Adaptive Wireless Sensor Networks for Aircraft

Pascale Minet∗ , Gerard Chalhoub† , ErwanLivolant∗ , Michel Misson† , Badr Rmili‡ and Jean-Francois Perelgritz§
∗
Inria, 78153 Le Chesnay Cedex, France
† Clermont University, 5 avenue Blaise Pascal, 63178 Aubiere, France
‡ CNES Launcher Directorate, 52 rue Jacques Hillairet, 75012 Paris, France
§ Airbus Group Innovations, 12 rue Pasteur, 92150 Suresnes, France

Abstract—In this paper, we unify the requirements of non- of the nodes, aircraft fuselage, empennage, wings,
critical and health monitoring applications in Aircraft and engine, landing gear; launcher stage, tank, fairing;
Launchers. We present different challenges faced by wireless
• Latency from a sensor node to the closest sink shall
sensor networks to meet these requirements. We also propose
a solution that provides an adaptive multichannel collision-free range from 100 to 500 ms (depending on the Aircraft
protocol for data gathering. application);
• Measurement dating accuracy from 1 to 100 µs (de-
I. C ONTEXT pending on the Aircraft application)...
Wireless Sensor Networks (WSNs) have been largely
Therefore, to meet all these application requirements, a multi-
adopted in various application domains: environmental appli-
channel WSN is required. In addition, if the IEEE 802.15.4 [2]
cations, reasonable use of fertilizers and pesticides in precision
technology is chosen, multihop routing is required to allow any
agriculture, fire detection, pollutant detection in industrial
sensor node to reach the sink.
worksites, exploration of natural resources, to name a few.
Are WSNs able to meet the requirements for Aircrafts and This paper is organized as follows. Section II describes
Launchers? For the sake of readability, this paper uses the the challenges resulting from multichannel use and presents
term ”Aircraft” to cover both Aircrafts and Launchers. different solutions of the state of the art to address them.
Section III presents an example of solution designed to be
Generally speaking, the benefits of WSNs in Aircraft would
compliant with the unified WSN requirements for Aircraft and
be a reduction of the complexity and mass of wiring, the
to IEEE 802.15.4 COTS (Commercial Off-The-Shelf). Finally,
facilitation of the addition or removal of sensors, the possibility
Section IV concludes this paper.
of installing sensors in locations inaccessible due to wiring
constraints and easier increases of the number of sensors. II. M ULTICHANNEL CHALLENGES
For instance, as authors of [1] have emphasized, WSNs can
achieve a dramatic saving of mass on Aircraft. Indeed, they In a multichannel network additional challenges arise for
have shown that wires represent 57% of the total weight of the network protocols. In this section we will discuss these chal-
Developmental Flight Instrumentation data system for Orion lenges and present solutions of the state of the art.
Exploration Flight Test 1. To cope with Aircraft constraints A. Network build-up
and specific environments, unified WSN requirements have
been defined by Aircraft manufacturers and End-Users. These During the network build-up phase, nodes try to be part
requirements cover different types of Aircraft and Launchers. of the network in order to be able to exchange and relay
data packets. When a node is activated it usually starts with
In terms of unified WSN requirements for Aircraft, this a network discovery phase during which it scans for existing
paper is focused only on non-critical and HMS (Health Mon- networks. The scanning procedure in a multichannel network
itoring System) measurements. These requirements can be is not trivial. A new node should be able to detect an activity
summarized as follows: on a certain channel and try to communicate with this network
• Static sensors such as temperature, pressure, etc.; in order to gain access and be part of it. Special advertisement
frames usually named beacons are used for signalling the
• Dynamic sensors such as vibration, shock, strain
presence of the network. This is the case in WiFi [3] and
gauges, etc.;
ZigBee [4] for example.
• Sampling rate: from a few samples per hour to 10
ksps; The challenge resides in making sure that the new node
is able to find the network. In other words, the node should
• Sensor network with heterogeneous sampling rates;
be able to be at the same time on the same channel as its
• Sensor nodes per network: 1 to 50; neighboring nodes having already joined the network. Without
• Expected battery capacity for a sensor node: from 40 a fixed and known control channel this procedure can last
minutes to 14 hours in active mode and 24 months in very long and, in some cases, drain a lot of the node energy
sleep mode; resources.
