Research On Blind Channel Equalization And Signal Identification In Complicated Electromagnetic Environments

As important parts in the radio monitoring field,channel blind equalization and signal identification have vital significances in both military and civil applications.Recently,with the explosive increase of wireless devices and wireless data,the spectrum resource has been more crowded,and the electromagnetic environment has been more complicated.Besides,the development of wireless communication also gives the rise to cheap and easy design of illegal interference source.In complicated electromagnetic environments,these illegal broadcasting stations severely interfere and disrupt both the military and civil wireless communication service.However,conventional radio monitoring methods usually degrade heavily or even fail in complicated electromagnetic environments,especially in non-Gaussian impulsive noise.Effective and intelligent blind equalization and signal identification still remain challenges in complicated electromagnetic environments.Hence,aiming at coping with these difficulties,in this paper,we carry out deep and systematical research on the channel blind equalization,automatic modulation classification in non-Gaussian noise,asynchronous modulation classification and unauthorized broadcasting identification,and then explore several novel channel blind equalization and signal identification methods.The main contributions are as follows:(1)We propose four novel blind equalization algorithms to improve the performance of existing methods in non-Gaussian impulsive noise.Firstly,we use the fractional low order statistics(FLOS)to suppress the impulsive noise,and then design a variable step size function to balance the convergence speed and steady-state error.Further,we propose a variable step-size modified blind equalization algorithm.Then,through introducing decision mechanism to solve the phase rotation problem,we design the cost function in parallel to form novel blind equalization algorithm on the basis of probability density function and FLOS.Thirdly,through the designed double threshold weighted decision method,we combine the Renyi entropy and FLOS to form cost function,and then propose a robust double mode blind equalization algorithm.Finally,based on bounded non-linear function,we design the cost function and employ quasi Newton method to perform fast blind equalization.These four blind equalization methods can effectively suppress non-Gaussian impulsive noise and meanwhile has fast convergence speed.(2)We propose three novel automatic modulation classification algorithms to improve the performance of existing methods in non-Gaussian noise.Firstly,we employ the cyclic correntropy spectrum to eliminate the impact of impulsive noise,and demonstrate that cyclic correntropy spectrum can be used to classify different modulation types.Besides,the principal component analysis method is used to optimize the extracted features and then radial basis neural network is utilized as classifier.Secondly,we employ the hyperbolic-tangent cyclic spectrum to suppress the impulsive noise,and then on the basis of deep residual networks,we propose an end-to-end automatic modulation classification method with great improvement in recognition accuracy.Finally,on the basis of bounded non-linear function,we propose the generalized cyclic spectrum to suppress the non-Gaussian impulsive noise,and the principal component analysis method is used to reduce the feature dimension.Then,support vector machine is used to perform modulation classification.The experiment results show that the aforementioned methods have an outstanding performance in non-Gaussian impulsive noise.(3)We propose a novel asynchronous automatic modulation classification method with Impulsive Noise via complex correntropy and one-dimensional convolutional neural network(Conv1D).Firstly,through using complex correntropy function,the non-Gaussian noise can be effectively eliminated and the corresponding features can be extracted for asynchronous modulation classification.Then,a detailed theoretical analysis is conducted to demonstrate the effectiveness of complex correntropy in feature extraction and impulsive noise suppression.Next,Conv1D is employed to identify different modulation schemes.Both simulation and implementation results show that the proposed methods have an outstanding performance.(4)We propose a novel identification approach based on long short-term memory(LSTM)recurrent neural network and Lab View software to develop an intelligent and efficient unauthorized broadcasting identification system.In this system,firstly,a series of Lab VIEW applications are developed to drive USRP 2930s for the acquisition of broadcasting signals.Then,the LSTM identification network is proposed to recognize unauthorized broadcasting,which can effectively extract the distinguishing features,such as channel state information,RF device fingerprinting and other features.Finally,the well trained LSTM network is invoked by LabVIEW applications to realize unauthorized broadcasting identification.Experimental implementation results show that the proposed method has an outstanding performance.The above work makes a beneficial exploration in the radio monitoring,which enriches the theory of blind equalization and signal identification in complicated electromagnetic environment,and improves the feasibility in the realistic system.Besides,it has theoretical significance and practical applications.