ENERGY EFFICIENT DATA DISSEMINATION IN WIRELESS SENSOR NETWORKS

Abstract

Wireless sensor networks (WSNs) have emerged as the recent wave of wireless technology which combines sensing, processing and communication through tiny low power and low cost Sensor Nodes (SNs). WSNs enable greatly distributed sensing across vast area through collaborating SNs. Due to very small size and disposable nature, SNs have various resource constraints. But, among all constraints, limited energy source onboard on a SN (in the form of fixed batteries) is the most challenging constraint that has created a major bottleneck in the design and deployment of WSNs. Due to this limited energy constraint, many of solutions prevalent in other ad hoc networks can not be used as such in WSNs. Since, in most of the scenarios it is not possible or economical to reclaim SNs for replenishing their energy sources, hence the only way to extend the operational lifetime of a WSN is to reduce the energy consumption while it is operational. Hence, the work presented in this thesis revolves around the central theme of reducing energy consumption within a WSN while it disseminates without compromising on granularity of required data. For the last few years, various attempts have been made to reduce energy consumption at different layers of WSN design by targeting different set of network activities. But, among all WSN activities, communication (i.e. inter-node transmissions) is the most power hungry activity and consumes major share of overall network energy. It is known that 3000 instructions can be executed for the same cost as the transmission of one bit over 100 meters. Moreover, the numbers of inter-node transmissions are proportional to the bulk of data to be disseminated from source to sink i.e. data collecting node. Hence, in current research we target at reducing energy consumption in a WSN by minimizing numbers of inter-node transmissions caused by voluminous data. Reduction in inter-node transmissions is achieved by various ways, like treating data in-network to eliminate undesired data present in the sensed data streams generated at SNs; by avoiding generation of unnecessary (i.e. redundant) data at SNs by estimating the current behaviour of an event and setting sensing frequency dynamically; by developing an effective clusteringand path setup scheme that yields smaller paths between source-sink pairs than set by existing methods; and by developing a caching strategy which increases proximity of the data nearer to sink which reduces query traversal path. First part of the thesis presents a novel Two-Ways Sliding Window (TWSW) data filtering scheme that exploits correlation among different data items at source SNs and at

cluster head nodes (CHNs) to block/suppress undesired data transmissions towards sink. Redundant Data Filtering (RDF) and Spurious Data Filtering (SDF) are two basic components ofTWSW data filtering that block data which either vary insignificantly from previously disseminated data or vary beyond maximum legitimate limit. Redundancy in data may be caused by different closely spaced SNs having same views of an event during a given sensing interval or by a SN observing same state during consecutive sensing intervals. Spurious data may be caused by temporary abnormal conditions (not related to event) prevailing around few SNs or due to malfunctioning of node software/hardware or caused by an adversary etc. TWSW efficiently isolates such unwanted data items and drops them without forwarding any further. Analytical study and simulation experiments have shown that little in-network processing done on data in the form of TWSW data filtering results in significant reduction in energy consumption by avoiding costlier and unnecessary inter-node transmissions. Second part of the thesis presents an approach called as Adaptive Sensing and Dynamic Data Reporting (ASDDR) which helps SNs to modify their sensing frequencies dynamically to adapt to changing behaviour of an event. Since, in most cases the rate of change in event behaviour is not consistent and is highly unpredictable, hence adjudging optimal sensing frequency at the start of network operation is extremely difficult. Improper sensing frequency results in either a lot of unnecessary sensing in relatively static event scenarios or may miss many important points of observation in more dynamic scenarios. Therefor, there is need for schemes like ASDDR that sets sensing frequency according to current event behaviour. Like TWSW data filtering, ASDDR exploits processing capabilities of nodes to treat data in-network. ASDDR basically augments TWSW data filtering with additional functionality in the form of a novel algorithm that runs at CHNs. It analyzes the flow of data reports received from SNs (i.e. reports that crossed TWSW data filtering check) to infer the current rate of change in event behaviour. Accordingly, it computes corrective adjustment values for active SNs to modify their sensing frequencies. Thus, sensing frequencies at SNs adapts automatically to event behaviour without manual intervention. ASDDR approach is very simple and easy to implement although very effective as revealed by simulation study. Third part of the thesis explains an efficient virtual grid formation/clustering strategy known as Grid Based Data Dissemination (GBDD). GBDD results in smaller paths between source-sink pairs and handles sink/event movements effectively. Apart from reducing inter-node transmissions, the other motive behind developing GBDD is to devise

a grid formation and clustering scheme that is tailor made for further proposed Cooperative Caching scheme for WSNs. GBDD exploits location awareness and dualradio modes of SNs to construct a virtual grid of square sized cell over entire sensor field. The size of grid cell is determined by dual radio range (i.e. high power and low power) of radio transceivers on SNs. The intuition behind this is that nodes use high power radio to transmit and receive data and queries, whereas other network management activities can be handled by low power radio. GBDD ensures automatic start and termination of the grid construction process as well as path setup between a source and a sink. Virtual grid created helps in defining clusters, setting better paths between source-sink pairs and handling movement as well as multiplicity of sinks/events effectively. Simulation results reveal that GBDD gives significant improvements in terms of overall network energy savings when compared with other existing schemes. Fourth part of thesis gives detailed Cooperative Caching (CC) scheme that solves the problem of temporarily holding large data in the network where individual SN storage is very limited. CC exploits cooperation among various SNs in a defined region to form cooperative zones in the network to form larger virtual cache. Apart from its own limited local storage, a selected coordinator node uses storages of nodes from certain region around it to realize larger cache known as cumulative cache (CMC). Based on availability of free storage locations, CC tries to cache most of data nearer to sink Thus, apart from handling excess data flow it reduces query traversal paths and hence reduces inter-node transmissions. As part of a complete CC scheme, various modules are devised. Token Based Cache Admission Control Scheme is devised to increase data proximity nearer to sink. A Cache Discovery Mechanism to fetch copy of data item in minimum time from a location nearest to query sender is developed. To avoid replication of data cached, a Single Copy Cache Rule is also developed which ensures that only singly copy of a data item exists in the network. Finally, a least utility based Item Replacement Policy is devised so that a data item with least utility is evicted out in case of required item replacement. Lastly, the contribution made in the dissertation is summarized and scope for future work is outlined.