Entropy-based Reward Design for Deep Reinforcement Learning-enabled Routing, Modulation and Spectrum Assignment of Elastic Optical Networks

We present a new reward design for the deep reinforcement learning (DRL)-based routing, modulation and spectrum assignment in the elastic optical networks (EONs). The performance of the EONs is highly related to the spectrum fragmentation, and entropy can be used to quantify the fragmentation level. In this paper, we propose entropy-based reward design for the DRL agent. By doing this, the impact on the spectrum fragmentation of different actions can be perceived to facilitate the learning of the agent. Simulation results show that our proposed approach outperforms previous simple reward-based DRL approaches.