Stochastic Hopfield Network for Multi-user Detection

In this paper a novel multi-user receiver is introduced, which unites fast convergence of neural networks with the asymptotically global optimization power of stochastic algorithms (e.g. Boltzmann machines). The proposed method is capable to achieve a 1..2 dB gain in performance over the traditional Hopfield neural network, while only 2 or 3 times more iterations is needed, which still does not prevent real-time application. Furthermore, the algorithmic complexity of the new method is the same as the original Hopfield type of multi-user detector, hence no additional hardware is required.