Minimum Latency Packet Coding and Scheduling in Multi-path Networks with Delayed Feedback

This paper investigates the minimum latency packet coding and scheduling scheme in multi-path communication, which is formulated as a Markov decision process (MDP) under instant feedback and as a Partially observable MDP (POMDP) under delayed feedback. We point out that the action space can be characterized by the number of repair packets, information packets and forward correction packets sent over the paths. Besides, an efficient greedy policy is proposed, where the actions are determined based on delayed feedback and historical actions. Numerical results show that the proposed policy greatly outperforms the benchmarks, especially when the feedback delay is large.