A hybrid receiver structure for DS-CDMA

This paper presents a hybrid receiver structure for DS-CDMA tailored for nonlinearly separable scenarios. It consists of two stages, a linear and a nonlinear network, and has reduced complexity when compared with the optimum nonlinear receiver. The filter weights of the linear network are determined by linear programming (LP) and this linear stage acts as a preprocessor receiver. The second network is responsible for resolving the nonlinearly separable signals and consists of a truncated radial basis function (RBF) network. Since both stages require the same a prior knowledge, the computational complexity is not increased in order to construct both networks. It was found that the separability is dependent on the length of the spreading code, the channel impulse response and the preprocessor stage at the front end, and on the distribution of the expected (noise free) signals. Despite the capability of classifying the signals correctly, a relative high SNR is required due to the nature of the classification problem