Scholarlink

<p>The power allocation in Massive Multiple Input Multiple Output (MIMO) reduce bit error rate and increase performance and capacity by distributing the power through a downlink between transmitted symbols in the wireless communication system. However dynamic power allocation encounters challenges in the Massive MIMO system. Therefore, the proposed method is introduced to address the challenge of accommodating various numbers of user equipment (UEs) using a deep learning MultiLayer Perceptron (MLP) Model for dynamic power allocation. Further, the proposed method ensures adaptive input allocation in the communication system and eliminates the need to compute statistical averages and the requirement of standard methods for achieving optimal performance. To handle varying input sizes within a fixed upper bound, the MLP incorporates a modified input structure that applies zero padding to the user equipment (UE) positions. The redundant zero inputs are unnecessarily processed in the zero padding method. Hence, the Recurrent Neural Network (RNN) is used to capture the temporal dependencies and sequential patterns inherited in the power allocation task. The RNN enhances adaptability to change the number of UEs and, improves performance and efficiency for handling dynamic input dimensions in the form of cell arrays by leveraging recurrent connections. For optimizing power allocation performance in dynamic wireless communication environments, the proposed encoder-decoder LSTM framework is systematically evaluated against conventional Feedforward Neural Network (FFNN) models with and without zero padding. Experimental results demonstrate that the proposed LSTM model achieves an achieved data rate of 16.84 Mbps, a spectral efficiency of 0.84 bits/s/Hz, a fairness index of 0.123 under the Max-Min MMSE strategy, and a minimum mean square error of 0.06. Compared with the FFNN-based approach, the proposed LSTM framework provides approximately 60% higher average spectral efficiency under dynamic user scenarios while naturally supporting variable-length UE inputs without redundant zero-padding operations. These results demonstrate that the proposed framework offers higher prediction accuracy, improved robustness, and lower computational overhead for real-time power allocation in heterogeneous Massive MIMO systems.</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

Deep Learning Techniques for Real-Time Power Allocation in Massive MIMO HetNets

  • 期刊论文

  • 发表日期：

    2026年7月19日

查看全文
  • P. Jeyakumar,
  • S. Subaselvi,
  • J. Kavitha,
  • J. Anandpushparaj

摘要

The power allocation in Massive Multiple Input Multiple Output (MIMO) reduce bit error rate and increase performance and capacity by distributing the power through a downlink between transmitted symbols in the wireless communication system. However dynamic power allocation encounters challenges in the Massive MIMO system. Therefore, the proposed method is introduced to address the challenge of accommodating various numbers of user equipment (UEs) using a deep learning MultiLayer Perceptron (MLP) Model for dynamic power allocation. Further, the proposed method ensures adaptive input allocation in the communication system and eliminates the need to compute statistical averages and the requirement of standard methods for achieving optimal performance. To handle varying input sizes within a fixed upper bound, the MLP incorporates a modified input structure that applies zero padding to the user equipment (UE) positions. The redundant zero inputs are unnecessarily processed in the zero padding method. Hence, the Recurrent Neural Network (RNN) is used to capture the temporal dependencies and sequential patterns inherited in the power allocation task. The RNN enhances adaptability to change the number of UEs and, improves performance and efficiency for handling dynamic input dimensions in the form of cell arrays by leveraging recurrent connections. For optimizing power allocation performance in dynamic wireless communication environments, the proposed encoder-decoder LSTM framework is systematically evaluated against conventional Feedforward Neural Network (FFNN) models with and without zero padding. Experimental results demonstrate that the proposed LSTM model achieves an achieved data rate of 16.84 Mbps, a spectral efficiency of 0.84 bits/s/Hz, a fairness index of 0.123 under the Max-Min MMSE strategy, and a minimum mean square error of 0.06. Compared with the FFNN-based approach, the proposed LSTM framework provides approximately 60% higher average spectral efficiency under dynamic user scenarios while naturally supporting variable-length UE inputs without redundant zero-padding operations. These results demonstrate that the proposed framework offers higher prediction accuracy, improved robustness, and lower computational overhead for real-time power allocation in heterogeneous Massive MIMO systems.