1-s2.0-s0140366415003230-main

ARTICLE IN PRESSJID: COMCOM [m5G;September 16, 2015;10:20]

Computer Communications xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Computer Communications

journal homepage: www.elsevier.com/locate/comcom

Beacon-based channel assignment and jammer mitigation for MANETs

with multiple interfaces and multiple channels✩

Yalew Zelalem Jembre∗, Young-June ChoiQ1

Department of Computer Engineering, Ajou University, Suwon 443-749, South Korea

a r t i c l e i n f o

Article history:

Received 30 January 2015

Revised 2 September 2015

Accepted 3 September 2015

Available online xxx

Keywords:

Ad-hoc networks

Channel assignment

Multiple interfaces

Jamming

a b s t r a c t

The capability of accessing multiple channels through multiple interfaces improve network capacity and is

desirable for future Mobile Ad-Hoc Networks (MANETs). However, due to the presence of jammers as well

as mobility and ad-hoc features, MANETs require distributed and efficient resource management for channel

assignment. To address the channel assignment problem, which is a non-deterministic polynomial-time hard

(NP-hard) problem, we propose a heuristic algorithm called Channel Assignment and JAmmer Mitigation (CA-

JAM). The CA-JAM algorithm assigns a distinct channel for every interface of one station, and then all stations

exchange the assignment information through beacon frames on every individual interface. When one station

receives a beacon, the station organizes the information into tables. Therefore, each station, distributively,

uses the table to reduce the number of neighboring stations using the same channel to avoid interference

which in turn improves the throughput. The tables are also used to learn the disconnected neighbors due

to jamming so as to mitigate the effect of jamming and maintain connectivity. CA-JAM is fully distributed

with no use of control channel or central entity; thus, it improves connectivity and reduces interference by

balancing stations over the available channels while mitigating jamming effects from multi channel multi

interface MANETs. We confirm that CA-JAM outperforms existing protocols using the OPNET simulator.

© 2015 Published by Elsevier B.V.

1. Introduction1

Mobile ad-hoc networks (MANETs) provide flexibility and scala-Q22

bility to set up a network compared to infrastructure networks. Such3

networks can be used in military combat, disaster relief or large con-4

struction sites. All MANET stations with a single interface or radio15

should belong to the same service set on a single channel to stay6

connected, even if the interface can switch between channels [1]. As7

the number of MANET stations increases, interference and collision8

among them increase as well. This degrades network capacity, flexi-9

bility and scalability. Unlike other networks, MANETs are designed to10

have minimal manual configuration, low cost of hardware, and toler-11

ance to jamming attacks and mobility [2].12

Jamming is caused by high powered devices intentionally de-13

signed to attack a wireless network, which is referred to as in-14

tentional jamming. Also, jamming can arise unintentionally from15

non-compatible standards, e.g. 802.11 and 802.15.1 operating in the16

✩ Part of this work has been presented in ACM International Conference on Ubiqui-

tous Information Management and Communication (IMCOM) 2014.∗ Corresponding author. Tel.: +82 312193618; fax: +82 312191688.

E-mail addresses: [email protected], [email protected] (Y.Z. Jembre),

[email protected] (Y.-J. Choi).1 Throughout the text, the terms interface and radio are used interchangeably.

2.4GHz ISM band, which is mostly referred to as interference. Another 17

type of jamming in MANETs is caused by denial-of-service (DoS) at- 18

tacks [3] where malicious stations transmit false messages to con- 19

sume network resource and starve other stations. To mitigate jam- 20

ming, first we need to detect it and then avoid using the jammed 21

channel. Under the assumption that such jamming attacks are de- 22

tectable, stations can dynamically switch from one channel to an- 23

other [4] as in dynamic spectrum access, where stations search for 24

a new channel when the current operating channel is unavailable. 25

Due to the reduction of radio cost, it is now easier to fit multi- 26

ple interfaces in one station, as seen in Wi-Fi devices that work for 27

both 2.4GHz and 5GHz. Therefore, our aim is to exploit this advantage 28

to enhance the network capacity and flexibility of MANETs with the 29

help of Multiple Interfaces of stations operating on Multiple Channels 30

(MIMC) [5] by proposing a new channel assignment scheme. In ad- 31

dition, the channel assignment algorithm should overcome the jam- 32

ming problem as well. 33

In this paper, we first formulate the channel assignment problem 34

as graph partitioning problem that minimizes the number of adja- 35

cent vertices on the same partition, and then propose a distributed 36

and heuristic channel assignment algorithm called Channel Assign- 37

ment and JAmmer Mitigation (CA-JAM), because the problem is found 38

to be non-deterministic polynomial-time hard (NP-hard). In CA-JAM, 39

first, each station determines a distinct random channel for all of 40

http://dx.doi.org/10.1016/j.comcom.2015.09.003

0140-3664/© 2015 Published by Elsevier B.V.

Please cite this article as: Y.Z. Jembre, Y.-J. Choi, Beacon-based channel assignment and jammer mitigation for MANETs with multiple inter-

faces and multiple channels, Computer Communications (2015), http://dx.doi.org/10.1016/j.comcom.2015.09.003


http://www.ScienceDirect.com

http://www.elsevier.com/locate/comcom

mailto:[email protected]





