An improved learning algorithm for training neural network based lattice equalizer

In nonlinear channel equalization, artificial neural networks (ANN) have attracted a significant interest. The ANN's primary drawback is their intensive training. We propose suggestions for enhancing their training capacities. The first involves applying a whitening technique to the input data by employing a lattice structure as the equalizer. Lattice equalizer therefore becomes insensitivity to the inputs correlation matrix. In the second strategy, we suggest modifying the slope of the activation function. Combining the two methods increases the ANN's nonlinear capabilities and adaptability. Through simulation tests, the offered methodologies efficacy is verified. The results demonstrate that the performance of the neural network-based lattice equalizer is greatly improved by whitening the received data using adaptive lattice channel equalization techniques in conjunction with an adjustable slope activation function.