An efficient architecture for multi-dimensional convolution

@article{Elnaggar2000AnEA,
  title={An efficient architecture for multi-dimensional convolution},
  author={Ayman Elnaggar and Mokhtar Aboelaze},
  journal={IEEE Transactions on Circuits and Systems Ii: Analog and Digital Signal Processing},
  year={2000},
  volume={47},
  pages={1520-1523}
}
  • A. Elnaggar, M. Aboelaze
  • Published 1 December 2000
  • Computer Science
  • IEEE Transactions on Circuits and Systems Ii: Analog and Digital Signal Processing
This paper presents modified parallel architectures for multidimensional (m-d) convolution. We show that for two-dimensional (2-d) convolutions, with careful design, the number of lower-order 2-d convolutions can be reduced from nine to six with a computation saving of 33%. However, the original speed of the computations is not affected. The proposed partitioning strategy results in a core of data-independent convolution computations, and can be generalized to the m-d convolution. The resulting… 
7 Citations

Figures and Tables from this paper

An efficient algorithm for multidimensional convolution
  • A. Elnaggar, M. Aboelaze
  • Computer Science
    2000 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. No.00CH37100)
  • 2000
TLDR
This paper presents a novel recursive algorithm for generating higher-order multidimensional convolution by combining the computation of 3/sup m/ identical lower-order convolution computations, and its implementation in parallel VLSI networks.
A Unified Framework for Mapping Multidimensional Separable DSP Transforms onto parallel VLSI Structures
  • A. Elnaggar, M. Aboelaze
  • Computer Science
    2007 IEEE International Symposium on Signal Processing and Information Technology
  • 2007
TLDR
The objective of the work is to derive a unified framework and a design methodology that allows direct mapping of the proposed algorithms into modular VLSI architectures.
Embedded Reconfigurable Architectures for Multidimensional Transforms
TLDR
The objective of the work is to derive a unified framework and a design methodology that allows direct mapping of the proposed algorithms into embedded reconfigurable architectures such as FPGAs.
A NEW MULTIDIMENSIONAL ARCHITECTURE FOR COMPUTING WHT
TLDR
A new recursive formulation for Walsh-Hadamard Transform (WHT) is presented that allows the generation of higher order (longer size) multidimensional WHT architecture from 2 m lower order (shorter sizes) WHT architectures.
A new recursive formulation for 2-D WHT
  • A. Elnaggar, M. Aboelaze
  • Computer Science
    Proceedings of the 2003 International Symposium on Circuits and Systems, 2003. ISCAS '03.
  • 2003
TLDR
A new recursive formulation for computing the Walsh-Hadamard Transform (WHT) is presented that allows the generation of higher order (longer size) 2-D WHT architecture from four lower order (shorter sizes) WHT architectures.
An Efficient Methodology for Mapping Algorithms to Scalable Embedded Architectures
This paper presents a general approach for generating higher order (longer size) multidimensional (m-d) architectures from
New systolic array processor architecture for simultaneous discrete convolution of an image plane with multiple filter coefficient sets
TLDR
Results from postimplementation virtual hardware prototype simulation testing and from testing a PLD-based experimental hardware prototype are shown that validate correct functional and performance operation of the new convolution architecture module.

References

SHOWING 1-6 OF 6 REFERENCES
A new recursive algorithm for multidimensional convolution
TLDR
A novel recursive algorithm for generating higher order m-dimensional (m-D) convolution by combining the computation of 3/sup m/ identical lower order convolution computations, and its implementation in parallel VLSI networks, which have very simple modular structure, highly regular topology, and use simple arithmetic units.
Higher radix aperiodic-convolution algorithms
TLDR
The contribution concludes with a comparison which clearly illustrates that the new methods lead to algorithms which are more efficient than algorithms based on the fast Fourier transform or rectangular transform for filter lengths up to 1296 for the case where the data block length is equal to the filter length.
A new tensor product formulation for Toom's convolution algorithm
This correspondence presents a new recursive formulation of Toom's algorithm that allows the generation of higher order (longer size) one-dimensional (1-D) convolution architectures from three lower
An advanced programmable 2D-convolution chip for, real time image processing
  • V. Hecht, K. Ronner
  • Computer Science
    1991., IEEE International Sympoisum on Circuits and Systems
  • 1991
TLDR
An advanced defect-tolerant systolic array implementation of the 2-D convolution algorithm for real-time image processing applications is presented and defect tolerance techniques are implemented in order to enhance yield and reliability, especially for future large area implementations.
High-speed systolic ladder structures for multidimensional recursive digital filters
TLDR
A new 1-D filter structure is introduced that proves the most suitable structure in terms of a systolic ladder implementation, because it leads to MD ladder filter structures possessing such important properties as the shortest critical path, the canonic number of high-level storage registers, and local interconnectivity.