Nonlinear Channel Estimation for OFDM System by Wavelet Transform Based Weighted TSVR

Abstract

An efficient nonlinear channel estimation method for pilot-aided orthogonal frequency division multiplexing system is proposed in this work. The considered channel is selective in time and frequency domain, that is doubly selective channel. Wavelet transform based weighted twin support vector regression is used for channel frequency response estimation, which is suitable for the regression of nonlinear system. Different from traditional support vector regression algorithm, the proposed algorithm gives samples different weights according to their variance calculated based on wavelet transform. The weights are added into both first and quadratic terms of the objective functions to reduce the impact of outliers, which is likely to appear in the received pilot signal polluted by noise. The proposed channel estimation algorithm has good generalization ability and can reduce the influence of overfitting problem. The results of computational tests show that the proposed algorithm is with better estimating performance compared to the classical pilot-aided channel estimation methods. INDEX TERMS Channel estimation, orthogonal frequency division multiplexing, twin support vector regression, wavelet transform.

Artificial intelligence for channel estimation in multicarrier systems for B5G/6G communications: a survey

“…Recently, the research on estimators based on regression algorithms has been scarce, which is understood due to the growth of NNs and RL solutions. However, some research has accomplished promising results regarding the SVR for OFDM and MIMO-OFDM systems [172,173]. Therefore, it is a research direction to apply these estimators for OFDM variations or other multicarrier systems, extending them to MIMO schemes.…”

“…The linear, polynomial, and nonlinear regression algorithms are early basic applications of ML concepts for channel estimation. Support vector regression (SVR) has recently been raised as a potential regression strategy in AI-aided channel estimation techniques [172,173]. The evolutionary algorithm has also been applied to channel estimation, whereas the genetic algorithm is more widely used than other evolutionary techniques.…”

“…The main concept is to find a function that best fits the training data behavior to perform predictions. For example, linear [175][176][177], polynomial [178], 2D nonlinear [99,179,180], and support vector [172,173,[181][182][183][184][185][186][187] regressions have been employed in channel estimation for multicarrier systems. Regression algorithms go under the supervised learning paradigm.…”

“…The algorithm leads to lower BER under the same SNR compared with the LS and feedback estimators, from a good approximation to a perfect estimation. The wavelet transform was used to obtain weights to improve twin SVR (TSVR) channel estimation in pilot-assisted OFDM systems operating in fast selective fading channels [172,173]. The training samples are weighted according to their distance from the mean values filtered by the wavelet transform.…”

Estimation of Rayleigh flat channel coefficients using deep learning

“…The CNN for channel classification (CNN-CC) is named as CNN-CCnet, and constructed based on the CNN model originally designed for image classification [26][27][28][29] which is found suitable for channel classification. Different layers of CNN-CCnet are given in Appendix C. modify and fine tune the hyper parameters like number of layers, maximum number of epochs and training options and so on and goto step 2 (retraining).…”