Based on the stepped annealing method used in transient chaotic neural network

a chaotic neural network model had been improved. The dividing point in the new model was chosen according to the change of the corresponding Lyapunov exponent

and made sure the network take good advantage of the property of chaos to search the global mini-mum and accelerate the convergence process. Various scales of the channel assignment problems had been simulated and researched based on channel assignment problem in mobile communication system. Kunz benchmark test problem was also used for the emulation of the channel assignment problem based on 25-district model. Simulation results show that compared with other transient chaotic neural networks

the new stepped annealing method decreases the computing time and enhances the searching ability

and has a better performance in the larger scale of network.