Automatic detection and classification of sounds generated by ocean vehicles via cyclic demodulation and beamformed coherence

Narrow-band tonal signals generated by ocean vessels are a dominant source of underwater sound, occurring at discrete frequencies ranging from several Hertz to several kiloHertz and continuously over long time intervals. Here we develop a two-stage approach for the automated passive acoustic detection and classification of underwater vehicles. The first stage utilizes the commonly used cyclic demodulation to extract the low frequency tonal sound modulated into high frequency noisy signal from which the fundamental frequency and blade pass frequency of ship propeller can be determined. The second applies magnitude-squared coherence to beamformed signals to directly extract tonal sounds spanning a wide range of frequencies from low (∼10 Hz) up to the array cutoff frequency and to estimate their bearings. Both the modulated and unmodulated tonal signals generated by a ship can be simultaneously determined using this two-stage approach leading to a more complete spectral characterization of the underwater sound radiated by an ocean vessel. Here, the approach is applied to analyze and characterize the underwater sound generated by research vessel RV Knorr, received on a large-aperture densely populated coherent hydrophone array.