity and structure to highlight the
advantages and disadvantages of our solution,
it is very good predictor of performance in even more random
topologies due to the method used for selecting alternative
parents in LFC (i.e., a nodes DGP must be in the set of parents
of a candidate AP).
At the routing layer, the RPL protocol [
9
] was used to build the
DODAG, while at the MAC layer, IEEE 802.15.4-2015 TSCH was
selected with slotframe length of
101
timeslots and one channel. The
number of timeslots is generally configurable but we selected
101
since it is the default value and it is big enough for our cases. Any
other big enough value would work as well. The duration of each
timeslot is
10ms
, the lowest value 6TiSCH supports. This value
provides the lower bound of the attainable delay for one network
hop. The payload size is configured at
20 bytes
. Each simulation
lasts
41.25 hours
(not including the
25 minutes
setup/initialisation
1http://www.contiki-os.org/start.html
TABLE II: Simulation Setup
Topology Value
Topology Multi-hop, see Fig. 3
Number of nodes 8 (including the root)
Number of sources 1 source
Node spacing 40 m(in average)
Simulation Value
Duration 41.25 hours
Traffic Pattern 1pkt/15 sec
Payload size 20 bytes
Routing model RPL [9]
MAC model TSCH [5]
TSCH Value
EB period 16 sec
LB period 30 sec
Slotframe length 101
Timeslot length 10 ms
phase) within which approximately
9900 packets
are sent, with
5 independent iterations per configuration. LFC was compared
against that of the retransmission-based RPL
+
TSCH (i.e., a single
copy traverses the network), as well as the state-of-the-art solution
LinkPeek [
11
]. Note that the maximum number of link-layer
retransmissions was implemented under various configurations: 0, 2,
4 and 8, namely RT0, RT2, RT4 and RT8 respectively. For example,
RT4 means that IEEE 802.15.4-TSCH configuration will re-transmit
at most four times if no acknowledgment was received at the link-
layer. In Table II the details of the network parameters are presented.
It is noted that in this configuration (
10
ms timeslot length,
R=5
,
n=2
and
m=2
), equations (9) and (10) indicate that the maximum
delay is 240 ms and the maximum jitter is 30 ms respectively.
B. Link environment
In real-world deployments, the radio link quality presents
dynamic behavior over time [
18
]. However, in critical industrial
environments, the deployed wireless network has to combat such
link quality variations to maintain ultra-high network performance.
In this study, the performance of LFC was evaluated under various
radio link qualities. In particular, the following four different
use-cases were studied: static link quality of 90%, 80%, 70%, and
link quality uniformly random between 70% to 100%. In the last
case all link qualities are reset to a randomly generated value in
the interval every
10 minutes
. Additionally, the links between the
sink node and its two children are always kept at
100%
. The reason
for this exception is because we would like to exclude these nodes
from the reliability calculations. These nodes cannot support the
LFC case due to having only one parent (the root node) instead
of two. As a result, a lower link quality than
100%
between these
nodes and the root would unfairly affect LFC in comparison to
the no-overhearing cases. Keeping the link qualities at
100%
does
affect latency, but in the same way for all the compared methods.
C. Simulation Results
In this section, the performance evaluation of LFC in terms of
delay, jitter, reliability and energy consumption is presented when
compared against LinkPeek and default single-path RPL
+
TSCH
with varying re-transmissions at the link-layer.
(a) Mean and standard deviation of end-to-end delay. (b) Mean end-to-end jitter.
(c) Packet Delivery Ratio performance. (d) Radio duty cycle performance for TX, RX, and Idle modes.
Fig. 9:
Performance evaluation of LFC in terms of end-to-end delay, jitter, reliability and radio duty cycle, when compared against single path
retransmission-based approaches of RPL+TSCH, state-of-the-art LinkPeek solution and the theoretical PDR performance of LFC (LFC-Th.An.).
1) Delay: In Fig. 9a the mean and the standard deviation of MAC
layer end-to-end, i.e., from the leaf node to the destination, delay
is depicted. Note that the end-to-end delay includes the propagation
time of the data packet, as well as the potential retransmission
delay. As it can be observed, LFC achieves end-to-end delay close
to
205ms
, which is in line with the proposed scheduler and with
the mathematical analysis. Furthermore, considering Fig.11, LFC
demonstrates a very stable delay performance. When compared to
the retransmission-based approaches of IEEE802.15.4-TSCH (i.e.,
RT2, RT4 and RT8) and Link Peek, LFC displays delay reduced
by up to
577%
,
669%
,
664%
, and
568%
respectively. Note that
the non-retransmitting IEEE802.15.4-TSCH RT0 approach does
achieve very low delay (
204 ms
) but at a very high cost in PDR.
The delay is low because only delivered packets are considered, and
if a packet has been delivered with RT0, by necessity, it will not
have missed any transmissions along its path to delay it.
2) Jitter: To provide deterministic communication in a wireless
network, it is necessary to obtain minimum jitter performance. Note
that jitter is the variation in latency from one packet to another, here
calculated as the standard deviation of end-to-end delay.
In Fig. 9b the average end-to-end jitter performance is illustrated.
As it can be observed, LFC achieves ultra-low jitter (i.e.,
15ms
):
lower than any other LFC or RPL
+
TSCH configuration and
LinkPeek solution. On the other hand, the more retransmissions are
configured for the following slotframe, the higher the jitter, i.e., see
RPL
+
TSCH-RT4, RPL
+
TSCH-RT8. As a result, LFC decreases
jitter by up to
7357%
,
8799%
,
8897%
and
7301%
when compared
to RPL
+
TSCH-RT2, RPL
+
TSCH-RT4, RPL
+
TSCH-RT8 and
LinkPeek, respectively. Finally, note that the delay and jitter metrics
were computed based only on successful packet receptions and
also removing delays more than 3 standard deviations (
3σ
) away
from the mean. This explains the performance of RPL
+
TSCH-RT0
(
14 ms
) that presents high performance of jitter but extremely
Fig. 10: Average radio power consumption on the Zolertia Z1 mote.
Fig. 11:
Detailed representation LFC’s delay performance under various
link qualities.
low PDR performance.
