A Combined Routing Layer for Wireless Sensor Networks and Mobile Ad-Hoc

Networks

Tobias Senner, Reinhardt Karnapke, Andreas Lagemann, Jörg Nolte

Distributed Systems/Operating Systems Group
Brandenburg University of Technology
Cottbus, Germany
{tsenner, karnapke, ae, jon}@informatik.tu-cottbus.de

Abstract shows an example of such a communication. The firefight-

ers have entered the building from 3 different entrances and
In the near future, first responders such as firefighters planted sensor nodes on the way. Now they meet in the
might be supported by sensor nodes, both installed before center, and try do determine a safe route back out. Three of
the incident and placed by the responders themselves. To them request temperature values from the nodes they placed
communicate with these sensor nodes, the first responders at the entrances. For simplicity the following communica-
could be equipped with hand held devices, e.g. PDAs with tion back to the firefighters is not shown here.
wireless LAN, additionally equipped with a radio module of Another application would be the communication among
the same type as used by the sensor nodes. This scenario the firefighters. Currently this communication and the com-
enables the usage of the PDAs as mobile communication munication between a firefighter and the operation con-
backbone by the sensor nodes, as well as allowing the first trollers is very lossy. If things go bad, and a house is about
responders to access nodes in the sensor network in an effi- to collapse, the controllers send an evacuation signal. But
cient way, both near and far. This paper presents a weighted sometimes, this signal is not received by all of the firefight-
routing protocol for heterogeneous networks, which could ers, leading to danger of injuries and sometimes even death.
be used in such a scenario.
In our approach the evacuation order would be sent over
multiple networks - the currently used one, the wireless
LAN of the PDAs and the sensor network. The specialty
1 Introduction is, that the communication can switch between networks at
any time. If specified, a message of high priority like an
In emergency scenarios such as fires or earthquakes, evacuation order received by a PDA will be flooded into the
quickly deployed sensor networks can lend vital support to sensor network as well as the wireless LAN. This way, even
first responders like firefighters and rescue teams. When firefighters who are not in direct communication range with
first responders arrive at the scene, they should be able any other can be reached through multiple sensor network
to communicate with all sensor nodes in their immediate hops. As this of course drains energy from the sensor nodes,
vicinity, or even nodes farther away, depending on the in- a multiple network flooding should only be used in extreme
tended use. An application for the former would be fire- cases, when it is absolutely vital to reach all firefighters. But
fighters, who measure the heat levels in the immediate area the situation mentioned before is of course such one, as the
they are moving to, using sensors that have been installed in life of the firefighters might be at stake. Figure 2 shows an
the building previously. An application for the latter would example for such a communication. The PDA in the left
be finding the way out of the burning building for the same upper corner just received an abort mission signal, which is
firefighters. If they dropped a few sensors on the way in, then flooded through both networks. All PDAs within range
they could determine a safe route back. When they reach receive it, and the corresponding firefighters evacuate. The
the trapped people, they would request the heat levels mea- firefighters carrying the PDAs in the right corner would nor-
sured by the nodes at the entrances/exits of the building. If mally not have received the signal, but are now evacuating,
the values returned from some sensors are too high, or there too, as the signal was relayed by the sensor nodes.
is no answer at all, they would know that this route was This paper is structured as follows: The weighted routing
not safe anymore and choose a different way out. Figure 1 protocol is described in section 2, while section 3 shows the
 Figure 1. Communication from PDAs to designated WSN nodes

and GPS receivers are still too expensive, if nodes should

be used in large quantities. Other localization algorithms
are not accurate enough or take too much time. A data cen-
tered approach is not useful either. A firefighter that looks
for the way out of a burning building will not be interested
in the heat levels from all nodes in the building. Rather, only
the nodes that are near a certain exit or on the way there are
of interest. Therefore, we choose to use ID-based address-
ing. When the firefighters drop nodes, their ID is recorded
automatically, or even preprogrammed before entering.

Support for Heterogeneous Networks and Architectures

Due to the scenario, communication must be possible be-
Figure 2. Emergency flooding through multi-
tween hand held devices and sensor nodes. We assume that
ple networks
the sensor nodes are equipped with a cheap communication
module which supplies only basic functionality with little
bandwidth and low data rate. The PDAs on the other hand
results of the simulations and experiments we made. Re- use wireless LAN to communicate among each other and a
lated work can be found in section 4. We finish with con- zigbee radio module to communicate with the sensor nodes.
clusion and future work in section 5. This leads to a number of problems that have to be solved,
e.g. the different size of basic data types (8 or 16 bit for an
2 A Routing Protocol for Mobile Heteroge- integer) or the internal representation of bytes (big or little
neous Networks endian).

