Research On Weak Signal Detection Technology Based On Neural Network

With the development of communication,the spectrum is getting more and more crowded,as a result,the power of the signal received from channel with strong background noise will be greatly reduced,which brings great difficulties and challenges to the detection of weak signals.The traditional methods usually require prior knowledge of the signal,have limitations on the type of signal detected,and the detection effect is not ideal.The rise of neural network technology give these problems a new method.This research introduces the neural network method into the weak signal detection problem,and conducts the modeling and method research.Firstly,a solution need no prior knowledge of signal and has no limitation on the type of signal detected is proposed in this paper,which is the weak signal detection method using Long Short-Term Memory(LSTM)network.Simulation experiment was implemented and a comprehensive performance comparison was implemented with the traditional detection method based on radial basis function(RBF)neural network.The experimental results show that our model has better performance.Secondly,to improve the limited detection accuracy achieved by the weak signal detection method based on the LSTM network when detecting the original signal directly,the solution based on the LSTM network is improved in this paper and wavelet decomposition was added in,propose a weak signal detection method based on wavelet-Long Short-Term Memory(WTLSTM)network.Wavelet decomposition is used to extract the approximate coefficients where the signal components concentrated,removes some noise components,and then inputs them into the LSTM network for feature learning.The performance of the trained model is tested at different strength of noise.The experimental results prove that this method is better than the detection method based on the LSTM model and the RBF neural network and has better results.Finally,to improve the limited detection accuracy of weak signal detection methods based on LSTM networks,this paper introduces the support vector machine(SVM)model which is a great classification method in machine learning into the detection model based on LSTM networks,and proposes a detection method with better performance,which is the solution based on Long Short-Term Memory-support vector machine(LSTM-SVM).This method uses two models for joint decision-making.Training two models separately,and input the distance between the sample and the classification plane obtained by the SVM model into the logistic regression model,which can map it to the detection posterior probability of each sample.And use the average of it and the posterior probability output by the LSTM model,and then to perform a threshold decision and acquire the final detection result.The experimental results prove that this method is better than the detection method based on the LSTM model and the RBF neural network and has better results.