Enhancing Energy Efficiency in Underlay Cellular Networks: Leveraging Deep Learning for Full-Duplex SWIPT-based D2D Communications

In densely packed small-cell networks, close proximity among device-to-device (D2D) user equipment (DUE) facilitates energy harvesting from the surroundings, thus improving device energy efficiency. While full-duplex (FD) communication theoretically increases spectral efficiency twofold compared to half-duplex (HD) communication, it introduces self-interference, impacting spectral capacity and energy efficiency. This paper examines FD D2D underlay cellular networks, where DUEs simultaneously decode information and harvest energy through SWIPT. We formulate an optimization problem with the aim of maximizing energy efficiency. Global and sub-optimal solutions are acquired through exhaustive search (ES) and gradient search (GS) with the barrier algorithm, respectively. Furthermore, we design a deep neural network (DNN) algorithm for the optimization model and assess its performance against ES and GS algorithms. The results derived from our study conclusively demonstrate the high performance of FD mode in energy efficiency and sum-rate compared to HD mode, and the proposed algorithm achieves solutions close to global optimality.