3) PDR: To further assess the performance of LFC, the network
reliability was evaluated. To this aim, the PDR was computed,
where packet loss is calculated as 1P DR and, thus, for example
packet loss 0% is the equivalent of 100% PDR.
In Fig. 9c, the PDR performance under various link qualities
(i.e., link qualities fixed at
90%
,
80%
,
70%
and variable random
between
70% 100%
), is illustrated. LFC presents ultra-high
PDR performance above
99.1%
in all cases and above
99.83%
for cases with links with link quality above
80%
. In fact, LFC
demonstrates results similar to the theoretical performance, i.e.,
dark green column. These results can be explained as follows. As
it can be observed, the additional features of LFC, overhearing and
over-provisioning, fundamentally improve the performance of the
LFC scheme, since by employing the over-provisioning option each
data packet has an additional timeslot as a back up to retransmit
in case of a failure of the original transmission, see Table I. As
a result, LFC provides reliability similar to the theoretical and
RPL
+
TSCH-RT8, while it minimizes the jitter performance as
well, contrary to the retransmission-based solutions.
4) Duty Cycle: Finally, this article evaluates the energy
consumption for all studied schemes.
Fig. 9d illustrates the average network-wide radio duty-cycle (
C
)
separately for the three modes (TX, RX, and Idle) that the radio can
be in. The upper panel in Fig. 9d shows TX (
CTX
) and RX (
CRX
)
stacked, while the lower panel shows only the Idle mode (
CIdle
).
For TX and RX the maximum duty duty cycle is
0.27%
, while
for the Idle mode the maximum is
8%
. The majority of the duty
cycle corresponds to the Idle mode, while TX and RX have very
low and similar duty cycles.
The evaluation results utilise the fact that the majority of the
energy consumption in IoT devices results from the operation of the
radio module [
19
]. This leads providing duty cycle results, which are
easily transferable to different hardware platforms, in lieu of power
consumption values. As an example of this process, we provide the
conversion from relative duty cycle to absolute power consumption
for the Zolertia Z1 mote which uses the CC2420 radio transceiver
module. Given the values for radio power consumption for the
Z1 [
20
] (
PTX = 52.2mW@3V
,
PRX = 56.4mW@3V
,
PIdle =
1.28mW@3V
) we receive the average radio power consumption
per mote as:
E(P)= PT X CT X +PRX CRX +PIdleCIdle (13)
The results for the Z1 mote using Eq. 13 are shown in Fig. 10 for
each transmission method. The results indicate that LFC induces
a straightforward impact on duty cycle performance, since LFC
allocates additional timeslots per scheduler transmission to guarantee
reliability above
99%
and bounded delay performance (by using
the over-provisioning feature). Moreover, by using the overhearing
option, the nodes remain active for more timeslots to listen to
neighboring transmissions, as compared to the default RPL
+
TSCH
operation. Therefore, LFC consumes more energy network-wide
when compared to LinkPeek and retransmission-based RPL
+
TSCH.
As shown in Fig. 10, for the Z1 mote the overhead that LFC adds is
195%
(
E(PLFC )0.25mW
,
E(PRPL+TS CH )0.085mW
).
For other hardware platforms, the overhead introduced may differ
since the reference energy consumption of the radio modes might
be different from the Z1 mote, but our results facilitate calculating
the overheads given the corresponding energy consumption values.
These results show that LFC provides determinism by bounding
delay at the cost of energy consumption.
It is noted that one of the ongoing works on LFC is to reduce
the unnecessary active timeslots by introducing a more intelligent
scheduler. For instance, in the over-provisioning feature, it may
not be necessary to explicitly allocate one additional timeslot per
transmission, instead it could be configured probabilistically based
on link quality, meaning that the more stable the link from A to B,
the lower the probability to allocate an additional timeslot.
D. Feasibility for industrial applications
LFC has been implemented on Contiki OS with the constraints
of low-power motes taken into consideration. The simulations
for network and energy performance were executed using the
COOJA target to allow examining a large set of parameters and
to allow multiple iterations to increase statistical confidence in the
results. However, to verify the feasibility of LFC for real industrial
applications, we have used the Zolertia Z1 mote.
The Z1 mote has been selected because it is one of the most
resource constrained motes available (
16
-bit RISC CPU
@16
MHz,
8
KB RAM,
92
KB internal flash memory,
2
MB external flash
memory, CC2420 radio transceiver module) which can still execute
in IPv6 mode to support 6TiSCH. It also has the added benefit that
it is possible to simulate the Z1 mote using the MSPSim software
package [
21
] within COOJA, thus we have accurately verified that
the operation is faithful to the scenarios examined in this work. We
are confident that since LFC using Contiki OS is operational on the
very constrained Z1 mote, then it will be applicable to almost any
other equivalent or more powerful hardware platform (e.g. CC2538,
CC2650).
More specifically, we have executed sample simulations to verify
that:
1)
LFC on Contiki OS compiles for the Z1 and fits within its
available external flash memory (
2
MB). More specifically,
RPL+TSCH compiles to
495960
bytes, while the LFC
implementation to
499600
bytes, so the overhead added is
3640 bytes or 0.73%.
2)
LFC on Contiki OS compiles for and executes normally
on the Z1, fitting within its available RAM (
8
KB). The
exact global RAM overhead is
99
bytes or
1.2%
of RAM,
augmented slightly by minimal stack use during function
execution. Out of the
99
bytes,
40
bytes are arrays of
configurable size and can be set to less to reduce memory
usage. LFC does not use any dynamically allocated memory.
3)
LFC on Contiki OS executes normally (no missed scheduled
events or other timing issues) through the MSPSim simulator
for the Z1. Thus, the computational overheads are acceptable.
In practice, our implementation mainly adds functionality
above the MAC layer, where timing is not so critical. It does
modify the operation of the MAC packet reception handler,
however our additions are executed after the packet has been
received and not during the critical timeslot operation. The
only modification inside the time critical timeslot operation
handler is to enable and disable MAC address filtering when
overhearing is required, which has no measurable impact on
performance since it is just a configuration option change.
