Machine Learning for Channel Estimation from Compressed Measurements

Michael Koller, Christoph Hellings, Michael Knoedlseder, Thomas Wiese, David Neumann, Wolfgang Utschick

Abstract: It has recently been proposed to employ convolutional neural networks for estimating structured channels, e.g., channels where the received power is concentrated around the centers of a small number of propagation paths. In simulations, the approach shows good performance also for systems with a high number of antennas, but it does not consider that such systems might have less receiver chains than receive antennas. In this case, an analog mixing network would connect the antennas to the receiver chains, which results in low-dimensional observations. In this paper, we study how the machine learning approach can be used to estimate the channel from such compressed measurements.