Topics on channel estimation and equalization for sparse channels with applications to digital TV systems

This research focuses on the channel estimation problem for sparse channels with long delay spreads which are usually encountered in Digital Television (DTV) systems. If a standard Least-Squares (LS) type channel estimator is used, the training sequence which is available in the current American DTV standard for frame synchronization purpose is generally quite short to provide useful channel state information where the DTV channels can normally span several hundred symbol intervals. In order to overcome this difficulty we propose a new method, called the Blended Least Squares (BLS), which is based on blending the Time-Of-Arrival (TOA) information (multipath locations) into the LS equations. We first obtain an initial raw channel estimate via iteratively updating the covariance matrix of the channel noise which includes random data as well as thermal noise. We use this initial channel estimate to obtain the TOA information. We also propose two different approaches to estimate the TOAs: the first one is based on thresholding the channel impulse response, and the second approach is based on frequency domain forward-only, or forward-backward, linear prediction based auto-regressive (AR) modeling of the frequency transform of the channel impulse response. Once the TOAs are obtained then the channel impulse response is estimated by blending the TOAs into the LS equations. We show the effectiveness of this method by studying the performance of the channel estimate based Decision Feedback Equalizers (DFE)