Vertical Federated Learning for Multicell Integrated Sensing and Communication Systems
Lai Jiang, Kaitao Meng, Christos Masouros · 2025
Beamforming is a crucial technique to enable dual-functionality for integrated sensing and communication (ISAC) systems by leveraging the multi-antenna arrays. Most existing beamforming strategies focus solely on single-base station (BS) scenarios, neglecting the impact of inter-cell interference (ICI). However, managing ICI often requires additional information exchange, which results in expensive communication overhead and extra latency. To address this problem, we investigate cooperative beamforming in multi-cell ISAC systems and formulate an optimization problem that jointly maximizes the weighted sum of communication rate and radar information rate. Considering the computational efficiency, we propose to apply the vertical federated learning (VFL) technique to solve the optimization problem with limited transmission overhead. In our approach, neural networks are distributed across BSs and jointly trained in an unsupervised manner, then the trained neural networks can perform real-time beamforming with local channel information. Meanwhile, we show that our method can achieve distributed ICI managing and find the globally optimized beamforming solutions for each BS. Through numerical simulations, we demonstrate that our beamforming solution outperforms the benchmarks in terms of both communication rate and radar information rate while significantly reducing computational and communication costs.