A spatial transformation scheme is proposed to be deployed at the radio-units (RUs) in cloud radio-access-networks (C-RAN) to alleviate the fronthaul burden and baseband signal processing complexity. The module transforms antenna domain signals to the beam domain, such that the spatial dimensionality is significantly reduced for RUs with large antenna arrays. A low-complexity transformation scheme is investigated with finite uplink receive samples. It leverages a combined discrete-Fourier-transform (C-DFT) approach, which shows significant signal power gain over the conventional DFT scheme by combining correlated DFT beams to form a more matched beam pattern. The optimality of the combination method is proved. The performance is validated by extensive simulations which are based on practical channel models and system parameters.