Deep Learning-Based Proactive Eavesdropping for Wireless Surveillance

In this work, we investigate a proactive eavesdropping system where a central monitor covertly wiretaps the communications between a pair of suspicious users via multiple intermediate nodes. For successful eavesdropping, it is required that the eavesdropping channel capacity is higher than the data rate of the suspicious users so that the central monitor can reliably decode the intercepted information. Hence, the intermediate nodes operate in two different modes, namely eavesdropping mode and jamming mode, to facilitate eavesdropping. Specifically, the eavesdropping nodes forward the intercepted data from the suspicious users to the central monitor, while the jamming nodes transmit jamming signals to proactively control the data rate of the suspicious users. We propose an efficient deep learning-based approach to identify the optimal mode selection for the intermediate nodes and the optimal transmit power for the jamming nodes. Numerical results confirm the significant performance gain of our proposed method both in terms of performance and time complexity over conventional schemes.