Impact of Time Slot Adjustment on a Multi-hop and Multi-channel
Solution for Dynamic WSN Topologies
Honoré Bizagwira
, Joël Toussaint
and Michel Misson
Université Blaise Pascal, Aubière, France
Université d’Auvergne, Aubière, France
LIMOS - CNRS, F-63175 Aubière, France
Wireless Sensor Networks, Adaptive Reception Window, Schedule-based Protocol, Dynamic Topology, Harsh
Ensure data delivery and extend the network lifetime are challenges that must be addressed when it comes to
deploy a WSN in outdoor harsh environment where it operates over unstable links. The instability of radio
links may induce a connectivity that is time-varying even when nodes are not moving. For these dynamic
topologies, resources allocated to the nodes must be adjusted to the local traffic conditions. Amongst the
proposed solutions, schedule-based protocols achieve an energy efficiency as they allow nodes to sleep during
their inactive periods. As the channel conditions are time-varying, and paths followed by the packets are
changing, inducing an instability in the distribution of the traffic load over the network. This paper is dealing
with the mastering of the size of the reception time slot of the next hop node. Our solution dynamically takes
in consideration the local traffic load to resize the time slots. This adaptive solution is compared with the
fixed-length window scheme, it obviously improves the performance of the WSN in terms of cycle length, idle
listening and collision reduction, and achieves suitable data delivery ratio.
Wireless sensor network (WSN) stands as a power-
ful solution for wide area environmental monitoring.
Sensor nodes equipped with data sensing device au-
tonomously organize themselves to deliver sampled
data to a particular node, the sink. As sensor nodes
have limited battery power, the transmission range
and the data rate have to be optimized according to the
deployment conditions. These characteristics impose
many challenges, especially for WSNs with fluctuat-
ing conditions of signal propagation. In this paper,
we focus on convergecast traffic in WSNs having dy-
namic topologies. It is the case for WSNs deployed
in harsh outdoor environment such as over freshwater
areas (lakes for example). For the WSNs we are deal-
ing with, the positions of nodes remain fairly stable
over time, but radio links may be volatile.
Our solution consists in a scheduling-based MAC
protocol where each node is assigned both time slots
and radio frequency for reception. In this proto-
col, each node selects periodically its next-hop node
among the neighbors according to the radio links
characteristics in order to route the traffic. Since we
deal with dynamic topologies, neighborhood of each
node is periodically updated. Instability of links has
an impact on the amount of traffic a particular node
has to route. In the paper we propose and evaluate
a mechanism to adjust the size of the receiving time
slots. We show that, coupled with a multi-channel so-
lution, it has a great impact on the time needed for
gathering data over such a WSN.
In this section, we investigate meanly the solutions
that have been proposed for dynamic topologies on
one hand and for TDMA-based MAC Protocols for
WSNs on the other hand.
When wireless links are intermittent, routing of
packets may become uncertain and/or induces large
end-to-end delays. WSN nodes are usually densely
deployed in order to ensure the global connectivity
and the coverage of the monitored area. Connectiv-
ity may be lost when some radio links on the path of
packets are temporarily disrupted. This variability of
Bizagwira H., Toussaint J. and Misson M.
Impact of Time Slot Adjustment on a Multi-hop and Multi-channel Solution for Dynamic WSN Topologies.
DOI: 10.5220/0006214701960201
In Proceedings of the 6th International Conference on Sensor Networks (SENSORNETS 2017), pages 196-201
ISBN: 421065/17
2017 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
the topology has a negative impact on the multi-hop
forwarding process and induces unpredictable end-to-
end delays. If this variability of topology happens
when some nodes are moving, the concept of data
mule can be used (Tseng et al., 2013). The goal of the
studies in (Nour Brinis, 2012) is to determine the opti-
mal number of data mules required to extend the cov-
ered network lifetime while meeting a given packet
delivery delay deadline.
In our contribution the hypothesis of having sev-
eral moving nodes is not necessary. Received or lo-
cally produced packets are stored in the node buffer
used with FIFO policy. When the radio link of the
next-hop is temporary broken, packets have to wait
for better propagation conditions or another next-hop
has to be chosen. The amount of traffic reaching
at a given node is depending on (i) the offered load
submitted by the source nodes, (ii) the instantaneous
shape of the WSN topology. Using a static capacity of
packet reception for nodes may lead to congestion ar-
eas and to FIFO overflow. This particular point leads
us to specify a dynamic capacity of reception. Our
objective is to adapt the size of the time slot dedicated
to reception of frames, to the local instantaneous load.
