[Paper Review] Artificial intelligence enabled radio propagation for communications-Part II: Scenario identification and channel modeling

Chen Huang, Ruisi He+9 more|arXiv (Cornell University)|Nov 23, 2021

Millimeter-Wave Propagation and Modeling123 references158 citations

TL;DR

This paper presents a comprehensive review of machine learning (ML)-based methods for wireless propagation scenario identification and channel modeling, emphasizing AI-driven classification of LoS/NLoS conditions and multi-dimensional channel prediction. It evaluates supervised and unsupervised ML techniques, highlights deep learning architectures like CNN-LSTM and DNN for CSI and geometry-based modeling, and identifies key challenges in generalization, dynamic environment adaptation, and integration with network optimization for 6G systems.

ABSTRACT

This two-part paper investigates the application of artificial intelligence (AI) and in particular machine learning (ML) to the study of wireless propagation channels. In Part I, we introduced AI and ML as well as provided a comprehensive survey on ML enabled channel characterization and antenna-channel optimization, and in this part (Part II) we review state-of-the-art literature on scenario identification and channel modeling here. In particular, the key ideas of ML for scenario identification and channel modeling/prediction are presented, and the widely used ML methods for propagation scenario identification and channel modeling and prediction are analyzed and compared. Based on the state-of-art, the future challenges of AI/ML-based channel data processing techniques are given as well.

Motivation & Objective

• To review state-of-the-art machine learning techniques for identifying wireless propagation scenarios, including LoS/NLoS and environmental classifications.
• To analyze and compare ML-based approaches for channel modeling and prediction across time, frequency, and spatial domains.
• To identify critical challenges in generalizing AI-based models across diverse propagation environments and dynamic scenarios.
• To explore the integration of scenario identification, channel prediction, and network optimization for future B5G/6G systems.
• To provide a roadmap for future research in AI/ML-driven radio propagation modeling and data processing.

Proposed method

• Classifies ML-based scenario identification into supervised and unsupervised learning, using features such as path loss, delay spread, and angle-of-arrival (AoA) statistics.
• Employs deep neural networks (DNN), convolutional neural networks (CNN), and recurrent neural networks (RNN/LSTM) to model and predict CSI, beamforming vectors, and geometry maps.
• Utilizes time-series modeling via RNN and LSTM to exploit temporal dynamics of channel characteristics for improved scenario classification.
• Integrates physical environmental factors (e.g., building density, material types, terrain) and propagation parameters as inputs to enhance model generalization.
• Applies meta-learning and transfer learning techniques to improve model adaptability across different scenarios with limited labeled data.
• Proposes a unified framework combining scenario identification, channel prediction, and network optimization using end-to-end ML pipelines.

Experimental results

Research questions

• RQ1 How can machine learning effectively classify LoS and NLoS propagation conditions using radio channel measurements?
• RQ2 What are the most effective deep learning architectures for predicting channel state information (CSI) across time, frequency, and spatial domains?
• RQ3 How can physical environment factors be incorporated into AI models to improve the generalization of channel models across diverse scenarios?
• RQ4 What are the key challenges in predicting rapidly changing channels during scenario transitions (e.g., urban to tunnel)?
• RQ5 How can scenario identification and channel prediction be jointly optimized to enhance network-level performance in B5G/6G systems?

Key findings

• Supervised ML methods achieve high accuracy in LoS/NLoS classification using features such as delay spread and power delay profiles, with reported F1-scores exceeding 0.9 in some studies.
• Hybrid deep learning models like CNN-LSTM outperform traditional methods in CSI prediction, especially in capturing both spatial and temporal channel dynamics.
• Incorporating physical environment factors (e.g., building height, material) into AI models improves generalization across different propagation scenarios.
• Unsupervised clustering techniques based on channel statistics can effectively identify unknown propagation environments without labeled data, though accuracy is generally lower than supervised approaches.
• AI-based channel prediction enables early CSI estimation, reducing latency in high-mobility scenarios and supporting ultra-reliable low-latency communications (URLLC).
• The integration of scenario identification, channel prediction, and network optimization via ML is a promising but underexplored direction for future 6G systems.