SINR Prediction in Mobile CDMA Systems by Linear and Nonlinear Artificial Neural-Network-Based Predictors

This article describes linear and nonlinear Artificial Neural Network(ANN)-based predictors as Autoregressive Moving Average models with Auxiliary input (ARMAX) process for Signal to Interference plus Noise Ratio (SINR) prediction in Direct Sequence Code Division Multiple Access (DS/CDMA) systems. The Multi Layer Perceptron (MLP) neural network with nonlinear function is used as nonlinear neural network and Adaptive Linear (Adaline) predictor is used as linear predictor. The problem of complexity of the MLP and Adaline structures is solved by using the Minimum Mean Squared Error (MMSE) principle to select the optimal numbers of input and hidden nodes by try and error role. Simulation results show that both of MLP and Adaline optimal neural networks can track the effect of deep fading due to using a 1.8 GHZ carrier frequency at the urban mobile speeds of 10 km/h, 50 km/h and 120 km/h with tolerable estimation errors. Therefore, the neural networkbased predictor is well suitable SINR-based predictor in closedloop power control to combat multi path fading in CDMA systems.