VII. RELATED WORK
Phinney et al. [
7
] investigates the applicability of RPL for
industrial applications and shows that RPL provides the baseline
requirements, but QoS still needs to be addressed. Thus, it highlights
the need for higher reliability and predictability. Attempting to
provide wire-like reliability in IIoT by utilizing the Automatic
Repeat reQuest (ARQ) method, commonly employed in general-
purpose wireless networks, leads to simple and easily implemented
solutions. However, while the value of simplicity and ease of
implementation is considerable, the drawbacks of approaches of
this nature are especially problematic for IIoT which have strict
constraints, i.e., increased delays and jitter metrics. Additionally,
they decrease spectrum utilization by reserving part of it for the
retransmissions.
In attempting to reduce the energy costs of reliability-enhancing
methods, overhearing has been used to implicitly acknowledge the
packet reception. In [
22
] Lee et al. use this approach when the
receiver in turn forwards the packet to provide an acknowledgement
of reception to the original sender. An enhanced version of this work
by Maalel et al. [
23
] adds spatial diversity by maintaining an ordered
list of neighboring nodes susceptible of retransmitting the packet.
Additionally, standardized methods of Packet Replication and
Elimination (PRE) targeting mission-critical and time-sensitive
applications have been created by the Time-Sensitive Networking
(TSN) Task Group (TG) at the IEEE the Deterministic Networking
(DetNet) Working Group (WG) at the IETF. Both are working on
redundancy-based methods wherein packet copies can follow non-
congruent paths through the network while ensuring single delivery
to the receiving end, even when one of the paths is interrupted [
24
].
In [
11
] the authors propose an algorithm to increase reliability in
IEEE 802.15.4 networks, by conditionally retransmitting packets to
fallback RPL parents when link failure is detected. Their solution,
however, does not aim at lowering end-to-end delay and by only
retransmitting when failure is detected misses an opportunity
for path diversification in cases where a node is completely
disconnected from the network.
In [
25
] the authors propose using data plane packet replication
(up to the network graph degree) at the originating nodes only and
routing the additional packet copies via disjoint paths, in order
to improve reliability and latency, at the cost of increased energy
consumption. They use a centralized scheduler to generate the
TSCH schedule and a custom simulation for the performance
evaluation, showing that packet replication significantly helps in
achieving wire-like reliability.
In [
26
], the authors propose the adaption of the IEEE 802.15.4 in
order to facilitate its cooperation with the RPL protocol to achieve
multipath routing. The adaptation is required due to a limitation
in the IEEE 802.15.4, which allows only one associated parent at
any time. Their work results in slightly improved end-to-end delay
and reliability. However, their work does not take advantage of the
multiple paths to replicate traffic in order to increase reliability.
In [
27
], the authors propose a distributed scheduling algorithm
for IEEE 802.15.4e-TSCH networks, which uses RPL ranks to
group nodes into layers which can be scheduled in overlapping
cells. Their aim is to use this to increase network capacity by taking
spatial separation into account and to create schedules which feature
an end-to-end upper bound to delay, specifically delivery within
the same slotframe.
VIII. CONCL USIO NS
Industrial networks require deterministic protocols to guarantee
that the transmitted data packet will traverse the wireless network in
a predefined and constant delay. This article proposes the LeapFrog
Collaboration mechanism to exploit spatial diversity and packet
redundancy to compensate for the inherently lossy wireless medium.
At its core, LFC computes two parallel paths for a single data flow,
thus, the nodes on one path may listen-in on the data transmissions
along the other parallel path. As a result, each data packet gets
multiple opportunities to be received at the upper DODAG level. The
performance evaluation results demonstrate that LFC achieves net-
work reliability above
99%
, while bounding the delay performance,
i.e., providing an ultra-low jitter performance of
15ms
. The ongoing
work consists of further investigating a more sophisticated scheduler
to reduce the unnecessarily active timeslots and, thus, to decrease
the energy consumption. Furthermore, it is planned to investigate the
performance of LFC in large-scale wireless networks. Finally, it is
planned to investigate the behavior of LFC under realistic conditions
by performing a set of experimental studies.
REFERENCES
[1]
E. Grossman, C. Gunther, P. Thubert, P. Wetterwald, J. Raymond, J. Korhonen,
Y. Kaneko, S. Das, Y. Zha, B. Varga, J. Farkas, F.-J. Goetz, J. Schmitt,
X. Vilajosana, T. Mahmoodi, S. Spirou, P. Vizarreta, D. Huang, X. Geng,
D. Dujovne, and M. Seewald, “Deterministic networking use cases,
Working Draft, IETF Secretariat, Internet-Draft draft-ietf-detnet-use-cases-13,
September 2017, https://tools.ietf.org/id/draft-ietf-detnet-use-cases-11.txt.
[Online]. Available: https://tools.ietf.org/html/draft-ietf-detnet-use-cases-13
[2]
L. D. Xu, W. He, and S. Li, “Internet of things in industries: A survey,IEEE
Transactions on Industrial Informatics, vol. 10, no. 4, pp. 2233–2243, Nov
2014.
[3]
S. Yamamoto, T. Emori, and K. Takai, “Field wireless solution based on isa100.
11a to innovate instrumentation,” Yokogawa Technical Report English Edition,
Tech. Rep., 2010.
[4]
P. Thubert, “Converging over deterministic networks for an industrial internet,
Ph.D. dissertation, 2017, th
`
ese de doctorat dirig
´
ee par Toutain, Laurent Infor-
matique Ecole nationale sup
´
erieure Mines-T
´
el
´
ecom Atlantique Bretagne Pays
de la Loire 2017. [Online]. Available: http://www.theses.fr/2017IMTA0011
[5]
“IEEE Standard for Low-RateWireless Personal AreaNetworks (LR-WPANs),
IEEE Std 802.15.4-2015 (Revision of IEEE Std 802.15.4-2011), April 2016.
[6]
G. Z. Papadopoulos, A. Gallais, G. Schreiner, and T. Noel, “Importance of
Repeatable Setups for Reproducible Experimental Results in IoT,” in Proc.
of the ACM PE-WASUN, 2016.
[7]
T. Phinney, P. Thubert, and
RA.
Assimiti, “RPL applicability in industrial
networks,” Working Draft, IETF Secretariat, Internet-Draft draft-ietf-roll-rpl-
industrial-applicability-02, October 2013.
