Using Nano-wireless Communications in Micro-Robots

Applications

Nicolas BOILLOT Dominique DHOUTAUT Julien BOURGEOIS

Institut FEMTO-ST∗ Institut FEMTO-ST Institut FEMTO-ST
UMR CNRS 6174 UMR CNRS 6174 UMR CNRS 6174
Université de Franche-Comté Université de Franche-Comté Université de Franche-Comté
Centre National de Recherche Centre National de Recherche Centre National de Recherche
Scientifique (CNRS) Scientifique (CNRS) Scientifique (CNRS)
25200 Montbéliard, FRANCE 1 Cours Leprince-Ringuet 1 Cours Leprince-Ringuet
nicolas.boillot@femto- [email protected]@femto-
st.fr st.fr st.fr

ABSTRACT tions of nano-communications using carbon nano-tubes [14]

The emergence of nano-electromagnetic communications ba- or nano-wireless communications [5] while some others have
sed on graphene nano-antennas has opened new perspec- defined the basis of nano-wireless communications using gra-
tives for communications between small things, referred as phene nano-antennas [6]. These works have changed the
to the internet of micro-things or even as the internet of perspectives for communicating between small things. The
nano-things. However, these antennas make use of the Ter- internet of things (IoT) changed his scale to go micro with
ahertz band which raises many problems like the absorption the internet of micro-things [10] or even smaller with the
of entire range of the available bandwidth by any molecule. internet of nano-things [28].
Meanwhile, recent advances have been made in the design However, these antennas make use of the Terahertz band
and fabrication of micro-robots enabling formation of micro- which raises many problems like the absorption of entire
robots networks. Nano-antennas are an interesting way of ranges of the available bandwidth by interactions with all
communicating between micro-robots. We envision two types molecules [25] or proposing an energy-efficient physical layer
of benefits using integrated nano-antennas in micro-robots. [29, 40]. These properties are varying regarding to the en-
First, nano-wireless communications could enhance exist- vironment and the application. Absorption is indeed not
ing applications enabling a greater communication distance the same if the transmission is made in an open environ-
in an ensemble and broadcasting facilities. Second, nano- ment composed of air, in a body or inside an ensemble of
wireless communications can create new usage and new ap- micro-robots. The huge bandwidth of the Terahertz band
plications. This article presents a simulation framework for offers many different possibilities for the design of the phys-
micro-robots using nano-wireless communications and an ical layer and diversity of potential applications will have
application being developed within our simulator. different needs ranging from multimedia transmission [16,
28], micro-robots communications [9], distributed intelligent
micro-electro-mechanical-systems (MEMS) [11, 12], to re-
Keywords placement of bus for core to memory communications [3].
nano-wireless communications, simulation, MEMS micro- We therefore think that a cross-layer approach is needed
robots in the design of the communication layers of nano-wireless
communications.
Recent advances in the field of MEMS, are enabling the
1. INTRODUCTION design and fabrication of distributed intelligent MEMS (Di-
These last years, different works have opened new perspec- MEMS). A node in a DiMEMS system is basically com-
tives for nano-electromagnetic communications. First stud- posed of actuators, sensors, a processing unit and commu-
ies have defined the research challenges and possible applica- nication capabilities, which makes it a micro-robot. It is
∗
This work has been funded by the Labex ACTION pro- the interconnection of these micro-robots that triggers the
gram (contract ANR-11-LABX-01-01) and ANR/RGC (con- emergence of a global behavior. All the current projects
tracts ANR-12-IS02-0004-01 and 3-ZG1F) and ANR (con- on DiMEMS systems are using wired or communications by
tract ANR-2011-BS03-005) contact [13, 21]. If nano-antennas are integrated in DiMEMS
Permission to make digital or hard copies of all or part of this work for systems, it would be an interesting way of communicate be-
personal or classroom use is granted without fee provided that copies are not tween micro-robots. We envision two types of benefits using
made or distributed for profit or commercial advantage and that copies bear integrated nano-antennas within a micro-robots ensemble.
this notice and the full citation on the first page. Copyrights for components First, nano-wireless communications could enhance exist-
of this work owned by others than ACM must be honored. Abstracting with ing applications, enabling a greater communication distance
credit is permitted. To copy otherwise, or republish, to post on servers or to
in an ensemble as well as broadcasting possibilities. Sec-
redistribute to lists, requires prior specific permission and/or a fee. Request
permissions from [email protected]. ond, nano-wireless communications can create new usage
NANOCOM’ 14, May 13 - 14 2014, Atlanta, GA, USA and new applications. In a previous work [9], we have stud-
Copyright 2014 ACM 978-1-4503-2979-8/14/05 ...$15.00 ied the integration of wireless capabilities in micro-robots of
http://dx.doi.org/10.1145/2619955.2619967.
the Claytronics project, showing the enhancement created Unfortunately, these simulators don’t offer physics and real-
by wireless communications, but nano-wireless communica- world modeling: mobile network nodes are only 2D objects
tions was out of the scope of this former article. with no support for collision and so on. However, it would be
This article presents a simulation framework for micro- possible to plug an external simulator which could take care
robots using nano-wireless communications as well as several of the physics but the network simulation is too detailed
applications being developed within this simulator. for our needs which limits the scalability to thousands of
node. The same comments apply to other simulators like
SSFNet [35], OPNET [1], QualNet [2] or J-Sim [31]. Nano-
2. RELATED WORKS wireless simulation is still in an early stage. Preliminary
works have been done in Nano-Sim [37] a plugin of NS-3.
2.1 Modular micro-robots Although this simulator is a very interesting achievement,
The programmable matter (i.e. figure 1) field makes use Nano-sim does not model physics nor real-world constraints
of small robots that fall in the ”‘modular robots”’ category. and scalability is limited to few thousands of nodes. It does
Here, robots have scale ranging from a few centimeter (pee- neither take into account the radio propagation delay that
ble [20] down to millimeter size (Claytronics [21, 22]). Those can be of utmost importance at the considered scales.
small to very small robots attach (or detach) to form virtu- Many modular robots simulators have been designed over
ally any macro-structure they are required to. the years. Very first works are simulators which are ded-
icated to specific hardware like Molecubes [41] or Cross-
Cube [34] and therefore lack of genericity. More recently,
there have been more generic simulators proposed like Player
/Stage [19], gazebo [32], USSR [15] or MORSE [18] but net-
work access is not sufficiently detailed, as focus is made on
the robots themselves. The most interesting initiative is AR-
GoS [36] which includes its own network simulator or that
could be plugged to NS2 or NS3 [33]. ARGoS is fast and
can simulate up to 100,000 of robots. However, the target
of ARGoS is swarm robots and not really modular robots.
Finally, DPRSim appears to be unbeatable for the scala-
(a) Robot Pebbles (b) Claytronics catom pro- bility as the most efficient simulators report 100x less simu-
totype
lated number of nodes.

