Power prediction in reverse link for mobile DS/CDMA systems using support vector regression

This paper presents an application of the support vector regression (SVR) in prediction of received signal power in the direct sequence code division multiple access (DS/CDMA) systems. The predictor selects the parameters by using five-fold cross-validation method. The results are evaluated in term of minimum mean square error (MMSE.) The inputs for the predictor are the past values of signal series and the output is the next step ahead value. The SVR-based predictor is compared to the previously proposed linear and nonlinear neural network-based predictors, i.e., the adaptive linear (Adaline) predictors and the multilayer perceptrons (MLP), respectively. A noisy Rayleigh fading channel with 1.8 GHz carrier frequency in an urban environment is simulated as the wireless channel. The results show that the SVR-based predictor can estimate the power better than the Adaline and MLP predictors by considering the signal-to-noise ratio (SNR).