2 Y.Z. Jembre, Y.-J. Choi / Computer Communications xxx (2015) xxx–xxx


its interfaces and exchanges beacons on each interface for a simple41

rendezvous process. To enhance the rendezvous process, algorithms42

described in [6] and [7] can be used. Up on receiving beacons from43

neighbors, stations organize the information into two tables: an in-44

terface table and a neighbor table. To avoid interference, stations look45

up the number of neighbors per interface from its interface table and46

check whether it has multiple links with all neighbors on this in-47

terface. If these conditions are satisfied, the station switches to an-48

other channel to organize a less congested network and seek more49

connectivity.50

When a channel is jammed, the following steps are performed: (1)51

neighbors that are exclusive to the jammed interface are selected, i.e.,52

neighbors with a single link; (2) channel information about the un-53

jammed interfaces of these neighbors is inferred; and (3) the chan-54

nel that is shared among most neighbors is selected and assigned to55

the jammed interface. Then, the interface switches to the assigned56

channel to re-establish communication with its neighbors to recover57

the lost connection due to the jammer presence. CA-JAM is a beacon-58

based fully distributed channel assignment scheme resistant to jam-59

ming attacks, where there is neither control channel nor a central60

unit. To make our algorithm off-the-shelf 802.11-compatible, we only61

slightly modified the beacon type that was defined in the IEEE 802.1162

standard.63

The rest of the paper is organized as follows. In Section 2, related64

work is presented, and in Section 3, our problem statement is de-65

scribed. In Sections 4 and 5, the proposed CA-JAM algorithm and our66

simulation results are described, respectively. Our conclusion is pre-67

sented in Section 6.68

2. Related work69

The use of multiple interfaces incurs a channel assignment is-70

sue. There exist several studies about channel assignment in litera-71

ture each from different perspective; in this section we assess pre-72

vious works in this issue. In [8] and [9], the authors provided ap-73

proaches on how graph theory can be used for channel assignment74

in MANETs. In [10], graphs are substantially applied to the channel75

allocation. First, the topology is determined based on the connectiv-76

ity and interference graph. The interference graph is used to deter-77

mine link interference, and an optimal algorithm is formulated based78

on the connectivity graph to reduce the multichannel link interfer-79

ence. Finally, authors proposed an approximation algorithm, because80

the coloring solution for the formulated graph is NP-hard. Moreover,81

interference among users is considered when designing channel al-82

location schemes. In [11], a centralized multi-radio conflict graph is83

used to model interference among stations, where a channel is as-84

signed using station intelligence to minimize interference through-85

out the network. In [12], another form of conflict graph called Multi-86

Dimensional Conflict Graph (MDCG) is proposed to find a possible87

non-interfering channel assignment.88

In [13], the authors proposed a static channel assignment algo-89

rithm, where each interface is assigned a distinct channel and that90

will determine the topology. The assignment strategy is to allocate91

interfaces in a common neighbor with as many distinct channels as92

possible such that the interference among connections is minimized.93

The authors of [16] suggested that one of the multiple interfaces is94

static while others are dynamic. In this method, HELLO messages are95

exchanged among the stations over the static interface and the infor-96

mation extracted from this message is used to mitigate interference97

among stations.98

The proposed scheme in [36], strives to minimize interference99

with the help of channel assignment. This is a greedy assignment100

which requires all the links in the network as an input. The goal101

is to maximize the number of links that operate simultaneously.102

First each link is mapped with the first channel then the upper and103

lower bound SINR of each link while using that channel is obtained.104

Following that links are prioritized based on the upper and lower 105

bound SINR on that channel; the link with highest priority will be 106

assigned to that channel and the rest of the links are mapped to the 107

next channel. This process is repeated until all the links are mapped 108

to channel. 109

Adaptive Dynamic Channel Allocation protocol (ADCA) is a hybrid 110

channel assignment protocol proposed in [25]. Like [16], one inter- 111

face of each station is static while the others are dynamic. The pur- 112

pose of the static interface is to enhance throughput between a sta- 113

tion and central node whereas the dynamic interfaces are designed to 114

work in on-demand mode. Time is divided into fixed intervals, each 115

having a control and data interval. In the control interval dynamic 116

interfaces negotiate a channel and choose the “least congested chan- 117

nel”. Each station has a queue associated with its neighbors and in 118

the data interval the algorithm takes the queue length into consid- 119

eration to choose to which group of neighbors it should communi- 120

cate first. In [15], IEEE 802.11 stations obtain neighbor information 121

through scanning and beacon broadcast. Then, they form a local co- 122

ordination group based on similarity of available channels. The group 123

votes on channels to select one channel as a coordination channel for 124

future channel assignment. 125

The work in [21] suggests a group-based channel assignment 126

(GCA) algorithm based on a divide and conquer approach. In GCA, sta- 127

tions are classified into a master and slaves, where the master is re- 128

sponsible for gathering new link formation and channel assignment. 129

When a slave node joins the network it will send a request to its 130

neighbors who acknowledge with a reply. However, it is the job of 131

the master station to decide whether the new station’s link should be 132

activated or not. If it decides to activate the new link, the master will 133

send a broadcast message to all neighbors about the activation of the 134

new link. In GCA, links are grouped based on the bandwidth require- 135

ment, and the groups are organized to form components. Then, links 136

of each component are assigned to different channels. The first and 137

second procedures guarantee a balance whereas the last procedure 138

enhances throughput and fairness. 139

Another group based channel assignment is found in [33]. In this 140

scheme, stations that are fully connected to one another belongs to 141

one group. Stations that overhear other groups communication are 142

called bridge stations. First each group will be assigned one channel 143

in such a way that there is no collision between neighboring groups. 144

One interface of each station in the group will use that for communi- 145

cation. Bridge stations will form fully connected group on their sec- 146

ond interface and that group will be assigned a channel in the same 147

manner. When stations communicate with in the group a Latin square 148

based scheduling is implemented to avoid collision. When stations 149

would like to communicate members of other group than its own 150

then it will forward its data to the bridge station; the group made 151

of bridge stations also employ Latin square method to avoid collision. 152

The authors in [14] investigate the joint effect of topology control 153

and channel assignment in two stages; first, every station adjusts and 154

checks its link until an undirected graph is constructed; second, ev- 155

ery station is assigned a transmission channel. In [17], a joint channel 156

assignment and routing protocol is proposed to minimize the maxi- 157

mum number of l-hop neighbors that share the same channel. Each 158

station builds the network topology from periodical HELLO messages 159

sent through the common control channel (CCC). Stations detect ac- 160

tive neighbors using the request-to-send (RTS)/ clear-to-send (CTS) 161

through CCC. Then, they choose a channel that is available for both 162

stations from their list. Other stations update their available channel 163

list upon listening to the RTS/CTS. The use of a CCC and 802.11-like 164

MAC protocol is also found in [18]. 165

Unlike [17], to detect channel availability, stations use spectrum 166

sensing in addition to HELLO messages. One transceiver is used for a 167

CCC whereas others are used for data exchange; Data Transmission 168

reServation (DTS) is used for control packet transmission in order 169

to announce spectrum reservation and transmit power to neighbors. 170




Y.Z. Jembre, Y.-J. Choi / Computer Communications xxx (2015) xxx–xxx 3


Table 1

Notations and their description.

C Number of channels

nI Number of interfaces per station

Cjmd Jammed channel number

NRC New random channel number

uic Interface i of station u operating on

channel c.

d(uic) Degree (i.e. number of neighbors) of

station u on interface i and channel c

S(uic) Set of distinct neighbors of station u

on interface i and channel c

dopt Optimal number of neighbors per

interface

k(G) Connectivity of graph G

�CG Group of neighbors that share the

same channel

Similarly, the work in [24] uses control information to implement171

channel assignment in MIMC environments. However, they reuse the172

802.11 RTS/CTS messages in pursuit of avoiding two-hop interference.173

The basic steps of this work are summarized as follows: when two174

stations want to communicate, they exchange RTS/CTS on the control175

channel and the neighboring stations avoid using the same channel176

by overhearing the control information.177

A channel assignment based on parallel rendezvous scheme with178

two interfaces was proposed by the authors in [34]. Stations uses179

hopping sequence to form a link with neighbors. The first interface180

is used for transmission and uses fast hopping whereas the second181

interface is used for reception and transmission of control messages182

such as HELLO packets; it follows slow hopping sequence. Station183

is required to be synchronized with its 1-hop neighbors with the184

second interface using the HELLO messages. Channel is divided into185

time slots, when two stations are in the same slot communication186

occur.187

In [19], the authors proposed an Ad hoc On Demand Distance Vec-188

tor (AODV) based joint channel assignment and routing algorithm.189

Similar to [17,18], this method uses a CCC for information control;190

however, it uses interference indices to weight channels to avoid191

interference in the network. The authors of [26] categorized chan-192

nels into control and data classes. This work also modifies the fa-193

mous AODV protocol to exchange control information. The authors194

in [23] add an extra routing layer to address channel assignment to-195

gether with routing. First, each interface of a station is indexed and196

the channels from the lowest to the highest are assigned to these197

interfaces. The interface with the lowest index and channel is used198

for best efforts; i.e. control information includes the channel and in-199

terface. This work modifies the traditional Open Link State Routing200

(OLSR) protocol for the purpose of enhancing the throughput and per-201

forming channel assignment. The control information is piggybacked202

in Topology Control (TC) message of OLSR. The channel assignment203

is then performed using TC messages which are received from the204

neighbor based on estimation of available bandwidth of each node.205

In [22], channel assignment using particle swarm optimization206

(PSO) is proposed. This is inspired by a social behavior, e.g. bird flock-207

ing or fish schooling. Due to the discrete nature of multi-hop net-208

works, the original PSO could not be directly applied for CA. For209

this reason, the authors first developed discrete PSO (DPSO) algo-210

rithm to form DPSO-CA. The main focus of the work is to minimize211

interference while maintaining a topology. To preserve a topology,212

two neighbors must be assigned at least one channel. Interference213

is avoided by applying crossover and mutation for the initial channel214

assignment of the network, which is supervised by a central node. A215

more comprehensive survey on channel assignment can be found in216

[27].217

Another bio-inspired joint work is found in [35]. The authors for-218

mulated an optimization problem to get the optimal channel and219

power allocation of the entire network such that throughput and220

fairness are enhanced. The problem is complex and NP-hard hard to 221

find the optimal solution, which made the authors to consider the hy- 222

bridization of two bio-inspired schemes, genetic algorithms (GA) and 223

PSO such that optimal channel and power allocation are achieved. 224

The authors used PSO to find the sub optimal solutions and use the 225

solutions from PSO in GA to find the optimal solution. 226

Most 802.11-based MAC protocols require the availability of a CCC 227

[17–19,23,24], topological view [13,14,22] or central entity [11,12,21] 228

for channel assignment. Although, CCC, central entity or the whole 229

network could easily be jammed or compromised, none of the above 230

works has taken jamming into consideration. In addition, some of 231

the works solve the channel assignment problem jointly with rout- 232

ing [19,23,26]. However, these works fail to work with multiple 233

routing protocols designed for MANETs. Taking the design princi- 234

ples outlined in [20] into account and also considering the weak- 235

ness of exiting work we have proposed CA-JAM; the contribution 236

of this work in comparison with existing works is summarized as 237

follows: 238

• We developed a channel assignment strategy that is scalable and 239

flexible for multi-interface multi-channel MANETs while being to- 240

tally distributed i.e. no CCC, central entity and no association with 241

network layer for routing. In addition, we develop a scheme that 242

enhance throughput and reduces interference. Our algorithm also 243

improves connectivity by balancing stations that are in the same 244

vicinity and using the same channel. Finally, CA-JAM is different 245

from existing works because it provides an adaptive channel as- 246

signment against jamming attacks.2 247

3. Problem definition 248

For MANETs, the topology of a network can be represented as con- 249

nectivity graph G(V, E), where a vertex in V represents a station in the 250

network and an edge in E is placed between two vertices (u, v), if they 251

are within the communication range. Let |V| denote the cardinality of 252

V. A conflict graph is consequently defined to account for interference 253

issues. In a conflict graph, vertices represent the communication links 254

and edges are placed between vertices corresponding to interfering 255

links based on the protocol model. However, the conventional con- 256

flict and connectivity graph does not capture the MIMC environment. 257

To support our system model, we discuss a multi dimensional conflict 258

graph (MDCG) in the following subsection. The notations used in our 259

system model are described in Table 1. 260

3.1. Conflict graph 261

An MDCG for MIMC MANETs is defined in [12]. To describe the 262

conflict relationship among stations, we modify radio-link-channel 263

(RLC) tuples used in [12] to simple-RLCM (S-RLCM); where M is the 264

number of tuples created by stations with a common channel and 265

can be calculated as M ≤ �(|V| × nI)�2�. S-RLC is defined as (uic, v j

c) 266

and indicates that there is a common channel c between interfaces 267

i and j of stations u and v that forms a link. Using these tuples, we 268

can describe all the possible conflicts in a MDCG. The conflicts can be 269

explained using three events: 270

• X: stations are interfering in a protocol model. 271

• Y: stations are associated with the same channel. 272

• Z: stations that share a common interface at one or two stations. 273

There is a conflict between tuples when the condition (X ∩ Y ∩ 274

Z) ∪ Z is true for two different S-RLCs. The first condition X ∩ Y ∩ Z 275

implies that concurrent transmissions within the same interference 276

2 This work is an extension of our previous work [32]. This work provides full de-

scription of the working principle of our proposed scheme, the proof on the NP-

hardness of the scheme, and further performance evaluation.