Figure 1: Programmable matter

3. CONTEXT
In the Claytronics project, MEMS micro-robots called 3.1 DPRSim
catoms (prototype photo 1) are covered with structures called
Dynamic Physical Rendering Simulator (DPRSim) [38, 7]
”features”. Those features are used as mean of both attach-
has been developed by Intel since 2006 for the Claytronics
ment (by electromagnetic or electrostatic force [30]) and di-
project of the Carnegie Mellon University.
rect communication.
The simulator is designed to support a potentially very
By actuating their features, catoms are able to move around
large number (up to several millions) of Claytronics micro-
their direct neighbors similarly as a stepper motor (see fig-
robots called catoms. DPRSim internally makes use of ODE
ure 2).
(Open Dynamics Engine, a rigid body dynamic library) and
can provide detailed physical simulation for its virtual world
(albeit activating physical simulation significantly reduces
the number of simulated elements). An optional 3D graphi-
cal interface is also provided through Drawstuff, the default
3D environment visualization tool included in ODE. For per-
formance considerations, DPRSim is written in C++ and
can also work in mono or multi-threaded mode.
In DPRSim, each catom is individually represented by
an object instance, and the code running on a catom is
called a ”CodeModule” which is also individually instanti-
ated. Someone designing a new Claytronics application has
Figure 2: Actuating the catoms like a stepper motor to write its own CodeModule and then load it into the sim-
(2 dimensional motion example) ulated catoms. CodeModules can be written in C++ but,
part of the claytronics project, is also the Meld language.
Meld is a declarative language aiming at easing the develop-
2.2 Wireless network simulation for modular ment of large scale distributed applications. One can thus
robots write Meld programs that are automatically converted into
There are many network simulators that are proposing CodeModules able to run in DPRSim.
wireless network simulations. The two most used in the re- In [9], we expanded DPRSIM by adding a wireless com-
search community are NS3 [23] and OMNeT++ [39]. They munication capability to the simulated catoms. Allowing
offer support for mobility and this characteristic could en- wireless transmission in such an environment open the path
able their usage for network simulation of modular robots. to new area of thrilling applications. This expansion to
DPRSIM had to integrate well with the inner working of sented in [5]. This article defines the different kinds of
the simulator, especially with the time-division system. nanonetworks media (nanomechanical, acoustic, electromag-
DPRSim uses a time-slicing approach, with an atomic du- netic and chemical or molecular) and defines the nanoma-
ration step (called a ”tick”), and processes the whole simula- chines in a similar way to Berlin in [8] but extended it to a
tion step by step. A tick is not a divisible duration and this bio-hybrid approach. In this article, the molecular commu-
has many repercussions, especially on the messaging system. nication is preferred to as the electromagnetic one. In fol-
If activated, the physical simulation (ODE library) uses lowing articles, Jornet, Akyildiz and al. present the concept
its own time-slicing - which is independent but has to be of CNT-based nano-antennas in [27] together with a first at-
configured conjointly with the main DPRSim’s time-slicing. tempt to define its characteristics [4]. In [24], a model of the
We developed a new simulator component called ”Vouivre” path loss is proposed using HIgh resolution TRANsmission
able to handle the radio channel and the its concurrent ac- molecular absorption database (HITRAN). HITRAN mea-
cess. Vouivre can be used as a standalone wireless network sure the absorption electro-magnetic radiation for different
simulator, but can also be used as a library in a slightly frequencies transmitted in different gas. Started in the 60s,
modified DPRSIM or other network simulator. Vouivre is with only seven gas in the infrared, they have now a wide
discrete-event based, and proposes an adaptable tradeoff be- range of gas and frequencies. The different molecules ab-
tween complexity and realism of the network. Going for mil- sorb the transmitted frequencies in discrete fashion which
lions of nodes sharing a medium is not possible at this time, means that each absorption can be calculated in isolation of
but it is still possible to tune it to go largely over the few the rest. The total absorption can therefore be computed
thousands most dedicated and detailed network simulators as the sum of absorption of the different molecules present
can do. on the transmission path. The noise, in the terahertz band,
To link DPRSim with our new network simulator, 3 ma- is mainly introduced by the molecular noise created by the
jors changes have been brought to DPRSim: absorption of the signal which is, in turn, exciting molecules.
