Si plate radius influence on the photoacoustic signal processed by neural networks

TL;DR

This study investigates how the radius of a silicon plate affects photoacoustic signals processed by neural networks, emphasizing the importance of sample size relative to microphone aperture for accurate predictions.

Contribution

It provides a detailed analysis of the influence of sample radius on neural network predictions of photoacoustic signals, highlighting the significance of matching sample size to microphone dimensions.

Findings

Optimal prediction accuracy occurs for sample radii of 2-3 mm.

Networks trained with both undistorted and distorted signals perform best near the microphone radius.

Using samples with radii similar to the microphone aperture improves model validity.

Abstract

The effect of the sample radius on the total photoacoustic signal processed by neural networks trained with undistorded and distorded signals is carefully analized for modulation frequencies from 20 to 20 kHz. This is done for signals generated for a 400micrometers thick Si n-type plate, whose radius varies from 2 to 7 mm. It is found that the networks trained with both undistorded or distorded signals yield the best predictions for sample radii between 2 and 3 mm, which is close to the used microphone aperture radius of 1.5 mm. The network trained only with undistorted signals gives the best results for sample radii comparable to the microphone dimensions. The obtained results justify the validity of a theoretical model derived for a sample radius equals to the microphone aperture and indicate the experimental necessity to use samples with radii comparable to this aperture.