• Sensor nodes integrated in confined environment lead- A solution to make this phase less time and energy con-
ing to propagation issues; suming would be to exchange control traffic on a fixed and
• Sensor nodes integrated at a distance from the closest known channel. This will allow new nodes to scan only one
sink that is greater than the communication range channel. On the other hand, nodes should periodically switch

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on August 19,2023 at 02:13:14 UTC from IEEE Xplore. Restrictions apply.
 to this control channel and send beaconing frames in order to the size, most of the nodes are equipped with one radio
be detected by new nodes. transceiver. Thus most of the protocols propose solutions to
allow nodes either to send or to receive at a time using only one
B. Multihop synchronization radio transceiver. Some assignment schemes allocate channels
When using multiple channels it is important to be able for the receivers, whereas others for the transmitters. Some
to know when to use a certain channel and for how long. protocols also combine channels allocation with time slots
Thus, maintaining multihop synchronization is a must. This allocation.
synchronization could be achieved using an external syn-
Some protocols such as [7] propose a static channel assign-
chronization device, but the difficult part would be to make
ment where nodes keep using the same channel until neighbor-
sure that this device can be reached by all the nodes of the
hood or interference conditions change and force them to seek
network. For example, synchronizing nodes using a Global
a more suitable channel. This approach is often lightweight
Positioning System is only possible when all nodes are able to
and does not waste energy due to frequent channel switchings.
communicate with satellites. In addition, this has consequences
Other protocols such as [8], [9], [10], use a semi-dynamic
on the nodes and energy consumption.
channel assignment where nodes switch channels according to
Another approach would consist in achieving relative syn- the destination. This approach is adaptive and allows more
chronization based on an internal reference that is part of the flexibility in choosing suitable channels. A more dynamic
network. For instance, a designated node in the network could channel assignment method consists in changing channels at
broadcast a synchronization beacon that is propagated by other each transmission such as [11], [12]. This approach is most
nodes in order to reach all the nodes of the network. Similar robust because it avoids bad channels and enables nodes to
approaches have been studied in [5] and [6]. use all available channels but at the cost of energy and time
wasting due to frequent channel switchings.
C. Selection of channels
Protocols such as [13] and [14] propose a multi-interface
Wireless standards based on IEEE 802.15.4 have 16 avail- sink in order to enhance the reception throughput of this
able channels in the 2.4 GHz frequency band. Other wireless particular node. Indeed, assigning different channels to the sink
standards such as Wi-Fi and Bluetooth also use the 2.4 GHz transceivers allows simultaneous receptions.
band. This makes the IEEE 802.15.4 networks vulnerable to in-
terference coming from other nearby networks including other E. Network connectivity
IEEE 802.15.4 networks. We call this external interference.
External interference happens when nodes of the network re- Multichannel assignment solutions may differ on the as-
ceive perturbations coming from sources that are not part of the sumptions made regarding network connectivity, as illustrated
network. Furthermore, internal interference is caused by nodes in Figure 1, where channel ch1 is represented by a black solid
that are part of the network. Depending on the modulation and line whereas channel ch2 is depicted in red dashed line:
the frequencies used on the physical layer, nodes using the
a) The same topology exists on all channels and connectivity
same channel are prone to generate interferences. In order to
is assumed on each channel.
avoid interference, orthogonal channels should be used.
b) The topology may differ from one channel to another but
Internal interferences can be avoided using channel assign- connectivity is assumed at least on one channel, usually
ment techniques that will be discussed in the following section the channel used for control messages; this assumption is
(Section II-D). In order to avoid external interferences, a scan frequently done. For instance in Figure 1, there are two
is usually achieved to identify the energy level that is detected disconnected parts on channel ch2.
on each channel. The scan procedure should result in what c) Several channels are needed to ensure network connectivity.
is called a channel blacklist, which is a list of channels that This is the most permissive assumption.
should not be used in the network.
