Signal Processing by Polynomial NN and Equivalent Polynomial Function

In this paper, the signal processing by using polynomial neural network (NN) and its equivalent polynomial function is studied and simulated. To demonstrate the superiority of the equivalent polynomial function proposed, the signal recognition in a two-dimension (NC2) non-convex system and system identification were simulated and discussed. All simulations were performed by using the conventional polynomial NN and its equivalent function as a comparison. From the simulation results shown, the error back-propagation learning rule based on the steepest descent algorithm easily makes the neural network get stuck in the local minimum. On the contrary, the equivalent polynomial function using least-mean-square (LMS) tuning method can easily reach the optimal solution.