A

B C

(a) Simple graph with 3 stations

,

,

,

,

,

,

,

,

(b) Resulting MDCG graph

Fig. 1. Conversion from a given graph to conflict graph.

range are not supported, whereas the second condition Z implies that277

a single interface cannot support multiple transmissions. Let S-RLC =278

{S-RLC1, S-RLC2, . . . , S-RLCm} (m ≤ K) be the set of tuples that transmit279

concurrently, then an indicator function is defined as follows:280

f (S-RLC) ={

1 if there is a conflict in S-RLC,

0 otherwise.(1)

To illustrate how the MDCG works, we assume a network with281

two available channels and three stations A, B and C. Here A has two282

interfaces while B and C have only one interface. A is within the trans-283

mission range of both B and C; however, B and C are not in the trans-284

mission range of each other as shown in Fig. 1a. This network can285

be mapped into four S-RLC tuples. For example, tuple (A11, B1

1) means286

that station A transmits to B on channel 1 and both stations use in-287

terface 1. Therefore, using (X ∩ Y ∩ Z) ∪ Z condition and considering288

all tuples, we obtain the MDCG as shown in Fig. 1b. According to the289

MDCG in Fig. 1b, there is no conflict in either {(A11, B1

1), (A22,C1

2)} or290

{(A291

ter292

one293

in t294

ing295

for296

ple297

for298

3.2299

300

num301

tion302

per303

cha304

ma305

nel306

of307

sam308

how309

the310

sha311

des312

can313

MD314

mit315

be stated as follows: 316

dopt = {min d(uic)}

subject to

uic �= uj

c, ∀(i, j) ∈ NI

k(G) ≥ 1,

f (S-RLC) ≤ 0,

k(G), f (S-RLC) ∈ {0, 1}i, j, c ∈ {0, 1, 2, . . .} (2)

where 317

k(G) ={

1 if there is at least one path for any sourceand destination,

0 otherwise.

(3)

Since the Finding the minimum degree of a vertex on each of its 318

interfaces is equivalent to finding the optimal number of neighbors 319

(do 20

con 21

sam 22

the 23

for 24

tion 25

tim 26

fere 27

pro 28

tha 29

eac 30

put 31

tha 32

ass 33

sig 34

Sin 35

in i 36

Pl

fa

22, B1

2), (A1

1,C1

1)} in S-RLC set. This implies (1) there will be no in-

ference between links if the network is configured according to

of these two sets and ; (2) there is no interface conflict occurs

he network. Although the MDCG finds the ultimate none interfer-

tuples, it has several deficiencies as discussed in Section 2. There-

e, we propose an algorithm that can find the non-interfering tu-

s while avoiding these drawbacks of MDCG. In the following, we

mulate the optimization problem.

. Optimization problem

The goal of channel assignment is to optimally allocate the limited

ber of channels available for MIMC MANETs. If the number of sta-

s is less than the available channels or if the number of interfaces

station is equivalent to the available channels, then the optimal

nnel assignment can be achieved easily by exhaustive resource

pping. However, in most practical scenarios, the available chan-

s are less than the number of stations and more than the number

interfaces. Minimizing the number of stations that are using the

e channel minimizes interference and enhances the throughput;

ever, it affects connectivity k(G). Therefore, we need to identify

optimal number of stations that are with in close proximity and

re the same channel while there is at least one path for any source

tination pair in the network. Also no two interfaces of one station

have the same channel and there is no conflict according to the

CG, i.e., two links on the same interference range do not trans-

simultaneously. Mathematically, the optimization problem can

pro 37

opt 38

The 39

eac 40

Pro 41

sam 42

sta

int

ery

tha

ease cite this article as: Y.Z. Jembre, Y.-J. Choi, Beacon-based channel assig

ces and multiple channels, Computer Communications (2015), http://dx.do

pt) that one station should have on each of its interfaces. The first 3

straint defines that two interfaces of one station cannot have the 3

e channel. The second constraint means that the connectivity of 3

graph should always be greater than or equal to one. This en- 3

ces that there should always be one path for any (source, destina- 3

) pair. The third constraint states that the final solution to the op- 3

ization problem should not contain conflicting tuples; i.e., inter- 3

nce should not be caused while trying to solve the optimization 3

blem. Therefore, if we can find the optimal number of neighbors 3

t one station has to have on each of its interfaces, then by limiting 3

h interface to that number, we can obtain the optimal through- 3

and connectivity because that is the optimal number of stations 3

t should share the same channel. Like the NP-hardness of channel 3

ignment problems in many literatures [10,28,29], the channel as- 3

nment of our work is also NP-hard for the MIMC ad-hoc networks. 3

ce the domain of all the variables of this optimization problem is 3

nteger, the nature of this optimization problem is of integer linear 3

gramming, which is NP-hard. The proof of the NP-hardness of the 3

imization problem is given as follows. 3

orem 1. It is NP-hard to find the optimal number of neighbors at 3

h interface. 3

of. We want to divide the network such that every group has the 3

e number of stations and there is connectivity between any two 3

tions (k(G) ≥ 1). Let d(uic) be the degree of a vertex of station u on 343

erface i and channel c, and let dopt be the optimal degree that ev- 344

station has, i.e., the minimum number of neighbors of one station 345

t share the same channel on a single interface. We show that it 346

nment and jammer mitigation for MANETs with multiple inter-

i.org/10.1016/j.comcom.2015.09.003




S

A

ATA

1 DCF

broadf thehanne

data

n-347

at348

G349

ub350

ity351

es352

ns353

e-354

es355

P-356

357

358

a-359

nt360

al-361

362

n-363

n-364

no365

s.366

a367

he368

ay369

ny370

371

be372

ng373

re374

od375

of376

377

378

ng379

ri-380

l-381

s-382

of383

to384

nd385

to386

de387

at388

M 389

e- 390

le 391

a- 392

-2. 393

ng 394

w 395

he 396

397

ng 398

399

ID 400

401

a- 402

x 403

404

er 405

n, 406

407

le- 408

409

410

Ds 411

on 412

ti- 413

414

n- 415

ed 416

be 417

le 418

ve 419

r- 420

in 421

422

423

C 424

rst 425

e- 426

ch 427

le 428

.e., 429j). 430

D RTS S

CTS

S DATA

NAV RTSNAV CTS

NAV D

B

B

B

X

Y

Z

802.1Beacon Duration

Beacon Interval

B: Beacon,CW:- Contention Window,D:- DIFS,S:-SIFS

STN-ID: beaconM-IDX:- index oC-1 to C-I:- the c

→→→→

→→→

Fig. 2. Beacon and

is NP-complete to find dopt using the reduction from graph partitio

ing problem [30,31,31]. The graph partitioning problem states th

given a graph G(V, E) and a constant k, it is NP-complete to induce

into k sub graphs G1, G2, . . . , Gk where the vertices of the induced s

graphs are equal, while minimizing the disconnection, i.e., capac

of edge cut, between sub graphs. Suppose we find dopt, which impli

d(uic)=dopt for all interfaces of one station on any channel. This mea

that the number of stations on a certain channel is equal to the d

gree of one station plus the station itself (dopt + 1). This solution giv

a valid graph partitioning with minimum edge cut. Therefore, it is N

complete to find an optimal degree for any dopt and d(uic). �

3.3. Design considerations

Before explaining our CA-JAM algorithm, we discuss some fe

tures that should be considered to design a good channel assignme

scheme. Later, the effectiveness of the proposed CA-JAM will be ev

uated based on these principles.

• Interface-channel mapping: in the MIMC MANETs, mapping an i

terface to a channel is not a simple task. It could create a disco

nected network if interfaces of communicating neighbors have

common channel, i.e., neighbors would not be able to rendezvou• Connectivity: channel assignment in MANETs is about having

connected network. Because there is no central unit to oversee t

delivery of a packet from a source to a destination, the only w

to achieve successful data delivery is to guarantee a path from a

source to any destination.• Interference and jamming avoidance: interference could

caused intentionally by stations that want to disrupt an ongoi

communication (jamming) or unintentionally by stations that a

not aware of an ongoing communication (interference). A go

channel assignment scheme should be able to avoid both types

interference.

