Meta-Learning Based Semi-blind Data Detection for Cell-Free Systems
Guanyang Fu, Qing Wang, Rui Zhu, Haozhi Wang · 2023
To detect data without channel estimation in cell-free systems, we propose a model-agnostic Meta-learning based semi-blind data detection Deep neural Network (MDNet) training via Model-Agnostic Meta-Learning (MAML) algorithm. Considering the block fading model, the different channel states between blocks are regarded as different tasks. A meta-training dataset consisting of the signals sent by the previous coherent block user is used to train the Deep Neural Network (DNN), so MDNet can quickly adapt to the new channel state. Furthermore, first-order MDNet (FMDNet) is also proposed based on the first-order MAML algorithm. The simulation results show that both the proposed MDNet and FMDNet can accurately detect user-transmitted data by using a few pilots, significantly improving the symbol error rate compared with the conventional learning methods.