Scholarlink

<p>Maximum spectrum utilization is a requirement for a successful provision of 5G and beyond mobile networks, especially for mmWave massive multiple-input, multiple-output non-orthogonal multiple access (MIMO-NOMA) systems. In the existing systems, complications arise, such as dynamic spectrum availability and complex user clustering, coupled with high computational time, which could inhibit optimal resource allocation. For this reason, this work presents a new hybrid framework, named Hybrid Attention-aware Spectrum Predictor (HASP), whereby Deep Learning (DL) and Machine Learning (ML) techniques are synergized toward optimizing spectrum utilization under millimeter wave (mmWave) massive MIMO-NOMA systems. The proposed framework consists of two key modules: (1) a spectrum-aware Long Short-Term Memory (LSTM) network for predicting temporal subchannel availability, and (2) an attention-enhanced clustering mechanism that groups users based on channel quality and QoS requirements using multi-head attention. Besides, this model embraces a dual-branch architecture: the DL branch uses LSTM layers for predicting spectrum occupancy, while the ML branch uses attention-weighted K-Means clustering for optimal user grouping. The attention mechanism provides automatic weighting of the channel features and spatial features for better resource allocation. Finally, an improved Support Vector Machine (iSVM) predicts the dynamic spectrum availability. Experimental results showed 35% prediction accuracy improvement and 28% enhancement in spectral efficiency compared to conventional approaches. The hybrid framework also yields a 45% reduction in computational load while meeting Quality of Service (QoS) requirements. This converged scheme speaks volumes for the upsurge toward spectrum-efficient communications for 5G and beyond networks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid Machine Learning Framework for Spectrum Prediction and User Clustering in mmWave Massive MIMO-NOMA Systems

  • 期刊论文

  • 发表日期：

    2025年6月1日

查看全文
  • Prameela Devi R.,
  • Srilakshmi Aouthu

摘要

Maximum spectrum utilization is a requirement for a successful provision of 5G and beyond mobile networks, especially for mmWave massive multiple-input, multiple-output non-orthogonal multiple access (MIMO-NOMA) systems. In the existing systems, complications arise, such as dynamic spectrum availability and complex user clustering, coupled with high computational time, which could inhibit optimal resource allocation. For this reason, this work presents a new hybrid framework, named Hybrid Attention-aware Spectrum Predictor (HASP), whereby Deep Learning (DL) and Machine Learning (ML) techniques are synergized toward optimizing spectrum utilization under millimeter wave (mmWave) massive MIMO-NOMA systems. The proposed framework consists of two key modules: (1) a spectrum-aware Long Short-Term Memory (LSTM) network for predicting temporal subchannel availability, and (2) an attention-enhanced clustering mechanism that groups users based on channel quality and QoS requirements using multi-head attention. Besides, this model embraces a dual-branch architecture: the DL branch uses LSTM layers for predicting spectrum occupancy, while the ML branch uses attention-weighted K-Means clustering for optimal user grouping. The attention mechanism provides automatic weighting of the channel features and spatial features for better resource allocation. Finally, an improved Support Vector Machine (iSVM) predicts the dynamic spectrum availability. Experimental results showed 35% prediction accuracy improvement and 28% enhancement in spectral efficiency compared to conventional approaches. The hybrid framework also yields a 45% reduction in computational load while meeting Quality of Service (QoS) requirements. This converged scheme speaks volumes for the upsurge toward spectrum-efficient communications for 5G and beyond networks.