Multi-Slot Distributed Measurement Selection: A Sparsity Learning Approach

When only a part of measurements are collected sequentially in large-scale Internet of Things (IoT) networks, it is motivated to extract significant information from the collected data for utilizing the communication resources over incoming slots more efficiently. In this article, we study the successive measurement selection problem for sparse parameter estimation and its distributed implementation in large-scale networks. Specifically, we transform the intractable multi-slot measurement selection problem into per-slot subproblems, and in each slot we select the uncollected measurements that are more “useful” for parameter estimation based on sparse Bayesian learning. In order to deal with the combinatorial problem with distributed implementation, we relax the subproblem and propose an enhanced subgradient method to solve the Boolean constraint. The proposed distributed scheme needs only a common information broadcast by the centralized controller, where the communication overheads only depend on the number of sparse active sources rather than the number of sensors which is always huge. Simulation results demonstrate the performance gain of our proposed measurement selection method compared to the conventional schemes. When the total communication overheads are limited with the network scale, our scheme presents extreme practicability and applicability for the large-scale networks.