Enhancing SONAR performance through statistically-informed sub-array processing

Ocean fluctuations affect acoustic propagation and reduce the performance of SONAR arrays. In particular, internal waves can lead to a loss of coherence in incident signals, reducing the efficiency of classical processing. Precise knowledge of the environment is necessary to predict the degradation it causes. However, it is impossible to know the exact state of the ocean, it is therefore necessary to use statistical methods to study the ocean and acoustic propagation. In this work, statistical method, canonical correlation analysis (CCA), is used to find linear relationships between acoustic variables of interest and oceanographic measurements. In particular, the coherence radius is inferred from empirical modes of temperature using the model learned by CCA. A sub-array processing scheme can then be informed by defining the length of the sub-arrays from the coherence radius. To compensate for the loss of angular resolution inherent in sub-arrays, we introduce the use of a Bayesian algorithm exploiting l0 norm regularization. We show on experimental data from the ALMA 2017 campaign that the proposed processing improves detection when signal coherence is low.