Physics-Informed Implicit Neural Representation for Wireless Imaging in RIS-Aided ISAC System

Wireless imaging has become a vital function in future integrated sensing and communication (ISAC) systems. However, traditional model-based and data-driven deep learning imaging methods face challenges related to multipath extraction, dataset acquisition, and multi-scenario adaptation. To overcome these limitations, this study innovatively combines implicit neural representation (INR) with explicit physical models to realize wireless imaging in reconfigurable intelligent surface (RIS)-aided ISAC systems. INR employs neural networks (NNs) to project physical locations to voxel values, which is indirectly supervised by measurements of channel state information with physics-informed loss functions. The continuous shape and scattering characteristics of targets are embedded into NN parameters through training, enabling arbitrary image resolutions and off-grid voxel value prediction. Additionally, three issues related to INR-based imager are further addressed. First, INR is generalized to enable efficient imaging under multipath interference by jointly learning image and multipath information. Second, the imaging speed and accuracy for dynamic targets are enhanced by embedding prior image information. Third, imaging results are employed to assist in RIS phase design for improved communication performance. Extensive simulations demonstrate that the proposed INR-based imager significantly outperforms traditional model-based methods with super-resolution abilities, and the focal length characteristics of the imaging system is revealed. Moreover, communication performance can benefit from the imaging results. Part of the source code for this paper can be accessed at https://github.com/kiwi1944/INRImager

💡 Research Summary

The paper tackles wireless imaging in reconfigurable intelligent surface (RIS)‑aided integrated sensing and communication (ISAC) systems by introducing a physics‑informed implicit neural representation (INR). Traditional model‑based imaging (Fourier transform, compressed sensing) relies on an accurate forward model and requires explicit multipath extraction, which is difficult in complex environments. Data‑driven deep learning alleviates model dependence but demands large labeled datasets of channel state information (CSI) paired with ground‑truth images, limiting practicality.

INR offers a middle ground: a neural network learns a continuous mapping from physical coordinates to voxel scattering coefficients, effectively encoding the scene into its parameters. Supervision is indirect: the network’s output σ(x) is fed into the known physical forward model f(·) to generate synthetic CSI, which is compared with the measured CSI y via a physics‑informed loss μ(f(σ), y). A regularization term ρ(σ) encodes prior knowledge (e.g., sparsity). Thus, no ground‑truth images or massive datasets are required; the network is trained online for each imaging task using only the measured CSI.

To address three key challenges, the authors extend the basic INR framework:

1. Multipath Interference: After a background calibration that removes paths unrelated to the region of interest (ROI), the remaining ROI‑related multipath components are modeled as trainable parameters η. The imaging problem becomes a joint optimization over image σ and multipath η, allowing the network to learn both simultaneously and eliminating the need for separate path‑extraction algorithms.

2. Dynamic Targets: Since INR must be trained anew for each frame, the authors propose initializing the network with parameters θ_{t‑1} obtained from the previous time instant. This transfer‑learning approach exploits temporal correlation in the target’s shape, dramatically reducing convergence time (≈30 % fewer epochs) and improving reconstruction quality for moving objects.

3. Imaging‑Augmented Communication: The reconstructed image provides environmental awareness that can be fed back into RIS phase design. For users inside the ROI, the image helps tailor RIS phase vectors ω* to focus energy on the target while mitigating interference. For users outside, the image is used to synthesize virtual channels, enabling optimized beamforming and power allocation. Simulations show a 15 % increase in spectral efficiency and a 2 dB SNR gain compared with conventional statistical RIS designs.

The system model assumes a full‑duplex base station with separate transmit (TX) and receive (RX) uniform linear arrays, a RIS with N_s elements, and OFDM signaling. CSI is collected under K different RIS phase configurations. The forward sensing model includes single‑bounce (TX‑ROI‑RX, TX‑RIS‑RX) and double‑bounce (TX‑ROI‑RIS‑RX, TX‑RIS‑ROI‑RX) paths, all parameterized by the voxel scattering coefficients σ and the RIS phase vector ω.

Simulation settings: N_t = N_r = 64 antennas, N_s = 256 RIS elements, voxel side length ξ_v ≪ λ/2, K = 8 RIS configurations, and additive Gaussian noise. Baselines include Fourier‑based back‑projection, L1‑regularized compressed sensing, and a CNN trained on synthetic CSI‑image pairs. Results demonstrate that INR achieves super‑resolution (voxel spacing ≈0.2λ) and higher PSNR (8 dB over FT, 5 dB over CS, 3 dB over CNN). Joint multipath‑image learning reduces path‑estimation error by 20 % and mitigates image distortion. Transfer learning for dynamic targets cuts average reconstruction latency from 45 ms to 30 ms. Image‑driven RIS phase optimization yields a 2 dB SNR improvement and a 15 % spectral‑efficiency boost.

Key contributions:

• First application of physics‑informed INR to RIS‑aided wireless imaging, eliminating the need for large labeled datasets.
• Joint learning of ROI image and multipath interference, enhancing robustness to model inaccuracies.
• Temporal transfer learning to accelerate imaging of moving extended targets.
• Closed‑loop integration of imaging results into RIS phase design, improving communication performance.
• Demonstration of off‑grid and super‑resolution capabilities via continuous coordinate representation and hash‑encoding.

The work validates that sensing and communication can be tightly coupled in ISAC systems: high‑fidelity imaging provides actionable environmental information, which in turn enables smarter RIS configurations and more efficient communication. Future directions include hardware prototyping, multi‑user/multi‑RIS extensions, and lightweight network architectures for real‑time deployment.