Abstract
Wireless sensor networks (WSN) is an emerging technology in recent years that has a wide variety of applications, such as environmental surveillance, target locating and tracking, and personal health care. Distributed estimation is one of the most interesting applications in WSNs, where spatially distributed sensors are deployed over the sensing area to estimate an unknown signal. Sensors are basically low-cost devices that are endowed with the sensing function and the wireless communication ability, and powered by the batteries. The battery-driven sensors are typically subject to stringent energy constraints; hence, energy conservation is a crucial issue for extending the network lifetime in WSNs. In this dissertation, we consider the effects of noisy observations and imperfect information transmissions, and propose energy-efficient cooperative information aggregation (CIA) schemes for distributed estimation in WSNs. In the CIA schemes, each node sends only one bit, representing its local observation, to the fusion center (FC) in each estimation procedure to reduce the transmission energy consumption. In addition, the concept of error correction codes is incorporated into the design of CIA schemes to provide the observation protection. In the first part of this dissertation, we use the technique of repetition codes for observation protection. Each node quantizes its local observation into multiple bits and then selects one of the resultant bits to forward to the FC. Two estimators, the hard decision estimator (HDE) and the maximum likelihood estimator (MLE), are proposed as the fusion rules performed at the FC in the CIA schemes. The design of the applied repetition codes corresponds to the arrangement for the forwarded bit at each node; the forwarded bit arrangements for the HDE and the MLE are formulated as a resource allocation problem to investigate. By solving the convex optimization problem transformed from the resource allocation problem, the suboptimal resource allocation (SORA) for HDE and the suboptimal hybrid resource allocation (SHRA) for MLE are derived, respectively. The simulation results show that the proposed SORA and SHRA are flexible resource allocation strategies that adaptively arrange the resource based on the system parameters, and can have better estimation performance than other compared resource allocation strategies. Also, the proposed CIA schemes are shown to outperform the other compared schemes under the scenarios that include imperfect communication channels between the nodes and the FC. Next, in the second part of the dissertation, we discuss the trade-off between energy conservation and estimation performance as it appears in the sequential CIA (SCIA) approach. In the SCIA approach, the strategy of sequential information forwarding is adopted instead of batch information forwarding. The fusion operation is based on the hard decision estimator and consisted of several sequentially executed voting processes. On the system initialization, each voting process is properly furnished with some ratio of resources, i.e. the information bits of the nodes, for the majority principle–based detection of its responsible bit in the quantized observation. Because the nodes sequentially forward their information bits to the FC, each voting process can be terminated when the majority is achieved after collecting partial information bits, and the nodes that are unneeded to forward their information bits are regarded as the unused resources in the voting process. Based on the different purposes regarding the management of the unused resources in each voting process, we design the energy-saving (ES) algorithm and the performance-improving (PI) algorithm for the SCIA approach, along with the analytical derivation for the total energy consumption of the ES algorithm. The presented simulation results reveal the effectiveness of the ES and PI algorithms on energy saving and performance improvement, respectively, in comparison with the CIA scheme using batch information forwarding. Also, the analytical solution for the total energy consumption of the ES algorithm is verified. In the last part of the dissertation, motivated by improving the estimation performance of the original repetition codes–based CIA scheme, we investigate the incorporation of other more robust techniques of error correction codes. In considering the trade-off between enhancement of the error correction capability and suppression of the impact of observation noise, we apply the low-density generator matrix (LDGM) codes and propose the LDGM codes–based CIA scheme. In the LDGM codes–based CIA scheme, the generation of the forwarded bit at each node corresponds to a predetermined observation encoding process with very low computational complexity. In addition, considering the degradation of the error correction performance caused by the presence of short cycles in the generator matrix, we devise the construction for the generator matrix of the applied LDGM codes so that it is free from cycle of length 4. The simulation results show that the LDGM codes–based CIA scheme significantly outperforms the original repetition codes–based CIA scheme. Also, the robustness of the LDGM codes–based CIA scheme against the random death of nodes in the system is presented in the simulation.