4. CA-JAM algorithm

In our system, stations exchange interface-to-channel mappi

information using beacons that are broadcasted by every station pe

odically. The data communication between consecutive beacons fo

lows the IEEE 802.11 protocol. Fig. 2 depicts beacon and data tran

mission in our system. It only shows the operation of one interface

three stations operating on the same channel. Time is divided in

Beacon Interval (BI), which consists of Beacon Duration (BD) a

802.11 DCF duration. In BD, stations randomly choose the time

broadcast a beacon. We only modify the beacon message to inclu

the interface/MAC index sending the beacon and the channels th

Please cite this article as: Y.Z. Jembre, Y.-J. Choi, Beacon-based channe

faces and multiple channels, Computer Communications (2015), http:/

CK

D / / / / / / / CW

/ / / / / / / CW

B

B

B

STN-ID M-IDX C-1 C-2 C-I

Time

...

caster station ID,interface the beacon initiated from,ls of each interface of broadcaster station,

transmission in CA-JAM.

are assigned to all the interfaces of a station, which makes CA-JA

easily implementable using off-the-shelf 802.11 interfaces. The r

ceived beacon is organized in two look-up tables, the Neighbor_Tab

and Interface_Table. When extracting the information from the be

con, if M-IDX is 2 the channel assigned to the sending interface is C

The rest C-i’s are used to re-establish a connection with the sendi

station in case this connection is lost. The detailed discussion on ho

to use the C-i’s to re-establish the lost connection is discussed in t

following subsections.

First, the Neighbor_Table is a collection of tuples with the followi

information.

• Neighbor_ID stores the information collected from STN-

field of the beacon.• MAC_Index stores information extracted from M-IDX of the be

con. Because a neighbor has multiple interfaces, MAC_Indeidentifies the MAC of the interface that sent the beacon.

• NbrChannel_i(iε{1,...,I}) stores the neighbor’s channel numb

for each interface and is collected from C_i fields of the beaco

where (iε{1, . . . , I}).• Expiry_Time stores the maximum time a tuple remains re

vant after it is created.

Second, the Interface_Table is made up of the following tuples:

• Nbrs_on_Interface_i(iε{1,...,I}) stores the list of neighbor I

that are exclusive to interface i for I interfaces. This helps a stati

track the number of neighbors on each interface and their iden

ties.

These tables are updated on three occasions: (1) when the cha

nel of an interface switches; (2) when the information has expir

and; (3) when a beacon is received. Note that a neighbor might

found on more than one interface, which means there are multip

links between any two stations. CA-JAM is solely based on the abo

look-up tables and composed of three modules: rendezvous, interfe

ence avoidance, and jammer mitigation, which will be explained

the following subsections.

4.1. Rendezvous

This module covers the channel assignment during the MIM

MANET initialization or when a station joins the network for the fi

time. It is assumed that interfaces are indexed from 1 to nI. Ther

fore, the station iterates through all of its interfaces and assigns ea

one a random channel that is different from others. This will enab

stations to satisfy the first condition of the optimization problem, i

two interfaces of one station cannot have the same channel (ui �= u
c c
l assignment and jammer mitigation for MANETs with multiple inter-

/dx.doi.org/10.1016/j.comcom.2015.09.003




Table 2

Rendezvous algorithm.

Input- C: number of channels, nI: number of interfaces

Procedure Rendezvous()

| 1: for i = ( 1 to nI ) do

| 2: channel_of_ui = newRandChNum()

| 3: i = i + 1

| 4: end for

| 5: while( beacon_duration )

| 6: for i = ( 1 to nI ) do

| 7: while( d(uic) == 0 )


| 9: i = i + 1

| 10: end while

| 11: end for

| 12: beacon_duration = next_duration

| 13: end while

End procedure

Ta

Ra

The431

in T432

433

3 o434

int435

lea436

the437

for438

tha439

sig440

ing441

the442

fro443

cha444

cha445

fac446

eac447

dit448

the449

the450

by451

fac452

is z453

4.2454

455

all456

eac457

on458

for459

tion460

one461

Tab

Inte

In

se

Pr

|

|

|

|

|

|

|

|

|

En

• 62

63

• 64

65

• 66

67

68

69

me 70

eas 71

int 72

ma 73

int 74

on 75

ma 76

ter 77

ber 78

wit 79

wit 80

dit 81

nel 82

and 83

Tab 84

ule 85

nei 86

the 87

nei 88

wh 89

fac 90

con 91

refe 92

Nb 93

in m 94

thi 95

me 96

The 97

line 98

and 99

to t 00

01

num 02

num 03

the 04

Sec 05

wil 06

4.3. Jammer mitigation 507

CA-JAM is able to switch channels to re-establish a link against 508

jamming. When an operating channel of an interface is jammed, the 509

Pl

fa

ble 3

ndom channel generator algorithm.

Input- Cjmd: jammed channel number NRC = 0: new random channel number

Procedure newRandChNum()

| 1: NRC = random( 1,...,C )

| 2: for i = ( 1 to nI ) do

| 3: if ( NRC == channel_of_ui OR NRC == Cjmd ) then

| 4: NRC = random( 1,...,C )

| 5: i = 1

| 6: end if

| 7: i = i + 1

| 8: end for

| 9: return NRC

End procedure

rendezvous and random channel generator algorithms are given

ables 2 and 3, respectively.

After all interfaces have a distinct channel according to lines 1 to

f the rendezvous algorithm, the station broadcasts beacons on all

erfaces and listens to its neighbors’ beacons. If a station hears at

st one beacon from neighbors, then a rendezvous is completed and

station starts to update the tables. However, if there is no beacon

the whole beacon duration, which means there is no neighbor on

t interface, the station waits until the next beacon duration, as-

ns a new channel, and broadcasts its new assignment while listen-

again (lines 5 to 13 of the rendezvous module). In other words,

station will stop rendezvous process when it receives a beacon

m other station, which mean there is at least one node in that

nnel. Line 3 in the random channel module ensures that the new

nnel is not jammed or has already been assigned to another inter-

e. This process is repeated until there is at least one neighbor on

h interface, i.e., d(uic) ≥ 1. Through this process, the second con-

ion of the optimization problem is fulfilled, i.e., k(G) ≥ 1. Unless

distance between one or a group of stations is much longer than

remaining network, there will be no disconnection. This is done

counting the entities of Nbrs_on_Interface_i from the Inter-

e_Table. In other words, if the count of Nbrs_on_Interface_iero, it means there is no neighbor on that interface.

. Interference avoidance

Now that the network has been established through rendezvous,

stations in the network have at least one or more neighbors on

h of their interfaces. Let �CG be the group of stations that reside

the transmission range of each other and use the same channel to

m a communicating group. However, for successful communica-

, the �CG group has to overcome interference that is caused by

of the following scenarios:



le 4

rference avoidance algorithm.

put- d(uic): degree of a station u on interface i, NI: number of interfaces, S(ui

c):

t of distinct neighbors on interface i,

ocedure interferenceAvoidance()

1: for i = ( 1 to NI ) do

2: S(uic) = getDistinctNbrs(i)

3: nbrsOn_Inti = Interface_Table.getNext()

4: d(uic) = nbrsOn_Inti .sizeOf()

5: if ( d(uic) > dopt AND S(ui

c) == ∅ ) then

6: channel_of_ui = newRandChNum()

7: end if

8: i = i + 1

9: end for

d Procedure

First, the hidden and exposed terminal problems occur because 4

they are common among the IEEE 802.11 networks. 4

Second, interference or congestion occurs, if the number of sta- 4

tions in �CG is high. 4

Third, interference occurs among �CG if there is any station that 4

has a neighbor within the interference range that uses the same 4

channel ( f (S-RLC) = 1). 4

Because stations that belong to the same �CG listen to the probe 4

ssages broadcasted in the group, the first type of interference is 4

ily avoided using the RTS/CTS mechanism. The second type of 4

erference is caused by capacity overload due to the existence of 4

ny stations in the same �CG or channel. Therefore, to avoid such 4

erference, a station needs to check if the number of its neighbors 4

each specific interface d(uic) is greater than dopt. Here, dopt is the 4

ximum number of neighbors a station can have on each of its in- 4

faces optimally. For that matter, we compare if the current num- 4

of neighbors on one interface is greater than dopt. For interfaces 4

h d(uic) ≥ dopt , stations check whether there are multiple links 4

h all neighbors on this interface (i.e. if S(uic) = ∅). When the con- 4

ion holds, S(uic) = ∅, the station re-assigns a new distinct chan- 4

to that interface such that the current channel is less congested 4

new connections are created. This enhances the connectivity. 4

le 4 shows the pseudo code of the interference avoidance mod- 4

, where lines 2 to 4 get the size of neighbors and the set of distinct 4

ghbors of an interface, and lines 5 to 7 check the conditions (that 4

number of neighbors on this interface has exceeded dopt and all 4

ghbors in this interface have multiple links with this station) and 4

en the condition is true station assign a new channel to this inter- 4

e. Otherwise, the station will remain on the same channel because 4

nectivity is more critical in MANETs. S(uic) is obtained by cross- 4

rencing both tables. Station matches the Neighbor_IDwith the 4

rs_on_Interface_i for iε{1, . . . , I}, if Neighbor_ID exists 4

ore than one interface , it means neighbor has multiple link with 4

s station so it will not be added to S(uic). This implies S(ui

c) = ∅, 4

ans all neighbors on interface i has multiple links with this station. 4

module to select distinct channels is shown in Table 5, where 4

s 3 to 9 count the number of interfaces on which neighbors exist 4

lines 10 to 12 check if the count is one i.e, if a neighbor is exclusive 4

he interface. 5

Eventually, when this process is repeated by every station, the 5

ber of neighbors on every interface converges to the minimum 5

ber that could communicate to that interface d(uic) = dopt . Thus 5

goal of the optimization problem is accomplished as stated in 5

tion 3. This process implicitly avoids the primary interference; this 5

l be discussed using an example in Section 4.4. 5


i.org/10.1016/j.comcom.2015.09.003




Table 5

Distinct neighbor selection algorithm.

