Optimization of Subspace Projection in Noncoherent Massive MIMO Systems

Abstract:
The combination of noncoherent multi-user detection with subspace projection was shown to be an attractive approach in multi-user massive MIMO systems. In previous work, the subspace projection was performed in principle by means of a singular value decomposition (SVD) or the signed URV (SURV) decomposition of the receive matrix. In a straightforward setting, the approaches first perform the decomposition, followed by nulling the non-signal dimensions. However, having knowledge of the number of relevant dimensions, it is possible to calculate the so-called truncated SVD (TSVD) which reduces the complexity of the problem. In this paper, we present an optimization of the subspace projection process, namely by employing a variant of the TSVD algorithms, the Krylov–Schur SVD, followed by efficient matrix operations. We compare the proposed approach to the straightforward one in terms of complexity in noncoherent massive MIMO systems. Numerical results cover the complexity reduction achieved by employing the TSVD in combination with efficient matrix calculations.