ACOUSTIC TOMOGRAPHY: SERIAL VERSUS PARALLEL COMPUTATION

Acoustic tomography is a numerical technique used to obtain material properties of a medium using ultrasonic measurements. The computationally intensive nature of tomography is further complicated by the bending of acoustic rays in heterogeneous media. Since both the ray paths and the material properties are unknown, a double iterative technique is required for the solution. A conventional serial approach to the problem can become unwieldy and restrict the ability to define sharp velocity contrasts in a heterogeneous material. In this paper, a parallel approach to the tomography problem is discussed. Parallel computations allow the calculation of all ray paths simultaneously along with a simultaneous update of all the velocities in the sample grid. Thus, the number of source-receiver measurements and grid size can increase dramatically without a significant increase in computation time. This work was implemented on a Connection Machine CM200, a supercomputer with a massively parallel architecture. The Connection Machine is located at the Naval Research Laboratory. ALGORITHM DESIGN Three major components comprise the tomographic algorithm, time delay calculations, ray tracing and the tomographic inversion. The time delay calculations replace experimental measurements in this numerical simulation and are part of the forward program