Q-MOHO: Q-Machine Learning Guided Multi-Objective Power Optimization for RIS-Assisted MIMO-NOMA Systems

Authors

  • P G Suprith P.E.S. Institute of Technology and Management, Shivamogga
  • H B Marulasiddappa GM University, Davangere
  • G S Prashanth Jawaharlal Nehru New College of Engineering (JNNCE), Shivamogga, and Karnataka
  • G Narayanaswamy P.E.S. Institute of Technology and Management, Shivamogga

Abstract

In this study, we introduce Q-MOHO (Q-Machine Learning Guided Multi-Objective Hybrid Optimization) for power allocation in MIMO-NOMA communication systems aided by RIS. To optimize throughput, improve equalization, and reduce power variation, the algorithm adjusts power distribution among three users facilitated by 64 RIS components. The Q-learning approach operates over an SNR spectrum of 0 dB to 30 dB and trains over 500 instances employing a separate power allocation strategy with five tiers. Simulation studies indicate that the proposed Q-MOHO results in an average reward increase of 32% over baseline methods after approximately 350 episodes. Dynamically adjusting with SNR, the optimal power distribution spans from 0.2 at elevated SNR levels to 0.8 at reduced SNRs. The Q-MOHO performance of 5.40 bps/Hz at an SNR of 25 dB is substantial when contrasted with equal power allocation (EPA) and difference of convex optimization (DCO). These results confirm that, in real-world wireless conditions, the Q-MOHO algorithm can effectively acquire optimal strategies and significantly enhance system performance.  

Additional Files

Published

2026-05-16

Issue

Section

Wireless and Mobile Communications