Reinforcement Learning-Assisted Transmit Signal Power Savings in Variable Bit-rate Fronthaul

The increasing bit-rate demands placed on the fronthaul from higher user rates and multiple antenna technologies will make the consideration of its power consumption an important issue. In this study, it is assumed that the fronthaul bit-rate can be reduced from the maximum required rate through prediction of the fronthaul traffic using deep reinforcement learning (DRL). Using such predictions, and benchmarked simulations of a discrete multitone (DMT) modulation electro-absorption modulator (EAM)-based optical fiber-link, as an example of a fronthaul transmission system, it is shown that the power reduction from reducing the transmitter signal power alongside the reduction in modulation level can be between 22.3% and 34.6% within a fixed bandwidth of 34 GHz and 18 GHz respectively. Such a transmitter could be built as a bandwidth variable transponder in a Flexible Ethernet fronthaul.