The scheduling-based MAC protocol we chose has to
be fitted in order to take into account the impact of
the dynamicity of the WSN topology. The use of a
unique or several frequencies has no significant im-
pact on this challenge. The question we are dealing
with here is: how to right-size the length of the re-
ception period without adversely impacting the traffic
flow through the network.
The problem of Time Slot Scheduling has been
deeply investigated for ad-Hoc Wireless Network. In
the WSN domain when TDMA is used, the con-
vergecast strategy requires both adequate coordina-
tion among the nodes and proper sizing of time slots
to avoid high packet collision rates or frequent FIFO
overflows near the sink node. Slot assignment and
slot adjustment are the two approaches usually ex-
plored (Pal and Chatterjee, 2014).
In (Cui et al., 2004), a variable length Time Di-
vision Multiple Access scheme is proposed for im-
proving total energy consumption. In (Cui et al.,
2005), delay and energy are taken into account to pro-
pose an algorithm for variable length TDMA sched-
ules. Most of the time, the available contributions are
driven by energy and delay considerations, for WSN
topologies supposed to be rather stable. Slot assign-
ment becomes very complex when the topology is re-
ally time-varying. The scheme allocation as presented
in (Ergen and Varaiya, 2010) seems to be proposed
for topologies having bidirectional and stable radio
links. These hypotheses are usually retained for slot
assignment using a coloring process (Mahfoudh et al.,
2010). Radio link quality measurements we done for
transmission over freshwater (Bizagwira et al., 2014)
justify another approach.
The protocol stack we propose has to be energeti-
cally efficient for WSNs having an unstable topol-
ogy. This objective leads us (i) to retain a cross lay-
ering approach for Network layer and MAC Layer,
(ii) and to use periodical sleeping periods to spare en-
ergy. The physical layer is inspired from the spectrum
allocation and channel specifications proposed in the
433 MHz frequency band by DASH7 Alliance (Tuset-
Peiró et al., 2014; Mode, 2013). This choice allows us
to have 8 non-overlapping channels for transmission.
Sleep period
Sleep period
Beacon frame
Single channel
A cycle
Figure 1: Mapping of time slot allocation within a cycle for
both single and multiple channel use.
3.1 Global Segmentation of Time
The operating cycle of the low layers of our proto-
col stack is divided into four periods as shown in Fig-
ure 1:
Synchronization and Build up Period : For multi-
hop deterministic synchronization of nodes using
Time guard period: Time guard allowing our so-
lution to be adaptive.
Data gathering period: For multi-hop forwarding
of data frames to the sink.
Sleeping Period: All the nodes sleep periodically
for energy saving.
Synchronization, neighborhood discovery and
routing strategy used to join the sink are done at
the beginning of each cycle using a beacon cascad-
ing. Synchronization is initiated by the sink that
Impact of Time Slot Adjustment on a Multi-hop and Multi-channel Solution for Dynamic WSN Topologies
broadcasts a beacon including the reference clock, the
length of current cycle and other information related
to the reception schedule.
3.2 Main Characteristics of MAC
The data gathering period is governed by a TDMA
schedule-based MAC protocol. It is a synchronous
and receiver-based protocol that applies the schedul-
ing decided from information obtained during the last
synchronization and build-up period. The neighbor-
hood has to be discovered again at the beginning of
each cycle. The results (as the quality of the radio
link) is used to choose the best next-hop to reach the
sink. The 8 frequency channels defined by DASH7
provides a multi-channel capability for data transmis-
The choice of channel scheduling is done dur-
ing the beacon cascading. When a node receives a
beacon, it reads the sender schedule and updates its
neighborhood knowledge table. Then, it selects its
next hop for data transmission before sending, in turn,
its own beacon. The neighborhood knowledge table
allows each node to choose a reception frequency and
time slot, taking into account the local use of the ra-
dio spectrum. As shown in Figure 1, the beginning
of the receiving Time Slot is deliberately chosen to
accelerate the convergecast traffic. The quality of the
path towards the sink is the main factor impacting the
choice of the next hop.
