Incorporation of prior knowledge of chip shapes in blind multiuser detection for DS-CDMA communication

In this paper we present a new method of compensating for complex multipath channel fading in blind multiuser detection for DS-CDMA communication. This compensation allows for the use of different chip pulse shapes. The compensation is based on the incorporation of priors derived from the FFT of the pulse shape. The system employs detectors that incorporate methods of independent component analysis (ICA), subspace estimation of channel noise, and Hopfield type neural networks. Several chip pulse shapes are tested, including the cosine, raised cosine, and the duobinary shapes, which are the pulse shapes that provide superior performance with more traditional detectors. They do not retain this superior performance when used with the uncompensated ICA detectors operating in a multipath fading channel environment. When compensated, the ICA neural network detectors displayed significantly better bit error rate performance than traditional detectors and the uncompensated ICA detectors.