Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transmit Weights, Not Features: Orthogonal-Basis Aided Wireless Point-Cloud Transmission

Junlin Chang, Yubo Han, Hnag Yue, John S Thompson, Rongke Liu

TL;DR

The paper tackles efficient wireless transmission of 3D point clouds by introducing a DeepJSCC-based semantic framework that leverages a receiver-side orthogonal feature pool and a folding-based decoder. It replaces raw features with learned combination weights over orthogonal bases, aided by an orthogonality regularizer and a folding prior to preserve geometry. Empirical results on ModelNet40 show parity with SEPT at high bandwidth and clear gains under bandwidth constraints, with ablations confirming the value of both the orthogonalization and folding components. This approach promises robust, geometry-preserving transmission suitable for wireless point-cloud applications, with potential extensions to fading/MIMO channels and richer attributes.

Abstract

The widespread adoption of depth sensors has substantially lowered the barrier to point-cloud acquisition. This letter proposes a semantic wireless transmission framework for three dimension (3D) point clouds built on Deep Joint Source - Channel Coding (DeepJSCC). Instead of sending raw features, the transmitter predicts combination weights over a receiver-side semantic orthogonal feature pool, enabling compact representations and robust reconstruction. A folding-based decoder deforms a 2D grid into 3D, enforcing manifold continuity while preserving geometric fidelity. Trained with Chamfer Distance (CD) and an orthogonality regularizer, the system is evaluated on ModelNet40 across varying Signal-to-Noise Ratios (SNRs) and bandwidths. Results show performance on par with SEmantic Point cloud Transmission (SEPT) at high bandwidth and clear gains in bandwidth-constrained regimes, with consistent improvements in both Peak Signal-to-Noise Ratio (PSNR) and CD. Ablation experiments confirm the benefits of orthogonalization and the folding prior.

Transmit Weights, Not Features: Orthogonal-Basis Aided Wireless Point-Cloud Transmission

TL;DR

The paper tackles efficient wireless transmission of 3D point clouds by introducing a DeepJSCC-based semantic framework that leverages a receiver-side orthogonal feature pool and a folding-based decoder. It replaces raw features with learned combination weights over orthogonal bases, aided by an orthogonality regularizer and a folding prior to preserve geometry. Empirical results on ModelNet40 show parity with SEPT at high bandwidth and clear gains under bandwidth constraints, with ablations confirming the value of both the orthogonalization and folding components. This approach promises robust, geometry-preserving transmission suitable for wireless point-cloud applications, with potential extensions to fading/MIMO channels and richer attributes.

Abstract

The widespread adoption of depth sensors has substantially lowered the barrier to point-cloud acquisition. This letter proposes a semantic wireless transmission framework for three dimension (3D) point clouds built on Deep Joint Source - Channel Coding (DeepJSCC). Instead of sending raw features, the transmitter predicts combination weights over a receiver-side semantic orthogonal feature pool, enabling compact representations and robust reconstruction. A folding-based decoder deforms a 2D grid into 3D, enforcing manifold continuity while preserving geometric fidelity. Trained with Chamfer Distance (CD) and an orthogonality regularizer, the system is evaluated on ModelNet40 across varying Signal-to-Noise Ratios (SNRs) and bandwidths. Results show performance on par with SEmantic Point cloud Transmission (SEPT) at high bandwidth and clear gains in bandwidth-constrained regimes, with consistent improvements in both Peak Signal-to-Noise Ratio (PSNR) and CD. Ablation experiments confirm the benefits of orthogonalization and the folding prior.

Paper Structure

This paper contains 23 sections, 15 equations, 5 figures.

Table of Contents

  1. Introduction
  2. System Model
  3. Point Cloud Semantic Communication System
  4. Evaluation Metrics for Point Cloud Transmission
  5. D1 (Point-to-Point Error)
  6. D2 (Point-to-Surface Error)
  7. PSNR for D1 and D2
  8. CD
  9. Proposed Method
  10. DeepJSCC Encoder
  11. Semantic Encoder
  12. Channel Encoder
  13. DeepJSCC Decoder
  14. Channel Decoder
  15. Semantic Decoder
  16. ...and 8 more sections

Figures (5)

  • Figure 1: Overview of the proposed wireless point cloud semantic transmission system.
  • Figure 2: Illustration of the folding mechanism. A 2D grid prior is deformed into a 3D point cloud.
  • Figure 3: Comparison with SEPT bian2024wireless and ablation results. Subfigures (a)–(c) plot D1 (point-to-point), D2 (point-to-surface), and Chamfer Distance (CD) versus SNR. Legend entries of the form m–n denote the method name (m) and the available channel bandwidth (n). (When D1/D2 are reported as PSNR, higher is better; for CD, lower is better.)
  • Figure 4: Visualization of feature pool orthogonalization. In order to better visualize orthogonality, all values are plotted in absolute magnitude and clipped to a fixed color scale.
  • Figure 5: Orthogonal metrics under different SNRs.