Spatio-Temporal Parameter Prediction for Dynamic ISAC Network Based on Diffusion Model

Jinlei Xu

doi:10.1109/mnet.2026.3725233

Spatio-Temporal Parameter Prediction for Dynamic ISAC Network Based on Diffusion Model

2026-08-28

0

PRE

AI

DOI：10.1109/mnet.2026.3725233

Original

Original request for help

Save

Abstract

En 中文

For highly dynamic integrated sensing and communication (ISAC) networks, modeling the evolution of target parameters can serve as a prior to reduce the latency and signaling overhead. However, the rapid variation and strong coupling pose significant challenges to accurate detection and prediction. Generative artificial intelligence (GAI), with its strong non-linear modeling generalization capabilities, can provide a probabilistic paradigm to calibrate uncertainty for spatio-temporal parameter prediction (STPP). Motivated by this, we first investigate the principles of several representative GAI models and explore their potentials in STPP through illustrative application cases. Particularly, by combining the sparse channel characteristics of orthogonal time, frequency and space, we highlight the advantages and adaptability of diffusion model (DM) for STPP in highly dynamic scenarios. On this basis, we develop a DM-based generative prediction network for highway-vehicle STPP, which leverages an iterative denoising process to effectively predict the target parameters. Simulation results validate the superiority of DM in terms of prediction accuracy and robustness. Finally, we outline the future research directions on applying DM for STPP in dynamic ISAC networks.

Keywords:

Modeling

Integrated sensing and communication

Noise reduction

Timing

Educational institutions

Evolution (biology)

Sequences

Sequential analysis

Transformers

Generative adversarial networks

Spatio-Temporal Parameter Prediction for Dynamic ISAC Network Based on Diffusion Model

Abstract

Journal

Organization

Cited Papers