Feature
OOppttiim
miissiinngg llooccaattiioonnss ooff
ssiinnkk nnooddeess
in wireless sensor
networks using genetic algorithms
Dr Lili Yang
The Business School, Loughborough University
W
ireless sensor network (WSN) is one of thefocus-ing topics in the realm of computer science and electronic engineering. The applications of WSN cover a wide range from natural monitoring to ambient awareness, from military to surveillance. The challenges in the research on WSN are the need for self-configuration and self-maintenance, and the extreme resource poverty of their individual sensor nodes in terms of memory, data pro-cessing capability, and life time. These new features mean that the design methodology for traditional networking is not suitable for WSN because of the severe and untradi-tional constraints1,2.
Although WSNs may take various topologies such as star, ring, mesh or tree, the signals from individual sensors of WSNs are often transferred to the Internet or other ter-minals via several sink nodes (in some cases it may be called a cluster centre). The sink nodes may have greater computing capability, i.e. storage capacity and processor speed, than an ordinary sensor node, but no greater trans-mitter power or receiver sensitivity. They perform gateway duties between WSNs and the Internet via wireless com-munication protocols such as IEEE 802.11b3-5. This paper will focus on optimising sink node locations in order to achieve certain objectives such as minimal average dis-tance from individual sensor nodes to the sink node. The rest of paper is organised as follows. The next section describes the features of WSN and the role of the sink nodes in WSN. Then the previous research in sink node locations is reviewed. Descriptions of the system assump-tion, system model and the optimal objectives in the opti-misation of sink node locations are given and then formu-late the optimal location problem of the sink node as a search problem which a genetic algorithm can be applied to. A simulation result is also given before concluding the paper.
Sink nodes in WSNs
Features of WSN
WSN typically consists of a large number of nodes with sensing and routing capabilities. The sensor nodes are usu-ally scattered in a sensor field. The position of sensor nodes need not be engineered or pre-determined. This allows random deployment in inaccessible areas such as disaster relief operations or battlefields. WSN has a
num-ber of features which distinguish it from ordinary wireless ad hoc networks, including6:
l The number of sensor nodes in a sensor network can be much bigger than the nodes in an ad hoc network; l Sensor nodes are densely deployed. Dense interconnec-tivity leads to redundancy in the range measurements; l Sensor nodes are prone to failures;
l Sensor nodes are limited in power, computational capac-ities and memory.
To reduce the power consumption attributed to commu-nication and to minimise interference, every sensor node can only communicate to its immediate neighbours result-ing in a mesh of connections. Hence, multi-hop communi-cation in sensor networks is expected to consume less power than the traditional single-hop communication since large numbers of sensor nodes are densely deployed.
One of the most important constraints on sensor nodes is the low power consumption requirement. The wireless sensor node, being a microelectronic device, can only be equipped with a limited power source. In some application scenarios, power sources of sensor nodes are irreplaceable. Sensor node lifetime, therefore, shows a strong depend-ence on battery lifetime. Sensor network protocols usually focus primarily on power conservation. Single-hop long range transmission is often prohibited.
Sink nodes in WSN
Sink nodes are designated devices connected with ordinary sensor nodes but more powerful than them and bridging sensor networks with the end users. The sink nodes may be envisaged as a laptop computer receiving data from the network, or a much smaller and dedicated micro-controller
Figure 1: Sink node with scattered sensor nodes.
A version of this work was published at the 2006 IEEE International Conference on Systems, Man & Cybernetics (SMC), Taipei, Taiwan, October 8-11. Yang 21/8/06 11:01 am Page 1
Feature
providing the gateway function. As illustrated in Figure 1, each sensor node has the capabilities to collect data and route data back to the sink node and the end users. Data are routed back to the end user by a multi-hop infrastructure-less architecture through the sink node.
There might be more than one sink nodes simultaneous-ly working in a sensor network, depending on application requests. Figure 2 shows two sink nodes X and Y express identical interests to an event occurring in the sensor node B. Sensor node A is one hop away from its nearest sink node Y, and many hops away from sink node X when two sink nodes are deployed. Thus, by employing two sink nodes instead of one sensor, node A will need fewer hops and less power to transmit its signal to the sink node. This is because the energy consumed in routing a message for any sensor node to its nearest base station is proportional to number of hops the message has to travel. Employing multiple base stations effectively reduces the energy con-sumption per message delivered. As the number of sink nodes is increased, the path length from a sensor node to a sink node is decreased and the lifetime of the sensor node is increased. However, the number of sink nodes is con-strained financially because the cost of the sink node is more expensive than the sensor node.
