**Protocols**

## **Chapter 1**

## **Data-Gathering and Aggregation Protocol for Networked Carrier Ad Hoc Networks: The Optimal and Heuristic Approach** Provisional chapter Data-Gathering and Aggregation Protocol

Chao Gao, Guorong Zhao and Jianhua Lu The Optimal and Heuristic Approach

Additional information is available at the end of the chapter Chao Gao, Guorong Zhao and Jianhua Lu

for Networked Carrier Ad Hoc Networks:

http://dx.doi.org/10.5772/66641 Additional information is available at the end of the chapter

## Abstract

In this chapter, we address the problem of data-gathering and aggregation (DGA) in navigation carrier ad hoc networks (NC-NET), in order to reduce energy consumption and enhance network scalability and lifetime. Several clustering algorithms have been presented for vehicle ad hoc network (VANET) and other mobile ad hoc network (MANET). However, DGA approach in harsh environments, in terms of long-range transmission, high dynamic topology and three-dimensional monitor region, is still an open issue. In this chapter, we propose a novel clustering-based DGA approach, namely, distributed multiple-weight data-gathering and aggregation (DMDG) protocol, to guarantee quality of service (QoS)-aware DGA for heterogeneous services in above harsh environments. Our approach is explored by the synthesis of three kernel features. First, the network model is addressed according to specific conditions of networked carrier ad hoc networks (NC-NET), and several performance indicators are selected. Second, a distributed multipleweight data-gathering and aggregation protocol (DMDG) is proposed, which contains allsided active clustering scheme and realizes long-range real-time communication by tactical data link under a time-division multiple access/carrier sense multiple access (TDMA/ CSMA) channel sharing mechanism. Third, an analytical paradigm facilitating the most appropriate choice of the next relay is proposed. Experimental results have shown that DMDG scheme can balance the energy consumption and extend the network lifetime notably and outperform LEACH, PEACH and DEEC in terms of network lifetime and coverage rate, especially in sparse node density or anisotropic topologies.

Keywords: navigation carrier ad hoc networks (NC-NET), clustering protocol, data-gathering and aggregation (DGA), QoS aware

© 2017 The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

© The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons

Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and eproduction in any medium, provided the original work is properly cited.

## 1. Introduction

The emerging advantages of wireless network inspire other technical fields to solve their own bottlenecks through the network approach [1]. In our researches, we devote to put forward a network-aware solution for a bottleneck of modern navigation technology, which restricts it from stage of a higher level [2]. That is, the level of navigation efficiency is closely related to selfcontained degree of navigation system; on one hand, the upgrade of its integrity will increase the burdens on economic investment and physical load; on the other hand, if we simplify the complexity of navigation equipment to alleviate the above burdens, its navigation capacity will be degraded accordingly. In previous works [2–6], we have proposed a novel network architecture, namely, navigation carrier ad hoc networks (NC-NET), to handle the navigation-related issues, in terms of cooperative navigation, localization, target tracking and multimedia data exchange, through a network approach. The proposed NC-NET, which is essentially ad hoc network between navigation carriers (NCs), is surveyed as a new network family. Hereinto, navigation carrier is defined as the carrier that has the demands of localization and/or navigation, such as aircraft, car, ship, submarine, buoyage system, satellite or pseudo satellite, etc. In Refs. [2, 3], we have finished partial protocols and mechanisms as follows: (i) the protocol framework and the models in physical layer [2]; (ii) a diffserv-based dynamic cooperative MAC protocol, i.e., DDC-MAC [3]; and (iii) the network-based localization mechanisms [4–6]. As part of a series of research work, the main objective of this chapter is to develop a data-gathering and aggregation (DGA) protocol with full account of the unique challenges in NC-NET.

In previous literature, the DGA problem has been investigated extensively [7–12], to prioritize and manage the resource sharing according to the unique requirements of each traffic class. However, in DGA of NC-NET, we should integrate into account several unique challenges that never handled in existing work, including (i) coexistence of multiple traffic services with heterogeneous QoS requests; (ii) coexistence of short-range and long-range traffic requests; and (iii) harsh routing environment, i.e., sparse network density, long-range transmission and high dynamic topology.

Several clustering algorithms are presented for VANET such as [13–26]. However, these algorithms do not show how the routing is performed according to their clustering algorithms after the cluster formation. Hence, they do not guarantee the network topology during the routing process. Their clustering algorithms ignore as well the quality of service requirements important for safety, emergency and multimedia services. On the other hand, QoS-based clustering algorithms take into consideration the quality of service metrics such as bandwidth, energy and end-to-end delay to group the nodes. However, they ignore the high speed mobility metrics which makes them inefficient to deal with NC-NET. The optimized link state routing (OLSR) [15] is a proactive routing protocol that has been modeled to cope with mobile ad hoc networks (MANETs). Its basic idea is to elect a cluster head for each group of neighbour nodes and divide hence the network into clusters. These heads then select a set of specialized nodes called multipoint relays (MPRs). The function of the MPR nodes is to reduce the overhead of flooding messages by minimizing the duplicate transmissions within the same zone. QoS-OLSR [17] is an enhanced version of OLSR that extends the MANET network lifetime taking into consideration the available bandwidth and the residual energy per node during cluster head election and MPR node selection. Nonetheless, this protocol does not consider the mobility of nodes while computing the QoS. Thus, nodes with high bandwidth, energy and mobility may be elected as cluster heads which leads to recurrent disconnections. Likewise, the MPRs selected according to this protocol do not satisfy both mobility constraints and routing parameters (end-to-end delay and packet delivery ratio). Moreover, the MPR selection algorithm according to QoS-OLSR is vulnerable to cheating in the sense that some nodes may claim bogus QoS values in order to ensure being selected as MPRs. Furthermore, QoS-OLSR does not advance any MPR recovery algorithm able to select quick alternatives and keep the network connected in case of link failures.

1. Introduction

4 Ad Hoc Networks

dynamic topology.

The emerging advantages of wireless network inspire other technical fields to solve their own bottlenecks through the network approach [1]. In our researches, we devote to put forward a network-aware solution for a bottleneck of modern navigation technology, which restricts it from stage of a higher level [2]. That is, the level of navigation efficiency is closely related to selfcontained degree of navigation system; on one hand, the upgrade of its integrity will increase the burdens on economic investment and physical load; on the other hand, if we simplify the complexity of navigation equipment to alleviate the above burdens, its navigation capacity will be degraded accordingly. In previous works [2–6], we have proposed a novel network architecture, namely, navigation carrier ad hoc networks (NC-NET), to handle the navigation-related issues, in terms of cooperative navigation, localization, target tracking and multimedia data exchange, through a network approach. The proposed NC-NET, which is essentially ad hoc network between navigation carriers (NCs), is surveyed as a new network family. Hereinto, navigation carrier is defined as the carrier that has the demands of localization and/or navigation, such as aircraft, car, ship, submarine, buoyage system, satellite or pseudo satellite, etc. In Refs. [2, 3], we have finished partial protocols and mechanisms as follows: (i) the protocol framework and the models in physical layer [2]; (ii) a diffserv-based dynamic cooperative MAC protocol, i.e., DDC-MAC [3]; and (iii) the network-based localization mechanisms [4–6]. As part of a series of research work, the main objective of this chapter is to develop a data-gathering and

aggregation (DGA) protocol with full account of the unique challenges in NC-NET.

In previous literature, the DGA problem has been investigated extensively [7–12], to prioritize and manage the resource sharing according to the unique requirements of each traffic class. However, in DGA of NC-NET, we should integrate into account several unique challenges that never handled in existing work, including (i) coexistence of multiple traffic services with heterogeneous QoS requests; (ii) coexistence of short-range and long-range traffic requests; and (iii) harsh routing environment, i.e., sparse network density, long-range transmission and high

Several clustering algorithms are presented for VANET such as [13–26]. However, these algorithms do not show how the routing is performed according to their clustering algorithms after the cluster formation. Hence, they do not guarantee the network topology during the routing process. Their clustering algorithms ignore as well the quality of service requirements important for safety, emergency and multimedia services. On the other hand, QoS-based clustering algorithms take into consideration the quality of service metrics such as bandwidth, energy and end-to-end delay to group the nodes. However, they ignore the high speed mobility metrics which makes them inefficient to deal with NC-NET. The optimized link state routing (OLSR) [15] is a proactive routing protocol that has been modeled to cope with mobile ad hoc networks (MANETs). Its basic idea is to elect a cluster head for each group of neighbour nodes and divide hence the network into clusters. These heads then select a set of specialized nodes called multipoint relays (MPRs). The function of the MPR nodes is to reduce the overhead of flooding messages by minimizing the duplicate transmissions within the same zone. QoS-OLSR [17] is an enhanced version of OLSR that extends the MANET network lifetime taking into consideration the available bandwidth and the residual energy per node during cluster head election and MPR node selection. Nonetheless, this protocol does not With these comparisons, it is clear that the results on challenge (i), challenge (ii) and their hybrid are extensive; however, in harsh environment that of challenge (iii), the integrated investigates on challenges (i) and (ii) are quite limited. From above analysis, we identify the missing properties in the existing work for QoS provisioning in NC-NET and introduce the design and implementation of a new distributed multiple-weight data-gathering and aggregation protocol (DMDG) protocol for NC-NET. DMDG is designed with key features to support the above three challenges. These features include the following: first, the network model is addressed according to specific conditions of networked carrier ad hoc networks (NC-NET), and several performance indicators are selected. Second, a distributed multiple-weight data-gathering and aggregation protocol (DMDG) is proposed, which contains all-sided active clustering scheme and realizes long-range real-time communication by tactical data link under a TDMA/CSMA channel sharing mechanism. Third, an analytical paradigm facilitating the most appropriate choice of the next relay is proposed. To optimize the performance of DMDG, the aforementioned contributions are theoretically analysed and evaluated. Finally, the results demonstrate the efficiency of our protocol by comparing to other contemporary MANET routing protocols. Experimental results have shown that DMDG scheme can balance the energy consumption and extend the network lifetime notably and outperform LEACH, PEACH and DEEC in terms of network lifetime and coverage rate, especially in sparse node density or anisotropic topologies.

The remainder of this chapter is organized as follows. Section 2 introduces the network model and evaluates indicators of NC-NET. Section 3 presents the details of the proposed DMDG protocol, including the clustering algorithm, data aggregation and communication scheme and the three-dimensional routing algorithm. Section 4 presents the numerical analysis and simulation of our scheme, along with the corresponding results. Finally, in Section 5, we summarize this chapter with conclusions and prospect.

## 2. Network model and problem formulations

In the area of cluster-based wireless sensor networks, the previous research is quite extensive, with energy efficiency and scalability being the main focus of many of the clustering protocols proposed so far [6–10]. Meanwhile, much research has been done on sensor activation protocols, which focus on selecting a subset of the active sensor nodes that are sufficient to satisfy the network's coverage requirements. Compare to traditional ad hoc network, NC-NET has special characters in a couple of aspects, including high-powered links, high-speed nodes and sparse and very-large-scale working range. In this section, we discuss the related work that has been done taking into account these points.

## 2.1. Network model

In this subsection, we first introduce the characteristics of ad hoc network and then define the network model of navigation carrier ad hoc network. We model the ad hoc network as a graph, G = (C, E, Rc), which consists of a set of mobile nodes, C = {C1, C2, ⋯, Cm}, and a set of wireless links, E; each sensor node consists of components of sensing, computing and wireless transmission. Assume that each node has the same transmission radius Rt and the same sensing radius Rs. The notations used in this chapter are defined in Table 1.

Before discussing the nature of this problem, we give the following assumptions for the feasibility of the NC-NET, which include monitor area, transmission channel and communication environment.

Assumption 1 (Monitor area). The monitor area of NC-NET Ac is a cylinder, which is equivalent with the tradition definition of monitoring area.

$$A\_{\varepsilon} = \left\{ (\mathbf{x}, y, z) | \mathbf{x}^2 + y^2 \le \mathbf{R}\_{\text{cov}}, \mathbf{0} \le z \le \mathbf{h}\_{\text{cov}} \right\} \tag{1}$$

where Rcov is monitor radius and hcov is monitor height.

Assumption 2 (Signal link). The transmission path adopts tactical data link. To meet the requirement of real-time performance and reliability, Rc is defined as 1/3 – 1/5 ratio to the effective transmission radius of the link and assumed that the bit error rate (BER) is invariable.

Assumption 3 (Communication environment). The statistical property of the NLOS propagation accords with exponential distribution, while the time delay accords with Bernoulli distribution. The coupling and interference in electromagnetic environment are mutually independent white noise with zero mean.

Figure 1 is an example of NC-NET, in which the type of area coverage Ac<sup>−</sup><sup>1</sup> is considered. Ac<sup>−</sup><sup>1</sup> ⊂ Ac and dch<sup>−</sup>ch, dch<sup>−</sup>M, dch<sup>−</sup>unk, dch<sup>−</sup>ag represent the distance between neighbour cluster heads, cluster head to cluster member, cluster head to uncertain node and cluster head to agent


Table 1. Notation and definition in NC-NET.

Data-Gathering and Aggregation Protocol for Networked Carrier Ad Hoc Networks: The Optimal... http://dx.doi.org/10.5772/66641 7

Figure 1. Sketch map of NC-NET model.

2.1. Network model

6 Ad Hoc Networks

tion environment.

noise with zero mean.

Link<sup>i</sup> j

Typei

Notation Definition

Ac Work area of the NC-NET C = {C1, C2, ⋯, Cm} Set of NC-NET mobile nodes

ð Þ<sup>τ</sup> Signal links between node <sup>i</sup> and <sup>j</sup> at time <sup>τ</sup>

<sup>k</sup> Type of the node i in cluster k,Typei

Rmsg(i) Maximum transmission range of node i

E = {(i, j)|s(i, j) ≤ 2Rc} Set of wireless links

Rc(i) Sensing radius of node i Tlife(i) Lifetime of network sensor i

Table 1. Notation and definition in NC-NET.

In this subsection, we first introduce the characteristics of ad hoc network and then define the network model of navigation carrier ad hoc network. We model the ad hoc network as a graph, G = (C, E, Rc), which consists of a set of mobile nodes, C = {C1, C2, ⋯, Cm}, and a set of wireless links, E; each sensor node consists of components of sensing, computing and wireless transmission. Assume that each node has the same transmission radius Rt and the same sensing

Before discussing the nature of this problem, we give the following assumptions for the feasibility of the NC-NET, which include monitor area, transmission channel and communica-

Assumption 1 (Monitor area). The monitor area of NC-NET Ac is a cylinder, which is equiva-

Ac <sup>¼</sup> ð Þj <sup>x</sup>; <sup>y</sup>; <sup>z</sup> <sup>x</sup><sup>2</sup> <sup>þ</sup> <sup>y</sup><sup>2</sup> <sup>≤</sup> <sup>R</sup>cov, <sup>0</sup> <sup>≤</sup> <sup>z</sup> <sup>≤</sup> <sup>h</sup>cov

Assumption 2 (Signal link). The transmission path adopts tactical data link. To meet the requirement of real-time performance and reliability, Rc is defined as 1/3 – 1/5 ratio to the effective transmission radius of the link and assumed that the bit error rate (BER) is invariable. Assumption 3 (Communication environment). The statistical property of the NLOS propagation accords with exponential distribution, while the time delay accords with Bernoulli distribution. The coupling and interference in electromagnetic environment are mutually independent white

Figure 1 is an example of NC-NET, in which the type of area coverage Ac<sup>−</sup><sup>1</sup> is considered. Ac<sup>−</sup><sup>1</sup> ⊂ Ac and dch<sup>−</sup>ch, dch<sup>−</sup>M, dch<sup>−</sup>unk, dch<sup>−</sup>ag represent the distance between neighbour cluster heads, cluster head to cluster member, cluster head to uncertain node and cluster head to agent

CMsg ∈ G Structure message of NC-NET nodes C = (ID, HD, Type, XYcord, Nbn, CP, Nbs[])

(1)

<sup>k</sup>∈f g Memb;CH; agent; NR; toCH; toNR

radius Rs. The notations used in this chapter are defined in Table 1.

lent with the tradition definition of monitoring area.

where Rcov is monitor radius and hcov is monitor height.

node, respectively. 2Rc is the cluster radius, and Rmsg is the searching radius of uncertain node. Each interested grid has at least one sensor node within it, or the grid can be sensed by neighbour sensors.

## 2.2. Evaluating indicators

## 2.2.1. Coverage relevant model

Coverage is an important issue of ad hoc networks and is closely related to energy saving, connectivity and network reconfiguration [11–14]. The following is the main methods, which is adopted to assess coverage, comprising 3D k-coverage model and the necessary and sufficient condition (NSC) (Table 2).

Definition 1 (3D k-covered problem). Given a set of N sensors in area P, the radius of sensor node is Rc, if there is Ci1, …, Cik ∈ {C1, …, CN}, k ≥ 1. In this case, every position v in area P can meet the following formula.

$$w \in \bigcap\_{j=1}^{k} \left\{ \mathbf{x} |dist(\mathbf{x}, \mathbb{C}\_{ij}) \le \mathbb{R}\_{\mathbf{c}} \right\} \tag{2}$$

Then area P is 3D k-coverage by {C1, C2, ⋯, CN}.

Theorem 1 is the necessary and sufficient condition used to determine the connected coverage set, which has been proved by Zhang and Hou in literature [23].

Theorem 1. Given a set of N sensors C in three-dimensional area P, C = {C1, C2, ⋯, CN}, the communicating radius of sensor node is Rc; the sensing radius of sensor node is Rc ≥ 2Rs.


Input: Sensing range Rs, maximum transmission range Rt, sensor nodes set V, and ε0.

Table 2. Main clustering procedure pseudo-code.

∃ Ci ⊂ C, Ci = {Ci1, ⋯, Cik}. If sensing area of Ci can complete cover P, the set Ci = {Ci1, Ci2, ⋯, Cik} is a connected coverage set.

## 2.2.2. Link consumption indicator

Link quality is one of the key factors which affected the application of ad hoc network technique. In this chapter, we adopt tactical data link to extend it to solving the fleet-networkbuild-up problem, in which the following link consumption model is adopted.

In relatively short distances, the propagation loss problem is modelled as being inversely proportional to d<sup>2</sup> , whereas for longer distance, it is inversely proportional to the propagation to d<sup>4</sup> . Power control can be used to invert this loss by setting the power amplifier to ensure a certain power at the receiver [15–17]. Therefore, to transmit and to receive an l-bit packet in distance d, the radio expends the following energy, respectively:

$$\begin{aligned} E\_{\rm Tx}(l, d) &= l \cdot E\_{\rm Tx-elec} + E\_{\rm Tx-amp}(l, d) \\ &= \begin{cases} l \times E\_{elec} + l \times \varepsilon\_{ft} \times d^2 & d < d\_0 \\ l \times E\_{elec} + l \times \varepsilon\_{amp} \times d^4 & d \ge d\_0 \end{cases} \end{aligned} \tag{3}$$

$$E\_{\rm Rx}(l, d) = E\_{\rm Rx-elec}(l) = l \times E\_{elec} \tag{4}$$

where d<sup>0</sup> is the free space model and multipath fading model. ETx<sup>−</sup>elec is the electronics energy and depends on factors such as digital coding, modulation, and filtering of the signal before it is sent to the transmit amplifier. The parameters εft and εamp depend on the required sensitivity and the noise figure of the receiver [12].

For the experiments described in this chapter, we adopted the parameters of the radio chips similar to those in [2] to determine the parameter values in (1) and (2). Then we have the representative values of the parameters εamp = 0.0013 pJ/bit/m<sup>4</sup> , Eelec = 50 nJ/bit and εft = 10 pJ/bit/m<sup>2</sup> .

## 3. Distributed multi-weight data-gathering and aggregation protocol

In this section, we present the distributed multiple-weight data-gathering and aggregation protocol (DMDG) after describing the network model and evaluating indicator adopted, which consists of three parts as described in the following subsection.

## 3.1. Clustering algorithm

∃ Ci ⊂ C, Ci = {Ci1, ⋯, Cik}. If sensing area of Ci can complete cover P, the set Ci = {Ci1,

Link quality is one of the key factors which affected the application of ad hoc network technique. In this chapter, we adopt tactical data link to extend it to solving the fleet-network-

In relatively short distances, the propagation loss problem is modelled as being inversely

. Power control can be used to invert this loss by setting the power amplifier to ensure a

, whereas for longer distance, it is inversely proportional to the propagation

build-up problem, in which the following link consumption model is adopted.

Input: Sensing range Rs, maximum transmission range Rt, sensor nodes set V, and ε0.

1. for all u,v∈V(G), i∈NEW(G) do

4. Subprogram GetFirst\_CHs();

5. end-while

13. end-if

15. CHs.{} ⇌ Ags.{};

21. end-while

23. end-for

2. while round==1//NC-NET initialization 3. C[v].{ID, Type} ←sensor[i].{PRI, Power}

6. while round==0//NC-NET regular circuit 7. if(C.new∩V(G)=Ø||C[i]∩V(G)=Ø) 8. Subprogram NEW\_DEAD(); 9. elseif(toCH[v]∪toAgent[v]≠Ø) 10. Subprogram RENEW\_agent(); 11. Subprogram RENEW\_CHs();

12. CHs[v]←NEW\_CHs[v]; AGs[v]←NEW\_AGs[v];

14. CHs[u] ←CHs[v]; AGs[u] ←AGs[v];

18. while(toPR[v]≠Ø&round==0) 19. Subprogram RENEW\_PRs();

16. BroadcastMsg(Ags[u].{ID,nbn,Level,CP}); 17. BroadcastMsg(CHs[u].{ID,nbn,Level,CP});

20. BroadcastMsg(C[i].{ID,nbn,Level,CP})

22. BroadcastMsg(C[i].{ID,nbn,Level,CP});

Output: agent, CCH, CPR.

8 Ad Hoc Networks

Ci2, ⋯, Cik} is a connected coverage set.

Table 2. Main clustering procedure pseudo-code.

2.2.2. Link consumption indicator

proportional to d<sup>2</sup>

to d<sup>4</sup>

The hierarchical clustered sensor network is composed of a number of clusters. The following are some important definitions used in the clustering algorithm.

Definition 2 (Initial probability for cluster head)

$$p\_{\rm init} = \beta \cdot ID\_i / ID\_{\rm max} + \left(1 - \beta\right) \cdot GDOP\_i / GDOP\_{\rm max} \tag{5}$$

where IDi is the initial priority of sensor node i, IDmax is the maximum priority within the spherical radius rc, GDOPi is the geometric distribution of position (GDOP) of sensor node i, GDOPmax is the optimum GDOP within the spherical radius rc, and β is a self-adapting weighing factor.

Definition 3 (Electing factor for cluster head, EFCH)

$$\mathcal{W}\_i = \alpha \cdot \left( E\_{i\uparrow} / \overline{E}' \right) + (1 - \alpha) / 2 \cdot \frac{d\_{\text{max}} - d\_{\text{toBS}}^i}{d\_{\text{max}} - d\_{\text{toBS}}} + (1 - \alpha) / 2 \cdot GDP\_k \tag{6}$$

where α is a self-adapting weighing factor, α = 1/(2 + β), (β = Eij/Ē<sup>r</sup> ). Eij is the mean energy value of communicated consumption between sensor node Ci and other nodes in the same cluster, Ē<sup>r</sup> = Etotal(r)/m. d<sup>i</sup> toBS is the distance between Ci and cluster head node CHk. dtoBS is the distance between CHk and the cluster centre, dtoBS ¼ ∑ m i¼1 di toBS=m. GDPk is the GDOP of cluster head node CHk.

Definition 4 (Correlation among sensor nodes). ∀ Ci, Cj ∈ G(V), ∃ R(Ci, Cj) ∈ R<sup>+</sup> , i ≠ j, if ‖D (Ci, Cj)‖ ≤ r, ‖R(Ci, Cj)‖ = 1; else if ‖D(Ci, Cj)‖ > r, ‖R(Ci, Cj)‖ = n + 1, where n is the sensor number between Ci and Cj, r is the average 1-hop distance, and then R(Ci, Cj) is called as the correlation between node Ci and Cj.

Now we describe our cluster formation algorithm in detail. The algorithm consists of three stages. In the initial stage, NC-NET nodes initiate the clustering procedure (Lines 2–5), round=1; NC-NCT nodes sensor[i] join into the network and obtain their own IDs and Types, which are proportional to PRI and power of their link terminal; then run the subprogram GetFirst\_CHs() to elect the first cluster head and agent, which is direct ratio to ID and Type.

