A Non-Linear Improved CNN Equalizer with Batch Gradient Decent In 5G Wireless Optical Communication

Abstract

An equalization scheme employing convolutional neural networks (CNN) is proposed in generalized frequency division multiplexing (GFDM) for propagation in a hybrid microwave-optical system. Batch Gradient Descent is used to train the CNN equalizer quickly. The approach suggested eliminates nonlinearity and produces improved performance. By incorporating the GFDM network with successive interference cancellation (SIC)-based receiver, this solution greatly increases the storage ability of consumers and also offers a larger coverage range. The suggested CNN equalizer is correlated with other approaches, including statistical models and estimator techniques. The experimental finding shows that CNN produces the highest value that is marginally better than the other equalizers with a numerical difficulty reduced to zero. The simulation findings suggest that QAM-GFDM can attain the same bit error rate (BER) as cyclic-prefix (CP) OFDM without compromising spectrum economy. The tests obtained have shown that the new approach works much better than traditional techniques.