DNN-Powered SIC-Free Receiver Artificial Noise Aided Terahertz Secure Communications With Randomly Distributed Eavesdroppers

Despite the narrowbeam nature of Terahertz (THz) communications, the physical layer security in the THz band is challenging when eavesdroppers are inside the beam radiation sector. In this paper, a deep neural network (DNN)-based self-interference cancellation (SIC)-free receiver artificial noise (AN) assisted mechanism is proposed to address the in-beam security challenge, by considering randomly distributed eavesdroppers in THz secure communications. By exploiting the different temporal broadening effects of the AN signals at distinct distances, the SIC can be saved with a proper signal detection design rather than using conventional high-complexity cancellation techniques. To combat the non-linearity and non-convexity of the optimization problem, the system parameters including carrier frequency, power of transmission signal and AN signal power, and frame time are designed by an efficient deep neural network (DNN) algorithm to minimize the secrecy outage probability. Numerical results demonstrate that the maximum secrecy rate of our proposed DNN-powered SIC-free receiver AN scheme is up to 3.3 bps/Hz over 10 m transmission when the eavesdropper is in close proximity. Moreover, the secrecy outage probability is less than 0.5% when the eavesdropper density is 0.001 per square meter, which is approximately 33% lower than that of conventional transmitter AN schemes.