Abstract
Training-based channel estimation involves probing of the channel in time, frequency, and space by the transmitter with known signals, and estimation of channel parameters from the output signals at the receiver. Traditional training-based methods, often comprising of maximum likelihood estimators, are known to be optimal under the assumption of rich multipath channels. Numerous measurement campaigns have shown, however, that physical multipath channels exhibit a sparse structure in angle-delay-Doppler, especially at large signal space dimensions. In this paper, key ideas from the emerging theory of compressed sensing are leveraged to: (i) propose new methods for efficient estimation of sparse multi-antenna channels, and (ii) show that explicitly accounting for multipath sparsity in channel estimation can result in significant performance improvements when compared with existing training-based methods.