Improved Low-Resolution Quantized SIMO Estimation via Deep Learning

The signal estimation problem in a single input multiple output (SIMO) system with low-resolution analog-to-digital converters (ADCs) is studied. The optimal minimum mean square error (MMSE) estimator is difficult to compute due to the nonlinearity introduced by ADCs, while linear MMSE (LMMSE) suffers from performance degradation. Motivated by the advances in deep learning, a convolutional neural network estimator (CNNE) is proposed, leveraging the highly successful TextCNN. We simulate the system under both accurate and inaccurate channel estimation, and the numerical results suggest that the normalized mean square error performance of CNNE outperforms LMMSE in majority of the cases with less runtime.