Center for Embedded Network Sensing

Wireless sensor networks using 802.11b long-distance links enable domain scientists to measure physical phenomena in real time and at a scientifically meaningful scale. Although deployments in this setting are increasing, a characterization of the network- throughput is lacking which makes it difficult to predict the performance of sensing applications on these types of networks. We present a study of the MASE seismic data collection network which includes 64 nodes spread across mountains, valleys, rural plains, and urban environments. The network spans 250km with links as long as 43km and has been operating continuously for more than 2 years. The contributions of this paper are three fold: (a) we present a throughput characterization of wireless links for a network in operational use, (b) we suggest settings for the wireless parameters to improve the performance of the network, and (c) we document the uptime of the nodes and the data yield of the network. Our findings are beneficial for the design of future wireless sensor network deployments using 802.11b.

In our networks widely varied settings, we find that while signal-to-noise ratio is not correlated with throughput or distance, throughput is roughly correlated with distance and does not change significantly over time. At the link-layer, enabling the RTS-CTS handshake decreases the throughput while enabling retries improves the throughput. In addition, the default threshold-based rate adaptation algorithm for the transmission bit-rate is suboptimal. We find that the majority of fault cases are due to system issues and not network issues. The data yield over the course of an eight month period was 78%.