**Autonomous Decentralized Control Scheme for Long-Term Operation of Large Scale and Dense Wireless Sensor Networks with Multiple Sinks**

Akihide Utani *Tokyo City University, Japan* 

#### **1. Introduction**

444 Environmental Monitoring

and use: developing tools to support management and policy. In print. Pahl-Wostl, C. (2007). The implications of complexity for integrated resources management.

Smit, A.M. (2003). Adaptive monitoring: an overview. *DOC Science Internal Series*, vol. 138.

Timmerman, J.G. and S. Langaas (2004). Conclusions. In *Environmental information in* 

Timmerman, J.G., J.J. Ottens, and R.C. Ward (2000). The information cycle as a framework

Walker, B. H., L. H. Gunderson, A. P. Kinzig, C. Folke, S. R. Carpenter, and L. Schultz

Walker, B. and J. A. Meyers (2004). Thresholds in ecological and social–ecological systems: a

Ward, R.C. (1995). Monitoring Tailor-made: what do you want to know? In *Proceeding of the* 

pp. 240-246. IWA Publishing, London, UK. ISBN: 1 84339 038 8.

in social-ecological systems. *Ecology and Society* 11(1): 13. [online]

developing database. *Ecology and Society* 9(2): 3. [online] URL:

URL:http://www.ecologyandsociety.org/vol11/iss1/art13/

23 September 1994, Beekbergen, The Netherlands.

http://www.ecologyandsociety.org/vol9/iss2/art3

*Environmental Modelling and Software*, 22: 561-569.

Department of Conservation, Wellington. 16 p.

*Management* 25(3): 229-239.

A, van Delden, H, Gaddis, E, Assaf, H. (2006). Bridging the gaps between design

*European transboundary water management.* Timmerman, J.G. and S. Langaas (eds.),

for defining information goals for water-quality monitoring. *Environmental* 

(2006). A handful of heuristics and some propositions for understanding resilience

*international workshop on monitoring and assessment in water management; Monitoring Tailor-Made.* Adriaanse M., J van der Kraats; P.G. Stocks, and R.C. Wards (eds), 20Various communication services have been provided. They include environmental monitoring and/or control, ad-hoc communication between mobile nodes, and inter-vehicle communication in intelligent transport systems. As a means of facilitating the above advanced communication services, autonomous decentralized networks, such as wireless sensor networks (Akyildiz et al., 2002; Rajagopalan & Varshney, 2006), mobile ad-hoc networks (Perkins & Royer, 1999; Johnson et al., 2003; Clausen & Jaquet, 2003; Ogier et al., 2003), and wireless mesh networks (Yamamoto et al., 2009), have been intensively researched with great interests. Especially, a wireless sensor network, which is a key network to construct ubiquitous information environments, has great potential as a means of realizing a wide range of applications, such as natural environmental monitoring, environmental control in residential spaces or plants, object tracking, and precision agriculture (Akyildiz et al., 2002). Recently, there is growing expectation for a new network service by a wireless sensor network consisting of a lot of static sensor nodes arranged in a service area and a few mobile robots as a result of the strong desire for the development of advanced systems that can flexibly function in dynamically changing environments (Matsumoto et al., 2009).

In this chapter, a large scale and dense wireless sensor network made up of many static sensor nodes with global positioning system, which is a representative network to actualize the above-mentioned sensor applications, is assumed. In a large scale and dense wireless sensor network, generally, hundreds or thousands of static sensor nodes limited resources, which are compact and inexpensive, are placed in a service area, and sensing data of each node is gathered to a sink node by inter-node wireless multi-hop communication. Each sensor node consists of a sensing function to measure the status (temperature, humidity, motion, etc.) of an observation point or object, a limited function of information processing, and a simplified wireless communication function, and it generally operates on a resource with a limited power-supply capacity such as a battery. Therefore, a data gathering scheme and/or a routing protocol capable of meeting the following requirements is mainly needed to prolong the lifetime of a large scale and dense wireless sensor network composed of hundreds or thousands of static sensor nodes limited resources.


Autonomous Decentralized Control Scheme for Long-Term

**2. Autonomous decentralized control scheme** 

nts from each sink node of its own neighborhood nodes.

accompanying the hop determined within the interval [0,1].

data transmission in neighborhood nodes of node (*i*).