2.1 Requirements Ad-Hoc Topologies and Mobility While the Sensor

nodes are assumed to be immobile, the firefighters are mov-
Our scenario produces a number of requirements for the ing through the burning building (that is their job after all).
routing protocol: This means that the logical topology, which denotes the
topology determined by radio neighborhood, is constantly
ID-based Addressing To communicate with nodes in the changing. While the connections between sensor nodes are
immediate vicinity a local broadcast is enough. To commu- assumed to be fairly static, the connections between sen-
nicate with distant nodes, a geographic approach might be sor nodes and firefighters’ PDAs change often, as do those
considered good. But GPS does not work inside buildings, between firefighters.
Low Overhead for the Sensor nodes All sensor nodes 2.3 The implemented Approach
currently available have severe resource limitations. This is
valid for the memory (e.g. 10kB of Ram and 48 kB of Rom To minimize the overhead in the sensor network, a flat
on a TMote Sky) as well as for the communication band- address space was chosen. The MANET nodes use the same
width and the energy reserves. All these make designing address type as the sensor nodes. Because the PDAs have
a routing protocol for wireless sensor networks much more stronger batteries and a longer range, most of the communi-
challenging then designing one for traditional networks. cation should take place in the MANET. In our implementa-
tion the connections between nodes are weighted with a cost
Limited Unicast and Broadcast To enable efficient col- metric, which can be individually tuned. Figure 3 shows an
lective operations it is necessary to supply the ability to ad- example where each connection involving a sensor node is
dress a single node within a certain range of hops (Unicast) assigned a cost of 2, and pure MANET connections a cost
and to distribute messages to all nodes within a certain num- of 1. In this example, the node in the upper left of the sensor
ber of hops (n-hop Broadcast). For efficiency purposes it is net needs to communicate with the one in the lower right.
also useful to enable different forms of broadcast for the As there is a MANET nearby, this can be used as a commu-
PDAs. As they could distribute a broadcast over different nication backbone. The communication through the sensor
networks, it is possible to transmit only into one of them network would involve 6 nodes for a cost of 12, whereas
(WSN/Manet) or into both. the communication through the MANET involves 2 sen-
sor nodes , 2 switches between the network and 3 MANET
Tolerance for Asymmetric and Unidirectional Links hops, leading to a total cost of 4+4+3 = 11. The difference
Experiments with real networks e.g. [12] have shown that becomes stronger when the difference of costs is higher.
asymmetric and even unidirectional links are quite common
in wireless sensor networks. Therefore, a way of dealing
with them has to be found.

2.2 Different Approaches to Connect Dif-

ferent Networks

There is a number of different ways to connect heteroge-

neous networks. The most prominent method is inter net-
working as used in the internet. In this approach, there can
be any number of different networks (autonomous systems,
AS) with different routing strategies, that are connected.
But this approach requires all nodes to have an IP-Address.
For wireless sensor networks, this approach has been shown
to be inefficient, because of the overhead of transmitting a
large IP-Header with every message, whereas the payload Figure 3. Routing in a weighted heteroge-
is often only a few bytes. neous network
Another solution would be to use IP-Addresses for the
PDAs, and simple IDs in the sensor network (WSN-ID).
This would require the usage of gateways for address trans- To further reduce the network load in the sensor network,
lation or heterogeneous addressing, where each sensor node the addresses are divided into sensor identities and MANET
needs to know the IP-Addresses of the PDAs it is communi- identities. This way, a node always knows in which network
cating with as well as the WSN-ID of its surrounding nodes. the destination can be found. This is also useful to limit the
This would lead to a lot of memory consumption on the range of multi hop broadcasts inside either the MANET or
sensor nodes. Also, in the case of communication between the sensor network.
WSN and PDAs, the IP-Address would have to be transmit- In some cases, like the afore mentioned emergency evac-
ted, in the worst case through the whole sensor network. uation, a spreading through both networks is wanted. To
A third possibility is to use a unified addressing scheme enable all these different scenarios, the addressing scheme
for both networks, that is based on WSN-IDs. In this case, depicted in figure 4 is used. The most significant bit decides
the IP-Addresses of the PDAs would be ignored, and the if one or both interfaces should be used on a MANET node,
unified addressing used on top of raw WLAN broadcast. on the sensor nodes only one is available. The second bit
This has the advantage of reducing the overhead in the is used as identifier for the network the destination can be
WSN. found in. The address of the node takes up the rest of the
 3.2.3 WSN-MANET-MANET-WSN
Table 1. Single Hop Scenario results
packets delivery ratio RTT min avg max The third row of experiments featured communication be-
100 94% 103 ms 299 ms 537 ms tween two Tmotes, using a MANET tunnel in between. In
figure 12 the sensor node on the left side wanted to transmit
to the senor node on the right side. As they were not within
Table 1 shows the results of our measurements. The 6% range, communication was only possible using the WLAN
packet loss can easily be explained by collisions between of the laptops in between.
route request- and acknowledgement packets. These exper-
iments were only meant to show that communication was
possible at all, but also already hint at a possible problem:
due to the usage of the Tmote as modem for the laptop, the
difference between minimal and maximal round-trip-time
(jitter) is very high. Figure 12. Tunnelling WSN packets through a
MANET

