A Deep Learning Based Receiver for Wireless Communications Systems With Unknown Channel Models.

In this paper, we consider the signal detection problem for practical communication systems where the transmission suffers from non-idealities effects such as I/Q imbalance, non-Gaussian noise, and amplifier nonlinearity. In this context, traditional model-based receiver algorithms are not applicable because the exact model for the end- to-end channel is unavailable. To deal with this challenge, a deep learning based receiver is developed, which learns the detection criterion from data and realizes signal detection without knowing the underlying channel model. To improve the generalizability of the proposed detector, domain knowledge in channel estimation and equalization is used to guide the design of the receiver structure. Moreover, the idea from transfer learning and adversarial learning is adopted to assist the training of the network. The performance of the proposed detector is evaluated in terms of bit error rate (BER), and the superiority of the proposed design is shown compared to the state-of-the-art technologies.