WO2008141719A1

WO2008141719A1 - Energy-driven cluster aggregation point re-election in a wireless sensor network

Info

Publication number: WO2008141719A1
Application number: PCT/EP2008/003480
Authority: WO
Inventors: Antonio Ruzzelli; Raja Jurdak; Gregory O'hare
Original assignee: University College Dublin, National University Of Ireland, Dublin
Priority date: 2007-05-21
Filing date: 2008-04-29
Publication date: 2008-11-27

Abstract

A sensor network comprises a plurality of nodes arranged to communicate wirelessly with one another, at least some of the nodes being arranged to operate as an aggregation point for the network in which the aggregation point communicates data between other nodes of the network and externally of the network. Potential aggregation points are arranged to: at some time after the node becomes an aggregation point, determine an energy state for the aggregation point; responsive to the energy state satisfying a criterion, transmit to adjacent nodes of the network a request to change aggregation point; receive from the adjacent nodes at least one energy state indicator for at least one other node in the network; determine in accordance with the received energy state indicators whether one of the other nodes should become the next aggregation point for the network; and responsive to a positive such determination, transmit to the one of the other nodes an indicator that the node is to become the next aggregation point for the network.

Description

ENERGY-DRIVEN CLUSTER AGGREGATION POINT RE-ELECTION IN A WIRELESS SENSOR NETWORK

Field of the Invention

The present invention relates to a cluster aggregation point and a method for load balancing in a sensor network.

Background of the Invention

Wireless sensor network applications typically rely upon the forwarding of data from network nodes to a node, entitled a cluster aggregation point (AP)₁ which collects, filters, processes, aggregates, and relays the relevant information beyond the local network as required. Typically, a sensor AP performs power intensive tasks such as in-network processing and longer-range transmissions for reaching other clusters, external networks, a base station or a personal computer (PC). This leads to asymmetry in node energy consumption due to (1) higher forwarding activity for nodes in the vicinity of the AP and (2) higher AP activity with respect to the rest of the nodes. AP consumption of more energy than the rest of the sensors potentially leads to premature network disconnection.

By way of example, sensor-based medical systems comprise (1) body-sensor nodes, located on patient(s), and (2) ambient-sensor nodes deployed in the environment. To collect data from the patient, the body-sensor nodes form a small tree-based network and nominate a node as AP to serve as a gateway and to communicate to the ambient sensors.

Rather than relaying all data, the AP performs data aggregation to relay only useful information to one or more nodes in the vicinity of the patient. The ambient sensor nodes eventually forward the data to the doctor for analysis. The AP in sensor-based medical systems is more active than any other body-sensor node, resulting in a higher energy consumption rate at the AP. Nodes closer to the AP also consume more energy than the average network node, as they have to forward more packets than nodes located further away. This imbalance in energy consumption rate causes the AP and nodes in its vicinity to deplete their batteries sooner than other nodes, compromising network connectivity.

Another application that shares the same problem is a ground-based sensor network communicating with an overhead unmanned aerial vehicle (UAV) is described in P. Vincent, M.Tummala, and J.McEachen "A new method of distributing power usage across a sensor network" IEEE Communication Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks (SECON 2006), Reston, VA, USA, 2006. Such an application requires some nodes to communicate with the flying device therefore consuming considerably more energy than other sensors.

Performing periodic AP reassignment can address energy consumption imbalance. However, the reassignment process involves a number of issues, such as how often the AP reassignment occurs, how to guarantee an autonomous and reliable handover, and upon which metrics to base the new AP nomination.

Some recent work has attempted to balance network energy consumption and prolong network lifetime through a regular reassignment of the AP.