When NC-NET is formed, round = 0, the program of regular circuit (Lines 7–21) will be performed. In the beginning of every cycle, cluster heads should judge whether there are new-coming requisitions C.new or new-dead node and judge whether there are new requisition for cluster head to CH[v] and for agent to Agent[v] and then execute relative treating subprograms NEW\_DEAD(), RENEW\_agent() and RENEW\_CHs(), in which the first one is related to IDs and Types and the last two subprograms are mainly determined by pinit and Wi which is defined in Definition 2 and Definition 3. When CHs[i] is selected as a cluster head, it broadcasts message Msg() to all other nodes to indicate its identity and adjust TDMA format. The agent and cluster head also communicate every cycle to back up information to preserve robustness and stability of the network. The procedure also handles the request for reference nodes RENEW\_PRs() to realize renew of the network and assure load balance of the clustering protocol.

## 3.2. Data aggregation and communication scheme

This subsection presents the data aggregation mechanism and the communication mode among NC-NET nodes. The data transmitted among network members mainly has four portions, including the communication between reference nodes CPR and cluster head CCH, cluster head CCH and the neighbour cluster head, cluster head CCH and the cluster member CCM and cluster member CCM and the 2-hop neighbour node. Each transmission process has its own packet mode, including broadcast, data-centric, local transmission and multicast. The data packets have a unified scheme, main portions of which are shown in Table 3. Mainly composed of synchronous head, node's structural information CMsg, sphere coordinate coefficient TSCA, relative velocity VC, heading angle ϕ<sup>C</sup> and system time tA. The received variables should accord with transmission demand and uniform formats, and the redundant information should be neglected.

• NR-CH: The data packet between reference node and cluster head node CCH has three kinds of process: (i) the position reference nodes CPR broadcast global coordinate information and the network distributing variable to cluster head nodes CCH; (ii) the time reference


Table 3. The format of data packet.

Ē<sup>r</sup> = Etotal(r)/m. d<sup>i</sup>

node CHk.

10 Ad Hoc Networks

protocol.

tion should be neglected.

between CHk and the cluster centre, dtoBS ¼ ∑

3.2. Data aggregation and communication scheme

correlation between node Ci and Cj.

toBS is the distance between Ci and cluster head node CHk. dtoBS is the distance

toBS=m. GDPk is the GDOP of cluster head

, i ≠ j, if ‖D

m i¼1 di

(Ci, Cj)‖ ≤ r, ‖R(Ci, Cj)‖ = 1; else if ‖D(Ci, Cj)‖ > r, ‖R(Ci, Cj)‖ = n + 1, where n is the sensor number between Ci and Cj, r is the average 1-hop distance, and then R(Ci, Cj) is called as the

Now we describe our cluster formation algorithm in detail. The algorithm consists of three stages. In the initial stage, NC-NET nodes initiate the clustering procedure (Lines 2–5), round=1; NC-NCT nodes sensor[i] join into the network and obtain their own IDs and Types, which are proportional to PRI and power of their link terminal; then run the subprogram GetFirst\_CHs() to elect the first cluster head and agent, which is direct ratio to ID and Type.

When NC-NET is formed, round = 0, the program of regular circuit (Lines 7–21) will be performed. In the beginning of every cycle, cluster heads should judge whether there are new-coming requisitions C.new or new-dead node and judge whether there are new requisition for cluster head to CH[v] and for agent to Agent[v] and then execute relative treating subprograms NEW\_DEAD(), RENEW\_agent() and RENEW\_CHs(), in which the first one is related to IDs and Types and the last two subprograms are mainly determined by pinit and Wi which is defined in Definition 2 and Definition 3. When CHs[i] is selected as a cluster head, it broadcasts message Msg() to all other nodes to indicate its identity and adjust TDMA format. The agent and cluster head also communicate every cycle to back up information to preserve robustness and stability of the network. The procedure also handles the request for reference nodes RENEW\_PRs() to realize renew of the network and assure load balance of the clustering

This subsection presents the data aggregation mechanism and the communication mode among NC-NET nodes. The data transmitted among network members mainly has four portions, including the communication between reference nodes CPR and cluster head CCH, cluster head CCH and the neighbour cluster head, cluster head CCH and the cluster member CCM and cluster member CCM and the 2-hop neighbour node. Each transmission process has its own packet mode, including broadcast, data-centric, local transmission and multicast. The data packets have a unified scheme, main portions of which are shown in Table 3. Mainly composed of synchronous head, node's structural information CMsg, sphere coordinate coefficient TSCA, relative velocity VC, heading angle ϕ<sup>C</sup> and system time tA. The received variables should accord with transmission demand and uniform formats, and the redundant informa-

• NR-CH: The data packet between reference node and cluster head node CCH has three kinds of process: (i) the position reference nodes CPR broadcast global coordinate information and the network distributing variable to cluster head nodes CCH; (ii) the time reference

Definition 4 (Correlation among sensor nodes). ∀ Ci, Cj ∈ G(V), ∃ R(Ci, Cj) ∈ R<sup>+</sup>

nodes CTR broadcast system time tA to CCH and other network member; (iii) CCH uploads its own sensing parameter and cluster member dynamic information to neighbour CPR, which can be used for the next reference nodes' election.


## 3.3. Three-dimensional routing algorithm

When the network is clustered, specific methods for intra-cluster and intercluster communications depend on applications. For intra-cluster communication, the nodes can directly send data to the cluster head using time-division multiple-access (TDMA) schedule, just as in LEACH [7].

As shown in Figure 2, when node B broadcasts a packet, all the one-hop neighbours receive that packet. In order to avoid recipient ambiguity in a local broadcast, the incoming label of a node must be unique at least in its 2-hop neighbours. In the label assignment phase, the cluster head must execute an algorithm to ensure that each node is allocated a unique incoming label. A 3D heuristic routing algorithm is proposed for this purpose, which can be expatiated in the following example.

• Step 1: When node A receives the request packet from the node B, firstly, estimate whether the distance between nodes B and A (dAB) satisfies the responding condition, dAB ≤ 2 Rc, and calculate the number of optimum hops (Khop) if the condition is satisfied; if not, ignore the request. As shown in Figure 2, Khop = 3.

Figure 2. Geographic routing algorithm in NC-NET.


## 4. Performance evaluation and simulation

To evaluate the performance of our scheme and compare it to both the association sponsor and central angle method, a set of simulations have been carried out, which are described in the following section.

## 4.1. Simulation setup

In our simulations, the NC-NET nodes are distributed schematically in a cubic region, and the number of nodes is varying from 5 to 50. In each simulation experiment, the deployment of sensor nodes is with a sparse setting and is in clusters based on DMDG. The default parameters used in both the simulations and the analysis are summarized in Table 4.

The proposed DMDG protocols were evaluated by extensive computer simulations by Matlab 2014 and compared with LEACH [7], DEEC and PEACH. The performance compared includes network lifetime, costs associated to clustering and backbone formation, as well as the Data-Gathering and Aggregation Protocol for Networked Carrier Ad Hoc Networks: The Optimal... http://dx.doi.org/10.5772/66641 13


Table 4. Main simulation parameters.

• Step 2: Node A computes the first trisection point O′ between A and B based on geographic position got from the request packet, and then confirm point O′ as the ideal next-hop

• Step 3: Node A compares all the distance between one-hop neighbour (nodes C and D) and point O′; then choose the next-hop node with the following equation, Hopi ¼

• Step 4: Node C repeats the steps (Steps (1)–(3)), and confirm the next-hop relay E, where Khop = 2; then the node F obtains its next-hop relay B, where Khop = 1. So the route is confirmed and the information transmitted through the route A-C-F to B; the routing

To evaluate the performance of our scheme and compare it to both the association sponsor and central angle method, a set of simulations have been carried out, which are described in the

In our simulations, the NC-NET nodes are distributed schematically in a cubic region, and the number of nodes is varying from 5 to 50. In each simulation experiment, the deployment of sensor nodes is with a sparse setting and is in clusters based on DMDG. The default parame-

The proposed DMDG protocols were evaluated by extensive computer simulations by Matlab 2014 and compared with LEACH [7], DEEC and PEACH. The performance compared includes network lifetime, costs associated to clustering and backbone formation, as well as the

ters used in both the simulations and the analysis are summarized in Table 4.

hop neighbour of node A is C, which obtains the information packet.

, in which <sup>i</sup> is ID of one-hop neighbours. So in the case of Step 2, the next-

position.

12 Ad Hoc Networks

min<sup>i</sup> Pi−Popt 

following section.

4.1. Simulation setup

procedure is finished.

Figure 2. Geographic routing algorithm in NC-NET.

4. Performance evaluation and simulation

properties of generated clusters [19–24]. We assume that all the messages received from the cluster members can be aggregated into a single message.

Three kinds of scenarios are conducted in this simulation: (i) the average power consumption per node obtained by using LEACH, DEEC and DMDG; (ii) the lifetime and stable periods of the related LEACH, DEEC, PEACH and DMDG protocols; and (iii) the coverage rate versus proportion of dead nodes in different sensing range Rc = 5, 8, 10 km. We choose these scenarios because they are key factors in reflecting the efficiency of major technical points of DMDG protocol.

## 4.2. Performance result and analysis

As an addressing basic of the clustering process, it is critical to answer two questions before inspecting the proposed clustering protocol: (i) how does the number of nodes impact on the average power consumption, and (ii) how does the number of nodes impact on the average number of hops. The objective of the first scenario is to answer them through a simulation approach. Figure 3 presents the average power consumption per node obtained by using LEACH, DEEC and DMDG and average number of hops versus the number of nodes when the maximum transmission range Rmsg = 15, 000 m, and the number of the node increase from 5 to 50.

It is clear in Figure 3(a) that DMDG always achieves the lowest power consumption and a relative stable hop number, Nhop ≈ 2, which is gradually increasing in LEACH and DEEC. The average power consumption using DMDG can be as low as 0.637 when compared to LEACH and 0.824 when compared to DEEC. The stability in average number of hops results in stable of end-to-end delay; this can prove to be an advantage since it is convenient for estimating endto-end delay and guarantee a data synchronism.

Figure 3. Average power consumption and average number of hops versus the number of nodes, N.

Figure 4. Comparison of lifetime and stable periods between LEACH, PEACH, DEEC and DMDG.

Data-Gathering and Aggregation Protocol for Networked Carrier Ad Hoc Networks: The Optimal... http://dx.doi.org/10.5772/66641 15

Figure 5. Coverage rate versus number of the dead nodes (Rc = 5, 8, 10 km).

Then, in the second scenario, we aim to investigate the lifetime and stable periods of the related LEACH, DEEC, PEACH and DMDG protocols, because the result is closely related with the topological balance of the NC-NET. The simulation results are presented in Figure 4. In the case of networks using DMDG with the maximum transmission range Rmsg = 5000 m, the proportion of dead nodes (renew for mission) is 26% in lifetime and increases slowly in stable period (about 51% in the 4th day), which is much better than other protocols.

As the last scenario, we investigate the coverage rate performance with the increase of dead node, and the results are depicted in Figure 5. In this simulation, the coverage rate performance has been evaluated under three kinds of sensing range, that is, Rc = 5, 8, 10 km. The tendency curves of coverage rate have been compared with the increase of the number of dead node. Besides, DMDG protocol adopts the 3D-coverage model, which is introduced in Definition 1, k = 2. The result has shown that DMDG can achieve a full two-coverage if the number of the dead nodes ndead ≤ 11 when Rc = 5 km, ndead ≤ 20 when Rc = 8 km and ndead ≤ 29 when Rc = 10 km; this conclusion approves that the DMDG protocol can guarantee a high-coverage environment in a fleet network with low-density in-motion nodes.

## 5. Conclusions

Figure 3. Average power consumption and average number of hops versus the number of nodes, N.

14 Ad Hoc Networks

Figure 4. Comparison of lifetime and stable periods between LEACH, PEACH, DEEC and DMDG.

This chapter presented a new clustering and routing scheme, namely, the distributed multipleweight data-gathering and aggregation protocol (DMDG), in order to work under long-distance sparse NC-NET and efficiently guarantee a long stable coverage rate through a unit circle test. Simulation results show that DMDG protocol can significantly extend the network lifetime when compared with other approaches. The NC-NET with DMDG protocol is equally applicable to the cases of three-dimensional localization and in-flight alignment for carrierbased aircraft. The detail performance evaluation of these cases will be resolved in the following research.

## Acknowledgements

This chapter was supported in part by the National Natural Science Foundation of China under Grant no. 61473306 and partly by the Defence Advanced Research Project of China under Grant nos. 9140A09040614JB14001.

## Author details

Chao Gao\*, Guorong Zhao and Jianhua Lu

\*Address all correspondence to: gaochao.shd@163.com

Department of Control Engineering, Naval Aeronautical and Astronautical University, Yantai, P.R. China

## References


[6] Gao C., Zhao G. R., and Pan S. DMDG: a distributed data-aggregation and routing protocol for fleet wireless sensor networks. In: The 2th International Conference on Inteligent Networks and Intelligent Systems, Tianjin, China. 2009. pp. 150–155.

test. Simulation results show that DMDG protocol can significantly extend the network lifetime when compared with other approaches. The NC-NET with DMDG protocol is equally applicable to the cases of three-dimensional localization and in-flight alignment for carrierbased aircraft. The detail performance evaluation of these cases will be resolved in the follow-

This chapter was supported in part by the National Natural Science Foundation of China under Grant no. 61473306 and partly by the Defence Advanced Research Project of China

Department of Control Engineering, Naval Aeronautical and Astronautical University, Yantai,

[1] Chong C. Y. and Kumar S. P. Sensor networks: evolution, opportunities and challenges.

[2] Gao C., Zhao G. R., Lu J. H., and Pan S. A grid-based cooperative QoS routing protocol with fading memory optimization for navigation carrier ad hoc networks. Computer

[3] Gao C., Zhao G. R., Lu J. H., and Liu T. Dynamic cooperative MAC protocol for navigation carrier ad hoc networks: a diffserv-based approach. International Journal of Anten-

[4] Gao C., Zhao G. R., Lu J. H., and Pan S. Decentralised moving-horizon state estimation for a class of networked spatial-navigation systems with random parametric uncertainties and communication link failures. IET Control Theory and Applications. 2015;9

[5] Lu J. H., Gao C., Zhao G. R., and Liu T. Decentralized state estimation for networked navigation systems with communication delay and packet loss: the receding horizon case. In: 17th IFAC Symposium on System Identification, Beijing, China. 2015. pp. 1094–

ing research.

16 Ad Hoc Networks

Acknowledgements

Author details

P.R. China

References

under Grant nos. 9140A09040614JB14001.

Chao Gao\*, Guorong Zhao and Jianhua Lu

\*Address all correspondence to: gaochao.shd@163.com

Proceedings of the IEEE. 2003;91(8):1247–1256.

nas and Propagation. Forthcoming. 2016.

Networks. 2015;76:294–316.

(18):2666–2677.

1099.


**Provisional chapter**

## **A Composite Trust Model for Secure Routing in Mobile Ad-Hoc Networks Ad-Hoc Networks**

**A Composite Trust Model for Secure Routing in Mobile** 

Rutvij H. Jhaveri, Narendra M. Patel and Devesh C. Jinwala Devesh C. Jinwala Additional information is available at the end of the chapter

Rutvij H. Jhaveri, Narendra M. Patel and

Additional information is available at the end of the chapter

http://dx.doi.org/10.5772/66519

## **Abstract**

[22] Chuang S., Chen C., and Jiang C. Minimum-delay energy-efficient source to multilink

[23] Zhang H. and Hou J. C. Maintaining sensing coverage and connectivity in large sensor

[24] Ci S., Guizani M., and Sharif H. Adaptive clustering in wireless sensor networks by

[25] Le T., Hu W., and Corke P. ERTP: energy-efficient and reliable transport protocol for data streaming in wireless sensor networks. Computer Communications. 2009;21:3235–3251.

[26] Abbasi A. and Younis M. A survey on clustering algorithm for wireless sensor networks.

routing in wireless sensor networks. Signal Processing. 2007;87:2934–2948.

mining sensor energy data. Computer Communications. 2007;30:3235–3251.

networks. Ad Hoc & Wireless Networks. 2005;1(1):89–124.

Computer Communications. 2007;30:2826–2841.

18 Ad Hoc Networks

It is imperative to address the issue of secure routing in mobile ad-hoc networks (MANETs) where the nodes seek for cooperative and trusted behaviour from the peer nodes in the absence of well-established infrastructure and centralized authority. Due to the inherent absence of security considerations in the traditional ad-hoc routing protocols, providing security and reliability in the routing of data packets is a major challenge. This work addresses this issue by proposing a composite trust metric based on the concept of social trust and quality-of-service (QoS) trust. Extended from the ad-hoc on-demand distance vector (AODV) routing protocol, we propose an enhanced trustbased model integrated with an attack-pattern discovery mechanism, which attempts to mitigate the adversaries craving to carry out distinct types of packet-forwarding misbehaviours. We present the detailed mode of operations of three distinct adversary models against which the proposed scheme is evaluated. Simulation results under different network conditions depict that the combination of social and QoS trust components provides significant improvement in packet delivery ratio, routing overhead, and energy consumption compared to an existing trust-based scheme.

**Keywords:** packet-forwarding misbehaviour, secure routing, composite trust model, attack pattern discovery, mobile ad-hoc networks

## **1. Introduction**

A mobile ad-hoc network (MANET) is an autonomous system of wireless mobile nodes that dynamically form a network in order to exchange information in the absence of centralized authority and fixed infrastructure. Mobile nodes communicate with each other in a multihop way to carry out data transmission due to limited communication range and resource

© 2016 The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. © 2017 The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

constraints of the nodes. In the absence of router, each node operates as a host as well as a wireless router to forward packets for other nodes that may be outside its communication range [1]. The network functions well if all nodes operate in an altruistic manner. Due to the openness in network topology, distributed nature and lack of central authority, MANETs are particularly vulnerable to different types of routing attacks launched by internal nodes [2]. As a result, routing in such dynamic networks faces inherent challenges as compared to the traditional wireless networks. The traditional routing protocols proposed for ad-hoc networks are inefficient in dealing with different routing attacks.

The security schemes based on traditional cryptographic systems are typically used to resist external attacks. However, they prove to be inefficient in resisting the attacks launched by internal malevolent nodes. Such malicious nodes may seriously influence the security of the network by performing distinct types of packet-forwarding misbehaviours. In such a hostile environment, introducing the concept of 'trust' would provide prediction about the behaviour of neighbour nodes [2]. The notion of trust would prove to be useful for dynamic environments where the nodes need to depend on each other to achieve their goals [3]. Recently, trust management schemes have been considered as a viable security solution to improve the routing decisions in MANETs by detecting and isolating distrusted nodes [4].

In our previous work [5], we devised a trusted routing scheme with pattern discovery (TRS-PD) that integrates a trust model (based on QoS trust components) with an attack-pattern discovery mechanism in order to detect the malicious nodes earlier than a solitary trust model. TRS-PD estimates the distrust degree of neighbour nodes using direct trust computation. On the top of this, the attack-pattern discovery mechanism is introduced, which predicts suspicious activities of the neighbour nodes by promiscuously monitoring and recording specific fields of the control packets which are transmitted by the neighbour nodes. This gives an idea about the neighbour nodes, which might be following certain attack patterns. In addition, the scheme carries out indirect computation using recommendations by the trusted neighbours in order to enhance the trust establishment process. In this chapter, we propose enhanced TRS-PD (ETRS-PD), which uses a composite trust model that combines social trust component along with QoS trust components. ETRS-PD attempts to improve the packet delivery ratio against the adversary models discussed in Ref. [5] by enhancing the routing process. The performance of ETRS-PD is compared with TRS-PD against these adversary models under different network conditions.

The main technical contributions of this work are as follows: (1) An enhanced trust model is proposed for AODV protocol to evaluate neighbours' distrust value using composite trust metric. (2) Simulations carried out to compare the performance of ETRS-PD with TRS-PD prove that the performance of MANETs employing ETRS-PD is superior to that of MANETs employing TRS-PD against distinct types of adversaries.

The rest of the chapter is organized as follows. Section 2 discusses relevant related work. In Section 3, the proposed trust model is discussed. The enhanced trust-based on-demand routing scheme incorporated into AODV protocol is discussed in Section 4. Section 5 presents operations performed by various adversary models. The simulation results depicting the performance of ETRS-PD are presented in Section 6. Finally, Section 7 concludes the chapter.

## **2. Related work**

constraints of the nodes. In the absence of router, each node operates as a host as well as a wireless router to forward packets for other nodes that may be outside its communication range [1]. The network functions well if all nodes operate in an altruistic manner. Due to the openness in network topology, distributed nature and lack of central authority, MANETs are particularly vulnerable to different types of routing attacks launched by internal nodes [2]. As a result, routing in such dynamic networks faces inherent challenges as compared to the traditional wireless networks. The traditional routing protocols proposed for ad-hoc networks

The security schemes based on traditional cryptographic systems are typically used to resist external attacks. However, they prove to be inefficient in resisting the attacks launched by internal malevolent nodes. Such malicious nodes may seriously influence the security of the network by performing distinct types of packet-forwarding misbehaviours. In such a hostile environment, introducing the concept of 'trust' would provide prediction about the behaviour of neighbour nodes [2]. The notion of trust would prove to be useful for dynamic environments where the nodes need to depend on each other to achieve their goals [3]. Recently, trust management schemes have been considered as a viable security solution to improve the

In our previous work [5], we devised a trusted routing scheme with pattern discovery (TRS-PD) that integrates a trust model (based on QoS trust components) with an attack-pattern discovery mechanism in order to detect the malicious nodes earlier than a solitary trust model. TRS-PD estimates the distrust degree of neighbour nodes using direct trust computation. On the top of this, the attack-pattern discovery mechanism is introduced, which predicts suspicious activities of the neighbour nodes by promiscuously monitoring and recording specific fields of the control packets which are transmitted by the neighbour nodes. This gives an idea about the neighbour nodes, which might be following certain attack patterns. In addition, the scheme carries out indirect computation using recommendations by the trusted neighbours in order to enhance the trust establishment process. In this chapter, we propose enhanced TRS-PD (ETRS-PD), which uses a composite trust model that combines social trust component along with QoS trust components. ETRS-PD attempts to improve the packet delivery ratio against the adversary models discussed in Ref. [5] by enhancing the routing process. The performance of ETRS-PD is compared with TRS-PD against these adversary models under

The main technical contributions of this work are as follows: (1) An enhanced trust model is proposed for AODV protocol to evaluate neighbours' distrust value using composite trust metric. (2) Simulations carried out to compare the performance of ETRS-PD with TRS-PD prove that the performance of MANETs employing ETRS-PD is superior to that of MANETs

The rest of the chapter is organized as follows. Section 2 discusses relevant related work. In Section 3, the proposed trust model is discussed. The enhanced trust-based on-demand routing scheme incorporated into AODV protocol is discussed in Section 4. Section 5 presents operations performed by various adversary models. The simulation results depicting the performance of ETRS-PD are presented in Section 6. Finally, Section 7 concludes the chapter.

routing decisions in MANETs by detecting and isolating distrusted nodes [4].

are inefficient in dealing with different routing attacks.

20 Ad Hoc Networks

different network conditions.

employing TRS-PD against distinct types of adversaries.

A substantial amount of research work has been carried out in the last few years to address the security requirements of routing protocols by means of trust management.

