2D regression channel estimation for equalizing OFDM signals

In this paper, we present a channel estimation method for Orthogonal Frequency Division Multiplexing (OFDM) signals. Our method is based on a two-dimensional nonlinear regression method, taking into account the correlations of the fading process in both time and frequency domains. We derive a general bit error rate (BER) expression which can also used to predict the performance of many other OFDM channel estimates. The performance of this new estimate is very close to the theoretical bit error probability lower bound that is obtained by assuming that the channel response is perfectly known. Unlike linear minimum mean-squared-error (LMMSE) channel estimates, it needs not to know or estimate channel statistics like channel correlation matrix and SNR hence is insensitive to additive white Gaussian noise (AWGN).