In this paper, we are interested in the data gathering for Wireless Sensor Networks (WSNs). In this context, we assume that only some nodes are active in the network, and that these nodes are not transmitting all the time. On the other side, the inactive nodes are considered to be inexistent or idle for a long time period. Henceforth, the sink should be able to recover the entire data matrix whie using the few received measurements. To this end, we propose a novel technique that is based on the Matrix Completion (MC) methodology. Indeed, the considered compression pattern, which is composed of structured and random losses, cannot be solved by existing MC techniques. When the received reading matrix contains several missing rows, corresponding to the inactive nodes, MC techniques are unable to recover the missing data. Thus, we propose a clustering technique that takes the inter-nodes correlation into account, and we present a complementary minimization problem based-interpolation technique that guarantees the recovery of the inactive nodes’ readings. The proposed reconstruction pattern, combined with the sampling one, is evaluated under extensive simulations. The results confirm the validity of each building block and the efficiency of the whole structured approach, and prove that it outperforms the closest scheme.
In this paper, we seek to address the data gathering in the continually growing Wireless Sensor Networks (WSNs) with the intention to save the nodes' energy. In order to address usual WSN problems, such as data losses, collisions and re-transmissions, a twofold data compression pattern is proposed. We consider that a restricted number of sensor nodes are selected to be active and represent the whole network, while the rest of nodes remain idle and do not participate at all in the data sensing and transmission. Furthermore, the set of active nodes' readings is efficiently reduced, in each time slot, according to the cluster scheduling. Relying on the existing Matrix Completion (MC) techniques, the sink node is unable to recover the entire data matrix due to the existence of completely empty rows that correspond to the inactive nodes, which can be considered as absent nodes for a very long period, or nodes that do not exist at all. Thereby, we propose a complementary interpolation technique, based on a minimization problem that benefits from sensor nodes inter-correlation, to guarantee the reconstruction of all the empty rows, despite their large number. The simulations confirm the efficiency of the proposed approach and show that it outperforms the existing one by up to 70.101% of Normalized Mean Absolute Error on all missed elements, when the number of active nodes is of about 10% of the total number of sensor nodes.