ROMANIS: An ambient noise imaging camera

Just like we use ambient light to see things around us, why not use ambient noise in the ocean for sensing? The principle underlying ambient noise imaging (ANI) makes use of the ensonification provided by the ambient noise field to create pictorial images of underwater objects. ROMANIS is a broadband acoustic camera built at the ARL, and successfully used for ANI, passive target detection/ranging, and mapping of the underwater environment. The ROMANIS sensor array features 508 simultaneously sampled acoustic sensors, each sampled at around 200 kSa/s, and streamed to the surface at a rate of about 1.6 Gbps. The data is processed in realtime on GPUs to form an acoustic “video” of the underwater environment. The ROMANIS system design uses novel techniques such as dry-coupled neoprene acoustic window, Helium-based thermal management, fractional-wavelength staggered sensor placement for optimal beamforming, and a hierarchical modular setup to reduce harnesses and electronic noise.

Most of the “illumination” we use in ROMANIS is from snapping shrimp. The noise made by snapping shrimp is compact in time, broadband, and highly impulsive. With a large number of these snaps arriving at a receiver, the generalized central limit theorem allows us to model the resulting pressure variation accurately using symmetric α-stable distributions. Signal processing based on these distributions leads to ANI algorithms that utilize fractional low-order moments (FLOM) and fractiles. These algorithms yield more stable and consistent results, as compared with previously developed methods that employ energy estimates and high-order statistics.

An alternative approach to using the statistics for imaging, is to localize loud snaps in space and time, and then to utilize the knowledge of the source to deterministically process the echoes. While this idea is attractive and offers the possibility of passive ranging and 3-dimensional imaging, it is difficult to implement in practice. The key difficulty lies in associating a snap with its echoes, as all snaps sound similar and arrive at the receiver overlapping in time with the echoes from other snaps. A voting algorithm we developed, solves the association problem in a quasi-static scenario using some physical constraints and an ensemble of arrivals. Using this algorithm, ROMANIS has been able to estimate range to targets passively. This demonstration opens a whole new exciting area of research, where existing techniques developed for multi-static sonar may be applied to ANI.