Hopfield neural network for AR spectral estimator

An autoregressive (AR) spectrum estimator using the Hopfield neural network (HNN) is introduced. The HNN is designed to minimize the mean squared error between a subject signal and the assumed AR model of the signal. The output of the HNN is the estimated AR coefficients; thus, the spectrum of the signal can be directly obtained in terms of the AR coefficients and the sampling interval. An odd symmetric soft-limiter-type neuron is selected for the HNN rather than the sigmoid or the hard limiter type. The idea is extended for the bearing estimation problem for sonar and radar. A hardware implementation of the model is discussed