Dual neural network models in acoustic propagation

Abstract

This paper presents multiple neural-network models to mimic a nonlinear dynamic system. The multiple neural network models consist of one or more simplified time-varying functions to dynamically approximate the nature of the physical phenomena to be interpolated and extrapolated. The purpose of using the multi-model function is to perform a real-time approximation for a complicated nonlinear system. The multi-model function was demonstrated using the underwater acoustic transmission loss data generated from the Navy-standard acoustic propagation-loss model ASTRAL. The interpolator-learning period for a 200 ft receiver interval, an 800 ft source interval, an 8000 Hz frequency range, and a 25 nautical time range window takes about 20 minutes (more or less time depends on the size of the parameter intervals and the complexity of the ocean environment). The interpolation speed is measured in fractions of a second, and the interpolation error is around 1% of the actual transmission-loss value in a root-mean-square (RMS) sense.