Design and Implementation Considerations for a 3D Beamformer on State-of-the-Art Multicore Processors

Underwater targets have become increasingly difficult to be detected by sonars. Improvement in ship and submarine design technology have made it less noisy, and hence difficult to detect. And there is a proliferation of small Autonomous Underwater Vehicles (AUVs) for conducting surveillance operations, and recently, also for carrying out offensive missions. AUV are more difficult to detect compared to a submarine.This has resulted in sonar designers to search for options to improve the detection capabilities of sonars. However, there exist little options for sonar designers to improve the detectability of such underwater objects. Sonar designers have resorted to bigger arrays to improve the detection, by improving the directive gain of the array resulting in huge data to be handled by the signal processors. Improvement in processing technology has made available enough resources to handle the multitude of data thus generated. The processor technology in turn brings its own issues to be handled. The signal processor designer must take care the architectural details of the processor as well as its memory architecture to get maximum through put. The paper presents different approaches to implementation for a multicore processor. The approaches are based on OpenMP and POSIX thread. In the thread-based approach different types of stride of data are considered. The results show that the pthread based implementation with stride carefully chosen to avoid the memory conflicts results in high processor utilization. A higher block size enables the memory controller to keep the computational resources from starving for data, there by getting better utilization.