Incorporating Deep Learning in Judicious Use of Power in 5G Wireless Communications
Year: 2026
Type: conference-paper
Abstract: The rise of 5G wireless technology on a global scale calls for unique levels of spectral efficiency, energy conservation, QoS, and multi-user connectivity. Achieving all four at once demands intelligent adaptive schemes of resource allocation that surpass the capabilities of traditional signal processing techniques. This research examines the use of DL techniques in power control of 5G technology, focusing specifically on the combination of Massive MIMO with NOMA and LDPC channel coding. A novel DNN-DLS-based algorithm is developed to enable adaptive power control in MIMO-NOMA systems. It helps tackle the near-far problem by allocating more transmission power to weaker users and less power to the stronger ones in an effort to provide uniform performance for different UE’s. The DNN element uses five layers of features extraction that take input from three different sources - UE positions, CSI, and SINR targets, resulting in compressed latent representation that undergoes further processing within the DLS algorithm consisting of four successive stages - power estimation, SIC decoding, SINR optimisation and power allocation based on priorities. Simulation results demonstrate that the proposed DNN-DLS approach achieves a Bit Error Rate (BER) of 0.022 at $\mathrm{SNR}=$ 10 dB, representing a 74\% reduction compared to conventional MIMO methods, and attains a spectral efficiency of 15.5 bps/Hz with ten simultaneous users, outperforming baseline NOMA and DNN-only approaches by a substantial margin. The proposed architecture is fully end-to-end trainable and offers a scalable, low-complexity solution for next-generation heterogeneous wireless networks.
Authors: Sandeep Soni, Rajvir Saini
Institution: Kalinga University
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7203682006