With the rapid advancement of 3D sensing technologies, point cloud data have emerged as a fundamental representation of 3D information. Unlike structured images, point clouds consist of irregular, unstructured spatial points that accurately capture object geometry while presenting unique challenges for efficient compression. We first examine essential deep learning (DL) architectures and their advantages in processing point cloud data, including Convolution Neural Networks and Sparse Convolutions. The discussion then analyzes various point cloud representation formats and their respective trade-offs in coding efficiency. Furthermore, we present typical DL frameworks for point cloud processing and systematically outline the general pipeline for DL-based 3D point cloud coding. Critical components including quantization techniques and entropy modeling are thoroughly investigated. The chapter ultimately establishes a cohesive theoretical framework for advancing point cloud coding research through deep learning methodologies.