Probabilistic neural networks for multi-user detection in code divisional multiple access communication channels

A probabilistic neural network is proposed and applied for implementation of a maximum likelihood detector and classifier. The network is trained using the algorithm based on Parzen probability density function estimation theory for detection of signals in CDMA multi-user communications Gaussian channel. By viewing these multi-user detector's problem as a nonlinear classification decision problem, we apply this algorithm which has the abilities of arbitrary nonlinear transformations, adaptive learning and tracking to implement this decision optimally and adaptively. The performance of the proposed neural networks detector is evaluated via extensive computer simulations and compared with other detectors and neural classifiers' schemes in a multi-user environment. The neural detector is shown to exhibits some desirable properties and significantly outperforms the conventional matched filter detector.