Learning a Low-Complexity Channel Estimator for One-Bit Quantization

Benedikt Fesl, Michael Koller, Nurettin Turan, Wolfgang Utschick

Abstract: A low-complexity convolutional neural network (CNN) channel estimator has been proposed recently, which was designed based on assumptions on the underlying channel model. In this work, we investigate how one-bit quantized observations affect this CNN estimator. In contrast to many other approaches, we propose a technique to obtain only one CNN estimator for a whole range of signal-to-noise ratio (SNR) values. We compare the performance of this estimator with a linear minimum mean square error (LMMSE) estimator based on the Bussgang decomposition and also with a state-of-the-art maximum a posteriori (MAP) approach, which exploits an approximate sparsity of the channels.