[Paper Review] Machine Learning Empowered Resource Allocation in IRS Aided MISO-NOMA Networks

Xinyu Gao, Yuanwei Liu+2 more|arXiv (Cornell University)|Mar 9, 2021

Advanced Wireless Communication Technologies38 references52 citations

TL;DR

This paper proposes a machine learning-empowered resource allocation framework for IRS-aided MISO-NOMA networks to maximize sum-rate under QoS constraints. It integrates LSTM for user mobility prediction, K-GMM for dynamic user clustering, and DQN for joint phase shift and power allocation optimization, achieving a 35% throughput gain over OMA schemes in simulations.

ABSTRACT

A novel framework of intelligent reflecting surface (IRS)-aided multiple-input single-output (MISO) non-orthogonal multiple access (NOMA) network is proposed, where a base station (BS) serves multiple clusters with unfixed number of users in each cluster. The goal is to maximize the sum rate of all users by jointly optimizing the passive beamforming vector at the IRS, decoding order, power allocation coefficient vector and number of clusters, subject to the rate requirements of users. In order to tackle the formulated problem, a three-step approach is proposed. More particularly, a long short-term memory (LSTM) based algorithm is first adopted for predicting the mobility of users. Secondly, a K-means based Gaussian mixture model (K-GMM) algorithm is proposed for user clustering. Thirdly, a deep Q-network (DQN) based algorithm is invoked for jointly determining the phase shift matrix and power allocation policy. Simulation results are provided for demonstrating that the proposed algorithm outperforms the benchmarks, while the throughput gain of 35% can be achieved by invoking NOMA technique instead of orthogonal multiple access (OMA).

Motivation & Objective

• To maximize sum-rate in IRS-aided MISO-NOMA networks with dynamic user clusters.
• To jointly optimize passive beamforming, decoding order, power allocation, and number of clusters under QoS constraints.
• To address the non-convex, high-complexity optimization problem using machine learning techniques.
• To enable efficient, real-time resource allocation in mobile, multi-user IRS-NOMA environments.

Proposed method

• Uses LSTM to predict user mobility for proactive resource adaptation.
• Proposes a K-GMM algorithm to cluster users based on predicted mobility and channel conditions.
• Employs a DQN-based algorithm to jointly optimize phase shift matrices and power allocation coefficients.
• Integrates the three-stage framework: mobility prediction → clustering → joint beamforming and power control.
• Formulates a sum-rate maximization problem with constraints on user rate requirements.
• Uses deep reinforcement learning to handle the non-convex, combinatorial optimization of beamforming and decoding order.

Experimental results

Research questions

• RQ1 Can LSTM-based mobility prediction improve resource allocation efficiency in IRS-NOMA networks?
• RQ2 How effective is the proposed K-GMM clustering method in handling dynamic user groupings?
• RQ3 Can DQN outperform conventional optimization and Q-learning in joint phase shift and power allocation?
• RQ4 What is the performance gain of NOMA over OMA in IRS-aided MISO-NOMA systems?
• RQ5 How does the number of IRS elements affect the sum-rate under the proposed framework?

Key findings

• The proposed DQN-based algorithm outperforms Q-learning and benchmark schemes in sum-rate performance.
• A 35% throughput gain is achieved by using NOMA instead of OMA in IRS-aided networks.
• Sum-rate increases with the number of IRS elements, though the gain growth rate slows at higher element counts.
• The optimal decoding order found by DQN significantly outperforms random decoding order, especially at high transmit power.
• The K-GMM clustering method effectively groups users based on mobility and channel state, improving system fairness and spectral efficiency.
• The integration of LSTM, K-GMM, and DQN enables real-time, adaptive resource allocation with low computational overhead.