ELEKTROTEHNIŠKI VESTNIK 79(4): 153-156, 2012 
ERK 2012 CONFERENCE ISSUE 
 
Simulation Model for Ant-Based Control Algorithm in Wireless 
Mesh Networks 
Erik Pertovt, Kemal Ali č, Ale š Š vigelj, Mihael M oh or č ič 
Institut "Jožef Stefan”, Jamova cesta 39, 1000 Ljubljana, Slovenija 
E-pošta: erik.pertovt@ijs.si 
 
 
Abstract 
In this paper, we present implementation of the ant-
based control algorithm (ABC) for load balancing in 
connection-oriented routing in the wireless mesh 
network (WMN) using the OPNET Modeler simulation 
tool. We investigate the time required for the initiali-
zation phase of the ABC algorithm within which the 
shortest path is found as well as network topologies 
with a different number of nodes and different number 
of neighbours per node. The resuts show that the 
algorithm initialization time increases with the number 
of nodes and decreases with the number of neighbours 
per node. 
1 Introduction 
In communication networks, heuristic/agents-based 
routing algorithms can be used to implement the 
adaptive routing functionalities capable of making the 
routing aware of dynamic changes of the network rather 
than being oblivious to them [1]. In this context, several 
routing algorithms have been developed inspired by the 
behaviour of the ant colonies and most of them are 
summarized in [2, 3]. Ants are social insects that 
collaboratively perform complex tasks, which can not 
be performed by one particular ant itself. In simulations, 
ants/agents are produced in the form of packets, which 
are used for network optimization and management. 
 In this paper, we present the implementation of the 
main reference ant optimization algorithm for 
connection-oriented networks, named Ant-based control 
(ABC) [1, 4, 5], in OPNET Modeler [6] simulation 
environment with some additional functionalities that 
can be used as an option. The algorithm is implemented 
in our network coding simulation model, presented in 
[7, 8]. We demonstrate the performance of the 
simulation model through the investigation of the 
initialization phase of the algorithm in wireless mesh 
network (WMN), which represents an underlying study 
for our work on routing being aware of network coding 
opportunities in WMNs. In the initialization time, the 
algorithm has to find, as quick as possible, the shortest 
paths in the sense of the minimum hop count from all 
nodes to all nodes, as Dijkstra optimal routing algorithm 
does, which is used as a reference. Several simulation 
runs are performed with different network topologies to 
evaluate the performance of ABC algorithm. 
2 Ant Algorithm 
In [1, 4, 5], the ABC algorithm was introduced as the 
first ant optimization algorithm applied to a routing 
problem. It was used to balance the network traffic of 
telephone calls in a telecommunication network. The 
purpose of ants in the ABC algorithm is to find paths 
with little hops and nodes with spare capacity.The paths 
are then used as routes to establish new connections. 
2.1 Algorithm description   
Ants are generated and lauched from all the nodes in the 
network at every simulation time step of the simulation, 
synchronously. The destination of each ant is randomly 
selected. Ants select their next node based on the 
pheromone tables located in the nodes. In pheromone 
tables, pheromone values represent probabilities with 
which ants select the next node. Higher the probability 
is, higher the chance of the node selection. 
 When an ant traverses a link between nodes A and B 
in the direction from A to B, it “lays” on it pheromone, 
which will be used by ants moving in the opposite 
direction of the pheromone laying ant, i.e. from node B 
to A. Pheromone laying is applied in the algorithm as 
ants updating the pheromone tables. When the ant from 
source node S travelling from node A to the next node 
B, reaches node B, it updates that part of the pheromone 
table of node B that concerns ants with destination S 
travelling from node B to A. The updating is done in 
such a way that the probability of choosing node A as 
the next node by ants with destination S is increased, 
while the probabilities of choosing other possible next 
nodes from node B is decreased. The increase and 
decrease of pheromone values (applied in the form of 
probabilities) is done according to equations in [1, 4, 5]. 
In the equations, pheromone decay (which corresponds 
to the pheromone evaporation in reality) is applied as a 
function of ant hops representing the ant age. In 
correspondence to this, ants which traversed many hops, 
will deposit less pheromone resulting in lower increase 
of corresponding probabilities. It means that paths with 
more hops will be selected with lower probability as the 
routes for connections. 
 For considering the degree of node congestion, ants 
are virtually delayed at nodes as a function of the node 
spare capacity. Lower the percentage of spare capacity, 
greater the delay at the node and later the update of the 
pheromone table by ants, which has for the consequence 
that the node will be selected with lower probability as 
part of the routes for connections.  
 Routing tables are built based on the pheromone 
(ant-decision) tables. For a destination D, next node B is 
selected at node A if pheromone value for B in 
pheromone table in node A is greater than for other 
neighbours of A concerning destination D. 

