A new design of channel denoiser using residual autoencoder

A joint neural network decoder and denoiser scheme demonstrated superior performance compared to individual modules. However, there is still a limitation that the existing denoisers cannot effectively learn patterns of encoded signals. To overcome the limitation, a novel denoiser based on a residual autoencoder structure is proposed. The proposed denoiser speeds up the training process and boosts the performance due to its structure effectively extracting compressed features. For the evaluation, a joint system model with a hyper‐graph‐network decoder that is known for outstanding decoding performance is considered. Simulation results show that this denoiser outperforms the existing denoisers. Furthermore, the proposed joint model shows significant performance improvement compared to the individual hyper‐graph‐network decoder with only 1% of the number of epochs for the training. A novel denoiser based on a residual autoencoder structure is proposed. The proposed denoiser speeds up training process and boosts the performance due to its structure effectively extracting compressed features. Simulation results show that this denoiser outperforms all the existing denoisers.