## Research

### Wireless sensor networks

The rapid advances in **micro-sensor technology** have enabled the development of distributed networks composed of small and inexpensive nodes, called **motes**, with capabilities of **sensing**, **computation** and **communication**. Since nodes are battery-powered, **energy consumption** is one of the critical aspects in the design of **wireless sensor networks**. Datasheets of commercial motes show that the radio on board nodes is the main cause of energy consumption. Thus, in order to save energy, communication among nodes should be reduced as much as possible without affecting the accuracy of the network. One of the most popular approaches to reduce energy consumption consists of aggregating the relevant data so as to limit transmission/reception of messages and therefore reducing the time in which the radio is on. Further, aggregation results to be particularly attractive when we are interested in collecting an aggregate, such as minimum or maximum, of the values collected from the single sensors, rather than the single values themselves.

In this framework, we have proposed a novel distributed approach based on **fuzzy numbers** and **ordered weighted aggregators** to perform an energy efficient aggregation in wireless sensor networks. A distributed approach is desirable because there is no need to create and maintain a hierarchical structure of the network. The basic point of our approach is that each node maintains an estimate of the aggregated value. For instance, if the objective of the application is to monitor that the temperature in the network does not exceed an alarm threshold, each node maintains the estimate of the maximum temperature. To take the measurement error into account, we represent the value measured by the sensor as a central triangular fuzzy number. It follows that the estimate of the aggregated value is a central triangular fuzzy number in its turn. The aggregation is performed by each node in the network when either a new value is available from the local sensor or a new value has been received from a neighbouring node.

The aggregation is implemented as a weighted average between the estimate and the new value, where weights are inversely proportional to the **degree of uncertainty** of the two fuzzy numbers. Actually, the aggregation is executed only if the two fuzzy numbers intersect each other. If there is no intersection, the new estimate corresponds to the largest or smallest fuzzy number depending on the objective of the application (estimating a maximum or a minimum).

Once the new estimate has been computed, the node has to decide whether the new estimate must be sent in broadcast to the neighbouring nodes or not. To take this decision, the node uses a table with an entry *i* for each neighbouring node *n _{i}* : the entry contains the last value received from node

*n*. This value coincides with the last aggregation that has been carried out in node

_{i}*n*and has generated a broadcast. In the case of maximum (minimum) estimation, if the new estimate is larger (lower) than at least one entry in the table and the degree of uncertainty of this entry is higher than the one of the new estimate, then the new estimate is sent in broadcast; otherwise no communication occurs. If the variable we are monitoring varies slowly, the amount of exchanged messages is very slow.

_{i}
We have evaluated the application of the algorithm to the monitoring of the maximum temperature in a 12-node real wireless sensor network deployed in a two hundred square meters flat. We used **Tmote Sky** motes from **MoteIv** as nodes. We have shown that our distributed algorithm promptly reacts to changes of temperature, maintaining the total number of messages exchanged among the nodes very low. In the hypothesis of using the **B-MAC protocol**, we have also computed an estimation of the node life-time.