[8]
K. Pister, P. Thubert, S. Dwars, and T. Phinney, “Industrial Routing
Requirements in Low-Power and Lossy Networks,” IETF, RFC 5673, 2009.
[9]
T. Winter, P. Thubert, A. Brandt, J. Hui, R. Kelsey, P. Levis, K. Pister, R. Struik,
J. Vasseur, and A. R., “RPL: IPv6 Routing Protocol for Low-Power and Lossy
Networks,” IETF, RFC 6550, 2012.
[10]
G. Z. Papadopoulos, T. Matsui, P. Thubert, G. Texier, T. Watteyne,
and N. Montavont, “Leapfrog Collaboration: Toward Determinism and
Predictability in Industrial-IoT applications,” in Proc. IEEE ICC, 2017.
[11]
Y. S. Lohith, T. S. Narasimman, S. V. R. Anand, and M. Hedge, “Link peek:
A link outage resilient ip packet forwarding mechanism for 6lowpan/rpl based
low-power and lossy networks (llns),” in 2015 IEEE International Conference
on Mobile Services, June 2015, pp. 65–72.
[12]
T. Watteyne, M. Palattella, and L. Grieco, “Using IEEE 802.15.4e Time-Slotted
Channel Hopping (TSCH) in the Internet of Things (IoT): Problem Statement,”
RFC 7554, 2015.
[13]
D. Stanislowski, X. Vilajosana, Q. Wang, T. Watteyne, and K. S. J. Pister,
Adaptive synchronization in ieee802.15.4e networks,IEEE Transactions on
Industrial Informatics, vol. 10, no. 1, pp. 795–802, Feb 2014.
[14]
T. Chang, T. Watteyne, K. Pister, and Q. Wang, “Adaptive
synchronization in multi-hop tsch networks,” Computer Networks,
vol. 76, pp. 165 – 176, 2015. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S1389128614003922
[15]
G. Z. Papadopoulos, A. Mavromatis, X. Fafoutis, N. Montavont, R. Piechocki,
T. Tryfonas, and G. Oikonomou, “Guard Time Optimisation and Adaptation
for Energy Efficient Multi-hop TSCH Networks,” in Proceedings of the IEEE
3rd World Forum on Internet of Things (WF-IoT), 2016, pp. 301–306.
[16]
G. Z. Papadopoulos, J. Beaudaux, A. Gallais, P. Chatzimisios, and T. Noel,
“Toward a Packet Duplication Control for Opportunistic Routing in WSNs,
in Proc. of the IEEE GLOBECOM, 2014, pp. 94–99.
[17]
A. Elsts, X. Fafoutis, J. Pope, G. Oikonomou, R. Piechocki, and I. Craddock,
“Scheduling High-Rate Unpredictable Traffic in IEEE 802.15.4 TSCH
Networks,” in 13th Int. Conf. on Distributed Comput. in Sensor Syst. (DCOSS),
2017.
[18]
G. Z. Papadopoulos, A. Gallais, G. Schreiner, E. Jou, and T. Noel,
“Thorough IoT testbed Characterization: from Proof-of-concept to Repeatable
Experimentations,” Elsevier Computer Networks, vol. 119, pp. 86–101, 2017.
[19]
J. Eriksson, F.
¨
Osterlind, N. Finne, A. Dunkels, N. Tsiftes, and T. Voigt,
Accurate network-scale power profiling for sensor network simulators,” in
Proceedings of the 6th European Conference on Wireless Sensor Networks
(EWSN), 2009, pp. 312–326.
[20] Zolertia, Z1 Datasheet, March 2010.
[21]
J. Eriksson, A. Dunkels, N. Finne, F.
¨
Osterlind, and T. Voigt, “MSPSim–an
extensible simulator for msp430-equipped sensor boards,” in Proceedings of
the European Conference on Wireless Sensor Networks (EWSN), Poster/Demo
session, vol. 118, 2007.
[22]
G.-W. Lee and E.-N. Huh, “Reliable data transfer using overhearing for
implicit ack,” in ICCAS-SICE, 2009. IEEE, 2009, pp. 1976–1979.
[23]
N. Maalel, P. Roux, M. Kellil, and A. Bouabdallah, “Adaptive Reliable Routing
Protocol for Wireless Sensor Networks,” in Proc. of the ICWMC, 2013, pp.
52–55.
[24]
G. Z. Papadopoulos, N. Montavont, and P. Thubert, “Exploiting packet repli-
cation and elimination in complex tracks in 6tisch llns,” Working Draft, IETF
Secretariat, Internet-Draft draft-papadopoulos-6tisch-pre-reqs-00, July 2017.
[25]
J. de Armas, P. Tuset, T. Chang, F. Adelantado, T. Watteyne, and X. Vilajosana,
“Determinism Through Path Diversity: Why Packet Replication Makes Sense,”
in Proc. of the INCoS, 2016.
[26]
B. Pavkovi
´
c, F. Theoleyre, and A. Duda, “Multipath opportunistic rpl routing
over ieee 802.15.4,” in Proc. ACM MSWiM, 2011, pp. 179–186.
[27]
I. Hosni and F. Th
´
eoleyre, “Self-healing distributed scheduling for end-to-end
delay optimization in multihop wireless networks with 6tisch,” Computer
Communications, vol. 110, pp. 103 – 119, 2017.
Remous-Aris Koutsiamanis
is a Postdoctoral
Researcher at IMT Atlantique in Rennes, France and
a Visiting Fellow at Bournemouth University, United
Kingdom. Previously he was a Postdoctoral Research
Assistant at Bournemouth University, United Kingdom.
He received his PhD from the Department of Electrical
and Computer Engineering of the Democritus University
of Thrace, Greece in 2016 with honours, his M.Sc. in
Artificial Intelligence from the University of Edinburgh,
United Kingdom with a full scholarship in 2006, and
his B.Sc. in Informatics from the University of Piraeus,
Greece in 2005 with distinction. Dr. Koutsiamanis has participated in numerous
international and national research projects. He has been involved in the organization
of an international event (AdHoc-Now’18) and has served as reviewer in multiple
conferences and journals. His research interests include the Industrial IoT, 6TiSCH
and network resource allocation within the wider networking field.
