Predictive Digital Twin-Assisted Geometry-Aware Uncertainty-Adaptive Beamforming for ISAC

Abstract

Integrated Sensing and Communication (ISAC) is becoming an important part of sixth-generation (6G) wireless networks, where the same radio resources are used for both data transmission and environmental sensing. In practical systems, however, beamforming performance depends strongly on accurate and frequently updated channel state information (CSI). This requirement increases signaling overhead and makes the system vulnerable to mobility, blockage, and multipath effects. To address this problem, this paper presents a Predictive Digital Twin-Assisted Geometry-Aware Uncertainty-Adaptive Beamforming (GA-UAB) framework for robust ISAC operation when digital twin information is imperfect. The proposed system uses a geometry-aware digital twin with ray-tracing-based line-of-sight (LoS) and non-line-of-sight (NLoS) path inference for a $16\times 16$ MIMO indoor ISAC scenario. Dual Kalman filters are used to separately track and predict the future states of the user and target, enabling the framework to account for mobility and outdated environmental information. The predicted state covariance is then used to estimate geometry uncertainty, which drives adaptive steering-angle correction and beamwidth control. The framework is evaluated through 2000 Monte Carlo realizations incorporating obstacles, reflections, mobility, and measurement dropouts, with conventional digital-twin beamforming, uncertainty-aware beamforming (UAB), and a Full-CSI (genie-aided) benchmark used for comparison. The proposed predictive GA-UAB achieves a median sensing SNR of 22.24 dB, compared with 14.04 dB for conventional digital-twin beamforming and 18.99 dB for UAB. The proposed method reduces the outage probability to 23.40%, compared with 87.15% for conventional digital-twin beamforming and 61.20% for UAB. The Full-CSI benchmark achieves a median sensing SNR of 23.99 dB, placing GA-UAB only 1.75 dB below the genie-aided reference. These results demonstrate that combining predictive geometry estimation, uncertainty adaptation, and geometry-aware beamforming improves sensing reliability and robustness in dynamic DT-assisted ISAC environments. The present evaluation is based on MATLAB Monte Carlo simulations and does not include over-the-air hardware validation.