This blacklisting technique can be achieved in a distributed 5 6 5 6 5 6
manner. Every node in the network scans its own environment
and builds its own blacklist. This local blacklist is then used 4 1 4 1 4 1
locally to choose a convenient channel, or is sent to a controller
node that is in charge of distributing channels to all nodes of
the network. This of course generates a significant delay before 3 2 3 2 3 2
being able to choose a channel. In addition, if interference
is not stable (which is most usually the case), the channel (a) Same topology. (b) Connectivity on (c) Connectivity with
blacklist should be updated frequently. Blacklisting channels one channel. two channels.
can also be achieved for the whole network. This allows
segmenting the frequency band over different networks in order Fig. 1: Different cases of connectivity.
to enhance performance while avoiding external interferences.
This can be achieved if nearby networks are manageable, but F. Neigborhood discovery
this is not often the case. For any node u, a node v is said one-hop neighbor of
u on channel c if and only if u is able to receive messages
D. Channel assignment
from v on channel c and v is able to receive messages from
Channel assignment represents one of the main challenges u on channel c: a symmetric link exists between them on
of multi-channel MAC protocols. Constrained by the cost and channel c. Some solutions do not check the symmetry of a link

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on August 19,2023 at 02:13:14 UTC from IEEE Xplore. Restrictions apply.
 before using it. This will cause useless retransmissions when preamble sending time, which results in increasing the protocol
an acknowledgment is required and the link is not symmetric overhead and the number of repetitions due to frequent channel
[15]. Notice that in a real multichannel environment, a node switchings.
may be a one-hop neighbor on a channel and not on another
In [7], authors describe TMCP, a centralized Tree-based
one [16]. There are different types of solutions according to the
Multi-Channel Protocol for data collection applications. It uses
assumptions made, assumptions similar with those on network
a fixed channel assignment approach for channel allocation.
connectivity. The simplest ones perform neighborhood discov-
The whole network is partitioned into multiple sub-trees
ery only on one channel, usually the control channel whereas
having the base station as a common root where each sub-
the most sophisticated ones perform as many neighborhood
tree is allocated a different channel. TMCP finds available
discoveries as channels used. Some solutions take benefit of
orthogonal channels, partitions the whole network into sub-
large similarities of links between channels to store efficiently
trees and allocates a different channel to each sub-tree. TMCP
this knowledge.
improves the throughput with regard to a single channel
G. Conflicting nodes solution, while keeping high packet delivery ratio and low
latency. However, TMCP blocks the direct communications
In the time slot and channel assignment problem, two between nodes belonging to different sub-trees.
conflicting nodes prevent a node from receiving a data or an
acknowledgement intended for it when they use the same chan- 2) Contention-free based protocols: In [12], authors pro-
nel during the same time slot. Assuming that the immediate posed MC-LMAC (Multi-Channel Lightweight MAC) proto-
acknowledgement is used at the MAC level (i.e. each unicast col which guarantees that the same slot/channel pair is not
data packet is acknowledged in the slot it is sent), we determine simultaneously used in the neighborhood up to two-hop. MC-
two types of collisions: data-data and data-acknowledgement. LMAC suffers from the overhead of the control messages
Taking into account that data sent by sensor nodes are collected that are exchanged in order to discover the channels used in
by the sink using a routing tree, we can show that the only the neighborhood up to two-hop. The problem increases with
possible conflicts are those given by Property 1. network density.
Property 1: For any node u, its conflicting nodes, when In [17], authors proposed TSMP (Time Synchronized Mesh
using the immediate acknowledgment, are: Protocol) on which industrial standards such as WirelessHART
and ISA100.11a are based, as well as IEEE 802.15.4e. It uses a
a) node u itself, channel hopping technique to enable nodes to switch channels
b) node P arent(u), at each transmission. It suffers from lack of topology changes
c) one-hop neighbors of u, support.
