Complex CNN-Based Equalization for Communication Signal

In this paper, we address the application of deep learning in signal equalization, presenting an end-to-end learned method based on convolutional neural network (CNN) to directly recover a communication signal from a noised signal influenced by wireless channel. Different from real-valued tasks, complex-valued tasks can be hardly resolved by normally sequential real-valued CNN. Alternatively, we propose a simply mixed cascade structure to replace the traditional equalization methods in communication systems, i.e., multi-modulus algorithm, least mean square method and recursive least square method. Additionally, we generate a noised dataset consisting of known modulation signals in digital communication signal by simulation. The effects of multi-path fading, additive white Gaussian noise (AWGN), frequency and phase offset and symbol rate are taken into consideration. Furthermore, we proved the proposed method obtaining better performance over the traditional equalization method.