Data-Driven Predictive Scheduling in Ultra-Reliable Low-Latency Industrial IoT: A Generative Adversarial Network approach

To date, model-based reliable communication with low latency is of paramount importance for time-critical wireless control systems. In this work, we study the downlink (DL), i.e., controller-to-actuator, scheduling problem in a wireless industrial network such that the outage probability is minimized. In contrast to the existing literature based on well-known stationary fading channel models, we assume an arbitrary and unknown channel fading model, which is available only via samples. To overcome the issue of limited data samples, we invoke the generative adversarial network (GAN) framework and propose an online data-driven approach to jointly schedule the DL transmissions and learn the channel distributions in an online manner. Numerical results show that the proposed approach can effectively learn the arbitrary channel distribution and further achieve the optimal performance by using the predicted outage probability.