d) one-hop neighbors of P arent(u), 3) Hybrid protocols: In [9] authors proposed EE-MAC
e) nodes whose parent is one-hop neighbor of u, (Energy Efficient hybrid MAC for WSN). It is a central-
ized protocol that uses a semi-dynamic channel assignment
f) nodes whose parent is one-hop neighbor of P arent(u), approach for channel allocation. EE-MAC operates in two
H. Medium access phases, a setup phase and a transmission phase. During the
first phase, neighbor discovery, slot assignment, and global
Multichannel MAC protocols that have been proposed synchronization are achieved. These operations run only during
in the literature use either TDMA (Time Division Multiple the set-up phase and every time a change in the topology oc-
Access), or CSMA/CA (Carrier Sense Multiple Access with curs. During the transmission phase, time is divided into time
Collision avoidance), or a combination of both techniques. slots. Each slot is divided into schedule subslots and contention
In what follows we will briefly describe the most known subslots. Each cycle starts with scheduled slots followed by
multichannel protocols. contention slots. Nodes use LPL (Low Power Listening) [18]
during contention slots and send Hello messages to the base
1) Contention based protocols: In [8] authors proposed
station. EE-MAC suffers from the overhead of Hello messages
MASN, a multi-channel protocol for hierarchical ZigBee net-
that are exchanged and sent to the base station.
works with many-to-one transmissions. Assignment of the
different channels is centralized by a coordinator and based In [11], authors proposed MuChMAC (Multi-Channel
on the hierarchical address assignment process used in Zig- MAC) protocol. It uses a dynamic channel assignment ap-
Bee. The main advantage of this solution is its simplicity of proach. Time is divided into slots. Each node is able to
integration in IEEE 802.15.4 devices with light modifications independently choose its receiving channel switching sequence
on the MAC layer. However, authors used a single topology for based on its identifier and the current slot number using a
the simulations which might be the most convenient topology pseudo-random generator. A broadcast slot is inserted every
for MASN. n slots. These common broadcast slots also follow a pseudo-
random channel hopping sequence. A sender is thus able to
In [10], authors proposed MMSN which uses a semi-
calculate the channel of the receiver. The main drawback of
dynamic channel assignment approach for channel allocation
MuChMAC is that channel allocation is based on a random
based on different strategies. Strategies differ according to
mechanism that does not take into consideration the channel
the level of overhead and the effectiveness of the channel
usage in the neighborhood.
allocation. At the beginning of each time slot, nodes contend
for the medium to broadcast control traffic on a common In [13], authors proposed HMC-MAC (Hybrid Multi-
broadcast channel. When a node transmits a packet, it switches Channel MAC) protocol that is based on TDMA for sig-
between its own channel and the destination channel during the nalling traffic and CSMA/CA combined with FDMA for

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on August 19,2023 at 02:13:14 UTC from IEEE Xplore. Restrictions apply.
 data exchange. A TDMA interval is dedicated for neighbor III. A N EXAMPLE OF SOLUTION
discovery and channel allocation process. HMC-MAC aims
at reducing control traffic overhead. It allows nodes to share The work described in this section is part of the larger
slots on the same channel to send data to the same destination. project SAHARA supported by the French aerospace cluster
Results in [19] show that it enhances network performance in ASTech. This project started at the end of 2011 and comprises
terms of number of collisions and also packet delivery rate. Academics, Research Institutes, SMB (Small and Medium-
However, it suffers from high end-to-end delays due to packet size Businesses) and Aircraft Manufacturers. The goal of this
accumulation in nodes close to the sink. project is to develop existing wireless technologies and COTS
and adapt them into a generic and adaptive WSN dedicated
to Aircraft. The scope of the project covers the definition
I. Multichannel routing of unified WSN requirements, the development of WSN
technologies and protocols, the development of TRL6 WSN
In convergecast scenarios, where all nodes send their traffic
demonstrators, tests in representative Aircraft environments
to one destination, namely the sink, the routing tree is generally
and the development of in-aircraft RF propagation models.