3.2.2 MANET-MANET-WSN
Table 3 shows the achieved packet delivery ratio and
The setup of the second row of experiments featured the round-trip-time. Note, that for all experiments no MAC
Tmotes and two laptops, one of which again used another layer was used. While this was no problem for those de-
Tmote as modem. The layout of the experiment can be seen picted above, it was devastating for this setup. The rate of
in figure 11. The laptop on top wanted to communicate with 86% was only achieved by disabling the data packets, be-
the sensor nodes, using the laptop on the middle as a gate- cause they kept colliding with the next route request.
way.
Table 3. Results of the tunnel scenario
packets delivery ratio RTT min avg max
100 86% 310 ms 514 ms 770 ms

3.2.4 WSN Multihop

In this last row of experiments only Tmotes were used,
which were arranged in a row of 2, 3 or 4 nodes. Table
4 shows the results of these experiments. As suggested ear-
lier, the communication between Tmotes is fairly fast, and
the jitter is fairly low. Each hop seems to add about 30 ms
to the round trip time.

Figure 11. Ping-Pong scenario setup

Table 4. Measured multihop round trip times
The results of these experiments are depicted in table 2. hops delivery ratio RTT min avg max
It can be seen that he additional MANET hop increased the 1 97% 20 ms 29 ms 30 ms
round-trip-time by at least 95 ms, up to 284 ms. This seems 2 81% 50 ms 56 ms 60 ms
to imply (as will be shown later) that the transition between 3 94% 70 ms 87 ms 100 ms
networks and the communication on the laptop takes much
more time than that between Tmotes.
4 Related Work
Table 2. Measured round trip time Ping-Pong The idea of adding firefighters with sensor networks has
Scenario been proposed in different ways. Sometimes the sensors are
packets delivery ratio RTT min avg max used to find trapped people or to monitor the health status of
the firefighters themselves [4]. Siren [2] uses a tuple-space
100 92% 198 ms 498 ms 821 ms
like abstraction to exchange data about measured heat val-
ues between firefighters. The main difference between their
approach and our is that we determine the heat levels from a References
distance, whereas in theirs one of the firefighters must have
received the values directly, before forwarding them to his [1] M. Brzozowski, R. Karnapke, and J. Nolte. Impact - a fam-
colleagues. ily of cross-layer transmission protocols for wireless sensor
The authors of [7] introduced TEEN and its successor networks. In The First International Workshop on Research
Challenges in Next Generation Networks for First Respon-
APTEEN [8]. Teen is a hierarchical cluster based routing
ders and Critical Infrastructures (NetCri 07), in conjunction
protocol for wireless sensor networks. In this protocol, sen-
with 26th IEEE IPCCC, 2007.
sors are gathered in groups (so called clusters), with one of [2] X. Jiang, N. Chen, K. Wang, L. Takayama, and J. Landay.
them as leader (cluster-head). Communication inside these Siren: Context-aware computing for firefighting.
groups always takes place between one of the nodes and the [3] R. Karnapke and J. Nolte. Copra - a communication process-
cluster-head. Communication between different clusters is ing architecture for wireless sensor networks. In Euro-Par
possible only from one cluster-head to another. In our ap- 2006 Parallel Processing, pages 951–960. Springer, 2006.
proach the higher cost assigned to the hops between sensor [4] K. Koumpis, L. Hanna, and S. Hailes. Tunnels of ter-
nodes leads to more of the communication taking place be- ror. IEE Computing and Control Engineering Magazine,
tween the PDAs than between sensor nodes. If the number Dec/Jan 2005/06.
[5] M. Krüger, R. Karnapke, and J. Nolte. In-network process-
of PDAs is high enough, and there is always a PDA within