**2.1 Routing algorithm** 

(*i*) is as follows:

routing table.

the following equation

Operation of Large Scale and Dense Wireless Sensor Networks with Multiple Sinks 447

To facilitate the long-term operation of an actual sensor network service, a recent approach has been to introduce multiple sinks in a wireless sensor network (Dubois-Ferriere et al., 20- 04; Oyman & Ersoy, 2004). In a wireless sensor network with multiple sinks, sensing data of each node is generally allowed to gather at any of the available sinks. Our scheme (Matsumoto et al., 2010) is a new data gathering scheme based on this assumption, which can be expected to produce a remarkable effect in a large scale and dense wireless sensor network with multiple sinks. In our scheme, each sensor node can select either of high power and low power for packet transmission, where high power corresponds to normal transmission power and low power is newly introduced to moreover balance the load of each sensor node.

Each sink node has a connective value named a "value to self", which is not updated by transmitting a control packet and receiving data packets. In the initial state of a large scale and dense wireless sensor network with multiple sinks, each sink node broadcasts a control packet containing its own location information, ID, hop counts(=*0*), and "value to self" by high power. This control packet is rebroadcast throughout the network with hop counts updated by high power. By receiving the control packet from each sink node, each sensor node can grasp the "value to self" of each sink node, their location information, IDs, and the hop cou-

Initial connective value of each sensor node, which is the connective value before starting data transmission, is generated by using the "value to self" of each sink node and the hop counts from each sink node. The procedure for computing initial connective value of a node

1. The value [*vij*(*0*)] on each sink node (*j*=1, … ,*S*) of node (*i*) is first computed according to

*v* (*0*) *vo dr* ( *j* 1*, ,S*) *ij hops*

2. Then, initial connective value [*vi*(*0*)] of node (*i*) is generated by the following equation

where this connective value [*vi*(*0*)] can be also conducted from the following equation

In the above Equation (3), *vmi*(*0*) represents the greatest connective value before starting

Before data transmission is started, each sensor node computes initial connective value of each neighborhood node based on the above Equations (1) and (2), and stores the computed connective value, which is used as the only index to evaluate the relay destination value of each neighborhood node, in each neighborhood node field of its own

where *voj*(*j*=1, … ,*S*) is the "value to self" of sink node (*j*), *hopsij*(*j*=1, … ,*S*) is the hop counts from sink node (*j*) of node (*i*). *dr* represents the value attenuation factor

*ij <sup>j</sup>* (1)

*<sup>v</sup>* (*0*) max *<sup>v</sup>* (*0*) ( *<sup>j</sup>* 1, *...* , *<sup>S</sup>*) *<sup>i</sup> ij* (2)

*<sup>v</sup> <sup>0</sup> vm <sup>0</sup> dr <sup>i</sup>*( ) *<sup>i</sup>*( ) (3)

As data gathering schemes for the long-term operation of a wireless sensor network, cluster-ing-based data gathering (Heinzelman et al., 2000; Dasgupta et al., 2003; Jin et al., 2008) and synchronization-based data gathering (Wakamiya & Murata, 2005; Nakano et al., 2009; Nak-ano et al., 2011) are under study, but not all the above requirements are satisfied. Recently, bio-inspired routing algorithms, such as ant-based routing algorithms, have attracted a sign-ificant amount of interest from many researchers as examples that satisfy the three require-ments above. In ant-based routing algorithms (Subramanian et al., 1998; Ohtaki et al., 2006), the routing table of each sensor node is generated and updated by applying the process in which ants build routes between their nest and food using chemical substances (pheromon-es). Advanced ant-based routing algorithm (Utani et al., 2008) is an efficient route learning algorithm which shares route information between control messages. In contrast to conven-tional ant-based routing algorithms, this can suppress the communication load of each sen-sor node and adapt itself to network topology changes. However, this does not positively ease the communication load concentration on specific sensor nodes, which is the source of problems in the long-term operation of a wireless sensor network. Gradient-based routing protocol (Xia et al., 2004) actualizes load-balancing data gathering. However, this cannot su-ppress the communication load concentration to sensor nodes around the set sink node. Int-ensive data transmission to specific sensor nodes results in concentrated energy consumpti-on by them, and causes them to break away from the network early. This makes long-term observation by a wireless sensor network difficult.

