Abstract
This paper considers uplink and downlink transmissions in a network with radio frequency-powered Internet of Things sensing devices. Unlike prior works, for uplinks, these devices use framed slotted Aloha for channel access. Another key distinction is that it considers uplinks and downlinks scheduling over multiple time slots using only causal information. As a result, the energy level of devices is coupled across time slots, where downlink transmissions in a time slot affect their energy and data transfers in future time slots. To this end, this paper proposes the first learning approach that allows a hybrid access point to optimize its power allocation for downlinks and frame size used for uplinks. Similarly, devices learn to optimize (1) their transmission probability and data slot in each uplink frame, and (2) power split ratio, which determines their harvested energy and data rate. The results show our learning approach achieved an average sum rate that is higher than non-learning approaches that employed Aloha, time division multiple access, and round-robin to schedule downlinks or/and uplinks.