Data-Driven Model-Predictive Communication for Resource-Efficient IoT Networks

Rapid growth of massive Internet-of-Things (IoT) technologies and a multitude of applications represent an ongoing paradigm shift within our traditional human-centric cellular communication towards expanding connectivity for billions of things. Nowadays, initial generations of connected IoT devices and applications enabled by Cellular-IoT (CIoT) and Low Power Wide Area Networks (LPWAN) technologies are deliberately kept simple and based on equidistant, regular communication intervals. However, this simple communication behavior does not meet 5G requirements of massive Machine Type Communication (mMTC) regarding high node density, long-term energy efficiency and low cost per data. In contrast, this paper presents an approach for an Artificial Intelligence (AI) based model-predictive communication to reduce the effort for continuous, regular data exchange-patterns. Training and test data for the generation of the underlying model are obtained from long-term environment sensor measurements. The derived model approaches are applied to an integrated Industrial-IoT field demonstrator covering a centralized heating control system and decentralized LoRa temperature sensors. Finally, results constitute that overall communication effort per day can be reduced by 60% to more than 95% depending on the required accuracy, significantly contributing to the achievement of mMTC performance targets.