The Research On OFDM System Channel Estimation Based On Support Vector Regression Algorithm

In the process of signal transmission of mobile communication system,due to the influence of channel fading,multi-path transmission,delay extension and other factors,the channel transmission characteristics are uncertain,and at the receiving end,there will be some problems such as inter-symbol interference and inter-code crosstalk,etc.Combined with the inevitable noise influence,the received signal may produce different degrees of error codes,which will affect the communication quality and even fail to normal communication.At the same time,the multi-path effect in propagation makes the wireless channel present frequency selection characteristics,which can make the communication quality worse.Orthogonal frequency division multiplexing technology is an effective way to solve the problem of inter-code interference caused by frequency selective channels.The adoption of coherent reception at the receiving end of OFDM system can reduce bit error rate,and obtaining channel parameters is a necessary condition for coherent demodulation.Therefore,this paper studies the channel estimation problem of OFDM system.Wireless channel presents strong nonlinear characteristics,and support vector machines can be used to estimate channel parameters due to their adaptability to nonlinear system regression.Performance is degraded because the traditional support vector regression machine algorithm gives the same weight to training samples with different degree of noise pollution.In order to improve the performance of OFDM channel estimation,a pilot based channel estimation algorithm based on Wavelet and twin support vector machine is proposed.In this paper,wavelet transform is used to preprocess the training data to get the weight matrix and weight vector,so as to establish twin support vector machine regression prediction model WTWTSVR.The proposed WTWTSVR algorithm is used to estimate the fading channel parameters of OFDM system.This algorithm is an improvement of the traditional TSVR algorithm.On the basis of TSVR,wavelet transform is introduced to calculate the distance between the training data and its expected value,providing the weight for the samples.The weight is determined by wavelet transform theory and inserted into the quadratic and primary terms of channel parameter regression objective function to reduce the influence of outliers.The weight matrix and the weight vector represent a representation of the distance between the sample polluted by noise and its expected value.For the sample greatly affected by noise,they are given smaller weight,and for the sample less affected by noise,they are given larger weight.The channel estimation method used in this paper is the estimation method with pilot.The least square method is used to estimate the channel frequency at the pilot,and the corresponding channel frequency at the pilot is obtained.Then the WTWTSVR algorithm is used to predict the channel frequency response of the subcarrier without pilot.The simulation results with BER and MSE as evaluation criteria show that under the condition of Jakes fast fading channel model,the improved twin support vector machine pilot channel estimation method has better prediction performance compared with the traditional twin support vector machine regression method and the traditional channel interpolation method.