Detection of Coherent Bioacoustic Signals in Underwater Noise

Challenges in signal detection in the presence of noise arise in many underwater acoustic applications such as active and passive sonar, geophysical surveying, and marine life monitoring. The detection problem is generally formulated within the framework of binary hypothesis testing. An important class of detectors, known as the locally optimum detectors, are designed based on the expectation that the signal of interest is relatively weak compared to noise. The West Indian manatee, added to the endangered species list in 1967, can be routinely found in the shallow Florida waterways and channels. These waterways are heavily utilized by recreational boaters, and a majority of the unnatural manatee deaths are attributed to watercraft collisions. Another major underwater acoustic noise source in Florida channels result from shrimp crackles. Recent research effort has been directed towards developing and implementing advanced signal processing techniques for far range passive acoustic detection of manatee vocalizations. In this paper, the problem of detecting manatee vocalizations in noisy underwater acoustic environments is addressed. Manatee vocalizations are modelled as narrowband, autoregressive, harmonic signals, and boat noise is modelled as a Gaussian process. Two impulsive noise models, the Middleton Class-A and the SαS PDF model for snapping shrimp crackles are investigated. Locally optimal or suboptimal detectors for each noise case are implemented in the form of Gaussian-Gaussian mixture and the Cauchy locally optimal detectors. The performance of these detectors is evaluated through Monte-Carlo simulations and results indicate that the locally optimum Gaussian detector outperforms the non-Gaussian detectors, even when noise contamination is significantly impulsive.