To Avoid or Not to Avoid CSI Leakage in Physical Layer Secret Communication Systems

Ta-Yuan Liu,Pin-Hsun Lin,Shih-Chun Lin,Y. -W. Peter Hong,Eduard Axel Jorswieck

DOI: https://doi.org/10.1109/mcom.2015.7355561

IF: 9.03

2015-01-01

IEEE Communications Magazine

Abstract:Physical layer secrecy has attracted much attention in recent years due to its ability to ensure communication secrecy with the use of channel coding and signal processing techniques (and without the explicit use of secret keys) in the physical layer. It serves as a promising technique for highly dynamic or ad hoc systems such as device-to-device and machine-type communication systems. However, the achievable secrecy performance depends highly on the level of CSI at the transmitter, the receiver, and the eavesdropper. In this article, we discuss how different levels of CSI resulting from conventional and unconventional ways of performing training and channel feedback may affect the confidentiality in terms of the information-theoretic (perfect) secrecy rate. The conventional approach refers to the emission of pilot signals from the transmitter and explicit channel feedback from the receiver. This approach is backward compatible with existing systems and allows the receiver to obtain accurate knowledge of the CSI, but may suffer from CSI leakage toward the eavesdropper. Unconventional approaches capitalize on reverse training to prevent CSI leakage and are shown to achieve significant improvements over conventional schemes in certain cases. For example, in a system with four transmit antennas and a single antenna at both the receiver and the eavesdropper, a secrecy rate gain of approximately 0.8 b/channel use at transmit SNR of 16 dB is observed over the full CSI case by providing CSI only to the transmitter (but not the receiver and the eavesdropper).