built using a gradient method: the sink broadcasts a message
including a cost. A node receiving this message selects the
A. Description of the solution proposed
transmitting node as parent if and only if it is the one-hop
neighbor that provides the smallest cost. In such a case, the We now present an example of solution performing joint
receiving node updates the cost before forwarding the message. routing, time slot and frequency assignment and ensuring
multihop synchronization by cascading beacons. A dedicated
A node may also select several potential parents. For
channel is used for signalling traffic. However, notice that it
instance, assuming that the cost is simply equal to the depth of
can also be used by data traffic. We assume that network
a node, any node u 6= sink selects as its potential parents its
connectivity is ensured on this control channel.
one-hop neighbors (i.e. nodes with which it has a symmetric
link) that have a smaller depth than itself. The depth of a node 1) Network deployment: As previously stated, we tackle
is recursively computed: the sink has a depth 0, its one-hop static deployments where nodes remain static after deployment.
neighbors have a depth 1, . . . Notice that the depth of a node The first node to be activated is the node that is responsible for
in a real multichannel network may differ from one channel creating the network and managing the time synchronization.
to another. Indeed, this specific node starts signalling its presence using
periodic beacon frames. The sink node can be chosen to play
The selection of potential parents may also take into that role. The beacon frame will help other nodes to detect
account statistics about link quality. A potential parent with a its presence when scanning on the specified control channel.
high link quality is more frequently used to transfer application When other nodes are activated, they will detect the beacon
messages. and send a join request to the sink. Once a node is allowed to
be part of the network (network admission process is out of
J. Traffic load scope of this paper) this node will propagate the beacon that
it has received. In order to avoid collisions between beacon
For any node u, let Gen(u) denote the number of data frames, the sink indicates in which order the beacons should be
packets generated by u in a slotframe. We can compute propagated. Thus, beacons are sent in a collision free TDMA
T rans(u), the number of data packets transmitted by u in manner where each node has its own slot for broadcasting the
a data gathering (assuming that a node is able to send all the beacon frame. Nodes that are multiple hops away from the
traffic that it generates). We have T rans(u) = Gen(u) + the sink will send their join request in a multihop manner to reach
number of data packets received by u f rom its children the sink. Nodes are informed about the propagation order in
and f orwarded to its parent. We can write: the join response messages. In order for other nodes to update
X their propagation order, the beacon will include the updates
T rans(u) = Gen(v), (1) during m consecutive beacon cycles (a beacon cycle is the
v∈subtree(u) period separating two successive transmissions of the beacon
by the sink). Indeed, including updates and not the complete
where subtree(u) denotes the subtree rooted at u in the routing list of nodes enables this mechanism to be scalable. For more
tree. details about the beacon propagation mechanism please refer
to [20].
With regard to traffic load we can distinguish two ap-
proaches for time slot and channel assignment. The first one 2) Slotframe: The slotframe describing the organization of
does not take into account heterogeneous traffic loads. All node activities consists of four periods:
nodes have the same number of slots assigned, although nodes
close to the sink have a considerably higher traffic load. As • a synchronization period that contains the multi-hop
a consequence, in the absence of message loss and message beacon propagation, using the control channel.
aggregation, all data sent by sensor nodes in a slotframe may • a control period that allows the transmissions of
need up to maxu∈W SN T rans(u) slotframes to reach the sink. network control messages in CSMA/CA, using the
control channel.
The second one, said traffic-aware, assigns the exact num-
ber of slots needed by each node per slotframe. Consequently, • a data period that collects the applicative data accord-
in the absence of message loss, a single slotframe is sufficient ing to a collision-free schedule. All available channels,
to enable all data gathered in this slotframe to reach the sink. including the control channel, are used.

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on August 19,2023 at 02:13:14 UTC from IEEE Xplore. Restrictions apply.
 • a sleep period where all network nodes sleep to save Broadcasting the Schedule to all nodes requires K · (N − 1)
node energy. messages in the worst case. Hence, the centralized assignment
requires (AverageDepth + K) · (N − 1) transmissions.
3) Multi-interface sink: As most of the wireless sensor net-
works deployed support data gathering applications. The sink In a distributed assignment, we assume that any node u
node is the destination of all the data generated in the network. uses T rans(u) as its priority for slot and channel assignment.
