Research On Physical Layer Security Theory And Method In Wireless Communication

With the popularity and development of wireless communication, wireless security has already become a crucial requirement. However, the broadcast nature of wireless channel makes it vulnerable to attacks, such as eavesdropping, spoofing and message tampering. Even though traditional security schemes are widely ap-plied to the existing wireless communication systems, these schemes can hardly meet the security requirement of future wireless communication caused by problems of key distribution difficulties, and inapplicability to resource-constrained large-scale wireless networks. Recently, physical layer security has become a hot re-search topic. It provides a novel security scheme from the perspective of information theory, which simplifies key distribution and management greatly. Thus, it is of great importance for both theory and practice to study theory and method of physical layer security in wireless communication.In the present dissertation, we study two main research contents in physical layer security, i.e. key-less security and key generation. To overcome the problems of imperfect channel state information (CSI) of the eavesdropper and to resist the attack based on blind source separation in key-less security, we propose two novel optimization designs of ANTR (Artificial Noise Transmitted by Receiver). Then, to overcome key gen-eration problems of the non-reciprocal channel measurement parameters and correlation of channel character-istics in time, space and frequency domain, we propose two key generation approaches based on log-domain differential transform (LDDT) and based on principle component analysis (PCA), respectively. Finally, we build a general mathematical model of Artificial Randomness (AR) aided key generation approaches and prove that the AR aided model can improve key generation rate.The innovative achievements of the thesis include:1. We propose a robust ANTR optimization approach under imperfect CSI.Existing ANTR approaches assume that Eve's CSI is perfectly known by Alice. However, in practical,Alice's estimation of Eve's CSI is always inaccurate. We jointly optimize covariance matrix of beamforming vector and ANTR to achieve user's quality of service (QoS). Employing upper bound of channel uncertainty,the original non-convex optimization problem is approximated to be a SDP problem. Then, we derive the optimal transmission approach. Furthermore, we prove that the optimal ANTR covariance must be of rank one. Simulation results show that under imperfect CSI, our proposed ANTR approach can improve the per-formance of security capacity by 2dB,compared with existing ANRT approaches.2. We study ANTR optimization approaches to resist blind source separation attack.When transmitted signal has temporal structure,Eve can steal it by using blind source separation at-tack in existing ANTR approaches. To resist this attack, we design a novel ANTR model which has certain correlation with transmitted signal. Considering conditions of blind source separation based on the second-order-statistics, we propose two ANTR approaches based on optimizing correlation coefficient and randomly changing the correlation coefficient. Simulation results show that when transmitted signal is speech signal and artificial noise is independent Gaussian noise, the correlation coefficients of reconstructed signal by us-ing blind source separation attack and transmitted signal is up to 90%. Our proposed AN-TR approaches can reduce the correlation coefficients to be less than 50%. When transmitted signal is BPSK signal, the bit error rate (BER) is kept around 0.5.3.We propose a novel log-domain differential (LDDT) transform to eliminate the hardware fingerprint(HF) deviationDifferent wireless devices have different HFs, and the HF deviation can cause the non-reciprocity of channel measurement parameters among legitimate users. We first turn the multiplicative HF into additive one, and then eliminate the HF deviation by using a differential operation in time domain according to the fact that HF is constant in frequency within a short time. The proposed LDDT approach could perfectly eliminate the effect of HF deviation in noise free cases. Considering LDDT's sensitivity to noise, we further improve LDDT by using zero-padding technique. Simulation results show that the BER performance of LDDT approaches is the same with ideal case (without HFD). Specifically, when HF deviation is 0.01, the BERs of exist approaches are around 10-2, while LDDT reduces BER to be 10-4.4. We propose an optimal linear preprocessing approach based on PCA.Channel measurement parameters are autocorrelated in time, space and frequency domain. Besides,reciprocity of channel measurement parameters is very weak in low SNR scenarios. To overcome these prob-lems, we build a general model for linear preprocessing approaches, derive the expression of key capacity and prove that PCA is the optimal linear preprocessing approach to achieve the maximal key capacity. Further-more, we show that using the same eigenvector can improve key agreement with low computation complexity and negligible information leakage in key generation approach based on PCA. Simulation results show that key generation approach based on PCA can improve key agreement efficiency. Specifically, when key gen-eration rate is 10 kb/s, SNR is 10 dB, the key error rate of PCA is around 10-3. Besides, PCA can achieve a pass rate of 0.825 in NIST tests, which outperforms other preprocessing approaches.5. We prove that AR aided key generation approach can improve key generation rate.Wireless key generation in quasi-static channels is difficult due to the limited channel variability and randomness. We build a general mathematical model for Artificial Randomness (AR) aided key generation approaches. From the perspective of information theory, we prove that AR can improve key generation rate in single antenna systems when Eve does not her CSI accurately. The improved key rate is bounded by the channel condition. Besides, key rate improved by AR is affected by the precoding method of transmitter in multiple antennas systems. The simulation results show that when Eve can estimate her CSI accurately, the improved key rate is 0. When Alice knows Eve's CSI, we can achieve the upper bound of the improved key rate.