GenAI-Enhanced Digital Twins for Predictive Interference Management in Ultra-Dense Networks

Ultra-dense indoor next-generation networks suffer severe interference from mobility-induced blockages and localized multi-user hotspots that conventional digital twins~(DTs) cannot anticipate. We propose a generative AI~(GenAI)-enhanced DT framework employing a conditional generative adversarial network~(cGAN) with a spatio-temporal generator and PatchGAN discriminator for proactive rare-event channel synthesis. A worst-case zero-forcing~(WC-ZF) beamformer driven by Monte Carlo synthetic trajectories realizes distributionally robust precoding, with control-channel overhead bounded to $\approx$2.1,kB per 10,ms slot. Sionna-based simulations confirm a 5–8,dB median signal-to-interference-plus-noise-ratio (SINR) gain, 60–70% packet-loss reduction, and 60–85% closure of the perfect channel state information (CSI) oracle gap within a 2.8–4.1,ms inference overhead.