Research On Adaptive Modulation And Automatic Modulation Classification Based On Deep Learning Techniques In Fading Channels

With the development of social economy,people gradually put forward the higher requirements for data communication,especially for the transmission rate and quality of communication.Different from economically and technologically developed areas,some developing countries and regions,like Pakistan and Afghanistan,don’t have rapid iteration of the communication system and have complex channel environment.How to maximize the data transfer rate on the limited bandwidth and how to improve the spectrum utilization of system in the fading channel become a key problem in these developing areas.And this case also raises a new problem on the communication theory and application technology of multi-generation communication system in the time-space overlapping scene.The adaptive modulation technology can adaptively change the modulation mode of the communication system according to the current channel state under the condition of the expected bit error rate,and make full use of wireless channel to improve the transmission reliability and transmission rate.However,in practical application,the performance of adaptive modulation is seriously affected by the complex fading channel environment.As one of the typical adaptive modulation technologies,Automatic Modulation Classification(AMC)can correctly identify and classify different modulation methods under complex time-varying channel conditions.However,the traditional AMC algorithm based on maximum likelihood has some disadvantages such as complex computation,too many unknown parameters and relies on transmitter prior information,which limit the performance of AMC algorithm.At the same time,how to develop the adaptive modulation technology and automatic classification technology under the scenario of multi-generation communication system,and seek new communication theory and technique become a core problem for realizing efficient communication in developing areas.Therefore,aiming at the actual communication environment where multi-generation communication systems coexist,this paper focuses on the adaptive modulation technology and AMC recognition technology of complex fading channel,and deeply studies the signal propagation principle and modulation recognition method of complex fading channel.At the same time,taking advantage of Deep Learning(DL)model in signal modulation recognition,a variety of AMC algorithms are proposed,which provides a feasible scheme for signal modulation recognition in complex fading channel environment.In this paper,the principles of various adaptive modulation techniques with different parameters(such as communication distance,signal-to-interference-plus-noise Ratio(SINR),etc.)are analyzed,which is the basis for studying signal propagation model and designing modulation recognition algorithm in complex environment.In order to realize high data rate communication in the multi-generation communication system,an adaptive modulation method is proposed based on the Standard Propagation Model(SPM),such as user SINR and communication distance.The bit error rate performance of modulation schemes in Additive White Gaussian Noise(AWGN)channel,Rayleigh channel and Rician channel with or without SPM technology is analyzed.The results show that the proposed method can effectively transmit data over a large distance with minimum bit error rate,alleviate the influence of fading channel,and optimize the available bandwidth.On this basis,the paper also proposes an adaptive modulation method based on the variable threshold of communication distance,and verifies the high data communication capability with low bit error rate in a multi-node cellular communication system with large range,thereby demonstrating the feasibility of the above method.Particularly in the developing areas with multi-generation communication systems,high-speed image communication becomes more and more important.The theoretical methods of multi-class automatic modulation and recognition are obviously suitable for practical scene.For solving the problem of multi-class automatic modulation classification in practical fading channels,this paper proposes a 2D Convolution Neural Network(2D-CNN)to implement multi-class AMC tasks.In this paper,we study a 2D image data set that can realize multi-class(24 class)communication modulation,and propose a new DL modulation classification framework based on 2D-CNN,which can analyze multi-class modulation recognition and classification with the minimum number of training samples.This paper studies the AWGN,Rayleigh and Rician fading channels,and considers Binary Phase Shift Keying(BPSK),Quadrature Phase Shift Keying(QPSK),16 Quadrature Amplitude Modulation(16-QAM)and 64 Quadrature Amplitude Modulation.Refer to the various modulation schemes,such as 64-QAM and 64 orthogonal amplitude modulation on 2D-CNN’s AMC technology,a 2D-CNN deep learning method combining batch normalization and Dropout is proposed.To evaluate the effectiveness of the proposed method,5-fold and 10-fold Cross Validation(CV)methods are used in independent data sets.In order to explore the impact of large amount of data on the AMC classification and 2D-CNN architecture,this paper proposes a robust method based on data augmentation model,such as translation,reflection,shear and random rotation,to improve the performance of the algorithm.The effectiveness of the method is verified by simulation experiment.Because the application scope of multi-class AMC method in 2D-CNN fading channel environment is limited.This paper proposes a novel AMC method based on 3Dconvolution Neural Network(3D-CNN)combining Discrete Cosine Transform(DCT)and spatial information.The method can learn from the results of data augmentation(such as random weak Gaussian blur,Random amplification/minification,random horizontal/vertical movement)or without data augmentation in the spatial domain.The recognition performance of the proposed method in AWGN,Rician and Rayleigh fading channels including BPSK,QPSK,16-QAM and 64-QAM,and the 3D-CNN multi-class recognition network trained with 10-fold CV was tested.The results show that the best performance is 3D-CNN AMC method without spatial augmentation,while the worst performance is 3D-CNN AMC method without frequency augmentation.In order to analyze the performance differences between 2D-CNN and 3D-CNN AMC methods under different fading channels,this paper uses CNNs to construct binary classification and multi-category classification systems,and uses unenhanced spatial learning method and pre-fuzzy technology before downsampling for binary classification and multi-category classification.Simulation experiments are performed to evaluate the performance of 16-QAM and 64-QAM modulated signals in Rayleigh and Rician fading channels with different parameters.Experimental results show that the performance of3D-CNN method trained by 10-fold CV is the best,while the performance of 2D-CNN method trained by 10-fold CV is the worst.In the case of multiple classification,the 3DCNN architecture trained by 10-fold CV has the best performance,while the 2D-CNN architecture trained by 10-fold CV has the worst performance.Therefore,you can see that3 D architectures perform better than 2D architectures.In this paper,the research provides the effective methods and techniques of classes modulation mode recognition in multi-generation communication system and complex fading channel environment.Hence,the proposed method can satisfy the user demand,high data rate and quality of communication,improve the performance of communication system,and achieve high data rate transmission in densely populated areas and multigeneration communication system.