Design of Transmission Lengths for IR-HARQ Scheme Using Q-Learning

In this paper, we apply the Q-learning algorithm to design the transmission lengths of incremental redundancy hybrid automatic repeat request (IR-HARQ). The incremental redundancy (IR) bits are generated by polar codes with the polarizing matrix extension. The IR-HARQ scheme is considered as an environment, where the number of transmissions, the number of information bits, code rates, and signal-to-noise ratio (SNR) are the state of Q-learning. The reward is obtained by calculating the throughput of the successful transmission. The simulation results show that our Q-learning-based transmission lengths outperforms the fixed transmission lengths in the IR-HARO scheme.