Cluster head Gateway Switching Routing (CGSR) as disclosed in P. Chowdavarapu and V. Puram, "An overview of routing protocols in ad-hoc networks", Technical Report CS-90, University of Kentucky Laboratory for Advanced Networking, December 2000 and Cluster Based Routing Protocol (CBRP) as disclosed in M. Jiang, J. Li, and Y. Tay "Cluster based routing protocol" Functional Specification Internet Draft, work in progress, June, 1999 exemplify a class of protocols which initially selects a AP according to lowest ID or highest connectivity. In order to avoid control overhead for frequent cluster leader changes, subsequent changes in AP for this class of protocols are only triggered for highly mobile situations where two APs move within range of each other or when node mobility causes certain nodes to wander out of the coverage areas of all APs. Another class of protocols, typified by Low Energy Adaptive Clustering Hierarchy (LEACH) discussed in W. Heinzelmann, A. Chandrakasan, and H. Balakrishnan "Energy-efficient communication protocol for wireless microsensor networks", In Proc. HICSS, 2000 and Power Efficient Gathering in Sensor Information Systems (PEGASIS) discussed in S. Lindsey and C. Raghavendra "Pegasis: Power-efficient gathering in sensor information systems", In Proc. ICC, 2001 attempt to balance energy consumption among a homogeneous set of sensor nodes through a periodic random or geographic rotation of the AP. This does not necessarily reduce network forwarding load.

Vincent et al mentioned above also adopts a periodic random rotation policy for a hierarchical sensor network that includes ground-based sensor nodes and mobile unmanned aerial vehicles (UAV). The ground-based nodes collect field data at a designated transmit cluster of nodes, which is responsible for aggregating and forwarding the network data to the UAV when it flies over the region. To avoid high energy consumption at the transmit cluster, Vincent et al propose the periodic random rotation of the transmit cluster.

Disclosure of the Invention

The present invention enables an AP to be re-assigned within a sensor network in order to balance energy consumption and extend the operative network longevity.

Preferably, re-assignment occurs either when the AP or when another sensor node experiences a significant energy depletion. The decision is made by means of a cost function that combines energy state parameters polled from the network. Preferably, the cost function combines node energy metrics that relate to battery level, forwarding activity, and data generation.

Nodes use the cost function to determine the best choice for the next AP through a set of control messages, resulting in a well-informed selection of the AP. Initial simulation results, confirm that preferred embodiments of the invention can balance energy and prolong the lifetime of a network by up to 50% for applications that require power-intensive tasks at the AP and for high traffic applications.

Although the invention is independent of any particular communication protocol, the invention is preferably implemented with the recent MERLIN protocol discussed in A.G. Ruzzelli, G. O'Hare, and R. Jurdak, "MERLIN: Cross-layer Integration of MAC and Routing for Low Duty-Cycle Sensor Networks," Special Issue of Elsevier Ad Hoc Networks Journal. February 2008, due to its low end-to-end delay and energy consumption features, and its provision of both medium access control (MAC) and routing support.

Preferably, the invention involves infrequent AP changes and limits these changes to situations where one node has consumed a significant portion of its energy resources.

By contrast with the prior art, instead of using AP rotation, the present invention employs energy state information from the nodes to make an informed selection of the next AP, which would balance network load.

Brief Description of the Drawings

An embodiment of the invention will now be described, by way of example, with reference to the accompanying drawings, in which:

Figure 1 illustrates AP re-assignment according to a preferred embodiment of the invention;

Figure 2 contrasts network lifetimes for different data rates for the embodiment with a single AP implementation; and Figure 3 contrasts network lifetimes for different AP transmission power for the embodiment with a single AP implementation, with a random periodic AP rotation implementation and with an AP rotation implementation based on a cost function. Description of the Preferred Embodiment

According to the preferred embodiment of the invention, each node of a wireless network which can become an AP, implements a protocol which comprises a series of control message for determining the identity of the next AP.

Referring now to Figure 1(a), which shows a network of sensors in operation, with one node acting as AP. In the embodiment, two events may trigger the selection of a new AP: (1) a significant energy depletion at the current AP; or (2) a significant energy depletion at any other node.