A trust-based source routing (TSR) scheme devised by Xia *et al.* [6] attempts to discover a shortest secure route for data transmission in MANETs. Neighbour nodes are evaluated based on the historical trust values using correct packet-forwarding ratios. In addition, fuzzy logic is used to estimate a node's current trust based on its capability and historical trust value. This estimated value is used to predict the misbehaving nodes in the neighbourhood. A trusted route is selected for data transmission by avoiding such untrustworthy nodes. Experimental results show the effectiveness of TSR against blackhole, grayhole and modification attacks. However, the scheme incurs high computational overhead in calculation of route trust after arrival of every data packet at the destination. Furthermore, the scheme requires buffering of the packets in a circular queue, which incurs significant overhead in searching the match for the packets in the buffer. Gharehkoolchian *et al.* [7] proposed a novel trust model, which uses different *trust levels (TL)* for nodes and imposes the limitations based on the trust level in order to mitigate the malicious nodes. When a node enters the network, it is assigned *TL* = *1*. It gains higher reputation if it acts normally by forwarding packets and thereby, it is assigned *TL = 2*. In case of malicious behaviour, it is assigned *TL = 0*. If the malicious behaviour of the node is observed for three times, it is assigned *TL = −1* and the node is permanently blocked. During the route discovery process, when a node receives a route reply from its neighbour node, it verifies its *TL* value. If the node is a non-malicious node *(TL = 2)*, the route reply is forwarded. Otherwise, a test route request is sent to the suspicious node *(TL = 1)* after the received route reply. If an abnormal reply is received from the suspicious node in response to the test route request, the route reply is discarded after assigning *TL = 0* to the node and the node is blocked for a specific time. Thus, the scheme attempts to isolate the malicious nodes during route discovery process. However, it does not have any reactive mechanism to cope up with sudden drops in packets during data transmission phase; instead, it just detects the adversary but attempts to isolate it during the next route discovery process. Airehrour *et al.* [8] proposed *GradeTrust* protocol to isolate blackhole adversaries by selecting a secure path, in addition to elimination of excessive routing computations and minimization of communication overhead. It classifies the nodes into three sets in order of the trust levels: *Trusted Friends*, *Friends* and *Possible Friends*. Trust level is assigned by monitoring neighbours' request packet-forwarding ratio. A source node selects the next hop from its Trusted Friends, and the process continues until the packet reaches the destination. In the case of unavailability of a Trusted Friend, a Friend is selected. A compromised node is dissociated swiftly from other trusted nodes, and it is pushed down to the lower trust level. However, the scheme does not consider the forwarding ratio of data packets in calculation of the trust level, which makes it susceptible to packet dropping adversaries during data transmission phase. In addition, simulation results showing comparison of the proposed protocol with the traditional protocols are not promising. Patel *et al.* [9] proposed a trust model for AODV-based MANETs, which attempts to increase network lifetime by uniform consumption of energy. A trust value is computed based on dropping ratios and delays of control and data packets as well as residual node energy. The scheme attempts to discover a trusted route during the route reply propagation towards the source node on the reverse path. All the intermediate nodes receiving the route reply packet update the path trust value in the packet using the available trust values of neighbours. If a node receives multiple route reply packets, it compares the trust of the newly received path with that of the current path and stores the path with the maximum trust value. However, the scheme does not have any reactive mechanism to fight against packet-dropping adversaries during data transmission phase. After identifying an adversary, it waits for the next route discovery process to isolate it. Chiejina *et al.* [10] proposed a solution to evaluate the trust of a node in the network, which ensures that nodes expending their energy in forwarding data and control packets for other nodes are allowed to carry on their activities while the malevolent nodes are isolated from the network. Trust values are computed by direct observations, which are aggregated at different time intervals to provide a historical reputation of the node. The total reputation value of a node is mapped with a grading criterion to decide the status of a node. Nodes with lower reputation value than the set threshold value are blacklisted and denied the network resources. Routes containing blacklisted nodes are discarded, and alternative routes are discovered. The solution attempts to mitigate selfish and deceitful nodes from the network with scarce resources. However, whenever the source sends a packet towards the destination, the solution generates additional overhead as *path administrator* has to check that the packet has not been sent via a path containing a blacklisted node. Mylsamy *et al.* [11] proposed a *preference-based protocol for trust and head selection (2PTH),* which takes four parameters to calculate a trust value: packet delivery ratio, packet misrouting ratio, packet alteration ratio, and packet injection ratio. Depending on the affected security parameters, weighing coefficients' values are determined. Trust values are classified into three different categories: high, medium and low. If trust value of a node goes below its relative threshold, it is not allowed to participate as a cluster member. A cluster-based routing mechanism is used which discovers a stable cluster head based on external factors such mobility, connectivity and distance as well as internal factors such as residual battery power, processing power and memory. When a cluster head of the cluster of the destination node receives a route request packet during the route discovery process, it verifies the trustworthiness of the node in order to establish a secure route. Simulation results show promising performance of the protocol as compared to some existing protocols. However, the protocol does not have any reactive mechanism for identifying packet-dropping adversaries during data transmission phase. Moreover, *weight assignment* and *cluster head election* consume a significant amount of computational resources. Indirani *et al.* [12] presented a *swarm-based distributed intrusion detection system (SDIDS)* with the objective to remove the complexity in the design of an IDS caused by the inherent MANET characteristics. Active nodes in a route are selected by *ant colony optimization (ACO)* technique based on a node's packet-forwarding activities, residual bandwidth, residual energy and connectivity. A *forward ant* reaches to every node in order to compute and update the pheromone value using the aforementioned parameters. When it reaches the destination, the information collected by the forward ant about all the hops is transferred to the *backward ant*. The backward ant then traverses on the reverse path and reaches to the source in order to deliver the status of all nodes. A routing decision is then made by selecting the optimal route to the destination. However, the scheme incurs high computational overhead in calculation of route trust. In addition, establishment of a trusted route should not be the sole responsibility of the source node. Xia *et al.* [13] proposed a *lightweight trust-enhanced routing protocol (TeAOMDV),* which attempts to provide an optimal twoway trusted route without containing the malicious entities. Its trust framework uses passive and local monitoring information to evaluate the trust values of neighbours. It considers activity, stability and historical trust record of a node in evaluation of a node trust. Moreover, the trust value is modified by collecting the recommendations from the trusted neighbours. It uses *hop count*, *forward path trust* and *reverse path trust* as the metrics to compose a threedimensional evaluation vector for taking routing decisions. The authors extend their work by proposing an improved *SCGM(1,1)-Markov chain prediction method* based on the *system cloud grey model* and *Markov stochastic chain theory* to forecast trust level of a node for future routing decisions. However, it holds similar drawbacks as the scheme proposed in [12] due to the consideration of route trust. Azer *et al.* [14] proposed a new reputation system for ad hoc networks, called *misbehaviour detection and control (MDAC),* which encourages the nodes to act in a trustworthy manner. It obtains first hand and second hand information about neighbouring nodes. Trust is evaluated based on number of incoming packets and total consumed time to deliver packets. The *MDAC modeller module* combines all collected information about a node into a meaningful reputation value. Based on the reputation values, nodes in the network are guided to take necessary actions such as trust/don't trust, cooperate/don't cooperate and forward/don't forward. A node is considered eligible for service only after verifying its reputation value. The mechanism shows better performance compared to an existing scheme in terms of throughput and delay. However, the mechanism does not consider control packets in the calculation of the reputation value which delays the detection of sequence number attacks. In addition, it adds significant computational overhead in making reputation decisions about neighbouring nodes. Rajkumar *et al.* [15] proposed a trust-based light-weight authentication routing protocol which adopts multipath route discovery technique to mitigate adversaries. A route is rated based on packet success rate after route reply is forwarded to the source node. An optimal path for data transmission is chosen based on its rating, and the next optimal path is stored as an alternative arrangement. The protocol calculates a trust value using *EigenTrust* algorithm, which is based on direct and indirect observations of neighbour nodes. A resolver is engaged for computing a global trust value of the node, which also executes trust noise cancellation mechanism. If the trust value of a node goes below the threshold value, it is authenticated using the *Shamir's secret sharing* technique. If a node is found to be malicious, all routes going through the node are discarded and the alternate optimal path is selected. However, cryptographic approaches add considerable amount of communication and memory overhead along with key distribution issues. In addition, the scheme involves high computational overhead in the estimation of packet success rate and calculation of the global trust value.

## **3. Trust model**

energy. The scheme attempts to discover a trusted route during the route reply propagation towards the source node on the reverse path. All the intermediate nodes receiving the route reply packet update the path trust value in the packet using the available trust values of neighbours. If a node receives multiple route reply packets, it compares the trust of the newly received path with that of the current path and stores the path with the maximum trust value. However, the scheme does not have any reactive mechanism to fight against packet-dropping adversaries during data transmission phase. After identifying an adversary, it waits for the next route discovery process to isolate it. Chiejina *et al.* [10] proposed a solution to evaluate the trust of a node in the network, which ensures that nodes expending their energy in forwarding data and control packets for other nodes are allowed to carry on their activities while the malevolent nodes are isolated from the network. Trust values are computed by direct observations, which are aggregated at different time intervals to provide a historical reputation of the node. The total reputation value of a node is mapped with a grading criterion to decide the status of a node. Nodes with lower reputation value than the set threshold value are blacklisted and denied the network resources. Routes containing blacklisted nodes are discarded, and alternative routes are discovered. The solution attempts to mitigate selfish and deceitful nodes from the network with scarce resources. However, whenever the source sends a packet towards the destination, the solution generates additional overhead as *path administrator* has to check that the packet has not been sent via a path containing a blacklisted node. Mylsamy *et al.* [11] proposed a *preference-based protocol for trust and head selection (2PTH),* which takes four parameters to calculate a trust value: packet delivery ratio, packet misrouting ratio, packet alteration ratio, and packet injection ratio. Depending on the affected security parameters, weighing coefficients' values are determined. Trust values are classified into three different categories: high, medium and low. If trust value of a node goes below its relative threshold, it is not allowed to participate as a cluster member. A cluster-based routing mechanism is used which discovers a stable cluster head based on external factors such mobility, connectivity and distance as well as internal factors such as residual battery power, processing power and memory. When a cluster head of the cluster of the destination node receives a route request packet during the route discovery process, it verifies the trustworthiness of the node in order to establish a secure route. Simulation results show promising performance of the protocol as compared to some existing protocols. However, the protocol does not have any reactive mechanism for identifying packet-dropping adversaries during data transmission phase. Moreover, *weight assignment* and *cluster head election* consume a significant amount of computational resources. Indirani *et al.* [12] presented a *swarm-based distributed intrusion detection system (SDIDS)* with the objective to remove the complexity in the design of an IDS caused by the inherent MANET characteristics. Active nodes in a route are selected by *ant colony optimization (ACO)* technique based on a node's packet-forwarding activities, residual bandwidth, residual energy and connectivity. A *forward ant* reaches to every node in order to compute and update the pheromone value using the aforementioned parameters. When it reaches the destination, the information collected by the forward ant about all the hops is transferred to the *backward ant*. The backward ant then traverses on the reverse path and reaches to the source in order to deliver the status of all nodes. A routing decision is then made by selecting the optimal route to the destination. However, the scheme incurs high computational overhead in calculation of route trust. In addition, establishment of a trusted

22 Ad Hoc Networks

As a part of the literature survey, we discover that a composite trust metric based on social and QoS trust components may successfully perform tasks to meet both performance and trust requirements [16, 17]. We have noticed some work in the literature moving in this direction. Cho *et al.* [17] considered honesty and intimacy, while Kohlas *et al.* [18] considered honesty, competency, reliability and maliciousness as social trust components to define trust relationships. In addition, we observe that *energy consumption* is an important QoS trust component for improving the network performance [17, 19]. Taking these notes into consideration, we devise an enhanced trust-based scheme, *ETRS-PD*.

ETRS-PD considers *ditch ratio* as a social trust component in estimation of distrust degree of the neighbours. This social trust component is utilized to know the magnitude of misbehaviour carried out by a node while residing in monitoring node's neighbourhood. In addition, *energy consumption* is considered as an additional QoS component along with *packet drop ratio*. Thus, a composite trust metric is constructed by including social trust along with QoS trust. Furthermore, the routing process of TRS-PD is modified to enhance the routing decisions. As aforementioned, ETRS-PD attempts to improve the packet delivery ratio against the adversary models discussed in our previous work [5].

In our trust model, we compute historical trust on a constant basis after a specific time interval called trust update interval. Overall, our trust model performs trust derivation and trust computation along with discovery of attack patterns. We modify the trust model of TRS-PD to perform trust derivation and trust computation in a different way.

## **3.1. Basic assumptions**

Our trust-based scheme makes the following assumptions: (i) all the mobile nodes have identical physical characteristics; (ii) the wireless links in the network are bidirectional; (iii) all the nodes operate in promiscuous mode in order to observe the neighbour nodes and (iv) the source and the destination are benevolent nodes. The above assumptions are fulfilled by wireless MAC layer protocols.

## **3.2. Trust derivation**

Our proposed trust model uses direct observations to derive distrust values of neighbour nodes by observing packet dropping ratios, energy consumption and ditch ratio of neighbour nodes. In addition to this, each node employs an attack pattern discovery mechanism, which detects malicious patterns generated by neighbour nodes in the transmitted control packets. We also consider recommendations of trusted neighbours for improving the routing decisions.

## **3.3. Trust computation**

In a routing process, neighbour node's distrust is evaluated by the sender by observing activities carried out by that neighbour. To be specific, a node *ni* will increase the distrust score of its neighbour *nj* if the *nj* does not forward the packet sent by *ni* [5].

**Definition 1.** *Control dropping ratio (CDR)*: It is the ratio of the number of control packets dropped to the number of control packets which are supposed to be forwarded. At time *t*, *CDR(t)* is computed as follows:

$$\text{CDR}(t) = \frac{\text{NC}\_{\text{s}}(t)}{\text{NC}\_{\text{s}}(t)} \tag{1}$$

where *NCd(t)* signifies the cumulative count of dropped control packets, and *NCa(t)* represents the total number of sent control packets from time *0* to *t*.

Cho *et al.* [17] considered honesty and intimacy, while Kohlas *et al.* [18] considered honesty, competency, reliability and maliciousness as social trust components to define trust relationships. In addition, we observe that *energy consumption* is an important QoS trust component for improving the network performance [17, 19]. Taking these notes into consideration, we

ETRS-PD considers *ditch ratio* as a social trust component in estimation of distrust degree of the neighbours. This social trust component is utilized to know the magnitude of misbehaviour carried out by a node while residing in monitoring node's neighbourhood. In addition, *energy consumption* is considered as an additional QoS component along with *packet drop ratio*. Thus, a composite trust metric is constructed by including social trust along with QoS trust. Furthermore, the routing process of TRS-PD is modified to enhance the routing decisions. As aforementioned, ETRS-PD attempts to improve the packet delivery ratio against the adver-

In our trust model, we compute historical trust on a constant basis after a specific time interval called trust update interval. Overall, our trust model performs trust derivation and trust computation along with discovery of attack patterns. We modify the trust model of TRS-PD

Our trust-based scheme makes the following assumptions: (i) all the mobile nodes have identical physical characteristics; (ii) the wireless links in the network are bidirectional; (iii) all the nodes operate in promiscuous mode in order to observe the neighbour nodes and (iv) the source and the destination are benevolent nodes. The above assumptions are fulfilled by

Our proposed trust model uses direct observations to derive distrust values of neighbour nodes by observing packet dropping ratios, energy consumption and ditch ratio of neighbour nodes. In addition to this, each node employs an attack pattern discovery mechanism, which detects malicious patterns generated by neighbour nodes in the transmitted control packets. We also consider recommendations of trusted neighbours for improving the routing

In a routing process, neighbour node's distrust is evaluated by the sender by observing activi-

**Definition 1.** *Control dropping ratio (CDR)*: It is the ratio of the number of control packets dropped to the number of control packets which are supposed to be forwarded. At time *t*,

does not forward the packet sent by *ni*

will increase the distrust score of

[5].

ties carried out by that neighbour. To be specific, a node *ni*

devise an enhanced trust-based scheme, *ETRS-PD*.

sary models discussed in our previous work [5].

**3.1. Basic assumptions**

24 Ad Hoc Networks

**3.2. Trust derivation**

**3.3. Trust computation**

if the *nj*

*CDR(t)* is computed as follows:

its neighbour *nj*

decisions.

wireless MAC layer protocols.

to perform trust derivation and trust computation in a different way.

**Definition 2.** *Data dropping ratio (CDR)*: It is the ratio of the number of data packets dropped to the number of data packets, which are supposed to be forwarded. At time *t*, *CDR(t)* is computed as follows:

$$DDR(t) = \frac{ND\_i(t)}{ND\_i(t)}\tag{2}$$

where *NDd(t)* signifies the cumulative count of dropped control packets, and *NDa(t)* represents the total number of sent control packets from time *0* to *t*.

**Definition 3.** *Energy consumption (EC)*: It is the ratio of the energy consumed by a node to the initial energy of that node. When a node possesses limited residual energy, it may not hold the capabilities to forward the packets of other nodes. At time *t*, *EC(t)* is computed as follows:

$$EC(t) = \frac{EI \cdot ER(t)}{EI} \tag{3}$$

where *EI* signifies the initial energy, and *ER(t)* signifies the residual energy of the node at time *t*.

**Definition 4.** *Ditch ratio (DTR)*: It is the ratio of the number of times a neighbour node is found to be distrusted while receiving its *HELLO* packets to the total number of *HELLO* packets received from that node. At time *t*, *DTR(t)* is computed as follows:

$$DTR(t) = \frac{NH\_4(t)}{NH\_4(t)}\tag{4}$$

where *NHd(t)* signifies the number of times a distrusted neighbor node has ditched the monitoring node while sending *HELLO* packets, and *NHa(t)* signifies the total number of *HELLO* packets received from that neighbour node.

The obtained distrust value of a node *nj* by a monitoring node *ni* is the measure of packet dropping activities, energy drain rate and magnitude of misbehaviour. The distrust value of node *nj* evaluated by node *ni*, denoted as *DTVij*, is calculated by the following formula:

$$\text{DTV}\_{\text{\tiny\text{\tiny\text{\tiny}}}}(\text{t}) = w1 \times \text{CDR}\_{\text{\tiny\text{\tiny\text{\tiny}}}}(\text{t}) + w2 \times \text{DDR}\_{\text{\tiny\text{\tiny}}}(\text{t}) + w3 \times \text{EC}\_{\text{\tiny\text{\tiny}}}(\text{t}) + w4 \times \text{DTR}\_{\text{\tiny\text{\tiny}}}(\text{t}) \tag{5}$$

where *w1, w2, w3* and *w4* (*w1*, *w2, w3, w4* ≥ 0 and *w1 + w2* + *w3* + *w4 = 1*) are the weights assigned to *CDR*, *DDR, EC* and *DTR*, respectively.

In our trust model, distrust values are restricted in the range from 0 to 1 (i.e., *0 ≤ DTVij ≤ 1*). The distrust value 0 indicates complete trust, whereas the distrust value 1 signifies complete distrust. We set the initial value of distrust to 0 as we assume all the nodes to be benevolent initially. Meanwhile, the distrust value constantly varies with the time as per the behaviour of neighbour nodes. We use a distrust threshold *η* to differentiate the malicious nodes from benign nodes.

As discussed in Ref. [5], we incorporate an attack pattern discovery mechanism on the top of the trust model, which employs the model of *method of common differences (MCD).* Thus, the pattern discovery mechanism attempts to identify the adversaries following attack patterns prior to conducting misbehaviours; on the other hand, the trust model detects other packetdropping adversaries during the trust update procedure.

## **4. Enhanced trust-based on-demand routing**

While any reactive routing protocol can be extended to incorporate ETRS-PD, we extend adhoc on-demand distance vector (AODV) protocol for this purpose. In addition to the modifications described in [5], we further modify the functionality of AODV in order to improve the routing decisions. The neighbour table is modified by appending the following fields: (i) *Energy consumption,* (ii) *Ditch count:* The number of times a neighbour node is found to be distrusted while receiving its *HELLO* packets, (iii) *HELLO count:* The total number of *HELLO* packets received from a neighbour node and (iv) *Ditch ratio*. The *distrust value* is calculated as per the formula (5). We modify the *HELLO* packets to include an additional field: (i) *Energy consumed*: Energy consumed by the node, which is provided as information to the neighbour nodes (calculated as per the formula (3)).

## **4.1. Routing strategy**

We further modify the routing strategy described in Ref. [5]. The modified routing strategy (by modifying *Step 4*, *Step 8* and *Step 9*) is described herewith:

*Step 1*: Before starting data transmission, the source node *ns* looks up in its local routing table for the destination node *nd* .

*Step 2*: If entry exists, it starts sending data through the trusted next hop to *nd* . Go to *Step 8*.

*Step 3*: If no such route exists, *ns* initiates a route discovery process by flooding route request (RREQ) packets to discover a route to *nd* .

*Step 4*: When an intermediate node *nk* receives a route reply (RREP) from its neighbour node *nj* , it accepts the reply only if *nj* is not a distrusted node (*nk* finds absence of attack patterns for *nj* with distrust value less than or equal to *η*) and *recommended as a trusted node*.

*Step 5*: If multiple route replies are received after the route discovery process, a route entry for the route with the highest destination sequence number and trusted next hop is created for *nd* and inserted into the routing table of *ns*.

*Step 6*: If no such route is discovered, go to *Step 3.*

*Step 7*: Node *ns* starts data transmission to *nd*.

*Step 8*: If an intermediate node *nk* finds a next hop *nm* distrusted (*by direct observation or by recommendation*) in its routing table for a destination *np* during the trust update procedure, the entry is discarded. A local route discovery process is initiated by *nk* to discover an alternate route to *np* .

*Step 9*: Even though an intermediate node *nk* finds a distrusted neighbor *nm* attempting to regain its trust by recuperating the distrust value less than or equal to *η*, *it is still considered as a distrusted node* (i.e. it is *not* reconsidered as a trusted node).

## **4.2. Routing procedures**

pattern discovery mechanism attempts to identify the adversaries following attack patterns prior to conducting misbehaviours; on the other hand, the trust model detects other packet-

While any reactive routing protocol can be extended to incorporate ETRS-PD, we extend adhoc on-demand distance vector (AODV) protocol for this purpose. In addition to the modifications described in [5], we further modify the functionality of AODV in order to improve the routing decisions. The neighbour table is modified by appending the following fields: (i) *Energy consumption,* (ii) *Ditch count:* The number of times a neighbour node is found to be distrusted while receiving its *HELLO* packets, (iii) *HELLO count:* The total number of *HELLO* packets received from a neighbour node and (iv) *Ditch ratio*. The *distrust value* is calculated as per the formula (5). We modify the *HELLO* packets to include an additional field: (i) *Energy consumed*: Energy consumed by the node, which is provided as information to the neighbour

We further modify the routing strategy described in Ref. [5]. The modified routing strategy

*Step 1*: Before starting data transmission, the source node *ns* looks up in its local routing table

initiates a route discovery process by flooding route request

finds a next hop *nm* distrusted (*by direct observation or by rec-*

receives a route reply (RREP) from its neighbour node

finds absence of attack patterns for

during the trust update procedure, the

to discover an alternate

. Go to *Step 8*.

*Step 2*: If entry exists, it starts sending data through the trusted next hop to *nd*

.

with distrust value less than or equal to *η*) and *recommended as a trusted node*.

is not a distrusted node (*nk*

*Step 5*: If multiple route replies are received after the route discovery process, a route entry for the route with the highest destination sequence number and trusted next hop is created for *nd*

dropping adversaries during the trust update procedure.

**4. Enhanced trust-based on-demand routing**

(by modifying *Step 4*, *Step 8* and *Step 9*) is described herewith:

.

nodes (calculated as per the formula (3)).

**4.1. Routing strategy**

26 Ad Hoc Networks

for the destination node *nd*

*nj*

*nj*

*Step 7*: Node *ns*

route to *np*

.

*Step 3*: If no such route exists, *ns*

, it accepts the reply only if *nj*

*Step 8*: If an intermediate node *nk*

(RREQ) packets to discover a route to *nd*

and inserted into the routing table of *ns*.

*Step 6*: If no such route is discovered, go to *Step 3.*

*ommendation*) in its routing table for a destination *np*

starts data transmission to *nd*.

entry is discarded. A local route discovery process is initiated by *nk*

*Step 4*: When an intermediate node *nk*

The procedures for sending RREQ, receiving RREQ and sending RREP remain unmodified as presented in **Figures 1**–**3**, respectively (as described in Ref. [5]).


**Figure 1.** *SendRREQ* procedure [5].


**Figure 2.** *RecvRREQ* procedure [5].


**Figure 3.** *SendRREP* procedure [5].

The modifications carried out in the receiving RREP procedure are highlighted in **Figure 4**.


**Figure 4.** *RecvRREP* procedure.

The procedure for route maintenance remains unmodified as presented in **Figure 5** (as described in Ref. [5]).


**Figure 5.** *Route maintenance* procedure [5].

## **4.3. Trust update and trust recommendation procedures**

The modifications carried out in the trust update and trust recommendation procedures are highlighted in **Figures 6** and **7**, respectively.


**Figure 6.** *Update trust* procedure.

The modifications carried out in the receiving RREP procedure are highlighted in **Figure 4**.

The procedure for route maintenance remains unmodified as presented in **Figure 5** (as

The modifications carried out in the trust update and trust recommendation procedures are

described in Ref. [5]).

**Figure 4.** *RecvRREP* procedure.

28 Ad Hoc Networks

**4.3. Trust update and trust recommendation procedures**

highlighted in **Figures 6** and **7**, respectively.

**Figure 5.** *Route maintenance* procedure [5].


**Figure 7.** *Recommend trust* procedure.

## **5. Adversary models**

It is obvious that the success of any security mechanism largely depends on the operations performed by the adversaries. In our work, we evaluate the performance of ETRS-PD against three adversary models described in Ref. [5].

## **5.1. Intelligent adversary model**

The operations performed by *intelligent adversary* (denoted as *Attack1*) are presented in **Figure 8** [5, 20]. The adversary follows a pattern in inserting the value of hop count (Hop\_Count = 2) while sending RREP packet.