The first one and most important for the work later pre- That is why short pulses should be preferred for transmit-
sented in this paper is to invoke at each Tick, a progression of ting information. These conclusions are confirmed and more
the simulated network time (see figure 3). From the point of detailed in [25]. In the mean time, the idea of nano-wireless
view of Vouivre, the duration of a DPRSim’s tick is defined sensor networks is investigated in [6] giving a first view of
as constant. It is important to configure this time jumping possible hardware and potential applications. Given that
duration to be the same as the duration of the intra-tick time short pulses would give better results, a new communica-
interval defined for the physics engine ODE. Consequently, tion scheme called Time Spread On-Off Keying (TS-OOK)
in the figure 3, the duration between t0 and t1 is the same is introduced in [26]. TS-OOK uses very short pulses in the
for Vouivre and ODE. Failing to do so, the network time range of femtoseconds where each pulse transmits a “1” and
would flow slower or faster than the physics time and the the absence of pulse transmits a “0” which is called on-off
data rates of network interfaces would not have sense any- keying. Pulses have to be transmitted using a period of time
more. During each tick processing, DPRSim lets Vouivre much bigger than the duration of a pulse. This ensure an
process the discrete events which have been scheduled for efficient medium sharing as well as reducing the possible in-
this time interval. From a different perspective, we should terferences and noise creation. It has to be noted that the
consider that the network simulation is periodically stalled transmission of “0” has to be preferred to “1” as it does not
to allow DPRSim to simulate CodeModules of catoms and create any perturbation of the channel and it is also pos-
to allow ODE to simulate physics. sible to detect it at a greater distance. Two amelioration
The second change was to deactivate the previous algo- of TS-OOK are proposed. First, using a low-weight chan-
rithm dedicated to communications on wired interfaces. The nel coding allows to reduce the influence of interference be-
existing mailbox system of DPRSim has still been preserved tween concurrent transmissions leading to better aggregated
to keep compatibility with already existing CodeModules bandwidth [26]. Second, Rate Division TS-OOK is proposed
and Meld applications. To avoid the synchronism barrier in- in [29] where the time interval between pulses can vary be-
duced by the time-slicing operating mode, the existing mail- tween transmitters. This is done to avoid series of collisions
box callback system have to be used in ”live mode” instead between two nodes. Indeed, in TS-OOK as the interval is
of ”delayed” mode. This allows for multiple small messages a pre-determined fixed value, if two nodes are emitting at
to be exchanged back and forth during the same DPRSim the same time, all their transmissions would in collision. To-
tick. A new type of interface has been added, as the Wire- gether with RD TS-OOk, a medium access control (MAC)
lessNetworkAdapter, which enables access to a shared ra- is proposed. PHysical Layer MAC Protocol for Electromag-
dio channel. NetworkAdapters corresponding to the normal netic nanonetworks (PHLAME) main proposition is a hand-
wired interfaces have also been modified to be managed by shaking protocol which allows multiple receptions, defines
our network simulator. allowing simulations of wired com- the interval length between two pulses and chooses the best
munication over shared mediums such as wired buses. channel coding scheme to increase transmission reliability.
The last important change is a mapping between DPRSim This coding scheme is chosen to match a pre-defined packet
structures representing catoms and the corresponding Vouivre error rate and consists in two mechanisms. First, the cod-
network nodes. Each network node can have several network ing is transformed reducing the number of “1”s in favor of
interfaces. They can be wired or wireless interfaces. This the “0”s taking into account that logical “1” will create per-
mapping is build at the creation of the DPRSim simulation turbation on the channel whereas “0”s does not. Further-
universe called ”CatomWorld”. more, “0”s can be detected at a further distance. Second,
a n-repetition code is used where each symbol is replicated
3.2 Nano-wireless communications n-times.The handshaking protocol consists in two phases: a
transmission request, starting the handshake and a trans-
The first idea of nano-wireless communications was pre-
 DPRSim timeline
Tick X Tick X+1
DPRSIM DPRSIM DPRSIM DPRSIM
Internal Advancing network Internal Advancing physics Internal Advancing network Internal Advancing physics
processing time (Vouivre). processing time (ODE). processing time (Vouivre). processing time (ODE).
(newTick…) (endTick…) (newTick…) (endTick…)
t0 t1 t1 t2
Vouivre
timeline
t0 t1 t1 t2
ODE timeline

