µSHARK: Software Radio Tools for Power-Efficient Programming of Multi-Core and Many-Core Digital Signal Processors

Modern multi-core and many-core digital signal processors (mDSPs) offer greater processing capabilities than any other reprogrammable device, with some even performing on par with an FPGA. mDSP development environments support standard languages such as C/C++, and provide application scheduling optimized for throughput or latency. In order to provide power-optimal application scheduling, e.g., on a handheld battery powered device utilizing an mDSP, a new tool must be created. A properly created software radio is an ideal candidate for this tool, as it can provide for implicit code parallelization and power minimization via scheduling of tasks. RFware and the University of Notre Dame propose to develop a tool that facilitates power-optimized programming of communications algorithm implementations on multi-core and many-core digital signal processors. RFware’s proposed tool consists of three primary components: an application for converting existing C/C++/VHDL code into a software radio-based data-flow representation; a power-aware scheduling algorithm for mapping of software radio tasks onto processing cores; and a tool for compiling software radio applications for deployment on embedded devices. Our solution transfers the capabilities of RFware’s µSHARK software radio onto mDSPs, simultaneously providing algorithm implementation parallelization and power consumption minimization.