SPAWN: Sensing, Probing and Accessing with Switching Energy Cost in Multichannel WSN.
Panlong Yang,Yubo Yan,Bowen Li,Deke Guo,Yafei Zhang
DOI: https://doi.org/10.1109/msn.2012.21
2012-01-01
Abstract:Multichannel accessing is a basic and important function in wireless sensor networks, where intelligent and low-cost methods are needed for energy-efficient spectrum utilization. Unfortunately, for energy constrained nodes, methods from cognitive radio technologies are not applicable in WSN multichannel systems due to computation and communication overhead, as well as extremely large energy consumptions. In this work, we propose a spectrum learning and utilizing paradigm, where sensor nodes can sense, probe, access and switch among channels automatically without prior knowledge. Under this basic and important learning method, we present a two-stage learning and utilization algorithm, which constitutes of two interleaving parts: 'transmitting while learning' and 'learning while transmitting'. Building these two parts is a non-trivial work, which need efficient channel learning and utilizing scheme working together seamlessly. In the first 'transmitting while learning' stage, we propose the 'non-call-back' policy and leverage the Lagrangian method for the additional energy cost considerations. Under this heuristic policy, an improved version for the traditional Gittins Index model is provided. After that, the heuristic algorithm is proved '2-approximate' to the optimal result. Furthermore, leveraging the channel correlations, the learning complexity is reduced from O(T) stages to O(log T) stages, where T is the number of time slots for channel learning. Secondly, we make a structural analysis on the ` learning while transmitting' stage, leveraging the 2-dimensional optimal stopping theory. Notably, we use the a 16-node TelosB sensor network and an 8-node USRP network for performance evaluation. The USRP nodes are playing as interfering and monitoring nodes for controllable network states design, making network performance evaluation convincible. The proposed algorithm is designed in real implementations with experimental results, showing the efficiency of the proposed scheme.