Low-complexity neural network based DOA estimation for wideband signals in massive MIMO systems

Abstract

Massive multiple-input multiple-output (MIMO) has become one of the most promising technologies for wireless communication systems, in which the direction-of-arrival (DOA) plays an important role in interference cancellation and transmission reliability. However, the challenge of conventional wideband DOA estimation algorithms is that their high computational complexity brings great difficulties for their effective application in massive MIMO systems. In this paper, we propose a wideband low-complexity DOA estimation algorithm based on a principal component analysis (PCA) neural network for massive MIMO systems. First, a new criterion for constructing a focusing matrix is proposed to avoid complex angle pre-estimation. To further reduce the complexity of the eigenvalue decomposition (EVD), we propose a signal subspace estimation algorithm based on PCA, which uses only a limited amount of self-organizing learning to estimate the weights of the network and the signal subspace and does not require prior sample training. Moreover, to increase the estimation accuracy, we use the Akaike information criterion (AIC) to divide signal and noise subspaces accurately. The theoretical analysis and simulation results show that the computational complexity of the proposed algorithm is effectively reduced, and the angles of wideband sources can be accurately estimated in massive MIMO systems.