## **5.2. Slow poison adversary model**

The operations performed by *slow poison adversary* (denoted as *Attack2*) are presented in **Figure 9** [5]. The adversary follows a pattern in inserting the value of destination sequence number (RREQ\_Dest\_Seqno + 1) while sending RREP packet.

**Figure 9.** Operations performed by a node launching *Attack2* [5].

## **5.3. Capricious adversary model**

**5. Adversary models**

30 Ad Hoc Networks

three adversary models described in Ref. [5].

**5.1. Intelligent adversary model**

while sending RREP packet.

**5.2. Slow poison adversary model**

number (RREQ\_Dest\_Seqno + 1) while sending RREP packet.

**Figure 8.** Operations performed by a node launching *Attack1* [5, 20].

It is obvious that the success of any security mechanism largely depends on the operations performed by the adversaries. In our work, we evaluate the performance of ETRS-PD against

The operations performed by *intelligent adversary* (denoted as *Attack1*) are presented in **Figure 8** [5, 20]. The adversary follows a pattern in inserting the value of hop count (Hop\_Count = 2)

The operations performed by *slow poison adversary* (denoted as *Attack2*) are presented in **Figure 9** [5]. The adversary follows a pattern in inserting the value of destination sequence The operations performed by *capricious adversary* (denoted as *Attack3*) are presented in **Figure 10** [5]. This adversary does not generate any attack pattern while sending RREP packet.


**Figure 10.** Operations performed by a node launching *Attack3* [5].

## **6. Simulation results and analysis**

NS-2 (ver. 2.34) simulator is used to evaluate the performance efficiency of ETRS-PD against the three adversary models, namely *Attack1*, *Attack2* and *Attack3*. To prove our claim that ETRS-PD provides enhanced routing process than our previous proposal, TRS-PD [5], the performance of ETRS-PD is compared with TRS-PD against all three adversary models. We employ IEEE 802.11 MAC to carry out simulations in an area of 1000 × 1000 m. The benign nodes were randomly distributed over the network, which employs either AODV, ETRS-PD or TRS-PD protocol. Randomly positioned malicious nodes selectively perform packet forwarding misbehaviours by employing either of the three adversary models, namely *Attack1, Attack2* and *Attack3*. It is considered that the wireless network interface consumes 1.65, 1.4, 1.15 and 0.045 W for the *Transmit*, *Receive* and *Idle* modes and the *Sleep* state, respectively [21]. We take 800 μs as the transition time from the Sleep state to Awake state and during this transition period, a mobile node will consume 2.3 W power. All the experimental data are obtained after performing 10 different simulations and taking their average values. The major simulation parameters are shown in **Table 1**.


**Table 1.** Simulation parameters.

In order to evaluate the performance of ETRS-PD, the following performance metrics are used: *packet delivery ratio (PDR)*, *normalized routing overhead (NRO)* and *average energy consumption (AEC)*. The following network parameters are varied: (1) *maximum speeds of nodes* and (2) *percentage of adversaries*.

The performance of AODV and TRS-PD in terms of PDR and NRO is already evaluated in Ref. [5], while their performance in terms of AEC is evaluated in Ref. [21].

## **6.1. Test 1: varying node mobility**

**6. Simulation results and analysis**

32 Ad Hoc Networks

**Parameter Value** Coverage area 1000 × 1000 m MAC layer protocol IEEE 802.11 Communication range of each node 250 m Channel bandwidth 2 Mbps Traffic type CBR-UDP Packet size 512 bytes

Mobility model Random way point

Routing protocols AODV, *Attack1*, *Attack2*, *Attack3*, TRS-PD, ETRS-PD

Simulation duration 240 s Number of nodes 50 Maximum mobility (varying) 4–20 m/s Pause time 5 s Number of connections 15 Percentage of malicious nodes (varying) 0–40%

Initial energy 1000 J Transmit power 1.65 W Receive power 1.4 W Idle power 1.15 W Sleep power 0.045 W Transition power 2.3 W Transition time 800 μs

**Table 1.** Simulation parameters.

NS-2 (ver. 2.34) simulator is used to evaluate the performance efficiency of ETRS-PD against the three adversary models, namely *Attack1*, *Attack2* and *Attack3*. To prove our claim that ETRS-PD provides enhanced routing process than our previous proposal, TRS-PD [5], the performance of ETRS-PD is compared with TRS-PD against all three adversary models. We employ IEEE 802.11 MAC to carry out simulations in an area of 1000 × 1000 m. The benign nodes were randomly distributed over the network, which employs either AODV, ETRS-PD or TRS-PD protocol. Randomly positioned malicious nodes selectively perform packet forwarding misbehaviours by employing either of the three adversary models, namely *Attack1, Attack2* and *Attack3*. It is considered that the wireless network interface consumes 1.65, 1.4, 1.15 and 0.045 W for the *Transmit*, *Receive* and *Idle* modes and the *Sleep* state, respectively [21]. We take 800 μs as the transition time from the Sleep state to Awake state and during this transition period, a mobile node will consume 2.3 W power. All the experimental data are obtained after performing 10 different simulations and taking their average values. The major simulation parameters are shown in **Table 1**.

In this test, the performance of the protocols is evaluated against *Attack1*, *Attack2* and *Attack3* by varying mobility of nodes from 4 to 20 m/s and keeping other parameters fixed. The percentage of malicious nodes is kept fixed to 20% for all three types of adversaries.

As shown in **Figure 11**, the PDR of AODV under *Attack1* declines from nearly 46 to 39% as the mobility increases from 4 to 20 m/s. The increase in packet loss at higher mobility is due to the increased number of link breakages at higher node speeds. Meanwhile, PDR of AODV under *Attack2* and *Attack3* declines from nearly 68 to 60% and 74 to 64%, respectively, as shown in **Figures 12** and **13**, respectively. When TRS-PD is employed, the PDR declines from nearly 73 to 57%, 79 to 69% and 80 to 69% under *Attack1*, *Attack2* and *Attack3,* respectively. This considerable rise in PDR is due to the integration of the attack-pattern discovery mechanism with the trust model. Meanwhile, when ETRS-PD is employed, it provides improvement in PDR over TRS-PD by an average of 6.21 under *Attack1*, 2.82 under *Attack2* and 4.03 under *Attack3*. The reasons behind improved results are as follows: (i) Construction of a composite trust metric using social trust and QoS trust. (ii) Enhanced routing decisions due to the modifications carried out in receive RREP, trust update and trust recommendation procedures.

**Figure 11.** PDR under *Attack1*.

**Figure 12.** PDR under *Attack2*.

**Figure 13.** PDR under *Attack3*.

As shown in **Figures 14**–**16**, as the node speed increases, the NRO of AODV increases from nearly 5.7 to 10.4, 1.8 to 4.1 and 2.8 to 5.1 under *Attack1*, *Attack2* and *Attack3,* respectively. Meanwhile, the TRS-PD provides improved performance over AODV by providing NRO from nearly 4.5 to 8.5 and 2.8 to 5.1 under *Attack1* and *Attack3,* respectively. On the other hand, due to the *Fibonacci dropping behaviour* of *Attack2* during the data transmission phase, the number of route hand-off mechanisms increases for TRS-PD as time goes on. As a result, resultant NRO is higher than that of AODV, which varies between nearly 3.2 and 5.5. Meanwhile, ETRS-PD provides improvement in NRO over TRS-PD by an average of 1.43 under *Attack1*, 0.30 under *Attack2* and 0.36 under *Attack3*. The reason behind this is, ETRS-PD leads to less number of route hand-off mechanisms than TRS-PD due to the inclusion of two more components in the overall trust composition as well as enhanced routing process.

**Figure 14.** NRO under *Attack1*.

As shown in **Figures 14**–**16**, as the node speed increases, the NRO of AODV increases from nearly 5.7 to 10.4, 1.8 to 4.1 and 2.8 to 5.1 under *Attack1*, *Attack2* and *Attack3,* respectively.

**Figure 12.** PDR under *Attack2*.

34 Ad Hoc Networks

**Figure 13.** PDR under *Attack3*.

In order to ensure the improvement in energy consumption, we compare the performance of ETRS-PD with TRS-PD. As depicted by the graph in **Figure 17**, the AEC under *Attack1* varies between 313.56 and 314.13 J when employing TRS-PD. Meanwhile, ETRS-PD improves the AEC of TRS-PD by an average of 1.6 J. As depicted by the graph in **Figure 18**, the AEC under *Attack2* varies in the range of 312.82–314.4 J when employing TRS-PD. Meanwhile, ETRS-PD improves the AEC of TRS-PD by an average of 0.57 J. As depicted by the graph in **Figure 19**, the AEC under *Attack3* varies between 312.79 and 313.41 J when employing TRS-PD. Meanwhile, ETRS-PD improves the AEC of TRS-PD by an average of 0.57 J.

**Figure 15.** NRO under *Attack2*.

**Figure 16.** NRO under *Attack3*.

A Composite Trust Model for Secure Routing in Mobile Ad-Hoc Networks http://dx.doi.org/10.5772/66519 37

**Figure 17.** AEC under *Attack1*.

**Figure 18.** AEC under *Attack2*.

**Figure 16.** NRO under *Attack3*.

**Figure 15.** NRO under *Attack2*.

36 Ad Hoc Networks

**Figure 19.** AEC under *Attack3*.

## **6.2. Test 2: varying percentage of malicious nodes**

In this test, the performance of the protocols is evaluated against *Attack1*, *Attack2* and *Attack3* by varying percentage of malicious nodes from 0 to 40% and keeping other parameters fixed. The mobility parameter is kept fixed to 20 m/s for all three types of adversaries.

As shown in **Figures 20**–**22**, due to the increased intensity of packet dropping activities with the percentage increase in malicious nodes, the PDR of AODV declines from nearly 79 to 32%, 79 to 54% and 79 to 56% under *Attack1*, *Attack2* and *Attack3,* respectively. On the other hand, TRS-PD proves provides improvement in PDR of nearly 12 to 18%, 8 to 9% and 4.5 to 7% in the presence of malicious nodes launching *Attack1*, *Attack2* and *Attack3,* respectively. Meanwhile, in the presence of adversaries, ETRS-PD provides improvement in PDR over TRS-PD by an average of 7.67 under *Attack1*, 2.14 under *Attack2* and 4.29 under *Attack3*.

The NRO of AODV varies in the range of nearly 4.8–12.1, 3.9–4.8 and 4.7–5.4 under *Attack1*, *Attack2* and *Attack3,* respectively, as shown in **Figures 23**–**25**. On the other hand, TRS-PD improves NRO by maximum of 2.2 and 0.5 under *Attack1* and *Attack3* respectively over AODV. Meanwhile, TRS-PD increases NRO from nearly 0.7 to 2.0 under *Attack2* as compared to AODV. On the other hand, in the presence of adversaries, ETRS-PD provides improvement in NRO over TRS-PD by an average of 2.22 under *Attack1*, 0.25 under *Attack2* and 0.46 under *Attack3* due to the aforementioned reasons.

As shown in **Figures 26**–**28**, when employing TRS-PD, the AEC of the network without the presence of adversaries is 313.84 J while that is 312.35 J when employing ETRS-PD. As shown in **Figure 26**, the AEC for the MANET employing TRS-PD under *Attack1* varies between 314.08 A Composite Trust Model for Secure Routing in Mobile Ad-Hoc Networks http://dx.doi.org/10.5772/66519 39

**Figure 20.** PDR under *Attack1*.

**6.2. Test 2: varying percentage of malicious nodes**

**Figure 19.** AEC under *Attack3*.

38 Ad Hoc Networks

*Attack3* due to the aforementioned reasons.

In this test, the performance of the protocols is evaluated against *Attack1*, *Attack2* and *Attack3* by varying percentage of malicious nodes from 0 to 40% and keeping other parameters fixed.

As shown in **Figures 20**–**22**, due to the increased intensity of packet dropping activities with the percentage increase in malicious nodes, the PDR of AODV declines from nearly 79 to 32%, 79 to 54% and 79 to 56% under *Attack1*, *Attack2* and *Attack3,* respectively. On the other hand, TRS-PD proves provides improvement in PDR of nearly 12 to 18%, 8 to 9% and 4.5 to 7% in the presence of malicious nodes launching *Attack1*, *Attack2* and *Attack3,* respectively. Meanwhile, in the presence of adversaries, ETRS-PD provides improvement in PDR over TRS-PD by an

The NRO of AODV varies in the range of nearly 4.8–12.1, 3.9–4.8 and 4.7–5.4 under *Attack1*, *Attack2* and *Attack3,* respectively, as shown in **Figures 23**–**25**. On the other hand, TRS-PD improves NRO by maximum of 2.2 and 0.5 under *Attack1* and *Attack3* respectively over AODV. Meanwhile, TRS-PD increases NRO from nearly 0.7 to 2.0 under *Attack2* as compared to AODV. On the other hand, in the presence of adversaries, ETRS-PD provides improvement in NRO over TRS-PD by an average of 2.22 under *Attack1*, 0.25 under *Attack2* and 0.46 under

As shown in **Figures 26**–**28**, when employing TRS-PD, the AEC of the network without the presence of adversaries is 313.84 J while that is 312.35 J when employing ETRS-PD. As shown in **Figure 26**, the AEC for the MANET employing TRS-PD under *Attack1* varies between 314.08

The mobility parameter is kept fixed to 20 m/s for all three types of adversaries.

average of 7.67 under *Attack1*, 2.14 under *Attack2* and 4.29 under *Attack3*.

**Figure 21.** PDR under *Attack2*.

**Figure 22.** PDR under *Attack3*.

**Figure 23.** NRO under *Attack1*.

A Composite Trust Model for Secure Routing in Mobile Ad-Hoc Networks http://dx.doi.org/10.5772/66519 41

**Figure 24.** NRO under *Attack2*.

**Figure 25.** NRO under *Attack3*.

**Figure 23.** NRO under *Attack1*.

**Figure 22.** PDR under *Attack3*.

40 Ad Hoc Networks

**Figure 26.** AEC under *Attack1*.

**Figure 27.** AEC under *Attack2*.

**Figure 28.** AEC under *Attack3*.

**Figure 26.** AEC under *Attack1*.

42 Ad Hoc Networks

**Figure 27.** AEC under *Attack2*.

and 314.25 J. Meanwhile, ETRS-PD improves the AEC of TRS-PD by an average of 1.83 J in the presence of adversaries. As shown in **Figure 27**, the AEC of the MANET employing TRS-PD under *Attack2* decreases from 314.08 to 313.2 J. Meanwhile, ETRS-PD improves the AEC of TRS-PD by an average of 0.34 J in the presence of adversaries. As shown in **Figure 28**, the AEC of the MANET employing TRS-PD under *Attack3* varies between 312.68 and 313.41 J. Meanwhile, ETRS-PD improves the AEC of TRS-PD by an average of 0.67 J in the presence of adversaries.

## **7. Conclusions**

As a part of the literature survey, we observe that integration of QoS trust and social trust in the composition of a trust metric would improve the performance of a trust-based scheme. Considering these notes, we modify our previous trust-based scheme, TRS-PD, such that it combines both the types of trust components. In addition, we suggest modifications in the route discovery, trust update and trust recommendation procedures of TRS-PD. The proposed trustbased approach, ETRS-PD, improves the routing decisions due to the suggested modifications. The performance comparison of ETRS-PD with TRS-PD under three distinct adversary models shows that ETRS-PD achieves remarkable improvement in packet delivery ratio due to the enhanced routing process and inclusion of two new trust components. Moreover, ETRS-PD reduces the generation of number of control packets due to the reduced number of route handoff mechanisms. As a result, ETRS-PD provides improved normalized routing overhead as well as energy consumption as compared to TRS-PD under different network scenarios.

## **Author details**

Rutvij H. Jhaveri<sup>1</sup> \*, Narendra M. Patel<sup>2</sup> and Devesh C. Jinwala3

\*Address all correspondence to: rhj\_svmit@yahoo.com


3 Sardar Vallabhbhai National Institute of Technology, Surat, India

## **References**


[11] Mylsamy, R. and Sankaranarayanan, S. A preference-based protocol for trust and head selection for cluster-based MANET. Wireless Personal Communications. 2016;**86**(3): 1611–1627.

**Author details**

44 Ad Hoc Networks

Rutvij H. Jhaveri<sup>1</sup>

**References**

2010;**98**(10):1755–1772.

\*, Narendra M. Patel<sup>2</sup>

\*Address all correspondence to: rhj\_svmit@yahoo.com

2 Birla Vishvakarma Mahavidyalaya, V.V. Nagar, India

3 Sardar Vallabhbhai National Institute of Technology, Surat, India

networks. IET Information Security. 2012;**6**(2):77–83.

Springer International Publishing; 2015. pp. 237–250.

Lecture Notes on Software Engineering. 2015;**3**(4):318–324.

for mobile ad hoc networks. Computers. 2015;**4**(2):87–112.

Conference (ITNAC); IEEE; 2015. pp. 65–70.

Systems. 2016; DOI: 10.1002/dac.3148.

1 SVM Institute of Technology, Bharuch, India

and Devesh C. Jinwala3

[1] Junhai, L., Danxia, Y., Liu, X. and Mingyu, F. A survey of multicast routing protocols for mobile ad-hoc networks. IEEE Communications Surveys & Tutorials. 2009;**11**(1):78–91.

[2] Xia, H., Jia, Z., Ju, L., Li, X. and Sha, E.H.M. Impact of trust model on on-demand multi-path routing in mobile ad hoc networks. Computer Communications. 2013;**36**(9):1078–1093.

[3] Yu, H., Shen, Z., Miao, C., Leung, C. and Niyato, D. A survey of trust and reputation management systems in wireless communications. Proceedings of the IEEE.

[4] Marchang, N. and Datta, R. Light-weight trust-based routing protocol for mobile ad hoc

[5] Jhaveri, R.H. and Patel, N.M. Attack-pattern discovery based enhanced trust model for secure routing in mobile ad-hoc networks. International Journal of Communication

[6] Xia, H., Jia, Z., Li, X., Ju, L. and Sha, E.H.M. Trust prediction and trust-based source routing in mobile ad hoc networks. Ad Hoc Networks. 2013;**11**(7):2096–2114.

[7] Gharehkoolchian, M., Hemmatyar, A.A. and Izadi, M. Improving security issues in MANET AODV routing protocol. In: International Conference on Ad Hoc Networks;

[8] Airehrour, D., Gutierrez, J. and Ray, S.K. Gradetrust: a secure trust based routing protocol for MANETs. In: International Telecommunication Networks and Applications

[9] Patel, V.H., Zaveri, M.A. and Rath, H.K. Trust based routing in mobile ad-hoc networks.

[10] Chiejina, E., Xiao, H. and Christianson, B. A dynamic reputation management system


**Provisional chapter**

## **Performance Analysis of Three Routing Protocols in MANET Using the NS-2 and ANOVA Test with Varying Speed of Nodes MANET Using the NS-2 and ANOVA Test with Varying Speed of Nodes**

**Performance Analysis of Three Routing Protocols in** 

Subhrananda Goswami, Subhankar Joardar, Chandan Bikash Das, Samarajit Kar and Dibyendu Kumar Pal Chandan Bikash Das, Samarajit Kar and Dibyendu Kumar Pal Additional information is available at the end of the chapter

Subhrananda Goswami, Subhankar Joardar,

Additional information is available at the end of the chapter

http://dx.doi.org/10.5772/66521

## **Abstract**

In this chapter, we analyzed ad hoc on demand distance vector (AODV), dynamic source routing (DSR), and destination-sequenced distance vector (DSDV) routing protocols using different parameters of QoS metrics such as packet delivery ratio (PDR), normalize routing overhead, throughput, and jitter. The aim of this chapter is to determine a difference between routing protocol performance when operating in a large-area MANET with high-speed mobile nodes. After the simulations, we use AWK to analyze the data and then Xgraph to plot the performance metric. After that we use one-way ANOVA tools to confirm the correctness of the result. We use NS-2 for the simulation work. The comparison analysis of these protocols will be carrying out and in the last, we conclude that which routing protocol is the best one for mobile ad hoc networks.

**Keywords:** AODV, DSR, DSDV, MANET, throughput, packet delivery ratio, jitter, NS-2, ANOVA

## **1. Introduction**

The existing literature on MANETs is very extensive. An extremely comprehensive work is presented in Refs. [1, 2], which extensively covers most issues related to the subject, whereas in Refs. [3, 4], authors provide a brief introduction. MANET design issues such as a routing architecture in the light of the nature of MANETS, unidirectional link support, QoS routing, and multicast support are discussed in Refs. [5, 6]. In Ref. [7], the authors cover some of the

© 2016 The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. © 2017 The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

same design issues as mentioned in Ref. [5], but they augment them with some additional ones, such as limited bandwidth, energy constrained operation, and limited physical security.

Communication networks are evolving with a great pace witnessing increase in infrastructure and applications too. A mobile ad hoc network is the latest outcome in this research. The mobile ad hoc network, also known as MANET [8], is a network without any available infrastructure.

Nodes are mobile and can move whenever and wherever they want, because there is no centralized control or any other infrastructure is needed in any MANET. Each node in an MANET must be capable of functioning as a router to relay the traffic of other nodes.

A number of protocols have been developed for accomplish this task. Various dedicated routing protocols have been proposed to the Internet Engineering Task Force (IETF) MANET Working Group [8]. Some of these protocols have been studied, and their performances have been analyzed in detail. Broch et al. [9] evaluated four protocols using mobility and traffic scenarios similar to those we used. They focused on packet loss, routing message overhead, and route length. In Ref. [10], Johansson et al. compare three routing protocols, over extensive scenarios, varying node mobility, and traffic load. They focus on packet loss, routing overhead, throughput, and delay, and introduce mobility measures in terms of node relative speed. Finally, in Ref. [11], Das and coworkers compare the performance of two protocols, focusing on packet loss, packet end-to-end delay, and routing load. They obtained simulation results consistent with previous works and conclude with some recommendations for improving protocols. In this chapter, we measure and compare three performance parameter behaviors of two routing protocols, respectively, ad hoc on demand distance vector (AODV) [12] and destination-sequenced distance vector (DSDV).

## **2. MANET routing protocols**

This is the leading routing protocol proposed so for in the category of on demand or reactive routing protocols. Unlike table-driven protocols, it does not maintain status of the network via continuous updates [13]. This approach assists in minimizing the flooded messages and also size of route tables. It was designed after a distance vector routing protocol (DSDV) but is much efficient than DSDV. Actually, AODV is a combination of DSDV and dynamic source routing (DSR). It has the actual on-demand technique of discovering the route and also route maintenance from DSR but uses sequence numbering and also the periodic beacons of DSDV. New routes are found through the process of RREQ and RREP where RREQ packets are broadcast and RREPs are unicast in nature. While route maintenance uses RERR packets for remedy of route breaks, routing information is kept afresh by the usage of sequence numbers, which is the idea borrowed from DSDV [14].

The DSDV [15] is a proactive routing algorithm based upon a well-known classical distance vector algorithm of Bellman-Ford. Routing tables are maintained and updated accordingly, so broadcast periodic routing table update packets consume the bandwidth. So, the main weakness of DSDV is that when network grows these packets also increase. The main improvement here to the Bellman-Ford algorithm is loop freedom, which is made possible by assigning the sequence number to each entry in the routing table, which avoids stale routes.

The dynamic source routing (DSR) [10] is an on-demand or reactive routing protocol. Therefore, unlike other proactive routing protocols, DSR involves no updates of whichever type at any stage inside the network. The DSR uses source routing for forwarding data packets, which distinguishes DSR from other reactive routing protocols. It is lightweight on inner routers due to source routing, the maintaining routing information is not needed at every host. The sender becomes aware of complete destination address before transmission and appends this address in the header of the routing data packet at the beginning. It is loop free due to source routing. Extensive use of cache and promiscuously listening are the main optimizations to DSR when network is at low mobility.

