Multipath Interference Suppression Based on on a Stable and Effective Neural Network

A multipath interference suppression model is proposed based on the experimental data in the multipath interference communication process. Based on the assumption that multipath interference is mostly of reflection origin, a reflector is placed in a simulation communication system to artificially generate interference. Through the neural network training for fitting the interference data to non-interference data, the multipath interference suppression model is build up. In order to simulate the diversity of interference sources as in the real communication process, three reflective materials of Chlorinated Polyethylene (PEC), aluminum and copper are used. Each of those materials is assumed with four different shapes, including plane, concave, convex, and waves, for comparison of various interference sources. The model parameters are optimized to adapt to the experimental data, mainly for the neural network from the aspects of model structure, data preprocessing, activation function selection, etc. So as to the simulation data under the experiment assumption, no many iterations are needed for achieving less than 0.0001 dB error. If communication scenarios and interference sources are replaced, the strong learning ability based on neural networks and adjustment of model parameters based on interference data distribution can provide rapid migration learning to fit a good effect.