For WSNs using convergecast, the nodes near the
sink have more traffic to store and forward than other
nodes in the network, and nodes located at the pe-
riphery of such networks have no forwarding activity.
Moreover, when a WSN operates using unstable links,
the traffic load of a given node is also time-varying. In
this section, we state the problem of dynamic resizing
of reception time slots of nodes in accordance with
the traffic of their potential child-nodes. The global
objective is to adjust dynamically the time slot size
of each node, to the instantaneous local load. As this
length is adapted to the traffic, (i) the wasted energy
due to the idle listening will be reduced, (ii) the risk of
FIFO overflow will be minimized and (iii) the length
of the sleeping period can be maximized.
4.1 Global Presentation of the Master
Process of the Time Slot Duration
Dynamic topologies using unreliable radio links in-
duce a kind of packet routing instability and a variable
risk of packet retransmission. This drawback is miti-
gated by the part played by the FIFO of the nodes. In
a given area the cumulative number of packets stored
in the local FIFOs is a local load indicator. This is the
parameter we retained as the main variable to master
the size of the time slot of nodes. In the closed loop
control system that we propose for the adjustment of
the time slot duration of the node i, our objective is to
empty the FIFOs of its child nodes for each cycle.
4.2 Specification of the Algorithm Used
for Resizing the Reception Period
Let us consider node N
a generic node of our WSN.
The Process Variable (PV ) is the number of packets
remaining in the FIFO of the neighboring nodes that
have chosen N
as the next hop. The Set Point (SP)
is corresponding to the state we want to reach for the
content of these FIFOs at the end of the receiving time
slot of N
. SP stands for the number of packets having
to wait for another cycle to be transmitted.
e(t) = PV SP (1)
represents the error we want to minimize by a
mastering process. S
is used to reduce the impact
of proportional component of the mastering process
when the local congestion is growing too fast.
In the Algorithm 1, SP = 1 packet. PV is used
to accumulate the remaining packets in the FIFOs of
nodes having N
as their next-hop. This accumula-
tion is bounded by the threshold S
. Let Trx
the length of current reception period of N
. It is re-
sized according to: Trx
+ K
e(t). 5 and
5 packets are empirically chosen, respectively for K
and S
. The value of Trx
is constraint within the
interval [Trx
]: interval bounded by the
initial value for the time slot, and a maximum value
allowing us to estimate the length of PA or PB of fig-
ure 1. Let ID
be the address of the node having
chosen N
as next hop, N
, and L
be respec-
tively the number of packets remaining in the FIFO
of node x and the list of packet sender nodes for the
current cycle.
SENSORNETS 2017 - 6th International Conference on Sensor Networks
while reception period not expired do
if data packet received then
records ID
and N
if ID
already exist in L
update the N
append (ID
, N
) to
PV 0
foreach neighbor node j in L
if PV > S
e(t) PV 1
+ K
Algorithm 1: The algorithm that is used for resizing the
reception period length.
We use NS-3 to simulate our proposal scheme and in-
corporate protocol in several scenarios. In this sec-
tion, we discuss our WSN protocol stack implementa-
tion and give the detail settings. In practice, this WSN
has been implemented on a TI CC430 based com-
ponent to be deployed on the surface of water, i.e.,
over the lac for monitoring the aquatic environment.
All the MAC layer and routing algorithms presented
above have been implemented in our NS-3 model. We
chose to use the Log-Distance Shadowing model ex-
perimental measurement (Bizagwira et al., 2014) as
our propagation model. The choice of parameter val-
ues of the model is justified by (Tuset-Peiró et al.,
2014; Tuset et al., 2013).
In this section, we evaluate the performance of the
proposed dynamic reception window resizing scheme
and its consequences on the network throughput as
well as energy efficiency of the sensor nodes. We
conduct two different simulation scenarios and com-
pare the performance based on three metrics: packet
delivery ratio, average energy consumption, and aver-
age latency. In the first scenario, we set a fixed inter-
val for reception period - as is done in SMAC (Sen-
sor MAC) protocol (Chang et al., 2013), for all nodes
in the network. Each node schedules its transmission
time according to the chosen next hop, and fixes the
duration of the reception period. In the second sce-
nario, we apply our dynamic resizing scheme for both
small and large size reception period. Table 1 shows
the parameters used in our simulation. The network
topology consists of 26 nodes, which are spread ran-
domly through the network field of 300 m x 300 m as
shown in figure 2. Each node generates 1 data packet
per cycle except for the sink node (green colored) that
collects the traffic. The cycle duration is 4 seconds.
