Abstract: In order to improve the channel estimation performance of MIMO-OFDM system， a channel frequency response estimation algorithm based on K-nearest neighbor twin support vector regression(KNN-TSVR)is proposed. The working process of the algorithm is that the least square algorithm is used to estimate the channel parameters of the pilot position， and the training samples are obtained. Then the KNN algorithm is used to preprocess the training samples， and the weights given to each sample are obtained. Finally， the weighted TSVR is used to interpolate the channel parameters of all the positions in the MIMO-OFDM system. The improved weighted TSVR channel estimation method proposed not only has the advantages of TSVR in nonlinear regression， but also has better regression performance and anti-noise ability compared with traditional TSVR due to the improvement of TSVR based on KNN. The simulation results of nonlinear MIMO-OFDM channel estimation confirm this conclusion.

Key words: channel estimation; KNN(K-nearest neighbor); MIMO(multiple-input multiple-out); OFDM(orthogonal frequency division multiplexing); TSVR(twin support vector regression)