This paper presents the average block error rate (BLER) performance for linear minimum mean-square error (LMMSE) based iterative decision-directed channel estimation (IDDCE) associated with the turbo frequency domain equalizer (FDE) for discrete Fourier transform (DFT)-precoded Orthogonal Frequency Division Multiple Access (OFDMA). In the proposed IDDCE, estimated channel responses, which are used for weights of feed-forward and decision feedback FDEs, are iteratively updated using an accurate 2-dimensional LMMSE algorithm at each iteration of the turbo FDE. Through computer simulations, we clarify the best window size for the 2-dimensional LMMSE algorithm assuming the 9-path Extended Typical Urban (ETU) channel model and the maximum Doppler frequency of up to approximately 220 Hz. The computer simulations show that the 2-dimensional LMMSE based IDDCE is effective in decreasing the required average received signal-to-noise power ratio (SNR) satisfying the target average BLER for the turbo FDE in a frequency-selective Rayleigh fading channel with low-to-high maximum Doppler frequencies and with a wide range of root mean square delay spread values.