Figure 2: Network topology.
Table 1: Simulation parameters.
Parameters Value
Prop. model Log-Distance Shadowing
Number of nodes 26
Topology Random
Size 300 m x 300 m
Channel Single, multiple (8 non-
overlapping frequencies)
Cycle duration 4 seconds
Simulation duration 600 seconds
Buffer length 45 packets
Data rate 1 data packet per node and
per cycle
5.1 Fixed-length Reception Period
As we explained before, the sink node triggers the net-
work activity by sending a new beacon as the start sig-
nal of the beginning of a cycle. For this scenario, the
sink includes in the beacon the fixed duration of re-
ception period. The beacon cascading process broad-
casts this value to all others nodes of the WSN.
Impact of Time Slot Adjustment on a Multi-hop and Multi-channel Solution for Dynamic WSN Topologies
5.2 Adaptive Length Reception Period
All the nodes are initialized with a same reception
period length of 27 ms. The figure 3 is an example
showing how the window size adjustment is carried
out over time. Our test utopology is given in the fig-
ure 2.
(a) (b)
Figure 3: The adjustment over time of the reception periods
for the sink neighbor nodes.
The test scenario consists in turning off node B
during a given amount of time (between 150
cycles). We are looking its impact on the win-
dow size of nodes C and E.
5.3 Discussions
As introduced earlier, we analyze the performance of
our MAC protocol scheme in terms of (i) Packet deliv-
ery ratio (PDR), and (ii) activity duration as defined
in figure 1 (length of the cycle - sleeping period PD).
It is important to emphasize that - in order to com-
pare both cases, fixed-length and adaptive reception
periods, on the same basis, we set the length value for
the first as the maximum allowed value for the second.
Packet Delivery Ratio. The ratio of packets that
are successfully delivered to the sink compared to
the number of packets that have been sent out by the
source nodes.
The results for single channel and multi-channel
are quite similar. For small values (35 and 70 ms),
the window size is not sufficient to allow all traffic,
especially around the sink where there is a bottle-
neck. This induces a large number of packets dropped
due to lack of space in node FIFOs. Concerning the
window size for our adaptive method, we also ob-
serve that for nodes closer to the sink, the max-size
is quickly reached and the size of the reception period
never diminishes.
For large window values, as it is illustrated in the
figure 4, the major part of the traffic is delivered to the
sink. A 100% delivery of packets is never reached be-
cause some nodes in the network periphery are too
far apart to have a permanent connectivity. So the
packet delivery rate for these nodes is relatively low.
There are also always collisions in congestion areas
and packets never reach the destination despite re-
On the other hand, the results are quite similar be-
cause (i) the mechanism for choosing the start dates
of the reception period ensures that nearby nodes win-
dows can be temporally disjointed, (ii) the CSMA/CA
mechanism is sufficiently effective to transmit frames
within the set time, (iii) the topology is sufficiently
sparse to have only few collisions.
35 70 105 140 175 210
P acket delivery ratio (%)
35 70 105 140 175 210
Maximum reception period length [ms]
Figure 4: Packet delivery ratio.
We note also that the delivery rate in the case of a
fixed window size can sometimes be slightly higher:
this is due to the transient phase at the beginning of
the simulation process. For the adaptive method, sim-
ulation process starts with nodes having a small re-
ception period (27 ms) which does not allow flowing
as much traffic as in the situation where the sizes of
the windows are fixed and larger. This also explains
the small difference that we have, for larger size win-
The period adjustment process has a transient pe-
riod during which the adaptive algorithm increments
the length of the reception period before reaching
the convenient value. During this time, nodes would
accumulate packets, some are later dropped when a
FIFO overflow.
Activity Duration. The activity duration is the time
between the date of the wake up of the first node and
the date of the switching to the sleeping mode of the
last node of our topology. As it is illustrated in the
figure 5, the average activity duration is lower with
the adaptive mechanism than with the fixed-length
scheme. In the first case, nodes spend around 50%
of their time in sleeping mode. This is an important
feature, as it would enable reducing the duty-cycle
and makes the protocol more reactive to the topology
changes. We note that the adaptive mechanism of the
window size associated with multi-channel scheme is
highly effective from an energy saving point of view.
