Variational Autoencoder-Based Intelligent Cooperative Integrated Sensing and Communication System

This paper proposes a collaborative variational autoencoder (VAE)-based target fusion method for sixth-generation (6G) intelligent cooperative integrated sensing and communication (IC-ISAC) systems using orthogonal frequency division multiplexing (OFDM) signals. A key challenge in multi-base station (multi-BS) ISAC networks is the target segmentation issue caused by multiple independent base stations (BSs) generating different measurements for the same target, which complicates the process of enhancing sensing performance for low-altitude aerial targets. To address this, the proposed system leverages the VAE-based intelligent network framework and distributed multiple-input multiple-output (MIMO) radar techniques to achieve robust target perception. By utilizing the compression and generation capabilities of the VAE-based cooperative ISAC architecture, raw sensing data from multiple cooperative BSs are encoded into latent vectors through a deep neural network. These latent representations are then transmitted via the backhaul and processed by a generative network at the fusion center to reconstruct the spatial extent and placement of extended targets. This approach effectively reduces backhaul resource consumption by compressing sensing data while preserving critical electromagnetic scattering features of the targets. Numerical simulations and network training results demonstrate that the proposed method significantly improves multi-target recognition accuracy and computational efficiency compared to traditional multistatic collaborative ISAC approaches.