A novel architecture for computing eigenvalues of matrix for high speed applications
Soumitr Sanjay Dubey, Rahul Shrestha, Shubhajit Roy Chowdhury · 2016
This paper presents a new approach of calculating eigenvalues of any matrix, using triangular systolic array (TSA) architecture which incorporates three angle complex-rotation (TACR) architecture, as it reduces area and latency of the design. This TACR architecture has been further modified to reduce the area utilized by the architecture. Subsequently, the very-large scale-integration (VLSI) architecture has been conceived to calculate eigenvalues of general matrix using QR algorithm. Additionally, comparative study regarding the resource utilization on complex Givens rotation (CGR) and TACR with the proposed architecture has been carried out. Thereafter, the TACR architecture has been used in TSA architecture for the implementation of QR algorithm. Finally, latency of our design is 2.7 mS which is approximately 28 χ faster than the general purpose computer based solution. Power consumed by the design is 577 mW at a maximum clock frequency of 891 MHz which gives an error in the range of 0.001.