SENSORNETS 2017 - 6th International Conference on Sensor Networks
35 70 105 140 175 210
Average activity duration [ms]
35 70 105 140 175 210
Maximum reception period length [ms]
Figure 5: Average activity duration within working cycle.
Despite the transitional phase of setting up the net-
work that induces FIFO overflows and slightly longer
end-to-end delays, the data delivery ratios at the sink
are equivalent and sometimes better.
In this paper, we proposed an adaptive approach based
on a schedule-based MAC protocol, to efficiently uti-
lize the resources of WSNs, especially deployed in
harsh environment and having an unstable topology.
Most of protocols proposed to deal with dynamic
topologies argue on the efficiency of the schedule-
based MAC protocol. They ensure an adequate deliv-
ery ratio while optimizing energy consumption is still
a challenging task. Idle listening, overhearing and
collision are the main sources of energy wasting in
WSNs. The presented approach uses the traffic infor-
mation to right-size the reception period and therefore
to minimize the idle listening. Moreover, it employs
multi-channel scheme to enable parallel transmission
within the network and consequently, it reduces both
collision and overhearing.
The results clearly show that it significantly im-
proves the performance in terms of average activity
duration while providing, at the same time, a good
packet delivery ratio. It also reduces by a half the la-
tency compared to traditional schemes. This meets
the requirements for our target application where the
nodes will communicate over intermittent radio links.
The global solution we are dealing with is able to
withstand the effect of unstable topologies in an in-
teresting energy-saving manner.
Bizagwira, H., Toussaint, J., and Misson, M. (2014). Ex-
perimental protocols and testbed for radio link quality
evaluation over the freshwater. In 2014 IFIP Wireless
Days (WD), pages 1–4. IEEE.
Chang, L.-W., Huang, Y.-M., and Lin, C.-C. (2013). Per-
formance analysis of s-mac protocol. International
Journal of Communication Systems.
Cui, S., Goldsmith, A. J., et al. (2004). Cross-layer opti-
mization in sensor networks with energy constraints.
IEEE Trans. on Wireless Communications.
Cui, S., Madan, R., Goldsmith, A., and Lall, S. (2005).
Joint routing, mac, and link layer optimization in sen-
sor networks with energy constraints. In IEEE Inter-
national Conference on Communications, 2005. ICC
2005. 2005, volume 2, pages 725–729. IEEE.
Ergen, S. C. and Varaiya, P. (2010). Tdma scheduling al-
gorithms for wireless sensor networks. Wireless Net-
works, 16(4):985–997.
Mahfoudh, S., Chalhoub, G., Minet, P., Misson, M., and
Amdouni, I. (2010). Node coloring and color conflict
detection in wireless sensor networks. Future Internet,
Mode, D. A. (2013). An Advanced Communication System
for Wide-Area Low Power Wireless Applications and
Active RFID. Morgan Hill, CA, USA.
Nour Brinis, Pascale Minet, L. A. S. (2012). Edgm: En-
ergy efficient data gathering with data mules in wire-
less sensor networks. The Sixth International Confer-
ence on Sensor Technologies and Applications.
Pal, P. and Chatterjee, P. (2014). A survey on tdma-based
mac protocols for wireless sensor network. Interna-
tional Journal of Emerging Technology and Advanced
Engineering, 4(6):219–230.
Tseng, Y.-C., Wu, F.-J., and Lai, W.-T. (2013). Oppor-
tunistic data collection for disconnected wireless sen-
sor networks by mobile mules. Ad Hoc Networks,
Tuset, P., Adelantado, F., Vilajosana, X., Gallego, F. V., and
Alonso-Zárate, J. (2013). On the use of the 433 mhz
band to improve the energy efficiency of m2m com-
munications. In PIMRC, pages 2813–2818. IEEE.
Tuset-Peiró, P., Anglès-Vazquez, A., López-Vicario, J.,
and Vilajosana-Guillén, X. (2014). On the suitabil-
ity of the 433 mhz band for m2m low-power wire-
less communications: propagation aspects. Transac-
tions on Emerging Telecommunications Technologies,
Impact of Time Slot Adjustment on a Multi-hop and Multi-channel Solution for Dynamic WSN Topologies