In a large scale and dense wireless sensor network, the communication load is generally concentrated on sensor nodes around the set sink node during the operation process. In cases where sensor nodes are not placed evenly in a large scale observation area, the communication load is concentrated on sensor nodes placed in an area of low node density. To solve this communication load concentration problem, a data gathering scheme for a wireless sensor network with multiple sinks has been proposed (Dubois-Ferriere et al., 2004; Oyman & Ersoy, 2004). In this scheme, each sensor node sends sensing data to the nearest sink node. In comparison with the case of one-sink wireless sensor networks, the communication load of sensor nodes around a sink node is reduced. In each sensor node, however, the destination sink node cannot be selected autonomously and adaptively. In cases where original data transmission rate from each sensor node is not even, therefore, the load of load-concentrated nodes is not sufficiently balanced. An autonomous load-balancing data transmission scheme is required.

This chapter represents a new data gathering scheme with transmission power control that adaptively reduces the load of load-concentrated nodes and facilitates the long-term operation of a large scale and dense wireless sensor network with multiple sinks (Matsumoto et al., 2010). This scheme has autonomous load-balancing data transmission devised by considering the application environment of a wireless sensor network as a typical example of complex systems where the adaptive adjustment of the entire system is realized from the local interactions of components of the system. In this scheme, the load of each sensor node is autonomously balanced. This chapter consists of four sections. In Section 2, the above data gathering scheme (Matsumoto et al., 2010) is detailed and its novelty and superiority are described. In Section 3, the results of simulation experiments are reported and the effectiveness of our scheme (Matsumoto et al., 2010) is demonstrated by comparing its performances with those of existing schemes. In Section 4, the overall conclusions of this work are given and future problems are discussed.

#### **2. Autonomous decentralized control scheme**

To facilitate the long-term operation of an actual sensor network service, a recent approach has been to introduce multiple sinks in a wireless sensor network (Dubois-Ferriere et al., 20- 04; Oyman & Ersoy, 2004). In a wireless sensor network with multiple sinks, sensing data of each node is generally allowed to gather at any of the available sinks. Our scheme (Matsumoto et al., 2010) is a new data gathering scheme based on this assumption, which can be expected to produce a remarkable effect in a large scale and dense wireless sensor network with multiple sinks. In our scheme, each sensor node can select either of high power and low power for packet transmission, where high power corresponds to normal transmission power and low power is newly introduced to moreover balance the load of each sensor node.

#### **2.1 Routing algorithm**

446 Environmental Monitoring

As data gathering schemes for the long-term operation of a wireless sensor network, cluster-ing-based data gathering (Heinzelman et al., 2000; Dasgupta et al., 2003; Jin et al., 2008) and synchronization-based data gathering (Wakamiya & Murata, 2005; Nakano et al., 2009; Nak-ano et al., 2011) are under study, but not all the above requirements are satisfied. Recently, bio-inspired routing algorithms, such as ant-based routing algorithms, have attracted a sign-ificant amount of interest from many researchers as examples that satisfy the three require-ments above. In ant-based routing algorithms (Subramanian et al., 1998; Ohtaki et al., 2006), the routing table of each sensor node is generated and updated by applying the process in which ants build routes between their nest and food using chemical substances (pheromon-es). Advanced ant-based routing algorithm (Utani et al., 2008) is an efficient route learning algorithm which shares route information between control messages. In contrast to conven-tional ant-based routing algorithms, this can suppress the communication load of each sen-sor node and adapt itself to network topology changes. However, this does not positively ease the communication load concentration on specific sensor nodes, which is the source of problems in the long-term operation of a wireless sensor network. Gradient-based routing protocol (Xia et al., 2004) actualizes load-balancing data gathering. However, this cannot su-ppress the communication load concentration to sensor nodes around the set sink node. Int-ensive data transmission to specific sensor nodes results in concentrated energy consumpti-on by them, and causes them to break away from the network early. This makes long-term