Georgios Z. Papadopoulos
(S’10-M’15) serves as an
Associate Professor at the IMT Atlantique in Rennes,
France. Previously, he was a Postdoctoral Researcher
at the University of Bristol. He received his Ph.D. from
University of Strasbourg, in 2015 with honors, his
M.Sc. in Telematics Engineering from University Carlos
III of Madrid in 2012 and his B.Sc. in Informatics
from Alexander T.E.I. of Thessaloniki in 2011. Dr.
Papadopoulos has participated in various international
and national (FP7 RERUM, FIT Equipex) research
projects. He has received the prestigious French national
ANR JCJC grant for young researchers. He has been involved in the organization
of many international events (AdHoc-Now’18, IEEE CSCN’18, IEEE ISCC’17).
Moreover, he has been serving as Editor for Wireless Networks journal and Internet
Technology Letters. His research interests include Industrial IoT, 6TiSCH, LPWAN,
Battery Management System and Smart Grid. He has received the Best Ph.D. Thesis
Award granted by the University of Strasbourg and he was a recipient of two Best
Paper Awards (IFIP Med-Hoc-Net’14 and IEEE SENSORS’14).
Xenofon Fafoutis
(S’09-M’14) received a PhD degree
in Embedded Systems Engineering from the Technical
University of Denmark in 2014; an MSc degree in Com-
puter Science from the University of Crete (Greece) in
2010; and a BSc in Informatics and Telecommunications
from the University of Athens (Greece) in 2007. From
2014 to 2018, he held various researcher positions at the
University of Bristol (UK), and he was a core member
of SPHERE: UK’s flagship Interdisciplinary Research
Collaboration on Healthcare Technologies. Since 2018
he is an Assistant Professor with the Embedded Systems
Engineering (ESE) section of the Department of Applied Mathematics and Computer
Science of the Technical University of Denmark (DTU Compute). His research
interests primarily lie in Networked Embedded Systems as an enabling technology
for Digital Health, Smart Cities, Industry 4.0, and the Internet of Things (IoT).
Juli´
an M. Del Fiore
is a first-year Ph.D. student in
the Icube Laboratory at the University of Strasbourg,
France. His research focuses on proposing solutions to
cope with routing insecurity on the Internet. Previously,
Juli
´
an obtained his Electronics Engineering diploma
with honors at the University of Buenos Aires. He also
did an internship in the area of Industrial IoT centered
on 6TiSCH networks at IMT Atlantique. His research
interests are the Internet and Machine Learning.
Pascal Thubert
has been actively involved in research,
development and standards efforts on Internet mobility
and wireless technologies since joining Cisco in 2000.
He currently works at Cisco’s Chief Technology and
Architecture office, where he focuses on products and
standards in the general context of IPv6, wireless, and
the Internet of Things. He co-chairs 6TiSCH, the IETF
Working Group focusing on IPv6 over the 802.15.4
TSCH deterministic MAC, and LPWAN, that applies
IETF protocols over low power wide area networking
technologies. Earlier, he specialized in IPv6 as applied
to mobility and wireless devices and developed routers and switches microcode
in Cisco’s core IPv6 product development group. In parallel with his R&D missions,
he has authored multiple IETF RFCs and draft standards dealing with IPv6, mobility
and the Internet of Things, including NEMO, 6LoWPAN and RPL. Pascal holds an
Engineering Degree from Ecole Centrale de Lyon and a PhD from IMT Atlantique.
Nicolas Montavont
is a full professor at IMT Atlantique,
in the SRCD department, and responsible of the IRISA
OCIF team since 2015. He received the M.Sc. degree
and PhD degree in Computer science from the University
of Strasbourg, France, in 2001 and 2004 respectively. He
also did a post-doc at National institute of Standard and
Technologies (NIST) in Gaithersburg, USA. He received
his HDR in 2015 from Universit
´
e Rennes 1. His research
topics are mobility and multihoming management in
IPv6 networks, wireless communications and Internet
of Things.
... With the objective to improve the packet delivery rate and delay of IoTH communications, Papadopoulos et al. [4] and Koutsiamanis et al. [5] propose a solution named Leapfrog Collaboration. When data reporting is demanded, this solution performs parallel transmissions over two paths, using promiscuous listening between the paths. ...
... The network topology is depicted in Fig. 2. This network topology has been used because Koutsiamanis et al. [5] used this same topology to conduct an evaluation of a solution proposed to deliver replicated traffic in an IoT-based e-health scenario. ...
... Regarding the occurrence of freezing state, node 8 is scheduled to assume a freezing state after 300 seconds of simulation. Koutsiamanis et al. [5] also evaluate their proposed solution under the same schedule. After entering in freezing state, the node is not able to transmit data. ...
... D. Latency Fig. 7 shows the average transmission, queuing and overall latencies of each device by different algorithms for achieving a moderate effective throughput of 1.8 bpcu at SNR γ = 20 dB. For a fair comparison between the shared-pilot and independent-pilot frame structures, it is assumed that the packets of all K devices arrive at the beginning of the frame, following the deterministic traffic arrival model in industrial IoT networks [3], [44], and that all packets have to wait until the allocated time slots. Let T α t = τ α T s /K denote the average transmission latency of each device, where τ α is the frame length given in (7). ...
... yielding the MSE for the shared pilot case in (3). By setting ψ in (25) to 0, the MSE of channel estimation for D k with the independent-pilot frame structure can be obtained as in (3). ...
Article
Full-text available
We investigate a multi-device ultra-reliable lowl-atency communication system with heterogeneous traffic and finite block length over temporally-correlated fading channels. In light of the challenging demand for accurate channel estimation with limited pilot in a short frame, two frame structures, which respectively adopt independent pilots and shared pilot, are investigated. Block lengths and pilot lengths are jointly optimized for the two frame structures, through instantaneous channel state information (CSI) based dynamic optimization and statistical CSI based static optimization, to strike the tradeoffs among performance, complexity and signaling overhead. The proposed joint optimization algorithms significantly outperform the existing approaches that solely optimize block lengths or pilot lengths. The dynamic optimization algorithms achieve near-optimal performance at dramatic complexity reduction over exhaustive search, and maintain robustness against traffic heterogeneity. Also, the static optimization algorithms are conducted offline, while still outperforming the previous instantaneous CSI based dynamic optimization approaches. It is demonstrated that the independent-pilot frame structure with dynamic optimization is preferable in the scenario with high traffic heterogeneity or high mobility, and that the shared-pilot frame structure with static optimization presents a comparable performance to the former in the case of low mobility, incurring negligible complexity and signaling overhead.
