Automatic Configuration of Controlled Interference Experiments in Sensornet Testbeds

Automatic configurationA group of researchers from the Institute for Technical Informatics Graz University of Technology and Computer Engineering and Networks Lab ETH Zurich have performed Automatic Configuration of Controlled Interference Experiments using Sensornet Testbeds. In a preliminary evaluation based on two testbeds, their result shows that their approach can find near-optimal solutions within at most a few hours.

In a paper that will be published this November in Proceedings of the 12th ACM Conference on Embedded Networked Sensor Systems (SenSys). They presented an automated configuration approach based on simulated annealing to overcome error-prone and time-consuming task in manually selecting sets of nodes acting as jammers and their individual transmit powers. The researchers believe their approach can facilitate the widespread adoption of controlled interference experiments by the sensornet community.

They argue that experiments under controlled radio interference are crucial to assess the robustness of low-power wireless protocols, while tools such as JamLab augment existing sensornet testbeds with realistic interference but it still remains a problem.  The research leading to their results has received funding from the European Union 7th Framework Programme (FP7-ICT-2011-8) under grant agreement n 317826 (RELYonIT) and from Nano-Tera, projects X-Sense and OpenSense

What the researchers did was to execute data “from a smaller testbed at TU Graz consisting of 17 nodes in a single room and the TWIST testbed. The latter consists of 102 nodes, out of which 77 were operational, set up in a grid-like network spread over several rooms on three floors. The optimization process was executed eight times for each testbed. Within a maximum of two hours, the current algorithm is able to find suitable configurations. The number of jammers is low enough to leave a sufficient number of nodes for the actual experimentation. While the simple solutions for the local testbed with just a single jammer do not suffer from cross-talk, this is an issue of the proposed configurations for the TWIST testbed. In the TWIST testbed, most nodes are affected by cross-talk, but typically only by one or two additional jammers with a weak signal. The synchronous variant of JamLab is able to handle cross-talk, so these configurations are still useful.

The next stage according to the researchers is to optimize the execution time and stability of the configuration process by systematically adjusting the parameters of the optimization process and by conducting a more thorough evaluation on a larger set of WSN testbeds.



ETH Zurich Publications