ing and collective operations using the cocos-framework. In
reach of any sensor node that wants to transmit, the PDAs 12th IEEE Conference on Emerging Technologies and Fac-
would behave like the cluster-heads in TEEN. tory Automation, 2007.
The routing protocol AODV, which was modified for this [6] A. Lagemann and J. Nolte. Csharpsimplemodule – writing
work was introduced in [11]. It was intended as a protocol omnet++ modules with c# and mono. In OMNeT++ 2008,
for MANETs, and is in its original form too heavy-weighted Marseille, March 2008.
[7] A. Manjeshwar and D. P. Agrawal. Teen: A routing pro-
to be used in sensor networks. However, with the modifica-
tocol for enhanced efficiency in wireless sensor networks.
tions we made, we were able to adapt it to run on a Tmote
In Proceedings 15th International Parallel and Distributed
Sky without running into memory or energy problems. Processing Symposium., pages 2009–2015, 2001.
Many other routing protocols for sensor networks use [8] A. Manjeshwar and D. P. Agrawaly. Apteen: A hybrid proto-
different metrics. These include among others the hop col for efficient routing and comprehensive information re-
count, the minimal energy consumed along the way or the trieval in wireless sensor networks. In Proceedings Interna-
remaining energy of each node [14, 10], or the load on par- tional, IPDPS Parallel and Distributed Processing Sympo-
ticular nodes, to name but a few. Our approach also keeps sium, pages 195–202, 2002.
[9] S. Mank, R. Karnapke, and J. Nolte. An adaptive tdma based
in mind the energy constraints on the sensor nodes, but not
mac protocol for mobile wireless sensor networks, best pa-
focused on the single sensor nodes. Rather, our approach per award. In International Conference on Sensor Technolo-
tries to shift as much traffic as possible to the much stronger gies and Applications, 2007.
(larger battery, higher bandwidth, greater range) PDAs. [10] O. Moussaoui and M. Naı̈mi. A distributed energy aware
routing protocol for wireless sensor networks. In PE-
WASUN ’05: Proceedings of the 2nd ACM international
5 Conclusion and Future Work workshop on Performance evaluation of wireless ad hoc,
sensor, and ubiquitous networks, pages 34–40, New York,
NY, USA, 2005. ACM.
In this paper we have presented a combined routing [11] C. E. Perkins and E. M. Royer. ”ad hoc on-demand distance
protocol for sensor networks and WLAN based MANETs. vector routing.”. In Proceedings of the 2nd IEEE Workshop
on Mobile Computing Systems and Applications, New Or-
The proposed approach uses weighted connections between
leans, LA, pages 90–100, FEB 1999.
nodes to determine which path to take. We have deliberately [12] Turau, Renner, and Venzke. The heathland experiment: Re-
chosen a simple metric that is not hard to compute and can sults and experiences. In Proceedings of the REALWSN’05
easily be implemented on resource constraint sensor nodes, Workshop on Real-World Wireless Sensor Networks., Jun
too. The results of our simulations show the advantages of 2005.
this approach and the feasibility of the approach has been [13] A. Varga. The omnet++ discrete event simulation system.
demonstrated using a prototype implementation on Tmote In Proceedings of the European Simulation Multiconference
Sky sensor nodes and laptops. However, more research with (ESM’2001), Prague, Czech Republic, June 2001.
a larger number of nodes is clearly necessary. In the future, [14] Y. Yu, R. Govindan, and D. Estrin. Geographical and energy
aware routing: A recursive data dissemination protocol for
we would like to combine this routing protocol with the col-
wireless sensor networks, 2001.
lective operations of C OCOS [5], a high level middleware
layer that provides data parallel operations on entire groups
of sensor nodes.