Abstract In this paper, applying the Deep Multilayer Perceptron Neural Network (MLPNN) to the Sum-Product Algorithm (SPA) for decoding the Modified Welch-Costas (MWC) coded Optical Code Division Multiple Access (OCDMA) system with Low-Density Parity-Check (LDPC) code is analyzed. The goal is to train the MLPNN-SPA through the stochastic gradient descent (SGD) to learn and optimize the weights to each edge of the neural network decoder. Once these parameters have been trained, the decoding complexity of the MLPNN-SPA is similar to that of the SPA. Furthermore, from the simulation results, it has been shown that the MLNN-SPA can improve the system performance without additional decoding complexity as compared to the SPA.
Read full abstract