Gated Neural Network Coordinate Transformation Blind Equalization Algorithm And SIMULINK Realization

In a wireless communication system,due to the time-varying,frequency-changing and environmental noise effects of the wireless channel,the signal at the receiving end is highly susceptible to inter symbol interference(ISI).Blind equalization technology saves channel bandwidth while effectively eliminating ISI.Therefore,this paper combines gated recursive unit neural network(GRUNN),long-term and short-term memory neural network(LSTM),bidirectional long-term and short-term memory neural network(BLSTM),coordinate transformation(CT),adaptive coordinate transformation(ACT)and adaptive triangle coordinates transformation(ATCT).A variety of new blind equalization algorithms have been proposed.The main research work of this paper is as follows:(1)For the problem that the feed-forward neural network blind equalization algorithm has poor equalization ability of the norm signal,and the statistical modulus value does not match the coordinate of the signal constellation in the cost function,a coordinate-transformation blind equalization algorithm based on gated recurrent neural network(GRUNN-CT-CMA)is proposed.The CT algorithm is added to the output of GRUNN.The number of input layer units of GRUNN is the equalizer tap length,and the network weight vector is updated by the output error value of the CT algorithm.Simulation experiments show that GRUNN-CT-CMA has excellent balance of very mode signals.(2)An adaptive coordinate transformation blind equalization algorithm based on long-term and short-term memory neural network(LSTM-ACT-CMA)is proposed.Because the feedforward neural network blind equalization algorithm has poor adaptability when equalizing high-order signals,the three-gate structure of LSTM has strong sensing capability for high-order signals and long-lasting memory.Combining LSTM with ACT algorithm enables the algorithm to automatically determine the QAM input signal of the order below 64 QAM,which is more applicable.Simulation experiments show that LSTM-ACT-CMA has a balanced effect when equalizing high-order very-mode signals.(3)The blind equalization algorithm for feedforward neural network can only be applied to the order of signal in a certain region when equalizing high-order multi-mode signals.An adaptive triangle coordinates transformation blind equalization algorithm based on bidirectional long-term and short-term memory neural network(BLSTM-ATCT-CMA)is proposed.First,BLSTM is highly aware of the sequence and treats it as a blind equalizer for high-order signals.Secondly,the ATCT algorithm is added to the BLSTM output to overcome the complexity of the traditional CT algorithm when the input signal order is high.Simulation experiments verify the excellent balance performance of the algorithm.(4)The system design and simulation of the algorithm are carried out through SIMULINK design platform.Firstly,the classical blind equalization system is designed and the parameter configuration is given.Secondly,different gated neural network modules and coordinate transformation modules are designed.According to the algorithm theory,the modules not in the SIMULINK module library are built and packaged,and finally through the SIMULINK platform.Simulations verify the effectiveness of the system.