Figure 3: Imbrication of Vouivre inside the DPRSIM and timelines

mission confirmation, confirming the handshake. Finally, over areas ranging from square centimeters to square meters)
in [40], a new MAC protocol is proposed aiming to optimize and because of the extremely small duration of the pulses,
the energy consumption in order to match the potential en- the signal propagation delay has to be taking into account.
ergy harvested. The propagation (3̃ nanoseconds per meter) delay can be
greater than the duration of a pulse and as such collisions
can easily occur. Figure 4 shows how symbols can overlap
4. NANO-WIRELESS COMMUNICATIONS from a receiver point of view even if they were not sent at the
IN DPRSIM same time. Sender S1 sends a ”1” (as a pulse) and S2 later
send a ”0” as an absence of pulse. Because receiver R1 is at
4.1 Implementation the same distance from S1 and S2, it received those symbols
In [9], we had already interfaced the time-slicing DPRSim one after the other and decodes them correctly. Receiver R2
simulator with our own discrete-event simulator called Vouivre. is much closer from S2 than from S1. As the propagation
In the current paper, we used this as a base to implement delay from S1 is longer, the two symbols overlap and the ”0”
nano-wireless communications into DPRSim. The way to from S2 is masked by the ”1” from S1.
imbricate the timelines (as shown on Figure 3) developed
for the previous version of Vouivre could be kept as is.
The timescale necessary to represent the events happening dBC

in nano-wireless communications has to be extremely small. S1 R1 S2

Nano-networks work in multiple THz wide band and the

duration of a TS-OOK pulse is bellow one picosecond.
This is not a problem for a discrete-event simulator such
as Vouivre. But the application code processing the received R2
messages is executing in DPRSim and is only called at each (a) Spatial positions
DPRSim TICK. This was causing an usually long latency
1
between the reception of a message and its processing. To
prevent this it was necessary to reduce the duration of a S1
Sent
0
TICK. As a side-effect, as DPRSim also call ODE (its physic S2
simulation engine) between each TICK to update position 1
and velocity, the simulation performances could have de- As ...
S1
graded a lot. To be able to correctly simulate TS-OOK seen ... 0
nano-communications and still have reasonably fast simula- by R1 S2
tion, we decided for a 100 picoseconds TICK duration as a 1 Lost
trade-off. Up to this duration is however added as a delay As S1
...
before processing a message. In most scenarii this can be seen
... 0
considered harmless. by R2 S2
In our previous work [9], we used the Friss propagation
model using a 2.4GHz centered frequency band. Such long (b) Chronogram
range radio transmitters were kept in Vouivre/DPRSim to
allow ulterior studies of hybrid systems. Figure 4: ”0” symbols can be masked by ”1” even if
Using nano-communications between nodes along with larger not sent at the same time
ones able to communicate over relatively long distance could
indeed pave to way to exciting applications. We imple- Moreover, catoms are mobile objects and the distances
mented new interfaces using TS-OOK over the same discrete- can vary over time.
event model we used before. The channel model can now be Figure 6 shows events as they are generated when catoms
changed on the fly. In the following work, only the TS-OOK represented on Figure 5 communicate. Catoms A and B
interface were used with logical ”1” being the first derivative transmit informations and points of view of A,B and C are
of a 100 femtosecond long Gaussian pulse and a duration represented on Figure 6. TS-OOK allows for the temporal
between symbols of 5 picoseconds. multiplexing of the radio channel. Multiple messages can
At the envisioned physical scale (nano-devices scattered be sent at the same time. As pulses are comparatively very
 DPRSim interruption DPRSim interruption
Vouivre timeline
... ... ... ...
(External (External
scheduling) Catom A is Catom A has 1 message to Catom A finished to scheduling)
queueing send ; Loading «Message 1» write «Message 1». (External
Discrete «Message from TX spooler ; scheduling) Catom B
events : 1» for Starting to write « Message dAB finished
Catom B : Starting to Features of
sending. 1» on the medium. to write
processing write «Message 2» on Catom B received catoms X and
and «Messag Z are
the medium. « Message1»
Running user’s e 2». disconnected.
scheduling
mailbox callbacks for
Catom B is receiving Catom B.
« Message1» Propagation delay
Transmission
between catoms A and C
Catom C received by feature
« Message1» failed on
dAC
Catom X.
Catom C is receiving
« Message1»

