Least Squares Support Vector Regression-Based Channel Estimation for OFDM Systems in the Presence of Impulsive Noise

This work aimed to investigate a multipath channel estimation technique for orthogonal frequency division multiplexing (OFDM) systems based on least squares support vector regression (LSSVR) in the presence of Gaussian and Impulse noise. Since impulsive noise can considerably affect the performance of communication systems and also due to the time-varying and frequency selectivity of wireless channels, it is unavoidable to have a proper channel estimation and interpolation technique. Therefore, in this contribution, an autoregressive modeled multipath channel was estimated using LSSVR in the presence of impulsive and Gaussian noises. The channel estimation method based on LSSVR contrasts with neural networks, as it does not require real data for training. It benefits from the LS estimator output for training, and due to the lack of need for solving a Quadratic Programming (QP) problem, has lower complexity than standard SVR. The simulation results illustrate that the proposed method outperforms standard SVR, multilayer perceptron neural networks, and cubic-spline interpolated LS in terms of bit error rate (BER) and mean square error (MSE) criteria.