Thus, in order to enhance the sink reception throughput, we use Any node u computes T rans(u) according to Equation 1 and
a multi-interface sink. Indeed, the sink node has multiple radio transmits it to its parent denoted P arent(u). This requires
interfaces enabling it to receive simultaneously on different N − 1 messages to enable all nodes to know their own value
channels. of T rans(u), where N is the total number of nodes in the
wireless sensor network. Any node u 6= sink should notify
4) Neighborhood discovery: Neighborhood Discovery in
its priority first and then its slot assignment to its conflicting
a multichannel environment is done successively on each
nodes. Hence, u notifies its slot assignment to its one-hop
channel in the channel list defined by the application. On
neighbors. This notification is forwarded by nodes that are
the channel to be discovered, each node broadcasts a Hello
parents and are one-hop away from u or P arent(u). Therefore
message containing the list of its one-hop neighbors. The sink
the slot assigned to u needs 1 + V + V = 2V + 1 messages,
initiates the Hello message cascade at the beginning of the
where V denotes the average number of neighbors per node.
first window. Each node at a depth d receives Hello messages
Since we have N − 1 sensor nodes and each node notifies first
from its closest neighbors of the sink in the window d and
its priority and then its slot assignment to its conflicting nodes,
sends its own Hello message in the window d + 1 with a
we need 2 · (2V + 1) · (N − 1) = (4V + 2) · (N − 1) messages
random jitter to avoid collisions between nodes of the same
to establish the conflict-free schedule.
depth. After several exchanges of Hello messages checking
the symmetry of links, and if the neighborhood of the node is Hence, the centralized assignment outperforms the dis-
stable, each node sends a N otif y message to the sink. The tributed one in terms of number of required messages if and
N otif y message sent by node u 6= sink contains its depth, only if: (AverageDepth + K) · (N − 1) ≤ (N − 1) · (4V + 2).
its neighborhood and some applicative information (e.g. the
number of slots needed, the number of radio interfaces, etc.). Property 2: The centralized assignment requires less con-
The N otif y messages are processed by the sink acting as the trol messages than the distributed one if and only if K ≤
entity in charge of computing the joint time slot and channel 4V + 2 − AverageDepth, where V is the average number of
assignment. neighbors per node and AverageDepth is the average of the
depth of all nodes different from the sink and K is the number
5) Centralized versus distributed computation of a of fragments of the Schedule.
collision-free schedule: The collision-free schedule consists of
a sequence of tuples (sender, receiver, channel, time slot) that Finding a conflict-free schedule with the minimum number
is reproduced periodically. This schedule can be computed in a of slots has been proved NP-hard in a single channel network
centralized or distributed way. We now evaluate the number of for arbitrary topologies [21]. That is why heuristics are gener-
control messages needed by each of them, assuming a traffic- ally used to compute the time slot and channel assignment.
aware slot and channel assignment. Such an assignment is
chosen because it ensures the smallest data gathering delays. 6) Collision-free schedule: Since in our application, the
We first observe that both the centralized and the distributed condition of Property 2 is met, we use a centralized algorithm
assignments use: for a joint time slot and channel assignment: MODESA [14].
The conflicting nodes and the potential parents of any node
• Neighborhood Discovery, where each node discovers u 6= sink are computed from the collected one-hop neigh-
its neighbors and checks the symmetry of the links. borhs, according to Property 1. The routes used for collecting
• Routing tree construction, where the routing tree used data packets through the whole network are selected jointly
for data gathering is built by exchanging messages with the channel and slot assignment.
including the depth of the sending node. The depth of
a node represents its distance to the sink, this distance During the initial discovery of channels, we want to reduce
is expressed in number of hops. the latency of the first data gathering. That is why, just after the
discovery of the first channel, a first schedule is built, making
In a centralized assignment, each node u 6= sink, which possible data gathering. The schedule is rebuilt each time a
depth in the routing tree is Depth(u) transmits the list of its new channel is discovered to take advantage of parallel trans-
neighbors, including its parent and its children in the routing missions. After that, any change in the information contained
tree, and its traffic demand Gen(u) to the sink. This message in the N otif y messages may cause updates in the current
needs
P Depth(u) hops to reach the sink. In total, we have schedule or the creation of a new schedule.