## **3. Simulation model**

same design issues as mentioned in Ref. [5], but they augment them with some additional ones, such as limited bandwidth, energy constrained operation, and limited physical security. Communication networks are evolving with a great pace witnessing increase in infrastructure and applications too. A mobile ad hoc network is the latest outcome in this research. The mobile ad hoc network, also known as MANET [8], is a network without any available

Nodes are mobile and can move whenever and wherever they want, because there is no centralized control or any other infrastructure is needed in any MANET. Each node in an

A number of protocols have been developed for accomplish this task. Various dedicated routing protocols have been proposed to the Internet Engineering Task Force (IETF) MANET Working Group [8]. Some of these protocols have been studied, and their performances have been analyzed in detail. Broch et al. [9] evaluated four protocols using mobility and traffic scenarios similar to those we used. They focused on packet loss, routing message overhead, and route length. In Ref. [10], Johansson et al. compare three routing protocols, over extensive scenarios, varying node mobility, and traffic load. They focus on packet loss, routing overhead, throughput, and delay, and introduce mobility measures in terms of node relative speed. Finally, in Ref. [11], Das and coworkers compare the performance of two protocols, focusing on packet loss, packet end-to-end delay, and routing load. They obtained simulation results consistent with previous works and conclude with some recommendations for improving protocols. In this chapter, we measure and compare three performance parameter behaviors of two routing protocols, respectively, ad hoc on demand distance

This is the leading routing protocol proposed so for in the category of on demand or reactive routing protocols. Unlike table-driven protocols, it does not maintain status of the network via continuous updates [13]. This approach assists in minimizing the flooded messages and also size of route tables. It was designed after a distance vector routing protocol (DSDV) but is much efficient than DSDV. Actually, AODV is a combination of DSDV and dynamic source routing (DSR). It has the actual on-demand technique of discovering the route and also route maintenance from DSR but uses sequence numbering and also the periodic beacons of DSDV. New routes are found through the process of RREQ and RREP where RREQ packets are broadcast and RREPs are unicast in nature. While route maintenance uses RERR packets for remedy of route breaks, routing information is kept afresh by the usage of sequence numbers,

The DSDV [15] is a proactive routing algorithm based upon a well-known classical distance vector algorithm of Bellman-Ford. Routing tables are maintained and updated accordingly, so broadcast periodic routing table update packets consume the bandwidth. So, the main weakness of DSDV is that when network grows these packets also increase. The main improvement

MANET must be capable of functioning as a router to relay the traffic of other nodes.

vector (AODV) [12] and destination-sequenced distance vector (DSDV).

**2. MANET routing protocols**

which is the idea borrowed from DSDV [14].

infrastructure.

48 Ad Hoc Networks

The simulation software used in this chapter is the network simulator, NS-2 [16, 17]. The software version used is the latest release at the time of the commencement of simulation, namely, ns-2.34, which can be downloaded from Ref. [17]. In addition, many existing ad hoc routing protocol modules have already been implemented in NS-2. Three such protocols are AODV, DSR, and DSDV. NS-2 is a discrete-event-driven simulation software targeted for network simulation. This software is currently maintained by the Information Science Institute of University of Southern California.

## **3.1. Simulation evaluation methodology**

In order to analyze and compare the performance of the three routing protocols AODV, DSR, and DSDV, simulation experiments were performed. The purpose of the simulations was to compare the efficiency of the routing protocols based on different simulation parameters. The focus was concentrated on four performance metrics:


## **3.2. Results**

Generated trace file that is (.tr)

r -t 2.046566484 -Hs 1 -Hd -1 -Ni 1 -Nx 454.33 -Ny 337.37 -Nz 0.00 -Ne 9.996194 -Nl RTR -Nw — -Ma 0 -Md ffffffff -Ms 4 -Mt 800 -Is 1.255 -Id 9.255 -It DSR -Il 48 -If 0 -Ii 17 -Iv 32 -P dsr -Ph 2 -Pq 1 -Ps 2 -Pp 0 -Pn 2 -Pl 0 -Pe 0->16 -Pw 0 -Pm 0 -Pc 0 -Pb 0->0

## **3.3. NAM file output**

NAM is a Tcl/TK-based animation tool for viewing network simulation traces and realworld packet traces. Taking data from network simulators (such as ns) or live networks, NAM was one of the first tools to provide general purpose, packet-level, and network animation, before starting to use NAM, a trace file needs to create [16]. This trace file is usually generated by NS. Once the trace file is generated, NAM can be used to animate it. A snapshot of the simulation topology in NAM for 15 mobile nodes is shown in **Figure 1**, which is visualized the traces of communication or packet movements between mobile nodes [17].

**Figure 1.** A simple NAM file output.

The NAM file output for packet dropping is shown in **Figure 2**.

**Figure 2.** A NAM output with packet dropping.

## **4. Simulation results and observation**

## **4.1. Packet delivery ratio (PDR)**

**3.3. NAM file output**

50 Ad Hoc Networks

nodes [17].

**Figure 1.** A simple NAM file output.

**Figure 2.** A NAM output with packet dropping.

NAM is a Tcl/TK-based animation tool for viewing network simulation traces and realworld packet traces. Taking data from network simulators (such as ns) or live networks, NAM was one of the first tools to provide general purpose, packet-level, and network animation, before starting to use NAM, a trace file needs to create [16]. This trace file is usually generated by NS. Once the trace file is generated, NAM can be used to animate it. A snapshot of the simulation topology in NAM for 15 mobile nodes is shown in **Figure 1**, which is visualized the traces of communication or packet movements between mobile

The NAM file output for packet dropping is shown in **Figure 2**.

Packet delivery ratio (PDR) is defined as the ratio of data packets delivered successfully to destination nodes and the total number of data packets generated for those destinations. PDR characterizes the packet loss rate, which limits the throughput of the network. The higher the delivery ratio, better the performance of the routing protocol. The ratio of the data delivered to the destination to the data sent out by the source. PDR is determined as

to the destination to the data sent out by the source. POK is determined as

$$\text{PDR} = \left(\frac{\text{Received packets}}{\text{Sent packets}}\right) \* 100\tag{1}$$

**Figures 3**–**6** clearly indicate that the AODV routing protocol outcomes are better with the CBR traffic. AODV protocol performs better in comparison of other two selected routing protocols in such a network environment with varying speeds of nodes. So, we conclude that AODV is better in most of the PDR cases.

**Figure 3.** Packet delivery ratio (PDR) at 3 m/s.

**Figure 4.** Packet delivery ratio (PDR) at 10 m/s.

**Figure 5.** Packet delivery ratio (PDR) at 25 m/s.

**Figure 6.** Packet delivery ratio (PDR) at 50 m/s.

### **4.2. Throughput**

Throughput is defined as the ratio of the total data reaches a receiver from the sender. The time it takes by the receiver to receive the last message is called as throughput. Throughput is expressed as bytes or bits per sec (byte/sec or bit/sec). Some factors affect the throughput as; if there are many topology changes in the network, unreliable communication between nodes, limited bandwidth available, and limited energy. A high throughput is absolute choice in every network. Throughput can be represented mathematically as in equation. This represents the number of packets received by the destination within a given time interval. It is a measure of effectiveness of a routing protocol. Throughput <sup>=</sup> File size \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ Transmission time (bps) (2)

$$\text{Throughput} = \frac{\text{File size}}{\text{Transmission time (bps)}} \tag{2}$$

$$\text{Throughput} = \frac{\text{File size}}{\text{Transmission time (bps)}} \tag{2}$$

$$\text{Transmission time (bps)} = \frac{\text{File size}}{\text{Bandwidth (sec)}} \tag{3}$$

The analysis of **Figures 7**–**10** shows that performance of AODV is better than DSR and DSDV. Another characteristic that has come to the notice is that pause time does not have significant bearing on the throughput, whereas the performance is dictated only by the density of the network.

Performance Analysis of Three Routing Protocols in MANET Using the NS-2 and ANOVA Test with Varying... http://dx.doi.org/10.5772/66521 53

### **Figure 7.** Throughput at 3 m/s.

**Figure 8.** Throughput at 10 m/s.

**4.2. Throughput**

52 Ad Hoc Networks

**Figure 6.** Packet delivery ratio (PDR) at 50 m/s.

**Figure 5.** Packet delivery ratio (PDR) at 25 m/s.

measure of effectiveness of a routing protocol.

Throughput is defined as the ratio of the total data reaches a receiver from the sender. The time it takes by the receiver to receive the last message is called as throughput. Throughput is expressed as bytes or bits per sec (byte/sec or bit/sec). Some factors affect the throughput as; if there are many topology changes in the network, unreliable communication between nodes, limited bandwidth available, and limited energy. A high throughput is absolute choice in every network. Throughput can be represented mathematically as in equation. This represents the number of packets received by the destination within a given time interval. It is a

Throughput <sup>=</sup> File size \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ Transmission time (bps) (2)

Transmission time (bps) <sup>=</sup> \_\_\_\_\_\_\_\_\_\_\_\_\_ File size Bandwidth (sec) (3)

The analysis of **Figures 7**–**10** shows that performance of AODV is better than DSR and DSDV. Another characteristic that has come to the notice is that pause time does not have significant bearing on the throughput, whereas the performance is dictated only by the density of the network.

**Figure 9.** Throughput at 25 m/s.

**Figure 10.** Throughput at 50 m/s.

## **4.3. Normalized routing overhead**

This is the ratio of routing-related transmissions (RREQ, RREP, RERR, etc.) to data transmissions in a simulation. A transmission is one node either sending or forwarding a packet. Either way, the routing load per unit data successfully delivered to the destination.

It is the total number of control or routing (RTR) packets generated by routing protocol during the simulation. All packets sent or forwarded at network layer is consider routing overhead.

$$\text{Routing overhead} \newline \text{= Number of RTR packets} \tag{4}$$

Based on the result of simulation, **Figures 11**–**14** show that the performance of DSDV is better than AODV and DSR. At all the considered mobility, DSDV is the best protocol as compared to other protocols.

**Figure 11.** Normalized routing overhead at 3 m/s.

Performance Analysis of Three Routing Protocols in MANET Using the NS-2 and ANOVA Test with Varying... http://dx.doi.org/10.5772/66521 55

**Figure 12.** Normalized routing overhead at 10 m/s.

**4.3. Normalized routing overhead**

**Figure 10.** Throughput at 50 m/s.

**Figure 11.** Normalized routing overhead at 3 m/s.

overhead.

54 Ad Hoc Networks

to other protocols.

This is the ratio of routing-related transmissions (RREQ, RREP, RERR, etc.) to data transmissions in a simulation. A transmission is one node either sending or forwarding a packet.

It is the total number of control or routing (RTR) packets generated by routing protocol during the simulation. All packets sent or forwarded at network layer is consider routing

 Routing overhead = Number of RTR packets (4) Based on the result of simulation, **Figures 11**–**14** show that the performance of DSDV is better than AODV and DSR. At all the considered mobility, DSDV is the best protocol as compared

Either way, the routing load per unit data successfully delivered to the destination.

**Figure 13.** Normalized routing overhead at 25 m/s.

**Figure 14.** Normalized routing overhead at 50 m/s.

## **4.4. Jitter**

The term jitter is often used as a measure of the variability over time of the packet latency across a network. A network with constant latency has no variation (or jitter). Packet jitter is expressed as an average of the deviation from the network mean latency. However, for this use, the term is imprecise [13]. Or in other words, jitter is the variation of the packet arrival time. In jitter calculation, the variation in the packet arrival time is expected to minimum. The delays between the different packets need to be low if we want better performance in mobile ad hoc networks.

Based on the result of simulation, **Figures 15** and **16** show that the performance of AODV and DSR gives the better result. **Figures 17** and **18** show that DSR gives the better performance.

**Figure 15.** Jitter at 3 m/s.

**Figure 16.** Jitter at 10 m/s.

Performance Analysis of Three Routing Protocols in MANET Using the NS-2 and ANOVA Test with Varying... http://dx.doi.org/10.5772/66521 57

**Figure 17.** Jitter at 25 m/s.

**4.4. Jitter**

56 Ad Hoc Networks

ad hoc networks.

performance.

**Figure 15.** Jitter at 3 m/s.

**Figure 16.** Jitter at 10 m/s.

The term jitter is often used as a measure of the variability over time of the packet latency across a network. A network with constant latency has no variation (or jitter). Packet jitter is expressed as an average of the deviation from the network mean latency. However, for this use, the term is imprecise [13]. Or in other words, jitter is the variation of the packet arrival time. In jitter calculation, the variation in the packet arrival time is expected to minimum. The delays between the different packets need to be low if we want better performance in mobile

Based on the result of simulation, **Figures 15** and **16** show that the performance of AODV and DSR gives the better result. **Figures 17** and **18** show that DSR gives the better

**Figure 18.** Jitter at 50 m/s.

## **5. ANOVA test**

Analysis of variance (ANOVA) is a collection of statistical models used to analyze the differences between group means and their associated procedures (such as "variation" among and between groups), in which the observed variance in a particular variable is partitioned into components attributable to different sources of variation [18].

In this chapter, we have use one-way ANOVA. One-way ANOVA is used to study the effect of (*k* > 2) levels of a single factor. A factor is defined as characteristics under consideration, thought to influence the measured observation. A level is defined as a value of a factor.

## **5.1. Output of the test for different parameters**

## *5.1.1. Packet delivery ratio*

The packet delivery ratio (PDR) is very much related to the throughput metric. The destination records the number of data packets it received and estimates the PDR delivery ratio in the network from the count of the data packets sent. The ANOVA hypothesis test is shown in **Table 1**, there is sufficient evidence to reject the null hypothesis. We see that there is a significant different in PDR performance when the network adopts different routing methods (*P-*value > 0.05).


**Table 1.** Summary of packet delivery ratio.

The one-way ANOVA test for PDR is shown in **Table 2**.

In this case, *F*crit = 3.135918 at *α* = 0.05. Since *F* = 0.560241875 < 3.135918, the results are significant at the 5% significance level. So, we will accept the null hypothesis, and conclusion can be drawn that there is strong evidence that the expected values in the three groups do not differ. The variation is quite small and can be eliminated at this significance level. The *P-*value for this test is 0.573763.


**Table 2.** ANOVA of packet delivery ratio.

## *5.1.2. Throughput*

Data throughput is defined as the total number of packets delivered over the total simulation time. ANOVA statistical computation shows that we do not reject the null hypothesis. That is, there is no significant difference for the different methods in terms of throughput performance (*P-*value > 0.05) (**Table 3**).

Performance Analysis of Three Routing Protocols in MANET Using the NS-2 and ANOVA Test with Varying... http://dx.doi.org/10.5772/66521 59


**Table 3.** Summary of throughput.

**5.1. Output of the test for different parameters**

The one-way ANOVA test for PDR is shown in **Table 2**.

Within groups 7670.3216 66 116.217

Total 7800.5409 68

**Table 1.** Summary of packet delivery ratio.

**Table 2.** ANOVA of packet delivery ratio.

The packet delivery ratio (PDR) is very much related to the throughput metric. The destination records the number of data packets it received and estimates the PDR delivery ratio in the network from the count of the data packets sent. The ANOVA hypothesis test is shown in **Table 1**, there is sufficient evidence to reject the null hypothesis. We see that there is a significant different in PDR performance when the network adopts different routing methods

**Groups Count Sum Average Variance** AODV 23 802.8693 34.90736 121.3164733 DSDV 23 744.1666 32.35507 171.2711624 DSR 23 729.8366 31.73203 56.06334723

In this case, *F*crit = 3.135918 at *α* = 0.05. Since *F* = 0.560241875 < 3.135918, the results are significant at the 5% significance level. So, we will accept the null hypothesis, and conclusion can be drawn that there is strong evidence that the expected values in the three groups do not differ. The variation is quite small and can be eliminated at this significance level. The *P-*value

**Source of variation SS Df MS** *F P***-value** *F***crit** Between groups 130.21925 2 65.10963 0.560241875 0.573763 3.135918

Data throughput is defined as the total number of packets delivered over the total simulation time. ANOVA statistical computation shows that we do not reject the null hypothesis. That is, there is no significant difference for the different methods in terms of throughput perfor-

*5.1.1. Packet delivery ratio*

58 Ad Hoc Networks

(*P-*value > 0.05).

for this test is 0.573763.

*5.1.2. Throughput*

mance (*P-*value > 0.05) (**Table 3**).

The one-way ANOVA test for throughput is shown in **Table 4**.


**Table 4.** ANOVA of throughput.

In this case, *F*crit = 3.135918 at *α* = 0.05. Since *F* = 0.778278814 < 3.135918, the results are significant at the 5% significance level. So, we will accept the null hypothesis, and conclusion can be drawn that there is strong evidence that the expected values in the three groups do not differ. The variation is quite small and can be eliminated at this significance level. The *P-*value for this test is 0.463364.

## *5.1.3. Normalized routing overhead*

Using the ANOVA hypothesis testing, the simulation results show a significant difference among methods used in terms of normalized routing overhead (*P-*value > 0.05). Thus, normalized routing overhead can be used as a metric to measure the performance of different algorithms (**Table 5**).


**Table 5.** Summary of normalized routing overhead.

The one-way ANOVA test for normalized routing overhead is shown in **Table 6**.


**Table 6.** ANOVA of normalized routing overhead.

In this case, *F*crit = 3.135918 at *α* = 0.05. Since *F* = 7.766781596 > 3.135918, the results are significant at the 5% significance level. So, we will reject the null hypothesis, and conclusion can be drawn that there is strong evidence that the expected values in the three groups differ significantly. The *P-*value for this test is 0.000935.

## *5.1.4. Jitter*

The term jitter often used as a measure of the packet of the variability over time of the packet latency across a network. A network with constant latency has no variation (or jitter). Packet jitter is expressed as an average of the derivation from the network mean latency. ANOVA statistical computation shows that we do not reject the null hypothesis. That is, there is no significant difference for the different methods in terms of throughput performance (*P-*value > 0.05) (**Table 7**).


**Table 7.** Summary of jitter.

The one-way ANOVA test for jitter is shown in **Table 8**.


**Table 8.** ANOVA of jitter.

In this case, *F*crit = 3.135918 at *α* = 0.05. Since *F* = 0.108241 < 3.135918, the results are significant at the 5% significance level. So, we will accept the null hypothesis, and conclusion can be drawn that there is strong evidence that the expected values in the three groups do not differ. The variation is quite small and can be eliminated at this significance level. The *P-*value for this test is 0.89757.

## **6. Conclusion**

In this case, *F*crit = 3.135918 at *α* = 0.05. Since *F* = 7.766781596 > 3.135918, the results are significant at the 5% significance level. So, we will reject the null hypothesis, and conclusion can be drawn that there is strong evidence that the expected values in the three groups differ

**Source of variation SS df MS** *F P***-value** *F***crit** Between groups 4.4470485 2 2.223524 7.766781596 0.000935 3.135918

The term jitter often used as a measure of the packet of the variability over time of the packet latency across a network. A network with constant latency has no variation (or jitter). Packet jitter is expressed as an average of the derivation from the network mean latency. ANOVA statistical computation shows that we do not reject the null hypothesis. That is, there is no significant difference for the different methods in terms of throughput performance (*P-*value > 0.05) (**Table 7**).

**Groups Count Sum Average Variance** AODV 23 0.129171 0.005616 2.07E−06 DSDV 23 0.125752 0.005467 5.9E−06 DSR 23 0.131442 0.005715 1.91E−06

In this case, *F*crit = 3.135918 at *α* = 0.05. Since *F* = 0.108241 < 3.135918, the results are significant at the 5% significance level. So, we will accept the null hypothesis, and conclusion can be drawn that

**Source of variation SS df MS** *F P***-value** *F***crit** Between Groups 7.14E−07 2 3.57E−07 0.108241 0.89757 3.135918

significantly. The *P-*value for this test is 0.000935.

Within groups 18.894905 66 0.286286

Total 23.341954 68

**Table 6.** ANOVA of normalized routing overhead.

The one-way ANOVA test for jitter is shown in **Table 8**.

Within Groups 0.000218 66 3.3E−06

Total 0.000218 68

*5.1.4. Jitter*

60 Ad Hoc Networks

**Table 7.** Summary of jitter.

**Table 8.** ANOVA of jitter.

The results indicate that the performance is better especially when the number of nodes in the network is higher. In this chapter, we have used a simulator that provides the virtual environment for the testing different parameters. Reactive routing protocol AODV performance is the best considering due to its ability to maintain connection by periodic exchange of information. Using NS-2 simulator we created the scenarios under which using tcl script, it is run. After analyzing the X-graphs, we concluded that AODV indicates its highest efficiency and performance under high mobility than DSR and DSDV, and the performance of TCP and UDP packets with respect to normalized routing overhead, jitter, throughput, and PDR, and the performance of AODV is better than DSDV and DSR routing protocol for real-time applications from the simulation results.

After that in one-way ANOVA test, AODV exhibits better routing performance compared with conventional routing methods such as DSDV and DSR. By performing an ANOVA analysis at the initial stage, we conclude that there is a significant difference in the performance metrics when using different routing algorithms. From there, we analyze the difference of the means and boundaries in 95% confidence interval. In all simulation scenarios, we see that AODV shows a lower packet loss and lower delay. It offers higher throughput and ensures higher packet delivery ratio.

## **Author details**

Subhrananda Goswami1,\*, Subhankar Joardar2 , Chandan Bikash Das<sup>3</sup> , Samarajit Kar<sup>4</sup> and Dibyendu Kumar Pal<sup>5</sup>

\*Address all correspondence to: subhrananda\_usca@yahoo.co.in

1 Department of Information Technology, Global Group of Institutions, Haldia, Purba Midnapore, West Bengal, India

2 Department of CSE, Haldia Institute of Technology, Haldia, Purba Medinipur, West Bengal, India

3 Department of Mathematics, Tamralipta Mahavidyalaya, Tamluk, Purba Midnapore, West Bengal, India

4 Department of Mathematics, National Institute of Technology, Durgapur, Burdwan, West Bengal, India

5 Department Of Computer Application, Asansol Engineering College, Asansol, Burdwan, West Bengal, India

## **References**


[14] Papadimitratos P. and Haas Z.J. Secure on-demand distance vector routing in ad hoc networks. *IEEE/Sarnoff Symposium on Advances in Wired and Wireless Communication.* Princeton, NJ, 2005, pp. 168–171.

**References**

62 Ad Hoc Networks

141p.

85–97.

16–28. 2001.

Upper Saddle River, NJ, 2004.

Computer Science. 2014;**6**(10):11–18.

Beijing, China. 2000, pp. 1298–1303.

*Network (MANET) Working Group, IETF*. 1998.

Internet Draft, MANET Working Group. 2000.

www.ietf,org/html.charters/manet-charter.html; 1999

2014;**3**(1):4908–4914.

[1] Murthy R.S.C. and Manoj B.S. Ad Hoc Wireless Networks- Architectures. Prentice Hall,

[2] Goswami S., Joardar S., Das C.B., and Das B. A simulation based performance comparison of AODV and DSDV mobile ad hoc networks. Information Technology and

[3] Remondo D. Tutorial on wireless ad hoc networks [Internet]. 2004. Available from:

[4] Goswami S., Joardar S., and Das C.B. Performance Comparison of Routing Protocols of MANET Using NS-2. 1st ed. Germany: LAP LAMBERT Academic Publishing; 2014,

[5] Chun Y.L. and Lin S.M. Routing protocols overview and design issues for self-organizednetwork [sic]. International Conference on Communication Technology Proceedings;

[6] Goswami S., Joardar S., and Das C.B. Reactive and proactive routing protocol performance metric comparison in mobile ad hoc networks using NS 2. International Journal of Advanced Research in Computer and Communication Engineering.

[7] Macker J.P. and Corson M.S. Mobile ad hoc networking (MANET):Routing protocol performance issues and evaluation considerations (Internet-draft), in: *Mobile Ad-hoc* 

[8] Internet Engineering Task Force MANET Working Group Charter. Available from:

[9] Broch J., Maltz D.A., Johnson D.B., Hu Y.C., and Jetcheva J. A performance comparisonof multi-hop wireless ad hoc network routing protocols. Proceedings of the Fourth Annual ACM; IEEE International Conference on Mobile Computing and Networking; 1998, pp.

[10] Broch J., Maltz D.A., and Johnson D.B. The dynamic source routing protocol for mobil-

[11] Perkins C.E., Royer E.M., and Das S.R. Performance comparison of two on-demand routing protocols for ad hoc networks. in *IEEE Personal Communications*. vol. 8. no. 1. pp.

[12] Perkins C.E. and Royer E.M. Ad hoc on-demand distance vector (AODV) routing.