Previous research on sink node location
Optimising sink node location has not been widely explored in the WSN communities. Heinzelman et al7 demonstrated through experimental results that the sensor nodes which are one-hop away from a sink node drain their energy faster than other nodes in the sensor network. It is because nodes which are one-hop away from a sink node need to forward messages originating from many other nodes, in addition to delivering their own generated mes-sages. The work load of these nodes which are near the sink node is much bigger than those far away from it. Therefore, these sensor nodes deplete their energy quickly and finish their lifetime ‘died’. If too many sensor nodes around the sink node die, i.e. become in-operational, other living sensor nodes will be unable to communicate with the sink node via these dying sensor nodes and the sensor network becomes in-operational.
In order to increase the lifetime of sensor networks, mul-tiple sink nodes, mobile sink nodes, and mulmul-tiple mobile sink nodes have been investigated3-5, 8-10. A first attempt on how to determine specific sink movements for energy min-imisation is presented in Gandham et al’s work4. The authors deployed multiple, mobile sink nodes to prolong the lifetime of the sensor network. They split the lifetime of the sensor network into equal periods of time known as
rounds. Sink nodes (they called base stations) are relocated at the start of a round. An integer linear program is used to determine feasible locations the sink nodes in the sensor net-work should travel to and a flow-based routing protocol to ensure energy efficient routing during each round. The model used aims at minimising the energy consumption per node and the total energy consumption during a given time. Wang et al.5 were concerned with the joint problems of determining the movements of the sink nodes and the times the sink nodes sojourn at certain sensor nodes so that sensor network lifetime is maximised. They considered WSNs where the homogeneous nodes are arranged in a bi-dimen-sional grid, i.e. a mesh connection and one mobile sink node travels through them. They defined the overall network life-time as the life-time till the first node ‘dies’ for energy depletion. The objective function for the network lifetime maximisation problem expressed network lifetime in term of sink sojourn time at the sensor nodes. Their simulation results show that the sink sojourns at the four corners for most of the time, and in the grid central area.
A similar problem for the cluster architecture of WSNs was discussed by Kim et al3. The cluster header in the clus-ter architecture has data forwarding functionality. A two-step procedure was implemented for addressing the loca-tion of sink nodes and the traffic flow rates of routing paths in WSNs. In the first step, it is assumed that the location and the number of the sink nodes are fixed. In the second step, instead of assuming the pre-fixed location and num-ber of the sink nodes, the constraints of the maximum number of sink nodes are only assumed. Two types of lin-ear programs are formulated to find both the optimal loca-tion of sink nodes and the optimal traffic flow rates of rout-ing paths in WSNs.
Optimal locations of sink nodes
System assumption
We restrict our work in a WSN with mesh topology, as shown in Figure 3. The following assumptions about WSNs are made in building the system model:
l All sensor nodes are stationary and located in a bi-dimensional square grid composed of same-size cells; l Multiple sink nodes are fixed on the grid;
l Data transmission and reception are the major energy consuming activities;
l All sensor nodes have an equal initial energy;
l The transmission range of each sensor node is fixed and equals the distance between two adjacent nodes in the grid, i.e. a hop is of one cell side length. Sensor nodes commu-nicate with the sink nodes by sending data via multiple hops along the shortest path;
l The number of sink nodes is fixed and known a priori; l Sink nodes can be located only at certain sites in the grid, called feasible sites.
l The energy consumed in a senor node when transmitting is constant, and is the same as the energy consumed for receiving a bit.
Formally, a sensor network is represented as a graph G(V,E), where V are the vertices representing sensor nodes and E edges representing one-hop connectivity (i, j) where i and j are neighbouring nodes. Considering the mesh topology in the sensor network, a sensor node i can com-municate directly with its four neighbouring nodes. If the sensor node is not linked with the sink node through
one-Figure 2: Multiple sink nodes with scattered sensor nodes.
hop connectivity, then data packages generated at this sen-sor node have to relay through multiple hops to reach the sink.
