1 Contributions to passive acoustic oceanic tomography – Inversion Algorithms based on Time Frequency Space Representation C. Gervaise*, S. Vallez*, E. Boumansour*, H. Le Floch*, A. Martin*, A. Khenchaf*, Y. Simard** * E3I2, ENSIETA, EA3876, 2 Rue François Verny, 29200, Brest, France ** Institut des Sciences de la Mer, Université de Québec à Rimouski, 310 Allée des Ursulines, Rimouski, Québec G5L 3A1, Canada, cedric.gervaise@ensieta.fr ABSTRACT This paper addresses the design of signal pre-processors dedicated to passive acoustic tomography. Blind processors of channel’s impulse response and spatial response are proposed and adapted with wide band transient signals such as marine mammals vocalizes. Full validation of blind channel’s impulse response processor is performed on real world data whereas validation of blind spatial response estimation is carried out on realistic synthetic data. 1. INTRODUCTION Acoustic tomography is a way to produce a fast, accurate and cheap monitoring of water mass. This monitoring requires an inversion procedure made with two steps. The first one is to estimate acoustic properties (such as the sound speed profile of the water column) from the measurement of a propagated known acoustic waveform between fixed sources and receivers. Then a second step consists in inferring some physical ocean parameters (temperature, bottom nature) from these previous estimated acoustic characteristics. Large scales deep water and small scales shallow water configurations were successfully studied and associated with matched delay, matched field and matched impulse response inversion processing. Accurate estimate of acoustic properties requires the emission of powerful and recurrent signals in the adapted bandwidth and in agreement with the scale of the monitoring. But we would rather not send these hard active sounds through the water column in a potential military underwater warfare context, or if mammal species health is considered. A recent solution has emerged in the community to tackle this problem with the passive tomography processing. Passive tomography processing consists in estimating acoustic properties by using opportunity sources present in the channel at the time of interest. Some experimentations were recently carried out using ships, marine mammals and surface noises. Different levels of complexity can be formulated to insure the discreetness of tomography processing. The first one is the Active Discreet Tomography (ADT) where active emission is allowed but with a waveform chosen to insure a low probability of interception using for instance a copy of a noise component or a spread spectrum signal. In that case, Signal to Noise Ratio (SNR) is reduced compared to classic active tomography. The second one is the Aided Passive Tomography (AiPT) where active emission is forbidden but where a cooperate entity of known position can produce an acoustic emission linked to its natural activity. Blind estimation of the impulse response of the channel is performed with the losses of absolute time and magnitude references, and a reduced SNR. The last one is the Autonomous Passive Tomography (AuPT) where active emission is forbidden but where an entity of unknown position can produce an acoustic emission linked to its natural activity. As in the case of the AiPT, blind estimation of the impulse response of the channel is performed with the losses of absolute time and magnitude references, SNR is reduced, and moreover, position of the source considered as a nuisance parameter has to be estimated jointly with the parameters of interest. For AiPT, preliminary works conducted on performance prediction based on lower Cramer Rao bound calculus have demonstrated that as soon as celerity profile or bottom parameters estimation are concerned, the blind estimation of impulse response of the channel between emitter and receiver in relative time (referenced to the first arrival) carries enough information to inverse the problem. Thus, the first part of the paper addresses the problem of Blind Impulse Response estimation using transient opportunity source with clear time frequency contents such as marine mammals vocalizes. Low- resolution and high-resolution are developed and applied with success to real data obtained from Laurentian channel experiment performed in summer 2003. For AuPT, the previous preliminary works have demonstrated that as soon as celerity profile or bottom parameters estimation are concerned at the same time that source position estimation, the measurement of direction of arrival associated with blind impulse response channel estimation carried enough information to inverse the problem. Then a second part of the paper deals with the development of a Spatial Time Frequency processor to estimate the temporal and spatial structure of the arrivals. This processor is applied with success to synthetic but realistic data in shallow water environment. 2. TIME FREQUENCY PROCESSOR FOR AiTP When the central frequency of the opportunity source is high enough, acoustic ray paths propagation takes place and signal at the receiver can be seen as a sum of attenuated and delayed versions of emission. Then, if the emission has a clear time frequency content