154 PERTOVT, ALIČ, ŠVIGELJ, MOHORČIČ 
 In one type of the algorithm, there is also a minority 
group of ants which select their next node randomly to 
prevent blocking problem by extreme node congestion 
or node failure, and to keep suddenly appearing better 
routes to be discovered more rapidly. 
2.2 Initialisation 
Before network operation begins and connection 
requests are placed in the network, routes from each 
source node to each destination node in the network 
have to be discovered by ants in the initialization phase.  
 At the beginning, all the pheromone tables are 
initialized with equal probabilities for neighbour nodes, 
meaning that each first ant in each node will select its 
next node among the neighbours with equal probability. 
Ants are not virtually delayed, as there is still no 
connection traffic in the network. The only influence on 
the routing are thus pheromone tables in the nodes. As 
pheromone tables are updated by ants as described 
above, the shortest paths (i.e. the minimum hop count) 
from sources to destinations will be discovered. As 
shown in [4], this scheme really tends to produce the 
shortest paths and gives a good enough result within the 
fixed period of the initialization phase time compared to 
a deterministic optimal algorithm as Dijkstra.  
3 Implementation of the Ant Algorithm in 
OPNET   
The ABC algorithm was implemented as ANT module 
in OPNET Modeler [6] simulation environment [7, 8] 
with all the algorithm functionalities. The state diagram 
of the implementation is presented in Figure 1. In 
addition, different types of the algorithm were 
introduced, which are desribed in Section 3.1. 
 