System notations
We adopt the notations and the mathematical formulation of power consumption at each sensor node from Wang et al’s work5, and extend it into a multiple static sink node case. The notations and the power consumption at each sensor node in this sub-section come from Wang et al’s work5.
Parameters
l e: Energy consumption coefficient for transmitting or receiving one bit (Joules/bit).
l e0: Initial energy (Joules) of each node minus the thresh-old energy required for node operation.
l r: Rate at which data packets are generated (bits/sec); for the homogeneous sensor nodes r is the same for all sen-sor nodes.
l Cik: Power consumption for receiving and transmitting packets at node I when the sink node is located at node k (Joules/sec).
l z: Network lifetime (secs).
Simplified routing protocol
There are rich literatures on routing algorithms of WSN1,2. In the WSN with the mesh topology we consider a very simple routing protocol as shown in Figure 4.
l When a sensor node lies on the same horizontal or ver-tical line of the sink node, a unique shortest path between the sensor node and the sink node is taken.
l Otherwise, the two paths along the perimeter of the rec-tangle with the sensor node and the sink node as the oppo-site corners will be taken at equal frequencies.
Three cases are illustrated in Figure 4. Two hops in a unique route are required in Figures 4(a) and 4(b) for trans-mitting a data package from node I to the sink node. Four hops in two symmetric routes are required in Figure 4(c).
Calculation of power consumption of individual sensor nodes
In Wang et al’s work5, each node’s position is represented using the ordered pair of the node’s column and row num-ber (x, y), x= 0, 1, …, L-1; y= 0, 1, …, L-1. L is the column and row number in the grid. A pair of horizontal and verti-cal dotted lines is drawn enclosing the nodes associated with the row and the column of the sink. These lines parti-tion the sensor field into nine subsets as shown in Figure 5: UL (Upper Left), UR (Upper Right), LL (Lower Left), LR (Lower Right), VA (Vertical Above), VB (Vertical
Below), HL (Horizontal Left), HR (Horizontal Right), and the node K with which the sink node is co-located.
Wang et al5gave the following formulas to calculate the energy consumption at node i:
Using node i in subset UR as an example, Node i trans-mits its own generated data packets to node j2and j4, one half each. Nodes j2and j4 relay these packets to the sink
Figure 3: WSN with a mesh topology
Figure 4: Simplified routing protocol for a WSN with the mesh topology.
Figure 5: Data flows received and transmitted at node i5
∈ ∈ − − − ∈ − + ∈ + − ∈ + + ∈ − + − ∈ − + + ∈ − + − ∈ − + + = k i er LR i y x L erer L x y i LL UR i y x L erer x y i UL VB i L y L erer y L i VA HR i L x L erer x L i HL ck i ) 1 2 ( ( ) ) ((1 ) ) 1 ) 1 )( (((( 1)(1 ) 1) ) 1 ) 1 )( (((( 1)(1 ) 1) (1)
Feature
node k. In addition, node i receives half of the packets gen-erated at nodes j1and j3, and half of the packets generated at nodes l1 and l2. Then, node i retransmits the packets originated at nodes j3and l2to node j2and those originat-ed at nodes j1and l1 to node j4. In summary, node i receives data packets at a rate 2r and transmits at a rate 3r, having therefore power consumption cki=5re.
If there are multiple sink nodes in the sensor field, the sensor field can be further partitioned. Fig. 6 illustrates the partition of the sensor field with two sink nodes k1 and k2. The subsets are denoted as UL1, UR1, LL1, LR1, VA1, VB1, HL1, HR1, and UL2, UR2, LL2, LR2, VA2, VB2, HL2, HR2. Depending on the locations of the two sink nodes, some of the subsets will be overlapped.
The formulas for calculating the energy consumption shown in Equation 1 will be still suitable for the multiple sink node cases shown in Figure 6 only if all the sensor nodes are assigned to their nearest sink nodes.
Optimal locations of multiple sink nodes
Assume there are N sink nodes k1, k2, Λ, kNfor a sensor network. The lifetime of a sensor node i which is assigned to the sink node kjis .
The lifetime of a sensor node i which is assigned to the nearest sink node is max , j=1, 2, Λ, Ν.
The lifetime of the sensor network is defined as the time till the first sensor node in the sensor network running out of battery, i.e. .
