The neural network multi-user detection based on MMSE

The problem about multiuser detection eventually is a combinatorial optimization problem. Hopfield neural network can get the near-optimal combinatorial optimization solution instantly by dynamic evolution of itself; and it has a fast convergence time. This is necessary for real-time multi-user detection. We remove the constraints because MMSE is a free minimization problem, and let the linear transfer matrix corresponds to the neural network connected matrix and bias current corresponds to spread sequences We get the HNN linear multiuser detection algorithm based on MMSE criteria, called the new MHNN. Simulation result shows that the error bit ratio (BER) decreases compared with the former MHNN and HNN algorithm and it increases system capacity. This is because the MHNN algorithm solves the local optimization problem of original neural network and using the optimal objective function based on MMSE.