Deep Learning Method of Polar Codes under Colored Noise
Danqiu Xiong, Bin Tian · 2019
Polar codes have been used as channel control codes in 5G mobile communications standard. Although polar codes yield good performance, the performance of their short codes is facing great challenges. In this paper, we cascade the full convolutional neural network model (FCN) in deep learning with the BP decoder of polar codes, so as to realize the decoding of polar codes under Additive colored Gaussian noise (ACGN), hence to improve the decoding accuracy of short polar codes under ACGN. The concatenated decoding structure is denoted as BP-FCN-BP decoder. The function of the first BP unit is to estimate the colored noise roughly, and the function of the FCN unit is to estimate the colored noise precisely. The second BP unit decodes with high accuracy according to the output of FCN.