Integrating Edge Intelligence and Industrial IoT via Learning-Communication Balancing Power Allocation

Edge intelligence is expected to revolutionize the industrial Internet of Things (IoT) by providing proximal intelligent services to massive low-cost IoT devices. However, integration of the two paradigms needs to simultaneously maximize the edge quality of training (QoT) and IoT quality of service (QoS) under time-varying co-channel interference, for which the existing edge or IoT resource allocation algorithms become ineffective, as they ignore the contradiction between learning performance and communication requirements. This paper proposes an edge intelligence industrial IoT (EI³) framework, which jointly maximizes QoT and QoS through a newly derived learning-communication balancing power allocation (LCBPA) formulation. An efficient algorithm is proposed to solve the non-convex and non-smooth LCBPA problem. Simulation results demonstrate that the proposed LCBPA scheme achieves superior performance compared to several benchmarks in terms of the desired learning accuracy and qualified transmission rate. It is also shown that EI³ can adapt to new scenarios by flexibly adjusting the importance factor between learning and communication.