Input- inum: Interface Number

Procedure getDistinctNbrs(inum)

| 1: count = 0

| 2: while ( Neighbor_Table.hasNext() )

| 3: nbr < − Neighbor_Table.getNext()

| 4: for i = ( 1 to NI ) do

| 5: nbrsOn_Inti = Interface_Table.getNext()

| 6: if ( nbr.isIn_List(nbrsOn_Inti) )

| 7: Increment count

| 8: end if

| 9: end for

| 10: if ( count == 1 )

| 11: S(uinumc ) < - nbr

| 12: end if

| 13: count = 0

| 14: end while

| 15: return S(uinumc )

End Procedure

Table 6

n-510

th511

is512

ed513

ls514

ed515

-516

he517

or518

h-519

of520

li-521

ly.522

ts.523

he524

to525

up526

en527

he528

. If529

ere

els

Table 7

Backup and candidate channel selection algorithm.

Input- inum: Interface Number BCH(ui) = 0: backup channel interface

i, CCH(ui) = 0: candidate channel interface i

Procedure getBackupCandidateCH(inum)

| 1: S(uinumc ) = getDistinctNbrs(inum)

| 2: while ( S(uinumc ).hasNext() )

| 3: distNbrCH < − S(uinumc ).getChannel()

| 4: if ( inum != nI and inum != NI−1 )

| 5: L-BCH < − distNbrCHinum+1

| 6: L-CCH < − distNbrCHinum+2

| 7: else if ( inum == NI−1 )

| 8: L-BCH < − distNbrCHnI

| 9: L-CCH < − distNbrCH1

| 10: else if ( inum == NI )

| 11: L-BCH < − distNbrCH1

| 12: L-CCH < − distNbrCH2

| 13: end if

| 14: end while

| 15: BCH(uinum ) = L-BCH.getFrequentCH()

| 16: CCH(uinum ) = L-BCH.getFrequentCH()

end Procedure

he 530

w 531

ce 532

533

ed 534

its 535

he 536

m 537

ts 538

nd 539

h- 540

d, 541

el 542

543

544

e- 545

- 546

ch 547

e, 548

th 549

on 550

= 551

r- 552

th 553

i- 554

on 555

th 55611, 557

re 558

)}. 559

a 560

ns 561

te 562

563

as 564

is 565

le. 566

is 567

ck 568

3. 569

E 570

es 571

Jammer mitigation algorithm.

Input- Cjmd: jammed channel number, /∗ flag is used to indicate whether BCH and

CCH can be used or not. If flag = 0 at least one of the can be used if flag = 1 then

the jammed channel should be assigned new random channel∗/ flag = 0 first it is

set to 0

Procedure jammingMitigation()

| 1: for i = ( 1 to nI ) do

| 2: if ( Cjmd == channel_of_ui ) then

| 3: getBackupCandidateCH(i)

| 4: channel_of_ui = BC(ui)

| 5: for j = ( 1 to NI ) do

| 6: if ( channel_of_ui == channel_of_uj and flag == 0 ) then

| 7: new_CH = CC(ui)

| 8: j = 1

| 9: f lag = 1

| 10: else if ( flag == 1 ) then


| 12: end if

| 13: end for

| 14: end if

| 15: end for

End Procedure

station selects a set of neighbors that are distinct to the jammed i

terface S(uic) from the table, i.e., stations will try to reconnect wi

neighbors that share a single link rather than multiple ones. Th

is described in Table 6. In line 2, it finds which channel is jamm

and in line 3, the preparation of backup and candidate channe

for that interface is requested by invoking the module as describ

in Table 7. Similar to the interference avoidance module, mem

bers of S(uic) are selected by matching the Neighbor_ID with t

NbrID_on_Interface_i for iε{1, . . . , I}. Then, for each neighb

in S(uic), stations compare the jammed channel Cjmd to the neig

bors’ channels. When a match is found , i.e., Cjmd = ci, iε{1, . . . , I}the neighbor, channels ci+1 and ci+2 will be added to the list of e

gible backup (L-BCH) and candidate (L-CCH) channels, respective

When Cjmd = cNI−1, then cnI

and c1 will be used in the respective lis

If Cjmd = cnI, then c1 and c2 will be used. This is done through t

backup and candidate channel module, lines 2 to 14. According

lines 15 and 16, the most common channel from S(uic) become back

(BCH) and candidate (CCH) channel using L-BCH and L-CCH. 3 Wh

the operating channel of an interface of a station is jammed due to t

presence of neighbor users, it independently switches to the BCH

3 Both BCH and CCH might be considered as backup channels. In our system wh

we have three interfaces, the BCH and CCH of a certain interface are the two chann

used in the other two interfaces.



the BCH is not available, that interface switches to the CCH. In t

circumstance when both BCH and CCH channels are jammed, a ne

channel other than the BCH and CCH will be assigned to the interfa

using the rendezvous process.

For instance, for stations with three interfaces, if the jamm

channel of a neighbor is the first interface, then the channels that

neighbor uses for its second and third interfaces will be added to t

list of eligible backup and candidate channels, respectively. Then fro

the list of eligible backup and candidate channels, the station selec

the most dominant ones from both lists and makes them the BCH a

CCH for the jammed interface. Next, the station updates the Neig

bor_Table and assigns the BCH to its interface. If this is also jamme

it tries the CCH. In case both are not available then a random chann

will be assigned according to the rendezvous process.

4.4. Example of CA-JAM algorithm

The CA-JAM algorithm is described using an example. We first d

scribe rendezvous, and then explain interference avoidance and jam

mer mitigation, concurrently. Assume a network of five stations, ea

with three interfaces. We chose the number of interfaces to be thre

because it is suitable to show the working principle of CA-JAM wi

backup and candidate channel, as discussed below. The simulati

follows this as well. Only one neighbor per interface is allowed (dopt

1). First, each station selects a distinct random channel for all its inte

faces as depicted in Fig. 3a. When stations pick the same channel wi

their interfaces and exchange beacons for rendezvous, a commun

cating group �CG is formed. Fig. 3b shows the processes of beac

exchange and rendezvous. For instance, station A on channel 1 wi

its interface 1, A11, and station B on channel 1 with its interface 1, B

form communicating group �CG. Similarly, other stations form mo

groups: �CG = {(A11, B1

1), (B3

5,C3

5), (C1

6, D1

6), (D2

8, E2

8), (C2

2, D3

2, E3

2During beacon reception, each station updates its Neighbor_Table

nd Interface_Table. In this paper, we only show the tables of statio

B and D due to the lack of space. Fig. 3c demonstrates the upda

procedure applied for stations B and D.

As a result of assignment and rendezvous process, station D h

more neighbors on its interface 3 than it is allowed. Station D

aware of the situation using the information from the Interface_Tab

In the mean time, suppose channel 1 between station A and B

jammed as depicted in Fig. 3d. The next step for station D is to che

whether there are multiple links with all neighbors on interface

This condition exists because, station D is connected with C and

through interface 1 and 2, respectively. Therefore, station D switch






1

2

8

7

5

6

2

5

6

8

2

1

8

2

A B C D E

1

(a) Stations select distinct random channels.

1

2

8

7

5

6

2

5

6

8

2

1

8

2

A B C D E

1

5

6

8

2

2

1BCNA

BCNB

BCNC

BCND

BCND

BCNE

(b) Stations exchange beacons for rendezvous.

NbrID_on_InterfaceInterface Table of B

Nbr_ID

MAC_Index

NbrChannel3

Expire_Time

Neighbor Table of B

A 1 0.5822

11

1 2 3A 0 C

C 3 0.5526

NbrID_on_InterfaceInterface Table of D

0 0 E

Nbr_ID

MAC_Index

NbrChannel3

Expire_Time

Neighbor Table of D

C 1 0.5522

16

1 2 3C E C

E 2 0.5281

StationName ALEGEND

A

X

Y

Z

Interface 1 of A on channel X

Interface 2 of A on channel Y

Interface 3 of A on channel Z

(c) Neighbor_Table and Interface_Table update from stations B and D. D has more thanone neighbor on 3rd interface.

