Ultrasonic distance sensor improvement using a two-level neural-network

Abstract

This paper discusses the performance improvement that a neural network can provide to a contactless distance sensor based on the measurement of the time of flight (TOF) of an ultrasonic (US) pulse. The sensor, which embeds a correction system for the temperature effect, achieves a distance uncertainty (rms) of less than 0.5 mm over 0.5 m by using a two-level neural network to process the US echo and determine the TOF in the presence of environmental acoustic noise. The network embeds a "guard" neuron that guards against gross measurement errors, which would be possible in the presence of high environmental noise.