Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robust Super-Capacity SRS Channel Inpainting via Diffusion Models

Usman Akram, Fan Zhang, Yang Li, Haris Vikalo

TL;DR

This work addresses robust uplink CSI acquisition for 5G NR with sparse, non-uniform SRS masking by introducing a diffusion-based inpainting framework. The method jointly denoises and inpaints masked CSI, and integrates system knowledge through a log-likelihood gradient during inference, enabling a single model to generalize across deployment-time distortions. Two diffusion formulations are explored (variance-exploding SDE and variance-preserving SDE), with a ViT-based masked autoencoder architecture trained on masked inputs. Empirical results on CDL channels show the diffusion-based approach outperforms a UNet baseline and the one-step MAE under distribution shifts, achieving up to 14 dB NMSE improvement in challenging conditions, while remaining competitive under matched conditions. This demonstrates a robust, generalizable pathway for high-coverage SRS design in future wireless systems, despite the higher iterative inference cost compared to one-shot methods.

Abstract

Accurate channel state information (CSI) is essential for reliable multiuser MIMO operation. In 5G NR, reciprocity-based beamforming via uplink Sounding Reference Signals (SRS) face resource and coverage constraints, motivating sparse non-uniform SRS allocation. Prior masked-autoencoder (MAE) approaches improve coverage but overfit to training masks and degrade under unseen distortions (e.g., additional masking, interference, clipping, non-Gaussian noise). We propose a diffusion-based channel inpainting framework that integrates system-model knowledge at inference via a likelihood-gradient term, enabling a single trained model to adapt across mismatched conditions. On standardized CDL channels, the score-based diffusion variant consistently outperforms a UNet score-model baseline and the one-step MAE under distribution shift, with improvements up to 14 dB NMSE in challenging settings (e.g., Laplace noise, user interference), while retaining competitive accuracy under matched conditions. These results demonstrate that diffusion-guided inpainting is a robust and generalizable approach for super-capacity SRS design in 5G NR systems.

Robust Super-Capacity SRS Channel Inpainting via Diffusion Models

TL;DR

This work addresses robust uplink CSI acquisition for 5G NR with sparse, non-uniform SRS masking by introducing a diffusion-based inpainting framework. The method jointly denoises and inpaints masked CSI, and integrates system knowledge through a log-likelihood gradient during inference, enabling a single model to generalize across deployment-time distortions. Two diffusion formulations are explored (variance-exploding SDE and variance-preserving SDE), with a ViT-based masked autoencoder architecture trained on masked inputs. Empirical results on CDL channels show the diffusion-based approach outperforms a UNet baseline and the one-step MAE under distribution shifts, achieving up to 14 dB NMSE improvement in challenging conditions, while remaining competitive under matched conditions. This demonstrates a robust, generalizable pathway for high-coverage SRS design in future wireless systems, despite the higher iterative inference cost compared to one-shot methods.

Abstract

Accurate channel state information (CSI) is essential for reliable multiuser MIMO operation. In 5G NR, reciprocity-based beamforming via uplink Sounding Reference Signals (SRS) face resource and coverage constraints, motivating sparse non-uniform SRS allocation. Prior masked-autoencoder (MAE) approaches improve coverage but overfit to training masks and degrade under unseen distortions (e.g., additional masking, interference, clipping, non-Gaussian noise). We propose a diffusion-based channel inpainting framework that integrates system-model knowledge at inference via a likelihood-gradient term, enabling a single trained model to adapt across mismatched conditions. On standardized CDL channels, the score-based diffusion variant consistently outperforms a UNet score-model baseline and the one-step MAE under distribution shift, with improvements up to 14 dB NMSE in challenging settings (e.g., Laplace noise, user interference), while retaining competitive accuracy under matched conditions. These results demonstrate that diffusion-guided inpainting is a robust and generalizable approach for super-capacity SRS design in 5G NR systems.

Paper Structure

This paper contains 7 sections, 2 equations, 4 figures, 1 table, 2 algorithms.

Table of Contents

  1. Introduction
  2. System Model
  3. Proposed Method
  4. Algorithm Design
  5. Model Architecture
  6. Experimental Evaluation
  7. Conclusion

Figures (4)

  • Figure 1: SRS allocation under comb-4 pattern with $N=32$ subcarriers. (a) A single UE is assigned 8 uniformly spaced tones. (b) With sparse non-uniform masking and channel inpainting, the same resources support 4 users simultaneously.
  • Figure 2: NMSE vs. SNR for CDL-A/B/C channels. Metrics are computed over sounded and reconstructed subcarriers.
  • Figure 3: NMSE vs. additional masking ratio $r$ (%) at 20 dB SNR, with extra masking applied along subcarriers or both subcarriers and antennas.
  • Figure 4: Effect of clipping on NMSE at 20 dB SNR vs. clipping threshold (in standard deviation units).