Deep Transfer Learning for Location-Aware Millimeter Wave Beam Selection

The main bottleneck for using deep neural networks in location-aided millimeter wave beam alignment procedures is the need large datasets to tune their set of trainable parameters. This letter proposes use transfer learning technique order reduce dataset size requirements deep-learning based selection. Information can be done from one environment another, or antenna configuration which we refer as domain and task adaptation, respectively. Numerical evaluations show a significant gain both adaptation scenarios, especially with limited datasets.