Energy-Efficient Data Fusion Techniques for Wireless Sensor Networks

Abstract

Wireless sensor networks (WSNs) are finding increasing use in applications such as environmental monitoring, military surveillance, and healthcare due to their low cost and ease of deployment. However, the nodes in a conventional WSN are constrained by their limited on-board battery energy capacity. Hence, devising energy-efficient schemes to increase the lifetime of WSNs is an active and important area of research.

Opportunistic transmission schemes enhance the lifetime of WSNs by curtailing the number of transmissions by the nodes. However, this causes an unwelcome degradation in the performance of the WSN. Energy harvesting (EH), on the other hand, is an alternate solution that altogether eliminates the problem of limited lifetime in WSNs. In it, the nodes can replenish their energy buffers by harvesting energy from renewable sources. However, since the energy available is random, the nodes can occasionally be unavailable due to lack of energy, which affects performance. In this thesis, we study the performance of energy-efficient opportunistic transmission schemes for estimation and detection problems using WSNs. First, we study these schemes for parameter estimation with EH WSNs. For a general model in which the nodes experience independent and non-identical fading and the EH process at a node is stationary and ergodic, we study two important classes of channel-based opportunistic transmission schemes, namely, censoring and opportunistic subset selection. We provide lower bounds on the mean squared error for them and show that the associated trade-offs are very different from that observed in conventional WSNs.

Next, we study a detection problem. In conventional WSNs, ordering transmissions based on node log-likelihood ratios (LLRs) reduces the number of nodes that transmit and yet achieves the same detection error probability as the conventional unordered transmissions scheme (UTS) in which all nodes transmit. However, this breaks down in EH WSNs when nodes can be unavailable due to lack of energy. For the Bayesian detection framework, we propose a novel scheme that addresses this challenge for the general case in which the LLRs are bounded and have a continuous distribution function. For truncated Gaussian statistics, we then propose a novel refinement that requires even fewer transmissions and that simultaneously lowers the detection error probability when the nodes miss their transmissions. The proposed schemes also achieve a lower error probability than sequential detection.

In the last part of our work, we study the detection problem for the general, practically relevant scenario in which the measurements at the sensor nodes are spatially correlated. We show that since the observations at the sensor nodes are dependent on each other, ordering transmissions based on individual node LLRs no longer works. Thus, even for conventional WSNs, ordering the transmissions becomes a challenge. We present a novel correlation-aware ordered transmissions scheme (CA-OTS) for the binary hypothesis testing problem with Gaussian statistics. CA-OTS applies to the general case in which the hypotheses differ in the mean vector and covariance matrix, and markedly reduces the number of transmissions as compared to UTS. When the mean vector or covariance matrix is the same for the two hypotheses, we propose novel refinements that require even fewer transmissions. We also derive insightful upper bounds for them that apply to a general product-correlation model.