Performance prediction of a coded digital communication system using cross-validation

This paper deals with the bit error rate (BER) semi-analytical prediction in a turbo coded digital communication system. We propose a new prediction method which is based on the kernel estimator of the probability density function (pdf). We assume that no knowledge on the distribution of the received soft samples is available. In the proposed method, we derived a new expression of the smoothing parameter which takes into account of the soft samples at the detector output. In this paper, we use the cross validation (CV) criterion for selecting the optimal smoothing parameter. We prove that the proposed method outperform MISE (mean integrated squared error) criterion in terms of squared error. After, application to BPSK modulation and coded SC-FDMA (Single carrier frequency division multiple access) system; Simulation results, shows that the BER obtained with the proposed method are close to that measured using Monte-Carlo method. Moreover, the use of cross validation criterion decreases the squared error between the true pdf and the estimated one.