Reduced complexity channel parameters estimation with multi-dimensional Unitary-SAGE algorithm

Estimating channel parameters from measured data using maximum-likelihood methods can be very time-consuming due to the iterative nature of these algorithms. Although the SAGE algorithm has a faster convergence speed compared to the classical EM algorithm, it could still be computational-intensive. Since most of the computations of the SAGE algorithm are concentrated in the M-step, we propose a new implementation of the algorithm, namely the Unitary-SAGE algorithm, that is able to implement the entire M-step throughout the iterations in the real-valued domain. This helps to reduce the processing time and memory consumption of the original SAGE algorithm since all computations in the M-step are performed in the real-valued domain with efficient matrix manipulation. We present the general implementation of the Unitary-SAGE algorithm in the frequency domain when applied in both the element-space and the newly developed beamspace domains.