A Heterogeneous System Architecture for Low-Power Wireless Sensor Nodes in Compute-Intensive Distributed Applications

Wireless Sensor Networks (WSNs) combine embedded sensing and processing capabilities with a wireless communication infrastructure, thus supporting distributed monitoring applications. WSNs have been investigated for more than three decades, and recent social and industrial developments such as home automation, or the Internet of Things, have increased the commercial relevance of this key technology. The communication bandwidth of the sensor nodes is limited by the transportation media and the restricted energy budget of the nodes. To still keep up with the ever increasing sensor count and sampling rates, the basic data acquisition and collection capabilities of WSNs have been extended with decentralized smart feature extraction and data aggregation algorithms. Energy-efficient processing elements are thus required to meet the ever-growing compute demands of the WSN motes within the available energy budget. The Hardware-Accelerated Low Power Mote (HaLoMote) is proposed and evaluated in this thesis to address the requirements of compute-intensive WSN applications. It is a heterogeneous system architecture, that combines a Field Programmable Gate Array (FPGA) for hardware-accelerated data aggregation with an IEEE 802.15.4 based Radio Frequency System-on-Chip for the network management and the top-level control of the applications. To properly support Dynamic Power Management (DPM) on the HaLoMote, a Microsemi IGLOO FPGA with a non-volatile configuration storage was chosen for a prototype implementation, called Hardware-Accelerated Low Energy Wireless Embedded Sensor Node (HaLOEWEn). As for every multi-processor architecture, the inter-processor communication and coordination strongly influences the efficiency of the HaLoMote. Therefore, a generic communication framework is proposed in this thesis. It is tightly coupled with the DPM strategy of the HaLoMote, that supports fast transitions between active and idle modes. Low-power sleep periods can thus be scheduled within every sampling cycle, even for sampling rates of hundreds of hertz. In addition to the development of the heterogeneous system architecture, this thesis focuses on the energy consumption trade-off between wireless data transmission and in-sensor data aggregation. The HaLOEWEn is compared with typical software processors in terms of runtime and energy efficiency in the context of three monitoring applications. The building blocks of these applications comprise hardware-accelerated digital signal processing primitives, lossless data compression, a precise wireless time synchronization protocol, and a transceiver scheduling for contention free information flooding from multiple sources to all network nodes. Most of these concepts are applicable to similar distributed monitoring applications with in-sensor data aggregation. A Structural Health Monitoring (SHM) application is used for the system level evaluation of the HaLoMote concept. The Random Decrement Technique (RDT) is a particular SHM data aggregation algorithm, which determines the free-decay response of the monitored structure for subsequent modal identification. The hardware-accelerated RDT executed on a HaLOEWEn mote requires only 43 % of the energy that a recent ARM Cortex-M based microcontroller consumes for this algorithm. The functionality of the overall WSN-based SHM system is shown with a laboratory-scale demonstrator. Compared to reference data acquired by a wire-bound laboratory measurement system, the HaLOEWEn network can capture the structural information relevant for the SHM application with less than 1 % deviation