Training of a channel estimator for frequency-selective fading channels

The authors deal with a fast start-up channel estimator based on a training signal consisting of a self-orthogonal sequence repeated several times. A process of maximum likelihood estimation of the sampled impulse response of the channel is carried out on the receipt of each linearly distorted self-orthogonal sequence. For the correct operation of the system, the channel impulse response must not change significantly over the duration of any one self-orthogonal sequence. Novel methods of processing the received training signal are studied. These techniques greatly reduce the error in the channel estimate but without involving any undue increase in equipment complexity or any change in the training signal itself. Results of computer simulation tests giving the performance of the training processes are presented.