Message 1 (sending) Message may be received or lost

Catom A
depending on the quality of the
Message 2 (receiving) transmission.
dAB

Network
layer and
messages Catom B Message 2 (sending)
transmission Message
Message 1 (receiving)
progress dAB finally
C start reading a message taking received
longer than remaining Tick’s time. on the
next tick.
Catom C Message 1 (receiving)
dAC
Message 2 (receiving)
dBC

Legend
Internal Vouivre
Vouivre Event User event
Data receiving event Data sending timeline is
event processing scheduling
scheduling stalled

Figure 6: Discrete events scheduling and consequences on the network layer

the ’1’ (pulse). Even if this way of coding increases the size
of the message, its ability to correctly receive the message
dBC exceeds its drawbacks.
Catom C DPRSim is meant to simulate networks of sizes ranging
dAC
from a few catoms to hundred of thousands of them. Deter-
dAB mining if a packet has been received correctly or nor has to
be done in a practical way. We need to be able to scale to
large numbers, but still capturing the peculiar behavior of
Catom A Catom B the nano-wireless radio channel.
Also, in the programmable matter and Claytronics specific
Figure 5: Propagation delay of electromagnetic sig- context, because of the application and control layers, there
nal depends on the distance between catoms. will be large variations in the communication needs over
time. Sometimes almost no communications are required as
the network reached a very stable state. Sometimes only a
short to the time between them, the probability of a collision few catoms may want to transmit, and sometimes events or
is small. Moreover as seen previously on Figure 4, only a ”1” user commands may trigger large and immediate urge for
can mask a ”0”. The ”1” are said to be dominant and the communications. These requirements are for example dif-
”0” recessive. If the symbol being received for the current ferent from those of nano-sensors networks, were the global
packet is a ”1”, other concurrent ”1” or ”0” will not directly network load is much more constant over time.
affect it. Only when receiving a ”0”, if a ”1” is transmited at To avoid computing the propagation of each pulse we only
the same there will be a data corruption. compute changes in the number of concurrent transmissions
For each receiver, a table is maintained with all incom- affecting the reception of an incoming packet. Those changes
ing messages. Each receiver gets a unique table as the re- define one to multiple periods for this incoming packet which
ceived packets are affected by the propagation delay previ- are affected by a coefficient we get from a precomputed table.
ously mentioned. The precomputed table has been generated from the sta-
As previously explained, the radio channel used for nano- tistical model of the network capacity done by Jornet and
wireless is very peculiar. In particular, a transmitted sig- Akyildiz.This model takes into account the number of con-
nal is not only impacted by a continuous attenuation with current transmissions along with the distance from the sender
the increasing distance, but also by the noise caused on the and the generated dataset is represented on Figure 7. The
channel by itself and other concurrent transmissions. We path-loss and noise in the Terahertz Band are computed by
have also seen that, especially when the number of concur- using models introduced in [24, 25] A standard medium with
rent transmissions is high (from thousands to hundred of 10% of water vapor is considered. This model takes into ac-
thousands), reducing the weight of the coding brings better count the number of concurrent transmissions along with
performances. Reducing the weight of the coding means in- the distance from the sender and the generated dataset is
creasing the proportion of ’0’ symbol (i.e. no pulse) against represented on Figure 7. As the number of catoms in the
 be received elsewhere. Using TMR in this context makes
the assumption that during a communication the network
condition will not vary dramatically. This assumption en-
sures the equiprobability of loosing a bit during the three
redundant transmissions of the message. Having different
network conditions during one message transmission will be
done in a future work.

4.2 Multi hop flooding in a dense environment

In the following scenario, an initial catom will send a 53
bytes packet to its neighbors. This packet will then be re-
transmitted hop by hop until the whole network has been
covered. To better illustrate the behavior of the protocol,
receiving this packet will trigger the coloring of the catom
on the screenshots. The color itself will depend on the last
re-transmitter. To prevent an everlasting flooding, the mes-
sage contains a sequence number and can by re-transmitted
only one time per catom.
Figure 7: Single-user achievable information rate in
bit/symbol as a function of the number of nano-
devices and the transmission distance. Matlab
model by Jornet and Akyildiz

