Wireless Personal Communications 27: 195–213, 2003. © 2003 Kluwer Academic Publishers. Printed in the Netherlands. A Neural Network-Based Blind Multiuser Receiver for DS-CDMA Communication Systems ROMANO FANTACCI, LEONARDO MANCINI, MAURO MARINI and DANIELE TARCHI Dipartimento di Elettronica e Telecomunicazioni, Università di Firenze, Via S. Marta, 3, 50139 Firenze, Italy E-mail: {fantacci; mancini; tarchi}@lenst.det.unifi.it and marini@ing.unifi.it Abstract. The Universal Mobile Telecommunications System (UMTS) which is based on Wideband-Code Di- vision Multiple Access (W-CDMA) techniques is one of the most important broadband wireless communication systems. Adaptive Blind Multiuser Detection was widely considered for mobile receivers. The main drawback of this approach is that it achieves the optimum solution after a certain number of bit times. This paper deals with a new neural network approach in order to reduce the convergence time in different application environments. In particular, a modified Kennedy-Chua neural network, based on the Hopfield model is proposed. The neural network stability was investigated by means of a suitable analytical approach, while the performance of the pro- posed receiver scheme was derived by means of computer simulations. The numerical results shown in this paper highlight a fast convergence behavior of the proposed scheme, in particular under multipath-fading conditions. Keywords: neural networks, CDMA, blind detection. 1. Introduction Mobile radio communications have experienced a larger development in the last ten years, es- pecially with the upcoming IMT-2000 (International Mobile Telecommunication, 2000) third generation mobile systems. Among the multiple access techniques proposed for this system, such as Wideband Code Division Multiple Access (W-CDMA) with Frequency Division Du- plexing (FDD) or Time Division Duplexing (TDD), and cdma-2000 [1, 2], FDD/W-CDMA was chosen by European Standardization groups for the future Universal Mobile Telecom- munications System (UMTS). Many research activities were focused on the most suitable receiving techniques for W-CDMA mobile systems. In particular, the multiuser adaptive blind detection algorithm proposed by Honig et al. [3], seems to be a suitable approach to improve bit error rate (BER) performance at the receiving end. This receiver has the main advantage of achieving minimum mean squared error (MMSE) performance without requiring any kind of training sequence, hence improving bandwidth use. However, blind receivers need a certain number of bit times to achieve the optimal so- lution, since the optimum filter coefficients definition is performed according to a step by step approximation method, based on the steepest gradient-descent technique. This behav- ior has a dramatic impact on the receiver BER performance, in particular under typical multipath-fading propagation conditions. An interesting approach to mitigate this drawback seems to be that of using neural networks that usually have a low convergence time, being based on parallel-processing algorithms. Neural network applications to CDMA systems have been previously considered in the liter- ature [4–7]. In particular, in [8], an adaptive CDMA multiuser detector with a self-organizing