2021/11/15 by Brenda Vilas Boas, Boas, Brenda Vilas, Wolfgang Zirwas +3
Computer Science · Engineering · #FOS: Computer and information sciences #FOS: Electrical engineering #Machine Learning (cs.LG) #Millimeter-Wave Propagation and Modeling #Signal Processing (eess.SP) #Speech and Audio Processing #Wireless Signal Modulation Classification #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2111.07858
openalex publication_date 2021/11/15 · openalex created_date 2023/02/16 · openalex updated_date 2026/07/28
Machine learning (ML) applications for wireless communications have gained\nmomentum on the standardization discussions for 5G advanced and beyond. One of\nthe biggest challenges for real world ML deployment is the need for labeled\nsignals and big measurement campaigns. To overcome those problems, we propose\nthe use of untrained neural networks (UNNs) for MIMO channel\nrecreation/estimation and low overhead reporting. The UNNs learn the\npropagation environment by fitting a few channel measurements and we exploit\ntheir learned prior to provide higher channel estimation gains. Moreover, we\npresent a UNN for simultaneous channel recreation for multiple users, or\nmultiple user equipment (UE) positions, in which we have a trade-off between\nthe estimated channel gain and the number of parameters. Our results show that\ntransfer learning techniques are effective in accessing the learned prior on\nthe environment structure as they provide higher channel gain for neighbouring\nusers. Moreover, we indicate how the under-parameterization of UNNs can further\nenable low-overhead channel state information (CSI) reporting.\n