CORDIC instruction set extensions for matrix decompositions on Software Defined Radio processors

Software Defined Radio (SDR) is favored by the wireless industry as the platform of choice for implementing physical layers of wireless protocols due to its significant benefits of reduced development costs and accelerated time-to-market. However, to attain high spectral efficiency, emerging wireless protocols use increasingly complex two-dimensional techniques that are extremely expensive to implement using conventional Digital Signal Processor (DSP) instruction sets. In this paper, we present COordinate Rotation DIgital Computer (CORDIC) instruction set extensions that speed up the QR Decomposition (QRD) and Singular Value Decomposition (SVD) of complex matrices that are used in several important communication algorithms. The performance benefits are evaluated on the Sandbridge Sandblaster SB3000 low-power, multithreaded SDR processor. The proposed instructions provide significant performance improvements with little hardware overhead, while improving the accuracy of the wireless algorithms under investigation.