7

2

8

A

(e) SstatiE, D

pireime0.50.5

pireime0.50.5

face_

algor

to572

to573

sta574

ma575

to t576

{ 8577

of j578

the579

CCH580

to581

cha582

Alt583

as584

mit585

a m586

ass587

of t588

589

int590

int591

ent592

alg593

pin594

sta595

tan 96

int 97

nei 98

how 99

sec 00

is n 01

5. 02

03

OP 04

tra 05

sta 06

tion 07

eac 08

by 09

of 10

tion 11

for 12

ma 13

the 14

sin 15

IEE 16

as a 17

Pl

fa

1

2

3

7

6

2

5

6

8

2

1

8

2

A B C D E

1 6

8

2

2

JAM 1

5 5

D has more than one neighbor on 2

(d) Channel 1 between stations A and B is jammedand station D has more than allowed neighbors oninterface 2.

Nbr_ID

MAC_Index

NbrChannel3

Ex_T

Neighbor Table of B

A 2 822

17

C 3 526

Nbr_ID

MAC_Index

NbrChannel3

Ex_T

Neighbor Table of D

C 1 522

16

E 2 281

(f) Neighbor_Table and Intertions B and D.

Fig. 3. The procedure followed in the CA-JAM

another channel to avoid interference and most likely switches

a channel that is common with one of its neighbors. Meanwhile,

tion B finds which interface of station A lost a link. Using this infor-

tion, station B adds the channel of the second and third interfaces

he backup and candidate channel list (L-BCH = { 2 } and L-CCH =}). Because there is no other neighbor on this channel at the time

amming, both lists only contain a single element, which implies

se channels directly become backup BCH(B1) = 2 and candidate

(B1) = 8 channels. Hence, station B switches its interface one

channel 2. It is important to note that, if station D had not left

nnel 2, the interference in that channel could have been severe.

hough it is not shown here station A follows the same procedure

B; therefore, interface 1 of station A switches to channel 7 to

igate the jammer effect. As depicted in Fig. 3e, CA-JAM results in

ore connected and less interfered network at the end of channel

ignment and jammer mitigation. Finally, Fig. 3f shows the updates

he Neighbor_Table and Interface_Table for stations B and D.

When the above process is performed repeatedly, neighbors per

erface will be equal to one. This shows that CA-JAM can find non-

erfering tuples, similar to the MDCG, without the help of a central

ity or use of CCC and manages to mitigate jamming. The CA-JAM

orithm fulfills the design considerations: interface-channel map-

g, connectivity, and interference avoidance. However, not every

tion transmits and receives all the times. Thus, it is not impor-



7

6

2

5

6

8

1

1

8

2

B C D E

6

8

22

2

5

7

5

1

tations A and B avoid jamming attack; Sinceon D has multiple links with neighbors C andavoids congestion.

NbrID_on_InterfaceInterface Table of B

1 2 3A A C

NbrID_on_InterfaceInterface Table of D

1 2 3C E 0

Table update from sta-

ithm with three interfaces per station.

t to have only one neighbor per each interface just to find non- 5

erfering tuples. In addition, stations can benefit from having a few 5

ghbors rather than only one. Therefore, it is essential to determine 5

many neighbors per each interface (dopt) is sufficient. In the next 5

tion, we will determine and show that one neighbor per interface 6

ot the ideal solution through simulation. 6

Performance evaluation 6

In this section, we discuss experimental results obtained from the 6

NET simulator. Table 8 presents our simulation parameters. As per 6

ffic, each station is configured with three traffic types and any 6

tion can choose to transmit at any time unless its beacon dura- 6

time. During transmission time stations transmit one packet of 6

h traffic type; however, the next transmission time is determined 6

traffic distribution. Unless specified otherwise, the distribution 6

packet inter arrival time remains uniform for the entire evalua- 6

. In constant distribution the inter arrival is fixed whereas in uni- 6

m distribution the inter-arrival time is varied between the min and 6

x value.Since we modified 802.11a standard in OPNET, to show 6

improvement by using multiple interfaces, we compared the 6

gle-interface scenario with the multi-interface using the legacy 6

E 802.11a protocol. In Fig. 4, we present the network throughput 6

function of the number of stations for the IEEE 802.11a protocol 6


i.org/10.1016/j.comcom.2015.09.003




30 35 40 45 50

5M

10M

15M

20M

25M

30M

35M

40M

Thr

ough

puti

nbp

s

Number of stations

Number of interfaces1 2 3

Fig. 4. Throughput comparison of a single and multiple interfaces of IEEE 802.11 sta-

tions as the number of stations increases.

0 1 2 3 4 5 6 7 8 9 10 11 12 1315M

20M

25M

30M

35M

40M

45M

50M

55M

60M

65M

Thr

ough

puti

nbp

s

Number of neighbors per interface

Number of stations30 35 40 45 50

Fig. 5. Throughput as the number of neighbors per interface increases for various

numbers of stations.

Table 8

Simulation parameters.

re618

ed619

n-620

621

of622

as623

ng624

-625

h-626

a-627

628

er629

nd630

in631

he632

0 1 2 3 4 5 6 7 8 9 10 11 1210M

15M

20M

25M

30M

35M

40M

45M

50M

55M

60M

65M

Thr

ough

puin

bps


Number of jammed channels0 1 2 3

Fig. 6. Throughput as the number of neighbors per interface increases for various

numbers of jammed channels (45 stations).

ber

ith

er, 633

is 634

al- 635

er 636

ps 637

e- 638

of 639

ve 640

of 641

u- 642

to 643

644

f- 645

u- 646

la- 647

ed 648

an 649

Parameters Values

Station type 802.11a (can be 802.11 n or ac)

Network deployment area 800 m x 800 m

Transmission range 150 m

Mobility Random way point

Number of channels 6

Number of interfaces 3

Simulation time 5 min

Routing protocol AODV

Traffic Type Email, VoIP, Low resolution video

Traffic generation pattern Uniform in [1,110]

Data rate 54 Mbps

Ground speed 5 m/sec

Beacon interval 0.5 s

Beacon duration 0.05 s

Jammer transmission power 20 W

Jammer packet inter-arrival time 1 ms

Jammer packet size 1024 bits

to show the impact of using multiple interfaces. When stations we

equipped with multiple interfaces, the network throughput increas

compared with a single interface. This ensures the use of multiple i

terfaces increases the capacity of the network.

Fig. 5 depicts network throughput as a function of the number

neighbors per interface. The result showed that the throughput w

maximum when the number of neighbors was three. It is interesti

to note that according to this result, in MANETs, increasing the num

ber of station for a given area did not always provide better throug

put, i.e., the throughput of 45 stations was better than that of 50 st

tions as shown in Fig. 5.

To see the effect of jamming attack on the ideal neighbor numb

per interface of the network, we kept the number of stations 45 a

varied the number of jammed channels. The results are compared

Fig. 6. As the number of jammers increased from zero to three, t



0 1 2 3 4 5 6 7 8 9 10 11 12

2.6

2.8

3.0

3.2

3.4

3.6

3.8

4.0

4.2

4.4

4.6

Hop

sfr

omso

urce

tode

stin

atio

n



Fig. 7. Number of hops a packet travels from a source to a destination as the num

of neighbors per interface increases.

30 35 40 45 50

10M

20M

30M

40M

50M

60M

70M

80M

90M

100M

110M

120M

Thr

ough

puti

nbp

s

Number of stations

Uni_AODVCon_AODVPois_AODVUni_OLSRCon_OLSRPois_OLSR

Fig. 8. Throughput of CA-JAM for uniform, constant, and poisson distributions w

OLSR and AODV routing protocols..

ideal number of neighbors per interface remained three; howev

the throughput decreased as the number of jammers increased. Th

explains that CA-JAM adapted to jamming attacks very well. To v

idate that a MIMC MANET benefits from assigning the ideal numb

of neighbors per interface, we measured the average number of ho

taken by data packet before it reach its destination (see Fig. 7). Mor

over, the connectivity of the network was better when the number

neighbors per interface was three (dopt = 3). From Figs. 5–7, we ha

learned that the network performs better when the ideal number

neighbors per interface is three. Therefore, for the remaining eval

ation, the number of neighbors per interface for CA-JAM was set

three, dopt = 3.

Then we evaluated the performance of CA-JAM for different tra

fic generation patterns and routing protocols as shown in Fig. 8. Int

itively, we had chosen AODV as a routing protocol for CA-JAM simu

tion, since it incurs less routing information exchange. We consider

the traffic patterns as Poisson and Constant distribution with me






30 35 40 45 50

10M

20M

30M

40M

50M

60M

Thr

ough

puti

nbp

s

Number of stations

CA-JAMMDCGRANDOM802.11 MAC

Fig. 9. Throughput comparison of CA-JAM with other schemes, as a function of num-

ber of stations.