In the ANT module, the red state represents the idle 
state, where the module waits for one of the interrupts. 
There are six main transitional states: (i) init state takes 
care of the algorithm parameters initialisation at the start 
of the simulation run; (ii) New ANT state creates and 
lauches new ants; (iii) Route ANT state receives ants 
coming from neighbouring nodes, updates pheromone 
tables based on the received ants pheromone 
information, virtually delays ants based on the 
percentage of the node’s spare capacity, and selects the 
next node for ants; (iv) Update Route state updates 
routing tables based on pheromone tables every time 
pheromone tables are updated; (v) Write Hops state 
computes the instantaneus shortest paths from sources 
to destinations based on routing tables created by ants 
every time step during the algorithm initialisation phase; 
(vi) Hops Dijkstra state compares the shortest paths (i.e. 
optimal paths) found by Dijkstra to the corresponding 
shortest paths found by ants during the algorithm 
initialisation phase. 
3.1 Propagation delay and ants dying 
In our simulation, the ant movement from node to node 
is not determined by the simulation time steps. The time 
an ant is sent from a node to its neighbour is determined 
by the time the ant is waiting in the node’s queue. In 
addition, we add to the present ABC algorithm some 
other functionalities as optional representing different 
types of the original algorithm. 
 In the first type of the ABC algorithm, we consider 
that an ant dies if it has performed a specific number of 
hops on the path. It means that ants which do not find 
their destination within a predefined number of hops are 
eliminated from the network. These ants are inefficient 
and do not really contribute to finding the shortest paths. 
All other functionalities are preserved from the ABC 
algorithm, presented above. We denote this type of the 
ABC algorithm as ABCdie. 
 In the second type of the ABC algorithm, besides all 
the ABC algorithm functionalities, we consider also 
propagation delay on the links between the nodes. This 
type of ABC algorithm considers the real network 
scenario case, where ants require the propagation time 
to travel on wireless links. We denote this type as 
ABCprop. The main property ABCprop introduces is 
that shorter links are preferable than longer links. On 
longer links, packets require more time to propagate 
through wireless medium and, therefore, there is more 
chance that they get lost, and vice versa. 
 The third type of the ABC algorithm comprises all 
the ABC algorithm functionalities and all additional 
functionalities introduced in ABCdie and ABCprop. It is 
denoted as ABCdie&prop. 
4 Performance Evaluation of the 
Initialization Phase  
We conduct several simulation runs of ABC algorithm 
in OPNET simulation model of WMN, presented in [7, 
8]. In this section, we present selected parameters in 
simulations and simulation results. In particular, we are 
interested in the time required by the initialization phase 
of the ABC algorithm to find the shortest paths which 
correspond to the shortest paths found by Dijkstra in the 
number of hops on the paths. In [1, 4, 5], the time 
required for the initialization is assumed to be fixed. In 
ABC paper [4], authors only test if the initialization 
phase of the algorithm finds paths close to the minimum 
number of hops as Dijkstra does. Moreover, they 
 
Figure 1. The state diagram of ABC OPNET Implementation. 

SIMULATION MODEL FOR ANT-BASED CONTROL ALGORITHM IN WIRELESS MESH NETWORKS 155 
performed simulation runs for a synchronized wired 
telecommunication network while, in our case, 
simulated networks are configured to present 
nonsynchronized WMNs. 
4.1 Simulation parameters 
In this paper, only the performance of the initialization 
part of the ABC algorithm is investigated, thus only the 
simulation parameters relevant for the initialization 
phase of the algorithm are discussed in the following. 
 We performed several simulation runs and 
investigate the required initialization time of ABC 
algorithm to find paths with the minimum number of 
hops in dependence of: (i) different number of nodes 
and different number of neighbours per node, and (ii) 
different types of the algorithm used during the 
simulation runs. 
 
 We assume that all wireless network nodes are the 
same and have identical configuration, representing 
homogeneous network. Each node is given a random 
location within a given area. Node locations remain the 
same in all simulation scenarios. Neighbour selection is 
mainly based on the node positions. For the simulation 
purposes all the links are symmetrical 1Mbit/1Mbit and 
are ideal, meaning that no packets get lost during 
transmissions. In Figure 2, a network topology with 30 
nodes each having 5 neighbours is presented. All other 
generated networks look similar. Wireless connections 
established between neighbours, which are represented 
in the network topology as wireless links, are 
graphically presented with dashed lines between nodes. 
Table 1. Parameters of simulation runs. 
Task Type Parameter Attrib. 
Ant creation all Time period 0.01 s 
Ant pheromone 
laying config. 
all 
d 0.08 
c 0.005 
Ants dying 
ABCdie 
ABCdie&prop 
10-30 nodes 15 hops 
40-70 nodes 20 hops 
 