following experiments is kept relatively small (up to a few

hundreds), we did not used a variable weight coding. Instead
we used an equal probability of ”0” and ”1” as it is mostly
optimal for low numbers of concurrent transmission. An
higher (thousands and above) numbers would have required
an adaptative coding, with catoms sensing the activity (or
being informed by some authority or centralized system) and
choosing a more appropriate weight.
Figure 8 shows the point of view - affected by the propa- Figure 9: A radio asynchronous multi-hop flooding
gation delay - of a node simultaneously receiving three mes- starting from the lower right corner in a ”checker-
sages. The impact of concurrent transmissions on the vari- board 20*20” macro-structure of 400 catoms.
ous parts of the messages has to be computed as the average (Sim1)
number of bits correctly received. During the first (small)
period, only ”message 1” is being received and got a factor Figure 9 shows 6 phases of retransmission on a square area
of 1 from the channel capacity table. Don’t forget that this covered by catoms. From phase to phase, we easily see how
table takes into account the distance and the equivocation new catoms are reached during each retransmission.
effects of the message itself on the channel ... getting a factor We consider submillimeter catoms of 0.06mm of radius.
of 1 is possible, but even alone on the channel, depending The used matlab nano-wireless communication model tar-
on the distance it could be less. geting ranges in the order of a few centimeters is conse-
The second period sees two concurrent transmissions, which quently ideally suited in terms of transmission power.
get different coefficients because the distance to the trans-
mitter is not the same. The third period sees more trans-
mitters and consequently even smaller factors.
When the reception of a packet ends, a global factor for
this packet is computed proportionally from the factors of
its individual periods.
On such an unfavorable environment, the probability for
a symbol to be lost or corrupted is usually great. Cod-
ing schemes adding resilience are necessary, but can’t get (a) Simu1: all (b) Simu2: 1 node (c) Simu3: 1 node
too complex because of limited processing power on nano- nodes are allowed among 9 can relay among 16 can relay
to relay
devices.
In order to reduce the number of errors, we use a Triple
Modular Redundancy (TMR) which increases threefold the Figure 10: 3 simulations by varying the number of
size of the messages. When the reception of a packet ends, a relays. (catoms allowed to relay are black at the
correction factor computed is applied. If a tolerance thresh- beginning of simulation)
old is met, the message is dispatch to the mailbox of the
catom. Is this threshold is not reached, the packet is lost In this scenario, in order to further reduce the number of
for this receiver. But as each catom will conduct this pro- retransmissions, only black catoms are allowed to retrans-
cedure within its own environment, the message could still mit. We performed three simulations with different relay
 ... Message 1

... Message 2

... Message 3

Numbers of
concurrent 1 2 3 2 1
senders

Bits received ...

Bit/symbol for
message 1, 1.0 0.9 0.9 0.7 0.7
0.85 0.85 0.85 0.65 0.65
message 2,
0.7 0.7 0.7
message 3
Period A B C D E

Figure 8: Calculating the number of bit received per message by using the information rate.

configurations as shown by figure 10.

Please note that the number of catoms (a 20x20 matrix)
and the nano-wireless range are voluntary kept small to en-
hance the readability of the pictures. But the benefits from
wireless retransmission in such data broadcasting applica-
tion s would only grow with increasing range and proper
retransmission management.

Figure 12: Number of catoms simultaneously send-

ing on the medium

in the Simu1 curve, corresponding to the 270 simultaneous

transmitters as seen on Figure 12 at the exact same time.

Figure 11: Progression of the colorization wave in

the time.

