AoI Minimization for Multi-user Networks with Delayed Feedback

Abstract

This paper considers the packet scheduling for minimizing the age of information (AoI) in multi-user networks under delayed feedback, and investigates the impact of feedback delay on the AoI performance. With delayed feedback, the transmitter is not aware of the real-time receiving status at the moment of making the scheduling decisions. Hence, this problem is modelled as a partially-observable Markov decision process (POMDP), which includes a belief vector describing the probability distribution of the receiving status for all the users. The belief vector is updated based on the delayed feedback and the historical actions. Solving the POMDP optimally is rather challenging due to the large state space. We further propose a low-complexity policy, which selects the action that maximizes the expected immediate reward at each time slot. Numerical results show that the proposed policy significantly outperforms the stationary random policy. By comparing with the scheduling algorithm under instantaneous feedback, we show that the performance degradation caused by feedback delay increases with the packet arrival rate, channel erasure probability and the number of users.