Wireless sensor networks (WSNs) are rapidly being adopted in a wide range of applications and soon will have a major environmental, financial and societal impact. Some of the main technical challenges in designing and deploying WSNs are meeting their communications reliability and energy consumption requirements. In order to address these two challenges, the thesis proposes new coding schemes and communication protocols, a novel paradigm for information acquisition, and the design and implementation of specific circuits architectures.
The reliability and energy efficiency trade-offs of splitting the inserted redundancy in multiple layers of the network stack are investigated through analysis and over-the-air experiments. The energy benefits of each approach are quantified by designing a low-power custom transmitter using a 65nm TSMC process, integrating the first hardware implementation of a multi-rate forward error correction (FEC) and random linear network coding (RLNC) accelerator. In addition, a physical layer (PHY) independent partial packet reception scheme is proposed for asymmetric networks, i.e. WSNs with star topology, called packetized rateless algebraic consistency (PRAC). Experiments with off-the-shelf transceivers validate our analysis results on the data reliability and energy consumption benefits of the proposed scheme.
Apart from communicating information, acquiring the signals of interest can account for a significant fraction of the power consumption of a sensor node. For this reason, the thesis investigates a nonuniform sampling scheme in order to exploit the inherent compressibility and sparse structure of typical signals encountered in many WSNs. Simulations results with real datasets and an energy comparison against the state-of-the-art sampling schemes demonstrate its rate and energy efficiency advantages. Finally, the thesis studies the fundamental performance bounds of jointly acquiring and transmitting sparse signals through noisy channels, and an integrated source representation-to-transmission scheme, called AdaptCast, is proposed. Using rate distortion analysis, its asymptotically optimal performance is proved and, based on simulation results in the context of a health monitoring application, AdaptCast's performance benefits are demonstrated against other coding schemes and PHY architectures.
Thesis supervisors: Prof. Muriel Medard, Prof. Anantha Chandrakasan
Thesis Committee: Prof. Andrea Goldsmith