In order to maximise the lifetime of the sensor network we need to maximise the shortest lifetime of all the sensor nodes by optimising the locations of the multiple sink nodes. Therefore the objective of optimal locations of mul-tiple sink nodes is given as:
(2) The obstacles in the sensor field are prohibited to place any sink nodes, which can be introduced into the above optimal problem as a constraint, where Ê is the set of the obstacles
k1, k2, Λ, kN ∉ φ (3) Reformulating optimal location problem of sink nodes as a GA search problem
In general, in order to represent any problem as a GA search problem, it is necessary to define11:
l A chromosome representation of the problem which is amenable symbolic manipulation;
l A fitness function defined in terms of this representation; l A set of manipulation operators such as crossover, muta-tion, and reproduction.
The optimal variables in the objective function shown in Equation 2 are the coordinates of N sink nodes, which are N pairs of integer coordinates
(ia, ja), a=(1, 2, Λ, N) (4) The chromosome is expressed as an integer string with a length 2N (double the number of sink nodes) as below:
i1 j1 i2 j2 Λ Λ iN jN (5) where (ia, ja) ≠ (ib, jb), if a ≠ b (6) Equation 6 represents that two different sink nodes can-not be located in the same location.
The fitness function in the GA is chosen from the objec-tive function shown in Equation 2.
(7) The desired value of the fitness function will be the max-imum value of Equation 7 for all possible combinations of the N sink nodes. The three operators of GA, crossover, mutation, and reproduction are applied into the chromo-some shown in Equation 5 with satisfaction of the con-straints represented in Equations 3 and 6.
A simple simulation has been done in this study. Two sink nodes are designed for an 8x8 sensor network with three prohibited locations (1,1), (5,5), (6,6). In the initial stage, these two sink nodes are selected randomly, so GA starts with several chromosomes that describe a number of random solutions to the optimal location problem. The parameters of GA are chosen as the probability of mutation is 0.08, the probability of crossover 0.6, and the size of the population 20. The values of the parameters of the 8x8 sen-sor network are chosen as r=1bit/sec, e=0.62µJ/bit, e0=1.35J The search results illustrate that the upper left corner (0, 0) and the bottom right corner (7, 7) are the opti-mal locations for the two sink nodes. This result is consis-tent with the results in Wang et al’s work5for mobile sink nodes.
Conclusions
In this paper we only consider the optimal problem of the multiple sink node locations in accordance with a simplified routing algorithm. If any other routing algo-rithm is employed in the sensor network, the system model will be expected to be much complex. Nevertheless, the methodology approached and the formulations of the search problem are, in general, applicable to any routing algorithm. We are working on experimental illustrations for this approach. In the future work, more realistic assumptions which may consider more complex circumstances should be made.
Figure 6: Partition of a sensor field with two sink nodes.
j k i c e0 )} ( { min 0 j k i j i c e x ma
)}},
(
max
{
min
{
max
0 , , , 2 1 k kN i j ikj kc
e
z
L=
L i=0, ,1 L, )} ( max { min 0 j k i j i fitness c e f = (Continued on page 219) e0 kj ci Yang 21/8/06 11:01 am Page 4tions, to include monitoring of responder location through mobile motes, floorplan provision and hydrant location. Information infrastructure
SAFETYNet provides an information infrastructure to enable buildings, firefighters, fire tenders, and their control centre to efficiently communicate during natural or man-made disasters by using sensor networks, wireless com-munications, DAB (Digital Audio Broadcasting) and 3G technologies. The novel information infrastructure com-prises three layers, as shown on page 218.
The bottom layer comprises a robust wireless sensor work installed in and around the building. The sensor net-work utilises robust sensor ‘motes’ to detect any changes in the environment at any specified locations. The sensor network can take the place of existing fire alarm networks, meaning no previous installations are required. Information collected flows over the sensor network and is then transmitted to the fire tender network.
The middle layer comprises a vehicle-mounted mobile network installed on the fire tenders. It is achieved by upgrading the newly introduced vehicle mounted mobile data systems (VMDS) and adding not only the up-link with the control centre but also the down-link with the sensor network. The real-time information about the building, occupants, and locations of the fire fighters is collected from the sensor network, transmitted to and presented at the fire tender network. Up-to-date information about the building such as the floor plan and hydrant status is down-loaded from the central database located at the control cen-tre to the fire tender’s network on the way to an incident. DAB is employed between the bottom and middle layers in order to maintain a time-critical one-way communication channel between the fire tenders and emergency personnel.
