Deep Learning Based MIMO Channel Prediction: An Initial Proof of Concept Prototype

Massive MIMO is a key component of current and future wireless communication systems. To harvest the multiplexing and beamforming gains of these large-scale MIMO systems, however, the channel knowledge needs to be acquired at the massive MIMO transmitters. This is typically associated with large training overhead, especially in FDD massive MIMO. Recent research showed that deep learning could lead to interesting gains for massive MIMO systems by mapping the channel knowledge from the uplink to downlink channels or between antennas at nearby locations. In this paper, we provide an initial proof-of-concept prototype for this concept, where we show using a sub-6GHz hardware setup promising results for channel mapping across frequency and space.