0 1 2 3 4 5 60

10M

20M

30M

40M

50M

60M

Thr

ough

puti

nbp

s

Number of jammed channels


Fig. 10. Throughput comparison of CA-JAM with other schemes, as a function of num-

ber of jammed channels (45 stations).

2 as well as Uniform distribution during interval [1,110]. We used650

routing protocols, AODV and OLSR. When we compared the routing651

protocols under the same traffic pattern, AODV showed a better per-652

formance than OLSR in all traffic distributions. This motivated us to653

choose AODV as a routing protocol for most of the simulation. How-654

ever, CA-JAM has flexibility to work with any routing protocol . On the655

other hand, we simulated with different traffic types to see if CA-JAM656

reacts to load variation. Fig. 8 evidently shows that CA-JAM can adapt657

to v658

loa659

wo660

its661

662

for663

as664

MIM665

tion666

ass667

the668

the669

the670

Fig671

Wh672

of s673

num674

dom675

wil676

afte677

678

sig679

eve680

thr681

as t682

30 35 40 45 50

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Del

ayin

seco

nd

Number of stations


Fig. 11. Delay comparison of CA-JAM with other schemes, as a function of number of

stations.

0.9

1.0

1.1

1.2

nco

unt


Fig.

tion

jam 83

sec 84

tho 85

put 86

net 87

88

sch 89

ble 90

sta 91

sys 92

foll 93

mo 94

the 95

fere 96

lea 97

cas 98

avo 99

diff 00

802 01

ver 02

a d 03

com 04

the 05

wit 06

bet 07

JAM 08

cha 09

Pl

fa

arious traffic types, which shows how CA-JAM could scale as the

d in the network increases or decreases. Thus, CA-JAM is proved to

rk under various traffic loads and routing protocols, which shows

flexibility and scalability.

The performance of CA-JAM was compared with other algorithms

multi-interfaced stations. We compared CA-JAM to MDCG as well

the IEEE 802.11a standard. This is because MDCG captures the

C network environment compared to the rest of works men-

ed in Section 2. Besides, CA-JAM was also compared with random

ignment to show that simple channel assignment does not give

best solution. For the random channel assignment, we modified

IEEE 802.11 system to assign channels randomly. We compared

ir throughput, number of re-transmissions, and delay. As shown in

. 9, the throughput of CA-JAM was higher than the other schemes.

ile random assignment showed a better result for a small number

tations, it was neither consistent nor better than CA-JAM as the

ber of stations increased. In addition, this does not mean ran-

is better than CA-JAM in terms of over all performance, which

l be shown when we discuss jamming, delay and retransmission

rwards.

Following that, we measured the performance of all channel as-

nment schemes under jamming attack; jammers jam the network

ry 1ms and send a packet of 1024 bits. As shown in Fig. 10, the

oughput of CA-JAM was always highest and decreased gradually

he number of jammed channels increased. Since jammers can not

thr 10

mis 11

dom 12

tac 13

con 14



30 35 40 45 50

0.4

0.5

0.6

0.7

0.8

Ret

rans

mis

sio

Number of stations

12. Retransmission caused by CA-JAM compared with other schemes, as a func-

of number of stations.

the entire area and due to the data transmitted between con- 6

utive jamming trials, some throughput was still achievable even 6

ugh all six available channels were jammed. In fact, the through- 6

of CA-JAM was twice higher than the other schemes, when the 6

work was under jamming attack. 6

We also compared the delay performance of CA-JAM with other 6

emes. As depicted in Fig. 11, the delay of CA-JAM was always sta- 6

and low, whereas the delay of IEEE 802.11 was very high and un- 6

ble because there is no means of controlling interference in 802.11 6

tem. When the number of stations are varied, 802.11 system only 6

ows the network setup, when the setup is suitable the network is 6

re connected and less interfered, which leads to lower delay. On 6

other hand, in unsuitable network set up stations in 802.11 inter- 6

most of the time since they reside in the same channel which 6

ds to transmission delay due to an unsent packet. This is not the 6

e for MDCG and CA-JAM because these schemes have interference 6

idance mechanism where as in random scheme stations reside in 6

erent channel, which makes it less prone to delay compared to 7

.11 systems. Moreover, the delay of CA-JAM only increased by a 7

y small margin, even when the number of stations increased. In 7

isconnected network, the number of re-transmission attempts be- 7

es very high, because stations did not have many paths. In Fig. 12, 7

number of re-transmission attempts showed similar behavior 7

h the delay. These results confirmed that the network achieved 7

ter connectivity and exhibited smaller interference under the CA- 7

assignment. In Fig. 9 the random assignment had a favorable 7

nnel selection when the number of nodes as small. However, the 7

oughput was enhanced at the cost of higher delay and retrans- 7

sion as shown in Figs. 11 and 12 respectively. In addition, ran- 7

assignment exhibits a lower performance under jamming at- 7

k. This means that, even when random selection has the favorable 7

ditions, the overall performance of CA-JAM is better in terms of 7


i.org/10.1016/j.comcom.2015.09.003




0.5 1.0 1.5 2.0 2.5 3.015M

20M

25M

30M

35M

40M

45M

50M

55M

60M

65M

Thr

ough

puti

nbp

s

Beacon interval in s


Fig. 13. Throughput as a function of beacon interval.

0 1 2 3 4 5 6 7 8 9 1025M

30M

35M

40M

45M

50M

55M

Thr

ough

puti

nbp

s

Number of channels


Fig. 14. Throughput as a function of number of available channels.

delay, retransmission and throughput as well as under jamming at-715

tack.716

Finally, since our work was based on beacon exchange and mul-717

tiple channels, we investigated the effect of these parameters on718

the network throughput. Although a longer beacon interval provided719

some benefit to the network with a higher number of stations, it720

did the opposite for a network with a smaller number of stations, as721

shown in Fig. 13. This means, the overhead created by short beacon722

interval is small. The longer intervals results in inconsistency rather723

a-724

725

nt726

n-727

a728

n-729

th730

a-731

s732

he733

en734

he735

as736

a737

ut738

er,739

r-740

il-741

n-742

to743

nt,744

is745

er746

747

6. Conclusion 748

MANETs can be easily deployed to provide essential help in a mil- 749

itary and disaster relief networks. However, the presence of jam- 750

mers, harsh environment of war zone or the absence of central en- 751

tity, such as access point, affect the performance of such networks. 752

The performance of MANETs can be enhanced by deploying stations 753

with multiple interfaces that operate on multiple channels. While 754

the network benefits from multiple interfaces, it still had limita- 755

tions due to poor channel utilization. To overcome this issue, we 756

proposed CA-JAM algorithm, which is a channel assignment scheme 757

that considers jammer mitigation as well. Our simulation results 758

showed that CA-JAM provided higher throughput with reasonable 759

delay and re-transmission. The results confirmed that the proposed 760

algorithm could greatly enhance the network performance; thus 761

making MANETs feasible to use in any desired environment. 762

Although our focus has been channel assignment and current 763

routing protocols are designed for single interfaced MANETS, there 764

is a room to enhance the performance by jointly integrating it with a 765

routing protocol. Our future work will integrate channel assignment 766

with routing to achieve a better performance and fully utilize MIMC 767

feature. Also, we plan to implement our work with off the shelf 802.11 768

products, to test the performance of CA-JAM in the real environments. 769

References 770

[1] Part 11: wireless lan medium access control (mac) and physical layer (phy) spec- 771ifications amendment 1: high-speed physical layer in the 5 ghz band, 2000, 772ISO/IEC 8802-11:1999/Amd 1:2000(E); IEEE Std 802.11a-1999, pp. i–83. 773

[2] J. Redi, R. Ramanathan, The DARPA WNaN network architecture, in: Military Com- 774munications Conference, 2011 - MILCOM 2011, 2011, pp. 2258–2263. 775

[3] G. Noubir, On connectivity in ad hoc networks under jamming using directional 776antennas and mobility, Wired/Wireless Internet Communications, Springer Berlin 777Heidelberg, 2004, pp. 186–200. 778

[4] I.F. Akyildiz, W.-Y. Lee, M.C. Vuran, S. Mohanty, Next generation/dynamic spec- 779trum access/cognitive radio wireless networks: A survey, Comput. Netw. vol. 50 780(2006) 2127–2159. 781

[5] N.V.M. Simone, M. Zorzi, Resource allocation in multi-radio multi-channel wire- 782less ad hoc networks, Online avaialble: http://www.dei.unipd.it/merlo/papers/ 783mc4.pdf. Q3784

[6] R. Paul, Y.Z. Jembre, Y.J. Choi, Multi-interface rendezvous in self-organizing cogni- 785tive radio networks. in dynamic spectrum access networks, in: IEEE International 786Symposium on (DYSPAN), 2014, pp. 531–540. 787

[7] H. Liu, Z. Lin, X. Chu, Y.W. Leung, Jump-stay rendezvous algorithm for cognitive 788radio networks, IEEE Trans. Parallel Distrib. Syst. 23 (10) (2012) 1867–1881. 789

[8] M.A. L. C. R. Rajan, M.G. Chandra, P. Hiremath, Concepts of graph theory relevant 790to ad-hoc networks, Int. J. Comput. Commun. Control 3 (2008) 465–469. 791

t. J. 792793

et- 794irst 795

796are 79706. 798gs, 799

800for 801ms 802

803in 804

na- 8058– 806

807as- 80809) 809

810lo- 81105. 8128. 813el, 814

nal 815ac- 816

817in 818

819nd 820gs 821

ion 822823

than smaller overheads. Therefore, the selection of making the be

con interval 1 ms was a desirable choice.

