Title: Traffic Prediction Based Fast Uplink Grant for Massive IoT

Abstract: This paper presents a novel framework for traffic prediction of IoT devices activated by binary Markovian events. First, we consider a massive set of IoT devices whose activation events are modeled by an On-Off Markov process with known transition probabilities. Next, we exploit the temporal correlation of the traffic events and apply the forward algorithm in the context of hidden Markov models (HMM) in order to predict the activation likelihood of each IoT device. Finally, we apply the fast uplink grant scheme in order to allocate resources to the IoT devices that have the maximal likelihood for transmission. In order to evaluate the performance of the proposed scheme, we define the regret metric as the number of missed resource allocation opportunities. The proposed fast uplink scheme based on traffic prediction outperforms both conventional random access and time division duplex in terms of regret and efficiency of system usage, while it maintains its superiority over random access in terms of average age of information for massive deployments.