The cumulative propagation over time is presented on Fig-

ure 11. As excepted, propagation is faster in simulation 2
and simulation 3 than simulation in 1 (it takes less time
for all of the 400 catoms to correctly get the broadcasted
packet). As we will see, the simulation 1 (in which all catoms
re-transmits) suffers from collisions. As expected also, to
number of simultaneous transmitters is much higher in sce-
nario 1. We can also observe that the number of concurrent Figure 13: Messages lost due to collisions
radio transmitters follows a pattern induced by the topol-
ogy of the network. As the time pass, the data propagation The statistic channel model we used (Figure 7) incorpo-
”wave” reach the widest area of the square and the number rate multiple characteritics of the channel and does not al-
of retransmitters is maximum. It then decreases as the mes- lows for a direct distinction between packets lost because
sage further propagate in the opposite corner (Figure 12). of collisions or for other reasons. Figure 13 shows the ratio
As we can see on Figure 7, the capacity of the channel de- of packets lost due to collisions over the overall number of
pending on the number of concurrent transmissions remain sent packets. The number of collisions is obtained by run-
almost maximal with up to around 200 transmitters, and ning again the scenarii in a special mode were the effects of
then experiences a strong decrease. This is perceivable on the concurrent transmissions are cancelled. The difference
Figure 11 at time=20 microseconds, with a clear inflection in packets losses thus can be attributed to the collisions.
We can observe that only the first simulation is affected by 2013, IEEE Int. Conf. on Internet of Things, pages
collision, this is of course due to the high number of retrans- 638–645, Beijing, China, Aug. 2013.
mitters. [10] J. Bourgeois and S. Goldstein. The internet of
[micro]-things. Keynote talk at iThings 2011, the 2011
IEEE Int. Conf. on Internet of Things. Dalian, China,
5. CONCLUSION Oct. 2011.
We presented a simulation framework for electromagnetic [11] J. Bourgeois and S. Goldstein. Distributed intelligent
nano-wireless network in the Terahertz band. This simulator mems: Progresses and perspectives. In L. Kocarev,
called Vouivre is plugged in a modular robots simulator en- editor, ICT Innovations 2011, volume 150 of Advances
vironment called DPRSim. In order to simulate the commu- in Intelligent and Soft Computing, pages 15–25.
nications we integrated the model proposed by Jornet and Springer Berlin / Heidelberg, 2012.
Akyildiz. We precomputed the transmission parameters to
[12] J. Bourgeois and S. C. Goldstein. Distributed
allow fast simulations and we took into account the propaga-
intelligent mems: Progresses and perspectives. IEEE
tion delay which can completely change the perturbations.
Systems Journal, 2013.
We keep the simulation as accurate as possible by count-
ing the number of simultaneous communications so as the [13] K. Boutoustous, G. J. Laurent, E. Dedu, L. Matignon,
probability of reception could be modified according to this J. Bourgeois, and N. L. Fort-Piat. Distributed control
number. One problem remains, as nano-wireless uses very architecture for smart surfaces. In IEEE/RSJ
short periods of time, the physics has to be cut in short pe- International Conference on Intelligent Robots and
riods of time as well slowing down the simulation. Our main Systems (IROS), pages 2018–2024, Taipei, Taiwan,
future work will be to use a complete discrete-event simula- October 2010. IEEE.
tor which will be able to manage nano-communications and [14] S. F. Bush, Y. Li, Y. Li, S. F. Bush, and Y. Li.
physics as events. The development of this simulator called Nano-communications: A new field? an exploration
VisibleSim [17] has already begun. into a carbon nanotube communication network, 2006.
[15] D. Christensen, D. Brandt, K. Stoy, and U. Schultz. A
unified simulator for self-reconfigurable robots. In
6. ACKNOWLEDGMENTS Intelligent Robots and Systems, 2008. IROS 2008.
We would like to thank Dr. Josep Miquel Jornet and IEEE/RSJ International Conference on, pages
Prof. Ian Akyildiz for sharing with us their models of the 870–876. IEEE, 2008.
Terahertz band and Dr. Hakim Mabed for is insight on [16] E. Dedu, J. Bourgeois, and M. Agus-Zainuddin. A first
Matlab and for helping getting the right parameter setup study on video transmission over a nanowireless
for TS-OOK. network. In NANOCOM 2014, 1st ACM Int. Conf. on
Nanoscale Computing and Communication, pages 1–4,
7. REFERENCES Atlanta, Georgia, USA, May 2014. ACM. Poster.
[17] D. Dhoutaut, B. Piranda, and J. Bourgeois. Efficient
[1] http://www.opnet.com/products/modeler/. simulation of distributed sensing and control
[2] Qualnet simulator. environments. In iThings 2013, IEEE Int. Conf. on
http://web.scalable-networks.com/content/qualnet. Internet of Things, pages 452–459, Beijing, China,
[3] S. Abadal, E. AlarcoÌAn, A. Cabellos-Aparicio, Aug. 2013.
,
M. Lemme, and M. Nemirovsky. Graphene-enabled [18] G. Echeverria, N. Lassabe, A. Degroote, and
wireless communication for massive multicore S. Lemaignan. Modular open robots simulation
architectures. Communications Magazine, IEEE, engine: Morse. In Robotics and Automation (ICRA),
51(11):137–143, 2013. 2011 IEEE International Conference on, pages 46 –51,
[4] I. Akyildiz, J. Jornet, and M. Pierobon. Propagation may 2011.
models for nanocommunication networks. In Antennas [19] B. Gerkey, R. Vaughan, and A. Howard. The
and Propagation (EuCAP), 2010 Proceedings of the player/stage project: Tools for multi-robot and
Fourth European Conference on, pages 1–5, 2010. distributed sensor systems. In Proceedings of the 11th
