Introduction

At the ARL, signal processing is an integral part of what we do. Signals recorded underwater using hydrophones always need some processing. Apart from fairly standard signal processing such as filtering, and beamforming data collected from a hydrophone array, we have developed several of our own techniques to process signals recorded underwater.

To give you a flavor of the signal processing research at ARL, we have listed (non-exhaustive) some of the key signal processing techniques below:

• Denoising

  Recorded acoustic signals are often contaminated by various kinds of noise. In order to obtain clean signals, we have developed several denoising techniques. One of the notable techniques, based on singular spectrum analysis (SSA), separates recordings into transients, spectrally smooth noise and tonals.

• Signal Processing in Non-Gaussian Noise

  As high-frequency noise in local waters is highly non-Gaussian, most signal processing techniques (which make Gaussian noise assumptions) perform poorly. We have shown that symmetric alpha-stable (SaS) distributions describe the local ambient noise well. We have developed several techniques based on p-norm metrics and Myriad statistics that provide near-optimal performance in SaS noise.

• Spectrogram Image Processing

  Usually acoustic signals are processed as 1D signals. A spectrogram is a rich 2D image representation of the acoustic signal. We apply many of the tools developed for digital image processing to process the spectrogram; this yields novel techniques for processing acoustic signals.

• Signal Classification

  In many acoustic applications, it is important to be able to classify signals. We have developed classifications systems based on neural networks and dynamic warping to successfully meet many application demands.