Thomas Holme and Karen Watkins, "Optimal Architectures for Massively Parallel Implementation of Hard Real-time Beamformers"
Contemporary sonar systems are required to produce very high temporal and spatial resolution information. In the spatial domain, this is achieved by using arrays with larger number of sensors that are combined to produce multiple beams simultaneously. This "beamformation" process results in a very large number of digital signal processors working concurrently. The available alternatives in beamformer architecture are divided into two main classes. The applicability of a class for a given problem depends on the number of elements, the number of beams, and the amount of real-time changes required in beam characteristics. In this analysis, the two architectures will be modeled using Ptolemy in order to characterize the applicability to a beamformer currently under development. The study of the modeling results will determine the optimal architecture to be used based on system constraints. Some of these constraints are data throughput, number of processors, and dynamic control of beam characteristics. For the first phase of the project, journal papers pertaining to time-delay beamforming and different modeling techniques using Ptolemy will be researched. In particular, the focus will be synchronous dataflow models. For the second phase of the project, the model will be used to simulate each algorithm to determine the most efficient one to implement. C code will be generated in Ptolemy to find execution bottlenecks. These areas within the code will then be written in assembly for improved real-time operation.
This work was conducted in collaboration with the Applied Research Laboratories at The University of Texas in Austin, Texas, where Karen Watkins is employed.