Collaborative computing in heterogeneous wireless sensor networks

Wireless Sensor Networks (WSNs) are nearly exclusively designed such as sensor nodes export sensed data to a base station in order to be processed due to their limited processing and memory capacities. This constitutes a critical energy issue for sensor nodes, which are often battery-powered, since radio transmission is much more energy consuming than data processing especially for packets traveling long distances. Thanks to technological progress, current WSNs integrate nodes with various processing, memory and battery capacities, and some of them offer even hardware acceleration and dynamic reconfiguration such as those based on FPGAs (Field-Programmable Gate Arrays). Using the processing power available in the network, in a localized way, appears thus more energy efficient than exporting data. In this paper, we propose a new collaborative data processing approach, which exploits node heterogeneity and dynamic hardware reconfiguration capabilities in WSNs in order to promote local processing. This aims at improving network lifetime and performance by reducing communication costs and increasing execution parallelism, respectively. Our approach uses a cluster-based network structure, where each node having data to be processed sends a request to its cluster-head. The cluster-head seeks in all its cluster members the best suited node, taking into account both real-time constraints and energy costs.