Although not limited to a particular MAC protocol, the embodiment has been tested with the recent MERLIN protocol mentioned above. MERLIN divides the network into time zones 1...3, which represent the number of links to the AP. Time zones streamline the flow of packets to and from the AP. Should the AP determine a significant energy depletion Ed in its energy resources with respect to an initial battery level, it broadcasts a State Request (S REQ) message to all the nodes in the cluster, Figure 1(b), and it sets a timer Tind whose duration is sufficiently long to accommodate responses from all nodes in the cluster. Upon receiving a S REQ message, Figure 1(c), a node N replies with an upstream State Indication (S IND) message that includes a timestamp and N's energy state (quantified by a cost function F described in more detail later). Preferably, the energy state parameters are expressed in consumption rate. Together with the timestamp, these parameters enable the node to account for the time lag between the transmission of S IND and its receipt at any other node.

Node N also forwards the S REQ message to downstream nodes that are further away from the AP.

The protocol filters upstream S IND messages in order to limit the control messaging overhead as follows: Every node keeps track of the node with the highest energy figure FMAX appearing in any S IND message in the current cluster reassignment process. Upon receiving a new S IND message, a node compares F in the new S IND message. If F > FMAX₁ then the node forwards the incoming S IND message and sets F to FMAX. Otherwise, the incoming S IND is discarded. Eventually, one or more S IND messages will have reached the AP before the expiry of the timer Tind. The current AP then selects the node with the highest figure F among all the nodes whose S IND was received at the AP, provided that the selected node has a higher F than the current AP. The AP then sends a Handover Request (H REQ) message to the AP designate node, Figure 1(d). Upon receiving H REQ, the AP designate node broadcasts a Handover Confirm (H CONF) message, Figure 1 (e), including its own ID and a time Tc at which the AP handover will take place. All nodes forward the H CONF message, including the current AP, to ensure that all nodes will re- synchronise to the new AP at time Tc.

At time Tc, all nodes reset their states and determine their route (timezone in the case of a MERLIN implementation) and synchronisation information relative to the new AP, Figure 1(f), so completing the handover process.

In certain cases, such as high frequency sampling applications, non-AP nodes may deplete their energy resources more quickly than APs. Thus in certain embodiments, there is provided an override function that enables nodes with high energy consumption to notify the AP of their state. Any node N that detects significant energy depletion in its energy resources can proactively send an S IND message to the AP. Such unsolicited S IND messages are always forwarded upstream by intermediate nodes towards the AP, unless a current handover process is already in progress. Once the S IND message reaches the AP, the latter compares the node N's energy figure FN to its own. If the AP determines that handing over its AP ability to node N will have a favourable impact on Ns energy profile, then the AP initiates a handover process by sending H REQ to node N. Node N then responds by broadcasting a H CONF message, as in the normal handover case. If the AP determines that handing over AP ability to node N wϋ! have adverse effects on the energy consumption of node N or other critical nodes, then the AP discards N's S IND message and continues normal operation.

It will be seen that when the time Tc arrives, all the nodes simultaneously switch to the new AP. Packets travelling across the network are seamlessly routed to the new destination, through reassignment of sequence number in each control packet to allow the detection of stale packets.

Preferred embodiments also employ acknowledgements of all packets to reduce the probability of loss. Should a communication failure happen (e.g. a failure of H CONF reception), a node could still re-synchronise to the new AP by requesting a H UPDATE packet from the neighbouring nodes. The node then synchronizes according to the H UPDATE packet payload.

As mentioned above, S IND packets from network nodes contain all the information required to compute the cost function at the AP. The objective of the cost function is to calculate an AP re-assignment figure F for a particular node. Before presenting the quantitative description of the equation for F, we identify the relevant components that characterize a node's energy consumption.

