Improved throughput of sensor networks with maths

PhD student Martijn Onderwater (CWI and VU) optimized the throughput of ZigBee sensor networks with mathematical techniques. This is important, as the number of sensors is growing and their networks become more crowded.

PhD student Martijn Onderwater (CWI and VU) optimized the throughput of ZigBee sensor networks with mathematical techniques. This is important, as the number of sensors is growing and their networks become more crowded. Onderwater also combined decision models for uncertain conditions with techniques from artificial intelligence. This resulted in a promising new technique, making sensor networks able to contribute to better decisions, such as choosing the best emergency exit. The researcher of Centrum Wiskunde & Informatica (CWI) defended his PhD thesis  'Network of Sensors - Operations and Control’ on 8 February at VU University Amsterdam.

Onderwater conducted his research with business partners who took indoor climate measurements at nurseries and in a sports hall. "For a nursery it is important that the carbon dioxide level is good, because people cannot concentrate well if the level is too high," the researcher explains. "In order to improve the indoor climate, sensors measure the CO2 level, humidity and temperature. If many of these sensors transmit their measurements to a central point, they sometimes have to wait for each other for a long time. One sensor can even dominate the network, which is not good for real-time applications such as active ventilation. With mathematical techniques it is now possible to configure the parameters of a ZigBee network in such a way, that an optimum throughput of data to a central point can be achieved. We are now even able to find the best settings without having to set up a large and expensive network." The results are relevant, for example, for applications in the Internet of Things.

Uncertainty
The researcher also studied mathematical models for decision making in uncertain conditions, so-called Markov decision models. Uncertainties are also common in sensor measurements. Onderwater explains: "In an emergency situation, for example on a train platform, you want to lead people to the correct exit, using sensor measurements. An advice to go to the nearest emergency exit can backfire if just a moment later a crowded train will arrive. It is important to bring uncertainties in the decision." Until now, a new strategy must be calculated for whenever the model differed from reality, which is computationally difficult to do. Onderwater developed a new, innovative method that does not have this disadvantage, in which he approaches value functions with evolutionary algorithms (dynamic strategy). This approach provides good strategies and has great potential. Onderwater: "The method is interesting for a broad range of problems that can be described by Markov decision models, such as the determination of future prices of hotel rooms."

The research was conducted at the Centrum Wiskunde & Informatica (CWI), the national research centre for mathematics and computer science in the Netherlands, and the Vrije Universiteit Amsterdam. It was partly funded by the Dutch ministry of Economic Affairs.


More information

Supervisor is Prof. R.D. van der Mei (CWI / VU) and co-supervisoris Dr. S. Bhulai (VU).

The research was carried out in the Stochastics research group of CWI  within the project RRR (Realisation or Reliable and Secure Residential Sensor Platforms) of the IOP program Generic Communication, number IGC1020, supported by the Strengths in Innovation Subsidy Scheme.

Picture: Sensor networks with devices like this Arduino sensor can achieve much faster throughput thanks to mathematical techniques.
Photographer/copyright: Ljubljana for goodcat / Shutterstock.com