... 15.4e standard is aimed at industrial, commercial, and healthcare applications [28], but still, it needs to gain more acceptation. Several amendments aimed at improving reliability and latency predictability were presented in [29,30]. WirelessHART, An independent communication channel was used for control communication operations. ...
... 15.4e standard is aimed at industrial, commercial, and healthcare applications [28], but still, it needs to gain more acceptation. Several amendments aimed at improving reliability and latency predictability were presented in [29,30]. WirelessHART, based over IEEE 802. ...
Article
Full-text available
The nature of wireless propagation may reduce the QoS of the applications, such that some packages can be delayed or lost. For this reason, the design of wireless control applications must be faced in a holistic way to avoid degrading the performance of the control algorithms. This paper is aimed at improving the reliability of wireless control applications in the event of communication degradation or temporary loss at the wireless links. Two controller levels are used: sophisticated algorithms providing better performance are executed in a central node, whereas local independent controllers, implemented as back-up controllers, are executed next to the process in case of QoS degradation. This work presents a reliable strategy for switching between central and local controllers avoiding that plants may become uncontrolled. For validation purposes, the presented approach was used to control a planar robot. A Fuzzy Logic control algorithm was implemented as a main controller at a high performance computing platform. A back-up controller was implemented on an edge device. This approach avoids the robot becoming uncontrolled in case of communication failure. Although a planar robot was chosen in this work, the presented approach may be extended to other processes. XBee 900 MHz communication technology was selected for control tasks, leaving the 2.4 GHz band for integration with cloud services. Several experiments are presented to analyze the behavior of the control application under different circumstances. The results proved that our approach allows the use of wireless communications, even in critical control applications.
... This introduces a hard timing constraint on the maximum allowable delay in message delivery between those participants. This is, however, a feature that commonly deployed communications protocols are not designed to meet, with 'best-effort' message delivery being the standard paradigm [48]. ...
Preprint
Purpose of Review. This review summarizes the broad roles that communication formats and technologies have played in enabling multi-robot systems. We approach this field from two perspectives: of robotic applications that need communication capabilities in order to accomplish tasks, and of networking technologies that have enabled newer and more advanced multi-robot systems. Recent Findings. Through this review, we identify a dearth of work that holistically tackles the problem of co-design and co-optimization of robots and the networks they employ. We also highlight the role that data-driven and machine learning approaches play in evolving communication pipelines for multi-robot systems. In particular, we refer to recent work that diverges from hand-designed communication patterns, and also discuss the "sim-to-real" gap in this context. Summary. We present a critical view of the way robotic algorithms and their networking systems have evolved, and make the case for a more synergistic approach. Finally, we also identify four broad Open Problems for research and development, while offering a data-driven perspective for solving some of them.
... s ∈ S, a ∈ A} is an optimal solution of problem (8). ...
Preprint
Full-text available
In this paper, we for the first time investigate the random access problem for a delay-constrained heterogeneous wireless network. We begin with a simple two-device problem where two devices deliver delay-constrained traffic to an access point (AP) via a common unreliable collision channel. By assuming that one device (called Device 1) adopts ALOHA, we aim to optimize the random access scheme of the other device (called Device 2). The most intriguing part of this problem is that Device 2 does not know the information of Device 1 but needs to maximize the system timely throughput. We first propose a Markov Decision Process (MDP) formulation to derive a model-based upper bound so as to quantify the performance gap of certain random access schemes. We then utilize reinforcement learning (RL) to design an R-learning-based random access scheme, called tiny state-space R-learning random access (TSRA), which is subsequently extended for the tackling of the general multi-device problem. We carry out extensive simulations to show that the proposed TSRA simultaneously achieves higher timely throughput, lower computation complexity, and lower power consumption than the existing baseline--deep-reinforcement learning multiple access (DLMA). This indicates that our proposed TSRA scheme is a promising means for efficient random access over massive mobile devices with limited computation and battery capabilities.
... In this regards, we choose 15 KB for the typical payload size of each data. 35 Regarding to the internal memory of proxy nodes, we carried out simulations with proxy nodes having equal cache capacity of 92 KB as well as proxy nodes having heterogeneous cache capacities randomly chosen from the range [67, 117] KB. 36 As these two scenarios showed quite similar outcome and in order to reduce the number of varying factors impacting the results, we then fixed the cache capacity to 92 KB as in Koutsiamanis et al 36 which represents the mean value of the interval [67, 117] KB. Also we consider expected latency on each node, which is the time that are needed for the data exchange from a node to another. ...
Article
Full-text available
Industry 4.0 is the emerging trend of manufacturing technologies towards smart factories. The industrial internet of things (IIoT) plays a primordial role to achieve the objectives of Industry 4.0, by enabling the cyber‐physical production systems to communicate and cooperate with each other and with humans both internally and across the participants of the supply chain. The Publish/Subscribe is the communication model used commonly in IIoT networks, in which the data generated by sensors are cached in a central controller to be subsequently consumed by actuators. However, the traditional centralized data management is inappropriate for real applications because of its communication overhead and inability to cope with strict delay requirements of IIoT applications and energy constraints of IIoT networks. To address these issues, we propose an energy‐efficient data management scheme (EDMS), in which a set of IoT nodes are selected to cache data distributedly, so that the network energy consumption is optimized while the constraints on data access latency and nodes' cache capacity are met. In this measure, the extra IoT nodes not involved in data transmission nor in data caching are switched off to prolong the network lifetime. We model the aforementioned optimization problem as an integer linear programming (ILP) considering the realistic constraints and characteristics of IIoT networks and solve it using the CPLEX tool. Moreover, we propose a metaheuristic algorithm based on ant colony optimization to solve the aforementioned problem in a timely manner. Simulation results show that EDMS outperforms significantly alternative solutions in terms of energy consumption, data latency, and cache utilization.