In a large scale and dense wireless sensor network, the communication load is generally concentrated on sensor nodes around the set sink node during the operation process. In cases where sensor nodes are not placed evenly in a large scale observation area, the communication load is concentrated on sensor nodes placed in an area of low node density. To solve this communication load concentration problem, a data gathering scheme for a wireless sensor network with multiple sinks has been proposed (Dubois-Ferriere et al., 2004; Oyman & Ersoy, 2004). In this scheme, each sensor node sends sensing data to the nearest sink node. In comparison with the case of one-sink wireless sensor networks, the communication load of sensor nodes around a sink node is reduced. In each sensor node, however, the destination sink node cannot be selected autonomously and adaptively. In cases where original data transmission rate from each sensor node is not even, therefore, the load of load-concentrated nodes is not sufficiently balanced. An autonomous load-balancing data transmission scheme

This chapter represents a new data gathering scheme with transmission power control that adaptively reduces the load of load-concentrated nodes and facilitates the long-term operation of a large scale and dense wireless sensor network with multiple sinks (Matsumoto et al., 2010). This scheme has autonomous load-balancing data transmission devised by considering the application environment of a wireless sensor network as a typical example of complex systems where the adaptive adjustment of the entire system is realized from the local interactions of components of the system. In this scheme, the load of each sensor node is autonomously balanced. This chapter consists of four sections. In Section 2, the above data gathering scheme (Matsumoto et al., 2010) is detailed and its novelty and superiority are described. In Section 3, the results of simulation experiments are reported and the effectiveness of our scheme (Matsumoto et al., 2010) is demonstrated by comparing its performances with those of existing schemes. In Section 4, the overall conclusions of this

observation by a wireless sensor network difficult.

work are given and future problems are discussed.

is required.

Each sink node has a connective value named a "value to self", which is not updated by transmitting a control packet and receiving data packets. In the initial state of a large scale and dense wireless sensor network with multiple sinks, each sink node broadcasts a control packet containing its own location information, ID, hop counts(=*0*), and "value to self" by high power. This control packet is rebroadcast throughout the network with hop counts updated by high power. By receiving the control packet from each sink node, each sensor node can grasp the "value to self" of each sink node, their location information, IDs, and the hop counts from each sink node of its own neighborhood nodes.

Initial connective value of each sensor node, which is the connective value before starting data transmission, is generated by using the "value to self" of each sink node and the hop counts from each sink node. The procedure for computing initial connective value of a node (*i*) is as follows:

1. The value [*vij*(*0*)] on each sink node (*j*=1, … ,*S*) of node (*i*) is first computed according to the following equation

$$\text{cov}\_{\boldsymbol{\eta}}(\boldsymbol{\mathcal{O}}) = \text{cov}\_{\boldsymbol{\eta}} \times d\boldsymbol{r}^{\text{hops}\_{\boldsymbol{\eta}}} \quad (j = 1, \ldots, S) \tag{1}$$

where *voj*(*j*=1, … ,*S*) is the "value to self" of sink node (*j*), *hopsij*(*j*=1, … ,*S*) is the hop counts from sink node (*j*) of node (*i*). *dr* represents the value attenuation factor accompanying the hop determined within the interval [0,1].

2. Then, initial connective value [*vi*(*0*)] of node (*i*) is generated by the following equation

$$\mathbf{v}\_{i}(\theta) = \max \mathbf{v}\_{i\circ}(\theta) \qquad \quad (j = 1, \ldots, S) \tag{2}$$

where this connective value [*vi*(*0*)] can be also conducted from the following equation

$$\text{v}\_{i}(\mathbf{0}) = \text{v}m\_{i}(\mathbf{0}) \times dr \tag{3}$$

In the above Equation (3), *vmi*(*0*) represents the greatest connective value before starting data transmission in neighborhood nodes of node (*i*).

Before data transmission is started, each sensor node computes initial connective value of each neighborhood node based on the above Equations (1) and (2), and stores the computed connective value, which is used as the only index to evaluate the relay destination value of each neighborhood node, in each neighborhood node field of its own routing table.

Autonomous Decentralized Control Scheme for Long-Term

*q*

*vq*

・・・ ・・・ *vms*

・・・ *node r node x* ・・・ Next Hop node *s* routing table

*r*

*vr*

(*t*) ・・・

Fig. 2. An example of autonomous load-balancing data transmission to multiple sinks

Our scheme (Matsumoto et al., 2010) requires the construction of a data gathering environment in the initial state of a large scale and dense wireless sensor network with multiple sinks, but needs no special communication for network control. The above-mentioned simple mechanism alone achieves autonomously adaptive load-balancing data transmission to multiple sinks, as in Fig.2. The lifetime of a wireless sensor network can be extended by reducing the communication load for network control and solving the node load concentration problem.

