In this thesis, we focused on urban wireless networks considered by the ANR project ARESA2. The networks considered by this project are heterogeneous networks. This heterogeneity is caused by the coexistence of sensor nodes with limited resources and actuator nodes with higher resources. Actuators nodes should be used differentially by the network. Hence designed protocols for WSANs should exploit resource-rich devices to reduce the communication burden on low power nodes. It is in this context that this thesis takes place in which we studied self-organizing and routing algorithms based on the heterogeneity. First, we are interested in self-organization protocols in a heterogeneous network. Based on the idea that resource-rich nodes must be exploited to reduce the communication load level on low-power nodes, we proposed self-organizing protocol called Far-Legos. Far-Legos uses the large transmit power of actuators to provide gradient information to sensor nodes. Actuators initiate and construct a logical topology. The nature of this logical topology is different inside and outside the transmission range of these resourceful nodes. This logical topology will be used to facilitate the data collection from sensor to actuator nodes. Second, we investigated the asymmetric links caused by the presence of heterogeneous nodes with different transmission ranges. The apparition of asymmetric links can dramatically decrease the performance of routing protocols that are not designed to support them. To prevent performance degradation of these routing protocols, we introduce a new metric for rank calculation. This metric will be useful to detect and avoid asymmetric links for RPL routing protocol. We also present an adaptation of data collection protocol based on Legos to detect and avoid these asymmetric links. Finally, we are interested in exploiting the asymmetric links present in the network. We proposed a new routing protocol for data collection in heterogeneous networks, called AsymRP. AsymRP, a convergecast routing protocol, assumes 2-hop neighborhood knowledge and uses implicit and explicit acknowledgment. It takes advantage of asymmetric links to ensure reliable data collection.
