Inception-ResNet Assisted Iterative Decoding Algorithm Based on Alternating Direction Multiplier Method

The application of deep learning to channel decoding methods has gradually become a hot research topic. However, the high complexity of deep neural network (DNN) parameters hinders the application of deep neural networks to long codes, and the difficulty of decoding increases exponentially with the increase of code length. To address this problem, this paper proposes a low-density parity-check (LDPC) decoding based on alternating direction multiplier method (ADMM) assisted by Inception-RestNet network. After passing through the channel with the noise codeword through the Inception-RestNet network to achieve the denoising process is input to the traditional ADMM decoder, and then according to the traditional this algorithm calculates the codeword with the original with the noise codeword to be processed and then go to reverse optimization of the denoising network, iterative operation between the Inception-RestNet and the hard verdict decoder can be reduce the effect of noise on the coded modulation system, thus enabling the decoder to obtain more accurate estimates. Experimental results show that the BER performance of this IR-ADMM decoder is improved by 0.5 dB with fewer iterations than the conventional LDPC decoder.