At the top layer is the central facility located at the control centre of a fire brigade. An emergency response management system at the control centre will provide the fire-fighters with up-to-date critical information and remotely monitor the lat-est development of incidents. The national FireLink radio communication system for the fire and rescue services will be connected with the top layer in the information
infrastruc-ture. The connection with FireLink allows the real-time situ-ation in emergency situsitu-ations to be available nsitu-ationally. Potential impacts
There are two main beneficiaries of the SafetyNET sys-tem: firefighters and building occupants. Every year fire-fighters are killed or injured due to inadequate information about the nature of the emergency they are attending. The Safetynet system will help prevent some of the annual fire-fighter casualties in the UK. The majority of non-residen-tial fire injuries occur in high-occupancy buildings such as hotels, public administration buildings and retirement homes. In high occupancy buildings, the SafetyNET sys-tem will reduce the risk of death or serious injury for occu-pants. In the case of earthquake or explosion, information about structural integrity and occupant location will also help rescuers to safely and quickly locate survivors. Environmentally, tackling a fire more quickly and effec-tively reduces the production of poisonous fumes, CO2and particulates. There is also a reduction in materials use through refurbishing and rebuilding.
Contact details
The DTI-funded project “Secure Adhoc Fire & Emergency safety NETwork (SAFETYNET)” formally started on 1st July 2006, led by Data Cables Services International. The consor-tium includes Loughborough University as the academic part-ner, PERAas the R&D organisation, Jennic, Sure, Arqiva, and Electronics Line as the sensor manufacturers, and Association for Specialist Fire Protection (ASFP) as the end user adviser. Derbyshire Fire and Rescue Service is informally involved in the consortium as the end-user partner. The project budget size is around £1.3 million over three years and 50% funded by DTI and the rest contributed by the industrial partners.
For more information visit the project website www.FireSafetyNet.org.uk or contact Mr John Lynch, Director of Data Cables Services International on 020 8533 3038, or Dr Shaun Yang at the Computer Science Department in Loughborough University, on 01509-635670 or via email s.h.yang@lboro.ac.uk
References
1. I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “Wireless sensor networks: a survey”. Computer Networks, 38, 393-422, 2002.
2. I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci. “A survey on sensor networks”. IEEE Communications Magazine, August, 102-114, 2002.
3. H. Kim, Y. Seok, N. Choi, Y. Choi, and T Kwon. “Optimal multi-sink positioning and energy-efficient routing in wireless sensor networks”, Information Networking: convergence in broadband and mobile networking, Lecture Notes in Computer Science, 3391: 264-274, 2005.
4. S. R. Gandham, M. Dawande, R. Parkash, and S. Venkatesan. “Energy efficient schemes for wireless sensor networks with multiple mobile base stations”, in the Proceedings of IEEE Globecom, 1, 377-381, 2003.
5. Z. M. Wang, S. Basagni, E. Melachrinoudis, and C. Petrioli. “Exploiting sink mobility for maximizing sensor networks life-time”, in the Proceedings of the 38th Hawaii International
Conferences on System Sciences, 1-9, 2005.
6. C. Perkins, Ad Hoc Networks, Addison-Wesley, Reading, MA, 2000.
7. W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “Energy-efficient communication protocol for wireless micro sensor networks”, in the Proceedings of the 33rd Annual Hawaii International Conference on System Sciences, 305-310, 2000. 8. E. I. Oyman and C. Ersoy. “Multiple sink network design prob-lem in large scale wireless sensor networks”, in the Proceedings of the International Conference on Communications, Paris, France, June, 2004.
9. L. Qiu, R. Chandra, K. Jain, and M. Mahdian. “Optimising the placement of Integration points in multi-hop wireless networks”, in the Proceedings of the International Conference on Network Protocols, Berlin, Germany, October, 2004.
10. M. Younis, M. Bangad, and K. Akkaya. “Base station reposi-tioning for optimized performance of sensor networks”, in the Proceedings of the Vehicular Technology Conference, Orlando, Florida, October, 2003.
11. L. Yang, B. F. Jones, S. H. Yang. “Genetic algorithm based software integration with minimum software risk”, Information and Software Technology, 48, 133-141, 2006.