[13] Belding-Royer E.M., Perkins C.E., and Chakeres I. *Ad hoc On-Demand Distance Vector (AODV) Routing: Work in progress*. July 2004. Internet Draft, RFC 3561bis-01, http://

ead hoc networks. Internet Draft, MANET Working Group. 1999.

moment. cs. ucsb. edu/pub/draft-perkins-manet-aodvbis-02. txt.

www.comp.brad.ac.uk/het-net/HET-NETs04/CameraPapers/T2.pdf


**Provisional chapter**

## **Cooperative Routing in Multi-Radio Multi-Hop Wireless Network Wireless Network**

**Cooperative Routing in Multi-Radio Multi-Hop** 

Kun Xie, Shiming He, Xin Wang, Dafang Zhang and Keqin Li Zhang and Keqin Li Additional information is available at the end of the chapter

Additional information is available at the end of the chapter

Kun Xie, Shiming He, Xin Wang, Dafang

http://dx.doi.org/10.5772/66414

## **Abstract**

There are many recent interests on cooperative communication (CC) in wireless networks. Despite the large capacity gain of CC in small wireless networks, CC can result in severe interference in large networks and even degraded throughput. The aim of this chapter is to concurrently exploit multi-radio and multi-channel (MRMC) and CC technique to combat co-channel interference and improve the performance of multi-hop wireless network. Our proposed solution concurrently considers cooperative routing, channel assignment, and relay selection and takes advantage of both MRMC technique and spatial diversity to improve the throughput. We propose two important metrics, contention-aware channel utilization routing metric (CACU) to capture the interference cost from both direct and cooperative transmission, and traffic aware channel condition metric (TACC) to evaluate the channel load condition. Based on these metrics, we propose three algorithms for interference-aware cooperative routing, local channel adjustment, and local path and relay adaptation, respectively, to ensure high-performance communications in dynamic wireless networks. Our algorithms are fully distributed and can effectively mitigate co-channel interference and achieve cooperative diversity gain. To our best knowledge, this is the first distributed solution that supports CC in MRMC networks. Our performance studies demonstrate that our algorithms can significantly increase the aggregate throughput.

**Keywords:** cooperative routing, relay assignment, channel assignment

## **1. Introduction**

As an emerging technique for future wireless networks, cooperative communication (CC) has been proposed to take advantage of the broadcast nature of wireless communications and spatial diversity to improve the network performance [1, 2]. More specifically, relay

nodes have been exploited to forward the replica of packets from the sources, and the destinations can combine multiple copies of the signal to better decode the original message. Taking advantage of spatial and multiuser diversities, CC can efficiently improve the network performance.

Despite the significant performance gain in small networks, recent research results show through both analysis and simulation that the use of cooperative relays (CRs) in largescale wireless networks can lead to severe interference, which in turn results in higher packet loss and consequent throughput reduction [3–5]. Although relay nodes may help to increase the throughput of a single source and destination pair, a cooperative transmission (CT) often involves three transmission links (i.e., from the source to the relay, from the source to the destination, and from the relay to the destination). The increase of transmission links in a neighborhood leads to higher interference, thus reducing the network-wise performance [6]. When the interference is severe, the performance can be even worse than without using cooperative transmissions. It is critical to reduce the interference for CC to work efficiently in a practical wireless network, especially when the network scale is large.

Another recent technique, multi-radio multi-channel (MRMC), has been exploited to alleviate the co-channel interference by supporting concurrent transmissions over orthogonal channels to improve the network capacity [7–9]. With the growth of modern wireless technologies, the cost of radio chips including those supporting 802.11 [10, 11] constantly reduces and more devices will be equipped with multiple radios.

In this chapter, we exploit MRMC to alleviate the interference in a network with cooperative communications for potentially much higher network performance. In cooperative networks, a routing path can be formed with a combination of cooperative transmissions and direct transmissions (DTs), and we call this kind of routing **cooperative routing**. The important and interesting question this chapter tries to answer is what is the maximum aggregate throughput of a multi-radio multi-channel network when the cooperative transmission is available? Current studies on cooperative communications in multi-hop wireless network generally assume that the network nodes are equipped with only a single antenna [12–20], and it is unclear what capacity and performance gain can be achieved if nodes are equipped with multiple antennas. Despite the large potential benefit, it is highly non-trivial to make both CC and MRMC techniques to work seamlessly together. Some of the challenges are as follows.

First, the coupled cooperative routing problem and relay selection problem should be solved together. Different from conventional routing in MRMC networks where every node just needs to find the next-hop node to forward packets toward the destination, with cooperative routing, a neighbor of the transmitter not only needs to serve as a multi-hop transmission relay (MR) for packet forwarding but may also act as a cooperative relay (CR) of the transmitter for cooperative transmission. The capability for a node or a radio interface to serve as two different types of relay makes multi-radio cooperative routing and relay node assignment inter-dependent.

Second, there is a trade-off between alleviating co-channel interference and exploiting cooperative diversity. In single-radio single-channel cooperative wireless networks, one-hop neighbors of a transmitter are candidate MR or CR nodes. A transmitter node can determine to use direct transmission and find an MR or cooperative transmission and find a CR to maximize the cooperative transmission gain. Although MRMC can largely relieve the co-channel interference, only the node, which tunes to the same channel as that of the transmitter, can act as an MR or a CR, which reduces the number of candidate relay nodes. This makes it important and challenging to consider radio-channel assignment along with cooperative communications.

nodes have been exploited to forward the replica of packets from the sources, and the destinations can combine multiple copies of the signal to better decode the original message. Taking advantage of spatial and multiuser diversities, CC can efficiently improve the net-

Despite the significant performance gain in small networks, recent research results show through both analysis and simulation that the use of cooperative relays (CRs) in largescale wireless networks can lead to severe interference, which in turn results in higher packet loss and consequent throughput reduction [3–5]. Although relay nodes may help to increase the throughput of a single source and destination pair, a cooperative transmission (CT) often involves three transmission links (i.e., from the source to the relay, from the source to the destination, and from the relay to the destination). The increase of transmission links in a neighborhood leads to higher interference, thus reducing the network-wise performance [6]. When the interference is severe, the performance can be even worse than without using cooperative transmissions. It is critical to reduce the interference for CC to work efficiently in a practical wireless network, especially when

Another recent technique, multi-radio multi-channel (MRMC), has been exploited to alleviate the co-channel interference by supporting concurrent transmissions over orthogonal channels to improve the network capacity [7–9]. With the growth of modern wireless technologies, the cost of radio chips including those supporting 802.11 [10, 11] constantly reduces and more

In this chapter, we exploit MRMC to alleviate the interference in a network with cooperative communications for potentially much higher network performance. In cooperative networks, a routing path can be formed with a combination of cooperative transmissions and direct transmissions (DTs), and we call this kind of routing **cooperative routing**. The important and interesting question this chapter tries to answer is what is the maximum aggregate throughput of a multi-radio multi-channel network when the cooperative transmission is available? Current studies on cooperative communications in multi-hop wireless network generally assume that the network nodes are equipped with only a single antenna [12–20], and it is unclear what capacity and performance gain can be achieved if nodes are equipped with multiple antennas. Despite the large potential benefit, it is highly non-trivial to make both CC and MRMC techniques to work seamlessly together. Some of the chal-

First, the coupled cooperative routing problem and relay selection problem should be solved together. Different from conventional routing in MRMC networks where every node just needs to find the next-hop node to forward packets toward the destination, with cooperative routing, a neighbor of the transmitter not only needs to serve as a multi-hop transmission relay (MR) for packet forwarding but may also act as a cooperative relay (CR) of the transmitter for cooperative transmission. The capability for a node or a radio interface to serve as two different types of relay makes multi-radio cooperative routing and relay node

work performance.

66 Ad Hoc Networks

the network scale is large.

lenges are as follows.

assignment inter-dependent.

devices will be equipped with multiple radios.

Third, the use of cooperative relays in cooperative communications makes the network interference condition more complicated than that in a network with only direct transmissions, and it desires careful design to reduce the interference along with the finding of the cooperative routing path and channel assignment in MRMC cooperative networks.

In summary, there is an inter-dependence among cooperative routing, channel assignment, and relay selection. To enable cooperative communications in MRMC wireless networks and fulfill the complete potential of both techniques, the three problems need to be systematically solved together. A few recent studies [21–25] have begun to investigate interference-aware cooperative routing algorithms to solve the challenge problems. This chapter presents a practical and distributed solution to effectively exploit both MRMC technique and cooperative diversity to ensure higher performance of a multi-hop network with dynamic channel conditions and traffic flows. In this design, the cooperative routing at the network layer, channel assignment at the MAC layer, and cooperative communication at the physical layer will work interactively and seamlessly together. The main techniques are as follows:


The remaining of this chapter is organized as follows. We introduce our system model in Section 2. Motivation example and solution overview are presented in Section 3. We present the detailed algorithms on cooperative routing, channel assignment, and local path and relay adjustment in Sections 4–6, respectively. The complete solution is presented in Section 7. Simulation results are given in Section 8. We conclude the work in Section 9.

## **2. System model**

We consider a multi-hop cooperative wireless network where a node can be equipped with multiple radios. We call this network MRMC cooperative wireless network. There are *N* nodes in the network. Each node *i* in the network is equipped with one or more radio interfaces (wireless NIC), represented by *I*(*i*). Each radio can serve as a transmitter or a receiver on a channel at a given time. We assume that there are a total of *K* orthogonal channels in the network, numbered *Ch*<sup>1</sup> , *Ch*<sup>2</sup> , …, *Ch*K*,* and there is no inter-channel interference. The channel assignment is trivial if the number of orthogonal channels available is at most the minimum number of radios per node *I*(*i*), since in this case every node must be assigned all the channels. This chapter assumes that *K* is larger than max*<sup>i</sup>*∈*<sup>N</sup> I*(*i*). A radio is capable of selecting a working channel from the set of orthogonal channels, and the set of working channels of node *i* is denoted as *w*(*i*). Due to the interference constraints, there is no capacity benefit in equipping two different radios of a node with the same channel. There are multiple concurrent flows, denoted by a set *F* = {*F*<sup>1</sup> *, F*<sup>2</sup> ,..., *F*M} of *M* flows. The data for each flow may traverse multiple hops in the network. A flow *F*<sup>i</sup> (*S*<sup>i</sup> →*D*<sup>i</sup> ) goes through a pair of source node and destination node, denoted as *S*<sup>i</sup> and *D*<sup>i</sup> , respectively.

There are two transmission modes between any two nodes in the network considered, direct transmission (DT) and cooperative transmission (CT), as shown in **Figure 1**. Direct transmission mode is widely employed in current wireless networks, where a source node transmits its signal directly to a destination node. The achievable rate of *C*DT(*S*, *D*) between *S* and *D* is expressed as follows:

$$\mathbb{C}\_{\text{Dir}}(\mathbf{S}, D) = W^\* \log\_2(1 + \text{SNR}(\mathbf{S}, D)). \tag{1}$$

A cooperative transmission involves three nodes and three links. Specifically, a collaborative neighbor *R* overhears the signal from source *S* and forwards the signal to the destination *D*, which then combines two signal streams, *S*→*D* and *R*→*D*, into a single stream that has a higher resistance to channel fading and noise and hence a higher probability of being successfully decoded. The mechanism to accomplish CT is not unique. In Ref. [3], the authors describe and compare the capacity of different cooperative transmission protocols and show

**Figure 1.** Two kinds of transmission modes.

that the AF-RAKE-based cooperative transmission protocol can achieve the maximum capacity. In AF-RAKE, *R* receives signals from *S* and amplifies and forwards them to *D* without demodulation or decoding. *D* uses a RAKE receiver to combine both signal streams of *S*→*D* and *R*→*D*. The achievable rate of *C*CT(*S*, *R*, *D*) between *S* and *D* with *R* as relay under AF-RAKE mode [3] is given by

and  $\kappa \to \nu$ . In the acutebare rate on  $\mathsf{C}\_{\rm crl}(S, \mathsf{A}, \mathsf{D})$  between  $\mathsf{S}$  and  $\nu$  will  $\kappa$  as relay under ArRAKE mode [3] is given by

$$\mathsf{C}\_{\rm crl}(S, \mathsf{R}, \mathsf{D}) = W^\* \log\_2(1 + \text{SNR}(S, \mathsf{D}) + \frac{\text{SNR}(S, \mathsf{R})^\* \text{SNR}(\mathsf{R}, \mathsf{D})}{\text{SNR}(S, \mathsf{R}) + \text{SNR}(\mathsf{R}, \mathsf{D}) + 1}) \tag{2}$$

$$\text{SNR(m, n)} = \frac{P\_m}{\sigma\_n^2} \left| h\_{m, n} \right|^2. \tag{3}$$

In the above equation, *P*m denotes the transmission power at node *m*. *h*m,n captures the effects of path loss, shadowing, and fading within the channel between *m* and *n*. The noise components are modeled as white Gaussian noise (AWGN). *σ<sup>n</sup>* 2 denotes variance of the background noise at nodes *n*.

Relay nodes can be categorized into two types based on their functions: a cooperative relay (CR), which operates at the physical layer for cooperative transmission (i.e., node *R*<sup>6</sup> in flow *F*<sup>3</sup> in **Figure 2**), and a multi-hop relay (MR), which operates at the network layer to relay packets from a source over multiple hops to its destination (i.e., node *R*<sup>1</sup> in flow *F*<sup>2</sup> in **Figure 2**). A node with multiple radios can serve as both CR and MR for multiple flows. For example, *R*<sup>3</sup> acts as CR relay in *F*<sup>3</sup> , but MR relay for *F*<sup>4</sup> . More complex function roles can be found on *R*<sup>7</sup> , which acts as CR relay in both *F*<sup>1</sup> and *F*<sup>3</sup> , and MR relay in *F*<sup>2</sup> .

A cooperative routing path could be a combination of cooperative transmissions and direct transmissions. For example, flow *F*<sup>3</sup> (*S*3 →*D*<sup>3</sup> ) = *S*<sup>3</sup> *Ch*2 ⎯→*R*2 (*R*<sup>3</sup> ) *Ch*3 ⎯→*R*4 (*R*<sup>7</sup> ) *Ch*1 ⎯→*R*5 *Ch* ⎯→2 *D*3 (*R*6 ), in **Figure 2**, where the first hop link *l S*3 *R*2 (*R*<sup>3</sup> ) *Ch*<sup>2</sup> , the second hop link *l R*2 *R*4 (*R*<sup>7</sup> ) *Ch*<sup>3</sup> , and the fourth hop link *l R*5 *D*3 (*R*<sup>6</sup> ) *Ch*2 adopt cooperative transmission mode with nodes *R*<sup>3</sup> , *R*<sup>7</sup> , and *R*<sup>6</sup> acting as CR relays, respectively, while the third hop link *l R*4 *R*5 *Ch*1 adopts the direct transmission mode with both *R*<sup>4</sup> and *R*<sup>5</sup> being MR relays.

**Figure 2.** The MRMC cooperative network.

relay adjustment in Sections 4–6, respectively. The complete solution is presented in Section 7.

We consider a multi-hop cooperative wireless network where a node can be equipped with multiple radios. We call this network MRMC cooperative wireless network. There are *N* nodes in the network. Each node *i* in the network is equipped with one or more radio interfaces (wireless NIC), represented by *I*(*i*). Each radio can serve as a transmitter or a receiver on a channel at a given time. We assume that there are a total of *K* orthogonal channels in the

assignment is trivial if the number of orthogonal channels available is at most the minimum number of radios per node *I*(*i*), since in this case every node must be assigned all the channels. This chapter assumes that *K* is larger than max*<sup>i</sup>*∈*<sup>N</sup> I*(*i*). A radio is capable of selecting a working channel from the set of orthogonal channels, and the set of working channels of node *i* is denoted as *w*(*i*). Due to the interference constraints, there is no capacity benefit in equipping two different radios of a node with the same channel. There are multiple concurrent flows,

There are two transmission modes between any two nodes in the network considered, direct transmission (DT) and cooperative transmission (CT), as shown in **Figure 1**. Direct transmission mode is widely employed in current wireless networks, where a source node transmits its signal directly to a destination node. The achievable rate of *C*DT(*S*, *D*) between *S* and *D* is

A cooperative transmission involves three nodes and three links. Specifically, a collaborative neighbor *R* overhears the signal from source *S* and forwards the signal to the destination *D*, which then combines two signal streams, *S*→*D* and *R*→*D*, into a single stream that has a higher resistance to channel fading and noise and hence a higher probability of being successfully decoded. The mechanism to accomplish CT is not unique. In Ref. [3], the authors describe and compare the capacity of different cooperative transmission protocols and show

, …, *Ch*K*,* and there is no inter-channel interference. The channel

,..., *F*M} of *M* flows. The data for each flow may traverse multiple

) goes through a pair of source node and destination

(1 + *SNR*(*S*, *D*)). (1)

Simulation results are given in Section 8. We conclude the work in Section 9.

**2. System model**

68 Ad Hoc Networks

network, numbered *Ch*<sup>1</sup>

denoted by a set *F* = {*F*<sup>1</sup>

node, denoted as *S*<sup>i</sup>

expressed as follows:

hops in the network. A flow *F*<sup>i</sup>

**Figure 1.** Two kinds of transmission modes.

, *Ch*<sup>2</sup>

*, F*<sup>2</sup>

and *D*<sup>i</sup>

*C*DT(*S*, *D*) = *W* \* log2

(*S*<sup>i</sup> →*D*<sup>i</sup>

, respectively.

## **3. Motivation example and solution overview**

To help understand the significance of our problem, we first give a motivation example to show that only channel assignment and cooperative routing cannot achieve the good performance. Expired from the example, we give an overview on our solution.

**Figure 3** is an MRMC cooperative wireless network consisting of 14 nodes. The small solid dots in each node denote the radios. There are three orthogonal channels available, denoted by *Ch*<sup>1</sup> , *Ch*2, and *Ch*<sup>3</sup> . For simplicity, we assume that all the links are free of transmission error, and the raw capacity of each link can be calculated by Eq. (1) or (2) depending on the transmission mode. The communication range and interference range are set to 250 and 550 m, respectively. The network is connected under an initial channel assignment [26] to guarantee the connectivity of network to transmit any possible flows over multiple hops.

Initially, there are two flows with their routing paths *F*<sup>1</sup> (*<sup>A</sup>* <sup>→</sup> *<sup>K</sup>* ) <sup>=</sup> *<sup>A</sup> Ch*<sup>2</sup> ⎯→*<sup>F</sup> Ch*<sup>3</sup> ⎯→*<sup>K</sup>* and *F*<sup>2</sup> (*<sup>D</sup>* <sup>→</sup> *<sup>E</sup>* ) <sup>=</sup> *<sup>D</sup> Ch*<sup>1</sup> ⎯→*<sup>B</sup> Ch*<sup>3</sup> ⎯→*<sup>E</sup>*, as shown in **Figure 3a**. According to Eq. (1), the raw capacity of links in these two flows can be calculated directly: *C*DT(*A*,*F*) = 61.189 Mbps, *C*DT(*F*,*K*) = 65.9819 Mbps, *C*DT(*D*,*B*) = 73.1874 Mbps, *C*DT(*B*,*E*) = 78.8452 Mbps. On channel *Ch*<sup>3</sup> , there are two co-channel links that interfere with each other, link *l FK Ch*3 passed by flow *F*<sup>1</sup> and link *l BE Ch*3 passed by flow *F*<sup>2</sup> . A straightforward way to avoid interference is to apply TDMA to fairly allocate time slots to different flows. As a result, the available capacity of these two links becomes *C*DT ′ (*F*, *K*) = *C*DT(*F*, *K*) /2 = 32.9909 Mbps, *C*DT ′ (*B*, *E*) = *C*DT(*B*, *E*) /2 = 39.4226 Mbps. Assume that all flows transmit a packet at the peak link data rate through a rough calculation that neglects the overhead cost. Constrained by the bottleneck rate of the path, the end-to-end throughput of *Flow*<sup>1</sup> and *Flow*<sup>2</sup> are min{*CDT* (*A*, *F*),*CDT* ' (*F*, *K*)} = 32.9909 Mbps and min{*C*DT(*D*, *B*),*C*DT ' (*B*, *E*)} = 39.4226 Mbps. The aggregate network throughput of these two flows is 32.9909 + 39.4226 = 72.4135 Mbps.

**Figure 3.** Motivation example.

In **Figure 3b**, a new flow *F*<sup>3</sup> (*G*→*J*) arrives. To obtain higher cooperative diversity in the network, the potential route for *F*<sup>3</sup> is chosen as *F*<sup>3</sup> (*G*→*J*) <sup>=</sup> *<sup>G</sup> Ch*<sup>3</sup> ⎯→*<sup>I</sup>*(*<sup>H</sup>* ) *Ch*<sup>1</sup> ⎯→*<sup>J</sup>*, where *H* acts as the CR relay in the flow path. Similarly, the raw capacity of each hop links *l GI*(*H*) *Ch*<sup>3</sup> , and *l IJ Ch*1 can be calculated, with *C*DT(*I*, *J*) = 91.8365 Mbps, *C*CT(*G*, *H*, *I*) = 51.3242 Mbps. However, there are two co-channel links that interfere with each other on the channel *Ch*<sup>1</sup> , and three co-channel links on the channel *Ch*<sup>3</sup> , respectively. The available capacity of these links is calculated as follows: *C*DT ′ (*A*, *F*) = *C*DT(*A*, *F*) = 61.189 Mbps, *C*DT ′ (*F*, *K*) = *C*DT(*F*, *K*) /3 = 21.90 Mbps, *C*DT ′ (*D*, *B*) = *C*DT(*D*, *B*) /2 = 36.5937 Mbps, *C*DT ′ (*B*, *E*) = *C*DT(*B*, *E*) /3 = 26.2817 Mbps, *C*CT ′ (*G*, *H*, *I*) = *C*CT(*G*, *H*, *I*) /3 = 17.1 Mbps, *C*DT ′ (*I*, *J*) = *C*DT(*I*, *J*) /2 = 45.9182 Mbps. As a result, the end-toend throughput of *Flow*<sup>1</sup> , *Flow*2, and *Flow*<sup>3</sup> are min{61.189, 21.90} = 21.90 Mbps, min{36.5937, 26.2871} = 26.2817 Mbps and min{17.1, 45.9182} = 17.1 Mbps. The aggregate throughput of the whole network is 21.90 + 26.2817 + 17.1 = 65.2637 Mbps. Although there are three concurrent transmission flows, the aggregate throughput decreases about 10% compared with that in **Figure 3a**.

**3. Motivation example and solution overview**

Initially, there are two flows with their routing paths *F*<sup>1</sup>

nel links that interfere with each other, link *l*

′

by *Ch*<sup>1</sup>

70 Ad Hoc Networks

, *Ch*2, and *Ch*<sup>3</sup>

32.9909 Mbps, *C*DT

'

**Figure 3.** Motivation example.

(*A*, *F*),*CDT*

mance. Expired from the example, we give an overview on our solution.

the connectivity of network to transmit any possible flows over multiple hops.

Mbps, *C*DT(*D*,*B*) = 73.1874 Mbps, *C*DT(*B*,*E*) = 78.8452 Mbps. On channel *Ch*<sup>3</sup>

ent flows. As a result, the available capacity of these two links becomes *C*DT

by the bottleneck rate of the path, the end-to-end throughput of *Flow*<sup>1</sup>

network throughput of these two flows is 32.9909 + 39.4226 = 72.4135 Mbps.

(*F*, *K*)} = 32.9909 Mbps and min{*C*DT(*D*, *B*),*C*DT

To help understand the significance of our problem, we first give a motivation example to show that only channel assignment and cooperative routing cannot achieve the good perfor-

**Figure 3** is an MRMC cooperative wireless network consisting of 14 nodes. The small solid dots in each node denote the radios. There are three orthogonal channels available, denoted

and the raw capacity of each link can be calculated by Eq. (1) or (2) depending on the transmission mode. The communication range and interference range are set to 250 and 550 m, respectively. The network is connected under an initial channel assignment [26] to guarantee

(*<sup>D</sup>* <sup>→</sup> *<sup>E</sup>* ) <sup>=</sup> *<sup>D</sup> Ch*<sup>1</sup> ⎯→*<sup>B</sup> Ch*<sup>3</sup> ⎯→*<sup>E</sup>*, as shown in **Figure 3a**. According to Eq. (1), the raw capacity of links in these two flows can be calculated directly: *C*DT(*A*,*F*) = 61.189 Mbps, *C*DT(*F*,*K*) = 65.9819

straightforward way to avoid interference is to apply TDMA to fairly allocate time slots to differ-

the peak link data rate through a rough calculation that neglects the overhead cost. Constrained

*Ch*3 passed by flow *F*<sup>1</sup>

(*B*, *E*) = *C*DT(*B*, *E*) /2 = 39.4226 Mbps. Assume that all flows transmit a packet at

'

*FK*

. For simplicity, we assume that all the links are free of transmission error,

(*<sup>A</sup>* <sup>→</sup> *<sup>K</sup>* ) <sup>=</sup> *<sup>A</sup> Ch*<sup>2</sup> ⎯→*<sup>F</sup> Ch*<sup>3</sup> ⎯→*<sup>K</sup>* and *F*<sup>2</sup>

and link *l*

*BE*

′

and *Flow*<sup>2</sup>

(*B*, *E*)} = 39.4226 Mbps. The aggregate

, there are two co-chan-

(*F*, *K*) = *C*DT(*F*, *K*) /2 =

are min{*CDT*

. A

*Ch*3 passed by flow *F*<sup>2</sup>

With the above-selected routes, we apply channel assignment to improve the network performance. In **Figure 3c**, we change the channel of node *G*, *H*, and *I* from the overloaded channel *Ch*<sup>3</sup> to the least-used one *Ch*<sup>2</sup> , which reduces the number of interfering links on channel *Ch*<sup>3</sup> from three to two. The raw capacity of links on the paths of the three flows becomes *C*DT ′ (*A*, *F*) = *C*DT (*A*, *F*) /2= 30.5945 Mbps, *C*DT ′ (*F*, *K*) = *C*DT(*F*, *K*) /2 = 32.9909 Mbps, *C*DT ′ (*D*, *B*) = *C*DT(*D*, *B*) /2= 36.5937 Mbps, *C*DT ′ (*B*, *E*) = *C* DT(*B*, *E*) /2 = 39.4226 Mbps, *C*CT ′ (*G*, *H*, *I*) = *C*CT(*G*, *H*, *I*) /2 = 25.6621 Mbps, *C*DT ′ (*I*, *J*) = *C*DT(*I*, *J*) /2 = 45.9182 Mbps. As a result, the end-to-end throughput of *Flow*<sup>1</sup> , *Flow*<sup>2</sup> , and *Flow*<sup>3</sup> are 30.594, 36.5937, and 25.6621 Mbps, respectively. The aggregate throughput of the three flows in the network is 30.5945 + 36.5937 + 25.6621 = 92.8503 Mbps. The performance is improved almost 42% compared with that in **Figure 3b**.

