A CORDIC-friendly FFT architecture

Fast Fourier Transform (FFT) is one of the basic building blocks in signal processing and communications systems. The butterflies-based structure of the FFT is the main reason for the reduced number of arithmetic operations required to implement the transform. From implementation point of view, the complex rotations used in butterflies can be implemented by using COordinate Rotation DIgital Computer (CORDIC). This implementation strategy reduces the hardware complexity compared to the direct implementation of the butterflies using complex multipliers. In this paper, we introduce a restructure of the butterflies of the radix-2 FFT to be more CORDIC friendly. This algorithm-level modification of the FFT is friendly towards all CORDIC types, including those introducing non-fixed gain. Compared with the conventional radix-2 FFT algorithm, the proposed algorithm introduces a substantial increase in performance. For example, it achieves superior signal to quantization noise ratio (SQNR), with around 14 dB gain for 8 to 1024 points FFT. In addition, in pipeline architectures the modification leads to an improvement in latency or a reduction in the total area, with an improvement in either of 38% for 1024 points FFT.