Methods and tools for the compilation and software optimization of wireless embedded systems dedicated to applications ; Méthodes et outils pour la compilation et l'optimisation logicielle des systèmes embarqués sans fil dédiés à des applications

Wireless Sensor Networks are are highly constrained in resources: computation, memory and energy. In this context, particular attention must be paid when designing applications, communication protocols and operating systems that will run on the sensor nodes. The purpose of this thesis was to adapt the software models conventionally used for programming applications to the field of sensor networks and to propose methods and tools to build a customized version of the software to run on the sensor nodes from a model of the application. One of the components of a micro-sensor that consumes the most energy is its radio interface. The behavior of a MAC layer protocol can be expressed as a timed automaton with fixed states (which correspond to specific states of the physical device) or free states (the choice for their mapping is left to the programmer). The behavior of a radio device itself can be expressed as an automaton consisting of transitional states (with fixed duration) and non-transitional states (with unlimited duration but with a minimum time duration constraint). The problem of mapping a free state of the software of fixed duration on a path of the physical device by minimizing the energy consumed is proven to be NP-complete. A heuristic to map the free states of an automaton corresponding to a software protocol to states or paths between the states of a physical device is proposed. The approach leads to theoretical gains of 60% for the mapping of B-MAC, a MAC layer protocol. For two categories of operating systems dedicated to Wireless Sensor Networks (event driven and multi-threaded), code skeletons are generated. The simulations under the Worldsens platform of the multi-threaded code skeleton generated for B-MAC and adapted to Mantis OS confirmed the theoretical gains. Experiments on the real platform SensLab have shown that the optimization in terms of energy does not change the functional performance of the protocol. ; Les réseaux de capteurs communicants sont fortement contraints en ressources: ...