... This strategy is recommended for IWSN [7], where coping with lossy link is even more challenging due to harsh and noisy industrial ennvironments. By adopting multipath routing, there is always an alternative path ready if a link or a node fails [74] or, alternatively, multiple copies of the same data packet traverse the network in parallel along different paths [75]. In addition, multipath routing may improve network performance in term of end-to-end delay and lifetime. ...
Thesis
Full-text available
With current low-power wireless standards, the mistrust about wireless technology for industrial applications is unjustifiable. For instance, a Medium Access Control (MAC) technique, called Time Slotted Channel Hopping (TSCH), has made wire-like end-to-end reliability, certified security, and over a decade of battery lifetime a reality in Industrial Wireless Sensor Networks (IWSNs). TSCH is integrated into the IEEE~802.15.4 standard, and it is a cornerstone of the open standardised protocol suite for the Industrial Internet of Thing (IIoT) proposed by IETF. Specifically, the IETF 6TiSCH stack combines the industrial performance of TSCH with a set of higher layers protocols providing IPv6-connectivity to constrained devices. Thus, it promises, above all, interoperability between vendors and seamlessly integration of IWSNs into the Internet. Despite these high potentials and the high reputation of 6TiSCH in industry and academia, challenges remain, and some of its limitations need to be understood. In this thesis, we focus on the issues related to the harmonisation of the asynchronous IPv6-based upper layers with the synchronous TSCH technique,which rely on control plane primitives such as the network bootstrap procedure, the management of communication resources and the collection of network statistics. We identify in which circumstances the 6TiSCH standardised control primitives, which should lay the foundations for a reliable operational 6TiSCH network, exhibit limitations. After explaining the shortcomings and their causes, we design refinements and validate them simulatively. The focus is not on data transmissions but on mechanisms for a dependable exchanging of control messages. Nevertheless, these have been designed without significantly reducing the available bandwidth for data applications and the lifetime of power-constrained nodes. Accordingly, we provide the following main contributions. First, we analyse the interplay between the scheduling of control messages and the multi-hop route computation during the network bootstrap phase, pointing out the limits of the current guidelines that may preclude or penalise the 6TiSCH network's operational state. Indeed, an improper choice of the protocol parameters may lead to a very long and energy-consuming network formation and stabilisation phase (e.g. more than 30 minutes even in small 5x5 grid networks). Second, we examine different resource allocation strategies for bootstrapping a 6TiSCH network. Here, we design, implement and evaluate a scheduling mechanism for coordinating the transmission of control messages among neighbouring nodes in a dynamic and distributed way. This mechanism has exhibited a significantly faster network formation than the default configuration, even in challenging chain network topologies, where it consumes at most 0.4% of the battery's charge of a sensor node. Finally, we investigate how to obtain an accurate Link Quality Estimation (LQE) in 6TiSCH. We demonstrate that state-of-the-art strategies, which are not designed having TSCH in mind, are too inaccurate for guaranteeing a reliable and stable 6TiSCH network setup. Indeed, internal interference hampers their link measurements. To overcome this issue, we propose a LQE strategy that allows a collision-free transmission of broadcast probe messages even during the network setup. This proposal improves the estimation accuracy dramatically, exhibiting a quite perfect estimation of at least 90% of the links in different network topologies and in a short time (i.e. in order of minutes) We are firmly convinced that 6TiSCH is an IIoT key enabler. Despite that, we forewarn the risk of its blind adoption as one-size-fits-all solutions in this work. Addressing some limitations in its control primitives and providing essential enhancements, we believe this thesis supports the future wide adoption of 6TiSCH in the industry.
... The guaranteed packet delivery results in a longer slotframe and hence the longer delays. LeapFrog collaboration scheduling [9] allows a packet transmitted by a node to be received by multiple receivers in the same cell. The transmission is considered to be successful if at least one receivers able to decode the packet. ...
Article
Internet of Things (IoT) is a technological concept bringing sustainability and sophistication to our lives and is a significant component of Industry 4.0. The main requirements of Industrial IoT are reliability, stringent latency and energy efficiency. The IEEE 802.15.4e standard has adapted a Medium Access Control (MAC) behavioural mode called Time Slotted Channel Hopping (TSCH) to address the requirements of Industry 4.0. The 6TiSCH (IPv6 over IEEE 802.15.4e TSCH) protocol stack enables us to schedule the transmissions in TSCH network to achieve application-specific guarantees. In this paper, we propose DPLLS, a conflict-free dynamic programming based low-latency scheduling algorithm for TSCH network. In particular, identifying non-interfering transmissions in the network is posed as a maximum weight independent set (MWIS) problem, which is solved using a dynamic programming method. We organise each slotframe into smaller parts called blocks in which the non-interfering transmissions are scheduled simultaneously using either a conservative or a greedy scheme. The blocks are repeated in a slotframe to accommodate the retransmissions of packets to ensure reliability and minimise the latency. The proposed scheme is evaluated using an example to find the suitable block length of the proposed schemes and compared with the existing scheduling algorithms.
Conference Paper
Full-text available
The upcoming Internet of Things (IoT) applications include real-time human activity monitoring with wearable sensors. Compared to the traditional environmental sensing with low-power wireless nodes, these new applications generate a constant stream of a much higher rate. Nevertheless, the wearable devices remain battery powered and therefore restricted to low-power wireless standards such as IEEE 802.15.4 or Bluetooth Low Energy (BLE). Our work tackles the problem of building a reliable autonomous schedule for forwarding this kind of dynamic data in IEEE 802.15.4 TSCH networks. Due to the a priori unpredictability of these data source locations, the quality of the wireless links, and the routing topology of the forwarding network, it is wasteful to reserve the number of slots required for the worst-case scenario; under conditions of high expected datarate, it is downright impossible. The solution we propose is a hybrid approach where dedicated TSCH cells and shared TSCH slots coexist in the same schedule. We show that under realistic assumptions of wireless link diversity, adding shared slots to a TSCH schedule increases the overall packet delivery rate and the fairness of the system.
