Multi-Task Learning and Complex Neural Network for Integrated Sensing and Communication

Deep learning is expected to be a viable solution for harnessing the capability of integrated sensing and communication (ISAC). Despite significant advances made, existing DL models primarily treat sensing and communication as separate tasks, missing opportunities for joint performance enhancement. Moreover, current DL methods process complex-valued wireless signals by treating the real and imaginary components independently, overlooking the inherent correlations within complexvalued data. Motivated by these challenges, this work develops a novel ISAC framework, MTL-CNET, which leverages multitask learning (MTL) and complex neural networks (CNET), to effectively exploit the properties of complex-valued signals and enable concurrent sensing and communication. Specifically, we design a novel three-dimensional complex convolutional neural network (3D-CCNN) module to handle complex-valued wireless signals. The 3D-CCNN captures the characteristics of real and imaginary components within signals and the intrinsic interrelations between them for comprehensive feature extraction. Next, we propose an MTL with information exchange (MTL-IE) module to effectively leverage extracted information and explore inherent connections across tasks, facilitating simultaneous sensing and communication. Extensive experimental results verify that our MTL-CNET outperforms prior state-of-the-art methods.