Research On Physical Layer Security For Finite Alphabet Input

Wireless physical layer security is to exploit physical characteristics of the channels between the transmitter and the legitimate user and the eavesdropper to achieve secure communications,such as the difference,randomness and so on.Nowadays a majority of researches on physical layer security from the information theoretic point is based on the constraints of Gaussian continuous input.In fact,in wireless communication systems,practical signals are non-Gaussian and are often of a finite alphabet,such as quadrature amplitude modulation(QAM)and phase-shift keying(PSK).Thus,existing results may substantially disagree with what are inferred from the Gaussian input assumption.In order to illustrate the important difference between finite alphabet input and Gaussian input,this paper has proven that the secrecy rate is finite and non-concave in the wiretap model.In the mean time,because of the lack of the closed-form expression of mutual information,this paper has also introduced the relationship between the information theory and the signal estimation theory,which has been expressed by the identity of mutul information and minimum mean square error(MMSE),for the sake of the research convenience.Then,according to the Multiple-Input Single-Output(MISO)wiretap channels,this paper has studied the aritificial noise scheme and corresponding power allocation issue under the constrains of finite alphabet input.The power allocation between the informationbearing signal and aritificial noise is to maximize ergodic secrecy rate.To avoid calculating mutual information directly,this paper has proposed a power allocation scheme with much lower computational complexity.By using the relationship between mutual information and received MMSE,the maximization problem can be converted into a zero-searching problem.Then the power allocation ratio can be obtained via gradient searching algorithm.Numerical examples show that the proposed scheme serves as a good approximation to numerical optimal solution.Furthermore,in order to illustrate how the power allocation rate influence the secrecy performance,we derived the approximate expressions of secrecy rate.For another aspect,precoding has also been studied in this paper in the constrains of finite alphabet input.It can be found that the precoding design based on generalized singular value decomposition is not optimal in the assumption of finite alphabet input.This paper also proved the necessary conditions of the optimal precoding,and proposed the precoding scheme based on the gradient decent algorithm.On the assumption that eavesdropper's CSI is unknown,we futher proposed a secure precoding scheme.