For data packet transmission, the transmission power of each sensor node is switched to low power if its own residual energy is less than the set threshold [*Te*]. In this case, each sensor node selects the neighboring node with the greatest connective value within range of radio wave of low power as a relay node, and transmits the data packet to this selected node by

*Sink1*

**2.3 Transmission power control** 

low power.

*p*

*vp*

Operation of Large Scale and Dense Wireless Sensor Networks with Multiple Sinks 449

nnective value of the source node based on the above Equation (4). Each neighborhood node that intercepts this packet stores the updated connective value in the source node field of its own routing table. Fig.1 shows an example of data packet transmission and its accompanying connective value update. In this example, node (*l*) refers to its own routing table and addresses the data packet to node (*r*), which has the greatest connective value [*vml*(*t*)]. When this data packet is intercepted, each neighboring node around node (*l*) stores the updated

*s*

*x*

: *data packet*

*Sink2*

connective value [*vl*(*t*)] in the data packet in the node (*l*) field of its own routing table.

#### **2.2 Data transmission and connective value update**

For a while from starting data transmission, each sensor node selects the neighboring node with the greatest connective value from its own routing table as a relay node, and transmits the data packet to this selected node by high power. In cases where more than one node shares the greatest connective value, however, the relay node is determined between them at random. The data packet in each sensor node is not sent to a specified sink node. By repetitive data transmission to the neighboring node with the greatest connective value, data gathering at any of the available sinks is completed. In our scheme, the connective value of each sensor node is updated by considering residual node energy. Therefore, by repetitive data transmission to the neighboring node with the greatest connective value, the data transmission routes are not fixed.

To realize autonomous load-balancing data transmission, in our scheme (Matsumoto et al., 2010), the data packet from each sensor node includes its own updated connective value. We assume that a node (*l*) receives a data packet at time (*t*). Before node (*l*) relays the data packet, it replaces the value in the connective value field of the data packet by its own renewal connective value computed according to the following connective value update equation

$$\mathbf{v}\_{l}(t) = \nu \mathbf{m}\_{l}(t) \times dr \times \begin{pmatrix} \mathbf{e}\_{l}(t) \\ \mathbf{E}\_{l} \end{pmatrix} \tag{4}$$

where *vml*(*t*) is the greatest connective value at time (*t*) in the routing table of node (*l*). *el*(*t*) and *El* represent the residual energy at time (*t*) of node (*l*) and the battery capacity of node (*l*), respectively.

Fig. 1. Data packet transmission and connective value update

In our scheme, the data packet addressed to the neighboring node with the greatest connective value is intercepted by all neighboring nodes. This data packet includes the updated co-

For a while from starting data transmission, each sensor node selects the neighboring node with the greatest connective value from its own routing table as a relay node, and transmits the data packet to this selected node by high power. In cases where more than one node shares the greatest connective value, however, the relay node is determined between them at random. The data packet in each sensor node is not sent to a specified sink node. By repetitive data transmission to the neighboring node with the greatest connective value, data gathering at any of the available sinks is completed. In our scheme, the connective value of each sensor node is updated by considering residual node energy. Therefore, by repetitive data transmission to the neighboring node with the greatest connective value, the data transmiss-

To realize autonomous load-balancing data transmission, in our scheme (Matsumoto et al., 2010), the data packet from each sensor node includes its own updated connective value. We assume that a node (*l*) receives a data packet at time (*t*). Before node (*l*) relays the data packet, it replaces the value in the connective value field of the data packet by its own renewal connective value computed according to the following connective value update equation

*<sup>l</sup> <sup>l</sup> <sup>l</sup> <sup>E</sup>*

where *vml*(*t*) is the greatest connective value at time (*t*) in the routing table of node (*l*). *el*(*t*) and *El* represent the residual energy at time (*t*) of node (*l*) and the battery capacity of node

> *l r s*

Fig. 1. Data packet transmission and connective value update

**Data Packet**

*l*

Next Hop

Next Hop ・・・ *node s* ・・・ ・・・

*node r vml*(*t*)

In our scheme, the data packet addressed to the neighboring node with the greatest connective value is intercepted by all neighboring nodes. This data packet includes the updated co-

node *l* routing table

node *s* routing table

・・・

・・・

・・・

・・・

・・・ *node l* ・・・ *vl*(*t*)

・・・ ・・・

*<sup>v</sup> <sup>t</sup> vm <sup>t</sup> dr* ( ) ( ) ( ) (4)

*e t*

**2.2 Data transmission and connective value update** 

ion routes are not fixed.

