THE PROBLEM
Even the smallest noise can make the crew of a submarine extremely vulnerable. Therefore, the acoustic profile of modern submarines is taken into account from the design phase. Nevertheless, for various reasons (wear and tear, aging, poor assembly) an unacceptable level of sound may still arise. That being the case, to be able to remedy this, there must be tools available that enable these sounds to be attributed to one or more devices operating in the vessel.

In this context, LABRADOR software (LArge Bande Recherche Analyse et Détermination des ORigines [Wide Band Search Analysis and Determination of Sources]) has been developed on Sun® and HP® workstations. This software is based on three main modules, which the user sees as three graphics windows, designed to be as simple to use as possible. The bulk of the software has been written in C, using IMSL C Numerical Library routines.

THE SOLUTION
The first window in the user interface allows the user to define the problem that they wish to resolve by specifying first the location of measurement sensors near potential noise sources and those measuring radiated noise, and second, the signal-processing parameters that are going to be used for the analysis.

In the second window, known as standard analysis, the user can select from menus that the display standard values for acoustic and vibration analysis (auto spectrum transfer and coherence). The calculation of these quantities relies heavily on the routines contained in the IMSL C Numerical Library, in particular with regard to Fourier transforms and all operations on complex vectors. The display itself is produced using Exponent Graphics. In addition to very careful presentation of the graphics, Exponent Graphics ensures interaction between the user and the graphics, allowing customized presentation using menus and buttons (colors, character sets, marker type, line thickness, etc.). Exponent Graphics also allows the generation of output files in HPGLTM and PostScriptTM, in particular guaranteeing a faithful and careful reproduction of the window graphics, facilitating incorporation into reports.

In the third window - known as identification of sources - and after the program has made all the necessary calculations, users can select the display of different graphics allowing them to diagnose the responsibility for the radiated noise. Here also, the majority of operations rely heavily on the contents of the IMSL C Numerical Library, in particular the formation of a spectral matrix for each discrete frequency using the FFT routine, its breakdown into distinct values and vectors and a set of product and addition calculations on the complex vectors resulting from this breakdown. This set of complex operations has the aim of constructing imaginary sensors independent of one another (i.e., representing exclusively one source), based on real sensors, of which each is a linear combination of all the sources in operation.

The most important of the various graphical outputs in this attribution phase is the spectral synthesis, allowing a simple glance to identify the device(s) responsible for the radiated noise output. In effect, on this representation, each of the devices is allocated a color; the area covered between the background noise measured and the auto spectrum of radiated noise is colored proportionately according to the influence of each of the devices with the color that has been attributed to it.

Another graphics output enables a comparison to be made between the measurement of spectral density of radiated noise for all the devices in operation and that calculated by the software for the noise radiated by just one of these devices. This function is particularly interesting, because for the majority of the time, it is not possible for a device to function independently of the others, and a signal-processing algorithm combined with linear algebra enables simulation of this operating mode.

RETURN ON INVESTMENT
A LABRADOR software function makes it possible to eliminate the contribution of one or more of the sources identified in the signals measured by each sensor. This enables a display of the spectral density of radiated noise when one or more incriminated devices are not in operation.

The use of the IMSL C Numerical Library and Exponent Graphics libraries has made it possible to reduce the cost of development by exploiting tried and tested graph-plotting calculation routines, with interactive possibilities and formatted print outputs, while concentrating efforts on the specific processes.