Scholarlink

Beamforming based on large-scale antenna arrays is an essential me-ans to achieve high-gain beams in millimeter wave (mmWave) communication systems. The considerable overhead of measuring beams presents a significant hurdle for real-world use. This paper exploits artificial intelligence (AI) techniques to forecast future beam pair signal quality using a limited set of past selections, thereby reducing measurement overhead and latency. Specifically, a recurrent residual model for temporal beam prediction (R2-TBP) is proposed, which consists of residual (Res) module, convolutional residual long short-term memory (CRLSTM) module and attention linear perceptron (ALP) module. The proposed R2TBP net could capture the high-dimensional spatial and temporal features of the historical selected beam pairs, ultimately predicting future reference signal received power (RSRP) values of all beam pairs through the ALP module. Experimental results demonstrate that the R2TBP net could achieve high accuracy in temporal beam prediction with low measurement overhead.

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

Temporal Beam Prediction for MmWave MIMO Systems Based on Deep Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yan Zhao,
  • Long Zhao,
  • Jiayi Xu,
  • Hongrui Shen

摘要

Beamforming based on large-scale antenna arrays is an essential me-ans to achieve high-gain beams in millimeter wave (mmWave) communication systems. The considerable overhead of measuring beams presents a significant hurdle for real-world use. This paper exploits artificial intelligence (AI) techniques to forecast future beam pair signal quality using a limited set of past selections, thereby reducing measurement overhead and latency. Specifically, a recurrent residual model for temporal beam prediction (R2-TBP) is proposed, which consists of residual (Res) module, convolutional residual long short-term memory (CRLSTM) module and attention linear perceptron (ALP) module. The proposed R2TBP net could capture the high-dimensional spatial and temporal features of the historical selected beam pairs, ultimately predicting future reference signal received power (RSRP) values of all beam pairs through the ALP module. Experimental results demonstrate that the R2TBP net could achieve high accuracy in temporal beam prediction with low measurement overhead.