.png){kind=icon}
.png){kind=icon}
Return
Deep Joint-Source Channel Coding-Based Receiver-Driven Replay Protocol for 3D Point Clouds
.png){kind=icon}
PRE
AI
DOI：10.1109/lsp.2026.3711525.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Wireless transmission of 3D point clouds must balance reconstruction quality and communication cost in the presence of channel noise. This paper proposes a receiver-driven replay protocol for learned point-cloud joint source-channel coding (JSCC). The encoder first maps the input point cloud into latent tokens and transmits a mandatory base set to produce a Round-0 reconstruction. The receiver then computes a self-consistency-based replay score from receiver-side cues available after Round-0 decoding and returns a one-bit ACK/NACK decision. If replay is requested, the transmitter sends the buffered replay tokens, allowing the receiver to refine the Round-0 reconstruction through residual correction using the Round-0 reconstruction and received replay tokens. The proposed protocol is evaluated on ShapeNet, KITTI, and 8iVFB for reconstruction. The results show that replay improves reconstruction quality over Round-0 transmission and provides a controllable trade-off between replay usage and quality under noisy channel conditions. Overall, these results support receiver-driven replay for adaptive point-cloud transmission under noisy channels.
Keywords:
3D point clouds
joint source-channel coding
receiver-driven replay
self-consistency
wireless transmission