Conference Paper
Full-text available
In traditional routing protocols designed for Wireless Sensor Networks, each sensor node is related to one or more neighbors that will forward its readings up to the sink. This technique performs well for static topologies with homogeneous configurations, but usually fails to cope with network dynamics such as mobility and node failures. Opportunistic routing is an approach to address this particular problem. In this context, the data packets are addressed to a set of potential forwarders and then forwarded by the neighbor that first acknowledges the message. Yet, several former studies demonstrated that in some cases, a single packet may be forwarded by multiple neighbors simultaneously. This situation leads to packet duplication and consequently to increased channel occupancy and energy consumption in the network. In this paper, we study to what extent the previously reported phenomenon depends on both the topology density and the nodes MAC configuration. We then introduce a mechanism that handles the potential deafness in the network through heterogeneous configuration among the nodes in the network. We do so through local, dynamic and automatic MAC parameters adaptation, in order to reduce unnecessary traffic, channel occupancy and energy consumption due to packet duplication in opportunistic networks. Finally, we provide both theoretical analysis and experimental campaign to detail the benefits of our approach.
Thesis
Based on time, resource reservation, and policy enforcement by distributed shapers, Deterministic Networking provides the capability to carry specified unicast or multicast data streams for real-time applications with extremely low data loss rates and bounded latency, so as to support time-sensitive and mission-critical applications on a converged enterprise infrastructure.As of today, deterministic Operational Technology (OT) networks are purpose-built, mostly proprietary, typically using serial point-to-point wires, and operated as physically separate networks, which multiplies the complexity of the physical layout and the operational (OPEX) and capital (CAPEX) expenditures, while preventing the agile reuse of the compute and network resources.Bringing determinism in Information Technology (IT) networks will enable the emulation of those legacy serial wires over IT fabrics and the convergence of mission-specific OT networks onto IP. The IT/OT convergence onto Deterministic Networks will in turn enable new process optimization by introducing IT capabilities, such as the Big Data and the network functions virtualization (NFV), improving OT processes while further reducing the associated OPEX.Deterministic Networking Solutions and application use-cases require capabilities of the converged network that is beyond existing QOS mechanisms.Key attributes of Deterministic Networking are: - Time synchronization on all the nodes, often including source and destination - The centralized computation of network-wide deterministic paths - New traffic shapers within and at the edge to protect the network- Hardware for scheduled access to the media.Through multiple papers, standard contribution and Intellectual Property publication, the presented work pushes the limits of wireless industrial standards by providing: 1. Complex Track computation based on a novel ARC technology 2. Complex Track signaling and traceability, extending the IETF BIER-TE technology 3. Replication, Retry and Duplicate Elimination along the Track 4. Scheduled runtime enabling highly reliable delivery within bounded time 5. Mix of IPv6 best effort traffic and deterministic flows within a shared 6TiSCH mesh structureThis manuscript presents enhancements to existing low power wireless networks (LoWPAN) such as Zigbee, WirelessHART¿and ISA100.11a to provide those new benefits to wireless OT networks. It was implemented on open-source software and hardware, and evaluated against classical IEEE Std. 802.15.4 and 802.15.4 TSCH radio meshes. This manuscript presents and discusses the experimental results; the experiments show that the proposed technology can guarantee continuous high levels of timely delivery in the face of adverse events such as device loss and transient radio link down.
Article
In this paper, we explore the role of simulators and testbeds in the development procedure of protocols or applications for Wireless Sensor Networks (WSNs) and Internet of Things (IoT). We investigate the complementarity between simulation and experimentation studies by evaluating latest features available among open testbeds (e.g., energy monitoring, mobility). We show that monitoring tools and control channels of testbeds allow for identification of crucial issues (e.g., energy consumption, link quality) and we identify some opportunities to leverage those real-life obstacles. In this context, we insist on how simulations and experimentations can be efficiently and successfully coupled with each other in order to obtain reproducible scientific results, rather than sole proofs-of-concept. Indeed, we especially highlight the main characteristics of such evaluation tools that allow to run multiple instances of a same experimental setup over stable and finely controlled components of hardware and real-world environment. For our experiments, we used and evaluated the FIT IoT-LAB facility. Our results show that such open platforms, can guarantee a certain stability of hardware and environment components over time, thus, turning the unexpected failures and changing parameters into core experimental parameters and valuable inputs for enhanced performance evaluation.
Conference Paper
Performance analysis of newly designed solutions is essential for efficient Internet of Things and Wireless Sensor Network (WSN) deployments. Simulation and experimental evaluation practices are vital steps for the development process of protocols and applications for wireless technologies. Nowadays, the new solutions can be tested at a very large scale over both simulators and testbeds. In this paper, we first discuss the importance of repeatable experimental setups for reproducible performance evaluation results. To this aim, we present FIT IoT-LAB, a very large-scale and experimental testbed, i.e., consists of 2769 low-power wireless devices and 127 mobile robots. We then demonstrate through a number of experiments conducted on FIT IoT-LAB testbed, how to conduct meaningful experiments under real-world conditions. Finally, we discuss to what extent results obtained from experiments could be considered as scientific, i.e., reproducible by the community.
Conference Paper
Packets traversing through a multi-hop Low-power and Lossy Network (LLN) experience various link outage conditions in transit that can lead to packet loss at an intermediate relay node. Motivated by the problem of providing high Packet Delivery Ratio (PDR) in the LLNs running RPL(Routing Protocol for Low-power and Lossy Networks) routing protocol, we implemented a lightweight add-on functionality to the network layer's packet forwarding mechanism, Link Peek, in which the forwarding node iteratively retransmits the packet to its alternate/next best parent belonging to the same RPL DODAG(Destination Oriented Directed A cyclic Graph) instance whenever IEEE 802.15.4 MAC layer retransmission count set by our mechanism for the current best parent is exceeded. With this simple reactive and successive retry mechanism, our experimentation has shown significant performance gain in terms of achieving high PDRs, even at higher data rates, over the LLNs under different operating environments. We implemented Link Peek over the Contiki RPL protocol stack. We describe the implementation details and discuss various tunable parameters that affect the network performance. The Cooja simulation and real-world physical test bed experimentation with link outage environment showed PDRs > 99.40% with Link Peek as against the PDRs of 76-90% obtained using the native network stack implementation. Through the experimental results we demonstrate the effectiveness of our approach in supporting higher data rates even in node mobility scenarios.