In addition, multiple channels with the right channel assignme

improved the capacity of MANETs. However, even with a good cha

nel assignment scheme, too many channels did not always create

better throughput. This was because stations took more time for re

dezvous. Even when a rendezvous was achieved it was done wi

small number of neighbors per interface because interfaces of st

tions were distributed over many available channels. Fig. 14 show

the effect of number of channels in the network. For CA-JAM, t

throughput increased until the number of channels was six, and th

it started to decrease again. For the MDCG and random schemes, t

highest throughput was achieved when the number of channels w

seven. Despite the fact that MDCG and random assignments had

maximum throughput at this number of channels, the throughp

of both schemes is still lower than the proposed CA-JAM. Howev

when the number of channels is fewer than the number of inte

faces, the random scheme outperforms CA-JAM. Although the ava

able channels cannot be able to accommodate any more partitio

ing, due to the working principles of CA-JAM, it requires stations

change channels frequently such that there is a balanced assignme

which is not the case in the random scheme. Therefore, CA-JAM

suitable for MANETs where the number of channels is always great

than the number of interfaces.



[9] N. Meghanathan, Graph theory algorithms for mobile ad hoc networks, Inf. In

Comput. Inf. 36 (2008) 185–199.[10] Y. Xu, X. Zhong, S. Zhou, Channel assignment in multi-radio wireless ad hoc n

works, in: Communications and Networking in China, 2006. ChinaCom ’06. F

International Conference on, 2006, pp. 1–5.[11] K. Ramachandran, E. Belding, K. Almeroth, M. Buddhikot, Interference-aw

channel assignment in multi-radio wireless mesh networks, in: INFOCOM 2025th IEEE International Conference on Computer Communications. Proceedin

2006, pp. 1–12.[12] C.Z.H. Li, Y. Cheng, P. Wan, Multi-dimensional conflict graph based computing

optimal capacity in mr-mc wireless networks, in: Distributed Computing Syste

(ICDCS), 2010 IEEE 30th International Conference on, 2010, pp. 774–783.[13] J. Tang, G. Xue, W. Zhang, Interference-aware topology control and qos routing

multi-channel wireless mesh networks, in: Proceedings of the 6th ACM Intertional Symposium on Mobile Ad Hoc Networking and Computing, 2005, pp. 6

77.[14] W. Li, P. Fan, K.B. Letaief, Joint processing of topology control and channel

signment in wireless ad hoc networks, Wirel. Commun. Mob. Comput. 9 (20

269–281.[15] J. Zhao, H. Zheng, G.-H. Yang, Distributed coordination in dynamic spectrum al

cation networks, in: New Frontiers in Dynamic Spectrum Access Networks, 20DySPAN 2005. 2005 First IEEE International Symposium on, 2005, pp. 259–26

[16] V. Raman, N.H. Vaidya, Interference aware channel allocation in a multichannmulti-interface wireless network, in: Proceedings of the Third ACM Internatio

Workshop on Wireless Network Testbeds, Experimental Evaluation and Charterization, 2008, pp. 111–112.

[17] M.X. Gong, S.F. Midkiff, S. Mao, A cross-layer approach to channel assignment

wireless ad hoc networks, Mob. Netw. Appl. 12 (2007) 43–56.[18] L. Ding, T. Melodia, S. Batalama, M.J. Medley, Rosa: Distributed joint routing a

dynamic spectrum allocation in cognitive radio ad hoc networks, in: Proceedinof the 12th ACM International Conference on Modeling, Analysis and Simulat

of Wireless and Mobile Systems, 2009, pp. 13–20.



http://www.dei.unipd.it/merlo/papers/mc4.pdf




[19] H.S. Chiu, K. Yeung, K.-S. Lui, J-car An efficient joint channel assignment and rout-824ing protocol for ieee 802.11-based multi-channel multi-interface mobile ad hoc825networks, Wirel. Commun. IEEE Trans. 8 (2009) 1706–1715.826

[20] M. Gong, S. Midkiff, S. Mao, Design principles for distributed channel assignment827in wireless ad hoc networks, Commun. ICC IEEE Int. Conf. 5 (2005) 3401–3406.828

[21] S.H. Kim, D.W. Kim, Y.J. Suh, A group-based channel assignment protocol for rate829separation in ieee 802.11-based multi-radio multi-rate ad hoc networks, Ad Hoc830Netw. 10 (1) (2012) 95–110.831

[22] H. Cheng, N. Xiong, A.V. Vasilakos, L.T. Yang, G. Chen, X. Zhuang, Nodes organiza-832tion for channel assignment with topology preservation in multi-radio wireless833mesh networks, Ad Hoc Netw. 10 (5) (2012) 760–773.834

[23] Y. Peng, Y. Yu, L. Guo, D. Jiang, Q. Gai, An efficient joint channel assignment and835qos routing protocol for ieee 802.11 multi-radio multi-channel wireless mesh net-836works, J. Netw. Comput. Appl. 36 (2) (2013) 843–857.837

[24] D.J. Deng, Y.S. Chen, Y.S. Wong, Adaptive channel allocation strategy for mobile ad838hoc networks, Math. Comput. Model. 57 (11) (2013) 2720–2730.839

[25] Y. Ding, K. Pongaliur, L. Xiao, Channel allocation and routing in hybrid multichan-840nel multiradio wireless mesh networks, Mobile Comput. IEEE Trans. 12 (2) (2013)841206–218.842

[26] Y. Wang, J.J. Garcia-Luna-Aceves, A distributed cross-layer routing protocol with843channel assignment in multi-channel MANET, Computing, Networking and Com-844munications (ICNC), International Conference on, IEEE, 2015, pp. 1050–1054.845

[27] Y. Ding, L. Xiao, Channel allocation in multi-channel wireless mesh networks,846Comput. Commun. 34 (7) (2011) 803–815.847

[28] L. Hu, Distributed code assignments for CDMA packet radio networks, in: INFO- 848COM ’91. Proceedings. Tenth Annual Joint Conference of the IEEE Computer and 849Communications Societies. Networking in the 90s., IEEE, 3, 1991, pp. 1500–1509. 850

[29] A.A. Bertossi, M.A. Bonuccelli, Code assignment for hidden terminal interference 851avoidance in multihop packet radio networks, IEEE/ACM Trans. Netw. 3 (1995) 852441–449. 853

[30] B.W. Kernighan, L. Shen, An efficient heuristic procedure for partitioning graphs, 854Bell Syst. Tech. J. 49 (2) (1970) 291–307. 855

[31] K. Andreev, R. Harald, Balanced graph partitioning, Theory Comput. Syst. 39 (6) 856(2006) 929–939. 857

[32] Y.Z. Jembre, R. Paul, Y.J. Choi, K.Y. Cheon, C.J. Kim, Channel assignment and jam- 858mer mitigation for military MANETs with multiple interfaces and multiple chan- 859nels, Proceedings of the 8th International Conference on Ubiquitous Information 860Management and Communication, ACM, 2014, pp. 66–72.January 861

[33] D. Wu, S.H. Yang, L. Bao, C.H. Liu, Joint multi-radio multi-channel assignment, 862scheduling, and routing in wireless mesh networks, Wirel. Netw. 20 (1) (2014) 86311–24. 864

[34] K.H. Almotairi, X.S. Shen, A distributed multi-channel mac protocol for ad hoc 865wireless networks, Mobile Comput. IEEE Trans. 14 (1) (2015) 1–13. 866

[35] J. Jia, J. Chen, J. Yu, X. Wang, Joint topology control and routing for multi-radio 867multi-channel WMNS under SINR model using bio-inspired techniques, Appl. Soft 868Comput. 32 (2015) 49–58. 869

[36] L.H. Yen, K.W. Huang, Link–preserving interference–minimisation channel as- 870signment in multi–radio wireless mesh networks, Int. J. Ad Hoc Ubiquitous Com- 871put. 18 (4) (2015) 222–233. 872

Pl

fa




i.org/10.1016/j.comcom.2015.09.003


1-s2.0-s0140366415003230-main

Documents

single channel

channel assignment problem

distinct channel

networks manets

jamming effects

effect of jamming

tentional jamming

use of control channel