Research On Detection Algorithm Of Information Hiding At Physical Layer In Wireless Communications

As people pay more attention to the security in communication systems,information hiding has attracted much attention from researchers.Information hiding is a secret communication technology that embeds secret message into a normal communication carrier,which cannot be detected by an attacker.The detection of information hiding is a reverse process of information hiding,which is used to detect the probability of the existence of hidden message.This thesis mainly researches on the detection algorithm of the information hiding scheme at physical layer,where there are two main applications: it can be used to evaluate the security level of an information hiding scheme;it can be used to detect the possibility that an adversary uses the information hiding scheme to transmit illegal information.So far,most of the research on information hiding and detection is concentrated in upper-layers,such as digital media.Since the physical-layer introduces randomness on the observed signal due to channel fading and receiver noise,it is more difficult to detect the information hiding scheme at physical-layer than that at upper-layer.There are two main approaches for detecting the information hiding scheme at physicallayer: heuristic based detection and machine learning based detection.However,they have the following drawbacks.For the heuristics-based detection,it lacks a complete framework of theoretical analysis,and it is difficult to give a quantitative theoretical analysis.For the machine learning based detection,it is trained only in a specific environment,so it cannot be directly applied to other environments,which lacks of robustness.This thesis focuses on two embedded strategies in the information hiding scheme at physical-layer: replacement strategy and superimposed strategy.This thesis first studies the replacement strategy.We assume that the secret communication parties share a key,the embedded location of the hidden signal is selected according to the key.If the bit information of the carrier signal is different from that of the hidden signal,the secret sender directly replaces the bit information of the carrier signal by the bit information of the hidden signal.This information hiding scheme method will inevitably increase the negative impact on the reception quality of the carrier signal after embedding the hidden signal,such as the increase of the demodulation error rate.We assume that a monitor neither knows the key nor whether there is a hidden signal in the received signal,but it can detect the communication parameters of the normal carrier signal.Using the fact that physical-layer information hiding scheme does not sacrifice the decoding error rate,we recover the carrier signal from the output of an equalizer without error through demodulating,decoding,re-encoding,and re-modulating.By comparing the recovered carrier signal and the received signal,we obtain the optimal detection solution for the replacement strategy by a matching method.However,there are two challenges for obtaining the optimal solution: accurate estimation of receiver noise variance and channel state information(CSI).For the receiver noise variance estimation,we address this challenge by utilizing the fact that the receiver noise and replacement operations have different effects on the carrier signal,and design the grouping estimation method to estimate the receiver noise variance with low complexity.For CSI estimation,we address this challenge by designing two detection algorithms: noise-grouping based detection algorithm and constellation-distance based detection algorithm.Both detection algorithms allow the presence of CSI estimation errors.Through experiments,we can see that the noise-grouping based detection algorithm has better detection performance,but the constellation-distance based detection algorithm can even pinpoint the embedded locations of the hidden signal.Since the superimposed strategy disguises the distribution characteristic of hidden signal as that of noise,it has a higher detection difficulty than the replacement strategy.This thesis uses the machine learning method to design the detection algorithm,e.g.,the support vector machine(SVM)detection algorithm.The key of like machine learning methods is to design the sequence of input features for training model.This thesis constructs two input features: based on the output signal of an equalizer and based on the output signal of a matching filter.After training,we obtain the classifier to detect the hidden signal in the superimposed strategy.Through experiments,we find that the SVM detection algorithm with the output signal of a matching filter has obvious advantage over that with the output signal of an equalizer.