HOPFIELD NEURAL-NETWORK IMPLEMENTATION OF THE OPTIMAL CDMA MULTIUSER DETECTOR

We investigate the application of Hopfield mural networks (HNN's) to t he problem of multiuser detection in spread spectrum/CDMA (code divisi on multiple access) communication systems, It is shown that the NP-com plete problem of minimizing the objective function of the optimal mult iuser detector (OMD) can be translated into minimizing an HNN ''energy '' function, thus allowing to take advantage of the ability of HNN's t o perform very fast gradient descent algorithms in analog: hardware an d produce in real-time suboptimal solutions to hard combinatorial opti mization problems. The performance of the proposed HNN receiver is eva luated via computer simulations and compared to that of other suboptim al schemes as well as to that of the OMD for both the synchronous and the asynchronous CDMA transmission cases. It is shown that the HNN det ector exhibits a number of attractive properties and that it provides in fact a powerful generalization of a well-known and extensively stud ied suboptimal scheme, namely the multistage detector (MSD).