With the prevalence of wireless signals, radio frequency (RF) energy harvesting (EH) has received substantial attention and becomes a candidate for self-sustainable energy supply. Various works show that the harvested energy from ambient RF sources co...
With the prevalence of wireless signals, radio frequency (RF) energy harvesting (EH) has received substantial attention and becomes a candidate for self-sustainable energy supply. Various works show that the harvested energy from ambient RF sources could range from micro-Watts to milli-Watts and it varies depending on environmental factors and energy receiver. In cognitive radio networks (CRNs), RF EH becomes a promising approach to support both spectral and energy utilization efficiencies. A considerable number of works focused on RF-powered CRNs from different perspectives.
The EH secondary transmitter (ST) equips with an RF transceiver and an RF energy harvester with separated antennas. Therefore, the EH ST can harvest energy from the same frequency band for information transmission and different from that for information transmission. Without additional energy supply, the ST has to harvest sufficient energy from ambient RF sources to support information transmission. Therefore, the achievable throughput of the secondary system significantly depends on the opportunity for accessing an idle channel and harvested energy supporting active transmission.
In this paper, we define an -slot charging system corresponding to the maximum number of energy harvesting slots to support the target signal-to-noise ratio. For opportunistic spectrum access, the ST needs to perform channel sensing to find an idle channel or can learn the optimal policy by interacting with the primary channel.
Firstly, in a single-slot charging system, the ST harvests energy from the frequency band different from information transmission. For analysis on achievable throughput, we develop a Markov based battery model in which the energy state varies depending on harvesting and transmission. From the Markov model, the probability that the energy outage occurs was computed, and the active transmission throughput was derived accordingly. Monte-Carlo simulation was performed to show an agreement between the analysis and simulation results. Also, we propose a double-slot sensing scheme in which the ST sequentially senses up to 2 different channels to find an idle channel. The improved throughput was validated through simulation results.
Secondly, in a multi-slot charging system, the ST harvests energy from a busy channel and transmit information when the primary channel becomes idle. We develop a quasi-birth-and-death (QBD) process-based energy variations to derive the stationary probability of energy deficit consisting of multiple energy states in which active transmission is unavailable. We also apply the ambient backscatter communications with which the ST can transmit information on a busy channel and consumes a negligible amount of energy. To support efficient energy replenishment, we propose a stochastic mode selection scheme so that the two modes are stochastically selected when the primary channel is busy. We propose an energy level-dependent mode selection scheme to flexibly control the probability at each energy level for further improvements. Monte-Carlo simulation was performed to validate the analysis and compare the achievable throughputs between level-independent and level-dependent schemes.
Lastly, different from opportunistic accessing based on channel sensing, we propose a reinforcement learning-based mode optimization scheme in which the ST can learn the optimal policy through rewards obtained by interacting with the primary channel. To be more specific, the ST performs harvesting mode observing the variations on energy state to predict the primary channel state. We formulate the proposed scheme with a Markov decision process and design a deep Q-network (DQN) for mode optimization. The achievable throughput was validated through simulations by considering a greedy policy based on the trained DQN model. It was compared with the ideal case when the complete information about the primary channel is provided. Simulation results demonstrated that the proposed scheme could achieve the throughput close to the ideal case in some conditions.