SDAM: A New data Aggregation Approach for Wireless Sensor NetworksHang Qin Software Engineering State KeyLaboratoryWuhan UniversityWuhan 430072, Chinahangqin100@Hui WangSoftware Engineering State KeyLaboratoryWuhan UniversityWuhan 430072, ChinaHuaibei ZhouInternational School of SoftwareWuhan UniversityWuhan 430072, ChinaAbstract—In this paper, a novel scalable scheme, Scalable Data Aggregation Monitoring (SDAM), is proposed f or ef f icient datagathering with aggregation (fusion) in Wireless Sensor Networks (WSN). Di f f erent f rom existing schemes, SDAM not only optimizes data transmission cost, but also incorporates the unction of data f usion, which is crucial f or emerging sensor networks with data management and high availability requirements. Employing a randomized algorithm that allows f usion points to be chosen by nodes’ data amount, SDAM can achieve high availability or sensor nodes, which results the optimal solution f or nowadays system setup. Simulation results demonstrate that SDAM scheme can re lect the node status promptly, and save the network throughput of sensor nodes. Therefore, the lifetime of WSN is significantly extended.Keywords-sensor networks, data aggregation, zone monitoring, timestampI.I NTRODUCTIONWSN is an accumulation of sensors interconnected by wireless communication channels. Under the control of the network, every sensor node is a small device that can collect data from surrounding area, communicate with each other, and carry out computation. Long distance communication is normally achieved in a multi-hop manner. Thanks to recent advances in remote monitoring system, such networks are progressing rapidly, and are expected to be popular in applications such as environment monitoring, intrusion detection, and earthquake warning.Nonetheless, existing strategies miss one vital measures in the optimization space for routing correlated data, namely, the data aggregation cost, which may not be negligible for certain applications. As the result, WSN monitoring field parameters need to seek simple functions (for instance, average, maximum, or minimum) with cost-effectiveness. Meanwhile other sensor nodes may require complex operations for data aggregation. In terms of network availability for monitoring algorithm, the sensor data fusion has been shown in the same order of data communication. Partitioning and recombining information are demonstrated as two aggregation process examples in experiment.II.B ACKGROUND AND R ELATED W ORK WSN have acquired much attention because of the potential application of the technology. Numerous data communication protocols have been proposed lately, such as: EPAS (Energy-Efficient Protocol for Aggregator Selection ) [1], TAG (Tiny AGgregation)[2], TTDD (Two-Tier Data Dissemination)[3], GRAB(GRAdient Broadcast)[4], MF ST (Minimum F usion Steiner Tree)[5]. Three types of data collections in WSN are discussed: (1)Event-based data, such as intrusion detection or object tracking, are accumulated when an event at a particular locale within the deployment region comes. The event is strengthened by detecting sensors with local agreement, and presented to the control authority.(2)Global state-based data, such as temperature or humidity, are accumulated by sensor nodes all over the deployment area and transmitted to the sink. (3)Focused state-based data is accumulated in response to a query sent to selected sensors requesting relevant data. Our interest here is in focused state-based data.III.S YSTEM M ODEL AND SDAM D ESIGNA.Problem Formulation and Sensor Node’s StateTransformationIn this paper, aggregation organizer determines the status of node machine from the data packet content and interval sent out by sensor node[6][7], and the conceivable node machine status may be: Down,Shift, Activate and Pause. When the data packet of node machine is received by aggregating mechanism, the aggregation organizer can search a new buffer region in the shared memory area to save this node’s information. Down is aggregation organizer has not received the data packet after waiting for a lot of time. At this time sensor node is closed down or has a fault, and aggregation organizer will inform this status to the administrator. Shift shows sensor node has been closed down for a long time, and aggregation organizer can shift this node machine from theshared memory area for the saving of shared memory area. Activate defines sensor node can work normally and aggregation organizer can activate sensor node’s previous data. Pause describes aggregation organizer can not receive sensor node’s data packet in the certain time, it represents sensor node is busy, fault or failure, and aggregation organizer can wait for the data packet continuously. Figure 1 depicts theaggregation states model of the above states:Figure 1. Aggregation States Model of Sensor NodeB.Finite Automata of Sensor Node’s State TransitionGive for the formal approach of sensor network nodes, the node machine’s states transformation diagram is enhanced in this section. In Figure 2, the start status and the terminal status are added. Ready depicts the initialization of aggregation organizer, double circles show the final state in this status diagram and one state transformation procedure of node machine. Thus the Deterministic Finite Automata (DF A) [6] can be constructed from sensor nodes’ state transformation diagram:Figure2. Transition