Distributed Learning Algorithms for Optimal Data Routing in IoT Networks

We consider the problem of joint lossy data compression and data routing in distributed Internet of Things (IoT). Heterogeneous sources compress their data using a source-specific lossy compression scheme, where heterogeneity is meant in terms of signal type and/or transmission rates. The compressed data is thus disseminated in a multi-hop fashion until it reaches a data collector (the IoT gateway). The problem we address is to compute a suitable rate-distortion working point for the compression scheme at the source nodes, while jointly assessing the most energy efficient routing paths for the data they transmit, under channel access, distortion and capacity constraints. This is formulated as a multi-objective optimization problem that is solved through distributed learning algorithms, where source coding and routing configurations emerge as the result of local interactions among the network devices. Our final algorithm is based on the alternating direction method of multipliers (ADMM), which is accelerated using the most recent findings from the literature. As a result, it has faster convergence (up to three times) to the global optimum than standard ADMM. Numerical results are discussed for selected network scenarios, emphasizing the interrelations that exist between signal reconstruction quality at the IoT gateway and total transport energy in the network.