Modulation Classification For MIMO Systems Based On Machine Learning Algorithm

Modulation classification techniques can identify the based-band modulation type of the received signal,which provide the necessary prior information for demodulation and decoding procedures.Hence,it is a significant issue in blind estimation of communication signal parameters.As the Multiple-input Multiple-output(MIMO)technology has been widely used in many communication systems,it is quite meaningful to investigate the modulation classification methods for MIMO systems.In MIMO systems,the aliasing phenomenon of signals from multiple transmit antennas disables the algorithms for single-antenna systems.Besides,the existing algorithms for MIMO systems suffer from high computational complexity and low noise robustness.To solve these problems,in this thesis,we introduce both the classifier and the feature selection algorithms from machine learning fields to balance the classification performance and computational complexity of the proposed algorithm.The main contributions and innovative results are as follows.When the channel state information is already known for the receiver,utilizing the Random Forest(RF),we propose a decision-fusion based modulation classification algorithm for MIMO systems.Different with the existing algorithms utilizing single or few classification features,the proposed algorithm adopts multiple Higher-order Cumulants(HOC)as the classification features,which extends feature dimension to enhance the noise robustness.Firstly,in order to recover the transmit signals,zero-forcing equalization technique is used.Subsequently,multiple HOCs are calculated then averaged as the feature vector to train the RF model.Finally,the classification results are obtained by judging the feature vector of the received signal utilizing the trained model.Simulation results demonstrate that the proposed algorithm improves the probability of correct classification under low signal-to-noise environment.When the channel state information is unknown for the receiver,based on the preselectedcumulants,we propose a low complexity modulation classification algorithm for MIMO systems.As the channel state information is unknown,to extract classification features,we use the Joint Approximate Diagonalization of Eigen-matrices(JADE)algorithm to recover the transmit signal.While the blind equalization algorithms require several rounds of iterations to converge,which are already computationally inefficient,it is quite unrealistic for practical engineering implementation if we still uses the proposed RF based algorithm.To reduce the overall complexity,we use Support Vector Machine based Recursive Feature Elimination(SVM-RFE)algorithm to perform feature selection from the chosen feature vector in the proposed RF based algorithm.The HOCs with comparatively higher feature importance ranks are selected as new features,and the classification results are obtained by comparing the feature values of the received signal with the the known threshold.Simulation results show that the proposed algorithm can maintain preferable classification performance and effectively reduce the computational complexity.