Based on the above channel assignment, transmitter nodes would further check whether there exists a better relay node which can be utilized to obtain a better cooperative capacity gain. As shown in **Figure 3d**, node *G* can select node *L* instead of *H* as the relay node to further improve the performance. The available capacity of cooperative transmission can be calculated as *C*CT ′ (*G*, *L*, *I*) = *C*CT(*G*, *L*, *I*) /2 = 66.5462/2 = 33.2731 Mbps. As a result, the end-to-end throughput of *Flow*<sup>1</sup> , *Flow*2, and *Flow*<sup>3</sup> are 30.594, 36.5937, and 33.2731 Mbps. After relay adjustment, the aggregate throughput of the whole network is 30.5945 + 36.5937 + 33.2731 = 100.4613 Mbps, which is nearly 1.5 times that in **Figure 3b**.

Existing studies have demonstrated that the joint optimization problem of routing and channel assignment in multi-radio multi-channel wireless network is NP [27], the joint optimization problem of relay selection and cooperative routing is also NP [19]. Compared to the above problems where each considers only two issues, our problem considers three issues and can be generally proven to be NP-hard. To solve the problem, we propose a solution framework which is formed with three important components. In the following sections, we introduce the detailed algorithms for each part.

First, to obtain cooperative gain and reduce co-channel interference, every node periodically calculates the routing metric of CACU (contention-aware channel utilization). Based on this metric, when new flow arrives, an interference-aware cooperative routing algorithm is run to find the cooperative routing path and select MR and CR nodes along the path.

Second, to adapt to dynamic traffic changes, every node periodically measures the channel condition and calculates TACC metric. When a node's working channel is detected to be overloaded according to the TACC value, a dynamic channel-adjustment algorithm is triggered to switch the highly loaded channel to a lightly loaded one to relieve the co-channel interference.

Third, as channel adjustment changes the network topology, a local path segment and relay adjustment algorithm are followed by switching the flow traffic to a new path segment locally according to the new topology.

## **4. Cooperative route**

To quantify the available capacity of a link, we first introduce a new routing metric. Based on the metric, we propose an interference-aware cooperative routing algorithm to better exploit the benefit of cooperative diversity.

There are several existing routing metrics proposed for multi-hop wireless networks. Hop count is a basic routing metric widely used. To further consider the wireless channel condition and interference, several improved metrics are also proposed, including ETX, WCETT, MIC, CCM [27], and MIPC [28]. The above routing metrics target for one-to-one direct transmissions between two nodes in conventional wireless networks. In cooperative wireless networks, the routing metric should consider multiple-to-one cooperative transmissions. As a cooperative transmission involves three links, it may cause more interference in the network, thus reducing the transmission performance. To facilitate the finding of more efficient cooperative routing path for higher throughput, the routing metric should concurrently consider the transmission mode selection and interference impact. To characterize radio transmissions in the presence of interference and identify the co-channel interference links of a given link, two receiver-driven interference models are proposed in the literature, the physical model [29] and the protocol model [30].

Our design does not depend on a specific model used. For the convenience of presentation and design, we simply apply a protocol model to illustrate our algorithms in this chapter. We consider link *l AB Chi* to be the co-channel interference link of another link *l CD Chi* if *A* and *B* work on the same channel of *C* and *D*, and at least one of node pairs (*A*,*C*), (*A*,*D*), (*B*,*C*), and (*B*,*D*) is within the interference range. A cooperative transmission may involve three nodes. We consider cooperative link *l AB*(*R*) *Chi* to interfere with another link *l CD Chi* if *A*, *B*, and *R* work on the same channel of *C* and *D*, and at least one of node pairs (*A*,*C*), (*A*,*D*), (*B*,*C*), (*B*,*D*), (*R*,*C*), and (*R*,*D*) is within the interference range.

If a node *x* transmits data to a node *y* through the direct transmission, the available capacity *CDTa* (*x*, *y*, *Chi* ) of the direct transmission link *l xy Chi* is equal to the link capacity *CDT*(*x*, *y*) (calculated using Eq. (1)) deducted by the traffic load of its co-channel interference links:

$$\mathbf{C}\_{\text{DT}\_s}(\mathbf{x}, \mathbf{y}, \mathbf{C}h) = \mathbf{C}\_{\text{Dry}}(\mathbf{x}, \mathbf{y}) - \sum\_{|\text{eff}\_\alpha(\mathbf{0})|} t(\mathbf{j}), \tag{4}$$

where *t*(*j*) is the traffic load on link *j*, *I Chi* (*l*) is the set of co-channel interference links of *l xy Chi* . Similarly, if node *x* transmits data to node *y* with the help of relay *z*, the available capacity *C*CT*<sup>a</sup>* (*x*, *z*, *y*, *Chi* ) of a cooperative transmission link *l xy*(*z*) *Chi* can be calculated as

metric, when new flow arrives, an interference-aware cooperative routing algorithm is run to

Second, to adapt to dynamic traffic changes, every node periodically measures the channel condition and calculates TACC metric. When a node's working channel is detected to be overloaded according to the TACC value, a dynamic channel-adjustment algorithm is triggered to switch the highly loaded channel to a lightly loaded one to relieve the co-channel interference. Third, as channel adjustment changes the network topology, a local path segment and relay adjustment algorithm are followed by switching the flow traffic to a new path segment locally

To quantify the available capacity of a link, we first introduce a new routing metric. Based on the metric, we propose an interference-aware cooperative routing algorithm to better exploit

There are several existing routing metrics proposed for multi-hop wireless networks. Hop count is a basic routing metric widely used. To further consider the wireless channel condition and interference, several improved metrics are also proposed, including ETX, WCETT, MIC, CCM [27], and MIPC [28]. The above routing metrics target for one-to-one direct transmissions between two nodes in conventional wireless networks. In cooperative wireless networks, the routing metric should consider multiple-to-one cooperative transmissions. As a cooperative transmission involves three links, it may cause more interference in the network, thus reducing the transmission performance. To facilitate the finding of more efficient cooperative routing path for higher throughput, the routing metric should concurrently consider the transmission mode selection and interference impact. To characterize radio transmissions in the presence of interference and identify the co-channel interference links of a given link, two receiver-driven interference models

are proposed in the literature, the physical model [29] and the protocol model [30].

*Chi* to interfere with another link *l*

using Eq. (1)) deducted by the traffic load of its co-channel interference links:

(*x*, *y*, *Chi*

Our design does not depend on a specific model used. For the convenience of presentation and design, we simply apply a protocol model to illustrate our algorithms in this chapter. We

the same channel of *C* and *D*, and at least one of node pairs (*A*,*C*), (*A*,*D*), (*B*,*C*), and (*B*,*D*) is within the interference range. A cooperative transmission may involve three nodes. We con-

channel of *C* and *D*, and at least one of node pairs (*A*,*C*), (*A*,*D*), (*B*,*C*), (*B*,*D*), (*R*,*C*), and (*R*,*D*)

If a node *x* transmits data to a node *y* through the direct transmission, the available capacity

) = *C*DT(*x*, *y*) − ∑

*j*∈*I Chi* (*l*)

*xy Chi* *CD Chi* *CD Chi*

is equal to the link capacity *CDT*(*x*, *y*) (calculated

if *A*, *B*, and *R* work on the same

*t*(*j*), (4)

if *A* and *B* work on

to be the co-channel interference link of another link *l*

find the cooperative routing path and select MR and CR nodes along the path.

according to the new topology.

the benefit of cooperative diversity.

**4. Cooperative route**

72 Ad Hoc Networks

consider link *l*

(*x*, *y*, *Chi*

*CDTa*

*AB Chi*

is within the interference range.

*C*DT*<sup>a</sup>*

*AB*(*R*)

) of the direct transmission link *l*

sider cooperative link *l*

$$\mathbf{C}\_{\text{cr}\_{\text{\textdegree}}}(\mathbf{x}, \mathbf{z}, y, \text{Ch}\_{\text{\textdegree}}) = \mathbf{C}\_{\text{cr}\uparrow}(\mathbf{x}, \mathbf{z}, y) - \sum\_{\text{\textdegree}\mathbf{d}\_{\text{\textdegree}}(\mathbf{0})} t(\mathbf{j}), \tag{5}$$

where *C*CT(*x*, *z*, *y*) is the capacity of the link *l xy*(*z*) *Chi* (calculated using Eq. (2)), *I Chi* (*l*) is the set of co-channel interference links, and *t*(*j*) denotes the traffic load on link *j*.

Therefore, the available capacity of a link (*x*, *y*, *Ch*<sup>i</sup> ) can be defined as the maximum available capacity among all possible transmission modes

$$\begin{aligned} \text{if all possible transmission modes} \\\\ \text{CACU}(\mathbf{x}, \mathbf{y}, \mathbf{C}h\_i) &= \max \left\{ \mathbb{C}\_{\text{DT}\_\circ}(\mathbf{x}, \mathbf{y}, \mathbf{C}h\_i), \max\_{\mathbf{z}} \left\{ \mathbb{C}\_{\text{CT}\_\circ}(\mathbf{x}, \mathbf{z}, \mathbf{y}, \mathbf{C}h\_i) \right\} \\ &: \mathbf{z} \in N\_{\text{cl}}(\mathbf{x}) \text{ and } N\_{\text{cl}}(\mathbf{y}) \right\}, \end{aligned} \tag{6}$$

where *NChi* (*x*) and *NChi* (*y*) denote the set of neighbors of nodes *x* and *y* on the channel *Ch*<sup>i</sup> . Obviously, *CACU*(*x*, *y*, *Chi* ) = *C*DT*<sup>a</sup>* (*x*, *y*, *Chi* ) if the available capacity of direct transmission is larger than the available capacity of the cooperative transmission, otherwise, *CACU*(*x*, *y*, *Chi* ) = max *z*{*C*CT*<sup>a</sup>* (*x*, *z*, *y*, *Chi* ) : *z* ∈ *NChi* (*x*) and *NChi* (*y*)}. Multiple radios on a node are generally assigned with orthogonal channels. A pair of nodes *x* and *y* may have multiple channels to be the same. Based on Eq. (6), the routing metric of link (*x*, *y*) is defined as the maximum available capacity among all common channels as follows:

$$\text{CACLI}(\mathbf{x}, y) = \max\_{\text{Cl}: \pi \omega(\mathbf{y}) = \omega(\mathbf{y})} \text{CACLI}(\mathbf{x}, y, \mathbf{C}h\_{\parallel}) \tag{7}$$

where *w*(*x*) and *w*(*y*) denote the working channel set of nodes *x* and *y*, respectively. CACU metric in Eq. (7) captures the interference cost from both direct transmission and cooperative transmission. Therefore, CACU can be applied to facilitate finding a transmission path with lower interference thus higher capacity. Based on the metric, a node *x* can decide that it will take direct transmission or cooperative transmission, and determine the channel to use for transmission. In the case that a cooperative transmission is needed, the selected relay node will be informed.

In this chapter, we modified ad hoc on-demand distance vector (AODV) routing to implement our distributed interference-aware cooperative routing algorithm to establish the maximum capacity path while considering the flow routing and relay selection, as shown in Algorithm 1. The derived CACU metric is applied to construct the cooperative path. When a source has data to transmit but does not have a path to the destination, it broadcasts a route request (RREQ) for that destination. When an intermediate node receives RREQ, if it is the destination or has a current route to the destination, it generates a route reply (RREP). Otherwise, the node needs to rebroadcast the RREQ with a set of parameters inserted: the CACU metric for each of its outgoing link is calculated based on Eq. (7), and the maximum capacity from the source to itself is calculated based on Algorithm 1 in **Figure 4**.


## **Figure 4.** Algorithm 1: interference-aware cooperative routing.

## **5. Channel adjustment**

As shown in the motivation example of Section 3, the channel adjustment can reduce cochannel interference and thus increase the aggregate throughput. The main function of channel adjustment is to switch one node's working channel from an overloaded one to a lightly loaded one to obtain better throughput. For practical implementation of the channel adjustment in a cooperative wireless network, we need to answer two basic questions: (1) Which channel to switch to? (2) How to keep the network stable and well connected during the channel adjustment?

Before presenting the detailed channel-adjustment algorithm, we first introduce a trafficaware channel condition metric (TACC) to evaluate the channel load condition. The TACC of node *i* on channel *Ch*m is defined as the channel utilization calculated as the summation of the co-channel traffic load within this node's two hops:

$$TAC\ C\_i(Ch\_w) = \sum\_{\forall \pi \in \mathcal{V}\_{c\_i}(0)} \left( t(l\_{i\bar{l}}^{Ch\_r}) + \sum\_{l \le \mathcal{V}\_{c\_i}(0)} t(l\_{j\bar{k}}^{Ch\_r}) \right), \tag{8}$$

where *NChm* (*i* )is node *i*'s neighbor set on channel *Ch*m, *t*( *l ij Chm* ) denotes the traffic load on link *l ij Chm*, while *j* and *k* represent the one-hop and two-hop neighbors of node *i*, respectively. The average traffic conditions may be obtained by attaching the information with periodical topology maintenance messages such as Hello over two hops.

When a node finds that the TACC of a working channel exceeds a threshold *θ*<sup>1</sup> , that is, *TAC Ci* (*C hm*) ≥ *θ*<sup>1</sup> , it will trigger a channel-adjustment process. The node needs to identify a set of feasible candidate channels and selects the best one to switch to. To improve the network throughput with channel switching, the condition of the candidate channel *Ch*<sup>b</sup> should be better than the condition of the current channel *Ch*<sup>a</sup> . However, the channel adjustment may lead channel *Ch*<sup>b</sup> to be overloaded and result in potential network instability. To avoid this problem, the following two conditions should be satisfied:

$$\text{TAC C}\_{\text{>}}(\text{Ch}\_{b}) \text{ + Tlo a } d\_{\text{>}}(\text{Ch}\_{o}) \le |\mathcal{O}\_{2^{\circ}}| \tag{9}$$

$$\text{TAC C}\_{\text{>}}\text{(Ch}\_{\text{b}}) \text{+Tloa } d\_{\text{l}}\text{(Ch}\_{\text{a}'} \le \Theta\_{\text{z}'} \tag{10}$$

where node *j* is within two hops of node *i*, *Tload*<sup>i</sup> *(Ch*<sup>a</sup> *)* is the total traffic load of all links on the original channel *Ch*<sup>a</sup> , expressed as

$$\text{Tloa } d\_{\parallel}(\text{Ch}\_{\omega}) = \sum\_{\neq \in \mathcal{N}\_{\text{cl}}(\emptyset)} t(l\_{\parallel}^{\text{Ch}}) . \tag{11}$$

We set *θ*<sup>2</sup> in Eq. (9) to *θ*<sup>2</sup> = 0.9 \* *θ*<sup>1</sup> so that the TACC of the new channel after the channel switching is less than 90% of TACC trigger threshold to avoid another channel switching and maintain the network stability.

**5. Channel adjustment**

74 Ad Hoc Networks

channel adjustment?

where *NChm*

(*C hm*) ≥ *θ*<sup>1</sup>

*Ci*

co-channel traffic load within this node's two hops:

**Figure 4.** Algorithm 1: interference-aware cooperative routing.

maintenance messages such as Hello over two hops.

*TAC Ci*

As shown in the motivation example of Section 3, the channel adjustment can reduce cochannel interference and thus increase the aggregate throughput. The main function of channel adjustment is to switch one node's working channel from an overloaded one to a lightly loaded one to obtain better throughput. For practical implementation of the channel adjustment in a cooperative wireless network, we need to answer two basic questions: (1) Which channel to switch to? (2) How to keep the network stable and well connected during the

Before presenting the detailed channel-adjustment algorithm, we first introduce a trafficaware channel condition metric (TACC) to evaluate the channel load condition. The TACC of node *i* on channel *Ch*m is defined as the channel utilization calculated as the summation of the

while *j* and *k* represent the one-hop and two-hop neighbors of node *i*, respectively. The average traffic conditions may be obtained by attaching the information with periodical topology

of feasible candidate channels and selects the best one to switch to. To improve the network

*ij*

, it will trigger a channel-adjustment process. The node needs to identify a set

), (8)

*ij Chm*,

, that is, *TAC*

should be

*Chm* ) denotes the traffic load on link *l*

(*Chm*) = ∑ *j*∈*NChm* (*i*)(*t*( *<sup>l</sup> ij Chm*) + ∑ *k*∈*NChm* (*j*) *t*( *l jk Chm*)

When a node finds that the TACC of a working channel exceeds a threshold *θ*<sup>1</sup>

throughput with channel switching, the condition of the candidate channel *Ch*<sup>b</sup>

(*i* )is node *i*'s neighbor set on channel *Ch*m, *t*( *l*

**Figure 5** shows an example of channel-adjustment procedure. There are four flows in the network, *F*<sup>1</sup> (*A*→*I*), *F*<sup>2</sup> *(A*→*L*), *F*<sup>3</sup> *(A*→*D*), and *F*<sup>4</sup> *(B*→*J*). When node *A* finds that *Ch*<sup>2</sup> is overloaded, it tries to find another channel to switch to. If node *A* uses *Ch*<sup>1</sup> as the new channel, then nodes *K*, *B*, and *F* need to switch to *Ch*<sup>1</sup> to get connected with node *A*. This will change the traffic load of *F*<sup>1</sup> , *F*2, and *F*<sup>3</sup> on original links *l AK Ch* 2, *l AB Ch* 2, and *l AF Ch* 2 to the links *l AK Ch* 1, *l AB Ch* 1, and *l AF Ch* 1 on the new channel *Ch*<sup>1</sup> . However, this will make *Ch*<sup>1</sup> overloaded and trigger another channel adjustment, which makes the network unstable. Therefore, *Ch*<sup>1</sup> is not the feasible candidate channel for node *A* because it does not satisfy condition in Eq. (9). Instead, according to Eq. (9), node *A* finds that *Ch*<sup>3</sup> is the candidate channel and the traffic load is switched from *Ch*<sup>2</sup> to *Ch*<sup>3</sup> as shown in **Figure 5c**.

Besides considering the condition in Eq. (9) to avoid network instability, to justify the extra channel switching overhead, the gains in terms of TACC should be larger than a given threshold *θ*<sup>3</sup> :

Hær swwhuung overneau, une ganus n'eulis ou 1:AC. Cusouau bez nager unan a gweril umsemu  $\upsilon\_{\mathcal{I}}$ .

$$\frac{\text{Befre\\_TAC.C(Ch.)}{\text{Afer\\_TAC.C(Ch.)}} \ge \mathcal{O}\_{\mathcal{Y}}\tag{12}$$

where *Before\_TACC*<sup>i</sup> *(Ch*<sup>a</sup> *)* is the original TACC value of the working channel *Ch*<sup>a</sup> before channel switching, while *After\_TACC*<sup>i</sup> *(Ch*<sup>a</sup> *) = TACC*<sup>i</sup> *(Ch*<sup>b</sup> *) + Tload*<sup>i</sup> *(Ch*<sup>a</sup> *)* is the TACC value of the working channel *Ch*<sup>b</sup> after the channel switching.

Obviously, to obtain a positive benefit of channel switching, *θ* <sup>3</sup> in Eq. (12) should be larger than 1. Moreover, channel adjustment may involve a significant switching overhead such as switching delay and traffic interruption, and frequent channel switching will result in oscillation and severely impact the network performance. Therefore, *θ* <sup>3</sup> should be set by well considering the tradeoff between the switching overhead and the benefit of channel switching. According to Ref. [31], in our simulation, *θ* <sup>3</sup> is set to 1.2. That is, after the channel adjustment, the TACC of the new channel should be at least 20% less than that of the original one. Only

**Figure 5.** An example of channel adjustment.

**Figure 6.** Chain-puzzle problem.

when conditions of Eqs. (9), (10), and (12) are satisfied, the node can switch its working channel from *Ch*<sup>a</sup> to *Ch*<sup>b</sup> . To preserve the current flows' connectivity and stability, connectivity checking should be applied to further identify the feasible channel, which is discussed in the next subsection.

In an MRMC cooperative network, two nodes may have more than one pair of radios connected. If nodes *i* and *j* have two pairs of radios directly connected with each other, the channel adjustment does not impact their connectivity. If nodes *i* and *j* have only one radio connected with each other over a channel, the channel adjustment may interrupt the transmission of an active flow carried over the link (*i, j*). To maintain the connectivity of the active flow, the channel switching may be carried by a chain of nodes, with each node on the chain having only a single common channel. We define this problem as *chain puzzle*.

Cooperative transmission may be more prone to the chain-puzzle problem. **Figure 6** gives two examples to illustrate the chain-puzzle problem under direct transmission and cooperative transmission, respectively. In **Figure 6a**, assume that flow *F*<sup>1</sup> (*M*→*A*) transmits over the link between nodes *M* and *F* using channel *Ch*<sup>2</sup> , if the channel needs to be switched to *Ch*<sup>3</sup> , node *F*'s single-channel neighbor *G* must switch to *Ch*<sup>3</sup> . Similarly, after node *G* switches its channel, node *G*'s single-channel neighbor *B* has to switch to *Ch*<sup>3</sup> too, which will also lead channel switching from node *A* and thus result in the chain-puzzle problem. In **Figure 6b**, *D, E*, and *J* work together for cooperative transmission. Assume that nodes *C* and *D* currently transmitting over channel *Ch*<sup>3</sup> want to switch to *Ch*<sup>1</sup> , nodes *E* and *J* are *D*'s single-channel neighbors. To maintain the flow's connectivity, nodes *E, J* should switch channel from *Ch*<sup>3</sup> to *Ch*<sup>1</sup> . As a result, *K*, *R*, *Q*, and *P* also need to switch their working channel from *Ch*<sup>3</sup> to *Ch*<sup>1</sup> . Chain puzzle again happens.




when conditions of Eqs. (9), (10), and (12) are satisfied, the node can switch its working chan-

checking should be applied to further identify the feasible channel, which is discussed in the

In an MRMC cooperative network, two nodes may have more than one pair of radios connected. If nodes *i* and *j* have two pairs of radios directly connected with each other, the channel adjustment does not impact their connectivity. If nodes *i* and *j* have only one radio connected with each other over a channel, the channel adjustment may interrupt the transmission of an active flow carried over the link (*i, j*). To maintain the connectivity of the active flow, the channel switching may be carried by a chain of nodes, with each node on the chain having only a

Cooperative transmission may be more prone to the chain-puzzle problem. **Figure 6** gives two examples to illustrate the chain-puzzle problem under direct transmission and coopera-

ing from node *A* and thus result in the chain-puzzle problem. In **Figure 6b**, *D, E*, and *J* work together for cooperative transmission. Assume that nodes *C* and *D* currently transmitting over

single common channel. We define this problem as *chain puzzle*.

tive transmission, respectively. In **Figure 6a**, assume that flow *F*<sup>1</sup>

the flow's connectivity, nodes *E, J* should switch channel from *Ch*<sup>3</sup>

and *P* also need to switch their working channel from *Ch*<sup>3</sup>

between nodes *M* and *F* using channel *Ch*<sup>2</sup>

single-channel neighbor *G* must switch to *Ch*<sup>3</sup>

want to switch to *Ch*<sup>1</sup>

*G*'s single-channel neighbor *B* has to switch to *Ch*<sup>3</sup>

. To preserve the current flows' connectivity and stability, connectivity

, if the channel needs to be switched to *Ch*<sup>3</sup>

, nodes *E* and *J* are *D*'s single-channel neighbors. To maintain

to *Ch*<sup>1</sup>

. Similarly, after node *G* switches its channel, node

to *Ch*<sup>1</sup>

too, which will also lead channel switch-

(*M*→*A*) transmits over the link

. Chain puzzle again happens.