In Table 1, the used parameters for the corresponding 
task and type of ABC algorithm are summarised. The 
time period of ants creation in nodes is presented in 
seconds; this time has to be increased after the 
initialization phase to reduce the network overhead. 
This parameter is used in all types of the algorithm. 
Parameters for updating ants routing tables are 
presented without units and are also used in all types of 
the algorithm. The parameters affect equations, 
presented in [1, 4, 5], for calculating the amount of 
pheromone laying on a specific link. The number of 
hops required that an ant dies is presented in hops for 
networks with different number of nodes and is used 
only in ABCdie and ABCdie&prop. 
4.2 Simulation results 
In the following, the time required to find the shortest 
paths, in terms of the minimum number of hops, by ants 
is investigated. Four types of ABC algorithm 
implementations are investigated: (i) ABC, (ii) 
ABCprop, (iii) ABCdie, and (iv) ABCdie&prop. The 
results of the initialization time are presented for 
networks with 10, 20, 30, 40, 50, 60, and 70 nodes with 
4, 5, 6, 7, and 8 neighbours per node. All the results are 
presented in Table 2. The time is presented in seconds. 
Table 2. Times of ABC initialization phase. 
Case Neigh. 
Nodes 
10 20 30 40 50 60 70 
ABC 
4 0.8 18.4 29.5 54 77 323 352 
5 1.4 10.3 20.8 24 258 249 312 
6 1.2 16.5 30 50 62 149 142 
7 0.5 6.2 9.4 23 45 73 133 
8 0.8 9 9.9 24 38 68 167 
ABC 
die 
4 1.2 16.4 33.1 70 91 143 309 
5 1.1 6.8 19.1 48 70 123 196 
6 0.5 11.1 18.8 47 81 126 129 
7 0.5 7 14.7 32 38 98 180 
8 0.3 7.9 8.4 31 62 80 110 
ABC 
prop 
4 1.3 10.6 54 53 144 294 462 
5 2.9 11.5 15.9 56 201 113 136 
6 0.5 10.1 20.9 59 94 91 130 
7 0.5 5.5 13.6 32 55 90 105 
8 0.4 5.9 9.2 25 66 89 121 
ABC 
die& 
prop 
4 1.2 10.6 56.8 36 143 148 390 
5 1.1 8.2 30.2 34 54 127 237 
6 0.5 12.4 15.3 76 60 132 166 
7 0.3 4.8 11 27 57 85 110 
8 1.3 7.6 7.1 21 40 87 157 
 
The initialization time in dependency of the number of 
nodes in the network for different number of neighbours 
per node for the ABC algorithm is depicted in Figure 3. 
The higher the number of nodes in the network, the 
greater the initialization time (as expected), and the 
lower the number of node neighbours, the greater the 
initialization time. The latter is due to the fact that in 
networks with less node neighbours the diameter of the 
network and average hop count to other nodes are 
greater. Thus, ants need to travel more time to achieve 
the desired results. Similar conclusions can be drawn for 
other types of ABC. 
 However, there are also cases where the 
initialization time for the network with more nodes is 
 
Figure 2. Network topology with 30 nodes and 5 neighbours 
per node. 

156 PERTOVT, ALIČ, ŠVIGELJ, MOHORČIČ 
smaller than for the network with fewer nodes. Similar, 
the initialization time for the network with fewer 
neighbours per node can be in some cases smaller than 
for the network with more neighbours per node. This is 
because of the reinforcement learning algorithm used by 
ants, which comprises also the random part of 
pheromone laying. An ant selects the next hop (i.e. one 
of the neighbours of the node at which the ant is located 
at the moment) based on the probabilities in the 
pheromone table; higher the probability value of the 
neighbour, higher the chance to be selected as ant’s next 
hop. But this also means that the node’s neighbours with 
lower values in the pheromone table can be also 
selected as next hop, only that with lower probability. 
Thus, this randomness which is introduced through the 
probabilities is the cause of deviation in some cases 
from the expected behaviour. 
 
