Vector and multiprocessor implementation of digital filtering algorithms
The efficient implementation of digital-filtering algorithms using parallel computer systems is investigated. This study presents a new direction for the parallel implementation by enabling an efficient use of a large number of processors. Parallel computation of linear recursive equations is treated first because these equations are used for describing many digital-filtering algorithms. These equations bring dependency problems in the computation of multiple output samples at the same time due to the feedback structure. A new method is developed to overcome these problems. With regard to vector-processor implementations, vectorization methods are reviewed, and applied to the implementation of digital-filtering algorithms. In particular, recursive and adaptive filtering algorithms are implemented very efficiently by applying the developed parallel computation method for recursive equations. A multiprocessor implementation method which only requires a ring network for interprocessor communication is studied. The method divides input data into blocks, and assigns a block of data to each processor in rotation. The final area is devoted to the architectural design of a VLSI single-chip processor. This processor is equipped with pipelined functional units for processing vector operations efficiently.
- Research Organization:
- California Univ., Santa Barbara (USA)
- OSTI ID:
- 6445756
- Country of Publication:
- United States
- Language:
- English
Similar Records
Implementation of digital filtering algorithms using pipelined vector processors
A multiprocessor architecture for two-dimensional digital filters