. As a result, *K*, *R*, *Q*,

, node *F*'s

nel from *Ch*<sup>a</sup>

76 Ad Hoc Networks

channel *Ch*<sup>3</sup>

next subsection.

to *Ch*<sup>b</sup>

**Figure 5.** An example of channel adjustment.

**Figure 6.** Chain-puzzle problem.

**Figure 7.** Algorithm 2: the chain-puzzle-checking algorithm and Algorithm 3: feasible channel selection algorithm.

Chain-puzzle checking becomes an important issue in channel-adjustment procedure because chain puzzle may cause a number of practical problems. First, a large number of nodes may be involved in a channel switch when chain puzzle happens, which could result in a high overhead. Second, it is difficult to synchronize the switching action among all nodes involved because the signaling used for negotiation needs to propagate through many hops. In the worst case, this may result in flow transmission interruption. Therefore, before a node switches its working channel, chain puzzle should be checked to identify feasible candidate channel to avoid switching channel sequentially, and maintain the network's stability. To facilitate chain-puzzle checking, we propose two different connectivity rules according to different transmission modes:

Connectivity Rule 1: if any node pair of a direct transmission link passed by an active flow is originally connected, the node pair should remain connected after the channel switching.

Connectivity Rule 2: if any three nodes in an active cooperative transmission are originally connected, they should remain directly connected under a new channel.

Based on the above two connectivity rules, we propose a local two-hop chain-puzzlechecking sub-algorithm, as shown in Algorithm 2 of **Figure 7**. When node *A* checks a new channel *Ch*<sup>i</sup> and finds that (through the communication with its neighbor which needs to switch channel with it) there exists a node which is not within its two-hop distance but also needs to switch channel to preserve the connectivity of the current flows, the chain puzzle may happen and *Ch*<sup>i</sup> is not a feasible channel for channel adjustment. The results of chain-puzzle-checking algorithm depend on the topology of the network. Based on the chain-puzzle-checking algorithm, the feasible channel selection algorithm can be presented as in Algorithm 3 of **Figure 7**.

**Figure 8.** Local path adaptation and relay adjustment.

**Figure 9.** Algorithm 4: local path adaptation and relay adjustment algorithm.

## **6. Local routing and relay adjustment**

After channel adjustment, the network topology may change. To make flow transmissions continuous under the new topology, some flows may need to adapt their path segments, channels, or relays locally.

As shown in **Figure 8a**, an active flow *F*<sup>1</sup> (*D*→*O*) exists in the network with its original route *F*1 (*D*→*O*) <sup>=</sup> *<sup>D</sup> Ch*<sup>1</sup> ⎯→*<sup>E</sup>*(*J*) *Ch*<sup>1</sup> ⎯→*<sup>L</sup> Ch*<sup>1</sup> ⎯→*<sup>O</sup>*. After nodes *D*, *E*, and *J* adjust their working channel from *Ch*<sup>1</sup> to *Ch*<sup>3</sup> , nodes *E* and *L* are not directly connected. Thus, flow *F*<sup>1</sup> should switch its path locally from *<sup>E</sup> Ch*<sup>1</sup> ⎯→*<sup>L</sup>* to *<sup>E</sup> Ch*<sup>3</sup> ⎯→*<sup>Q</sup> Ch*<sup>1</sup> ⎯→*<sup>L</sup>*. Moreover, after *D* and *E* switch their working channel from *Ch*<sup>1</sup> to *Ch*<sup>3</sup> , besides node *J*, node *N* may have the opportunity to support cooperative transmission and provide a higher cooperative gain. As a result, node *D* would select node *N* as the relay node instead of *J*. To make the network stable, the path adaptation and relay adjustment should be performed locally and the corresponding traffic flows should also switch to new path segments. Based on the cooperative routing algorithm in Algorithm 1, we design a local path adaptation and relay adjustment algorithm as shown in Algorithm 4 of **Figure 9**.

## **7. Completed solution**

The complete algorithm of joint cooperative routing, channel adjustment, and relay selection is shown in Algorithm 5 of **Figure 10**. To handle the dynamic wireless environment, nodes in the network execute the algorithm locally as follows.

When a new flow arrives, the interference-aware cooperative routing algorithm is applied to find the cooperative routing path with the maximum end-to-end available capacity and with the MR and CR relays selected along the path. Every node periodically evaluates the traffic conditions of a channel by calculating the TCAA metric according to Eq. (8) and checking whether its working channel is overloaded. If so, the node first applies Algorithm 3 to identify


**Figure 10.** Algorithm 5: complete algorithm of joint cooperative routing, channel assignment, and relay selection.

the feasible channel to switch to. Then, the channel adjustment will be triggered, which is followed by the local path adaptation and relay adjustment through Algorithm 4 for uninterrupted transmissions and better performance.

If each node independently makes a local channel-adjustment decision, multiple channeladjustment requests may be received simultaneously by a node, which either leads to request message collisions or inconsistent requests (if all messages are successfully received). To reduce the chance of simultaneous transmissions of channel-adjustment messages, we design a channel-adjustment timer which introduces a random delay before the message sending according to the channel load, and the timer can be set as follows:

$$T\_{\rm c\Omega\_\*} = \frac{1}{\text{TAC C}\_{\rm{\gamma}}(\text{Ch}\_n)}.\tag{13}$$

From Eq. (13), obviously, the node with a higher channel load, that is, a larger TACC value, has a lower average timer value, thus an earlier chance of adjusting its overloaded channel. When the channel load is high, the data transmissions should be switched from an overloaded channel to a light-loaded channel. In Algorithm 5, the channel load is measured with the metric TACC and updated when the TACC timer goes off. A smaller TACC timer would allow for more frequent update of the TACC value at high measurement cost, while a larger TACC timer for smaller measurement cost would make the TACC metric less accurate. In this chapter, we set the TACC timer adaptively according to the traffic pattern in the network taking into account the tradeoff between the accuracy of TACC metric and the measurement cost. If the traffic load is high, the TACC timer reduces but remains above a minimum timeout value *T*min. If the traffic load is low, the TACC timer increases but not beyond a maximum timeout value *T*max. In this chapter, we set *T*min = 20 ms and *T*max = 300 ms according to the channel-monitoring duration mentioned in [32].

## **8. Simulation**

**6. Local routing and relay adjustment**

**Figure 9.** Algorithm 4: local path adaptation and relay adjustment algorithm.

**Figure 8.** Local path adaptation and relay adjustment.

As shown in **Figure 8a**, an active flow *F*<sup>1</sup>

channels, or relays locally.

to *Ch*<sup>3</sup>

to *Ch*<sup>3</sup>

**7. Completed solution**

the network execute the algorithm locally as follows.

*F*1

from *Ch*<sup>1</sup>

78 Ad Hoc Networks

nel from *Ch*<sup>1</sup>

After channel adjustment, the network topology may change. To make flow transmissions continuous under the new topology, some flows may need to adapt their path segments,

(*D*→*O*) <sup>=</sup> *<sup>D</sup> Ch*<sup>1</sup> ⎯→*<sup>E</sup>*(*J*) *Ch*<sup>1</sup> ⎯→*<sup>L</sup> Ch*<sup>1</sup> ⎯→*<sup>O</sup>*. After nodes *D*, *E*, and *J* adjust their working channel

locally from *<sup>E</sup> Ch*<sup>1</sup> ⎯→*<sup>L</sup>* to *<sup>E</sup> Ch*<sup>3</sup> ⎯→*<sup>Q</sup> Ch*<sup>1</sup> ⎯→*<sup>L</sup>*. Moreover, after *D* and *E* switch their working chan-

transmission and provide a higher cooperative gain. As a result, node *D* would select node *N* as the relay node instead of *J*. To make the network stable, the path adaptation and relay adjustment should be performed locally and the corresponding traffic flows should also switch to new path segments. Based on the cooperative routing algorithm in Algorithm 1, we design a local path adaptation and relay adjustment algorithm as shown in Algorithm 4 of **Figure 9**.

The complete algorithm of joint cooperative routing, channel adjustment, and relay selection is shown in Algorithm 5 of **Figure 10**. To handle the dynamic wireless environment, nodes in

When a new flow arrives, the interference-aware cooperative routing algorithm is applied to find the cooperative routing path with the maximum end-to-end available capacity and with the MR and CR relays selected along the path. Every node periodically evaluates the traffic conditions of a channel by calculating the TCAA metric according to Eq. (8) and checking whether its working channel is overloaded. If so, the node first applies Algorithm 3 to identify

, besides node *J*, node *N* may have the opportunity to support cooperative

, nodes *E* and *L* are not directly connected. Thus, flow *F*<sup>1</sup>

(*D*→*O*) exists in the network with its original route

should switch its path

In the simulation, unless otherwise specified, the simulation setting is as follows. Thirty nodes are generated one by one in random locations in a 1000 × 1000 m area. Each new node is ensured to get connected with existing nodes in the network, and the initial channel assignment is done according to [26] to guarantee the connectivity of network to transmit possible flows over multiple hops. A node is equipped with two radio interfaces, and has the maximum transmission range set to 250 m. There are a total of 11 orthogonal channels, and the default number of flows in the network is set as *M* = 5.

Although ns-3 and Omnet++ are widespread simulator, cooperative communication is a physical layer technique, and can be very hard to simulate in ns-3 and Omnet++ if it is not completely impossible. Following the simulation setup in Refs. [17, 19], we evaluate the performance of our proposed algorithm through extensive simulations using MATLAB. Specifically, following the parameter setting in Ref. [19], we set the bandwidth of each channel to *W =* 22 MHz, the maximum transmission power at every node to 1 W. For simplicity, we assume that *h*m,n only considers the distance between nodes *m* and *n* and is given by <sup>|</sup>*hm*,*<sup>n</sup>*<sup>|</sup> <sup>=</sup> ||*m*, *<sup>n</sup>*||<sup>−</sup><sup>4</sup> , where ||*m*, *n*|| is the distance (in m) between *m* and *n* and the path loss index is 4. We assume the variance of noise is 10−10 W at all nodes. TACC trigger threshold *θ*<sup>1</sup> is set to 200. We set *θ*<sup>2</sup> to *θ*<sup>2</sup> = 0.9 \* *θ*<sup>1</sup> = 180 to help maintain network stability, and *θ*<sup>3</sup> to 1.2 to balance the switching overhead and benefit of channel switching.

We evaluate the effectiveness of our algorithms by comparing the results from six different implementation schemes. We implement two different cooperative transmission (CT) schemes. The first is our proposed Algorithm 5, denoted as CT\_adjustment. In the second scheme, we apply the cooperative routing algorithm in Algorithm 1 to find the cooperative path, without applying channel adjustment or local path adaptation and relay adjustment, which is denoted as CT\_No-adjustment. We also implement four additional schemes based on direct transmission (DT). The first DT scheme is denoted as DT\_No-adjustment, where we use the available capacity calculated in Eq. (7) as the routing metric and apply Algorithm 1 to find the path with the maximum available capacity for each flow. In the second scheme, denoted as DT\_ adjustment, channel and local path segment adjustment in Sections 5 and 6 are applied periodically to obtain better performance. The third and fourth schemes take two different routing metrics proposed in the literature to find the routing path: a HOP scheme, which uses hop count as routing metric and finds the shortest path, and an ETT scheme [27], which captures the packet transmission time in a time unit.

Two metrics are used to evaluate the performance. One is aggregate throughput which is the aggregative throughput of all flows. The other is minimum throughput, which is the minimum of all flows' throughput. Various factors affect the performance. We perform two set of simulations to analyze the effect of node density and number of orthogonal channel. As follows, we

**Figure 11.** Throughput results under network with different node density.

**Figure 12.** The impact of number of orthogonal channels.

show the simulation results, respectively. With the same topology and flows, we run the six schemes orderly to obtain the two metrics.

## **8.1. Impact of node density**

maximum transmission range set to 250 m. There are a total of 11 orthogonal channels, and

Although ns-3 and Omnet++ are widespread simulator, cooperative communication is a physical layer technique, and can be very hard to simulate in ns-3 and Omnet++ if it is not completely impossible. Following the simulation setup in Refs. [17, 19], we evaluate the performance of our proposed algorithm through extensive simulations using MATLAB. Specifically, following the parameter setting in Ref. [19], we set the bandwidth of each channel to *W =* 22 MHz, the maximum transmission power at every node to 1 W. For simplicity, we assume that *h*m,n only considers the distance between nodes *m* and *n* and is given by

index is 4. We assume the variance of noise is 10−10 W at all nodes. TACC trigger threshold *θ*<sup>1</sup>

We evaluate the effectiveness of our algorithms by comparing the results from six different implementation schemes. We implement two different cooperative transmission (CT) schemes. The first is our proposed Algorithm 5, denoted as CT\_adjustment. In the second scheme, we apply the cooperative routing algorithm in Algorithm 1 to find the cooperative path, without applying channel adjustment or local path adaptation and relay adjustment, which is denoted as CT\_No-adjustment. We also implement four additional schemes based on direct transmission (DT). The first DT scheme is denoted as DT\_No-adjustment, where we use the available capacity calculated in Eq. (7) as the routing metric and apply Algorithm 1 to find the path with the maximum available capacity for each flow. In the second scheme, denoted as DT\_ adjustment, channel and local path segment adjustment in Sections 5 and 6 are applied periodically to obtain better performance. The third and fourth schemes take two different routing metrics proposed in the literature to find the routing path: a HOP scheme, which uses hop count as routing metric and finds the shortest path, and an ETT scheme [27],

Two metrics are used to evaluate the performance. One is aggregate throughput which is the aggregative throughput of all flows. The other is minimum throughput, which is the minimum of all flows' throughput. Various factors affect the performance. We perform two set of simulations to analyze the effect of node density and number of orthogonal channel. As follows, we

, where ||*m*, *n*|| is the distance (in m) between *m* and *n* and the path loss

to *θ*<sup>2</sup> = 0.9 \* *θ*<sup>1</sup> = 180 to help maintain network stability, and *θ*<sup>3</sup>

to 1.2

the default number of flows in the network is set as *M* = 5.

to balance the switching overhead and benefit of channel switching.

which captures the packet transmission time in a time unit.

**Figure 11.** Throughput results under network with different node density.

<sup>|</sup>*hm*,*<sup>n</sup>*<sup>|</sup> <sup>=</sup> ||*m*, *<sup>n</sup>*||<sup>−</sup><sup>4</sup>

80 Ad Hoc Networks

is set to 200. We set *θ*<sup>2</sup>

To investigate how the node density impacts the network performance, we vary the number of nodes *D* from 30 to 120. When the number of nodes increases, the set of candidate relay nodes for cooperative relay (CR) and multi-hop relay (MR) becomes larger. Therefore, the aggregate rate and minimum rate under all routing schemes increase, as shown in **Figure 11**.

Among all the routing schemes, the aggregate throughput and the minimum throughput increase the fastest under our CT\_adjustment. With well-designed algorithms, our CT\_adjustment can more effectively exploit the resources of relay nodes and multiple channels to achieve high cooperative gain when the number of nodes is large. Our CT\_adjustment has the largest aggregate throughput and minimum throughput. At the node density 120, the aggregate throughput of our CT\_adjustment is 382, 270, 276, 127, 36, and 112% higher than those of HOP, ETT, DT\_No-adjustment, DT-adjustment, CT\_No-adjustment, and CT-adjustment, respectively. The minimum throughput of CT\_adjustment is 527, 317, 235, 124, 74, and 124% higher than those of HOP, ETT, DT\_No-adjustment, DT-adjustment, and CT\_No-adjustment, respectively.

Although the performance under CT\_No-adjustment is much better than that under DT\_Noadjustment, the performance under DT\_adjustment is better than that under CT\_No-adjustment. As discussed in "Introduction", under CT\_No-adjustment, although relay nodes can help to increase the capacity of a transmission pair, cooperative transmissions may also cause interference to more network nodes and consequently significant performance degradation. Compared with CT\_No-adjustment, CT adjustment can obtain much larger cooperative transmission gain, which demonstrates the effectiveness of our algorithms in relieving the interference raised by cooperative relays. The performance gain is also attributed to our algorithms for channel adjustment and adaptation of local path segments and relays. By exploiting the MRMC technique, the co-channel interference is alleviated, which is the key reason for the throughput improvement.

## **8.2. Impact of the number of orthogonal channel**

We vary the number of orthogonal channels from 1 to 15 in the network while setting other parameters to the default values. As shown in **Figure 12**, the aggregate network throughput and minimum flow throughput achieved by all the routing schemes increase when the number of orthogonal channels increases initially, while remaining the same when the number of channels is large enough for the tested five flows. As each node has only two radio interfaces and the traffic in the network is limited, extra channels cannot be fully utilized. Therefore, increasing the number of channels cannot unboundedly increase the performance of the routing schemes for the tested five flows. We observe that CT\_adjustment can exploit available orthogonal channels to increase the throughput and achieve the best performance. When the number of channels is larger than 5, CT\_adjustment achieves 251, 228, 180, 22, and 126% higher aggregate throughput compared to HOP, ETT, DT\_No-adjustment, DT-adjustment, and CT\_No-adjustment, respectively.

## **9. Conclusion**

To fulfill the complete potential of cooperative transmission in MRMC cooperative networks, a solution is proposed where cooperative routing at the network layer, channel assignment at the MAC layer, and cooperative communication at the physical layer can work coherently together to maximize the throughput. The simulation results demonstrate that cooperative communication can achieve a large capacity gain in MRMC wireless networks under welldesigned algorithms. Compared to direct transmission in multi-radio multi-channel, cooperative transmission in multi-radio multi-channel can increase the aggregate throughput more than 1.8 times when there are at least five orthogonal channels.

## **Acknowledgements**

This work is supported by the National Natural Science Foundation of China under grant nos.6157218, 61472130, 71331001, and 71420107027, U.S. National Science Foundation under grant no. CNS 1526843, the Science and Technology Projects of Hunan Province (No. 2016JC2075), and the Research Foundation of Education Bureau of Hunan Province, China (No. 16C0047).

## **Author details**

Kun Xie1,\*, Shiming He2 , Xin Wang3 , Dafang Zhang1 and Keqin Li4

\*Address all correspondence to: xiekun@hnu.edu.cn

1 College of Computer Science and Electronics Engineering, Hunan University, Changsha, China

2 School of Computer and Communication Engineering, Changsha University of Science and Technology, Changsha, China

3 Department of Electrical and Computer Engineering, State University of New York at Stony Brook, Stony Brook, USA

4 Department of Computer Science, State University of New York, New York, USA

## **References**

minimum flow throughput achieved by all the routing schemes increase when the number of orthogonal channels increases initially, while remaining the same when the number of channels is large enough for the tested five flows. As each node has only two radio interfaces and the traffic in the network is limited, extra channels cannot be fully utilized. Therefore, increasing the number of channels cannot unboundedly increase the performance of the routing schemes for the tested five flows. We observe that CT\_adjustment can exploit available orthogonal channels to increase the throughput and achieve the best performance. When the number of channels is larger than 5, CT\_adjustment achieves 251, 228, 180, 22, and 126% higher aggregate throughput compared to

To fulfill the complete potential of cooperative transmission in MRMC cooperative networks, a solution is proposed where cooperative routing at the network layer, channel assignment at the MAC layer, and cooperative communication at the physical layer can work coherently together to maximize the throughput. The simulation results demonstrate that cooperative communication can achieve a large capacity gain in MRMC wireless networks under welldesigned algorithms. Compared to direct transmission in multi-radio multi-channel, cooperative transmission in multi-radio multi-channel can increase the aggregate throughput more

This work is supported by the National Natural Science Foundation of China under grant nos.6157218, 61472130, 71331001, and 71420107027, U.S. National Science Foundation under grant no. CNS 1526843, the Science and Technology Projects of Hunan Province (No. 2016JC2075), and

the Research Foundation of Education Bureau of Hunan Province, China (No. 16C0047).

, Dafang Zhang1

1 College of Computer Science and Electronics Engineering, Hunan University, Changsha,

2 School of Computer and Communication Engineering, Changsha University of Science and

3 Department of Electrical and Computer Engineering, State University of New York at Stony

4 Department of Computer Science, State University of New York, New York, USA

and Keqin Li4

HOP, ETT, DT\_No-adjustment, DT-adjustment, and CT\_No-adjustment, respectively.

than 1.8 times when there are at least five orthogonal channels.

, Xin Wang3

\*Address all correspondence to: xiekun@hnu.edu.cn

**9. Conclusion**

82 Ad Hoc Networks

**Acknowledgements**

**Author details**

China

Kun Xie1,\*, Shiming He2

Technology, Changsha, China

Brook, Stony Brook, USA


[27] Wu H, Yang F, Tan K, Chen J, Zhang Q, Zhang Z. Distributed channel assignment and routing in multiradio multichannel multihop wireless networks. IEEE J. Sel. Areas Commun. 2006; **24**(11): 1972–1983. DOI: 10.1109/JSAC.2006.881638

[14] Ding H, Costa DB, Liu W, Ge J. Enhancing cooperative diversity gains in dual-hop oneway/two-way AF relaying systems: a fully opportunistic role selection strategy. IEEE

[15] Elhawary M, Haas Z. Energy-efficient protocol for cooperative networks. IEEE/ACM

[16] Xu H, Huang L, Qiao C, Zhang Y, Sun Q. Bandwidth power aware cooperative multipath routing for wireless multimedia sensor networks. IEEE Trans. Wireless Commun.

[17] Guo Y, Duan L, Zhang R. Optimal pricing and load sharing for energy saving with cooperative communications. IEEE Trans. Wireless Commun. 2016;**15**(2):951–964. DOI:

[18] Jayakody DNK, Flanagan MF. A soft decode–compress–forward relaying scheme for cooperative wireless networks. IEEE Trans. Veh. Technol. 2016;65(5):3033–3041. DOI:

[19] Jiang W, Kaiser T, Vinck AJH. A robust opportunistic relaying strategy for co-operative wireless communications. IEEE Trans. Wireless Commun. 2016;**15**(4):2642–2655. DOI:

[20] Nazari B, Jamalipour A. Contract-auction based distributed resource allocation for cooperative communications. IET Commun. 2016;10(9):1087–1095. DOI: 10.1049/iet-com.2015.

[21] Xie K, Wang X, Liu X, Wen J, Cao J. Interference-aware cooperative communication in multi-radio multi-channel wireless networks. IEEE Trans. Comput. 2016;**65**(5):1528–

[22] Xie K, Wang X, Wen J, Cao J. Cooperative routing with relay assignment in multi-radio multihop wireless networks. IEEE/ACM Trans. Netw. 2016;**24**(2):859–872. DOI: 10.1109/

[23] Xie K, Li H, Wang X, He S, Wen J, Guizani M. Joint cooperative routing and channel assignment in multi-flow multi-hop wireless networks. In: IWCMC; 4–8 August. 2014;

[24] Qiao H, Zhang D, Xie K, Zhang J, He S. Power-bandwidth aware cooperative routing in multi-radio multi-channel wireless network. In: IEEE ISPA; 23-26 Aug; Tianjin China.

[25] Qiao H, Zhang D, Xie K, Zhang J, He S. A distributed joint cooperative routing and channel assignment in multi-radio wireless mesh network. In: ICA3PP; 18–20 November

[26] Raniwala A, Gopalan K, Chiueh T-C. Centralized channel assignment and routing algorithms for multi-channel wireless mesh networks. ACM SIGMOBILE Mobile Comput.

Trans. Veh. Technol. 2015;64(8): 3440–3457. DOI: 10.1109/TVT.2014.2356506

Trans. Netw. 2011;**19**(2):561–574. DOI: 10.1109/TNET.2010.2089803

2012;**11**(4):1532–1543. DOI: 10.1109/TWC.2012.020812.111265

10.1109/TWC.2015.2480771

10.1109/TVT.2015.2442459

10.1109/TWC.2015.2506574

1542. DOI: 10.1109/TC.2015.2448089

Nicosia. New York, NY: IEEE; 2014, p. 1188–1193.

2015; Zhangjiajie China. Springer Verlag; 2015, pp. 552–566.

Commun. Rev. 2004; **8**(2):50–65. DOI: 10.1145/997122.997130

New York, NY: IEEE; 2016, pp. 1342–1349.

0764

84 Ad Hoc Networks

TC.2015.2448089


**Section 2**