u Depth(u) transmissions=AverageDepth · (N − 1) trans-
missions. Then, the sink computes the conflict-free schedule B. Illustrative example
and broadcasts it to all sensor nodes. Thus, the total number
of messages required to establish the conflict-free schedule in We consider a network with eight nodes and three chan-
centralized mode is AverageDepth · (N − 1) transmissions nels where, for simplicity reasons, the topology is the same
+ the transmissions to broadcast the Schedule to all nodes. on all channels and is depicted in Figure 2. Each node ∈
Let us assume that the message including the Schedule must {3, 4, 6, 7, 8} generates one data packet per slotframe, whereas
be fragmented into K fragments to be compliant with the nodes 2 and 5 generate two packets per slotframe. The sink
maximum frame size allowed by the standard MAC protocol. has three radio interfaces.

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on August 19,2023 at 02:13:14 UTC from IEEE Xplore. Restrictions apply.
 For this network example, where AverageDepth = 1.57 requirements. In this paper, after a state of the art of existing
and V = 3, the centralized assignment of time slots and multichannel WSNs, we propose an adaptive solution that
channels outperforms the distributed one as long as K ≤ 12.43 provides a multichannel conflict-free schedule of data trans-
fragments. missions. It also supports topology changes and traffic changes.
A more intensive performance evaluation will be conducted
1 using network simulators.

R EFERENCES
2 6 7
[1] E. R. Martinez, J. A. Santos, R. David, M. Mojarradi, L. del Castillo,
S. P. Jackson, Challenge of Developmental Flight Instrumentation for
4 Orion Exploration Flight Test 1: Potential benefit of wireless technology
5 3 8 for future Orion missions, IEEE WISEE, October 2014.
[2] IEEE Standard for Local and Metropolitan Area Networks-Part 15.4:
Fig. 2: An example of topology. Low Rate Wireless Personal Area Networks (LR-WPANs) Amendment 1:
MAC Sublayer, IEEE Std 802.15.4e-2012, IEEE, April 2012.
[3] IEEE Standard for Local and Metropolitan Area Networks-Part 11:
MODESA provides the schedule given in Table I. We Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY)
observe that the three channels are simultaneously used in Specifications, IEEE Std 802.11-2012, IEEE, March 2012.
slots 1 and 3. The three sink interfaces are active in slot 1: [4] Zigbee Specification, Document 053474r17, ZigBee Alliance, January
2008.
they are receiving from nodes 2, 6 and 7. In slot 2, we notice
[5] G. Chalhoub, A Guitton, F. Jacquet, A. Freitas, M. Misson, Medium Ac-
spatial reuse on channel 2: nodes 3 and 8, that are 4-hop away, cess Control for a Tree-Based Wireless Sensor Network: Synchronization
transmit simultaneously on the same channel. Management, IFIP Wireless Days, November 2008.
TABLE I: The schedule obtained with MODESA. [6] G. Huang, A. Y. Zomaya, F. C. Delicato, P. F. Pires, Long term and
large scale time synchronization in wireless sensor networks, Computer
Ch\S 1 2 3 4 5 6 Communications 2014, 37:77-91.
1 2→1 2→1 6→1 2→1 6→1 2→1
[7] Y. Wu, J. A. Stankovic, T. He, Sh. Lin, Realistic and Efficient Multi-
2 6→1 3→6 7→1 4→6 5→2 Channel Communications in Wireless Sensor Networks, IEEE Infocom,
8→7
3 7→1 5→2 April 2008.
[8] S. Lohier, A. Rachedi, I. Salhi, E. Livolant, Multichannel Access for
C. Adaptivity of the proposed solution Bandwidth Improvement in IEEE 802.15.4 Wireless Sensor Network,
IFIP/IEEE Wireless Days, November 2012.
