Scholarlink

Next-generation 5G/6G systems rely on mmWave bands to deliver multi-gigabit links, but urban microcell deployments introduce complex multipath that makes accurate channel impulse-response (CIR) estimation via ray-tracing too slow for real-time beamforming. We present a hybrid Physics-Informed Neural Network that embeds the Helmholtz PDE into its loss and learns residual multipath effects from ray-traced DeepMIMO data. Given only transmitter/receiver coordinates and baseline path-loss features, our model predicts the CIR (path delays and complex gains) in under one millisecond, ensuring both physical consistency and rapid edge-ready inference. Experimental results on DeepMIMO scenarios show a 20% reduction in delay-estimation error versus pure data-driven baselines and a 100 \(\times \) speed-up compared to full ray-tracing. This approach unites high-fidelity physics with fast machine learning for practical, real-time mmWave channel modeling.

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

Physics-Informed Neural Network for Real-Time MmWave Channel Impulse-Response Prediction in Urban Microcells

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sharv Murgai

摘要

Next-generation 5G/6G systems rely on mmWave bands to deliver multi-gigabit links, but urban microcell deployments introduce complex multipath that makes accurate channel impulse-response (CIR) estimation via ray-tracing too slow for real-time beamforming. We present a hybrid Physics-Informed Neural Network that embeds the Helmholtz PDE into its loss and learns residual multipath effects from ray-traced DeepMIMO data. Given only transmitter/receiver coordinates and baseline path-loss features, our model predicts the CIR (path delays and complex gains) in under one millisecond, ensuring both physical consistency and rapid edge-ready inference. Experimental results on DeepMIMO scenarios show a 20% reduction in delay-estimation error versus pure data-driven baselines and a 100 \(\times \) speed-up compared to full ray-tracing. This approach unites high-fidelity physics with fast machine learning for practical, real-time mmWave channel modeling.