FFT implementation with Multi-operand floating point units

In this paper we propose two new design techniques for floating point arithmetic units used in DSP applications, and apply them in the design of two multi-operand units for high-radix FFTs. These two multi-operand arithmetic units are a Modified two term Dot Product (MDP) unit and a Multi-operand Add-Sub unit (MAS). A radix-4 FFT butterfly computation block is implemented efficiently with these multi-operand units. Synthesis results show that the butterfly unit built with our design is about 43% faster and 9.5% smaller than a conventional implementation.