(*l*), respectively.

nnective value of the source node based on the above Equation (4). Each neighborhood node that intercepts this packet stores the updated connective value in the source node field of its own routing table. Fig.1 shows an example of data packet transmission and its accompanying connective value update. In this example, node (*l*) refers to its own routing table and addresses the data packet to node (*r*), which has the greatest connective value [*vml*(*t*)]. When this data packet is intercepted, each neighboring node around node (*l*) stores the updated connective value [*vl*(*t*)] in the data packet in the node (*l*) field of its own routing table.

Fig. 2. An example of autonomous load-balancing data transmission to multiple sinks

Our scheme (Matsumoto et al., 2010) requires the construction of a data gathering environment in the initial state of a large scale and dense wireless sensor network with multiple sinks, but needs no special communication for network control. The above-mentioned simple mechanism alone achieves autonomously adaptive load-balancing data transmission to multiple sinks, as in Fig.2. The lifetime of a wireless sensor network can be extended by reducing the communication load for network control and solving the node load concentration problem.

### **2.3 Transmission power control**

For data packet transmission, the transmission power of each sensor node is switched to low power if its own residual energy is less than the set threshold [*Te*]. In this case, each sensor node selects the neighboring node with the greatest connective value within range of radio wave of low power as a relay node, and transmits the data packet to this selected node by low power.

Autonomous Decentralized Control Scheme for Long-Term

respectively.

nodes

and 0.5J, respectively.

Table 1. Conditions of simulation

Operation of Large Scale and Dense Wireless Sensor Networks with Multiple Sinks 451

detected abnormal data set were assumed to transmit the measurement data. The conditions of the si-mulation which were used in the experiments performed are shown in Table1. In the initial state of the simulation experiments, static sensor nodes are randomly arranged in the set ex-perimental area, and multiple sinks are placed on the boundaries containing the corners of this area. The network configuration is shown in Fig.4. In the experiments performed, the value attenuation factor accompanying hop (*dr*) and the "value to self" of each sink node in-troduced in our scheme were set to 0.5 and 100.0,

Number of sinks *2 or 3*

Size of each control packet *6* [*bytes*] Size of each data packet *18* [*bytes*]

Fig. 4. Large scale and dense wireless sensor network consisting of many static sensor

results in a preliminary investigation were adopted in preference to existing ones.

**3.2 Experimental results on simulation model with two sinks** 

In the experimental results reported, our scheme (Matsumoto et al., 2010) is evaluated through a comparison with existing ones (Dubois-Ferriere et al., 2004; Oyman & Ersoy, 2004; Ohtaki et al., 2006; Utani et al., 2008) where the parameter settings that produced good

In this subsection, experimental results on the simulation model with two sinks of our scheme without transmission power control are shown, where the number of sensor nodes was 1000, the range of radio wave and the battery capacity of each sensor node were set to 150m

Battery capacity of each sensor node *0.2* [*J*] *or 0.5*[*J*]

*evaluation node*

Range of radio wave *150m or 200m* Number of sensor nodes *750*, *1000*, *1250* Simulation size *2400m* × *2400m*

Fig. 3. An example of transmission power control

Fig.3 shows an example of the above transmission power control, which means that the tra-nsmission power of each sensor node is switched to low power according to the above con-dition. In this example, node (*m*) is a load concentration node. Node (*m*) has autonomously transmitted the data packet to node (*r*) with the greatest connective value within low power range by low power because its own residual energy has become less than the set threshold [*Te*]. By switching to low power, the energy consumption of node (*m*) is saved, but node (*k*) and node (*l*) may continue to transmit the data packet to node (*m*) because they cannot grasp the updated connective value of node (*m*). In our scheme, therefore, every tenth data packet from the node switched to low power is transmitted by high power.
