Linear Transforms: From Math to Efficient Hardware

Introduction. Linear transforms (such as the discrete Fourier transform) are widely used building blocks in signal processing applications. The domain of linear transforms and their algorithms is well understood mathematically. In this work, we utilize this math-level knowledge to construct a formula driven domain specific high-level synthesis tool. This tool is able to generate efficient register transfer level designs from a mathematical description of the problem. The tool’s flexibility allows its generated designs to span a wide cost/performance tradeoff space.