A Novel Neural Network Denoiser for BCH Codes

Hongfei Zhu, Zhiwei Cao, Yuping Zhao, Dou Li

Abstract: Traditional filters are dedicated to reducing the out-of-band noise while the in-band noise is beyond their capability. With the development of deep learning, the deep neural network (DNN) provides a more powerful and effective approach to denoising. In this paper, we propose a novel neural network denoiser for BCH codes. The denoiser directly learns an end-to-end mapping from a noisy codeword to its corresponding denoised codeword. Simulation results show that the signal to noise ratio (SNR) improvement and the symbol error rate (SER) reduction of the denoiser is significant. Consequently, the denoiser assists the traditional decoder in achieving far better bit error rate (BER) and frame error rate (FER) performance.