diagram of DFA to node via aggregationDef inition 3.1DFA node =(T, Q, δ, F, S ). T = {p, d, s, r }, pdefines having not received data packet in t p senonds, d stands for having not received data packet in t d seconds, r depicts having not received data packet in t s seconds, r shows receive one data packet. Q is the state set, Q = {R, S , D, P, A }, R stands for the ready status, S stands for shift, D stands for Down, P stands for pause, A stands for activate. And T is transaction set. δstands for a state transformation function:Q ×T → Q , and the state transformation formula can be showed in Table 1. F is the final states set, F = {S }, and F ⊆Q .Table 1. the sensor nodes’ transformation aggregation rule r p d s R A S A D A S P A D A A PThe sensor nodes’ one action set from the above equation set can be got, this action depicts the process from start state to failure state, or any states from the start status to the shutting down status. And this set can be described by the regular expression: R node =r(r + pr + pdr + pdsr)* pds . Any sentence produced by the regular expression R node is one action of node machine, such as rpds depicts the failure for some reason after the sensor nodes send one information packet to the monitoring mechanism. Therefore finite automata DFA node and regular expression R node provide the formal description for the research of sensor nodes’ state transformation.C.SDAM Algorithm DesignIn this Section, the data aggregation algorithm is proposed as below.Epoch 1COUNT packet is the number of 8K data packets, LENGTH total is the overall length of the data to be sent, LENGTH data is the length of data to be hold in one packet, MOD is the arithmetic operation, << is the shift left operation, packet is one data packet, TS is timestamp, and sequence is the current sending packet serial number. Whereas LENGTH total MOD LENGTH data = 0, all the data can be hold in the COUNT packet data packets:1.COUNT packet ←LENGTH total / LENGTH data ,TS ←current system time, LENGTH data ← 8K – the length of information packet header, id ← 0;2.if LENGTH total MOD LENGTH data = 0, then cover← (1 << COUNT packet ) – 1, else cover ← (1 << (COUNT packet + 1)) – 1;3.if COUNT packet = 0, then goto 7;4.run distribute( ), packet_cover ← cover, packet_id← 1 << id, packet_timestamp ←TS ,LEGTH packet ← 8K5.copy LENGTH data data to the information message atthe beginning of data’s id ×LENGTH data offset address, send information packet;6.id ← id + 1, if id = COUNT packet ,then goto 4;7.if LENGTH total MOD LENGTH data = 0, then goto 10;8.packet_cover ←cover, packet_id ← id,packet_timestamp ←TS ,LEGTH packet ←LENGTH total MOD LENGTH data + the length of information packet;9.copy LENGTH data data to the information message atthe beginning of data’s id ×LENGTH data offset address, send information packet;10.return .Epoch 2:packet is the information packets received by aggregation organizer, TS is the timestamp when aggregation organizer receives the former information packet, and the initialization is TS ← 0.Aggregation organizer allocates 256K buffer region to save the received data, LENGTH data is he length of data to be hold in one packet, | is the arithmetic OR operation. The current packet will be discarded when it is older than the former packet:1.if packet_timestamp < TS ,goto 5;2.copy the packet data with the length of packet.LENGTH data to the buffer region at the position of corresponding offset address;3.run election( ), if packet_timestamp > TS ,thencover ← 0, TS ←packet_timestamp; 4.cover ← cover | packet_id;5.if cover = packet_cover, then clean up theinformation in the buffer region, and save it in the shared memory area; 6.return .IV.T RANSMISSION AND P ACKET S ALVAGING A NALYSIS OFSDMAPerformance of this sensor monitoring system can be evaluated from two aspects: the response time and the sensor network bandwidth in occupancy. In order to describe the performance of remote monitoring system more accurately, it is modeled as: Definition 4.1 Node machine performance 4-tuple PF= (F com ,V max , TP, LT ), F com (Communication F requency) is the bandwidth of communication, which is the number of sensor nodes’ request from monitoring node to sensor nodes per second, V max is the transmission speed maximum of node machine sensor network device, TP (Throughput Percentage) describes the percentage of sensor network bandwidth in theresponding process of node machine, LT (Latency Time)defines the monitoring request time of node machine responsemonitoring time. Therefore the responding time LT is the sumof sensor network communication time and the sensor nodes data aggregation time, and LT = T agg +T com . Sensor network throughput TP is closely related with the communicationquantity, communication frequency and the maximum sensor network transmission ratio of sensor nodes and monitoring node, and as is show in formula (1), where S cmd and S data are the size of command packet and information packetrespectively:max()100%com cmd data F S S TP V +=× (1)De inition 4.2 Sensor aggregation monitoring performancemodel SDAM PF = (M, E, N, ATP, ALT ), and M (Medium)represents what wireless network medium the remotemonitoring system runs, M ∈{L, X, P }, L (WLAN) is thewireless local area network, X (WiMAX) is the wirelessmetropolitan area network or the Internet, P (WPAN) is thewireless personal area networks. E (Environment) describes thehardware