The proposed solution supports any network topology that [9] B. Priya, S. S. Manohar, EE-MAC: Energy Efficient Hybrid MAC for
is connex on the control channel. The topology may vary WSN, International Journal of Distributed Sensor Networks 2013.
from one channel to another. Additional links, that are not [10] G. Zhou, Ch. Huang, T. Yan, T. He, J. A. Stankovic, MMSN: Multi-
used in the routing graph, may exist. These links may add Frequency Media Access Control for Wireless Sensor Networks, IEEE
interferences that are avoided by our solution. The number of Infocom, 2006.
radio interfaces of the sink is a parameter that may vary. To [11] J. Borms, K. Steenhaut, B. Lemmens, Low-Overhead Dynamic Multi-
cope with transmission errors, the number of slots granted to channel MAC for Wireless Sensor Networks, EWSN, February 2010.
each node is higher than the number requested, taking into [12] O. D. Incel, P. Jansen, S. Mullender, MC-LMAC: A Multi-Channel MAC
Protocol for Wireless Sensor Networks, Technical Report TR-CTIT-08-
account the error rate of the link considered. 61, Centre for Telematics and Information Technology, 2008.
This solution is able to adapt to topology changes. Af- [13] R. Diab, G. Chalhoub, M. Misson, Evaluation of a hybrid multi-channel
ter joining the network, a node discovers its neighbors and MAC protocol for periodic and burst traffic, IEEE LCN, September 2014.
among them its potential parents. It then sends a N otif y [14] R. Soua, P. Minet, E. Livolant, MODESA: an Optimized Multichannel
Slot Assignment for Raw Data Convergecast in Wireless Sensor Net-
message to the sink that allocates time slots and channels works, IPCCC 2012, December 2012.
for the transmissions originated at this node. This update of
[15] P. Minet, S. Mahfoudh, G. Chalhoub, A. Guitton, Node Coloring in
the schedule is then broadcast to all nodes to be applied a Wireless Sensor Network with Unidirectional Links and Topology
in the next slotframe. When a node leaves the network, the Changes, IEEE WCNC, April 2010.
conflict-free schedule is updated to free the slots allocated [16] T. Watteyne, A. Mehta, K. Pister, Reliability Through Frequency Diver-
to the transmissions originated at this node. If a link with sity: Why Channel Hopping Makes Sense, ACM PE-WASUN, October
a parent is broken, the conflict-free schedule is updated to 2009.
select another parent among the potential parents. If a new [17] K. Pister, L. Doherty, TSMP: Time Synchronised Mesh Protocol, PDCS,
link is created, the conflicting nodes are updated. If the sink November 2008.
detects no conflict in the current schedule, this schedule is [18] J. Polastre, J. Hill, D. Culler, Versatile Low Power Media Access for
Wireless Sensor Networks, ACM Sensys, November 2004.
kept. Otherwise, it is updated to avoid conflicts. Furthermore,
[19] G. Chalhoub, R. Diab, M. Misson, HMC-MAC Protocol for High Data
statistics about link quality may be used by the sink to select Rate Wireless Sensor Networks, Electronics 2015, 4(2):359-379.
the preferred parent among the potential ones. In addition, [20] G. Chalhoub, M. Misson, Cluster-tree based energy efficient protocol
changes in the application needs (e.g. generation of more for wireless sensor networks, IEEE ICNSC, April 2010.
traffic) are taken into account [22], leading to the update of [21] H. Choi, J. Wang, E. A. Hughes, Scheduling on Sensor Hybrid Network,
the existing schedule. IEEE ICCCN, October 2005.
IV. C ONCLUSION [22] R. Soua, E. Livolant, P. Minet, Adaptive Strategy for an Optimized
Collision-Free Slot Assignment in Multichannel Wireless Sensor Net-
Aircraft has strong requirements on throughput and latency. works, Journal of Sensor and Actuator Networks 2013, 2(3), 449-485.
Multichannel wireless sensor networks are able to meet such

Authorized licensed use limited to: J.R.D. Tata Memorial Library Indian Institute of Science Bengaluru. Downloaded on August 19,2023 at 02:13:14 UTC from IEEE Xplore. Restrictions apply.