A node's energy consumption is primarily due to the transceiver communication activity (Ec), the processing activity (Ep) and the sensing activity (Es). In general, Ep is negligible relative to Ec and Es. Furthermore, Ec at a node comprises the energy for transmitting the node's own generated data, Et, and the energy consumed in forwarding other nodes' data, Ef. For the AP, Ef is substituted with Eext that represents the energy consumed to transmit to an external network (e.g. wifi), where Eext involves a higher transmission power to reach the external network. Finally, we define the data generation energy Eg as the sum of the energies spent in sensing and transmitting such data, Eg = Es +Et. The seieciion of the next AP favours nodes with a higher value of the cost function F. Assume B as the initial battery level of a node and Br(i) the battery level of a node at the beginning of an i-th AP leadership reassignment.

1. The battery level of a node at time t, B(t), normalized relative the initial battery level B, positively affects the reassignment cost function;

2. Ef varies every time a reassignment procedure takes place. The procedure entails rebuilding the routing tree to route packets towards the new AP. Assigning the AP to a node with high Ef would reduce the overall forwarding activity of the network. Therefore, Ef also accounts positively in the cost function and it is normalized relative to Br(i);

3. Eg, which includes Et and Es, is affected by the communication protocol and the AP vicinity. For example, in CSMA techniques the closer the node is to the AP the higher the competition to access the channel for transmission. The overall network forwarding activity can be reduced by assigning the AP as the node with the highest Eg. Hence, Eg also accounts positively in the cost function, and it is normalized relative to Br(i);

4. The reassignment process assumes a request, an indication, and a confirmation to and from all nodes in the network before an AP handover takes place. Therefore, assigning the AP leadership to an adjacent node does not necessarily save more energy than assigning the AP to a far node.

As a result of such considerations, the cost figure F is based on a weighted sum of (1) the total battery level B(t) at time t normalized to the initial battery level B; (2) the forwarding energy Ef and the external energy (Eext), and (3) the sensing energy Es and transmitting energy Et normalized to Br(i) as follows:

F(t) = α[B(t)/B] + β[Ef(t)/Br(i)]+ γ[(Es + Et)/Br(i)] + δ[Eext/Br(i)] where Eext≠O and Ef = O for the AP; and furthermore, α+β+γ = 1 ; γ= δ.

The weights can be tuned in accordance with application requirements, network topology and user's choice of either privileging node balancing or single node lifetime (e.g. a bigger α corresponds to a more energy greedy node).

An initial evaluation of the above embodiment uses the object oriented OmNet++ network simulator discussed in A. Varga "The OMNet++ discrete event simulation system" In European Simulation Multiconference, Prague, Czech Republic, June 2002. http://www.omnetpp.org. The sensor network framework is based on templates disclosed in E. WSN "The codesign project" University of Twente, The Netherlands, 2003 (http://wwwhome.cs.utwente.nl/~dulman/codesign/) and E. WSN "The eu eyes project", European Commission, 2003 (http://www.eves.eu.org/). 4.1. Performance Metrics and Setup

The main objective of the embodiment is to provide longer sensor operative lifespan through balancing the energy consumption across the network. Therefore, an initial assessment of the embodiment is obtained considering the network lifetime.

Although the definition of network lifetime varies according to network disconnection and application requirements, we define network lifetime as the time when the first node in the network is depleted.

We now describe relevant parameters for the simulation setup. The significant energy depletion, Ed, is the percentage of energy decrease with respect to the full battery and it triggers the reassignment procedure. Tuning Ed changes the control overhead of the protocol. The value of Ed depends on the network topology, the node data rate and the number of nodes within the network. The simulations here consider a value of 10% for Ed. The AP aggregation factor, Ag, specifies the portion of incoming sensor data that is relayed by the AP. The current evaluation considers a value of 40% for Ag. The external transmitting power ratio Eext/Et is a tuneable parameter that represents the ratio between the transmission power to relay data to the external network and the node transmission power.

The simulations use the IEEE 802.15.4 standard for the physical and MAC layers and the transceiver parameters from the Chipcon transceiver CC2420 data sheet (C. AS. CC2420 datasheet. Technical report, Chipcon AS, Oslo, Norway, 2005). The rest of the protocol parameters are in line with those disclosed by Ruzzelli et al.

