An Implementation of the MVDR Beamformer on the Intel iWarp System
Abstract
This report discusses the implementation of the minimum Variance Distortionless Response (MVDR) beamformer on the Intel iWarp system. A unidirectional ring and a full connection of processors were used in the implementations. The MVDR computation basically consists of a matrix factorization and a triangular system solver. A parallel Cholesky Factorization and a parallel QR factorization were implemented. A fan-in parallel algorithm was used for the triangle system solver. Using the unidirectional ring connection, the QR factorization achieved a 90% efficiency on 4 processors and a 64% efficiency on 16 processors: the Cholesky factorization achieved a 83% efficiency on 4 processors and a 44% efficiency on 16 processors. Using the unidirectional ring connection and QR factorization, the MVDR beamformer achieved a 83% efficiency on 4 processors and 34% on 16 processors. Using the full connection of 4 processors, a 20% performance improvement has been obtained on MVDR as compared with the unidirectional ring connection. The parallel processing performance in terms of MFLOPS is still low, which is expected considering the low node performance on a C code using software release 2.3.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1992
- Accession Number
- ADA253641
Entities
People
- J. Z. Lou
Organizations
- Naval Command, Control and Ocean Surveillance Center