Tree-based VLSI architecture with applications to signal processing
Signal processing has been a major contributor in the field of pattern recognition, computer vision, image processing, speech analysis, etc. Special purpose computer structures specifically designed on VLSI are being tried for DSP applications. VLSI implementable structures have certain advantages like lower maintenance cost, high fault tolerance, modular design, etc. Most of the VLSI architectures designed are application specific. In this research, tree-based architecture is proposed and evaluated for several DSP applications.
This research is by far an extension of the original work done by S.K. Base and J. Dutta Gupta titled “Parallel processing with Lateral Hybridization of Trees” in 1990. Referred to as Ring Connected Trees or RCT, a two-dimensional lattice of (NxN) PEs is used for an (NxN) RCT. In the proposed architecture, a parallel processing environment for signal processing is provided by the multiple processing elements working in a coordinated way to do the computation. Multiprocessing along with pipelining results in lower computation time and faster results. The machine was tested for the parallel working of a number of computation problems from the signal processing domain at the same time. Finally, the research also mentions the layout of the structure for VLSI implementation to find the area-complexity of the implementation under the grid model inspired by C.D. Thompson’s “Area-time complexity for VLSI” published in 1984.
RCT has better VLSI area complexity as compared to that of Mesh-of-Tree and has linear time performance for signal processing computations. This research largely misses out on signal processing applications like auto-correlation, IIR filtering, Hadamard transform, Walsh transform and also towards multi-dimensional cases. However, a tree-based architecture has some obvious advantages like simplicity and regularity, area-efficient VLSI design, embedded hierarchic organization, ease of mapping signal processing algorithms to Tree-based structures, etc. These advantages along with the application-specific ones make it a commendable research in the field of VLSI in DSP.
S. K. Basu, J. D. Gupta and R. D. Gupta, “Tree-based VLSI architecture with applications to signal processing,” 1991., IEEE International Symposium on Circuits and Systems, Singapore, 1991, pp. 2105–2107 vol.4, DOI: 10.1109/ISCAS.1991.176699.
