Abstract

Cloud Radio Access Network systems with mmWave Massive MIMO framework can be considered as a potential candidate for next generation wireless communications due to its promise of increased spectral efficiency and distributed signal processing capability. State-of-the-art compressive sensing algorithms like sparse Bayesian learning can exploit the inherent sparsity of the mm-Wave wireless channels to estimate the channel connecting the remote radio head and the user equipment in the wireless access link. The performance of the sparse Bayesian learning based channel estimation can be adversely affected by impairments due to optical fiber based front-haul channel and quantization noise. As a result, it is necessary to compensate for the performance degradation by applying methods which can combat the effects of the front-haul channel. Contemporary research has demonstrated the capability of deep learning algorithms in signal enhancement under low signal-to-noise ratio conditions, such as hybrid beamforming design, channel estimation as well as feedback of channel state in heterogeneous multi-antenna wireless systems. Motivated by their de-noising and signal prediction capabilities, convolutional long short term memory networks are employed in this work to jointly remove the quantization noise and optical fiber impairment due to the front-haul channel, which can improve the performance of sparse Bayesian learning in estimating the wireless access channel at the base-band processing unit. Computer simulation results show that the proposed methodology performs well under low signal-to-noise ratio conditions.