Learning-Based Wireless Powered Secure Transmission

In this letter, we propose a learning-based wireless powered secure transmission, in which a source utilizes energy harvested from a power beacon to communicate with a legitimate receiver, in the presence of an eavesdropper. In order to confuse the eavesdropper, we assume that the source transmits the artificial noise signals, in addition to the information signals. We first characterize the effective secrecy throughput of our system, showing its dependence on the transmission parameters, including the fraction of time allocated for wireless power transfer, the fraction of power allocated to the information signals, as well as the wiretap code rates. We then leverage the deep feedforward neural network to learn how the optimal transmission parameters that jointly maximize the effective secrecy throughput can be obtained. Through numerical results, we demonstrate that our learning-based scheme can achieve almost the same secrecy performance as the optimal solution obtained from the exhaustive search, while requiring much less computational complexity.