The embodiment has been evaluated for a series of multihop random networks of 30 nodes obtained by 10 different seeds for each run. For the present evaluation we set _α = β = γ = o.33 so that all factors of the cost function present the same weights.

Figure 2 shows the network lifetime achieved by varying the average node data rate. The results clearly show an extension of the operative network lifetime from 20% to about 40% with respect to a single AP. In fact as the average node data rate increases, the AP is exposed to a higher load. Therefore reassigning the AP has a greater impact on the network lifetime.

In communication theory, the transmission power is defined as the energy spent in transmitting a signal in a time unit, e.g. Pt = Et/T, where T is equal to a unit of time and Pt is the transmission power. Figure 3 shows the network longevity benefits of the embodiment relative to the single AP case as the ratio of external AP transmitting power Pext and sensor transmitting power Pt varies between about 11 and 40.

Recalling that the transmitting power (Pt) is affected exponentially by the distance, i.e. Pt ~ D^α, 2 < α < 4 in air, the range of ratios corresponds to an external transmitting distance of AP that is between about 3.3 and 6.2 times the sensor transmitting distance. Such a range reflects real application scenarios. The results show an improvement of the operative network lifetime with respect to the static AP that ranges from about 9% to 47% and with respect to the random periodic AP rotation that ranges from about 22% to 40%. Thus, as expected, the embodiment is more effective when the system presents a large external transmitting power ratio. In fact_; a higher ratio results in a much higher depletion ratθ of the AP than other sensors.

It will therefore be seen that the embodiment provides for energy/load balancing of wireless sensor networks and enables the reassignment of the cluster aggregation point according to a node cost function. The embodiment provides a flexible technique that provides several tunable parameters, such as significant energy depletion Ed, aggregation factor Ag, network size and network topology, for tailoring network behaviour according to application requirements. Investigating the impact of two such parameters, the simulation results have shown that network lifetime benefits over the static AP case increase greatly with respect to both the data rate and the external transmission power ratio.

As mentioned above, the present invention finds particular application in sensor- based medical systems comprising body-sensor nodes, located on patients, and ambient-sensor nodes deployed in the environment. To collect data from the patient, the body-sensor nodes form a small tree-based network and nominate a node as an AP to serve as a gateway and to communicate to the ambient sensors. Preferably, rather than relaying all data, the AP performs data aggregation to relay only useful information to one or more nodes in the vicinity of the patient. The ambient sensor nodes eventually forward the data to a remote computing device operated by a doctor for analysis. The AP in sensor-based medical systems is more active than any other body-sensor node, resulting in a higher energy consumption rate at the AP. Nodes closer to the AP also consume more energy than the average network node, as they have to forward more packets than nodes located further away. This imbalance in energy consumption rate can cause the AP and nodes in its vicinity to deplete their batteries sooner than other nodes, and the present invention mitigates these problems in such networks.

Also, the invention finds application in a ground-based sensor network communicating with an overhead unmanned aerial vehicle (UAV). Such an application requires some nodes to communicate with the flying device therefore consuming considerably more energy than other sensors Re-assignment of AP involves a number of issues, such as how often the AP reassignment occurs, how to guarantee an autonomous and reliable handover, and upon which metrics to base the new AP. Again, the present invention addresses these problems in such networks.

Claims

Claims:

1. A method operable in a node of a sensor network comprising a plurality of nodes arranged to communicate wirelessly with one another, at least some of said plurality of nodes being arranged to operate as an aggregation point for said network in which said aggregation point communicates data between other nodes of said sensor network and externally of said sensor network, said method comprising: a) at some time after said node becomes an aggregation point, determining an energy state for said aggregation point; b) responsive to said energy state satisfying a criterion, transmitting to adjacent nodes of said network a request to change aggregation point; c) receiving from said adjacent nodes at least one energy state indicator for at least one other node in said sensor network; d) determining in accordance with said received energy state indicators whether one of said other nodes should become the next aggregation point for said sensor network; and e) responsive to a positive such determination, transmitting to said one of said other nodes an indicator that said node is to become the next aggregation point for said sensor network.

