Underwater acoustic networks operate in an inhomogeneous and dynamic environment, which makes it difficult to model the propagation path of signals. In essence acoustic signals experience reflection and refraction due to sound speed variation, based on many parameters such as salinity, temperature, and depth. To enable modeling of signal propagation, the sound speed profile (SSP) has to be accurately estimated. The most famous SSP equation has been proposed by Mackenzie and has been widely used among others like the Coppens’ and UNESCO equations. The drawback of these equations is that they yield different accuracy levels for various setups. They are also valid only for certain limits of salinity, depth and temperature. Moreover, the SSP estimation method should suit both deep and shallow water environments. In this paper, we use machine learning algorithms to predict sound speed in both deep and shallow waters and compare our results with data collected from acoustic tomography measurements. For training we have considered sound speed measurements across various oceans like Pacific Ocean, Arctic Ocean, Indian Ocean, etc. Our results show that our model achieves 99.99% accuracy and outperforms Leroy and Mackenzie equations.
Paper
Full text
Machine Learning Based Sound Speed Prediction for Underwater Networking Applications
Semantic Scholar · Environmental Science · 2021
Abstract
Underwater acoustic networks operate in an inhomogeneous and dynamic environment, which makes it difficult to model the propagation path of signals. In essence acoustic signals experience reflection and refraction due to sound speed variation, based on many parameters such as salinity, temperature, and depth. To enable modeling of signal propagation, the sound speed profile (SSP) has to be accurately estimated. The most famous SSP equation has been proposed by Mackenzie and has been widely used among others like the Coppens’ and UNESCO equations. The drawback of these equations is that they yield different accuracy levels for various setups. They are also valid only for certain limits of salinity, depth and temperature. Moreover, the SSP estimation method should suit both deep and shallow water environments. In this paper, we use machine learning algorithms to predict sound speed in both deep and shallow waters and compare our results with data collected from acoustic tomography measurements. For training we have considered sound speed measurements across various oceans like Pacific Ocean, Arctic Ocean, Indian Ocean, etc. Our results show that our model achieves 99.99% accuracy and outperforms Leroy and Mackenzie equations.