One-Bit Quantized Channel Prediction with Neural Networks

Abstract

We study the problem of predicting channel coefficients from one-bit quantized observations in an environment of a moving user who sends pilots to a base station. To start with, we propose a prediction algorithm which consists of two stages. The first stage aims at reconstructing the high-resolution (pre-quantization) receive signal. The second stage then predicts channel coefficients from this reconstructed signal. A drawback of this algorithm is that certain second moments of the channel statistics are required. In case of high-resolution (no quantization) observations, a recently introduced neural network based approach was able to predict channels even without the use of second order statistics. A low-SNR formulation of the proposed two stage algorithm motivates us to employ the neural network based method also in the case of one-bit quantization. Numerical simulations demonstrate the validity of this approach. We observe that the obtained channel predictor can compete with the algorithm that makes use of the second order statistics.