2. A method according to claim 1 further comprising: f) while said node is not an aggregation point, receiving a request to change aggregation point from a first adjacent node of said wireless network; g) forwarding said request to change aggregation point to other adjacent nodes of said network; h) forwarding to said first adjacent node an indicator of the energy state of said node; and i) responsive to receipt of energy state indicators from said other adjacent nodes, forwarding said energy state indicators to said first adjacent node.

3. A method according to claim 2 wherein said step of forwarding i) comprises: for each energy state indicator received from other adjacent nodes: responsive to said indicator being greater than an energy state indicator stored at said node, storing said energy state indicator at said and forwarding said energy state indicator to said first adjacent node; and responsive to said indicator not being greater than an energy state indicator stored at said node, discarding said energy state indicator.

4. A method according to claim 1 further comprising: j) while said node is not an aggregation point, determining an energy state for said node; k) responsive to said energy state satisfying a criterion, transmitting to adjacent nodes of said network a request for said node to become the next aggregation point;

I) responsive to receipt of a positive response to said request received from an adjacent node, setting said node as the next aggregation point for said sensor network.

5. A method according to claim 1 wherein said energy state is a function F(t) of one or more of: remaining battery power (B(t)); energy (Ef, Et) required to communicate data between said node and adjacent nodes; energy (Es) required to sense data at said node; and energy (Eext) required to communicate data externally of said sensor network.

6. A method according to claim 5 wherein F(t) is normalised relative to battery level.

7. A method according to claim 6 wherein F(t) is: F(t) = α[B(t)/B] + β[Ef(t)/Br(i)]+ γ[(Es + Et)/Br(i)] + δ[Eext/Br(i)] where in an aggregation point Eext≠O and Ef = 0; B is initial battery levs!; B(t) is battery at time t; Br(i) is battery level at the beginning of a request to change aggregation point; and wherein α,β,γ and δ are weighting factors.

8. A method according to claim 7 wherein: α+β+γ = 1 ; and γ= δ.

9. A method according to claim 5 wherein the node having the highest energy state is determined to be the next aggregation point in said sensor network.

10. A method according to claim 1 further comprising: m) while said node is not an aggregation point and responsive to receipt of said indicator that said node is to become the next aggregation point, setting said node as the next aggregation point for said sensor network.

11. A method according to claim 4 or 10 wherein said setting said node as the next aggregation point for said sensor network comprises transmitting to adjacent nodes confirmation of when said node is to become said aggregation point.

12. A method according to claim 11 comprising: n) while said node is not an aggregation point and responsive to receiving confirmation that a node is to become the next aggregation point from an adjacent node, transmitting said confirmation to other adjacent nodes.

13. A node in a sensor network comprising a plurality of nodes arranged to communicate wirelessly with one another, at least some of said plurality of nodes being arranged to operate as an aggregation point for said network in which said aggregation point communicates data between other nodes of said sensor network and externally of said sensor network, said node being arranged to: a) at some time after said node becomes an aggregation point, determine an energy state for said aggregation point; b) responsive to said energy state satisfying a criterion, transmit to adjacent nodes of said network a request to change aggregation point; c) receive from said adjacent nodes at least one energy state indicator for at least one other node in said sensor network; d) determine in accordance with said received energy state indicators whether one of said other nodes should become the next aggregation point for said sensor network; and e) responsive to a positive such determination, transmit to said one of said other nodes an indicator that said node is to become the next aggregation point for said sensor network.

14. A sensor network comprising a plurality of nodes according to claim 13, at least some of said nodes being arranged to be mounted about the body of a patient and to sense respective attributes of said patient, said aggregation point being arranged to relay data corresponding to said attributes externally of said sensor network.

15. A sensor network comprising a plurality of nodes according to claim 13, at least some of said nodes being arranged to be located in an environment and to sense respective attributes of said environment, said aggregation point being arranged to relay data corresponding to said attributes externally of said sensor network.