I. Introduction
Iterative channel estimation [1], which processes the inputs from the equalizer or decoder and feeds back the outputs to the equalizer in an iterative manner, has been shown to effectively track channel variations. It can be categorized into hard-input and soft-input iterative channel estimation. Soft-input iterative channel estimation, which is also known as Turbo channel estimation, exploits the soft estimates of signals from the decoder as the input of the channel estimator and has been shown to be more robust against channel variations than the hard-decision-based iterative channel estimation [1]. Turbo channel estimation is usually incorporated with Turbo equalization [2] to allow iterative channel estimation, equalization, and decoding. Turbo channel estimation was applied to flat-fading channels in [3] and frequency-selective fading channels in [4], respectively, based on the minimum mean square error (MMSE) criterion. In [1], a Kalman-filtering-based Turbo channel estimation and a weighted Turbo recursive least-squares (RLS)-based channel estimation schemes were proposed. In [5], a simplified Turbo RLS-based channel estimation scheme was proposed for phase-shift keying (PSK) modulation, which has almost no performance loss and achieves a tremendous complexity reduction compared with Turbo RLS-based channel estimation. However, most previous work on Turbo channel estimation was symbolwise and was incorporated with time-domain equalization (TDE), which requires a prohibitive complexity for highly dispersive channels [6].