 Similar as above, initialization time in dependency 
of the number of neighbours per node in the network for 
all four types of ABC algorithms for the network with 
20 and 70 nodes is investigated in Figure 4. Please note 
different time scales in both graphs. It can be seen that 
times change when ABCprop is used compared with 
ABC. In the first case, ants require the propagation time 
to travel from a node to a node, while in the second case 
it is assumed that the time for travelling on links 
between nodes is zero. There is no tendency of the 
simulation time to be greater or lower in one of the 
cases. When ABCdie algorithm is used, times tend to be 
smaller compared with ABC, as the ants which are not 
efficient and do not find their destinations after 15 or 20 
hops are eliminated from the network. However, there 
are also cases when ABCdie does not perform better. 
Again, this is because of some randomness of the 
reinforcement learning algorithm. In some cases, it is 
better to leave all ants in the network to learn but it is 
difficult to predict which cases are these. 
5 Conclusion and future work 
In this paper, we have described the implementation of 
the ABC [1, 4, 5] algorithm in the OPNET Modeler [6] 
simulation environment. We have implemented the 
algorithm in our network coding simulation model [7, 8] 
for WMNs. The time of the initialization phase required 
to find the shortest paths by ants, in terms of the 
minimum number of hops, is investigated for different 
types of ABC algorithm. The required initialization time 
increases with the number of nodes in the network and 
decreases with the number of neighbours per node, as 
expected. However, the additional functionalities of the 
ABCdie and the ABCprop tend to decrease the 
initialization time, which is favourable for routing. 
As a further work, we will adapt the ABC algorithm 
for adaptive and dynamic routing that is aware of 
network coding opportunities in WMNs. 
Literature 
[1] R. Schoonderwoerd, “Collective intelligence for 
network control,” M.S.c Thesis, Delft University of 
Technology, May 1996. 
[2] M. Dorigo, G. Di Caro, L. M. Gambardella, “Ant 
Algorithms for Discrete Optimization,” Artificial Life, 
5(2), pp. 137-172, 1999. 
[3] M. Farooq, G. A. Di Caro, “Routing Protocols for Next 
Generation Networks Inspired by Collective Behaviors 
of Insect Societies: An Overview,” Swarm Intelligence 
(Natural Computing Series), Springer, pp. 101-160, 
2008.  
[4] R. Schoonderwoerd, O. Holland, J. Bruten, “Ant-Like 
Agents for Load Balancing in Telecommunication 
Networks,” in Proceedings of the First Int. Conf. on 
Autonomous Agents, pp. 209-216, ACM Press, 1997.  
[5] R. Schoonderwoerd, O. Holland, J. Bruten, L. 
Rothkrantz, “Ant-based load balancing in 
telecommunications networks,” Adaptive Behavior, 
5(2), pp. 169–207, 1996. 
[6]  OPNET web page, available at http://www.opnet.com/, 
retrieved July 2012. 
[7] K. Alic, E. Pertovt, and A. Svigelj, “Simulation 
Environment for Network Coding,” IEEE Jordan 
Conference on Applied Electrical Engineering and 
Computing Technologies 2011 (AEECT 2011) Amman, 
Jordan, 2011. 
[8] K. Alic, E. Pertovt, A. Svigelj, “Network coding 
simulation model in OPNET Modeler,” OPNETWORK 
2012,  Washnigton, NW, USA, August 2012. 
 
 
Figure 4. Initialization time in dependency of the number of 
neighbours per node for different types of ABC algorithm with 
20 and 70 nodes in the network. 
4 5 6 7 8
0
5
10
15
20
Number of neighbours per node
Initialization time (s)
Number of nodes: 20
 
 
ABC
ABCdie
ABCprop
ABCdie&prop
4 5 6 7 8
0
50
100
150
200
250
300
350
400
450
500
Number of neighbours per node
Initialization time (s)
Number of nodes: 70
 
 
ABC
ABCdie
ABCprop
ABCdie&prop
 
Figure 3. Initilization time in dependency of the number of 
nodes in the network for different number of neighbours per 
node for ABC algorithm. 
10 20 30 40 50 60 70
0
50
100
150
200
250
300
350
400
Number of nodes
Initialization time (s)
Type: ABC
 
 
4 neighbours per node
5 neighbours per node
6 neighbours per node
7 neighbours per node
8 neighbours per node