High-speed sensing for object detection in underwater bi-static acoustic paths Angela Digulescu *† , Cindy Bernard * , Elena Lungu , Ion Candel * and Cornel Ioana * * University of Grenoble-Alpes, GIPSA-lab, Saint Martin D’Heres, France Email: [angela.digulescu, cindy.bernard, ion.candel, cornel.ioana]@gipsa-lab.grenoble-inp.fr Sigintec, 334 rue Tour de l’Eveque, 30000, Nimes, France Email: elungu@sigintec.fr Abstract—Detection of objects in underwater environment is an important operation that can be aimed to assess the pollution in the region under study as well as to prevent intrusion of undesired objects in off shore power production facilities or in the restricted entrance area. The use of acoustic methods has been proven to be very powerful for the detection and localization of the underwater objects. This paper proposes a new method for the detection of an underwater object using bi-static acoustic paths. The method uses the dynamic evolution the phase space of the received signals, namely the phase space trajectory, without any synchronization and it computes the point wise Euclidean norm. The detection map obtained gives both the amplitude and time of arrival information. The major advantage brought by this approach (rather than conventional techniques) is that the signals need no synchronization, their characterization being based only on their wave form. Keywordsphase space; recurrence matrix; detection curve; time of arrival I. I NTRODUCTION Underwater object detection is an important task for the marine community and it represents a heavy duty for the signal processing field. Its purpose can relate to the monitoring of the underwater fauna as least intrusive as possible, as well as to supervise the pollution in the studied area. While the detection part has been studied using vari- ous methods like correlation techniques [1], time-frequency methods [2], sonar imaging [3], etc., our approach is based on high speed sensing in order to determine the time of arrival (TOA) of the object relative to a reference response. This implies that there is no synchronization between the emission and reception. Moreover, our approach also proposes to characterize the trajectory of the object. Therefore, the application is based on a wide-band active acoustic system, these types of signals being more robust to on-site drawbacks like: ambient noise, electronic noise, interference sources, etc. [2]. Then, the TOAs are determined using the Recurrence Plot Analysis (RPA) method. The method is applied for each response of a fixed length and the result is stored in a detection map which contains the time bins of the responses and the time distributed recurrences (TDR) computed for each time bin. Next, the TOAs of each response are determined relative to the first one and a comparison with the classical signal envelope approach is done. The paper is organized as follows: in section II, the RPA method will be presented; section III will describe the exper- imental setup and in section IV, the results are highlighted. Then, in section V, the most important ideas and future work aspects are presented. II. RECURRENCE PLOT ANALYSIS METHOD The Recurrence Plot Analysis (RPA) is derived from the dynamical systems theory and it considers the evolution of a measurement in its phase space [4], [5]. Therefore, we firstly consider a received signal as the time series [6], [7] expressed in (2): x = {x[1],x[2], ..., x[N ]} (1) where N is the length of the received signal. Then, the time series is represented in a m dimensional phase space. The vectors that describe the trajectory of the dynamical system have as coordinates m values of the time series which equally spaced: v i = m k=1 x[i +(k - 1)d] · e k ,i = 1,M (2) where v i are the vectors from the phase space corresponding to a state of the system, m is the embedding dimension, d is the the delay (lag) between the samples of the received signal, M = N - (m - 1)d and e k are the axis unit vectors. Afterwards, the next step of the RPA method is to compute the distance/ recurrence matrix. The distance matrix determines the pointwise distance between the points of the phase space. These points are given by the position vectors expressed in (2): D i,j = Δ(v i ,v j ) (3) R i,j = Θ(ε - Δ(v i ,v j )) (4) where Δ(v i ,v j ) from (3) is a distance considered between the points from the phase space (the Euclidean distance, the L 1 norm, the angular distance [7], etc.). The operator Θ(·) from (4) is the Heaviside step function and ε(·) is the chosen threshold that discriminates if one distance is a states of recurrence or not (usually is value is constant). The most important parameters of the RPA method are the delay, d and the embedding dimension, m. The choice of the delay [8] is critical for the ”quality” of the trajectory in the phase space. If the delay is chosen too small, then the trajectory is over folded around the main diagonal of the phase space, hereby the representation is redundant. But, if the delay is 978-0-933957-43-5 ©2015 MTS