and soft configuration environment of the whole sensor networks. Hardware environment includes the configuration of sensor nodes, monitoring nodes and wireless network connectivity, such as the number of sensor nodes, the transmission ratio of wireless network communication and the capacity of data storage device. Software environment includes the version of operating system, communication protocol and so on. N (Nodes) is the set of performance 4-tuple in the monitored sensor nodes, thus N={x | x=PF }. ATP (Average Throughput Percentage) defines the network bandwidth percentage of monitoring system in the whole wireless sensor networks. ALT (Average Latency Time) describes the average responding time of sensor nodes.max cmd datacom n Nn NS S F n TPnV ATP NN∈∈+××==¦¦ (2)()comagg n Nn Nn LT n TT ALT NN∈∈××+==¦¦ (3)As far as the average latency time ALT is concerned, it is responsible for the N ,T com and T agg in every sensor nodes. With the increasing of sensor nodes number ΊN Ί,T com inevery node is becoming larger and larger. It is because monitoring node will process more information packet from the sensor nodes, while the sensor nodes’ processing time T agg is invariable. With the increasing of sensor nodes’ number, the average responding time will decrease. As the percentage ATP of the monitoring system in the whole wireless sensor network, it is closely related with the number of sensor node VΊN Ί, the size of command packet S cmd and information packet S data, the transmission speedmaximum of wireless network device V max, and thecommunication frequency F combetween the monitoring nodesand sensor nodes. When the number of sensor nodes isincreasing, ATP is invariable practically, although theresponding time in the sensor nodes is becoming larger. It isbecause S cmd ,S data ,V max and F com are invariable. At the sametime, ATP is invariable practically when the overload of the sensor nodes system becomes larger. Due to the invariability of S cmd and S data , the change of ATP depends on F com and V max .V.S IMULATION S TUDY To verify the algorithm introduced above, a simulationprogram is run and shows following results.Suppose the surveillance region have 2500 square meters with a foursquare shape. Comparing the results of two different groups shown in Figure 3 and Figure 4, the overheadof WSN increase slowly. Moreover, if increasing theprobability of edge fails, the overhead decreases on thecontrary.When the system overload as the sensor nodes increase,the processing time T agg is becoming larger, and T com isbecoming larger even if these overloads focus on the wirelessnetwork communication. Otherwise T com will has a littlevariation, triggered by the increasing of sensor nodes overload. Therefore the average responding time becomes larger when the system overload is increasing.Number of Sensor NodesL a t e n c y T i m e (m S e c )Figure 3. Data aggregation and Latency101010101010Packet CountL a t e n c y T i m e (m S e c )Figure 4. Transmission Cost and LatencyIt is obviously the ALT is increasing with the increase of F com and the decrease of V max . This provides an approach for the administrator to reduce ALT when the wireless network overload is increasing. We can not only configure the sensornetworks monitoring system for the increase of collecting information interval and the decrease of F com , but also deploythe faster sensor networks device for the increase of V max .VI.C ONCLUSIONIn this paper, a scalable data fusion server based on WSN architecture is proposed. Techniques such as parallel data storing and distribution policy, adaptive timestamp data aggregation algorithm, are discussed extensively. Simulation results based on existing data aggregation monitoring system demonstrate the expected advantages. Also, the results will lead to further evaluation of these techniques later. In the future, the impacts on the performance of the data fusion system will be explored in details, especially it is induced by communication operation reduction. The principle of dynamic transmitting rate at aggregator to reduce the user perceived latency needs further study in depth.R EFERENCES[1] Yuanzhu, Peter Chen, and Arthur L. Liestman, “A Hierarchical Energy-Efficient F ramework for Data Aggregation in Wireless Sensor Networks”, IEEE Transactions on vehicular technology, Vol. 55, No. 3, May 2006, 789–796[2] S. Madden, M. J. Franklin, J. M. Hellerstein, and W. Hong, “TAG: A tinyaggregation service for ad-hoc sensor networks”, in Proc.5th S ymp. OSDI, Boston, MA, Dec. 2002.[3] F.Ye, H.Luo, J.Cheng, S.Lu and L.Zhang,“A two-tier data disseminationmodel for large-scale wireless sensor networks”, in Proc. MobiCom, Atlanta, GA, Sep. 2002[4] F. Ye, G. Zhong, S. Lu, and L. Zhang, “Robust data delivery protocol forlarge scale sensor networks”, in Proc. IEEE Int. Workshop IPS N, Palo Alto, CA, Apr. 2003.[5] Hong Luo, Yonghe Liu and Sajal K. Das, “Routing Correlated Data withFusion Cost in Wireless Sensor Networks”, IEEE Transactions on mobile computing , Vol. 5, No. 11, November 2006, pp. 1620–1632.[6] Harry R. Lewis and Christos H. Papadimitrious, Elements of the Theoryof Computation, Prentice Hall, Second Edition, 2006, pp.34–39.[7] Alberto Cerpa and Deborah Estrin, “ASCENT: Adaptive Self-ConfiguringSensor Networks Topologies”, in Proc. 11th Joint Conf on IEEE Computer and Communications Societies (INFOCOM), New York, NY, June 2002.。
