Hebbian learning based blind adaptive multiuser detection in DS-CDMA systems

Nonlinear Hebbian learning rules have been instrumental in independent component analysis (ICA) and blind source separation (BSS) problems. We show that in a CDMA system, where the users can be assumed independent, application of these rules give accurate estimates of user signature codes and data from the observations alone. However, the solutions are obtained within a sign and permutation ambiguity. A novel method is then proposed for the resolution of these ambiguities. It is shown via numerical examples that this technique is robust in near-far situations and can be used to estimate various other parameters such as user amplitudes and cross-correlation values. Computer simulations indicate improvements in terms of reduced error probability in detection and increased signal to interference plus noise ratio (SINR). Convergence of the proposed detector is also comparable to that of the blind minimum output energy (MOE) detector