Learning-Based Channel Estimation for Various Antenna Array Configurations

Recently, a neural-network-based method for massive MIMO uplink channel estimation was introduced. The derivations assumed a uniform linear array (ULA) with half-wavelength antenna spacing at the base station. In this work, we show that the estimator can also be used in case of ULAs and uniform rectangular arrays (URAs) with antenna spacings given by integer multiples of half the wavelength. We then investigate how the antenna spacing and certain parameters of the channel model influence the estimation performance.