Analog capabilities of the BSB model as applied to the anti-radiation homing missile problem

The binary versus analog issues for inputs and outputs using the BSB model for the antiradiation homing sensor application, i.e. relatively unstructured data, is discussed. Using a fine-grain analog to binary code and weak positive feedback recall, the authors find that a fully interconnected BSB model produces analog as well as the traditional outputs. The analog outputs represent the shape and the magnitude of the binary feature distributions. The feature average (binary output) and the feature distribution can be used to more accurately classify the source of the signals than the average information alone. The authors believe that this distribution demonstration is a first for neural network technology. They have also found that, under unity feedback, the fully connected BSB model is capable of learning analog input shapes. This capability is helpful for the radar emitter classification goal, e.g. permitting storage of frequency spectra.<>