Multi-View Wireless Sensing via Conditional Generative Learning: Framework and Model Design

TL;DR

This work addresses high-precision target sensing in ISAC by leveraging multi-view uplink CSI from multiple BSs and UEs. It proposes Gen-MV, a conditional generative framework that fuses multi-view CSI through a family of encoders and uses a diffusion-based generator conditioned on a latent target code, with a shape-EM weighted loss to jointly recover geometry and EM properties. Key contributions include four multi-view encoders (VS-MLP, MV-BiLSTM, MVT, IVT), a multiplicative view-position embedding, and a diffusion-based reconstruction with a normalizing-flow prior, plus ablations confirming the value of physical priors and loss balancing. The results show flexible handling of varying view counts, robustness to CSI disturbances, and improved reconstruction quality, highlighting the practical potential of integrating physics-informed generative learning for multi-view wireless sensing.

Abstract

In this paper, we incorporate physical knowledge into learning-based high-precision target sensing using the multi-view channel state information (CSI) between multiple base stations (BSs) and user equipment (UEs). Such kind of multi-view sensing problem can be naturally cast into a conditional generation framework. To this end, we design a bipartite neural network architecture, the first part of which uses an elaborately designed encoder to fuse the latent target features embedded in the multi-view CSI, and then the second uses them as conditioning inputs of a powerful generative model to guide the target's reconstruction. Specifically, the encoder is designed to capture the physical correlation between the CSI and the target, and also be adaptive to the numbers and positions of BS-UE pairs. Therein the view-specific nature of CSI is assimilated by introducing a spatial positional embedding scheme, which exploits the structure of electromagnetic(EM)-wave propagation channels. Finally, a conditional diffusion model with a weighted loss is employed to generate the target's point cloud from the fused features. Extensive numerical results demonstrate that the proposed generative multi-view (Gen-MV) sensing framework exhibits excellent flexibility and significant performance improvement on the reconstruction quality of target's shape and EM properties.

Figures (12)

• Figure 1: The considered multi-view uplink sensing scenario.
• Figure 2: The proposed Gen-MV sensing framework, which consists of a multi-view channel encoder $q_{\boldsymbol{\varphi}}(\boldsymbol{z}|\mathcal{H})$, a latent space prior $p_{\phi}(\boldsymbol{z})$, and a conditional generative model $p_{\boldsymbol{\theta}}(\boldsymbol{x}|\boldsymbol{z})$. Here, we employ a flow-based model $F_{\boldsymbol{\phi}}$ to parameterize the prior $p_{\boldsymbol{\phi}}(\boldsymbol{z})$, and implement a conditional point cloud diffusion model $p_{\boldsymbol{\theta}}(\mathcal{X}^{(0)}|\boldsymbol{z})$ for target reconstruction, where $\mathcal{X}^{(0)}$ is the point cloud representation of the target properties $\boldsymbol{x}$.
• Figure 3: Four structural designs for the multi-view channel feature encoder $E$. All feed-forward networks are two-layer MLPs with a hidden dimension of $2d_{\mathrm{v}}$.
• Figure 4: Generative reconstruction process of a target based on the conditional point cloud diffusion model, where the multi-view channel encoder is implemented with IVT structure.
• Figure 5: Reconstruction results of several targets using four multi-view channel encoder architectures (i.e. VS-MLP, MV-BiLSTM, MVT, and IVT). The relative permittivity $\varepsilon_r$ and conductivity $\sigma$ at each point position are visualized by color mapping.