Active Secure Coding Based on Eavesdropper Behavior Learning
The secrecy capacity achieving problem of the wiretap channel against an active eavesdropper with unlimited computational power over is an important foresighted task for secure communication. For active wiretap channel, the effectiveness of cryptography embedded secure coding schemes are limited due to the passive problem of physical layer coding. Thus in this paper, a novel solution called active secure coding scheme is proposed which combines physical secure coding with machine learning to implement an active defense against the active eavesdropper. To construct an universal active method for secure coding, an abstract active wiretap channel model is constructed under the detectable precondition, in which hidden Markov model is employed to build the internal eavesdropper behavior pattern (the stochastic process for eavesdropper behavior states) with stochastically mapped external observations. Based on the abstract model, an eavesdropper behavior pattern learning and eavesdropper behavior states decoding are constructed for estimating the optimal eavesdropper behavior states, which enables the secure coding scheme to respond the eavesdropper actively. Then this active secure coding method is performed to the general varying wiretap channel to construct an explicit active secure polar coding scheme. As proofed, the proposed active secure polar coding scheme can theoretically achieve the average secrecy capacity of t times secure transmissions under the reliability and strong security criterions.
READ FULL TEXT