Estimation of frequency-selective CDMA channels with large possible delay and Doppler spreads

The problem of estimating channels with a large number of unknown coefficients is considered. A matching pursuit (MP) followed by cancellation (MPC) algorithm is presented. The algorithm consists of two steps. In the first step, a basic MP algorithm that identifies the strongest channel taps and successively cancels them is used. The second step of the algorithm uses the already identified channel taps to estimate the channel coefficients more accurately. The algorithm has several benefits: low complexity, significantly improved performance over conventional least squares (LS) in sparse channels, and handling of time-varying channels including carrier offsets. Methods to successfully terminate the MP step of the algorithm are presented and are shown to perform well. The statistical behavior of MPC is studied. The MPC estimator is found to achieve a performance close to the Cramer-Rao lower bound. Further, MPC is found to outperform conventional LS in terms of bit error rate when low training overhead is used