Low Complexity Distributed Estimation for IoT Sensor Networks

We propose a new low-complexity distributed estimation technique based on the LMS algorithm for applications with low available power, such as sensor networks and internet of things (IoT). The nodes that compose the network must estimate the environment state where they are inserted, using local measurements and shared information. Our algorithm, named Fixed Regressor Distributed LMS (FRD-LMS), shows a significantly reduced complexity when the input regressors at nodes are fixed, which is a practical situation that arises, for example, in source localization problems, where the regressor is related to the node position. We show that the network correctly performs estimation provided that the fixed regressors span the state space. We prove convergence in the mean for fully connected networks, and show through simulations that the algorithm also converges when the nodes are connected as a simple ring. The low-complexity and simplicity of the FRD-LMS make it suitable to IoT contexts, where such features are greatly desirable.