[5] I. F. Akyildiz, F. Brunetti, and C. BlÃazquez. international conference on advanced robotics,
,
Nanonetworks: A new communication paradigm. volume 1, pages 317–323. Portugal, 2003.
Computer Networks, 52(12):2260 – 2279, 2008. [20] K. Gilpin, A. Knaian, and D. Rus. Robot pebbles:
[6] I. F. Akyildiz and J. M. Jornet. Electromagnetic One centimeter modules for programmable matter
wireless nanosensor networks. Nano Communication through self-disassembly. In IEEE International
Networks, 1(1):3 – 19, 2010. Conference on Robotics and Automation (ICRA),
[7] M. P. Ashley-Rollman, P. Pillai, and M. L. Goodstein. pages 2485–2492., 3–7 May 2010.
Simulating multi-million-robot ensembles. In ICRA, [21] S. C. Goldstein, J. D. Campbell, and T. C. Mowry.
pages 1006–1013, 2011. Programmable matter. IEEE Computer, 38(6):99–101,
[8] A. Berlin and K. Gabriel. Distributed mems: New June 2005.
challenges for computation. IEEE Computational [22] S. C. Goldstein, T. C. Mowry, J. D. Campbell, M. P.
Science and Engineering Journal, 4(1):12–16, March Ashley-Rollman, M. De Rosa, S. Funiak, J. F. Hoburg,
1997. M. E. Karagozler, B. Kirby, P. Lee, P. Pillai, J. R.
[9] N. Boillot, D. Dhoutaut, and J. Bourgeois. Efficient Reid, D. D. Stancil, and M. P. Weller. Beyond audio
simulation environment of wireless radio and video: Using claytronics to enable pario. AI
communications in mems modular robots. In iThings Magazine, 30(2), July 2009.
[23] T. Henderson, S. Roy, S. Floyd, and G. Riley. ns-3 and M. Dorigo. Argos: a modular, parallel,
project goals. In Proceeding from the 2006 workshop on multi-engine simulator for multi-robot systems. Swarm
ns-2: the IP network simulator, page 13. ACM, 2006. Intelligence, 6(4):271–295, 2012.
[24] J. Jornet and I. Akyildiz. Channel capacity of [37] G. Piro, L. A. Grieco, G. Boggia, and P. Camarda.
electromagnetic nanonetworks in the terahertz band. Nano-sim: Simulating electromagnetic-based
In Communications (ICC), 2010 IEEE International nanonetworks in the network simulator 3. In
Conference on, pages 1–6, 2010. Proceedings of the 6th International ICST Conference
[25] J. Jornet and I. Akyildiz. Channel modeling and on Simulation Tools and Techniques, SimuTools ’13,
capacity analysis for electromagnetic wireless pages 203–210, ICST, Brussels, Belgium, Belgium,
nanonetworks in the terahertz band. Wireless 2013. ICST (Institute for Computer Sciences,
Communications, IEEE Transactions on, Social-Informatics and Telecommunications
10(10):3211–3221, 2011. Engineering).
[26] J. Jornet and I. Akyildiz. Low-weight channel coding [38] B. D. Rister, J. Campbell, P. Pillai, and T. C. Mowry.
for interference mitigation in electromagnetic Integrated debugging of large modular robot
nanonetworks in the terahertz band. In ensembles. In ICRA, pages 2227–2234, 2007.
Communications (ICC), 2011 IEEE International [39] A. Varga and R. Hornig. An overview of the omnet++
Conference on, pages 1–6, 2011. simulation environment. In Proceedings of the 1st
[27] J. M. Jornet and I. F. Akyildiz. Graphene-based international conference on Simulation tools and
nano-antennas for electromagnetic techniques for communications, networks and systems
nanocommunications in the terahertz band. In & workshops, Simutools ’08, pages 60:1–60:10. ICST,
Antennas and Propagation (EuCAP), 2010 2008.
Proceedings of the Fourth European Conference on, [40] P. Wang, J. M. Jornet, M. Abbas Malik, N. Akkari,
pages 1–5. IEEE, 2010. and I. F. Akyildiz. Energy and spectrum-aware mac
[28] J. M. Jornet and I. F. Akyildiz. The internet of protocol for perpetual wireless nanosensor networks in
multimedia nano-things. Nano Communication the terahertz band. Ad Hoc Networks,
Networks, 3(4):242–251, December 2012. 11(8):2541–2555, 2013.
[29] J. M. Jornet, J. Capdevila Pujol, and J. Solé Pareta. [41] V. Zykov, P. William, N. Lassabe, and H. Lipson.
Phlame: A physical layer aware mac protocol for Molecubes extended: Diversifying capabilities of
electromagnetic nanonetworks in the terahertz band. open-source modular robotics. In IROS-2008
Nano Communication Networks, 3(1):74–81, 2012. Self-Reconfigurable Robotics Workshop, 2008.
[30] M. E. Karagozler, S. C. Goldstein, and J. R. Reid.
Stress-driven mems assembly+ electrostatic forces=
1mm diameter robot. In Intelligent Robots and
Systems, 2009. IROS 2009. IEEE/RSJ International
Conference on, pages 2763–2769. IEEE, 2009.
[31] J. Kačer. J-Sim – a Java-based tool for discrete
simulations. In Proceedings of the 23rd International
Autumn Colloquium ASIS-2001: Advanced Simulation
of Systems, pages 135–141, Náměstı́ Msgre Šrámka 6,
70200 Ostrava, Czech Republic, September 2001.
MARQ.
[32] N. Koenig and A. Howard. Design and use paradigms
for gazebo, an open-source multi-robot simulator. In
Intelligent Robots and Systems, 2004.(IROS 2004).
Proceedings. 2004 IEEE/RSJ International Conference
on, volume 3, pages 2149–2154. IEEE, 2004.
[33] M. Kudelski, M. Cinus, L. Gambardella, and
G. Di Caro. A framework for realistic simulation of
networked multi-robot systems. In Intelligent Robots
and Systems (IROS), 2012 IEEE/RSJ International
Conference on, pages 5018–5025, 2012.
[34] Y. Meng, Y. Zhang, and Y. Jin. Autonomous
self-reconfiguration of modular robots by evolving a
hierarchical mechanochemical model. Computational
Intelligence Magazine, IEEE, 6(1):43 –54, feb. 2011.
[35] D. M. Nicol, B. Premore, and A. Ogielski. Using
simulation to understand dynamic connectivity at the
core of the internet. In Proceedings of UKSim 2003,
Cambridge University, England, April 2003.
[36] C. Pinciroli, V. Trianni, R. O’Grady, G. Pini,
A. Brutschy, M. Brambilla, N. Mathews, E. Ferrante,
G. Caro, F. Ducatelle, M. Birattari, L. Gambardella,