Transfer Learning Capabilities of Untrained Neural Networks for MIMO CSI Recreation

Machine learning (ML) applications for wireless communication has gained momentum on the standardization discussions for 5G advanced and beyond. One of the biggest challenges for real world ML deployment is the need for labelled signals and big measurement campaigns. To overcome those problems, we propose the use of untrained neural networks (UNNs) for MIMO channel recreation/estimation and exploit their learned prior to provide higher channel estimation gains. Moreover, we present an UNN for simultaneous channel recreation for multiple users in which estimation gain is traded by reduction on the overall number of parameters. Our results show that transfer learning technique is effective in accessing the learned prior on the environment structure as it provides higher channel gain for neighbouring users. Moreover, we indicate how the underparametrization of UNNs can further enable low-overhead channel state information (CSI) reporting.