Research On Adaptive Equalization Algorithm Based On Heuristic Neural Network

With the advent of the era of big data and 5G,the Internet of Everything has gradually become a reality,and wireless communication has become very important.In actual wireless communication,there are few pure linear channels.At the same time,due to the existence of multiple users,the problems of delay and intersymbol interference(ISI)are very serious.In order to solve the above problem,channel equalization technology is usually adopted at the receiving end,and the equalizer is formed at the receiving end through the equalization algorithm to compensate for the distortion caused by the nonlinear characteristics of the channel.However,when the channel conditions are bad,the performance of traditional equalizer is not ideal.Neural networks have strong non-linearity and have outstanding performance in applications such as signal analysis and image processing.Compared with the traditional linear equalizer,the bit error rate of the equalizer realized by the neural network is lower.However,neural networks are prone to local minima,slow training speed,and lack of methods to select the network structure,which limits the application of neural networks in the field of equalization.Heuristic algorithms have become a very popular algorithm for solving optimization problems,which can find high-quality solutions in a short period of time.The introduction of heuristic algorithms can speed up the training of neural networks and improve the quality of solutions.Aiming at the problems of long training time and slow convergence speed of existing adaptive equalization algorithms based on WNN-SOS(Wavelet Neural Network trained by Symbiotic Organisms Search algorithm),this thesis proposes a method based on WNNm SOS adaptive equalization algorithm.Based on the realization of the equalizer based on wavelet neural network,the algorithm modified the structure of the wavelet neural network,added the direct connection structure from the input layer to the output layer,and deleted the threshold part of the neurons in the output layer.Then the improved symbiotic search algorithm is used to optimize the combination of weights and wavelet parameters of the wavelet neural network.Finally,the optimal parameter combination is used to construct a wavelet neural network to predict the output of the test data.As the proposed algorithm simplifies the SOS while increasing the optimization parameters,it can reduce the training time of the algorithm as a whole and improve the convergence accuracy.In order to balance the exploration and development capabilities of the m SOS algorithm,a WNN-m SOS1 equalization algorithm is proposed.This algorithm modifies the m SOS algorithm and changes the global search and local search in the algorithm to a gradual transition from global search to local search.While maintaining the advantages of the m SOS algorithm,it can improve the convergence speed of the algorithm.In order to test the performance of the adaptive equalization algorithm proposed in this thesis,a comparative experiment is performed on the data set generated by the non-linear channel.The results indicate that the WNN-m SOS equalization algorithm proposed in this thesis reduces the training time of the neural network equalizer while ensuring the convergence accuracy and equalization performance.WNN-m SOS1 balance algorithm improves the convergence speed of the algorithm while ensuring the best solution quality.It is proved that using a heuristic algorithm to optimize the neural network can not only enhance the performance of the neural network and obtain the optimal network structure,but also improve the quality of the feasible solution and prevent the neural network from falling into a local minimum.